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 | ||
0a835c4f MW |
25 | #include <linux/bitmap.h> |
26 | #include <linux/bitops.h> | |
460488c5 | 27 | #include <linux/bug.h> |
e157b555 | 28 | #include <linux/cpu.h> |
1da177e4 | 29 | #include <linux/errno.h> |
0a835c4f MW |
30 | #include <linux/export.h> |
31 | #include <linux/idr.h> | |
1da177e4 LT |
32 | #include <linux/init.h> |
33 | #include <linux/kernel.h> | |
0a835c4f | 34 | #include <linux/kmemleak.h> |
1da177e4 | 35 | #include <linux/percpu.h> |
0a835c4f MW |
36 | #include <linux/preempt.h> /* in_interrupt() */ |
37 | #include <linux/radix-tree.h> | |
38 | #include <linux/rcupdate.h> | |
1da177e4 | 39 | #include <linux/slab.h> |
1da177e4 | 40 | #include <linux/string.h> |
1da177e4 LT |
41 | |
42 | ||
c78c66d1 KS |
43 | /* Number of nodes in fully populated tree of given height */ |
44 | static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
45 | ||
1da177e4 LT |
46 | /* |
47 | * Radix tree node cache. | |
48 | */ | |
e18b890b | 49 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 50 | |
55368052 NP |
51 | /* |
52 | * The radix tree is variable-height, so an insert operation not only has | |
53 | * to build the branch to its corresponding item, it also has to build the | |
54 | * branch to existing items if the size has to be increased (by | |
55 | * radix_tree_extend). | |
56 | * | |
57 | * The worst case is a zero height tree with just a single item at index 0, | |
58 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
59 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
60 | * Hence: | |
61 | */ | |
62 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
63 | ||
0a835c4f MW |
64 | /* |
65 | * The IDR does not have to be as high as the radix tree since it uses | |
66 | * signed integers, not unsigned longs. | |
67 | */ | |
68 | #define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1) | |
69 | #define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \ | |
70 | RADIX_TREE_MAP_SHIFT)) | |
71 | #define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1) | |
72 | ||
7ad3d4d8 MW |
73 | /* |
74 | * The IDA is even shorter since it uses a bitmap at the last level. | |
75 | */ | |
76 | #define IDA_INDEX_BITS (8 * sizeof(int) - 1 - ilog2(IDA_BITMAP_BITS)) | |
77 | #define IDA_MAX_PATH (DIV_ROUND_UP(IDA_INDEX_BITS, \ | |
78 | RADIX_TREE_MAP_SHIFT)) | |
79 | #define IDA_PRELOAD_SIZE (IDA_MAX_PATH * 2 - 1) | |
80 | ||
1da177e4 LT |
81 | /* |
82 | * Per-cpu pool of preloaded nodes | |
83 | */ | |
84 | struct radix_tree_preload { | |
2fcd9005 | 85 | unsigned nr; |
1293d5c5 | 86 | /* nodes->parent points to next preallocated node */ |
9d2a8da0 | 87 | struct radix_tree_node *nodes; |
1da177e4 | 88 | }; |
8cef7d57 | 89 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 90 | |
148deab2 MW |
91 | static inline struct radix_tree_node *entry_to_node(void *ptr) |
92 | { | |
93 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE); | |
94 | } | |
95 | ||
a4db4dce | 96 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 97 | { |
30ff46cc | 98 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
99 | } |
100 | ||
a4db4dce | 101 | #define RADIX_TREE_RETRY node_to_entry(NULL) |
afe0e395 | 102 | |
db050f29 MW |
103 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
104 | /* Sibling slots point directly to another slot in the same node */ | |
35534c86 MW |
105 | static inline |
106 | bool is_sibling_entry(const struct radix_tree_node *parent, void *node) | |
db050f29 | 107 | { |
d7b62727 | 108 | void __rcu **ptr = node; |
db050f29 MW |
109 | return (parent->slots <= ptr) && |
110 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
111 | } | |
112 | #else | |
