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 | ||
afe0e395 MW |
83 | #define RADIX_TREE_RETRY ptr_to_indirect(NULL) |
84 | ||
db050f29 MW |
85 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
86 | /* Sibling slots point directly to another slot in the same node */ | |
87 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
88 | { | |
89 | void **ptr = node; | |
90 | return (parent->slots <= ptr) && | |
91 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
92 | } | |
93 | #else | |
94 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
95 | { | |
96 | return false; | |
97 | } | |
98 | #endif | |
99 | ||
100 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
101 | void **slot) | |
102 | { | |
103 | return slot - parent->slots; | |
104 | } | |
105 | ||
106 | static unsigned radix_tree_descend(struct radix_tree_node *parent, | |
107 | struct radix_tree_node **nodep, unsigned offset) | |
108 | { | |
109 | void **entry = rcu_dereference_raw(parent->slots[offset]); | |
110 | ||
111 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
112 | if (radix_tree_is_indirect_ptr(entry)) { | |
113 | unsigned long siboff = get_slot_offset(parent, entry); | |
114 | if (siboff < RADIX_TREE_MAP_SIZE) { | |
115 | offset = siboff; | |
116 | entry = rcu_dereference_raw(parent->slots[offset]); | |
117 | } | |
118 | } | |
119 | #endif | |
120 | ||
121 | *nodep = (void *)entry; | |
122 | return offset; | |
123 | } | |
124 | ||
612d6c19 NP |
125 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
126 | { | |
127 | return root->gfp_mask & __GFP_BITS_MASK; | |
128 | } | |
129 | ||
643b52b9 NP |
130 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
131 | int offset) | |
132 | { | |
133 | __set_bit(offset, node->tags[tag]); | |
134 | } | |
135 | ||
136 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
137 | int offset) | |
138 | { | |
139 | __clear_bit(offset, node->tags[tag]); | |
140 | } | |
141 | ||
142 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
143 | int offset) | |
144 | { | |
145 | return test_bit(offset, node->tags[tag]); | |
146 | } | |
147 | ||
148 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
149 | { | |
150 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
151 | } | |
152 | ||
153 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
154 | { | |
155 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
156 | } | |
157 | ||
158 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
159 | { | |
160 | root->gfp_mask &= __GFP_BITS_MASK; | |
161 | } | |
162 | ||
163 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
164 | { | |
165 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
166 | } | |
167 | ||
168 | /* | |
169 | * Returns 1 if any slot in the node has this tag set. | |
170 | * Otherwise returns 0. | |
171 | */ | |
172 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
173 | { | |
174 | int idx; | |
175 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
176 | if (node->tags[tag][idx]) | |
177 | return 1; | |
178 | } | |
179 | return 0; | |
180 | } | |
78c1d784 KK |
181 | |
182 | /** | |
183 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
184 | * | |
185 | * @addr: The address to base the search on | |
186 | * @size: The bitmap size in bits | |
187 | * @offset: The bitnumber to start searching at | |
188 | * | |
189 | * Unrollable variant of find_next_bit() for constant size arrays. | |
190 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
191 | * Returns next bit offset, or size if nothing found. | |
192 | */ | |
193 | static __always_inline unsigned long | |
194 | radix_tree_find_next_bit(const unsigned long *addr, | |
195 | unsigned long size, unsigned long offset) | |
196 | { | |
197 | if (!__builtin_constant_p(size)) | |
198 | return find_next_bit(addr, size, offset); | |
199 | ||
200 | if (offset < size) { | |
201 | unsigned long tmp; | |
202 | ||
203 | addr += offset / BITS_PER_LONG; | |
204 | tmp = *addr >> (offset % BITS_PER_LONG); | |
205 | if (tmp) | |
206 | return __ffs(tmp) + offset; | |
207 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
208 | while (offset < size) { | |
209 | tmp = *++addr; | |
210 | if (tmp) | |
211 | return __ffs(tmp) + offset; | |
212 | offset += BITS_PER_LONG; | |
213 | } | |
214 | } | |
215 | return size; | |
216 | } | |
217 | ||
7cf19af4 MW |
218 | #if 0 |
219 | static void dump_node(void *slot, int height, int offset) | |
220 | { | |
221 | struct radix_tree_node *node; | |
222 | int i; | |
223 | ||
224 | if (!slot) | |
225 | return; | |
226 | ||
227 | if (height == 0) { | |
228 | pr_debug("radix entry %p offset %d\n", slot, offset); | |
229 | return; | |
230 | } | |
231 | ||
232 | node = indirect_to_ptr(slot); | |
233 | pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n", | |
234 | slot, offset, node->tags[0][0], node->tags[1][0], | |
235 | node->tags[2][0], node->path, node->count, node->parent); | |
236 | ||
237 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) | |
238 | dump_node(node->slots[i], height - 1, i); | |
239 | } | |
240 | ||
241 | /* For debug */ | |
242 | static void radix_tree_dump(struct radix_tree_root *root) | |
243 | { | |
244 | pr_debug("radix root: %p height %d rnode %p tags %x\n", | |
245 | root, root->height, root->rnode, | |
246 | root->gfp_mask >> __GFP_BITS_SHIFT); | |
247 | if (!radix_tree_is_indirect_ptr(root->rnode)) | |
248 | return; | |
249 | dump_node(root->rnode, root->height, 0); | |
250 | } | |
251 | #endif | |
252 | ||
1da177e4 LT |
253 | /* |
254 | * This assumes that the caller has performed appropriate preallocation, and | |
255 | * that the caller has pinned this thread of control to the current CPU. | |
256 | */ | |
257 | static struct radix_tree_node * | |
258 | radix_tree_node_alloc(struct radix_tree_root *root) | |
259 | { | |
e2848a0e | 260 | struct radix_tree_node *ret = NULL; |
612d6c19 | 261 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 262 | |
5e4c0d97 JK |
263 | /* |
264 | * Preload code isn't irq safe and it doesn't make sence to use | |
265 | * preloading in the interrupt anyway as all the allocations have to | |
266 | * be atomic. So just do normal allocation when in interrupt. | |
267 | */ | |
d0164adc | 268 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
269 | struct radix_tree_preload *rtp; |
270 | ||
58e698af VD |
271 | /* |
272 | * Even if the caller has preloaded, try to allocate from the | |
273 | * cache first for the new node to get accounted. | |
274 | */ | |
275 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
276 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
277 | if (ret) | |
278 | goto out; | |
279 | ||
e2848a0e NP |
280 | /* |
281 | * Provided the caller has preloaded here, we will always | |
282 | * succeed in getting a node here (and never reach | |
283 | * kmem_cache_alloc) | |
284 | */ | |
7c8e0181 | 285 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 286 | if (rtp->nr) { |
9d2a8da0 KS |
287 | ret = rtp->nodes; |
288 | rtp->nodes = ret->private_data; | |
289 | ret->private_data = NULL; | |
1da177e4 LT |
290 | rtp->nr--; |
291 | } | |
ce80b067 CM |
292 | /* |
293 | * Update the allocation stack trace as this is more useful | |
294 | * for debugging. | |
295 | */ | |
296 | kmemleak_update_trace(ret); | |
58e698af | 297 | goto out; |
1da177e4 | 298 | } |
58e698af VD |
299 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
300 | gfp_mask | __GFP_ACCOUNT); | |
301 | out: | |
