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 |
6b053b8e MW |
7 | * Copyright (C) 2016 Intel, Matthew Wilcox |
8 | * Copyright (C) 2016 Intel, Ross Zwisler | |
1da177e4 LT |
9 | * |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License as | |
12 | * published by the Free Software Foundation; either version 2, or (at | |
13 | * your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, but | |
16 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with this program; if not, write to the Free Software | |
22 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
25 | #include <linux/errno.h> | |
26 | #include <linux/init.h> | |
27 | #include <linux/kernel.h> | |
8bc3bcc9 | 28 | #include <linux/export.h> |
1da177e4 LT |
29 | #include <linux/radix-tree.h> |
30 | #include <linux/percpu.h> | |
31 | #include <linux/slab.h> | |
ce80b067 | 32 | #include <linux/kmemleak.h> |
1da177e4 LT |
33 | #include <linux/notifier.h> |
34 | #include <linux/cpu.h> | |
1da177e4 LT |
35 | #include <linux/string.h> |
36 | #include <linux/bitops.h> | |
7cf9c2c7 | 37 | #include <linux/rcupdate.h> |
92cf2118 | 38 | #include <linux/preempt.h> /* in_interrupt() */ |
1da177e4 LT |
39 | |
40 | ||
1da177e4 LT |
41 | /* |
42 | * Radix tree node cache. | |
43 | */ | |
e18b890b | 44 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 45 | |
55368052 NP |
46 | /* |
47 | * The radix tree is variable-height, so an insert operation not only has | |
48 | * to build the branch to its corresponding item, it also has to build the | |
49 | * branch to existing items if the size has to be increased (by | |
50 | * radix_tree_extend). | |
51 | * | |
52 | * The worst case is a zero height tree with just a single item at index 0, | |
53 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
54 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
55 | * Hence: | |
56 | */ | |
57 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
58 | ||
1da177e4 LT |
59 | /* |
60 | * Per-cpu pool of preloaded nodes | |
61 | */ | |
62 | struct radix_tree_preload { | |
2fcd9005 | 63 | unsigned nr; |
9d2a8da0 KS |
64 | /* nodes->private_data points to next preallocated node */ |
65 | struct radix_tree_node *nodes; | |
1da177e4 | 66 | }; |
8cef7d57 | 67 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 68 | |
a4db4dce | 69 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 70 | { |
30ff46cc | 71 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
72 | } |
73 | ||
a4db4dce | 74 | #define RADIX_TREE_RETRY node_to_entry(NULL) |
afe0e395 | 75 | |
db050f29 MW |
76 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
77 | /* Sibling slots point directly to another slot in the same node */ | |
78 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
79 | { | |
80 | void **ptr = node; | |
81 | return (parent->slots <= ptr) && | |
82 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
83 | } | |
84 | #else | |
85 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
86 | { | |
87 | return false; | |
88 | } | |
89 | #endif | |
90 | ||
91 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
92 | void **slot) | |
93 | { | |
94 | return slot - parent->slots; | |
95 | } | |
96 | ||
97 | static unsigned radix_tree_descend(struct radix_tree_node *parent, | |
98 | struct radix_tree_node **nodep, unsigned offset) | |
99 | { | |
100 | void **entry = rcu_dereference_raw(parent->slots[offset]); | |
101 | ||
102 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 103 | if (radix_tree_is_internal_node(entry)) { |
db050f29 MW |
104 | unsigned long siboff = get_slot_offset(parent, entry); |
105 | if (siboff < RADIX_TREE_MAP_SIZE) { | |
106 | offset = siboff; | |
107 | entry = rcu_dereference_raw(parent->slots[offset]); | |
108 | } | |
109 | } | |
110 | #endif | |
111 | ||
112 | *nodep = (void *)entry; | |
113 | return offset; | |
114 | } | |
115 | ||
612d6c19 NP |
116 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
117 | { | |
118 | return root->gfp_mask & __GFP_BITS_MASK; | |
119 | } | |
120 | ||
643b52b9 NP |
121 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
122 | int offset) | |
123 | { | |
124 | __set_bit(offset, node->tags[tag]); | |
125 | } | |
126 | ||
127 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
128 | int offset) | |
129 | { | |
130 | __clear_bit(offset, node->tags[tag]); | |
131 | } | |
132 | ||
133 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
134 | int offset) | |
135 | { | |
136 | return test_bit(offset, node->tags[tag]); | |
137 | } | |
138 | ||
139 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
140 | { | |
141 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
142 | } | |
143 | ||
2fcd9005 | 144 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
145 | { |
146 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
147 | } | |
148 | ||
149 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
150 | { | |
151 | root->gfp_mask &= __GFP_BITS_MASK; | |
152 | } | |
153 | ||
154 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
155 | { | |
2fcd9005 | 156 | return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
643b52b9 NP |
157 | } |
158 | ||
7b60e9ad MW |
159 | static inline unsigned root_tags_get(struct radix_tree_root *root) |
160 | { | |
161 | return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT; | |
162 | } | |
163 | ||
643b52b9 NP |
164 | /* |
165 | * Returns 1 if any slot in the node has this tag set. | |
166 | * Otherwise returns 0. | |
167 | */ | |
168 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
169 | { | |
2fcd9005 | 170 | unsigned idx; |
643b52b9 NP |
171 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
172 | if (node->tags[tag][idx]) | |
173 | return 1; | |
174 | } | |
175 | return 0; | |
176 | } | |
78c1d784 KK |
177 | |
178 | /** | |
179 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
180 | * | |
181 | * @addr: The address to base the search on | |
182 | * @size: The bitmap size in bits | |
183 | * @offset: The bitnumber to start searching at | |
184 | * | |
185 | * Unrollable variant of find_next_bit() for constant size arrays. | |
186 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
187 | * Returns next bit offset, or size if nothing found. | |
188 | */ | |
189 | static __always_inline unsigned long | |
190 | radix_tree_find_next_bit(const unsigned long *addr, | |
191 | unsigned long size, unsigned long offset) | |
192 | { | |
193 | if (!__builtin_constant_p(size)) | |
194 | return find_next_bit(addr, size, offset); | |
195 | ||
196 | if (offset < size) { | |
197 | unsigned long tmp; | |
198 | ||
199 | addr += offset / BITS_PER_LONG; | |
200 | tmp = *addr >> (offset % BITS_PER_LONG); | |
201 | if (tmp) | |
202 | return __ffs(tmp) + offset; | |
203 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
204 | while (offset < size) { | |
205 | tmp = *++addr; | |
206 | if (tmp) | |
207 | return __ffs(tmp) + offset; | |
208 | offset += BITS_PER_LONG; | |
209 | } | |
210 | } | |
211 | return size; | |
212 | } | |
213 | ||
0796c583 | 214 | #ifndef __KERNEL__ |
d0891265 | 215 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 216 | { |
0796c583 | 217 | unsigned long i; |
7cf19af4 | 218 | |
c12e51b0 | 219 | pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d parent %p\n", |
0c7fa0a8 | 220 | node, node->offset, |
0796c583 | 221 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
c12e51b0 | 222 | node->shift, node->count, node->parent); |
0796c583 RZ |
223 | |
224 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
225 | unsigned long first = index | (i << node->shift); |
226 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
227 | void *entry = node->slots[i]; |