35534c86 MW |
113 | static inline |
114 | bool is_sibling_entry(const struct radix_tree_node *parent, void *node) | |
db050f29 MW |
115 | { |
116 | return false; | |
117 | } | |
118 | #endif | |
119 | ||
d7b62727 MW |
120 | static inline unsigned long |
121 | get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot) | |
db050f29 | 122 | { |
76f070b4 | 123 | return parent ? slot - parent->slots : 0; |
db050f29 MW |
124 | } |
125 | ||
35534c86 | 126 | static unsigned int radix_tree_descend(const struct radix_tree_node *parent, |
9e85d811 | 127 | struct radix_tree_node **nodep, unsigned long index) |
db050f29 | 128 | { |
9e85d811 | 129 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
d7b62727 | 130 | void __rcu **entry = rcu_dereference_raw(parent->slots[offset]); |
db050f29 MW |
131 | |
132 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 133 | if (radix_tree_is_internal_node(entry)) { |
8d2c0d36 | 134 | if (is_sibling_entry(parent, entry)) { |
d7b62727 MW |
135 | void __rcu **sibentry; |
136 | sibentry = (void __rcu **) entry_to_node(entry); | |
8d2c0d36 LT |
137 | offset = get_slot_offset(parent, sibentry); |
138 | entry = rcu_dereference_raw(*sibentry); | |
db050f29 MW |
139 | } |
140 | } | |
141 | #endif | |
142 | ||
143 | *nodep = (void *)entry; | |
144 | return offset; | |
145 | } | |
146 | ||
35534c86 | 147 | static inline gfp_t root_gfp_mask(const struct radix_tree_root *root) |
612d6c19 | 148 | { |
fa290cda | 149 | return root->gfp_mask & (__GFP_BITS_MASK & ~GFP_ZONEMASK); |
612d6c19 NP |
150 | } |
151 | ||
643b52b9 NP |
152 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
153 | int offset) | |
154 | { | |
155 | __set_bit(offset, node->tags[tag]); | |
156 | } | |
157 | ||
158 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
159 | int offset) | |
160 | { | |
161 | __clear_bit(offset, node->tags[tag]); | |
162 | } | |
163 | ||
35534c86 | 164 | static inline int tag_get(const struct radix_tree_node *node, unsigned int tag, |
643b52b9 NP |
165 | int offset) |
166 | { | |
167 | return test_bit(offset, node->tags[tag]); | |
168 | } | |
169 | ||
35534c86 | 170 | static inline void root_tag_set(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 171 | { |
0a835c4f | 172 | root->gfp_mask |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
173 | } |
174 | ||
2fcd9005 | 175 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 176 | { |
0a835c4f | 177 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
178 | } |
179 | ||
180 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
181 | { | |
0a835c4f | 182 | root->gfp_mask &= (1 << ROOT_TAG_SHIFT) - 1; |
643b52b9 NP |
183 | } |
184 | ||
35534c86 | 185 | static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag) |
643b52b9 | 186 | { |
0a835c4f | 187 | return (__force int)root->gfp_mask & (1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
188 | } |
189 | ||
35534c86 | 190 | static inline unsigned root_tags_get(const struct radix_tree_root *root) |
643b52b9 | 191 | { |
0a835c4f | 192 | return (__force unsigned)root->gfp_mask >> ROOT_TAG_SHIFT; |
643b52b9 NP |
193 | } |
194 | ||
0a835c4f | 195 | static inline bool is_idr(const struct radix_tree_root *root) |
7b60e9ad | 196 | { |
0a835c4f | 197 | return !!(root->gfp_mask & ROOT_IS_IDR); |
7b60e9ad MW |
198 | } |
199 | ||
643b52b9 NP |
200 | /* |
201 | * Returns 1 if any slot in the node has this tag set. | |
202 | * Otherwise returns 0. | |
203 | */ | |
35534c86 MW |
204 | static inline int any_tag_set(const struct radix_tree_node *node, |
205 | unsigned int tag) | |
643b52b9 | 206 | { |
2fcd9005 | 207 | unsigned idx; |
643b52b9 NP |
208 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