c0bc9875 | 302 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
303 | return ret; |
304 | } | |
305 | ||
7cf9c2c7 NP |
306 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
307 | { | |
308 | struct radix_tree_node *node = | |
309 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 310 | int i; |
643b52b9 NP |
311 | |
312 | /* | |
313 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
314 | * can leave us with a non-NULL entry in the first slot, so clear | |
315 | * that here to make sure. | |
316 | */ | |
b6dd0865 DC |
317 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
318 | tag_clear(node, i, 0); | |
319 | ||
643b52b9 NP |
320 | node->slots[0] = NULL; |
321 | node->count = 0; | |
322 | ||
7cf9c2c7 NP |
323 | kmem_cache_free(radix_tree_node_cachep, node); |
324 | } | |
325 | ||
1da177e4 LT |
326 | static inline void |
327 | radix_tree_node_free(struct radix_tree_node *node) | |
328 | { | |
7cf9c2c7 | 329 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
330 | } |
331 | ||
332 | /* | |
333 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
334 | * ensure that the addition of a single element in the tree cannot fail. On | |
335 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
336 | * with preemption not disabled. | |
b34df792 DH |
337 | * |
338 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 339 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 340 | */ |
5e4c0d97 | 341 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
342 | { |
343 | struct radix_tree_preload *rtp; | |
344 | struct radix_tree_node *node; | |
345 | int ret = -ENOMEM; | |
346 | ||
347 | preempt_disable(); | |
7c8e0181 | 348 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 349 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 350 | preempt_enable(); |
488514d1 | 351 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
352 | if (node == NULL) |
353 | goto out; | |
354 | preempt_disable(); | |
7c8e0181 | 355 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
356 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
357 | node->private_data = rtp->nodes; | |
358 | rtp->nodes = node; | |
359 | rtp->nr++; | |
360 | } else { | |
1da177e4 | 361 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 362 | } |
1da177e4 LT |
363 | } |
364 | ret = 0; | |
365 | out: | |
366 | return ret; | |
367 | } | |
5e4c0d97 JK |
368 | |
369 | /* | |
370 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
371 | * ensure that the addition of a single element in the tree cannot fail. On | |
372 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
373 | * with preemption not disabled. | |
374 | * | |
375 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 376 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
377 | */ |
378 | int radix_tree_preload(gfp_t gfp_mask) | |
379 | { | |
380 | /* Warn on non-sensical use... */ | |
d0164adc | 381 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
382 | return __radix_tree_preload(gfp_mask); |
383 | } | |
d7f0923d | 384 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 385 | |
5e4c0d97 JK |
386 | /* |
387 | * The same as above function, except we don't guarantee preloading happens. | |
388 | * We do it, if we decide it helps. On success, return zero with preemption | |
389 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
390 | */ | |
391 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
392 | { | |
d0164adc | 393 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
394 | return __radix_tree_preload(gfp_mask); |
395 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
396 | preempt_disable(); | |
397 | return 0; | |
398 | } | |
399 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
400 | ||
1da177e4 LT |
401 | /* |
402 | * Return the maximum key which can be store into a | |
403 | * radix tree with height HEIGHT. | |
404 | */ | |
405 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
406 | { | |
407 | return height_to_maxindex[height]; | |
408 | } | |
409 | ||
1456a439 MW |
410 | static inline unsigned long node_maxindex(struct radix_tree_node *node) |
411 | { | |
412 | return radix_tree_maxindex(node->path & RADIX_TREE_HEIGHT_MASK); | |
413 | } | |
414 | ||
415 | static unsigned radix_tree_load_root(struct radix_tree_root *root, | |
416 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
417 | { | |
418 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
419 | ||
420 | *nodep = node; | |
421 | ||
422 | if (likely(radix_tree_is_indirect_ptr(node))) { | |
423 | node = indirect_to_ptr(node); | |
424 | *maxindex = node_maxindex(node); | |
425 | return (node->path & RADIX_TREE_HEIGHT_MASK) * | |
426 | RADIX_TREE_MAP_SHIFT; | |
427 | } | |
428 | ||
429 | *maxindex = 0; | |
430 | return 0; | |
431 | } | |
432 | ||
1da177e4 LT |
433 | /* |
434 | * Extend a radix tree so it can store key @index. | |
435 | */ | |
e6145236 | 436 | static int radix_tree_extend(struct radix_tree_root *root, |
49ea6ebc | 437 | unsigned long index) |
1da177e4 LT |
438 | { |
439 | struct radix_tree_node *node; | |
e2bdb933 | 440 | struct radix_tree_node *slot; |
1da177e4 | 441 | unsigned int height; |
1da177e4 LT |
442 | int tag; |
443 | ||
444 | /* Figure out what the height should be. */ | |
445 | height = root->height + 1; | |
446 | while (index > radix_tree_maxindex(height)) | |
447 | height++; | |
448 | ||
49ea6ebc | 449 | if (root->rnode == NULL) { |
1da177e4 LT |
450 | root->height = height; |
451 | goto out; | |
452 | } | |
453 | ||
1da177e4 | 454 | do { |
7cf9c2c7 | 455 | unsigned int newheight; |
1da177e4 LT |
456 | if (!(node = radix_tree_node_alloc(root))) |
457 | return -ENOMEM; | |
458 | ||
1da177e4 | 459 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 460 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 461 | if (root_tag_get(root, tag)) |
1da177e4 LT |
462 | tag_set(node, tag, 0); |
463 | } | |
464 | ||
e2bdb933 | 465 | /* Increase the height. */ |
7cf9c2c7 | 466 | newheight = root->height+1; |
449dd698 JW |
467 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
468 | node->path = newheight; | |
1da177e4 | 469 | node->count = 1; |
e2bdb933 HD |
470 | node->parent = NULL; |
471 | slot = root->rnode; | |
49ea6ebc | 472 | if (radix_tree_is_indirect_ptr(slot)) { |
e2bdb933 HD |
473 | slot = indirect_to_ptr(slot); |
474 | slot->parent = node; | |
339e6353 | 475 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
476 | } |
477 | node->slots[0] = slot; | |
27d20fdd | 478 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
479 | rcu_assign_pointer(root->rnode, node); |
480 | root->height = newheight; | |
1da177e4 LT |
481 | } while (height > root->height); |
482 | out: | |
49ea6ebc | 483 | return height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
484 | } |
485 | ||
486 | /** | |
139e5616 | 487 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
488 | * @root: radix tree root |
489 | * @index: index key | |
e6145236 | 490 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
491 | * @nodep: returns node |
492 | * @slotp: returns slot | |
1da177e4 | 493 | * |
139e5616 JW |
494 | * Create, if necessary, and return the node and slot for an item |
495 | * at position @index in the radix tree @root. | |
496 | * | |
497 | * Until there is more than one item in the tree, no nodes are | |
498 | * allocated and @root->rnode is used as a direct slot instead of | |
499 | * pointing to a node, in which case *@nodep will be NULL. | |