228 | if (!entry) | |
229 | continue; | |
230 | if (is_sibling_entry(node, entry)) { | |
231 | pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n", | |
232 | entry, i, | |
4dd6c098 | 233 | *(void **)entry_to_node(entry), |
0796c583 | 234 | first, last); |
b194d16c | 235 | } else if (!radix_tree_is_internal_node(entry)) { |
0796c583 RZ |
236 | pr_debug("radix entry %p offset %ld indices %ld-%ld\n", |
237 | entry, i, first, last); | |
238 | } else { | |
4dd6c098 | 239 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
240 | } |
241 | } | |
7cf19af4 MW |
242 | } |
243 | ||
244 | /* For debug */ | |
245 | static void radix_tree_dump(struct radix_tree_root *root) | |
246 | { | |
d0891265 MW |
247 | pr_debug("radix root: %p rnode %p tags %x\n", |
248 | root, root->rnode, | |
7cf19af4 | 249 | root->gfp_mask >> __GFP_BITS_SHIFT); |
b194d16c | 250 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 251 | return; |
4dd6c098 | 252 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 MW |
253 | } |
254 | #endif | |
255 | ||
1da177e4 LT |
256 | /* |
257 | * This assumes that the caller has performed appropriate preallocation, and | |
258 | * that the caller has pinned this thread of control to the current CPU. | |
259 | */ | |
260 | static struct radix_tree_node * | |
261 | radix_tree_node_alloc(struct radix_tree_root *root) | |
262 | { | |
e2848a0e | 263 | struct radix_tree_node *ret = NULL; |
612d6c19 | 264 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 265 | |
5e4c0d97 | 266 | /* |
2fcd9005 MW |
267 | * Preload code isn't irq safe and it doesn't make sense to use |
268 | * preloading during an interrupt anyway as all the allocations have | |
269 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 270 | */ |
d0164adc | 271 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
272 | struct radix_tree_preload *rtp; |
273 | ||
58e698af VD |
274 | /* |
275 | * Even if the caller has preloaded, try to allocate from the | |
276 | * cache first for the new node to get accounted. | |
277 | */ | |
278 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
279 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
280 | if (ret) | |
281 | goto out; | |
282 | ||
e2848a0e NP |
283 | /* |
284 | * Provided the caller has preloaded here, we will always | |
285 | * succeed in getting a node here (and never reach | |
286 | * kmem_cache_alloc) | |
287 | */ | |
7c8e0181 | 288 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 289 | if (rtp->nr) { |
9d2a8da0 KS |
290 | ret = rtp->nodes; |
291 | rtp->nodes = ret->private_data; | |
292 | ret->private_data = NULL; | |
1da177e4 LT |
293 | rtp->nr--; |
294 | } | |
ce80b067 CM |
295 | /* |
296 | * Update the allocation stack trace as this is more useful | |
297 | * for debugging. | |
298 | */ | |
299 | kmemleak_update_trace(ret); | |
58e698af | 300 | goto out; |
1da177e4 | 301 | } |
58e698af VD |
302 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
303 | gfp_mask | __GFP_ACCOUNT); | |
304 | out: | |
b194d16c | 305 | BUG_ON(radix_tree_is_internal_node(ret)); |
1da177e4 LT |
306 | return ret; |
307 | } | |
308 | ||
7cf9c2c7 NP |
309 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
310 | { | |
311 | struct radix_tree_node *node = | |
312 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 313 | int i; |
643b52b9 NP |
314 | |
315 | /* | |
316 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
317 | * can leave us with a non-NULL entry in the first slot, so clear | |
318 | * that here to make sure. | |
319 | */ | |
b6dd0865 DC |
320 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
321 | tag_clear(node, i, 0); | |
322 | ||
643b52b9 NP |
323 | node->slots[0] = NULL; |
324 | node->count = 0; | |
325 | ||
7cf9c2c7 NP |
326 | kmem_cache_free(radix_tree_node_cachep, node); |
327 | } | |
328 | ||
1da177e4 LT |
329 | static inline void |
330 | radix_tree_node_free(struct radix_tree_node *node) | |
331 | { | |
7cf9c2c7 | 332 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
333 | } |
334 | ||
335 | /* | |
336 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
337 | * ensure that the addition of a single element in the tree cannot fail. On | |
338 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
339 | * with preemption not disabled. | |
b34df792 DH |
340 | * |
341 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 342 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 343 | */ |
5e4c0d97 | 344 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
345 | { |
346 | struct radix_tree_preload *rtp; | |
347 | struct radix_tree_node *node; | |
348 | int ret = -ENOMEM; | |
349 | ||
350 | preempt_disable(); | |
7c8e0181 | 351 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 352 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 353 | preempt_enable(); |
488514d1 | 354 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
355 | if (node == NULL) |
356 | goto out; | |
357 | preempt_disable(); | |
7c8e0181 | 358 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
359 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
360 | node->private_data = rtp->nodes; | |
361 | rtp->nodes = node; | |
362 | rtp->nr++; | |
363 | } else { | |
1da177e4 | 364 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 365 | } |
1da177e4 LT |
366 | } |
367 | ret = 0; | |
368 | out: | |
369 | return ret; | |
370 | } | |
5e4c0d97 JK |
371 | |
372 | /* | |
373 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
374 | * ensure that the addition of a single element in the tree cannot fail. On | |
375 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
376 | * with preemption not disabled. | |
377 | * | |
378 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 379 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
380 | */ |
381 | int radix_tree_preload(gfp_t gfp_mask) | |
382 | { | |
383 | /* Warn on non-sensical use... */ | |
d0164adc | 384 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
385 | return __radix_tree_preload(gfp_mask); |
386 | } | |
d7f0923d | 387 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 388 | |
5e4c0d97 JK |
389 | /* |
390 | * The same as above function, except we don't guarantee preloading happens. | |
391 | * We do it, if we decide it helps. On success, return zero with preemption | |
392 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
393 | */ | |
394 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
395 | { | |
d0164adc | 396 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
397 | return __radix_tree_preload(gfp_mask); |
398 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
399 | preempt_disable(); | |
400 | return 0; | |
401 | } | |
402 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
403 | ||
1da177e4 | 404 | /* |
d0891265 | 405 | * The maximum index which can be stored in a radix tree |
1da177e4 | 406 | */ |
c12e51b0 MW |
407 | static inline unsigned long shift_maxindex(unsigned int shift) |
408 | { | |
409 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
410 | } | |
411 | ||
1456a439 MW |
412 | static inline unsigned long node_maxindex(struct radix_tree_node *node) |
413 | { | |
c12e51b0 | 414 | return shift_maxindex(node->shift); |
1456a439 MW |
415 | } |
416 | ||
417 | static unsigned radix_tree_load_root(struct radix_tree_root *root, | |
418 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
419 | { | |
420 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
421 | ||
422 | *nodep = node; | |
423 | ||
b194d16c | 424 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 425 | node = entry_to_node(node); |
1456a439 | 426 | *maxindex = node_maxindex(node); |