209 | if (node->tags[tag][idx]) | |
210 | return 1; | |
211 | } | |
212 | return 0; | |
213 | } | |
78c1d784 | 214 | |
0a835c4f MW |
215 | static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag) |
216 | { | |
217 | bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE); | |
218 | } | |
219 | ||
78c1d784 KK |
220 | /** |
221 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
222 | * | |
223 | * @addr: The address to base the search on | |
224 | * @size: The bitmap size in bits | |
225 | * @offset: The bitnumber to start searching at | |
226 | * | |
227 | * Unrollable variant of find_next_bit() for constant size arrays. | |
228 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
229 | * Returns next bit offset, or size if nothing found. | |
230 | */ | |
231 | static __always_inline unsigned long | |
bc412fca MW |
232 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
233 | unsigned long offset) | |
78c1d784 | 234 | { |
bc412fca | 235 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 236 | |
bc412fca | 237 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
238 | unsigned long tmp; |
239 | ||
240 | addr += offset / BITS_PER_LONG; | |
241 | tmp = *addr >> (offset % BITS_PER_LONG); | |
242 | if (tmp) | |
243 | return __ffs(tmp) + offset; | |
244 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 245 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
246 | tmp = *++addr; |
247 | if (tmp) | |
248 | return __ffs(tmp) + offset; | |
249 | offset += BITS_PER_LONG; | |
250 | } | |
251 | } | |
bc412fca | 252 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
253 | } |
254 | ||
268f42de MW |
255 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
256 | { | |
257 | return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK; | |
258 | } | |
259 | ||
218ed750 MW |
260 | /* |
261 | * The maximum index which can be stored in a radix tree | |
262 | */ | |
263 | static inline unsigned long shift_maxindex(unsigned int shift) | |
264 | { | |
265 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
266 | } | |
267 | ||
35534c86 | 268 | static inline unsigned long node_maxindex(const struct radix_tree_node *node) |
218ed750 MW |
269 | { |
270 | return shift_maxindex(node->shift); | |
271 | } | |
272 | ||
0a835c4f MW |
273 | static unsigned long next_index(unsigned long index, |
274 | const struct radix_tree_node *node, | |
275 | unsigned long offset) | |
276 | { | |
277 | return (index & ~node_maxindex(node)) + (offset << node->shift); | |
278 | } | |
279 | ||
0796c583 | 280 | #ifndef __KERNEL__ |
d0891265 | 281 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 282 | { |
0796c583 | 283 | unsigned long i; |
7cf19af4 | 284 | |
218ed750 MW |
285 | pr_debug("radix node: %p offset %d indices %lu-%lu parent %p tags %lx %lx %lx shift %d count %d exceptional %d\n", |
286 | node, node->offset, index, index | node_maxindex(node), | |
287 | node->parent, | |
0796c583 | 288 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
218ed750 | 289 | node->shift, node->count, node->exceptional); |
0796c583 RZ |
290 | |
291 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
292 | unsigned long first = index | (i << node->shift); |
293 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
294 | void *entry = node->slots[i]; |
295 | if (!entry) | |
296 | continue; | |
218ed750 MW |
297 | if (entry == RADIX_TREE_RETRY) { |
298 | pr_debug("radix retry offset %ld indices %lu-%lu parent %p\n", | |
299 | i, first, last, node); | |
b194d16c | 300 | } else if (!radix_tree_is_internal_node(entry)) { |
218ed750 MW |
301 | pr_debug("radix entry %p offset %ld indices %lu-%lu parent %p\n", |
302 | entry, i, first, last, node); | |
303 | } else if (is_sibling_entry(node, entry)) { | |