500 | * | |
501 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 502 | */ |
139e5616 | 503 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
504 | unsigned order, struct radix_tree_node **nodep, |
505 | void ***slotp) | |
1da177e4 | 506 | { |
201b6264 | 507 | struct radix_tree_node *node = NULL, *slot; |
49ea6ebc | 508 | unsigned long maxindex; |
139e5616 | 509 | unsigned int height, shift, offset; |
49ea6ebc MW |
510 | unsigned long max = index | ((1UL << order) - 1); |
511 | ||
512 | shift = radix_tree_load_root(root, &slot, &maxindex); | |
1da177e4 LT |
513 | |
514 | /* Make sure the tree is high enough. */ | |
49ea6ebc MW |
515 | if (max > maxindex) { |
516 | int error = radix_tree_extend(root, max); | |
517 | if (error < 0) | |
1da177e4 | 518 | return error; |
49ea6ebc MW |
519 | shift = error; |
520 | slot = root->rnode; | |
521 | if (order == shift) { | |
522 | shift += RADIX_TREE_MAP_SHIFT; | |
523 | root->height++; | |
524 | } | |
1da177e4 LT |
525 | } |
526 | ||
1da177e4 | 527 | height = root->height; |
1da177e4 LT |
528 | |
529 | offset = 0; /* uninitialised var warning */ | |
e6145236 | 530 | while (shift > order) { |
201b6264 | 531 | if (slot == NULL) { |
1da177e4 | 532 | /* Have to add a child node. */ |
201b6264 | 533 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 534 | return -ENOMEM; |
449dd698 | 535 | slot->path = height; |
e2bdb933 | 536 | slot->parent = node; |
201b6264 | 537 | if (node) { |
339e6353 MW |
538 | rcu_assign_pointer(node->slots[offset], |
539 | ptr_to_indirect(slot)); | |
1da177e4 | 540 | node->count++; |
449dd698 | 541 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 542 | } else |
339e6353 MW |
543 | rcu_assign_pointer(root->rnode, |
544 | ptr_to_indirect(slot)); | |
e6145236 MW |
545 | } else if (!radix_tree_is_indirect_ptr(slot)) |
546 | break; | |
1da177e4 LT |
547 | |
548 | /* Go a level down */ | |
e6145236 | 549 | height--; |
0070e28d | 550 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 551 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e6145236 | 552 | node = indirect_to_ptr(slot); |
201b6264 | 553 | slot = node->slots[offset]; |
e6145236 MW |
554 | } |
555 | ||
57578c2e | 556 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 557 | /* Insert pointers to the canonical entry */ |
3b8c00f6 MW |
558 | if (order > shift) { |
559 | int i, n = 1 << (order - shift); | |
e6145236 MW |
560 | offset = offset & ~(n - 1); |
561 | slot = ptr_to_indirect(&node->slots[offset]); | |
562 | for (i = 0; i < n; i++) { | |
563 | if (node->slots[offset + i]) | |
564 | return -EEXIST; | |
565 | } | |
566 | ||
567 | for (i = 1; i < n; i++) { | |
568 | rcu_assign_pointer(node->slots[offset + i], slot); | |
569 | node->count++; | |
570 | } | |
612d6c19 | 571 | } |
57578c2e | 572 | #endif |
1da177e4 | 573 | |
139e5616 JW |
574 | if (nodep) |
575 | *nodep = node; | |
576 | if (slotp) | |
577 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
578 | return 0; | |
579 | } | |
580 | ||
581 | /** | |
e6145236 | 582 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
583 | * @root: radix tree root |
584 | * @index: index key | |
e6145236 | 585 | * @order: key covers the 2^order indices around index |
139e5616 JW |
586 | * @item: item to insert |
587 | * | |
588 | * Insert an item into the radix tree at position @index. | |
589 | */ | |
e6145236 MW |
590 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
591 | unsigned order, void *item) | |
139e5616 JW |
592 | { |
593 | struct radix_tree_node *node; | |
594 | void **slot; | |
595 | int error; | |
596 | ||
597 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
598 | ||
e6145236 | 599 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
600 | if (error) |
601 | return error; | |
602 | if (*slot != NULL) | |
1da177e4 | 603 | return -EEXIST; |
139e5616 | 604 | rcu_assign_pointer(*slot, item); |
201b6264 | 605 | |
612d6c19 NP |
606 | if (node) { |
607 | node->count++; | |
139e5616 JW |
608 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
609 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 610 | } else { |
612d6c19 NP |
611 | BUG_ON(root_tag_get(root, 0)); |
612 | BUG_ON(root_tag_get(root, 1)); | |
613 | } | |
1da177e4 | 614 | |
1da177e4 LT |
615 | return 0; |
616 | } | |
e6145236 | 617 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 618 | |
139e5616 JW |
619 | /** |
620 | * __radix_tree_lookup - lookup an item in a radix tree | |
621 | * @root: radix tree root | |
622 | * @index: index key | |
623 | * @nodep: returns node | |
624 | * @slotp: returns slot | |
625 | * | |
626 | * Lookup and return the item at position @index in the radix | |
627 | * tree @root. | |
628 | * | |
629 | * Until there is more than one item in the tree, no nodes are | |
630 | * allocated and @root->rnode is used as a direct slot instead of | |
631 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 632 | */ |
139e5616 JW |
633 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
634 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 635 | { |
139e5616 | 636 | struct radix_tree_node *node, *parent; |
1da177e4 | 637 | unsigned int height, shift; |
139e5616 | 638 | void **slot; |
612d6c19 | 639 | |
2676a58c | 640 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 641 | if (node == NULL) |
1da177e4 LT |
642 | return NULL; |
643 | ||
c0bc9875 | 644 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
645 | if (index > 0) |
646 | return NULL; | |
139e5616 JW |
647 | |
648 | if (nodep) | |
649 | *nodep = NULL; | |
650 | if (slotp) | |
651 | *slotp = (void **)&root->rnode; | |
652 | return node; | |
7cf9c2c7 | 653 | } |
27d20fdd | 654 | node = indirect_to_ptr(node); |
7cf9c2c7 | 655 | |
449dd698 | 656 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
657 | if (index > radix_tree_maxindex(height)) |
658 | return NULL; | |
612d6c19 | 659 | |
1da177e4 | 660 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 661 | |
7cf9c2c7 | 662 | do { |
139e5616 JW |
663 | parent = node; |
664 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 665 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 666 | if (node == NULL) |
1da177e4 | 667 | return NULL; |
e6145236 MW |
668 | if (!radix_tree_is_indirect_ptr(node)) |
669 | break; | |
339e6353 | 670 | node = indirect_to_ptr(node); |
1da177e4 | 671 | |
1da177e4 LT |
672 | shift -= RADIX_TREE_MAP_SHIFT; |
673 | height--; | |
7cf9c2c7 | 674 | } while (height > 0); |
1da177e4 | 675 | |
139e5616 JW |
676 | if (nodep) |
677 | *nodep = parent; | |
678 | if (slotp) | |
679 | *slotp = slot; | |
680 | return node; | |
b72b71c6 HS |
681 | } |
682 | ||
683 | /** | |
684 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
685 | * @root: radix tree root | |
686 | * @index: index key | |
687 | * | |
688 | * Returns: the slot corresponding to the position @index in the | |
689 | * radix tree @root. This is useful for update-if-exists operations. | |
690 | * | |
691 | * This function can be called under rcu_read_lock iff the slot is not | |
692 | * modified by radix_tree_replace_slot, otherwise it must be called | |
693 | * exclusive from other writers. Any dereference of the slot must be done | |
694 | * using radix_tree_deref_slot. | |
695 | */ | |
696 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
697 | { | |
139e5616 JW |
698 | void **slot; |
699 | ||
700 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
701 | return NULL; | |
