Commit | Line | Data |
---|---|---|
de6cc651 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Copyright (C) 2001 Momchil Velikov | |
4 | * Portions Copyright (C) 2001 Christoph Hellwig | |
cde53535 | 5 | * Copyright (C) 2005 SGI, Christoph Lameter |
7cf9c2c7 | 6 | * Copyright (C) 2006 Nick Piggin |
78c1d784 | 7 | * Copyright (C) 2012 Konstantin Khlebnikov |
6b053b8e MW |
8 | * Copyright (C) 2016 Intel, Matthew Wilcox |
9 | * Copyright (C) 2016 Intel, Ross Zwisler | |
1da177e4 LT |
10 | */ |
11 | ||
0a835c4f MW |
12 | #include <linux/bitmap.h> |
13 | #include <linux/bitops.h> | |
460488c5 | 14 | #include <linux/bug.h> |
e157b555 | 15 | #include <linux/cpu.h> |
1da177e4 | 16 | #include <linux/errno.h> |
0a835c4f MW |
17 | #include <linux/export.h> |
18 | #include <linux/idr.h> | |
1da177e4 LT |
19 | #include <linux/init.h> |
20 | #include <linux/kernel.h> | |
0a835c4f | 21 | #include <linux/kmemleak.h> |
1da177e4 | 22 | #include <linux/percpu.h> |
cfa6705d | 23 | #include <linux/local_lock.h> |
0a835c4f MW |
24 | #include <linux/preempt.h> /* in_interrupt() */ |
25 | #include <linux/radix-tree.h> | |
26 | #include <linux/rcupdate.h> | |
1da177e4 | 27 | #include <linux/slab.h> |
1da177e4 | 28 | #include <linux/string.h> |
02c02bf1 | 29 | #include <linux/xarray.h> |
1da177e4 | 30 | |
1da177e4 LT |
31 | /* |
32 | * Radix tree node cache. | |
33 | */ | |
58d6ea30 | 34 | struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 35 | |
55368052 NP |
36 | /* |
37 | * The radix tree is variable-height, so an insert operation not only has | |
38 | * to build the branch to its corresponding item, it also has to build the | |
39 | * branch to existing items if the size has to be increased (by | |
40 | * radix_tree_extend). | |
41 | * | |
42 | * The worst case is a zero height tree with just a single item at index 0, | |
43 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
44 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
45 | * Hence: | |
46 | */ | |
47 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
48 | ||
0a835c4f MW |
49 | /* |
50 | * The IDR does not have to be as high as the radix tree since it uses | |
51 | * signed integers, not unsigned longs. | |
52 | */ | |
53 | #define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1) | |
54 | #define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \ | |
55 | RADIX_TREE_MAP_SHIFT)) | |
56 | #define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1) | |
57 | ||
1da177e4 LT |
58 | /* |
59 | * Per-cpu pool of preloaded nodes | |
60 | */ | |
cfa6705d SAS |
61 | DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { |
62 | .lock = INIT_LOCAL_LOCK(lock), | |
1da177e4 | 63 | }; |
cfa6705d | 64 | EXPORT_PER_CPU_SYMBOL_GPL(radix_tree_preloads); |
1da177e4 | 65 | |
148deab2 MW |
66 | static inline struct radix_tree_node *entry_to_node(void *ptr) |
67 | { | |
68 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE); | |
69 | } | |
70 | ||
a4db4dce | 71 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 72 | { |
30ff46cc | 73 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
74 | } |
75 | ||
02c02bf1 | 76 | #define RADIX_TREE_RETRY XA_RETRY_ENTRY |
db050f29 | 77 | |
d7b62727 MW |
78 | static inline unsigned long |
79 | get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot) | |
db050f29 | 80 | { |
76f070b4 | 81 | return parent ? slot - parent->slots : 0; |
db050f29 MW |
82 | } |
83 | ||
35534c86 | 84 | static unsigned int radix_tree_descend(const struct radix_tree_node *parent, |
9e85d811 | 85 | struct radix_tree_node **nodep, unsigned long index) |
db050f29 | 86 | { |
9e85d811 | 87 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
d7b62727 | 88 | void __rcu **entry = rcu_dereference_raw(parent->slots[offset]); |
db050f29 | 89 | |
db050f29 MW |
90 | *nodep = (void *)entry; |
91 | return offset; | |
92 | } | |
93 | ||
35534c86 | 94 | static inline gfp_t root_gfp_mask(const struct radix_tree_root *root) |
612d6c19 | 95 | { |
f8d5d0cc | 96 | return root->xa_flags & (__GFP_BITS_MASK & ~GFP_ZONEMASK); |
612d6c19 NP |
97 | } |
98 | ||
643b52b9 NP |
99 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
100 | int offset) | |
101 | { | |
102 | __set_bit(offset, node->tags[tag]); | |
103 | } | |
104 | ||
105 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
106 | int offset) | |
107 | { | |
108 | __clear_bit(offset, node->tags[tag]); | |
109 | } | |
110 | ||
35534c86 | 111 | static inline int tag_get(const struct radix_tree_node *node, unsigned int tag, |
643b52b9 NP |
112 | int offset) |
113 | { | |
114 | return test_bit(offset, node->tags[tag]); | |
115 | } | |
116 | ||
35534c86 | 117 | static inline void root_tag_set(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 118 | { |
f8d5d0cc | 119 | root->xa_flags |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
120 | } |
121 | ||
2fcd9005 | 122 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 123 | { |
f8d5d0cc | 124 | root->xa_flags &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
125 | } |
126 | ||
127 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
128 | { | |
f8d5d0cc | 129 | root->xa_flags &= (__force gfp_t)((1 << ROOT_TAG_SHIFT) - 1); |
643b52b9 NP |
130 | } |
131 | ||
35534c86 | 132 | static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag) |
643b52b9 | 133 | { |
f8d5d0cc | 134 | return (__force int)root->xa_flags & (1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
135 | } |
136 | ||
35534c86 | 137 | static inline unsigned root_tags_get(const struct radix_tree_root *root) |
643b52b9 | 138 | { |
f8d5d0cc | 139 | return (__force unsigned)root->xa_flags >> ROOT_TAG_SHIFT; |
643b52b9 NP |
140 | } |
141 | ||
0a835c4f | 142 | static inline bool is_idr(const struct radix_tree_root *root) |
7b60e9ad | 143 | { |
f8d5d0cc | 144 | return !!(root->xa_flags & ROOT_IS_IDR); |
7b60e9ad MW |
145 | } |
146 | ||
643b52b9 NP |
147 | /* |
148 | * Returns 1 if any slot in the node has this tag set. | |
149 | * Otherwise returns 0. | |
150 | */ | |
35534c86 MW |
151 | static inline int any_tag_set(const struct radix_tree_node *node, |
152 | unsigned int tag) | |
643b52b9 | 153 | { |
2fcd9005 | 154 | unsigned idx; |
643b52b9 NP |
155 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
156 | if (node->tags[tag][idx]) | |
157 | return 1; | |
158 | } | |
159 | return 0; | |
160 | } | |
78c1d784 | 161 | |
0a835c4f MW |
162 | static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag) |
163 | { | |
164 | bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE); | |
165 | } | |
166 | ||
78c1d784 KK |
167 | /** |
168 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
169 | * | |
170 | * @addr: The address to base the search on | |
171 | * @size: The bitmap size in bits | |
172 | * @offset: The bitnumber to start searching at | |
173 | * | |
174 | * Unrollable variant of find_next_bit() for constant size arrays. | |
175 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
176 | * Returns next bit offset, or size if nothing found. | |
177 | */ | |
178 | static __always_inline unsigned long | |
bc412fca MW |
179 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
180 | unsigned long offset) | |
78c1d784 | 181 | { |
bc412fca | 182 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 183 | |
bc412fca | 184 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
185 | unsigned long tmp; |
186 | ||
187 | addr += offset / BITS_PER_LONG; | |
188 | tmp = *addr >> (offset % BITS_PER_LONG); | |
189 | if (tmp) | |
190 | return __ffs(tmp) + offset; | |
191 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 192 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
193 | tmp = *++addr; |
194 | if (tmp) | |
195 | return __ffs(tmp) + offset; | |
196 | offset += BITS_PER_LONG; | |
197 | } | |
198 | } | |
bc412fca | 199 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
200 | } |
201 | ||
268f42de MW |
202 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
203 | { | |