304 | pr_debug("radix sblng %p offset %ld indices %lu-%lu parent %p val %p\n", | |
305 | entry, i, first, last, node, | |
306 | *(void **)entry_to_node(entry)); | |
0796c583 | 307 | } else { |
4dd6c098 | 308 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
309 | } |
310 | } | |
7cf19af4 MW |
311 | } |
312 | ||
313 | /* For debug */ | |
314 | static void radix_tree_dump(struct radix_tree_root *root) | |
315 | { | |
d0891265 MW |
316 | pr_debug("radix root: %p rnode %p tags %x\n", |
317 | root, root->rnode, | |
0a835c4f | 318 | root->gfp_mask >> ROOT_TAG_SHIFT); |
b194d16c | 319 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 320 | return; |
4dd6c098 | 321 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 | 322 | } |
0a835c4f MW |
323 | |
324 | static void dump_ida_node(void *entry, unsigned long index) | |
325 | { | |
326 | unsigned long i; | |
327 | ||
328 | if (!entry) | |
329 | return; | |
330 | ||
331 | if (radix_tree_is_internal_node(entry)) { | |
332 | struct radix_tree_node *node = entry_to_node(entry); | |
333 | ||
334 | pr_debug("ida node: %p offset %d indices %lu-%lu parent %p free %lx shift %d count %d\n", | |
335 | node, node->offset, index * IDA_BITMAP_BITS, | |
336 | ((index | node_maxindex(node)) + 1) * | |
337 | IDA_BITMAP_BITS - 1, | |
338 | node->parent, node->tags[0][0], node->shift, | |
339 | node->count); | |
340 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) | |
341 | dump_ida_node(node->slots[i], | |
342 | index | (i << node->shift)); | |
3159f943 | 343 | } else if (xa_is_value(entry)) { |
d37cacc5 MW |
344 | pr_debug("ida excp: %p offset %d indices %lu-%lu data %lx\n", |
345 | entry, (int)(index & RADIX_TREE_MAP_MASK), | |
346 | index * IDA_BITMAP_BITS, | |
3159f943 MW |
347 | index * IDA_BITMAP_BITS + BITS_PER_XA_VALUE, |
348 | xa_to_value(entry)); | |
0a835c4f MW |
349 | } else { |
350 | struct ida_bitmap *bitmap = entry; | |
351 | ||
352 | pr_debug("ida btmp: %p offset %d indices %lu-%lu data", bitmap, | |
353 | (int)(index & RADIX_TREE_MAP_MASK), | |
354 | index * IDA_BITMAP_BITS, | |
355 | (index + 1) * IDA_BITMAP_BITS - 1); | |
356 | for (i = 0; i < IDA_BITMAP_LONGS; i++) | |
357 | pr_cont(" %lx", bitmap->bitmap[i]); | |
358 | pr_cont("\n"); | |
359 | } | |
360 | } | |
361 | ||
362 | static void ida_dump(struct ida *ida) | |
363 | { | |
364 | struct radix_tree_root *root = &ida->ida_rt; | |
7ad3d4d8 MW |
365 | pr_debug("ida: %p node %p free %d\n", ida, root->rnode, |
366 | root->gfp_mask >> ROOT_TAG_SHIFT); | |
0a835c4f MW |
367 | dump_ida_node(root->rnode, 0); |
368 | } | |
7cf19af4 MW |
369 | #endif |
370 | ||
1da177e4 LT |
371 | /* |
372 | * This assumes that the caller has performed appropriate preallocation, and | |
373 | * that the caller has pinned this thread of control to the current CPU. | |
374 | */ | |
375 | static struct radix_tree_node * | |
0a835c4f | 376 | radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent, |
d58275bc | 377 | struct radix_tree_root *root, |
e8de4340 MW |
378 | unsigned int shift, unsigned int offset, |
379 | unsigned int count, unsigned int exceptional) | |
1da177e4 | 380 | { |
e2848a0e | 381 | struct radix_tree_node *ret = NULL; |
1da177e4 | 382 | |
5e4c0d97 | 383 | /* |
2fcd9005 MW |
384 | * Preload code isn't irq safe and it doesn't make sense to use |
385 | * preloading during an interrupt anyway as all the allocations have | |
386 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 387 | */ |
d0164adc | 388 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
389 | struct radix_tree_preload *rtp; |
390 | ||
58e698af VD |
391 | /* |
392 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