702 | return slot; | |
a4331366 | 703 | } |
a4331366 HR |
704 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
705 | ||
706 | /** | |
707 | * radix_tree_lookup - perform lookup operation on a radix tree | |
708 | * @root: radix tree root | |
709 | * @index: index key | |
710 | * | |
711 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
712 | * |
713 | * This function can be called under rcu_read_lock, however the caller | |
714 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
715 | * them safely). No RCU barriers are required to access or modify the | |
716 | * returned item, however. | |
a4331366 HR |
717 | */ |
718 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
719 | { | |
139e5616 | 720 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
721 | } |
722 | EXPORT_SYMBOL(radix_tree_lookup); | |
723 | ||
724 | /** | |
725 | * radix_tree_tag_set - set a tag on a radix tree node | |
726 | * @root: radix tree root | |
727 | * @index: index key | |
728 | * @tag: tag index | |
729 | * | |
daff89f3 JC |
730 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
731 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
732 | * the root all the way down to the leaf node. |
733 | * | |
734 | * Returns the address of the tagged item. Setting a tag on a not-present | |
735 | * item is a bug. | |
736 | */ | |
737 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 738 | unsigned long index, unsigned int tag) |
1da177e4 LT |
739 | { |
740 | unsigned int height, shift; | |
201b6264 | 741 | struct radix_tree_node *slot; |
1da177e4 LT |
742 | |
743 | height = root->height; | |
4c91c364 | 744 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 745 | |
27d20fdd | 746 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 747 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
748 | |
749 | while (height > 0) { | |
750 | int offset; | |
751 | ||
752 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
753 | if (!tag_get(slot, tag, offset)) |
754 | tag_set(slot, tag, offset); | |
201b6264 CL |
755 | slot = slot->slots[offset]; |
756 | BUG_ON(slot == NULL); | |
e6145236 MW |
757 | if (!radix_tree_is_indirect_ptr(slot)) |
758 | break; | |
339e6353 | 759 | slot = indirect_to_ptr(slot); |
1da177e4 LT |
760 | shift -= RADIX_TREE_MAP_SHIFT; |
761 | height--; | |
762 | } | |
763 | ||
612d6c19 NP |
764 | /* set the root's tag bit */ |
765 | if (slot && !root_tag_get(root, tag)) | |
766 | root_tag_set(root, tag); | |
767 | ||
201b6264 | 768 | return slot; |
1da177e4 LT |
769 | } |
770 | EXPORT_SYMBOL(radix_tree_tag_set); | |
771 | ||
772 | /** | |
773 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
774 | * @root: radix tree root | |
775 | * @index: index key | |
776 | * @tag: tag index | |
777 | * | |
daff89f3 JC |
778 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
779 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
780 | * this causes the leaf node to have no tags set then clear the tag in the |
781 | * next-to-leaf node, etc. | |
782 | * | |
783 | * Returns the address of the tagged item on success, else NULL. ie: | |
784 | * has the same return value and semantics as radix_tree_lookup(). | |
785 | */ | |
786 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 787 | unsigned long index, unsigned int tag) |
1da177e4 | 788 | { |
e2bdb933 | 789 | struct radix_tree_node *node = NULL; |
612d6c19 | 790 | struct radix_tree_node *slot = NULL; |
1da177e4 | 791 | unsigned int height, shift; |
e2bdb933 | 792 | int uninitialized_var(offset); |
1da177e4 LT |
793 | |
794 | height = root->height; | |
795 | if (index > radix_tree_maxindex(height)) | |
796 | goto out; | |
797 | ||
e2bdb933 | 798 | shift = height * RADIX_TREE_MAP_SHIFT; |
339e6353 | 799 | slot = root->rnode; |
1da177e4 | 800 | |
e2bdb933 | 801 | while (shift) { |
201b6264 | 802 | if (slot == NULL) |
1da177e4 | 803 | goto out; |
e6145236 MW |
804 | if (!radix_tree_is_indirect_ptr(slot)) |
805 | break; | |
339e6353 | 806 | slot = indirect_to_ptr(slot); |
1da177e4 | 807 | |
e2bdb933 | 808 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 809 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 810 | node = slot; |
201b6264 | 811 | slot = slot->slots[offset]; |
1da177e4 LT |
812 | } |
813 | ||
612d6c19 | 814 | if (slot == NULL) |
1da177e4 LT |
815 | goto out; |
816 | ||
e2bdb933 HD |
817 | while (node) { |
818 | if (!tag_get(node, tag, offset)) | |
d5274261 | 819 | goto out; |
e2bdb933 HD |
820 | tag_clear(node, tag, offset); |
821 | if (any_tag_set(node, tag)) | |
6e954b9e | 822 | goto out; |
e2bdb933 HD |
823 | |
824 | index >>= RADIX_TREE_MAP_SHIFT; | |
825 | offset = index & RADIX_TREE_MAP_MASK; | |
826 | node = node->parent; | |
612d6c19 NP |
827 | } |
828 | ||
829 | /* clear the root's tag bit */ | |
830 | if (root_tag_get(root, tag)) | |
831 | root_tag_clear(root, tag); | |
832 | ||
1da177e4 | 833 | out: |
612d6c19 | 834 | return slot; |
1da177e4 LT |
835 | } |
836 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
837 | ||
1da177e4 | 838 | /** |
32605a18 MT |
839 | * radix_tree_tag_get - get a tag on a radix tree node |
840 | * @root: radix tree root | |
841 | * @index: index key | |
daff89f3 | 842 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 843 | * |
32605a18 | 844 | * Return values: |
1da177e4 | 845 | * |
612d6c19 NP |
846 | * 0: tag not present or not set |
847 | * 1: tag set | |
ce82653d DH |
848 | * |
849 | * Note that the return value of this function may not be relied on, even if | |
850 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
851 | * from concurrency. | |
1da177e4 LT |
852 | */ |
853 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 854 | unsigned long index, unsigned int tag) |
1da177e4 LT |
855 | { |
856 | unsigned int height, shift; | |
7cf9c2c7 | 857 | struct radix_tree_node *node; |
1da177e4 | 858 | |
612d6c19 NP |
859 | /* check the root's tag bit */ |
860 | if (!root_tag_get(root, tag)) | |
861 | return 0; | |
862 | ||
2676a58c | 863 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
864 | if (node == NULL) |
865 | return 0; | |
866 | ||
c0bc9875 | 867 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 868 | return (index == 0); |
27d20fdd | 869 | node = indirect_to_ptr(node); |
7cf9c2c7 | 870 | |
449dd698 | 871 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
872 | if (index > radix_tree_maxindex(height)) |
873 | return 0; | |
612d6c19 | 874 | |
1da177e4 | 875 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
876 | |
877 | for ( ; ; ) { | |
878 | int offset; | |
879 | ||
7cf9c2c7 | 880 | if (node == NULL) |
1da177e4 | 881 | return 0; |
339e6353 | 882 | node = indirect_to_ptr(node); |
1da177e4 LT |
883 | |
884 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 885 | if (!tag_get(node, tag, offset)) |
3fa36acb | 886 | return 0; |
ce82653d | 887 | if (height == 1) |
3fa36acb | 888 | return 1; |
2676a58c | 889 | node = rcu_dereference_raw(node->slots[offset]); |
e6145236 MW |
890 | if (!radix_tree_is_indirect_ptr(node)) |
891 | return 1; | |
1da177e4 LT |
892 | shift -= RADIX_TREE_MAP_SHIFT; |
893 | height--; | |
894 | } | |
895 | } | |
896 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 897 | |
78c1d784 KK |
898 | /** |
899 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
900 | * | |
901 | * @root: radix tree root | |
902 | * @iter: iterator state | |
903 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
904 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