3a08cd52 | 204 | return iter->index & RADIX_TREE_MAP_MASK; |
268f42de MW |
205 | } |
206 | ||
218ed750 MW |
207 | /* |
208 | * The maximum index which can be stored in a radix tree | |
209 | */ | |
210 | static inline unsigned long shift_maxindex(unsigned int shift) | |
211 | { | |
212 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
213 | } | |
214 | ||
35534c86 | 215 | static inline unsigned long node_maxindex(const struct radix_tree_node *node) |
218ed750 MW |
216 | { |
217 | return shift_maxindex(node->shift); | |
218 | } | |
219 | ||
0a835c4f MW |
220 | static unsigned long next_index(unsigned long index, |
221 | const struct radix_tree_node *node, | |
222 | unsigned long offset) | |
223 | { | |
224 | return (index & ~node_maxindex(node)) + (offset << node->shift); | |
225 | } | |
226 | ||
1da177e4 LT |
227 | /* |
228 | * This assumes that the caller has performed appropriate preallocation, and | |
229 | * that the caller has pinned this thread of control to the current CPU. | |
230 | */ | |
231 | static struct radix_tree_node * | |
0a835c4f | 232 | radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent, |
d58275bc | 233 | struct radix_tree_root *root, |
e8de4340 | 234 | unsigned int shift, unsigned int offset, |
01959dfe | 235 | unsigned int count, unsigned int nr_values) |
1da177e4 | 236 | { |
e2848a0e | 237 | struct radix_tree_node *ret = NULL; |
1da177e4 | 238 | |
5e4c0d97 | 239 | /* |
2fcd9005 MW |
240 | * Preload code isn't irq safe and it doesn't make sense to use |
241 | * preloading during an interrupt anyway as all the allocations have | |
242 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 243 | */ |
d0164adc | 244 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
245 | struct radix_tree_preload *rtp; |
246 | ||
58e698af VD |
247 | /* |
248 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
249 | * cache first for the new node to get accounted to the memory |
250 | * cgroup. | |
58e698af VD |
251 | */ |
252 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 253 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
254 | if (ret) |
255 | goto out; | |
256 | ||
e2848a0e NP |
257 | /* |
258 | * Provided the caller has preloaded here, we will always | |
259 | * succeed in getting a node here (and never reach | |
260 | * kmem_cache_alloc) | |
261 | */ | |
7c8e0181 | 262 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 263 | if (rtp->nr) { |
9d2a8da0 | 264 | ret = rtp->nodes; |
1293d5c5 | 265 | rtp->nodes = ret->parent; |
1da177e4 LT |
266 | rtp->nr--; |
267 | } | |
ce80b067 CM |
268 | /* |
269 | * Update the allocation stack trace as this is more useful | |
270 | * for debugging. | |
271 | */ | |
272 | kmemleak_update_trace(ret); | |
58e698af | 273 | goto out; |
1da177e4 | 274 | } |
05eb6e72 | 275 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 276 | out: |
b194d16c | 277 | BUG_ON(radix_tree_is_internal_node(ret)); |
e8de4340 | 278 | if (ret) { |
e8de4340 MW |
279 | ret->shift = shift; |
280 | ret->offset = offset; | |
281 | ret->count = count; | |
01959dfe | 282 | ret->nr_values = nr_values; |
d58275bc | 283 | ret->parent = parent; |
01959dfe | 284 | ret->array = root; |
e8de4340 | 285 | } |
1da177e4 LT |
286 | return ret; |
287 | } | |
288 | ||
58d6ea30 | 289 | void radix_tree_node_rcu_free(struct rcu_head *head) |
7cf9c2c7 NP |
290 | { |
291 | struct radix_tree_node *node = | |
292 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
293 | |
294 | /* | |
175542f5 MW |
295 | * Must only free zeroed nodes into the slab. We can be left with |
296 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
297 | * and tags here. | |
643b52b9 | 298 | */ |
175542f5 MW |
299 | memset(node->slots, 0, sizeof(node->slots)); |
300 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 301 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 302 | |
7cf9c2c7 NP |
303 | kmem_cache_free(radix_tree_node_cachep, node); |
304 | } | |
305 | ||
1da177e4 LT |
306 | static inline void |
307 | radix_tree_node_free(struct radix_tree_node *node) | |
308 | { | |
7cf9c2c7 | 309 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
310 | } |
311 | ||
312 | /* | |
313 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
314 | * ensure that the addition of a single element in the tree cannot fail. On | |
315 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
316 | * with preemption not disabled. | |
b34df792 DH |
317 | * |
318 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 319 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 320 | */ |
bc9ae224 | 321 | static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr) |
1da177e4 LT |
322 | { |
323 | struct radix_tree_preload *rtp; | |
324 | struct radix_tree_node *node; | |
325 | int ret = -ENOMEM; | |
326 | ||
05eb6e72 VD |
327 | /* |
328 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
329 | * they should never be accounted to any particular memory cgroup. | |
330 | */ | |
331 | gfp_mask &= ~__GFP_ACCOUNT; | |
332 | ||
cfa6705d | 333 | local_lock(&radix_tree_preloads.lock); |
7c8e0181 | 334 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 335 | while (rtp->nr < nr) { |
cfa6705d | 336 | local_unlock(&radix_tree_preloads.lock); |
488514d1 | 337 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
338 | if (node == NULL) |
339 | goto out; | |
cfa6705d | 340 | local_lock(&radix_tree_preloads.lock); |
7c8e0181 | 341 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 342 | if (rtp->nr < nr) { |
1293d5c5 | 343 | node->parent = rtp->nodes; |
9d2a8da0 KS |
344 | rtp->nodes = node; |
345 | rtp->nr++; | |
346 | } else { | |
1da177e4 | 347 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 348 | } |
1da177e4 LT |
349 | } |
350 | ret = 0; | |
351 | out: | |
352 | return ret; | |
353 | } | |
5e4c0d97 JK |
354 | |
355 | /* | |
356 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
357 | * ensure that the addition of a single element in the tree cannot fail. On | |
358 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
359 | * with preemption not disabled. | |
360 | * | |
361 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 362 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
363 | */ |
364 | int radix_tree_preload(gfp_t gfp_mask) | |
365 | { | |
366 | /* Warn on non-sensical use... */ | |
d0164adc | 367 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 368 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 369 | } |
d7f0923d | 370 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 371 | |
5e4c0d97 JK |
372 | /* |
373 | * The same as above function, except we don't guarantee preloading happens. | |
374 | * We do it, if we decide it helps. On success, return zero with preemption | |
375 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
376 | */ | |
377 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
378 | { | |
d0164adc | 379 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 380 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 381 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
cfa6705d | 382 | local_lock(&radix_tree_preloads.lock); |
5e4c0d97 JK |
383 | return 0; |
384 | } | |
385 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
386 | ||
35534c86 | 387 | static unsigned radix_tree_load_root(const struct radix_tree_root *root, |
1456a439 MW |
388 | struct radix_tree_node **nodep, unsigned long *maxindex) |
389 | { | |
f8d5d0cc | 390 | struct radix_tree_node *node = rcu_dereference_raw(root->xa_head); |
1456a439 MW |
391 | |
392 | *nodep = node; | |
393 | ||
b194d16c | 394 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 395 | node = entry_to_node(node); |
1456a439 | 396 | *maxindex = node_maxindex(node); |
c12e51b0 | 397 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
398 | } |
399 | ||
400 | *maxindex = 0; | |
401 | return 0; | |
402 | } | |
403 | ||
1da177e4 LT |
404 | /* |
405 | * Extend a radix tree so it can store key @index. | |
406 | */ | |