c12e51b0 | 427 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
428 | } |
429 | ||
430 | *maxindex = 0; | |
431 | return 0; | |
432 | } | |
433 | ||
1da177e4 LT |
434 | /* |
435 | * Extend a radix tree so it can store key @index. | |
436 | */ | |
e6145236 | 437 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 438 | unsigned long index, unsigned int shift) |
1da177e4 | 439 | { |
e2bdb933 | 440 | struct radix_tree_node *slot; |
d0891265 | 441 | unsigned int maxshift; |
1da177e4 LT |
442 | int tag; |
443 | ||
d0891265 MW |
444 | /* Figure out what the shift should be. */ |
445 | maxshift = shift; | |
446 | while (index > shift_maxindex(maxshift)) | |
447 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 448 | |
d0891265 MW |
449 | slot = root->rnode; |
450 | if (!slot) | |
1da177e4 | 451 | goto out; |
1da177e4 | 452 | |
1da177e4 | 453 | do { |
2fcd9005 MW |
454 | struct radix_tree_node *node = radix_tree_node_alloc(root); |
455 | ||
456 | if (!node) | |
1da177e4 LT |
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 | ||
d0891265 MW |
465 | BUG_ON(shift > BITS_PER_LONG); |
466 | node->shift = shift; | |
0c7fa0a8 | 467 | node->offset = 0; |
1da177e4 | 468 | node->count = 1; |
e2bdb933 | 469 | node->parent = NULL; |
b194d16c | 470 | if (radix_tree_is_internal_node(slot)) |
4dd6c098 | 471 | entry_to_node(slot)->parent = node; |
e2bdb933 | 472 | node->slots[0] = slot; |
a4db4dce MW |
473 | slot = node_to_entry(node); |
474 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 475 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 476 | } while (shift <= maxshift); |
1da177e4 | 477 | out: |
d0891265 | 478 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
479 | } |
480 | ||
481 | /** | |
139e5616 | 482 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
483 | * @root: radix tree root |
484 | * @index: index key | |
e6145236 | 485 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
486 | * @nodep: returns node |
487 | * @slotp: returns slot | |
1da177e4 | 488 | * |
139e5616 JW |
489 | * Create, if necessary, and return the node and slot for an item |
490 | * at position @index in the radix tree @root. | |
491 | * | |
492 | * Until there is more than one item in the tree, no nodes are | |
493 | * allocated and @root->rnode is used as a direct slot instead of | |
494 | * pointing to a node, in which case *@nodep will be NULL. | |
495 | * | |
496 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 497 | */ |
139e5616 | 498 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
499 | unsigned order, struct radix_tree_node **nodep, |
500 | void ***slotp) | |
1da177e4 | 501 | { |
89148aa4 MW |
502 | struct radix_tree_node *node = NULL, *child; |
503 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 504 | unsigned long maxindex; |
89148aa4 | 505 | unsigned int shift, offset = 0; |
49ea6ebc MW |
506 | unsigned long max = index | ((1UL << order) - 1); |
507 | ||
89148aa4 | 508 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
509 | |
510 | /* Make sure the tree is high enough. */ | |
49ea6ebc | 511 | if (max > maxindex) { |
d0891265 | 512 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 513 | if (error < 0) |
1da177e4 | 514 | return error; |
49ea6ebc | 515 | shift = error; |
89148aa4 | 516 | child = root->rnode; |
d0891265 | 517 | if (order == shift) |
49ea6ebc | 518 | shift += RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
519 | } |
520 | ||
e6145236 | 521 | while (shift > order) { |
c12e51b0 | 522 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 523 | if (child == NULL) { |
1da177e4 | 524 | /* Have to add a child node. */ |
89148aa4 MW |
525 | child = radix_tree_node_alloc(root); |
526 | if (!child) | |
1da177e4 | 527 | return -ENOMEM; |
89148aa4 MW |
528 | child->shift = shift; |
529 | child->offset = offset; | |
530 | child->parent = node; | |
531 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
532 | if (node) | |
1da177e4 | 533 | node->count++; |
89148aa4 | 534 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 535 | break; |
1da177e4 LT |
536 | |
537 | /* Go a level down */ | |
89148aa4 | 538 | node = entry_to_node(child); |
8a14f4d8 | 539 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
89148aa4 MW |
540 | offset = radix_tree_descend(node, &child, offset); |
541 | slot = &node->slots[offset]; | |
e6145236 MW |
542 | } |
543 | ||
57578c2e | 544 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 545 | /* Insert pointers to the canonical entry */ |
3b8c00f6 | 546 | if (order > shift) { |
89148aa4 | 547 | unsigned i, n = 1 << (order - shift); |
e6145236 | 548 | offset = offset & ~(n - 1); |
89148aa4 MW |
549 | slot = &node->slots[offset]; |
550 | child = node_to_entry(slot); | |
e6145236 | 551 | for (i = 0; i < n; i++) { |
89148aa4 | 552 | if (slot[i]) |
e6145236 MW |
553 | return -EEXIST; |
554 | } | |
555 | ||
556 | for (i = 1; i < n; i++) { | |
89148aa4 | 557 | rcu_assign_pointer(slot[i], child); |
e6145236 MW |
558 | node->count++; |
559 | } | |
612d6c19 | 560 | } |
57578c2e | 561 | #endif |
1da177e4 | 562 | |
139e5616 JW |
563 | if (nodep) |
564 | *nodep = node; | |
565 | if (slotp) | |
89148aa4 | 566 | *slotp = slot; |
139e5616 JW |
567 | return 0; |
568 | } | |
569 | ||
570 | /** | |
e6145236 | 571 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
572 | * @root: radix tree root |
573 | * @index: index key | |
e6145236 | 574 | * @order: key covers the 2^order indices around index |
139e5616 JW |
575 | * @item: item to insert |
576 | * | |
577 | * Insert an item into the radix tree at position @index. | |
578 | */ | |
e6145236 MW |
579 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
580 | unsigned order, void *item) | |
139e5616 JW |
581 | { |
582 | struct radix_tree_node *node; | |
583 | void **slot; | |
584 | int error; | |
585 | ||
b194d16c | 586 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 587 | |
e6145236 | 588 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
589 | if (error) |
590 | return error; | |
591 | if (*slot != NULL) | |
1da177e4 | 592 | return -EEXIST; |
139e5616 | 593 | rcu_assign_pointer(*slot, item); |
201b6264 | 594 | |
612d6c19 | 595 | if (node) { |
7b60e9ad | 596 | unsigned offset = get_slot_offset(node, slot); |
612d6c19 | 597 | node->count++; |
7b60e9ad MW |
598 | BUG_ON(tag_get(node, 0, offset)); |
599 | BUG_ON(tag_get(node, 1, offset)); | |
600 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 601 | } else { |
7b60e9ad | 602 | BUG_ON(root_tags_get(root)); |
612d6c19 | 603 | } |
1da177e4 | 604 | |
1da177e4 LT |
605 | return 0; |
606 | } | |
e6145236 | 607 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 608 | |
139e5616 JW |
609 | /** |
610 | * __radix_tree_lookup - lookup an item in a radix tree | |
611 | * @root: radix tree root | |
612 | * @index: index key | |
613 | * @nodep: returns node | |
614 | * @slotp: returns slot | |
615 | * | |
616 | * Lookup and return the item at position @index in the radix | |
617 | * tree @root. | |
618 | * | |
619 | * Until there is more than one item in the tree, no nodes are | |
620 | * allocated and @root->rnode is used as a direct slot instead of | |
621 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 622 | */ |
139e5616 JW |
623 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
624 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 625 | { |
139e5616 | 626 | struct radix_tree_node *node, *parent; |
85829954 MW |
627 | unsigned long maxindex; |
628 | unsigned int shift; | |
139e5616 | 629 | void **slot; |
612d6c19 | 630 | |
85829954 MW |
631 | restart: |
632 | parent = NULL; | |
633 | slot = (void **)&root->rnode; | |
634 | shift = radix_tree_load_root(root, &node, &maxindex); | |