905 | */ | |
906 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
907 | struct radix_tree_iter *iter, unsigned flags) | |
908 | { | |
909 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
910 | struct radix_tree_node *rnode, *node; | |
449dd698 | 911 | unsigned long index, offset, height; |
78c1d784 KK |
912 | |
913 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
914 | return NULL; | |
915 | ||
916 | /* | |
917 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
918 | * during iterating; it can be zero only at the beginning. | |
919 | * And we cannot overflow iter->next_index in a single step, | |
920 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
921 | * |
922 | * This condition also used by radix_tree_next_slot() to stop | |
923 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
924 | */ |
925 | index = iter->next_index; | |
926 | if (!index && iter->index) | |
927 | return NULL; | |
928 | ||
929 | rnode = rcu_dereference_raw(root->rnode); | |
930 | if (radix_tree_is_indirect_ptr(rnode)) { | |
931 | rnode = indirect_to_ptr(rnode); | |
932 | } else if (rnode && !index) { | |
933 | /* Single-slot tree */ | |
934 | iter->index = 0; | |
935 | iter->next_index = 1; | |
936 | iter->tags = 1; | |
937 | return (void **)&root->rnode; | |
938 | } else | |
939 | return NULL; | |
940 | ||
941 | restart: | |
449dd698 JW |
942 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
943 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
944 | offset = index >> shift; |
945 | ||
946 | /* Index outside of the tree */ | |
947 | if (offset >= RADIX_TREE_MAP_SIZE) | |
948 | return NULL; | |
949 | ||
950 | node = rnode; | |
951 | while (1) { | |
e6145236 | 952 | struct radix_tree_node *slot; |
78c1d784 KK |
953 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
954 | !test_bit(offset, node->tags[tag]) : | |
955 | !node->slots[offset]) { | |
956 | /* Hole detected */ | |
957 | if (flags & RADIX_TREE_ITER_CONTIG) | |
958 | return NULL; | |
959 | ||
960 | if (flags & RADIX_TREE_ITER_TAGGED) | |
961 | offset = radix_tree_find_next_bit( | |
962 | node->tags[tag], | |
963 | RADIX_TREE_MAP_SIZE, | |
964 | offset + 1); | |
965 | else | |
966 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
967 | if (node->slots[offset]) | |
968 | break; | |
969 | } | |
970 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
971 | index += offset << shift; | |
972 | /* Overflow after ~0UL */ | |
973 | if (!index) | |
974 | return NULL; | |
975 | if (offset == RADIX_TREE_MAP_SIZE) | |
976 | goto restart; | |
977 | } | |
978 | ||
979 | /* This is leaf-node */ | |
980 | if (!shift) | |
981 | break; | |
982 | ||
e6145236 MW |
983 | slot = rcu_dereference_raw(node->slots[offset]); |
984 | if (slot == NULL) | |
78c1d784 | 985 | goto restart; |
e6145236 MW |
986 | if (!radix_tree_is_indirect_ptr(slot)) |
987 | break; | |
988 | node = indirect_to_ptr(slot); | |
78c1d784 KK |
989 | shift -= RADIX_TREE_MAP_SHIFT; |
990 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
991 | } | |
992 | ||
993 | /* Update the iterator state */ | |
994 | iter->index = index; | |
995 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
996 | ||
997 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
998 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
999 | unsigned tag_long, tag_bit; | |
1000 | ||
1001 | tag_long = offset / BITS_PER_LONG; | |
1002 | tag_bit = offset % BITS_PER_LONG; | |
1003 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1004 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1005 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1006 | /* Pick tags from next element */ | |
1007 | if (tag_bit) | |
1008 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1009 | (BITS_PER_LONG - tag_bit); | |
1010 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1011 | iter->next_index = index + BITS_PER_LONG; | |
1012 | } | |
1013 | } | |
1014 | ||
1015 | return node->slots + offset; | |
1016 | } | |
1017 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1018 | ||
ebf8aa44 JK |
1019 | /** |
1020 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
1021 | * tag if item has another tag set | |
1022 | * @root: radix tree root | |
1023 | * @first_indexp: pointer to a starting index of a range to scan | |
1024 | * @last_index: last index of a range to scan | |
1025 | * @nr_to_tag: maximum number items to tag | |
1026 | * @iftag: tag index to test | |
1027 | * @settag: tag index to set if tested tag is set | |
1028 | * | |
1029 | * This function scans range of radix tree from first_index to last_index | |
1030 | * (inclusive). For each item in the range if iftag is set, the function sets | |
1031 | * also settag. The function stops either after tagging nr_to_tag items or | |
1032 | * after reaching last_index. | |
1033 | * | |
144dcfc0 DC |
1034 | * The tags must be set from the leaf level only and propagated back up the |
1035 | * path to the root. We must do this so that we resolve the full path before | |
1036 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
1037 | * we can get to the leaf node and find that the index that has the iftag | |
1038 | * set is outside the range we are scanning. This reults in dangling tags and | |
1039 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1040 | * | |
ebf8aa44 JK |
1041 | * The function returns number of leaves where the tag was set and sets |
1042 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
1043 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1044 | * be prepared to handle that. | |
ebf8aa44 JK |
1045 | */ |
1046 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1047 | unsigned long *first_indexp, unsigned long last_index, | |
1048 | unsigned long nr_to_tag, | |
1049 | unsigned int iftag, unsigned int settag) | |
1050 | { | |
144dcfc0 | 1051 | unsigned int height = root->height; |
e2bdb933 | 1052 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
1053 | struct radix_tree_node *slot; |
1054 | unsigned int shift; | |
1055 | unsigned long tagged = 0; | |
1056 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
1057 | |
1058 | last_index = min(last_index, radix_tree_maxindex(height)); | |
1059 | if (index > last_index) | |
1060 | return 0; | |
1061 | if (!nr_to_tag) | |
1062 | return 0; | |
1063 | if (!root_tag_get(root, iftag)) { | |
1064 | *first_indexp = last_index + 1; | |
1065 | return 0; | |
1066 | } | |
1067 | if (height == 0) { | |
1068 | *first_indexp = last_index + 1; | |
1069 | root_tag_set(root, settag); | |
1070 | return 1; | |
1071 | } | |
1072 | ||
1073 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 1074 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
1075 | |
1076 | for (;;) { | |
e2bdb933 | 1077 | unsigned long upindex; |
ebf8aa44 JK |
1078 | int offset; |
1079 | ||
1080 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1081 | if (!slot->slots[offset]) | |
1082 | goto next; | |
1083 | if (!tag_get(slot, iftag, offset)) | |
1084 | goto next; | |
e2bdb933 | 1085 | if (shift) { |
e2bdb933 | 1086 | node = slot; |
144dcfc0 | 1087 | slot = slot->slots[offset]; |
e6145236 MW |
1088 | if (radix_tree_is_indirect_ptr(slot)) { |
1089 | slot = indirect_to_ptr(slot); | |
1090 | shift -= RADIX_TREE_MAP_SHIFT; | |
1091 | continue; | |
1092 | } else { | |
1093 | slot = node; | |
1094 | node = node->parent; | |
1095 | } | |
144dcfc0 DC |
1096 | } |
1097 | ||
1098 | /* tag the leaf */ | |