0a835c4f | 407 | static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp, |
d0891265 | 408 | unsigned long index, unsigned int shift) |
1da177e4 | 409 | { |
d7b62727 | 410 | void *entry; |
d0891265 | 411 | unsigned int maxshift; |
1da177e4 LT |
412 | int tag; |
413 | ||
d0891265 MW |
414 | /* Figure out what the shift should be. */ |
415 | maxshift = shift; | |
416 | while (index > shift_maxindex(maxshift)) | |
417 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 418 | |
f8d5d0cc | 419 | entry = rcu_dereference_raw(root->xa_head); |
d7b62727 | 420 | if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE))) |
1da177e4 | 421 | goto out; |
1da177e4 | 422 | |
1da177e4 | 423 | do { |
0a835c4f | 424 | struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL, |
d58275bc | 425 | root, shift, 0, 1, 0); |
2fcd9005 | 426 | if (!node) |
1da177e4 LT |
427 | return -ENOMEM; |
428 | ||
0a835c4f MW |
429 | if (is_idr(root)) { |
430 | all_tag_set(node, IDR_FREE); | |
431 | if (!root_tag_get(root, IDR_FREE)) { | |
432 | tag_clear(node, IDR_FREE, 0); | |
433 | root_tag_set(root, IDR_FREE); | |
434 | } | |
435 | } else { | |
436 | /* Propagate the aggregated tag info to the new child */ | |
437 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
438 | if (root_tag_get(root, tag)) | |
439 | tag_set(node, tag, 0); | |
440 | } | |
1da177e4 LT |
441 | } |
442 | ||
d0891265 | 443 | BUG_ON(shift > BITS_PER_LONG); |
d7b62727 MW |
444 | if (radix_tree_is_internal_node(entry)) { |
445 | entry_to_node(entry)->parent = node; | |
3159f943 | 446 | } else if (xa_is_value(entry)) { |
01959dfe MW |
447 | /* Moving a value entry root->xa_head to a node */ |
448 | node->nr_values = 1; | |
f7942430 | 449 | } |
d7b62727 MW |
450 | /* |
451 | * entry was already in the radix tree, so we do not need | |
452 | * rcu_assign_pointer here | |
453 | */ | |
454 | node->slots[0] = (void __rcu *)entry; | |
455 | entry = node_to_entry(node); | |
f8d5d0cc | 456 | rcu_assign_pointer(root->xa_head, entry); |
d0891265 | 457 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 458 | } while (shift <= maxshift); |
1da177e4 | 459 | out: |
d0891265 | 460 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
461 | } |
462 | ||
f4b109c6 JW |
463 | /** |
464 | * radix_tree_shrink - shrink radix tree to minimum height | |
465 | * @root radix tree root | |
466 | */ | |
1cf56f9d | 467 | static inline bool radix_tree_shrink(struct radix_tree_root *root) |
f4b109c6 | 468 | { |
0ac398ef MW |
469 | bool shrunk = false; |
470 | ||
f4b109c6 | 471 | for (;;) { |
f8d5d0cc | 472 | struct radix_tree_node *node = rcu_dereference_raw(root->xa_head); |
f4b109c6 JW |
473 | struct radix_tree_node *child; |
474 | ||
475 | if (!radix_tree_is_internal_node(node)) | |
476 | break; | |
477 | node = entry_to_node(node); | |
478 | ||
479 | /* | |
480 | * The candidate node has more than one child, or its child | |
3a08cd52 | 481 | * is not at the leftmost slot, we cannot shrink. |
f4b109c6 JW |
482 | */ |
483 | if (node->count != 1) | |
484 | break; | |
12320d0f | 485 | child = rcu_dereference_raw(node->slots[0]); |
f4b109c6 JW |
486 | if (!child) |
487 | break; | |
f4b109c6 | 488 | |
66ee620f MW |
489 | /* |
490 | * For an IDR, we must not shrink entry 0 into the root in | |
491 | * case somebody calls idr_replace() with a pointer that | |
492 | * appears to be an internal entry | |
493 | */ | |
494 | if (!node->shift && is_idr(root)) | |
495 | break; | |
496 | ||
f4b109c6 JW |
497 | if (radix_tree_is_internal_node(child)) |
498 | entry_to_node(child)->parent = NULL; | |
499 | ||
500 | /* | |
501 | * We don't need rcu_assign_pointer(), since we are simply | |
502 | * moving the node from one part of the tree to another: if it | |
503 | * was safe to dereference the old pointer to it | |
504 | * (node->slots[0]), it will be safe to dereference the new | |
f8d5d0cc | 505 | * one (root->xa_head) as far as dependent read barriers go. |
f4b109c6 | 506 | */ |
f8d5d0cc | 507 | root->xa_head = (void __rcu *)child; |
0a835c4f MW |
508 | if (is_idr(root) && !tag_get(node, IDR_FREE, 0)) |
509 | root_tag_clear(root, IDR_FREE); | |
f4b109c6 JW |
510 | |
511 | /* | |
512 | * We have a dilemma here. The node's slot[0] must not be | |
513 | * NULLed in case there are concurrent lookups expecting to | |
514 | * find the item. However if this was a bottom-level node, | |
515 | * then it may be subject to the slot pointer being visible | |
516 | * to callers dereferencing it. If item corresponding to | |
517 | * slot[0] is subsequently deleted, these callers would expect | |
518 | * their slot to become empty sooner or later. | |
519 | * | |
520 | * For example, lockless pagecache will look up a slot, deref | |
521 | * the page pointer, and if the page has 0 refcount it means it | |
522 | * was concurrently deleted from pagecache so try the deref | |
523 | * again. Fortunately there is already a requirement for logic | |
524 | * to retry the entire slot lookup -- the indirect pointer | |
525 | * problem (replacing direct root node with an indirect pointer | |
526 | * also results in a stale slot). So tag the slot as indirect | |
527 | * to force callers to retry. | |
528 | */ | |
4d693d08 JW |
529 | node->count = 0; |
530 | if (!radix_tree_is_internal_node(child)) { | |
d7b62727 | 531 | node->slots[0] = (void __rcu *)RADIX_TREE_RETRY; |
4d693d08 | 532 | } |
f4b109c6 | 533 | |
ea07b862 | 534 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 535 | radix_tree_node_free(node); |
0ac398ef | 536 | shrunk = true; |
f4b109c6 | 537 | } |
0ac398ef MW |
538 | |
539 | return shrunk; | |
f4b109c6 JW |
540 | } |
541 | ||
0ac398ef | 542 | static bool delete_node(struct radix_tree_root *root, |
1cf56f9d | 543 | struct radix_tree_node *node) |
f4b109c6 | 544 | { |
0ac398ef MW |
545 | bool deleted = false; |
546 | ||
f4b109c6 JW |
547 | do { |
548 | struct radix_tree_node *parent; | |
549 | ||
550 | if (node->count) { | |
12320d0f | 551 | if (node_to_entry(node) == |
f8d5d0cc | 552 | rcu_dereference_raw(root->xa_head)) |
1cf56f9d | 553 | deleted |= radix_tree_shrink(root); |
0ac398ef | 554 | return deleted; |
f4b109c6 JW |
555 | } |
556 | ||
557 | parent = node->parent; | |
558 | if (parent) { | |
559 | parent->slots[node->offset] = NULL; | |
560 | parent->count--; | |
561 | } else { | |
0a835c4f MW |
562 | /* |
563 | * Shouldn't the tags already have all been cleared | |
564 | * by the caller? | |
565 | */ | |
566 | if (!is_idr(root)) | |
567 | root_tag_clear_all(root); | |
f8d5d0cc | 568 | root->xa_head = NULL; |
f4b109c6 JW |
569 | } |
570 | ||
ea07b862 | 571 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 572 | radix_tree_node_free(node); |
0ac398ef | 573 | deleted = true; |
f4b109c6 JW |
574 | |
575 | node = parent; | |
576 | } while (node); | |
0ac398ef MW |
577 | |
578 | return deleted; | |
f4b109c6 JW |
579 | } |
580 | ||
1da177e4 | 581 | /** |
139e5616 | 582 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
583 | * @root: radix tree root |
584 | * @index: index key | |
139e5616 JW |
585 | * @nodep: returns node |
586 | * @slotp: returns slot | |
1da177e4 | 587 | * |
139e5616 JW |
588 | * Create, if necessary, and return the node and slot for an item |
589 | * at position @index in the radix tree @root. | |
590 | * | |
591 | * Until there is more than one item in the tree, no nodes are | |
f8d5d0cc | 592 | * allocated and @root->xa_head is used as a direct slot instead of |
139e5616 JW |
593 | * pointing to a node, in which case *@nodep will be NULL. |
594 | * | |
595 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 596 | */ |
74d60958 | 597 | static int __radix_tree_create(struct radix_tree_root *root, |
3a08cd52 MW |
598 | unsigned long index, struct radix_tree_node **nodep, |
599 | void __rcu ***slotp) | |
1da177e4 | 600 | { |
89148aa4 | 601 | struct radix_tree_node *node = NULL, *child; |
f8d5d0cc | 602 | void __rcu **slot = (void __rcu **)&root->xa_head; |
49ea6ebc | 603 | unsigned long maxindex; |