635 | if (index > maxindex) | |
1da177e4 LT |
636 | return NULL; |
637 | ||
b194d16c | 638 | while (radix_tree_is_internal_node(node)) { |
85829954 | 639 | unsigned offset; |
1da177e4 | 640 | |
85829954 MW |
641 | if (node == RADIX_TREE_RETRY) |
642 | goto restart; | |
4dd6c098 | 643 | parent = entry_to_node(node); |
1da177e4 | 644 | shift -= RADIX_TREE_MAP_SHIFT; |
85829954 MW |
645 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
646 | offset = radix_tree_descend(parent, &node, offset); | |
647 | slot = parent->slots + offset; | |
648 | } | |
1da177e4 | 649 | |
139e5616 JW |
650 | if (nodep) |
651 | *nodep = parent; | |
652 | if (slotp) | |
653 | *slotp = slot; | |
654 | return node; | |
b72b71c6 HS |
655 | } |
656 | ||
657 | /** | |
658 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
659 | * @root: radix tree root | |
660 | * @index: index key | |
661 | * | |
662 | * Returns: the slot corresponding to the position @index in the | |
663 | * radix tree @root. This is useful for update-if-exists operations. | |
664 | * | |
665 | * This function can be called under rcu_read_lock iff the slot is not | |
666 | * modified by radix_tree_replace_slot, otherwise it must be called | |
667 | * exclusive from other writers. Any dereference of the slot must be done | |
668 | * using radix_tree_deref_slot. | |
669 | */ | |
670 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
671 | { | |
139e5616 JW |
672 | void **slot; |
673 | ||
674 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
675 | return NULL; | |
676 | return slot; | |
a4331366 | 677 | } |
a4331366 HR |
678 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
679 | ||
680 | /** | |
681 | * radix_tree_lookup - perform lookup operation on a radix tree | |
682 | * @root: radix tree root | |
683 | * @index: index key | |
684 | * | |
685 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
686 | * |
687 | * This function can be called under rcu_read_lock, however the caller | |
688 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
689 | * them safely). No RCU barriers are required to access or modify the | |
690 | * returned item, however. | |
a4331366 HR |
691 | */ |
692 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
693 | { | |
139e5616 | 694 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
695 | } |
696 | EXPORT_SYMBOL(radix_tree_lookup); | |
697 | ||
698 | /** | |
699 | * radix_tree_tag_set - set a tag on a radix tree node | |
700 | * @root: radix tree root | |
701 | * @index: index key | |
2fcd9005 | 702 | * @tag: tag index |
1da177e4 | 703 | * |
daff89f3 JC |
704 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
705 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
706 | * the root all the way down to the leaf node. |
707 | * | |
2fcd9005 | 708 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
709 | * item is a bug. |
710 | */ | |
711 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 712 | unsigned long index, unsigned int tag) |
1da177e4 | 713 | { |
fb969909 RZ |
714 | struct radix_tree_node *node, *parent; |
715 | unsigned long maxindex; | |
716 | unsigned int shift; | |
1da177e4 | 717 | |
fb969909 RZ |
718 | shift = radix_tree_load_root(root, &node, &maxindex); |
719 | BUG_ON(index > maxindex); | |
1da177e4 | 720 | |
b194d16c | 721 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 722 | unsigned offset; |
1da177e4 | 723 | |
1da177e4 | 724 | shift -= RADIX_TREE_MAP_SHIFT; |
fb969909 RZ |
725 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
726 | ||
4dd6c098 | 727 | parent = entry_to_node(node); |
fb969909 RZ |
728 | offset = radix_tree_descend(parent, &node, offset); |
729 | BUG_ON(!node); | |
730 | ||
731 | if (!tag_get(parent, tag, offset)) | |
732 | tag_set(parent, tag, offset); | |
1da177e4 LT |
733 | } |
734 | ||
612d6c19 | 735 | /* set the root's tag bit */ |
fb969909 | 736 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
737 | root_tag_set(root, tag); |
738 | ||
fb969909 | 739 | return node; |
1da177e4 LT |
740 | } |
741 | EXPORT_SYMBOL(radix_tree_tag_set); | |
742 | ||
d604c324 MW |
743 | static void node_tag_clear(struct radix_tree_root *root, |
744 | struct radix_tree_node *node, | |
745 | unsigned int tag, unsigned int offset) | |
746 | { | |
747 | while (node) { | |
748 | if (!tag_get(node, tag, offset)) | |
749 | return; | |
750 | tag_clear(node, tag, offset); | |
751 | if (any_tag_set(node, tag)) | |
752 | return; | |
753 | ||
754 | offset = node->offset; | |
755 | node = node->parent; | |
756 | } | |
757 | ||
758 | /* clear the root's tag bit */ | |
759 | if (root_tag_get(root, tag)) | |
760 | root_tag_clear(root, tag); | |
761 | } | |
762 | ||
1da177e4 LT |
763 | /** |
764 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
765 | * @root: radix tree root | |
766 | * @index: index key | |
2fcd9005 | 767 | * @tag: tag index |
1da177e4 | 768 | * |
daff89f3 | 769 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
770 | * corresponding to @index in the radix tree. If this causes |
771 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
772 | * next-to-leaf node, etc. |
773 | * | |
774 | * Returns the address of the tagged item on success, else NULL. ie: | |
775 | * has the same return value and semantics as radix_tree_lookup(). | |
776 | */ | |
777 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 778 | unsigned long index, unsigned int tag) |
1da177e4 | 779 | { |
00f47b58 RZ |
780 | struct radix_tree_node *node, *parent; |
781 | unsigned long maxindex; | |
782 | unsigned int shift; | |
e2bdb933 | 783 | int uninitialized_var(offset); |
1da177e4 | 784 | |
00f47b58 RZ |
785 | shift = radix_tree_load_root(root, &node, &maxindex); |
786 | if (index > maxindex) | |
787 | return NULL; | |
1da177e4 | 788 | |
00f47b58 | 789 | parent = NULL; |
1da177e4 | 790 | |
b194d16c | 791 | while (radix_tree_is_internal_node(node)) { |
e2bdb933 | 792 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 793 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
00f47b58 | 794 | |
4dd6c098 | 795 | parent = entry_to_node(node); |
00f47b58 | 796 | offset = radix_tree_descend(parent, &node, offset); |
1da177e4 LT |
797 | } |
798 | ||
d604c324 MW |
799 | if (node) |
800 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 801 | |
00f47b58 | 802 | return node; |
1da177e4 LT |
803 | } |
804 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
805 | ||
1da177e4 | 806 | /** |
32605a18 MT |
807 | * radix_tree_tag_get - get a tag on a radix tree node |
808 | * @root: radix tree root | |
809 | * @index: index key | |
2fcd9005 | 810 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 811 | * |
32605a18 | 812 | * Return values: |
1da177e4 | 813 | * |
612d6c19 NP |
814 | * 0: tag not present or not set |
815 | * 1: tag set | |
ce82653d DH |
816 | * |
817 | * Note that the return value of this function may not be relied on, even if | |
818 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
819 | * from concurrency. | |
1da177e4 LT |
820 | */ |
821 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 822 | unsigned long index, unsigned int tag) |
1da177e4 | 823 | { |
4589ba6d RZ |
824 | struct radix_tree_node *node, *parent; |
825 | unsigned long maxindex; | |
826 | unsigned int shift; | |
1da177e4 | 827 | |
612d6c19 NP |
828 | if (!root_tag_get(root, tag)) |
829 | return 0; | |
830 | ||
4589ba6d RZ |
831 | shift = radix_tree_load_root(root, &node, &maxindex); |
832 | if (index > maxindex) | |
833 | return 0; | |
7cf9c2c7 NP |
834 | if (node == NULL) |
835 | return 0; | |
836 | ||
b194d16c | 837 | while (radix_tree_is_internal_node(node)) { |
4589ba6d | 838 | int offset; |