e6145236 | 1099 | tagged += 1 << shift; |
ebf8aa44 | 1100 | tag_set(slot, settag, offset); |
144dcfc0 DC |
1101 | |
1102 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
1103 | upindex = index; |
1104 | while (node) { | |
1105 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
1106 | offset = upindex & RADIX_TREE_MAP_MASK; | |
1107 | ||
144dcfc0 | 1108 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 1109 | if (tag_get(node, settag, offset)) |
144dcfc0 | 1110 | break; |
e2bdb933 HD |
1111 | tag_set(node, settag, offset); |
1112 | node = node->parent; | |
ebf8aa44 | 1113 | } |
144dcfc0 | 1114 | |
e2bdb933 HD |
1115 | /* |
1116 | * Small optimization: now clear that node pointer. | |
1117 | * Since all of this slot's ancestors now have the tag set | |
1118 | * from setting it above, we have no further need to walk | |
1119 | * back up the tree setting tags, until we update slot to | |
1120 | * point to another radix_tree_node. | |
1121 | */ | |
1122 | node = NULL; | |
1123 | ||
ebf8aa44 JK |
1124 | next: |
1125 | /* Go to next item at level determined by 'shift' */ | |
1126 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1127 | /* Overflow can happen when last_index is ~0UL... */ |
1128 | if (index > last_index || !index) | |
ebf8aa44 JK |
1129 | break; |
1130 | if (tagged >= nr_to_tag) | |
1131 | break; | |
1132 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
1133 | /* | |
1134 | * We've fully scanned this node. Go up. Because | |
1135 | * last_index is guaranteed to be in the tree, what | |
1136 | * we do below cannot wander astray. | |
1137 | */ | |
e2bdb933 | 1138 | slot = slot->parent; |
ebf8aa44 JK |
1139 | shift += RADIX_TREE_MAP_SHIFT; |
1140 | } | |
1141 | } | |
1142 | /* | |
ac15ee69 TO |
1143 | * We need not to tag the root tag if there is no tag which is set with |
1144 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1145 | */ |
ac15ee69 TO |
1146 | if (tagged > 0) |
1147 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1148 | *first_indexp = index; |
1149 | ||
1150 | return tagged; | |
1151 | } | |
1152 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1153 | ||
1da177e4 LT |
1154 | /** |
1155 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1156 | * @root: radix tree root | |
1157 | * @results: where the results of the lookup are placed | |
1158 | * @first_index: start the lookup from this key | |
1159 | * @max_items: place up to this many items at *results | |
1160 | * | |
1161 | * Performs an index-ascending scan of the tree for present items. Places | |
1162 | * them at *@results and returns the number of items which were placed at | |
1163 | * *@results. | |
1164 | * | |
1165 | * The implementation is naive. | |
7cf9c2c7 NP |
1166 | * |
1167 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1168 | * rcu_read_lock. In this case, rather than the returned results being | |
1169 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1170 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1171 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1172 | */ |
1173 | unsigned int | |
1174 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1175 | unsigned long first_index, unsigned int max_items) | |
1176 | { | |
cebbd29e KK |
1177 | struct radix_tree_iter iter; |
1178 | void **slot; | |
1179 | unsigned int ret = 0; | |
7cf9c2c7 | 1180 | |
cebbd29e | 1181 | if (unlikely(!max_items)) |
7cf9c2c7 | 1182 | return 0; |
1da177e4 | 1183 | |
cebbd29e | 1184 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1185 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1186 | if (!results[ret]) |
1187 | continue; | |
46437f9a MW |
1188 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1189 | slot = radix_tree_iter_retry(&iter); | |
1190 | continue; | |
1191 | } | |
cebbd29e | 1192 | if (++ret == max_items) |
1da177e4 | 1193 | break; |
1da177e4 | 1194 | } |
7cf9c2c7 | 1195 | |
1da177e4 LT |
1196 | return ret; |
1197 | } | |
1198 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1199 | ||
47feff2c NP |
1200 | /** |
1201 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1202 | * @root: radix tree root | |
1203 | * @results: where the results of the lookup are placed | |
6328650b | 1204 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1205 | * @first_index: start the lookup from this key |
1206 | * @max_items: place up to this many items at *results | |
1207 | * | |
1208 | * Performs an index-ascending scan of the tree for present items. Places | |
1209 | * their slots at *@results and returns the number of items which were | |
1210 | * placed at *@results. | |
1211 | * | |
1212 | * The implementation is naive. | |
1213 | * | |
1214 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1215 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1216 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1217 | */ | |
1218 | unsigned int | |
6328650b HD |
1219 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1220 | void ***results, unsigned long *indices, | |
47feff2c NP |
1221 | unsigned long first_index, unsigned int max_items) |
1222 | { | |
cebbd29e KK |
1223 | struct radix_tree_iter iter; |
1224 | void **slot; | |
1225 | unsigned int ret = 0; | |
47feff2c | 1226 | |
cebbd29e | 1227 | if (unlikely(!max_items)) |
47feff2c NP |
1228 | return 0; |
1229 | ||
cebbd29e KK |
1230 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1231 | results[ret] = slot; | |
6328650b | 1232 | if (indices) |
cebbd29e KK |
1233 | indices[ret] = iter.index; |
1234 | if (++ret == max_items) | |
47feff2c | 1235 | break; |
47feff2c NP |
1236 | } |
1237 | ||
1238 | return ret; | |
1239 | } | |
1240 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1241 | ||
1da177e4 LT |
1242 | /** |
1243 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1244 | * based on a tag | |
1245 | * @root: radix tree root | |
1246 | * @results: where the results of the lookup are placed | |
1247 | * @first_index: start the lookup from this key | |
1248 | * @max_items: place up to this many items at *results | |
daff89f3 | 1249 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1250 | * |
1251 | * Performs an index-ascending scan of the tree for present items which | |
1252 | * have the tag indexed by @tag set. Places the items at *@results and | |
1253 | * returns the number of items which were placed at *@results. | |
1254 | */ | |
1255 | unsigned int | |
1256 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1257 | unsigned long first_index, unsigned int max_items, |
1258 | unsigned int tag) | |
1da177e4 | 1259 | { |
cebbd29e KK |
1260 | struct radix_tree_iter iter; |
1261 | void **slot; | |
1262 | unsigned int ret = 0; | |
612d6c19 | 1263 | |
cebbd29e | 1264 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1265 | return 0; |
1266 | ||
cebbd29e | 1267 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1268 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1269 | if (!results[ret]) |
1270 | continue; | |
46437f9a MW |
1271 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1272 | slot = radix_tree_iter_retry(&iter); | |
1273 | continue; | |
1274 | } | |
cebbd29e | 1275 | if (++ret == max_items) |
1da177e4 | 1276 | break; |
1da177e4 | 1277 | } |
7cf9c2c7 | 1278 | |
1da177e4 LT |
1279 | return ret; |
1280 | } | |
1281 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1282 | ||
47feff2c NP |
1283 | /** |
1284 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1285 | * radix tree based on a tag | |