89148aa4 | 604 | unsigned int shift, offset = 0; |
3a08cd52 | 605 | unsigned long max = index; |
0a835c4f | 606 | gfp_t gfp = root_gfp_mask(root); |
49ea6ebc | 607 | |
89148aa4 | 608 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
609 | |
610 | /* Make sure the tree is high enough. */ | |
49ea6ebc | 611 | if (max > maxindex) { |
0a835c4f | 612 | int error = radix_tree_extend(root, gfp, max, shift); |
49ea6ebc | 613 | if (error < 0) |
1da177e4 | 614 | return error; |
49ea6ebc | 615 | shift = error; |
f8d5d0cc | 616 | child = rcu_dereference_raw(root->xa_head); |
1da177e4 LT |
617 | } |
618 | ||
3a08cd52 | 619 | while (shift > 0) { |
c12e51b0 | 620 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 621 | if (child == NULL) { |
1da177e4 | 622 | /* Have to add a child node. */ |
d58275bc | 623 | child = radix_tree_node_alloc(gfp, node, root, shift, |
e8de4340 | 624 | offset, 0, 0); |
89148aa4 | 625 | if (!child) |
1da177e4 | 626 | return -ENOMEM; |
89148aa4 MW |
627 | rcu_assign_pointer(*slot, node_to_entry(child)); |
628 | if (node) | |
1da177e4 | 629 | node->count++; |
89148aa4 | 630 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 631 | break; |
1da177e4 LT |
632 | |
633 | /* Go a level down */ | |
89148aa4 | 634 | node = entry_to_node(child); |
9e85d811 | 635 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 636 | slot = &node->slots[offset]; |
e6145236 MW |
637 | } |
638 | ||
175542f5 MW |
639 | if (nodep) |
640 | *nodep = node; | |
641 | if (slotp) | |
642 | *slotp = slot; | |
643 | return 0; | |
644 | } | |
645 | ||
175542f5 MW |
646 | /* |
647 | * Free any nodes below this node. The tree is presumed to not need | |
648 | * shrinking, and any user data in the tree is presumed to not need a | |
649 | * destructor called on it. If we need to add a destructor, we can | |
650 | * add that functionality later. Note that we may not clear tags or | |
651 | * slots from the tree as an RCU walker may still have a pointer into | |
652 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
653 | * but we'll still have to clear those in rcu_free. | |
654 | */ | |
655 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
656 | { | |
657 | unsigned offset = 0; | |
658 | struct radix_tree_node *child = entry_to_node(node); | |
659 | ||
660 | for (;;) { | |
12320d0f | 661 | void *entry = rcu_dereference_raw(child->slots[offset]); |
02c02bf1 | 662 | if (xa_is_node(entry) && child->shift) { |
175542f5 MW |
663 | child = entry_to_node(entry); |
664 | offset = 0; | |
665 | continue; | |
666 | } | |
667 | offset++; | |
668 | while (offset == RADIX_TREE_MAP_SIZE) { | |
669 | struct radix_tree_node *old = child; | |
670 | offset = child->offset + 1; | |
671 | child = child->parent; | |
dd040b6f | 672 | WARN_ON_ONCE(!list_empty(&old->private_list)); |
175542f5 MW |
673 | radix_tree_node_free(old); |
674 | if (old == entry_to_node(node)) | |
675 | return; | |
676 | } | |
677 | } | |
678 | } | |
679 | ||
d7b62727 | 680 | static inline int insert_entries(struct radix_tree_node *node, |
3a08cd52 | 681 | void __rcu **slot, void *item, bool replace) |
175542f5 MW |
682 | { |
683 | if (*slot) | |
684 | return -EEXIST; | |
685 | rcu_assign_pointer(*slot, item); | |
686 | if (node) { | |
687 | node->count++; | |
3159f943 | 688 | if (xa_is_value(item)) |
01959dfe | 689 | node->nr_values++; |
175542f5 MW |
690 | } |
691 | return 1; | |
692 | } | |
139e5616 JW |
693 | |
694 | /** | |
e6145236 | 695 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
696 | * @root: radix tree root |
697 | * @index: index key | |
698 | * @item: item to insert | |
699 | * | |
700 | * Insert an item into the radix tree at position @index. | |
701 | */ | |
3a08cd52 MW |
702 | int radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
703 | void *item) | |
139e5616 JW |
704 | { |
705 | struct radix_tree_node *node; | |
d7b62727 | 706 | void __rcu **slot; |
139e5616 JW |
707 | int error; |
708 | ||
b194d16c | 709 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 710 | |
3a08cd52 | 711 | error = __radix_tree_create(root, index, &node, &slot); |
139e5616 JW |
712 | if (error) |
713 | return error; | |
175542f5 | 714 | |
3a08cd52 | 715 | error = insert_entries(node, slot, item, false); |
175542f5 MW |
716 | if (error < 0) |
717 | return error; | |
201b6264 | 718 | |
612d6c19 | 719 | if (node) { |
7b60e9ad | 720 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
721 | BUG_ON(tag_get(node, 0, offset)); |
722 | BUG_ON(tag_get(node, 1, offset)); | |
723 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 724 | } else { |
7b60e9ad | 725 | BUG_ON(root_tags_get(root)); |
612d6c19 | 726 | } |
1da177e4 | 727 | |
1da177e4 LT |
728 | return 0; |
729 | } | |
3a08cd52 | 730 | EXPORT_SYMBOL(radix_tree_insert); |
1da177e4 | 731 | |
139e5616 JW |
732 | /** |
733 | * __radix_tree_lookup - lookup an item in a radix tree | |
734 | * @root: radix tree root | |
735 | * @index: index key | |
736 | * @nodep: returns node | |
737 | * @slotp: returns slot | |
738 | * | |
739 | * Lookup and return the item at position @index in the radix | |
740 | * tree @root. | |
741 | * | |
742 | * Until there is more than one item in the tree, no nodes are | |
f8d5d0cc | 743 | * allocated and @root->xa_head is used as a direct slot instead of |
139e5616 | 744 | * pointing to a node, in which case *@nodep will be NULL. |
7cf9c2c7 | 745 | */ |
35534c86 MW |
746 | void *__radix_tree_lookup(const struct radix_tree_root *root, |
747 | unsigned long index, struct radix_tree_node **nodep, | |
d7b62727 | 748 | void __rcu ***slotp) |
1da177e4 | 749 | { |
139e5616 | 750 | struct radix_tree_node *node, *parent; |
85829954 | 751 | unsigned long maxindex; |
d7b62727 | 752 | void __rcu **slot; |
612d6c19 | 753 | |
85829954 MW |
754 | restart: |
755 | parent = NULL; | |
f8d5d0cc | 756 | slot = (void __rcu **)&root->xa_head; |
9e85d811 | 757 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 758 | if (index > maxindex) |
1da177e4 LT |
759 | return NULL; |
760 | ||
b194d16c | 761 | while (radix_tree_is_internal_node(node)) { |
85829954 | 762 | unsigned offset; |
1da177e4 | 763 | |
4dd6c098 | 764 | parent = entry_to_node(node); |
9e85d811 | 765 | offset = radix_tree_descend(parent, &node, index); |
85829954 | 766 | slot = parent->slots + offset; |
eff3860b MW |
767 | if (node == RADIX_TREE_RETRY) |
768 | goto restart; | |
66ee620f MW |
769 | if (parent->shift == 0) |
770 | break; | |
85829954 | 771 | } |
1da177e4 | 772 | |
139e5616 JW |
773 | if (nodep) |
774 | *nodep = parent; | |
775 | if (slotp) | |
776 | *slotp = slot; | |
777 | return node; | |
b72b71c6 HS |
778 | } |
779 | ||
780 | /** | |
781 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
782 | * @root: radix tree root | |
783 | * @index: index key | |
784 | * | |
785 | * Returns: the slot corresponding to the position @index in the | |
786 | * radix tree @root. This is useful for update-if-exists operations. | |
787 | * | |
788 | * This function can be called under rcu_read_lock iff the slot is not | |
789 | * modified by radix_tree_replace_slot, otherwise it must be called | |
790 | * exclusive from other writers. Any dereference of the slot must be done | |
791 | * using radix_tree_deref_slot. | |
792 | */ | |
d7b62727 | 793 | void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root, |
35534c86 | 794 | unsigned long index) |
b72b71c6 | 795 | { |
d7b62727 | 796 | void __rcu **slot; |
139e5616 JW |
797 | |
798 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
799 | return NULL; | |
800 | return slot; | |
a4331366 | 801 | } |
a4331366 HR |
802 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
803 | ||
804 | /** | |
805 | * radix_tree_lookup - perform lookup operation on a radix tree | |
806 | * @root: radix tree root | |
807 | * @index: index key | |
808 | * | |
809 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
810 | * |
811 | * This function can be called under rcu_read_lock, however the caller | |