393 | * cache first for the new node to get accounted to the memory |
394 | * cgroup. | |
58e698af VD |
395 | */ |
396 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 397 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
398 | if (ret) |
399 | goto out; | |
400 | ||
e2848a0e NP |
401 | /* |
402 | * Provided the caller has preloaded here, we will always | |
403 | * succeed in getting a node here (and never reach | |
404 | * kmem_cache_alloc) | |
405 | */ | |
7c8e0181 | 406 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 407 | if (rtp->nr) { |
9d2a8da0 | 408 | ret = rtp->nodes; |
1293d5c5 | 409 | rtp->nodes = ret->parent; |
1da177e4 LT |
410 | rtp->nr--; |
411 | } | |
ce80b067 CM |
412 | /* |
413 | * Update the allocation stack trace as this is more useful | |
414 | * for debugging. | |
415 | */ | |
416 | kmemleak_update_trace(ret); | |
58e698af | 417 | goto out; |
1da177e4 | 418 | } |
05eb6e72 | 419 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 420 | out: |
b194d16c | 421 | BUG_ON(radix_tree_is_internal_node(ret)); |
e8de4340 | 422 | if (ret) { |
e8de4340 MW |
423 | ret->shift = shift; |
424 | ret->offset = offset; | |
425 | ret->count = count; | |
426 | ret->exceptional = exceptional; | |
d58275bc MW |
427 | ret->parent = parent; |
428 | ret->root = root; | |
e8de4340 | 429 | } |
1da177e4 LT |
430 | return ret; |
431 | } | |
432 | ||
7cf9c2c7 NP |
433 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
434 | { | |
435 | struct radix_tree_node *node = | |
436 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
437 | |
438 | /* | |
175542f5 MW |
439 | * Must only free zeroed nodes into the slab. We can be left with |
440 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
441 | * and tags here. | |
643b52b9 | 442 | */ |
175542f5 MW |
443 | memset(node->slots, 0, sizeof(node->slots)); |
444 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 445 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 446 | |
7cf9c2c7 NP |
447 | kmem_cache_free(radix_tree_node_cachep, node); |
448 | } | |
449 | ||
1da177e4 LT |
450 | static inline void |
451 | radix_tree_node_free(struct radix_tree_node *node) | |
452 | { | |
7cf9c2c7 | 453 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
454 | } |
455 | ||
456 | /* | |
457 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
458 | * ensure that the addition of a single element in the tree cannot fail. On | |
459 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
460 | * with preemption not disabled. | |
b34df792 DH |
461 | * |
462 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 463 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 464 | */ |
bc9ae224 | 465 | static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr) |
1da177e4 LT |
466 | { |
467 | struct radix_tree_preload *rtp; | |
468 | struct radix_tree_node *node; | |
469 | int ret = -ENOMEM; | |
470 | ||
05eb6e72 VD |
471 | /* |
472 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
473 | * they should never be accounted to any particular memory cgroup. | |
474 | */ | |
475 | gfp_mask &= ~__GFP_ACCOUNT; | |
476 | ||
1da177e4 | 477 | preempt_disable(); |
7c8e0181 | 478 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 479 | while (rtp->nr < nr) { |
1da177e4 | 480 | preempt_enable(); |
488514d1 | 481 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
482 | if (node == NULL) |
483 | goto out; | |
484 | preempt_disable(); | |
7c8e0181 | 485 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 486 | if (rtp->nr < nr) { |
1293d5c5 | 487 | node->parent = rtp->nodes; |
9d2a8da0 KS |
488 | rtp->nodes = node; |
489 | rtp->nr++; | |
490 | } else { | |
1da177e4 | 491 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 492 | } |