1286 | * @root: radix tree root | |
1287 | * @results: where the results of the lookup are placed | |
1288 | * @first_index: start the lookup from this key | |
1289 | * @max_items: place up to this many items at *results | |
1290 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1291 | * | |
1292 | * Performs an index-ascending scan of the tree for present items which | |
1293 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1294 | * returns the number of slots which were placed at *@results. | |
1295 | */ | |
1296 | unsigned int | |
1297 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1298 | unsigned long first_index, unsigned int max_items, | |
1299 | unsigned int tag) | |
1300 | { | |
cebbd29e KK |
1301 | struct radix_tree_iter iter; |
1302 | void **slot; | |
1303 | unsigned int ret = 0; | |
47feff2c | 1304 | |
cebbd29e | 1305 | if (unlikely(!max_items)) |
47feff2c NP |
1306 | return 0; |
1307 | ||
cebbd29e KK |
1308 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1309 | results[ret] = slot; | |
1310 | if (++ret == max_items) | |
47feff2c | 1311 | break; |
47feff2c NP |
1312 | } |
1313 | ||
1314 | return ret; | |
1315 | } | |
1316 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1317 | ||
e504f3fd HD |
1318 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1319 | #include <linux/sched.h> /* for cond_resched() */ | |
1320 | ||
1321 | /* | |
1322 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1323 | */ | |
1324 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1325 | unsigned long index, unsigned long *found_index) | |
1326 | { | |
1327 | unsigned int shift, height; | |
1328 | unsigned long i; | |
1329 | ||
449dd698 | 1330 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1331 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1332 | ||
1333 | for ( ; height > 1; height--) { | |
1334 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1335 | for (;;) { | |
1336 | if (slot->slots[i] != NULL) | |
1337 | break; | |
1338 | index &= ~((1UL << shift) - 1); | |
1339 | index += 1UL << shift; | |
1340 | if (index == 0) | |
1341 | goto out; /* 32-bit wraparound */ | |
1342 | i++; | |
1343 | if (i == RADIX_TREE_MAP_SIZE) | |
1344 | goto out; | |
1345 | } | |
1346 | ||
e504f3fd HD |
1347 | slot = rcu_dereference_raw(slot->slots[i]); |
1348 | if (slot == NULL) | |
1349 | goto out; | |
e6145236 MW |
1350 | if (!radix_tree_is_indirect_ptr(slot)) { |
1351 | if (slot == item) { | |
1352 | *found_index = index + i; | |
1353 | index = 0; | |
1354 | } else { | |
1355 | index += shift; | |
1356 | } | |
1357 | goto out; | |
1358 | } | |
339e6353 | 1359 | slot = indirect_to_ptr(slot); |
e6145236 | 1360 | shift -= RADIX_TREE_MAP_SHIFT; |
e504f3fd HD |
1361 | } |
1362 | ||
1363 | /* Bottom level: check items */ | |
1364 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1365 | if (slot->slots[i] == item) { | |
1366 | *found_index = index + i; | |
1367 | index = 0; | |
1368 | goto out; | |
1369 | } | |
1370 | } | |
1371 | index += RADIX_TREE_MAP_SIZE; | |
1372 | out: | |
1373 | return index; | |
1374 | } | |
1375 | ||
1376 | /** | |
1377 | * radix_tree_locate_item - search through radix tree for item | |
1378 | * @root: radix tree root | |
1379 | * @item: item to be found | |
1380 | * | |
1381 | * Returns index where item was found, or -1 if not found. | |
1382 | * Caller must hold no lock (since this time-consuming function needs | |
1383 | * to be preemptible), and must check afterwards if item is still there. | |
1384 | */ | |
1385 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1386 | { | |
1387 | struct radix_tree_node *node; | |
1388 | unsigned long max_index; | |
1389 | unsigned long cur_index = 0; | |
1390 | unsigned long found_index = -1; | |
1391 | ||
1392 | do { | |
1393 | rcu_read_lock(); | |
1394 | node = rcu_dereference_raw(root->rnode); | |
1395 | if (!radix_tree_is_indirect_ptr(node)) { | |
1396 | rcu_read_unlock(); | |
1397 | if (node == item) | |
1398 | found_index = 0; | |
1399 | break; | |
1400 | } | |
1401 | ||
1402 | node = indirect_to_ptr(node); | |
449dd698 JW |
1403 | max_index = radix_tree_maxindex(node->path & |
1404 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1405 | if (cur_index > max_index) { |
1406 | rcu_read_unlock(); | |
e504f3fd | 1407 | break; |
5f30fc94 | 1408 | } |
e504f3fd HD |
1409 | |
1410 | cur_index = __locate(node, item, cur_index, &found_index); | |
1411 | rcu_read_unlock(); | |
1412 | cond_resched(); | |
1413 | } while (cur_index != 0 && cur_index <= max_index); | |
1414 | ||
1415 | return found_index; | |
1416 | } | |
1417 | #else | |
1418 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1419 | { | |
1420 | return -1; | |
1421 | } | |
1422 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1423 | |
a5f51c96 NP |
1424 | /** |
1425 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1426 | * @root radix tree root | |
1427 | */ | |
1428 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1429 | { | |
1430 | /* try to shrink tree height */ | |
c0bc9875 | 1431 | while (root->height > 0) { |
a5f51c96 | 1432 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1433 | struct radix_tree_node *slot; |
a5f51c96 | 1434 | |
c0bc9875 | 1435 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1436 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1437 | |
1438 | /* | |
1439 | * The candidate node has more than one child, or its child | |
e6145236 MW |
1440 | * is not at the leftmost slot, or it is a multiorder entry, |
1441 | * we cannot shrink. | |
c0bc9875 NP |
1442 | */ |
1443 | if (to_free->count != 1) | |
1444 | break; | |
339e6353 MW |
1445 | slot = to_free->slots[0]; |
1446 | if (!slot) | |
c0bc9875 | 1447 | break; |
afe0e395 MW |
1448 | if (!radix_tree_is_indirect_ptr(slot) && (root->height > 1)) |
1449 | break; | |
1450 | ||
1451 | if (radix_tree_is_indirect_ptr(slot)) { | |
1452 | slot = indirect_to_ptr(slot); | |
1453 | slot->parent = NULL; | |
1454 | slot = ptr_to_indirect(slot); | |
1455 | } | |
c0bc9875 | 1456 | |
7cf9c2c7 NP |
1457 | /* |
1458 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1459 | * moving the node from one part of the tree to another: if it |
1460 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1461 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1462 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1463 | */ |
e2bdb933 | 1464 | root->rnode = slot; |
a5f51c96 | 1465 | root->height--; |
27d20fdd NP |
1466 | |
1467 | /* | |
1468 | * We have a dilemma here. The node's slot[0] must not be | |
1469 | * NULLed in case there are concurrent lookups expecting to | |
1470 | * find the item. However if this was a bottom-level node, | |
1471 | * then it may be subject to the slot pointer being visible | |
1472 | * to callers dereferencing it. If item corresponding to | |
1473 | * slot[0] is subsequently deleted, these callers would expect | |
1474 | * their slot to become empty sooner or later. | |
1475 | * | |
1476 | * For example, lockless pagecache will look up a slot, deref | |
1477 | * the page pointer, and if the page is 0 refcount it means it | |
1478 | * was concurrently deleted from pagecache so try the deref | |
1479 | * again. Fortunately there is already a requirement for logic | |
1480 | * to retry the entire slot lookup -- the indirect pointer | |
1481 | * problem (replacing direct root node with an indirect pointer | |