812 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
813 | * them safely). No RCU barriers are required to access or modify the | |
814 | * returned item, however. | |
a4331366 | 815 | */ |
35534c86 | 816 | void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index) |
a4331366 | 817 | { |
139e5616 | 818 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
819 | } |
820 | EXPORT_SYMBOL(radix_tree_lookup); | |
821 | ||
d7b62727 | 822 | static void replace_slot(void __rcu **slot, void *item, |
01959dfe | 823 | struct radix_tree_node *node, int count, int values) |
f7942430 | 824 | { |
01959dfe | 825 | if (node && (count || values)) { |
f4b109c6 | 826 | node->count += count; |
01959dfe | 827 | node->nr_values += values; |
f4b109c6 | 828 | } |
f7942430 JW |
829 | |
830 | rcu_assign_pointer(*slot, item); | |
831 | } | |
832 | ||
0a835c4f MW |
833 | static bool node_tag_get(const struct radix_tree_root *root, |
834 | const struct radix_tree_node *node, | |
835 | unsigned int tag, unsigned int offset) | |
a90eb3a2 | 836 | { |
0a835c4f MW |
837 | if (node) |
838 | return tag_get(node, tag, offset); | |
839 | return root_tag_get(root, tag); | |
840 | } | |
a90eb3a2 | 841 | |
0a835c4f MW |
842 | /* |
843 | * IDR users want to be able to store NULL in the tree, so if the slot isn't | |
844 | * free, don't adjust the count, even if it's transitioning between NULL and | |
845 | * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still | |
846 | * have empty bits, but it only stores NULL in slots when they're being | |
847 | * deleted. | |
848 | */ | |
849 | static int calculate_count(struct radix_tree_root *root, | |
d7b62727 | 850 | struct radix_tree_node *node, void __rcu **slot, |
0a835c4f MW |
851 | void *item, void *old) |
852 | { | |
853 | if (is_idr(root)) { | |
854 | unsigned offset = get_slot_offset(node, slot); | |
855 | bool free = node_tag_get(root, node, IDR_FREE, offset); | |
856 | if (!free) | |
857 | return 0; | |
858 | if (!old) | |
859 | return 1; | |
a90eb3a2 | 860 | } |
0a835c4f | 861 | return !!item - !!old; |
a90eb3a2 MW |
862 | } |
863 | ||
6d75f366 JW |
864 | /** |
865 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
866 | * @root: radix tree root |
867 | * @node: pointer to tree node | |
868 | * @slot: pointer to slot in @node | |
869 | * @item: new item to store in the slot. | |
6d75f366 JW |
870 | * |
871 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
872 | * across slot lookup and replacement. | |
873 | */ | |
874 | void __radix_tree_replace(struct radix_tree_root *root, | |
875 | struct radix_tree_node *node, | |
1cf56f9d | 876 | void __rcu **slot, void *item) |
6d75f366 | 877 | { |
0a835c4f | 878 | void *old = rcu_dereference_raw(*slot); |
01959dfe | 879 | int values = !!xa_is_value(item) - !!xa_is_value(old); |
0a835c4f MW |
880 | int count = calculate_count(root, node, slot, item, old); |
881 | ||
6d75f366 | 882 | /* |
01959dfe | 883 | * This function supports replacing value entries and |
f4b109c6 | 884 | * deleting entries, but that needs accounting against the |
f8d5d0cc | 885 | * node unless the slot is root->xa_head. |
6d75f366 | 886 | */ |
f8d5d0cc | 887 | WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->xa_head) && |
01959dfe MW |
888 | (count || values)); |
889 | replace_slot(slot, item, node, count, values); | |
f4b109c6 | 890 | |
4d693d08 JW |
891 | if (!node) |
892 | return; | |
893 | ||
1cf56f9d | 894 | delete_node(root, node); |
6d75f366 JW |
895 | } |
896 | ||
897 | /** | |
898 | * radix_tree_replace_slot - replace item in a slot | |
899 | * @root: radix tree root | |
900 | * @slot: pointer to slot | |
901 | * @item: new item to store in the slot. | |
902 | * | |
7b8d046f | 903 | * For use with radix_tree_lookup_slot() and |
6d75f366 JW |
904 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked |
905 | * across slot lookup and replacement. | |
906 | * | |
907 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
01959dfe | 908 | * regular entries, and value entries, as that requires accounting |
f4b109c6 | 909 | * inside the radix tree node. When switching from one type of entry or |
e157b555 MW |
910 | * deleting, use __radix_tree_lookup() and __radix_tree_replace() or |
911 | * radix_tree_iter_replace(). | |
6d75f366 JW |
912 | */ |
913 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
d7b62727 | 914 | void __rcu **slot, void *item) |
6d75f366 | 915 | { |
1cf56f9d | 916 | __radix_tree_replace(root, NULL, slot, item); |
6d75f366 | 917 | } |
10257d71 | 918 | EXPORT_SYMBOL(radix_tree_replace_slot); |
6d75f366 | 919 | |
e157b555 MW |
920 | /** |
921 | * radix_tree_iter_replace - replace item in a slot | |
922 | * @root: radix tree root | |
923 | * @slot: pointer to slot | |
924 | * @item: new item to store in the slot. | |
925 | * | |
2956c664 MW |
926 | * For use with radix_tree_for_each_slot(). |
927 | * Caller must hold tree write locked. | |
e157b555 MW |
928 | */ |
929 | void radix_tree_iter_replace(struct radix_tree_root *root, | |
d7b62727 MW |
930 | const struct radix_tree_iter *iter, |
931 | void __rcu **slot, void *item) | |
e157b555 | 932 | { |
1cf56f9d | 933 | __radix_tree_replace(root, iter->node, slot, item); |
e157b555 MW |
934 | } |
935 | ||
30b888ba MW |
936 | static void node_tag_set(struct radix_tree_root *root, |
937 | struct radix_tree_node *node, | |
938 | unsigned int tag, unsigned int offset) | |
939 | { | |
940 | while (node) { | |
941 | if (tag_get(node, tag, offset)) | |
942 | return; | |
943 | tag_set(node, tag, offset); | |
944 | offset = node->offset; | |
945 | node = node->parent; | |
946 | } | |
947 | ||
948 | if (!root_tag_get(root, tag)) | |
949 | root_tag_set(root, tag); | |
950 | } | |
951 | ||
1da177e4 LT |
952 | /** |
953 | * radix_tree_tag_set - set a tag on a radix tree node | |
954 | * @root: radix tree root | |
955 | * @index: index key | |
2fcd9005 | 956 | * @tag: tag index |
1da177e4 | 957 | * |
daff89f3 JC |
958 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
959 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
960 | * the root all the way down to the leaf node. |
961 | * | |
2fcd9005 | 962 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
963 | * item is a bug. |
964 | */ | |
965 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 966 | unsigned long index, unsigned int tag) |
1da177e4 | 967 | { |
fb969909 RZ |
968 | struct radix_tree_node *node, *parent; |
969 | unsigned long maxindex; | |
1da177e4 | 970 | |
9e85d811 | 971 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 972 | BUG_ON(index > maxindex); |
1da177e4 | 973 | |
b194d16c | 974 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 975 | unsigned offset; |
1da177e4 | 976 | |
4dd6c098 | 977 | parent = entry_to_node(node); |
9e85d811 | 978 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
979 | BUG_ON(!node); |
980 | ||
981 | if (!tag_get(parent, tag, offset)) | |
982 | tag_set(parent, tag, offset); | |
1da177e4 LT |
983 | } |
984 | ||
612d6c19 | 985 | /* set the root's tag bit */ |
fb969909 | 986 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
987 | root_tag_set(root, tag); |
988 | ||
fb969909 | 989 | return node; |
1da177e4 LT |
990 | } |
991 | EXPORT_SYMBOL(radix_tree_tag_set); | |
992 | ||
d604c324 MW |
993 | static void node_tag_clear(struct radix_tree_root *root, |
994 | struct radix_tree_node *node, | |
995 | unsigned int tag, unsigned int offset) | |
996 | { | |
997 | while (node) { | |
998 | if (!tag_get(node, tag, offset)) | |
999 | return; | |
1000 | tag_clear(node, tag, offset); | |
1001 | if (any_tag_set(node, tag)) | |
1002 | return; | |
1003 | ||
1004 | offset = node->offset; | |
1005 | node = node->parent; | |
1006 | } | |
1007 | ||
1008 | /* clear the root's tag bit */ | |
1009 | if (root_tag_get(root, tag)) | |
1010 | root_tag_clear(root, tag); | |
1011 | } | |
1012 | ||
1da177e4 LT |
1013 | /** |
1014 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1015 | * @root: radix tree root | |