612d6c19 | 839 | |
4589ba6d RZ |
840 | shift -= RADIX_TREE_MAP_SHIFT; |
841 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1da177e4 | 842 | |
4dd6c098 | 843 | parent = entry_to_node(node); |
4589ba6d | 844 | offset = radix_tree_descend(parent, &node, offset); |
1da177e4 | 845 | |
4589ba6d | 846 | if (!node) |
1da177e4 | 847 | return 0; |
4589ba6d | 848 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 849 | return 0; |
4589ba6d RZ |
850 | if (node == RADIX_TREE_RETRY) |
851 | break; | |
1da177e4 | 852 | } |
4589ba6d RZ |
853 | |
854 | return 1; | |
1da177e4 LT |
855 | } |
856 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 857 | |
21ef5339 RZ |
858 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
859 | unsigned int shift) | |
860 | { | |
861 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
862 | iter->shift = shift; | |
863 | #endif | |
864 | } | |
865 | ||
78c1d784 KK |
866 | /** |
867 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
868 | * | |
869 | * @root: radix tree root | |
870 | * @iter: iterator state | |
871 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
872 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
873 | */ | |
874 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
875 | struct radix_tree_iter *iter, unsigned flags) | |
876 | { | |
877 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
8c1244de | 878 | struct radix_tree_node *node, *child; |
21ef5339 | 879 | unsigned long index, offset, maxindex; |
78c1d784 KK |
880 | |
881 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
882 | return NULL; | |
883 | ||
884 | /* | |
885 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
886 | * during iterating; it can be zero only at the beginning. | |
887 | * And we cannot overflow iter->next_index in a single step, | |
888 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
889 | * |
890 | * This condition also used by radix_tree_next_slot() to stop | |
891 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
892 | */ |
893 | index = iter->next_index; | |
894 | if (!index && iter->index) | |
895 | return NULL; | |
896 | ||
21ef5339 | 897 | restart: |
8c1244de | 898 | shift = radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
899 | if (index > maxindex) |
900 | return NULL; | |
8c1244de MW |
901 | if (!child) |
902 | return NULL; | |
21ef5339 | 903 | |
8c1244de | 904 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 905 | /* Single-slot tree */ |
21ef5339 RZ |
906 | iter->index = index; |
907 | iter->next_index = maxindex + 1; | |
78c1d784 | 908 | iter->tags = 1; |
8c1244de | 909 | __set_iter_shift(iter, 0); |
78c1d784 | 910 | return (void **)&root->rnode; |
8c1244de | 911 | } |
21ef5339 | 912 | |
8c1244de MW |
913 | do { |
914 | node = entry_to_node(child); | |
915 | shift -= RADIX_TREE_MAP_SHIFT; | |
916 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
917 | offset = radix_tree_descend(node, &child, offset); | |
21ef5339 | 918 | |
78c1d784 | 919 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 920 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
921 | /* Hole detected */ |
922 | if (flags & RADIX_TREE_ITER_CONTIG) | |
923 | return NULL; | |
924 | ||
925 | if (flags & RADIX_TREE_ITER_TAGGED) | |
926 | offset = radix_tree_find_next_bit( | |
927 | node->tags[tag], | |
928 | RADIX_TREE_MAP_SIZE, | |
929 | offset + 1); | |
930 | else | |
931 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
932 | void *slot = node->slots[offset]; |
933 | if (is_sibling_entry(node, slot)) | |
934 | continue; | |
935 | if (slot) | |
78c1d784 KK |
936 | break; |
937 | } | |
8c1244de | 938 | index &= ~node_maxindex(node); |
78c1d784 KK |
939 | index += offset << shift; |
940 | /* Overflow after ~0UL */ | |
941 | if (!index) | |
942 | return NULL; | |
943 | if (offset == RADIX_TREE_MAP_SIZE) | |
944 | goto restart; | |
8c1244de | 945 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
946 | } |
947 | ||
8c1244de | 948 | if ((child == NULL) || (child == RADIX_TREE_RETRY)) |
78c1d784 | 949 | goto restart; |
8c1244de | 950 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
951 | |
952 | /* Update the iterator state */ | |
8c1244de MW |
953 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
954 | iter->next_index = (index | node_maxindex(node)) + 1; | |
21ef5339 | 955 | __set_iter_shift(iter, shift); |
78c1d784 KK |
956 | |
957 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
958 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
959 | unsigned tag_long, tag_bit; | |
960 | ||
961 | tag_long = offset / BITS_PER_LONG; | |
962 | tag_bit = offset % BITS_PER_LONG; | |
963 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
964 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
965 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
966 | /* Pick tags from next element */ | |
967 | if (tag_bit) | |
968 | iter->tags |= node->tags[tag][tag_long + 1] << | |
969 | (BITS_PER_LONG - tag_bit); | |
970 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
971 | iter->next_index = index + BITS_PER_LONG; | |
972 | } | |
973 | } | |
974 | ||
975 | return node->slots + offset; | |
976 | } | |
977 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
978 | ||
ebf8aa44 JK |
979 | /** |
980 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
981 | * tag if item has another tag set | |
982 | * @root: radix tree root | |
983 | * @first_indexp: pointer to a starting index of a range to scan | |
984 | * @last_index: last index of a range to scan | |
985 | * @nr_to_tag: maximum number items to tag | |
986 | * @iftag: tag index to test | |
987 | * @settag: tag index to set if tested tag is set | |
988 | * | |
989 | * This function scans range of radix tree from first_index to last_index | |
990 | * (inclusive). For each item in the range if iftag is set, the function sets | |
991 | * also settag. The function stops either after tagging nr_to_tag items or | |
992 | * after reaching last_index. | |
993 | * | |
144dcfc0 DC |
994 | * The tags must be set from the leaf level only and propagated back up the |
995 | * path to the root. We must do this so that we resolve the full path before | |
996 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
997 | * we can get to the leaf node and find that the index that has the iftag | |
998 | * set is outside the range we are scanning. This reults in dangling tags and | |
999 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1000 | * | |
2fcd9005 | 1001 | * The function returns the number of leaves where the tag was set and sets |
ebf8aa44 | 1002 | * *first_indexp to the first unscanned index. |
d5ed3a4a JK |
1003 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1004 | * be prepared to handle that. | |
ebf8aa44 JK |
1005 | */ |
1006 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1007 | unsigned long *first_indexp, unsigned long last_index, | |
1008 | unsigned long nr_to_tag, | |
1009 | unsigned int iftag, unsigned int settag) | |
1010 | { | |
a8e4da25 | 1011 | struct radix_tree_node *parent, *node, *child; |
070c5ac2 | 1012 | unsigned long maxindex; |
a8e4da25 | 1013 | unsigned int shift = radix_tree_load_root(root, &child, &maxindex); |
144dcfc0 DC |
1014 | unsigned long tagged = 0; |
1015 | unsigned long index = *first_indexp; | |
ebf8aa44 | 1016 | |
070c5ac2 | 1017 | last_index = min(last_index, maxindex); |
ebf8aa44 JK |
1018 | if (index > last_index) |
1019 | return 0; | |
1020 | if (!nr_to_tag) | |
1021 | return 0; | |
1022 | if (!root_tag_get(root, iftag)) { | |
1023 | *first_indexp = last_index + 1; | |
1024 | return 0; | |
1025 | } | |
a8e4da25 | 1026 | if (!radix_tree_is_internal_node(child)) { |