1da177e4 LT |
493 | } |
494 | ret = 0; | |
495 | out: | |
496 | return ret; | |
497 | } | |
5e4c0d97 JK |
498 | |
499 | /* | |
500 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
501 | * ensure that the addition of a single element in the tree cannot fail. On | |
502 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
503 | * with preemption not disabled. | |
504 | * | |
505 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 506 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
507 | */ |
508 | int radix_tree_preload(gfp_t gfp_mask) | |
509 | { | |
510 | /* Warn on non-sensical use... */ | |
d0164adc | 511 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 512 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 513 | } |
d7f0923d | 514 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 515 | |
5e4c0d97 JK |
516 | /* |
517 | * The same as above function, except we don't guarantee preloading happens. | |
518 | * We do it, if we decide it helps. On success, return zero with preemption | |
519 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
520 | */ | |
521 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
522 | { | |
d0164adc | 523 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 524 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
525 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
526 | preempt_disable(); | |
527 | return 0; | |
528 | } | |
529 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
530 | ||
2791653a MW |
531 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
532 | /* | |
533 | * Preload with enough objects to ensure that we can split a single entry | |
534 | * of order @old_order into many entries of size @new_order | |
535 | */ | |
536 | int radix_tree_split_preload(unsigned int old_order, unsigned int new_order, | |
537 | gfp_t gfp_mask) | |
538 | { | |
539 | unsigned top = 1 << (old_order % RADIX_TREE_MAP_SHIFT); | |
540 | unsigned layers = (old_order / RADIX_TREE_MAP_SHIFT) - | |
541 | (new_order / RADIX_TREE_MAP_SHIFT); | |
542 | unsigned nr = 0; | |
543 | ||
544 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); | |
545 | BUG_ON(new_order >= old_order); | |
546 | ||
547 | while (layers--) | |
548 | nr = nr * RADIX_TREE_MAP_SIZE + 1; | |
549 | return __radix_tree_preload(gfp_mask, top * nr); | |
550 | } | |
551 | #endif | |
552 | ||
c78c66d1 KS |
553 | /* |
554 | * The same as function above, but preload number of nodes required to insert | |
555 | * (1 << order) continuous naturally-aligned elements. | |
556 | */ | |
557 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
558 | { | |
559 | unsigned long nr_subtrees; | |
560 | int nr_nodes, subtree_height; | |
561 | ||
562 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
563 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
564 | preempt_disable(); | |
565 | return 0; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Calculate number and height of fully populated subtrees it takes to | |
570 | * store (1 << order) elements. | |
571 | */ | |
572 | nr_subtrees = 1 << order; | |
573 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
574 | subtree_height++) | |
575 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
576 | ||
577 | /* | |
578 | * The worst case is zero height tree with a single item at index 0 and | |
579 | * then inserting items starting at ULONG_MAX - (1 << order). | |
580 | * | |
581 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
582 | * 0-index item. | |
583 | */ | |
584 | nr_nodes = RADIX_TREE_MAX_PATH; | |
585 | ||
586 | /* Plus branch to fully populated subtrees. */ | |
587 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
588 | ||