1016 | * @index: index key | |
2fcd9005 | 1017 | * @tag: tag index |
1da177e4 | 1018 | * |
daff89f3 | 1019 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1020 | * corresponding to @index in the radix tree. If this causes |
1021 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1022 | * next-to-leaf node, etc. |
1023 | * | |
1024 | * Returns the address of the tagged item on success, else NULL. ie: | |
1025 | * has the same return value and semantics as radix_tree_lookup(). | |
1026 | */ | |
1027 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1028 | unsigned long index, unsigned int tag) |
1da177e4 | 1029 | { |
00f47b58 RZ |
1030 | struct radix_tree_node *node, *parent; |
1031 | unsigned long maxindex; | |
e2bdb933 | 1032 | int uninitialized_var(offset); |
1da177e4 | 1033 | |
9e85d811 | 1034 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1035 | if (index > maxindex) |
1036 | return NULL; | |
1da177e4 | 1037 | |
00f47b58 | 1038 | parent = NULL; |
1da177e4 | 1039 | |
b194d16c | 1040 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1041 | parent = entry_to_node(node); |
9e85d811 | 1042 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1043 | } |
1044 | ||
d604c324 MW |
1045 | if (node) |
1046 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1047 | |
00f47b58 | 1048 | return node; |
1da177e4 LT |
1049 | } |
1050 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1051 | ||
30b888ba MW |
1052 | /** |
1053 | * radix_tree_iter_tag_clear - clear a tag on the current iterator entry | |
1054 | * @root: radix tree root | |
1055 | * @iter: iterator state | |
1056 | * @tag: tag to clear | |
1057 | */ | |
1058 | void radix_tree_iter_tag_clear(struct radix_tree_root *root, | |
1059 | const struct radix_tree_iter *iter, unsigned int tag) | |
1060 | { | |
1061 | node_tag_clear(root, iter->node, tag, iter_offset(iter)); | |
1062 | } | |
1063 | ||
1da177e4 | 1064 | /** |
32605a18 MT |
1065 | * radix_tree_tag_get - get a tag on a radix tree node |
1066 | * @root: radix tree root | |
1067 | * @index: index key | |
2fcd9005 | 1068 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1069 | * |
32605a18 | 1070 | * Return values: |
1da177e4 | 1071 | * |
612d6c19 NP |
1072 | * 0: tag not present or not set |
1073 | * 1: tag set | |
ce82653d DH |
1074 | * |
1075 | * Note that the return value of this function may not be relied on, even if | |
1076 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1077 | * from concurrency. | |
1da177e4 | 1078 | */ |
35534c86 | 1079 | int radix_tree_tag_get(const struct radix_tree_root *root, |
daff89f3 | 1080 | unsigned long index, unsigned int tag) |
1da177e4 | 1081 | { |
4589ba6d RZ |
1082 | struct radix_tree_node *node, *parent; |
1083 | unsigned long maxindex; | |
1da177e4 | 1084 | |
612d6c19 NP |
1085 | if (!root_tag_get(root, tag)) |
1086 | return 0; | |
1087 | ||
9e85d811 | 1088 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1089 | if (index > maxindex) |
1090 | return 0; | |
7cf9c2c7 | 1091 | |
b194d16c | 1092 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1093 | unsigned offset; |
1da177e4 | 1094 | |
4dd6c098 | 1095 | parent = entry_to_node(node); |
9e85d811 | 1096 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1097 | |
4589ba6d | 1098 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1099 | return 0; |
4589ba6d RZ |
1100 | if (node == RADIX_TREE_RETRY) |
1101 | break; | |
1da177e4 | 1102 | } |
4589ba6d RZ |
1103 | |
1104 | return 1; | |
1da177e4 LT |
1105 | } |
1106 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1107 | |
148deab2 MW |
1108 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1109 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1110 | struct radix_tree_node *node, unsigned offset, | |
1111 | unsigned tag) | |
1112 | { | |
1113 | unsigned tag_long = offset / BITS_PER_LONG; | |
1114 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1115 | ||
0a835c4f MW |
1116 | if (!node) { |
1117 | iter->tags = 1; | |
1118 | return; | |
1119 | } | |
1120 | ||
148deab2 MW |
1121 | iter->tags = node->tags[tag][tag_long] >> tag_bit; |
1122 | ||
1123 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1124 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1125 | /* Pick tags from next element */ | |
1126 | if (tag_bit) | |
1127 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1128 | (BITS_PER_LONG - tag_bit); | |
1129 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1130 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1131 | } | |
1132 | } | |
1133 | ||
d7b62727 MW |
1134 | void __rcu **radix_tree_iter_resume(void __rcu **slot, |
1135 | struct radix_tree_iter *iter) | |
148deab2 | 1136 | { |
148deab2 MW |
1137 | slot++; |
1138 | iter->index = __radix_tree_iter_add(iter, 1); | |
148deab2 MW |
1139 | iter->next_index = iter->index; |
1140 | iter->tags = 0; | |
1141 | return NULL; | |
1142 | } | |
1143 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1144 | ||
78c1d784 KK |
1145 | /** |
1146 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1147 | * | |
1148 | * @root: radix tree root | |
1149 | * @iter: iterator state | |
1150 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1151 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1152 | */ | |
d7b62727 | 1153 | void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root, |
78c1d784 KK |
1154 | struct radix_tree_iter *iter, unsigned flags) |
1155 | { | |
9e85d811 | 1156 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1157 | struct radix_tree_node *node, *child; |
21ef5339 | 1158 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1159 | |
1160 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1161 | return NULL; | |
1162 | ||
1163 | /* | |
1164 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1165 | * during iterating; it can be zero only at the beginning. | |
1166 | * And we cannot overflow iter->next_index in a single step, | |
1167 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1168 | * |
1169 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1170 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1171 | */ |
1172 | index = iter->next_index; | |
1173 | if (!index && iter->index) | |
1174 | return NULL; | |
1175 | ||
21ef5339 | 1176 | restart: |
9e85d811 | 1177 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1178 | if (index > maxindex) |
1179 | return NULL; | |
8c1244de MW |
1180 | if (!child) |
1181 | return NULL; | |
21ef5339 | 1182 | |
8c1244de | 1183 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1184 | /* Single-slot tree */ |
21ef5339 RZ |
1185 | iter->index = index; |
1186 | iter->next_index = maxindex + 1; | |
78c1d784 | 1187 | iter->tags = 1; |
268f42de | 1188 | iter->node = NULL; |
f8d5d0cc | 1189 | return (void __rcu **)&root->xa_head; |
8c1244de | 1190 | } |
21ef5339 | 1191 | |
8c1244de MW |
1192 | do { |
1193 | node = entry_to_node(child); | |
9e85d811 | 1194 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1195 | |
78c1d784 | 1196 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1197 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1198 | /* Hole detected */ |
1199 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1200 | return NULL; | |
1201 | ||
1202 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1203 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1204 | offset + 1); |
1205 | else | |
1206 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
12320d0f MW |
1207 | void *slot = rcu_dereference_raw( |
1208 | node->slots[offset]); | |
21ef5339 | 1209 | if (slot) |
78c1d784 KK |
1210 | break; |
1211 | } | |
8c1244de | 1212 | index &= ~node_maxindex(node); |
9e85d811 | 1213 | index += offset << node->shift; |
78c1d784 KK |
1214 | /* Overflow after ~0UL */ |
1215 | if (!index) | |
1216 | return NULL; | |
1217 | if (offset == RADIX_TREE_MAP_SIZE) | |
1218 | goto restart; | |
8c1244de | 1219 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1220 | } |