ebf8aa44 JK |
1027 | *first_indexp = last_index + 1; |
1028 | root_tag_set(root, settag); | |
1029 | return 1; | |
1030 | } | |
1031 | ||
a8e4da25 | 1032 | node = entry_to_node(child); |
070c5ac2 | 1033 | shift -= RADIX_TREE_MAP_SHIFT; |
ebf8aa44 JK |
1034 | |
1035 | for (;;) { | |
a8e4da25 MW |
1036 | unsigned offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
1037 | offset = radix_tree_descend(node, &child, offset); | |
1038 | if (!child) | |
ebf8aa44 | 1039 | goto next; |
070c5ac2 | 1040 | if (!tag_get(node, iftag, offset)) |
ebf8aa44 | 1041 | goto next; |
070c5ac2 | 1042 | /* Sibling slots never have tags set on them */ |
a8e4da25 MW |
1043 | if (radix_tree_is_internal_node(child)) { |
1044 | node = entry_to_node(child); | |
070c5ac2 MW |
1045 | shift -= RADIX_TREE_MAP_SHIFT; |
1046 | continue; | |
144dcfc0 DC |
1047 | } |
1048 | ||
1049 | /* tag the leaf */ | |
070c5ac2 MW |
1050 | tagged++; |
1051 | tag_set(node, settag, offset); | |
144dcfc0 DC |
1052 | |
1053 | /* walk back up the path tagging interior nodes */ | |
a8e4da25 MW |
1054 | parent = node; |
1055 | for (;;) { | |
1056 | offset = parent->offset; | |
1057 | parent = parent->parent; | |
1058 | if (!parent) | |
1059 | break; | |
144dcfc0 | 1060 | /* stop if we find a node with the tag already set */ |
a8e4da25 | 1061 | if (tag_get(parent, settag, offset)) |
144dcfc0 | 1062 | break; |
a8e4da25 | 1063 | tag_set(parent, settag, offset); |
ebf8aa44 | 1064 | } |
070c5ac2 | 1065 | next: |
ebf8aa44 JK |
1066 | /* Go to next item at level determined by 'shift' */ |
1067 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1068 | /* Overflow can happen when last_index is ~0UL... */ |
1069 | if (index > last_index || !index) | |
ebf8aa44 | 1070 | break; |
070c5ac2 MW |
1071 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
1072 | while (offset == 0) { | |
ebf8aa44 JK |
1073 | /* |
1074 | * We've fully scanned this node. Go up. Because | |
1075 | * last_index is guaranteed to be in the tree, what | |
1076 | * we do below cannot wander astray. | |
1077 | */ | |
070c5ac2 | 1078 | node = node->parent; |
ebf8aa44 | 1079 | shift += RADIX_TREE_MAP_SHIFT; |
070c5ac2 | 1080 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
ebf8aa44 | 1081 | } |
070c5ac2 MW |
1082 | if (is_sibling_entry(node, node->slots[offset])) |
1083 | goto next; | |
1084 | if (tagged >= nr_to_tag) | |
1085 | break; | |
ebf8aa44 JK |
1086 | } |
1087 | /* | |
ac15ee69 TO |
1088 | * We need not to tag the root tag if there is no tag which is set with |
1089 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1090 | */ |
ac15ee69 TO |
1091 | if (tagged > 0) |
1092 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1093 | *first_indexp = index; |
1094 | ||
1095 | return tagged; | |
1096 | } | |
1097 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1098 | ||
1da177e4 LT |
1099 | /** |
1100 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1101 | * @root: radix tree root | |
1102 | * @results: where the results of the lookup are placed | |
1103 | * @first_index: start the lookup from this key | |
1104 | * @max_items: place up to this many items at *results | |
1105 | * | |
1106 | * Performs an index-ascending scan of the tree for present items. Places | |
1107 | * them at *@results and returns the number of items which were placed at | |
1108 | * *@results. | |
1109 | * | |
1110 | * The implementation is naive. | |
7cf9c2c7 NP |
1111 | * |
1112 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1113 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1114 | * an atomic snapshot of the tree at a single point in time, the |
1115 | * semantics of an RCU protected gang lookup are as though multiple | |
1116 | * radix_tree_lookups have been issued in individual locks, and results | |
1117 | * stored in 'results'. | |
1da177e4 LT |
1118 | */ |
1119 | unsigned int | |
1120 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1121 | unsigned long first_index, unsigned int max_items) | |
1122 | { | |
cebbd29e KK |
1123 | struct radix_tree_iter iter; |
1124 | void **slot; | |
1125 | unsigned int ret = 0; | |
7cf9c2c7 | 1126 | |
cebbd29e | 1127 | if (unlikely(!max_items)) |
7cf9c2c7 | 1128 | return 0; |
1da177e4 | 1129 | |
cebbd29e | 1130 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1131 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1132 | if (!results[ret]) |
1133 | continue; | |
b194d16c | 1134 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1135 | slot = radix_tree_iter_retry(&iter); |
1136 | continue; | |
1137 | } | |
cebbd29e | 1138 | if (++ret == max_items) |
1da177e4 | 1139 | break; |
1da177e4 | 1140 | } |
7cf9c2c7 | 1141 | |
1da177e4 LT |
1142 | return ret; |
1143 | } | |
1144 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1145 | ||
47feff2c NP |
1146 | /** |
1147 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1148 | * @root: radix tree root | |
1149 | * @results: where the results of the lookup are placed | |
6328650b | 1150 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1151 | * @first_index: start the lookup from this key |
1152 | * @max_items: place up to this many items at *results | |
1153 | * | |
1154 | * Performs an index-ascending scan of the tree for present items. Places | |
1155 | * their slots at *@results and returns the number of items which were | |
1156 | * placed at *@results. | |
1157 | * | |
1158 | * The implementation is naive. | |
1159 | * | |
1160 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1161 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1162 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1163 | */ | |
1164 | unsigned int | |
6328650b HD |
1165 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1166 | void ***results, unsigned long *indices, | |
47feff2c NP |
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; | |
47feff2c | 1172 | |
cebbd29e | 1173 | if (unlikely(!max_items)) |
47feff2c NP |
1174 | return 0; |
1175 | ||
cebbd29e KK |
1176 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1177 | results[ret] = slot; | |
6328650b | 1178 | if (indices) |
cebbd29e KK |
1179 | indices[ret] = iter.index; |
1180 | if (++ret == max_items) | |
47feff2c | 1181 | break; |
47feff2c NP |
1182 | } |
1183 | ||
1184 | return ret; | |
1185 | } | |
1186 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1187 | ||
1da177e4 LT |
1188 | /** |
1189 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1190 | * based on a tag | |
1191 | * @root: radix tree root | |
1192 | * @results: where the results of the lookup are placed | |
1193 | * @first_index: start the lookup from this key | |
1194 | * @max_items: place up to this many items at *results | |
daff89f3 | 1195 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1196 | * |
1197 | * Performs an index-ascending scan of the tree for present items which | |
1198 | * have the tag indexed by @tag set. Places the items at *@results and | |
1199 | * returns the number of items which were placed at *@results. | |
1200 | */ | |
1201 | unsigned int | |
1202 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1203 | unsigned long first_index, unsigned int max_items, |
1204 | unsigned int tag) | |
1da177e4 | 1205 | { |
cebbd29e KK |
1206 | struct radix_tree_iter iter; |
1207 | void **slot; | |
1208 | unsigned int ret = 0; | |
612d6c19 | 1209 | |
cebbd29e | 1210 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1211 | return 0; |
1212 | ||
cebbd29e | 1213 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1214 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1215 | if (!results[ret]) |
1216 | continue; | |