589 | /* Root node is shared. */ | |
590 | nr_nodes--; | |
591 | ||
592 | /* Plus nodes required to build subtrees. */ | |
593 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
594 | ||
595 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
596 | } | |
597 | ||
35534c86 | 598 | static unsigned radix_tree_load_root(const struct radix_tree_root *root, |
1456a439 MW |
599 | struct radix_tree_node **nodep, unsigned long *maxindex) |
600 | { | |
601 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
602 | ||
603 | *nodep = node; | |
604 | ||
b194d16c | 605 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 606 | node = entry_to_node(node); |
1456a439 | 607 | *maxindex = node_maxindex(node); |
c12e51b0 | 608 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
609 | } |
610 | ||
611 | *maxindex = 0; | |
612 | return 0; | |
613 | } | |
614 | ||
1da177e4 LT |
615 | /* |
616 | * Extend a radix tree so it can store key @index. | |
617 | */ | |
0a835c4f | 618 | static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp, |
d0891265 | 619 | unsigned long index, unsigned int shift) |
1da177e4 | 620 | { |
d7b62727 | 621 | void *entry; |
d0891265 | 622 | unsigned int maxshift; |
1da177e4 LT |
623 | int tag; |
624 | ||
d0891265 MW |
625 | /* Figure out what the shift should be. */ |
626 | maxshift = shift; | |
627 | while (index > shift_maxindex(maxshift)) | |
628 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 629 | |
d7b62727 MW |
630 | entry = rcu_dereference_raw(root->rnode); |
631 | if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE))) | |
1da177e4 | 632 | goto out; |
1da177e4 | 633 | |
1da177e4 | 634 | do { |
0a835c4f | 635 | struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL, |
d58275bc | 636 | root, shift, 0, 1, 0); |
2fcd9005 | 637 | if (!node) |
1da177e4 LT |
638 | return -ENOMEM; |
639 | ||
0a835c4f MW |
640 | if (is_idr(root)) { |
641 | all_tag_set(node, IDR_FREE); | |
642 | if (!root_tag_get(root, IDR_FREE)) { | |
643 | tag_clear(node, IDR_FREE, 0); | |
644 | root_tag_set(root, IDR_FREE); | |
645 | } | |
646 | } else { | |
647 | /* Propagate the aggregated tag info to the new child */ | |
648 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
649 | if (root_tag_get(root, tag)) | |
650 | tag_set(node, tag, 0); | |
651 | } | |
1da177e4 LT |
652 | } |
653 | ||
d0891265 | 654 | BUG_ON(shift > BITS_PER_LONG); |
d7b62727 MW |
655 | if (radix_tree_is_internal_node(entry)) { |
656 | entry_to_node(entry)->parent = node; | |
3159f943 | 657 | } else if (xa_is_value(entry)) { |
f7942430 | 658 | /* Moving an exceptional root->rnode to a node */ |
e8de4340 | 659 | node->exceptional = 1; |
f7942430 | 660 | } |
d7b62727 MW |
661 | /* |
662 | * entry was already in the radix tree, so we do not need | |
663 | * rcu_assign_pointer here | |
664 | */ | |
665 | node->slots[0] = (void __rcu *)entry; | |
666 | entry = node_to_entry(node); | |
667 | rcu_assign_pointer(root->rnode, entry); | |
d0891265 | 668 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 669 | } while (shift <= maxshift); |
1da177e4 | 670 | out: |
d0891265 | 671 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
672 | } |
673 | ||
f4b109c6 JW |
674 | /** |
675 | * radix_tree_shrink - shrink radix tree to minimum height | |
676 | * @root radix tree root | |
677 | */ | |
0ac398ef | 678 | static inline bool radix_tree_shrink(struct radix_tree_root *root, |
c7df8ad2 | 679 | radix_tree_update_node_t update_node) |
f4b109c6 | 680 | { |
0ac398ef MW |
681 | bool shrunk = false; |
682 | ||
f4b109c6 | 683 | for (;;) { |
12320d0f | 684 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); |
f4b109c6 JW |
685 | struct radix_tree_node *child; |
686 | ||