1221 | ||
e157b555 | 1222 | if (!child) |
78c1d784 | 1223 | goto restart; |
e157b555 MW |
1224 | if (child == RADIX_TREE_RETRY) |
1225 | break; | |
66ee620f | 1226 | } while (node->shift && radix_tree_is_internal_node(child)); |
78c1d784 KK |
1227 | |
1228 | /* Update the iterator state */ | |
3a08cd52 | 1229 | iter->index = (index &~ node_maxindex(node)) | offset; |
8c1244de | 1230 | iter->next_index = (index | node_maxindex(node)) + 1; |
268f42de | 1231 | iter->node = node; |
78c1d784 | 1232 | |
148deab2 MW |
1233 | if (flags & RADIX_TREE_ITER_TAGGED) |
1234 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1235 | |
1236 | return node->slots + offset; | |
1237 | } | |
1238 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1239 | ||
1da177e4 LT |
1240 | /** |
1241 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1242 | * @root: radix tree root | |
1243 | * @results: where the results of the lookup are placed | |
1244 | * @first_index: start the lookup from this key | |
1245 | * @max_items: place up to this many items at *results | |
1246 | * | |
1247 | * Performs an index-ascending scan of the tree for present items. Places | |
1248 | * them at *@results and returns the number of items which were placed at | |
1249 | * *@results. | |
1250 | * | |
1251 | * The implementation is naive. | |
7cf9c2c7 NP |
1252 | * |
1253 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1254 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1255 | * an atomic snapshot of the tree at a single point in time, the |
1256 | * semantics of an RCU protected gang lookup are as though multiple | |
1257 | * radix_tree_lookups have been issued in individual locks, and results | |
1258 | * stored in 'results'. | |
1da177e4 LT |
1259 | */ |
1260 | unsigned int | |
35534c86 | 1261 | radix_tree_gang_lookup(const struct radix_tree_root *root, void **results, |
1da177e4 LT |
1262 | unsigned long first_index, unsigned int max_items) |
1263 | { | |
cebbd29e | 1264 | struct radix_tree_iter iter; |
d7b62727 | 1265 | void __rcu **slot; |
cebbd29e | 1266 | unsigned int ret = 0; |
7cf9c2c7 | 1267 | |
cebbd29e | 1268 | if (unlikely(!max_items)) |
7cf9c2c7 | 1269 | return 0; |
1da177e4 | 1270 | |
cebbd29e | 1271 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1272 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1273 | if (!results[ret]) |
1274 | continue; | |
b194d16c | 1275 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1276 | slot = radix_tree_iter_retry(&iter); |
1277 | continue; | |
1278 | } | |
cebbd29e | 1279 | if (++ret == max_items) |
1da177e4 | 1280 | break; |
1da177e4 | 1281 | } |
7cf9c2c7 | 1282 | |
1da177e4 LT |
1283 | return ret; |
1284 | } | |
1285 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1286 | ||
1da177e4 LT |
1287 | /** |
1288 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1289 | * based on a tag | |
1290 | * @root: radix tree root | |
1291 | * @results: where the results of the lookup are placed | |
1292 | * @first_index: start the lookup from this key | |
1293 | * @max_items: place up to this many items at *results | |
daff89f3 | 1294 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1295 | * |
1296 | * Performs an index-ascending scan of the tree for present items which | |
1297 | * have the tag indexed by @tag set. Places the items at *@results and | |
1298 | * returns the number of items which were placed at *@results. | |
1299 | */ | |
1300 | unsigned int | |
35534c86 | 1301 | radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results, |
daff89f3 JC |
1302 | unsigned long first_index, unsigned int max_items, |
1303 | unsigned int tag) | |
1da177e4 | 1304 | { |
cebbd29e | 1305 | struct radix_tree_iter iter; |
d7b62727 | 1306 | void __rcu **slot; |
cebbd29e | 1307 | unsigned int ret = 0; |
612d6c19 | 1308 | |
cebbd29e | 1309 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1310 | return 0; |
1311 | ||
cebbd29e | 1312 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1313 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1314 | if (!results[ret]) |
1315 | continue; | |
b194d16c | 1316 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1317 | slot = radix_tree_iter_retry(&iter); |
1318 | continue; | |
1319 | } | |
cebbd29e | 1320 | if (++ret == max_items) |
1da177e4 | 1321 | break; |
1da177e4 | 1322 | } |
7cf9c2c7 | 1323 | |
1da177e4 LT |
1324 | return ret; |
1325 | } | |
1326 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1327 | ||
47feff2c NP |
1328 | /** |
1329 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1330 | * radix tree based on a tag | |
1331 | * @root: radix tree root | |
1332 | * @results: where the results of the lookup are placed | |
1333 | * @first_index: start the lookup from this key | |
1334 | * @max_items: place up to this many items at *results | |
1335 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1336 | * | |
1337 | * Performs an index-ascending scan of the tree for present items which | |
1338 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1339 | * returns the number of slots which were placed at *@results. | |
1340 | */ | |
1341 | unsigned int | |
35534c86 | 1342 | radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root, |
d7b62727 | 1343 | void __rcu ***results, unsigned long first_index, |
35534c86 | 1344 | unsigned int max_items, unsigned int tag) |
47feff2c | 1345 | { |
cebbd29e | 1346 | struct radix_tree_iter iter; |
d7b62727 | 1347 | void __rcu **slot; |
cebbd29e | 1348 | unsigned int ret = 0; |
47feff2c | 1349 | |
cebbd29e | 1350 | if (unlikely(!max_items)) |
47feff2c NP |
1351 | return 0; |
1352 | ||
cebbd29e KK |
1353 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1354 | results[ret] = slot; | |
1355 | if (++ret == max_items) | |
47feff2c | 1356 | break; |
47feff2c NP |
1357 | } |
1358 | ||
1359 | return ret; | |
1360 | } | |
1361 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1362 | ||
0ac398ef | 1363 | static bool __radix_tree_delete(struct radix_tree_root *root, |
d7b62727 | 1364 | struct radix_tree_node *node, void __rcu **slot) |
0ac398ef | 1365 | { |
0a835c4f | 1366 | void *old = rcu_dereference_raw(*slot); |
01959dfe | 1367 | int values = xa_is_value(old) ? -1 : 0; |
0ac398ef MW |
1368 | unsigned offset = get_slot_offset(node, slot); |
1369 | int tag; | |
1370 | ||
0a835c4f MW |
1371 | if (is_idr(root)) |
1372 | node_tag_set(root, node, IDR_FREE, offset); | |
1373 | else | |
1374 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1375 | node_tag_clear(root, node, tag, offset); | |
0ac398ef | 1376 | |
01959dfe | 1377 | replace_slot(slot, NULL, node, -1, values); |
1cf56f9d | 1378 | return node && delete_node(root, node); |
0ac398ef MW |
1379 | } |
1380 | ||
1da177e4 | 1381 | /** |
0ac398ef MW |
1382 | * radix_tree_iter_delete - delete the entry at this iterator position |
1383 | * @root: radix tree root | |
1384 | * @iter: iterator state | |
1385 | * @slot: pointer to slot | |
1da177e4 | 1386 | * |
0ac398ef MW |
1387 | * Delete the entry at the position currently pointed to by the iterator. |
1388 | * This may result in the current node being freed; if it is, the iterator | |
1389 | * is advanced so that it will not reference the freed memory. This | |
1390 | * function may be called without any locking if there are no other threads | |
1391 | * which can access this tree. | |
1392 | */ | |
1393 | void radix_tree_iter_delete(struct radix_tree_root *root, | |
d7b62727 | 1394 | struct radix_tree_iter *iter, void __rcu **slot) |
0ac398ef MW |
1395 | { |
1396 | if (__radix_tree_delete(root, iter->node, slot)) | |
1397 | iter->index = iter->next_index; | |
1398 | } | |
d1b48c1e | 1399 | EXPORT_SYMBOL(radix_tree_iter_delete); |
0ac398ef MW |
1400 | |
1401 | /** | |
1402 | * radix_tree_delete_item - delete an item from a radix tree | |
1403 | * @root: radix tree root | |
1404 | * @index: index key | |
1405 | * @item: expected item | |
1da177e4 | 1406 | * |