b194d16c | 1217 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1218 | slot = radix_tree_iter_retry(&iter); |
1219 | continue; | |
1220 | } | |
cebbd29e | 1221 | if (++ret == max_items) |
1da177e4 | 1222 | break; |
1da177e4 | 1223 | } |
7cf9c2c7 | 1224 | |
1da177e4 LT |
1225 | return ret; |
1226 | } | |
1227 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1228 | ||
47feff2c NP |
1229 | /** |
1230 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1231 | * radix tree based on a tag | |
1232 | * @root: radix tree root | |
1233 | * @results: where the results of the lookup are placed | |
1234 | * @first_index: start the lookup from this key | |
1235 | * @max_items: place up to this many items at *results | |
1236 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1237 | * | |
1238 | * Performs an index-ascending scan of the tree for present items which | |
1239 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1240 | * returns the number of slots which were placed at *@results. | |
1241 | */ | |
1242 | unsigned int | |
1243 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1244 | unsigned long first_index, unsigned int max_items, | |
1245 | unsigned int tag) | |
1246 | { | |
cebbd29e KK |
1247 | struct radix_tree_iter iter; |
1248 | void **slot; | |
1249 | unsigned int ret = 0; | |
47feff2c | 1250 | |
cebbd29e | 1251 | if (unlikely(!max_items)) |
47feff2c NP |
1252 | return 0; |
1253 | ||
cebbd29e KK |
1254 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1255 | results[ret] = slot; | |
1256 | if (++ret == max_items) | |
47feff2c | 1257 | break; |
47feff2c NP |
1258 | } |
1259 | ||
1260 | return ret; | |
1261 | } | |
1262 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1263 | ||
e504f3fd HD |
1264 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1265 | #include <linux/sched.h> /* for cond_resched() */ | |
1266 | ||
0a2efc6c MW |
1267 | struct locate_info { |
1268 | unsigned long found_index; | |
1269 | bool stop; | |
1270 | }; | |
1271 | ||
e504f3fd HD |
1272 | /* |
1273 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1274 | */ | |
1275 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
0a2efc6c | 1276 | unsigned long index, struct locate_info *info) |
e504f3fd | 1277 | { |
0c7fa0a8 | 1278 | unsigned int shift; |
e504f3fd HD |
1279 | unsigned long i; |
1280 | ||
c12e51b0 | 1281 | shift = slot->shift + RADIX_TREE_MAP_SHIFT; |
e504f3fd | 1282 | |
0a2efc6c MW |
1283 | do { |
1284 | shift -= RADIX_TREE_MAP_SHIFT; | |
e504f3fd | 1285 | |
0a2efc6c MW |
1286 | for (i = (index >> shift) & RADIX_TREE_MAP_MASK; |
1287 | i < RADIX_TREE_MAP_SIZE; | |
1288 | i++, index += (1UL << shift)) { | |
1289 | struct radix_tree_node *node = | |
1290 | rcu_dereference_raw(slot->slots[i]); | |
1291 | if (node == RADIX_TREE_RETRY) | |
1292 | goto out; | |
b194d16c | 1293 | if (!radix_tree_is_internal_node(node)) { |
0a2efc6c MW |
1294 | if (node == item) { |
1295 | info->found_index = index; | |
1296 | info->stop = true; | |
1297 | goto out; | |
1298 | } | |
1299 | continue; | |
e6145236 | 1300 | } |
4dd6c098 | 1301 | node = entry_to_node(node); |
0a2efc6c MW |
1302 | if (is_sibling_entry(slot, node)) |
1303 | continue; | |
1304 | slot = node; | |
1305 | break; | |
e6145236 | 1306 | } |
0a2efc6c MW |
1307 | if (i == RADIX_TREE_MAP_SIZE) |
1308 | break; | |
1309 | } while (shift); | |
e504f3fd | 1310 | |
e504f3fd | 1311 | out: |
0a2efc6c MW |
1312 | if ((index == 0) && (i == RADIX_TREE_MAP_SIZE)) |
1313 | info->stop = true; | |
e504f3fd HD |
1314 | return index; |
1315 | } | |
1316 | ||
1317 | /** | |
1318 | * radix_tree_locate_item - search through radix tree for item | |
1319 | * @root: radix tree root | |
1320 | * @item: item to be found | |
1321 | * | |
1322 | * Returns index where item was found, or -1 if not found. | |
1323 | * Caller must hold no lock (since this time-consuming function needs | |
1324 | * to be preemptible), and must check afterwards if item is still there. | |
1325 | */ | |
1326 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1327 | { | |
1328 | struct radix_tree_node *node; | |
1329 | unsigned long max_index; | |
1330 | unsigned long cur_index = 0; | |
0a2efc6c MW |
1331 | struct locate_info info = { |
1332 | .found_index = -1, | |
1333 | .stop = false, | |
1334 | }; | |
e504f3fd HD |
1335 | |
1336 | do { | |
1337 | rcu_read_lock(); | |
1338 | node = rcu_dereference_raw(root->rnode); | |
b194d16c | 1339 | if (!radix_tree_is_internal_node(node)) { |
e504f3fd HD |
1340 | rcu_read_unlock(); |
1341 | if (node == item) | |
0a2efc6c | 1342 | info.found_index = 0; |
e504f3fd HD |
1343 | break; |
1344 | } | |
1345 | ||
4dd6c098 | 1346 | node = entry_to_node(node); |
0a2efc6c MW |
1347 | |
1348 | max_index = node_maxindex(node); | |
5f30fc94 HD |
1349 | if (cur_index > max_index) { |
1350 | rcu_read_unlock(); | |
e504f3fd | 1351 | break; |
5f30fc94 | 1352 | } |
e504f3fd | 1353 | |
0a2efc6c | 1354 | cur_index = __locate(node, item, cur_index, &info); |
e504f3fd HD |
1355 | rcu_read_unlock(); |
1356 | cond_resched(); | |
0a2efc6c | 1357 | } while (!info.stop && cur_index <= max_index); |
e504f3fd | 1358 | |
0a2efc6c | 1359 | return info.found_index; |
e504f3fd HD |
1360 | } |
1361 | #else | |
1362 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1363 | { | |
1364 | return -1; | |
1365 | } | |
1366 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1367 | |
a5f51c96 | 1368 | /** |
d0891265 | 1369 | * radix_tree_shrink - shrink radix tree to minimum height |
a5f51c96 NP |
1370 | * @root radix tree root |
1371 | */ | |
fb209019 | 1372 | static inline bool radix_tree_shrink(struct radix_tree_root *root) |
a5f51c96 | 1373 | { |
fb209019 MW |
1374 | bool shrunk = false; |
1375 | ||
d0891265 | 1376 | for (;;) { |
af49a63e MW |
1377 | struct radix_tree_node *node = root->rnode; |
1378 | struct radix_tree_node *child; | |
a5f51c96 | 1379 | |
af49a63e | 1380 | if (!radix_tree_is_internal_node(node)) |
d0891265 | 1381 | break; |
af49a63e | 1382 | node = entry_to_node(node); |
c0bc9875 NP |
1383 | |
1384 | /* | |
1385 | * The candidate node has more than one child, or its child | |
d0891265 MW |
1386 | * is not at the leftmost slot, or the child is a multiorder |
1387 | * entry, we cannot shrink. | |
c0bc9875 | 1388 | */ |
af49a63e | 1389 | if (node->count != 1) |
c0bc9875 | 1390 | break; |
af49a63e MW |
1391 | child = node->slots[0]; |
1392 | if (!child) | |
c0bc9875 | 1393 | break; |
af49a63e | 1394 | if (!radix_tree_is_internal_node(child) && node->shift) |
afe0e395 MW |
1395 | break; |
1396 | ||
af49a63e MW |
1397 | if (radix_tree_is_internal_node(child)) |
1398 | entry_to_node(child)->parent = NULL; | |
c0bc9875 | 1399 | |
7cf9c2c7 NP |
1400 | /* |
1401 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1402 | * moving the node from one part of the tree to another: if it |
1403 | * was safe to dereference the old pointer to it | |
af49a63e | 1404 | * (node->slots[0]), it will be safe to dereference the new |
27d20fdd | 1405 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1406 | */ |
af49a63e | 1407 | root->rnode = child; |
27d20fdd NP |
1408 | |
1409 | /* | |
1410 | * We have a dilemma here. The node's slot[0] must not be | |
1411 | * NULLed in case there are concurrent lookups expecting to | |
1412 | * find the item. However if this was a bottom-level node, | |
1413 | * then it may be subject to the slot pointer being visible | |
1414 | * to callers dereferencing it. If item corresponding to | |
1415 | * slot[0] is subsequently deleted, these callers would expect | |
1416 | * their slot to become empty sooner or later. | |
1417 | * | |
1418 | * For example, lockless pagecache will look up a slot, deref | |