1482 | * also results in a stale slot). So tag the slot as indirect | |
1483 | * to force callers to retry. | |
1484 | */ | |
afe0e395 MW |
1485 | if (!radix_tree_is_indirect_ptr(slot)) |
1486 | to_free->slots[0] = RADIX_TREE_RETRY; | |
27d20fdd | 1487 | |
a5f51c96 NP |
1488 | radix_tree_node_free(to_free); |
1489 | } | |
1490 | } | |
1491 | ||
139e5616 JW |
1492 | /** |
1493 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1494 | * @root: radix tree root | |
139e5616 JW |
1495 | * @node: node containing @index |
1496 | * | |
1497 | * After clearing the slot at @index in @node from radix tree | |
1498 | * rooted at @root, call this function to attempt freeing the | |
1499 | * node and shrinking the tree. | |
1500 | * | |
1501 | * Returns %true if @node was freed, %false otherwise. | |
1502 | */ | |
449dd698 | 1503 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1504 | struct radix_tree_node *node) |
1505 | { | |
1506 | bool deleted = false; | |
1507 | ||
1508 | do { | |
1509 | struct radix_tree_node *parent; | |
1510 | ||
1511 | if (node->count) { | |
1512 | if (node == indirect_to_ptr(root->rnode)) { | |
1513 | radix_tree_shrink(root); | |
1514 | if (root->height == 0) | |
1515 | deleted = true; | |
1516 | } | |
1517 | return deleted; | |
1518 | } | |
1519 | ||
1520 | parent = node->parent; | |
1521 | if (parent) { | |
449dd698 | 1522 | unsigned int offset; |
139e5616 | 1523 | |
449dd698 JW |
1524 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1525 | parent->slots[offset] = NULL; | |
139e5616 JW |
1526 | parent->count--; |
1527 | } else { | |
1528 | root_tag_clear_all(root); | |
1529 | root->height = 0; | |
1530 | root->rnode = NULL; | |
1531 | } | |
1532 | ||
1533 | radix_tree_node_free(node); | |
1534 | deleted = true; | |
1535 | ||
1536 | node = parent; | |
1537 | } while (node); | |
1538 | ||
1539 | return deleted; | |
1540 | } | |
1541 | ||
57578c2e MW |
1542 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1543 | void *ptr, unsigned offset) | |
1544 | { | |
1545 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1546 | int i; | |
1547 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1548 | if (node->slots[offset + i] != ptr) | |
1549 | break; | |
1550 | node->slots[offset + i] = NULL; | |
1551 | node->count--; | |
1552 | } | |
1553 | #endif | |
1554 | } | |
1555 | ||
1da177e4 | 1556 | /** |
53c59f26 | 1557 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1558 | * @root: radix tree root |
1559 | * @index: index key | |
53c59f26 | 1560 | * @item: expected item |
1da177e4 | 1561 | * |
53c59f26 | 1562 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1563 | * |
53c59f26 JW |
1564 | * Returns the address of the deleted item, or NULL if it was not present |
1565 | * or the entry at the given @index was not @item. | |
1da177e4 | 1566 | */ |
53c59f26 JW |
1567 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1568 | unsigned long index, void *item) | |
1da177e4 | 1569 | { |
139e5616 | 1570 | struct radix_tree_node *node; |
57578c2e | 1571 | unsigned int offset; |
139e5616 JW |
1572 | void **slot; |
1573 | void *entry; | |
d5274261 | 1574 | int tag; |
1da177e4 | 1575 | |
139e5616 JW |
1576 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1577 | if (!entry) | |
1578 | return NULL; | |
1da177e4 | 1579 | |
139e5616 JW |
1580 | if (item && entry != item) |
1581 | return NULL; | |
1582 | ||
1583 | if (!node) { | |
612d6c19 NP |
1584 | root_tag_clear_all(root); |
1585 | root->rnode = NULL; | |
139e5616 | 1586 | return entry; |
612d6c19 | 1587 | } |
1da177e4 | 1588 | |
29e0967c | 1589 | offset = get_slot_offset(node, slot); |
53c59f26 | 1590 | |
1da177e4 | 1591 | /* |
e2bdb933 HD |
1592 | * Clear all tags associated with the item to be deleted. |
1593 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1594 | */ |
daff89f3 | 1595 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1596 | if (tag_get(node, tag, offset)) |
612d6c19 | 1597 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1598 | } |
1da177e4 | 1599 | |
57578c2e | 1600 | delete_sibling_entries(node, ptr_to_indirect(slot), offset); |
139e5616 JW |
1601 | node->slots[offset] = NULL; |
1602 | node->count--; | |
e2bdb933 | 1603 | |
449dd698 | 1604 | __radix_tree_delete_node(root, node); |
612d6c19 | 1605 | |
139e5616 | 1606 | return entry; |
1da177e4 | 1607 | } |
53c59f26 JW |
1608 | EXPORT_SYMBOL(radix_tree_delete_item); |
1609 | ||
1610 | /** | |
1611 | * radix_tree_delete - delete an item from a radix tree | |
1612 | * @root: radix tree root | |
1613 | * @index: index key | |
1614 | * | |
1615 | * Remove the item at @index from the radix tree rooted at @root. | |
1616 | * | |
1617 | * Returns the address of the deleted item, or NULL if it was not present. | |
1618 | */ | |
1619 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1620 | { | |
1621 | return radix_tree_delete_item(root, index, NULL); | |
1622 | } | |
1da177e4 LT |
1623 | EXPORT_SYMBOL(radix_tree_delete); |
1624 | ||
1625 | /** | |
1626 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1627 | * @root: radix tree root | |
1628 | * @tag: tag to test | |
1629 | */ | |
daff89f3 | 1630 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1631 | { |
612d6c19 | 1632 | return root_tag_get(root, tag); |
1da177e4 LT |
1633 | } |
1634 | EXPORT_SYMBOL(radix_tree_tagged); | |
1635 | ||
1636 | static void | |
449dd698 | 1637 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1638 | { |
449dd698 JW |
1639 | struct radix_tree_node *node = arg; |
1640 | ||
1641 | memset(node, 0, sizeof(*node)); | |
1642 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1643 | } |
1644 | ||
1645 | static __init unsigned long __maxindex(unsigned int height) | |
1646 | { | |
430d275a PL |
1647 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1648 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1649 | ||
1650 | if (shift < 0) | |
1651 | return ~0UL; | |
1652 | if (shift >= BITS_PER_LONG) | |
1653 | return 0UL; | |
1654 | return ~0UL >> shift; | |
1da177e4 LT |
1655 | } |
1656 | ||
1657 | static __init void radix_tree_init_maxindex(void) | |
1658 | { | |
1659 | unsigned int i; | |
1660 | ||
1661 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1662 | height_to_maxindex[i] = __maxindex(i); | |
1663 | } | |
1664 | ||
1da177e4 LT |
1665 | static int radix_tree_callback(struct notifier_block *nfb, |
1666 | unsigned long action, | |
1667 | void *hcpu) | |
1668 | { | |
1669 | int cpu = (long)hcpu; | |
1670 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1671 | struct radix_tree_node *node; |
1da177e4 LT |
1672 | |
1673 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1674 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1675 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1676 | while (rtp->nr) { | |
9d2a8da0 KS |
1677 | node = rtp->nodes; |
1678 | rtp->nodes = node->private_data; | |
1679 | kmem_cache_free(radix_tree_node_cachep, node); | |
1680 | rtp->nr--; | |
1da177e4 LT |
1681 | } |
1682 | } | |
1683 | return NOTIFY_OK; | |
1684 | } | |
1da177e4 LT |
1685 | |
1686 | void __init radix_tree_init(void) | |
1687 | { | |
1688 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1689 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1690 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1691 | radix_tree_node_ctor); | |
1da177e4 LT |
1692 | radix_tree_init_maxindex(); |
1693 | hotcpu_notifier(radix_tree_callback, 0); | |
1694 | } |