0ac398ef | 1407 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1408 | * |
0ac398ef MW |
1409 | * Return: the deleted entry, or %NULL if it was not present |
1410 | * or the entry at the given @index was not @item. | |
1da177e4 | 1411 | */ |
53c59f26 JW |
1412 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1413 | unsigned long index, void *item) | |
1da177e4 | 1414 | { |
0a835c4f | 1415 | struct radix_tree_node *node = NULL; |
7a4deea1 | 1416 | void __rcu **slot = NULL; |
139e5616 | 1417 | void *entry; |
1da177e4 | 1418 | |
139e5616 | 1419 | entry = __radix_tree_lookup(root, index, &node, &slot); |
7a4deea1 MW |
1420 | if (!slot) |
1421 | return NULL; | |
0a835c4f MW |
1422 | if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE, |
1423 | get_slot_offset(node, slot)))) | |
139e5616 | 1424 | return NULL; |
1da177e4 | 1425 | |
139e5616 JW |
1426 | if (item && entry != item) |
1427 | return NULL; | |
1428 | ||
0ac398ef | 1429 | __radix_tree_delete(root, node, slot); |
612d6c19 | 1430 | |
139e5616 | 1431 | return entry; |
1da177e4 | 1432 | } |
53c59f26 JW |
1433 | EXPORT_SYMBOL(radix_tree_delete_item); |
1434 | ||
1435 | /** | |
0ac398ef MW |
1436 | * radix_tree_delete - delete an entry from a radix tree |
1437 | * @root: radix tree root | |
1438 | * @index: index key | |
53c59f26 | 1439 | * |
0ac398ef | 1440 | * Remove the entry at @index from the radix tree rooted at @root. |
53c59f26 | 1441 | * |
0ac398ef | 1442 | * Return: The deleted entry, or %NULL if it was not present. |
53c59f26 JW |
1443 | */ |
1444 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1445 | { | |
1446 | return radix_tree_delete_item(root, index, NULL); | |
1447 | } | |
1da177e4 LT |
1448 | EXPORT_SYMBOL(radix_tree_delete); |
1449 | ||
1450 | /** | |
1451 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1452 | * @root: radix tree root | |
1453 | * @tag: tag to test | |
1454 | */ | |
35534c86 | 1455 | int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1456 | { |
612d6c19 | 1457 | return root_tag_get(root, tag); |
1da177e4 LT |
1458 | } |
1459 | EXPORT_SYMBOL(radix_tree_tagged); | |
1460 | ||
0a835c4f MW |
1461 | /** |
1462 | * idr_preload - preload for idr_alloc() | |
1463 | * @gfp_mask: allocation mask to use for preloading | |
1464 | * | |
1465 | * Preallocate memory to use for the next call to idr_alloc(). This function | |
1466 | * returns with preemption disabled. It will be enabled by idr_preload_end(). | |
1467 | */ | |
1468 | void idr_preload(gfp_t gfp_mask) | |
1469 | { | |
bc9ae224 | 1470 | if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE)) |
cfa6705d | 1471 | local_lock(&radix_tree_preloads.lock); |
0a835c4f MW |
1472 | } |
1473 | EXPORT_SYMBOL(idr_preload); | |
1474 | ||
460488c5 | 1475 | void __rcu **idr_get_free(struct radix_tree_root *root, |
388f79fd CM |
1476 | struct radix_tree_iter *iter, gfp_t gfp, |
1477 | unsigned long max) | |
0a835c4f MW |
1478 | { |
1479 | struct radix_tree_node *node = NULL, *child; | |
f8d5d0cc | 1480 | void __rcu **slot = (void __rcu **)&root->xa_head; |
0a835c4f | 1481 | unsigned long maxindex, start = iter->next_index; |
0a835c4f MW |
1482 | unsigned int shift, offset = 0; |
1483 | ||
1484 | grow: | |
1485 | shift = radix_tree_load_root(root, &child, &maxindex); | |
1486 | if (!radix_tree_tagged(root, IDR_FREE)) | |
1487 | start = max(start, maxindex + 1); | |
1488 | if (start > max) | |
1489 | return ERR_PTR(-ENOSPC); | |
1490 | ||
1491 | if (start > maxindex) { | |
1492 | int error = radix_tree_extend(root, gfp, start, shift); | |
1493 | if (error < 0) | |
1494 | return ERR_PTR(error); | |
1495 | shift = error; | |
f8d5d0cc | 1496 | child = rcu_dereference_raw(root->xa_head); |
0a835c4f | 1497 | } |
66ee620f MW |
1498 | if (start == 0 && shift == 0) |
1499 | shift = RADIX_TREE_MAP_SHIFT; | |
0a835c4f MW |
1500 | |
1501 | while (shift) { | |
1502 | shift -= RADIX_TREE_MAP_SHIFT; | |
1503 | if (child == NULL) { | |
1504 | /* Have to add a child node. */ | |
d58275bc MW |
1505 | child = radix_tree_node_alloc(gfp, node, root, shift, |
1506 | offset, 0, 0); | |
0a835c4f MW |
1507 | if (!child) |
1508 | return ERR_PTR(-ENOMEM); | |
1509 | all_tag_set(child, IDR_FREE); | |
1510 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
1511 | if (node) | |
1512 | node->count++; | |
1513 | } else if (!radix_tree_is_internal_node(child)) | |
1514 | break; | |
1515 | ||
1516 | node = entry_to_node(child); | |
1517 | offset = radix_tree_descend(node, &child, start); | |
1518 | if (!tag_get(node, IDR_FREE, offset)) { | |
1519 | offset = radix_tree_find_next_bit(node, IDR_FREE, | |
1520 | offset + 1); | |
1521 | start = next_index(start, node, offset); | |
b7e9728f | 1522 | if (start > max || start == 0) |
0a835c4f MW |
1523 | return ERR_PTR(-ENOSPC); |
1524 | while (offset == RADIX_TREE_MAP_SIZE) { | |
1525 | offset = node->offset + 1; | |
1526 | node = node->parent; | |
1527 | if (!node) | |
1528 | goto grow; | |
1529 | shift = node->shift; | |
1530 | } | |
1531 | child = rcu_dereference_raw(node->slots[offset]); | |
1532 | } | |
1533 | slot = &node->slots[offset]; | |
1534 | } | |
1535 | ||
1536 | iter->index = start; | |
1537 | if (node) | |
1538 | iter->next_index = 1 + min(max, (start | node_maxindex(node))); | |
1539 | else | |
1540 | iter->next_index = 1; | |
1541 | iter->node = node; | |
0a835c4f MW |
1542 | set_iter_tags(iter, node, offset, IDR_FREE); |
1543 | ||
1544 | return slot; | |
1545 | } | |
1546 | ||
1547 | /** | |
1548 | * idr_destroy - release all internal memory from an IDR | |
1549 | * @idr: idr handle | |
1550 | * | |
1551 | * After this function is called, the IDR is empty, and may be reused or | |
1552 | * the data structure containing it may be freed. | |
1553 | * | |
1554 | * A typical clean-up sequence for objects stored in an idr tree will use | |
1555 | * idr_for_each() to free all objects, if necessary, then idr_destroy() to | |
1556 | * free the memory used to keep track of those objects. | |
1557 | */ | |
1558 | void idr_destroy(struct idr *idr) | |
1559 | { | |
f8d5d0cc | 1560 | struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.xa_head); |
0a835c4f MW |
1561 | if (radix_tree_is_internal_node(node)) |
1562 | radix_tree_free_nodes(node); | |
f8d5d0cc | 1563 | idr->idr_rt.xa_head = NULL; |
0a835c4f MW |
1564 | root_tag_set(&idr->idr_rt, IDR_FREE); |
1565 | } | |
1566 | EXPORT_SYMBOL(idr_destroy); | |
1567 | ||
1da177e4 | 1568 | static void |
449dd698 | 1569 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1570 | { |
449dd698 JW |
1571 | struct radix_tree_node *node = arg; |
1572 | ||
1573 | memset(node, 0, sizeof(*node)); | |
1574 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1575 | } |
1576 | ||
d544abd5 | 1577 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1578 | { |
2fcd9005 MW |
1579 | struct radix_tree_preload *rtp; |
1580 | struct radix_tree_node *node; | |
1581 | ||
1582 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1583 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1584 | while (rtp->nr) { | |
1585 | node = rtp->nodes; | |
1293d5c5 | 1586 | rtp->nodes = node->parent; |
d544abd5 SAS |
1587 | kmem_cache_free(radix_tree_node_cachep, node); |
1588 | rtp->nr--; | |
2fcd9005 | 1589 | } |
d544abd5 | 1590 | return 0; |
1da177e4 | 1591 | } |
1da177e4 LT |
1592 | |
1593 | void __init radix_tree_init(void) | |
1594 | { | |
d544abd5 | 1595 | int ret; |
7e784422 MH |
1596 | |
1597 | BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32); | |
fa290cda | 1598 | BUILD_BUG_ON(ROOT_IS_IDR & ~GFP_ZONEMASK); |
02c02bf1 | 1599 | BUILD_BUG_ON(XA_CHUNK_SIZE > 255); |
1da177e4 LT |
1600 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
1601 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1602 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1603 | radix_tree_node_ctor); | |
d544abd5 SAS |
1604 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
1605 | NULL, radix_tree_cpu_dead); | |
1606 | WARN_ON(ret < 0); | |
1da177e4 | 1607 | } |