2fcd9005 | 1419 | * the page pointer, and if the page has 0 refcount it means it |
27d20fdd NP |
1420 | * was concurrently deleted from pagecache so try the deref |
1421 | * again. Fortunately there is already a requirement for logic | |
1422 | * to retry the entire slot lookup -- the indirect pointer | |
1423 | * problem (replacing direct root node with an indirect pointer | |
1424 | * also results in a stale slot). So tag the slot as indirect | |
1425 | * to force callers to retry. | |
1426 | */ | |
af49a63e MW |
1427 | if (!radix_tree_is_internal_node(child)) |
1428 | node->slots[0] = RADIX_TREE_RETRY; | |
27d20fdd | 1429 | |
af49a63e | 1430 | radix_tree_node_free(node); |
fb209019 | 1431 | shrunk = true; |
a5f51c96 | 1432 | } |
fb209019 MW |
1433 | |
1434 | return shrunk; | |
a5f51c96 NP |
1435 | } |
1436 | ||
139e5616 JW |
1437 | /** |
1438 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1439 | * @root: radix tree root | |
139e5616 JW |
1440 | * @node: node containing @index |
1441 | * | |
1442 | * After clearing the slot at @index in @node from radix tree | |
1443 | * rooted at @root, call this function to attempt freeing the | |
1444 | * node and shrinking the tree. | |
1445 | * | |
1446 | * Returns %true if @node was freed, %false otherwise. | |
1447 | */ | |
449dd698 | 1448 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1449 | struct radix_tree_node *node) |
1450 | { | |
1451 | bool deleted = false; | |
1452 | ||
1453 | do { | |
1454 | struct radix_tree_node *parent; | |
1455 | ||
1456 | if (node->count) { | |
4dd6c098 | 1457 | if (node == entry_to_node(root->rnode)) |
fb209019 | 1458 | deleted |= radix_tree_shrink(root); |
139e5616 JW |
1459 | return deleted; |
1460 | } | |
1461 | ||
1462 | parent = node->parent; | |
1463 | if (parent) { | |
0c7fa0a8 | 1464 | parent->slots[node->offset] = NULL; |
139e5616 JW |
1465 | parent->count--; |
1466 | } else { | |
1467 | root_tag_clear_all(root); | |
139e5616 JW |
1468 | root->rnode = NULL; |
1469 | } | |
1470 | ||
1471 | radix_tree_node_free(node); | |
1472 | deleted = true; | |
1473 | ||
1474 | node = parent; | |
1475 | } while (node); | |
1476 | ||
1477 | return deleted; | |
1478 | } | |
1479 | ||
57578c2e MW |
1480 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1481 | void *ptr, unsigned offset) | |
1482 | { | |
1483 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1484 | int i; | |
1485 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1486 | if (node->slots[offset + i] != ptr) | |
1487 | break; | |
1488 | node->slots[offset + i] = NULL; | |
1489 | node->count--; | |
1490 | } | |
1491 | #endif | |
1492 | } | |
1493 | ||
1da177e4 | 1494 | /** |
53c59f26 | 1495 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1496 | * @root: radix tree root |
1497 | * @index: index key | |
53c59f26 | 1498 | * @item: expected item |
1da177e4 | 1499 | * |
53c59f26 | 1500 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1501 | * |
53c59f26 JW |
1502 | * Returns the address of the deleted item, or NULL if it was not present |
1503 | * or the entry at the given @index was not @item. | |
1da177e4 | 1504 | */ |
53c59f26 JW |
1505 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1506 | unsigned long index, void *item) | |
1da177e4 | 1507 | { |
139e5616 | 1508 | struct radix_tree_node *node; |
57578c2e | 1509 | unsigned int offset; |
139e5616 JW |
1510 | void **slot; |
1511 | void *entry; | |
d5274261 | 1512 | int tag; |
1da177e4 | 1513 | |
139e5616 JW |
1514 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1515 | if (!entry) | |
1516 | return NULL; | |
1da177e4 | 1517 | |
139e5616 JW |
1518 | if (item && entry != item) |
1519 | return NULL; | |
1520 | ||
1521 | if (!node) { | |
612d6c19 NP |
1522 | root_tag_clear_all(root); |
1523 | root->rnode = NULL; | |
139e5616 | 1524 | return entry; |
612d6c19 | 1525 | } |
1da177e4 | 1526 | |
29e0967c | 1527 | offset = get_slot_offset(node, slot); |
53c59f26 | 1528 | |
d604c324 MW |
1529 | /* Clear all tags associated with the item to be deleted. */ |
1530 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1531 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1532 | |
a4db4dce | 1533 | delete_sibling_entries(node, node_to_entry(slot), offset); |
139e5616 JW |
1534 | node->slots[offset] = NULL; |
1535 | node->count--; | |
e2bdb933 | 1536 | |
449dd698 | 1537 | __radix_tree_delete_node(root, node); |
612d6c19 | 1538 | |
139e5616 | 1539 | return entry; |
1da177e4 | 1540 | } |
53c59f26 JW |
1541 | EXPORT_SYMBOL(radix_tree_delete_item); |
1542 | ||
1543 | /** | |
1544 | * radix_tree_delete - delete an item from a radix tree | |
1545 | * @root: radix tree root | |
1546 | * @index: index key | |
1547 | * | |
1548 | * Remove the item at @index from the radix tree rooted at @root. | |
1549 | * | |
1550 | * Returns the address of the deleted item, or NULL if it was not present. | |
1551 | */ | |
1552 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1553 | { | |
1554 | return radix_tree_delete_item(root, index, NULL); | |
1555 | } | |
1da177e4 LT |
1556 | EXPORT_SYMBOL(radix_tree_delete); |
1557 | ||
d604c324 MW |
1558 | struct radix_tree_node *radix_tree_replace_clear_tags( |
1559 | struct radix_tree_root *root, | |
1560 | unsigned long index, void *entry) | |
1561 | { | |
1562 | struct radix_tree_node *node; | |
1563 | void **slot; | |
1564 | ||
1565 | __radix_tree_lookup(root, index, &node, &slot); | |
1566 | ||
1567 | if (node) { | |
1568 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1569 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1570 | node_tag_clear(root, node, tag, offset); | |
1571 | } else { | |
1572 | /* Clear root node tags */ | |
1573 | root->gfp_mask &= __GFP_BITS_MASK; | |
1574 | } | |
1575 | ||
1576 | radix_tree_replace_slot(slot, entry); | |
1577 | return node; | |
1578 | } | |
1579 | ||
1da177e4 LT |
1580 | /** |
1581 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1582 | * @root: radix tree root | |
1583 | * @tag: tag to test | |
1584 | */ | |
daff89f3 | 1585 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1586 | { |
612d6c19 | 1587 | return root_tag_get(root, tag); |
1da177e4 LT |
1588 | } |
1589 | EXPORT_SYMBOL(radix_tree_tagged); | |
1590 | ||
1591 | static void | |
449dd698 | 1592 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1593 | { |
449dd698 JW |
1594 | struct radix_tree_node *node = arg; |
1595 | ||
1596 | memset(node, 0, sizeof(*node)); | |
1597 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1598 | } |
1599 | ||
1da177e4 | 1600 | static int radix_tree_callback(struct notifier_block *nfb, |
2fcd9005 | 1601 | unsigned long action, void *hcpu) |
1da177e4 | 1602 | { |
2fcd9005 MW |
1603 | int cpu = (long)hcpu; |
1604 | struct radix_tree_preload *rtp; | |
1605 | struct radix_tree_node *node; | |
1606 | ||
1607 | /* Free per-cpu pool of preloaded nodes */ | |
1608 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
1609 | rtp = &per_cpu(radix_tree_preloads, cpu); | |
1610 | while (rtp->nr) { | |
9d2a8da0 KS |
1611 | node = rtp->nodes; |
1612 | rtp->nodes = node->private_data; | |
1613 | kmem_cache_free(radix_tree_node_cachep, node); | |
1614 | rtp->nr--; | |
2fcd9005 MW |
1615 | } |
1616 | } | |
1617 | return NOTIFY_OK; | |
1da177e4 | 1618 | } |
1da177e4 LT |
1619 | |
1620 | void __init radix_tree_init(void) | |
1621 | { | |
1622 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1623 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1624 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1625 | radix_tree_node_ctor); | |
1da177e4 LT |
1626 | hotcpu_notifier(radix_tree_callback, 0); |
1627 | } |