687 | if (!radix_tree_is_internal_node(node)) | |
688 | break; | |
689 | node = entry_to_node(node); | |
690 | ||
691 | /* | |
692 | * The candidate node has more than one child, or its child | |
693 | * is not at the leftmost slot, or the child is a multiorder | |
694 | * entry, we cannot shrink. | |
695 | */ | |
696 | if (node->count != 1) | |
697 | break; | |
12320d0f | 698 | child = rcu_dereference_raw(node->slots[0]); |
f4b109c6 JW |
699 | if (!child) |
700 | break; | |
701 | if (!radix_tree_is_internal_node(child) && node->shift) | |
702 | break; | |
703 | ||
66ee620f MW |
704 | /* |
705 | * For an IDR, we must not shrink entry 0 into the root in | |
706 | * case somebody calls idr_replace() with a pointer that | |
707 | * appears to be an internal entry | |
708 | */ | |
709 | if (!node->shift && is_idr(root)) | |
710 | break; | |
711 | ||
f4b109c6 JW |
712 | if (radix_tree_is_internal_node(child)) |
713 | entry_to_node(child)->parent = NULL; | |
714 | ||
715 | /* | |
716 | * We don't need rcu_assign_pointer(), since we are simply | |
717 | * moving the node from one part of the tree to another: if it | |
718 | * was safe to dereference the old pointer to it | |
719 | * (node->slots[0]), it will be safe to dereference the new | |
720 | * one (root->rnode) as far as dependent read barriers go. | |
721 | */ | |
d7b62727 | 722 | root->rnode = (void __rcu *)child; |
0a835c4f MW |
723 | if (is_idr(root) && !tag_get(node, IDR_FREE, 0)) |
724 | root_tag_clear(root, IDR_FREE); | |
f4b109c6 JW |
725 | |
726 | /* | |
727 | * We have a dilemma here. The node's slot[0] must not be | |
728 | * NULLed in case there are concurrent lookups expecting to | |
729 | * find the item. However if this was a bottom-level node, | |
730 | * then it may be subject to the slot pointer being visible | |
731 | * to callers dereferencing it. If item corresponding to | |
732 | * slot[0] is subsequently deleted, these callers would expect | |
733 | * their slot to become empty sooner or later. | |
734 | * | |
735 | * For example, lockless pagecache will look up a slot, deref | |
736 | * the page pointer, and if the page has 0 refcount it means it | |
737 | * was concurrently deleted from pagecache so try the deref | |
738 | * again. Fortunately there is already a requirement for logic | |
739 | * to retry the entire slot lookup -- the indirect pointer | |
740 | * problem (replacing direct root node with an indirect pointer | |
741 | * also results in a stale slot). So tag the slot as indirect | |
742 | * to force callers to retry. | |
743 | */ | |
4d693d08 JW |
744 | node->count = 0; |
745 | if (!radix_tree_is_internal_node(child)) { | |
d7b62727 | 746 | node->slots[0] = (void __rcu *)RADIX_TREE_RETRY; |
4d693d08 | 747 | if (update_node) |
c7df8ad2 | 748 | update_node(node); |
4d693d08 | 749 | } |
f4b109c6 | 750 | |
ea07b862 | 751 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 752 | radix_tree_node_free(node); |
0ac398ef | 753 | shrunk = true; |
f4b109c6 | 754 | } |
0ac398ef MW |
755 | |
756 | return shrunk; | |
f4b109c6 JW |
757 | } |
758 | ||
0ac398ef | 759 | static bool delete_node(struct radix_tree_root *root, |
4d693d08 | 760 | struct radix_tree_node *node, |
c7df8ad2 | 761 | radix_tree_update_node_t update_node) |
f4b109c6 | 762 | { |
0ac398ef MW |
763 | bool deleted = false; |
764 | ||
f4b109c6 JW |
765 | do { |
766 | struct radix_tree_node *parent; | |
767 | ||
768 | if (node->count) { | |
12320d0f MW |
769 | if (node_to_entry(node) == |
770 | rcu_dereference_raw(root->rnode)) | |
c7df8ad2 MG |
771 | deleted |= radix_tree_shrink(root, |
772 | update_node); | |
0ac398ef | 773 | return deleted; |
f4b109c6 JW |
774 | } |
775 | ||
776 | parent = node->parent; | |
777 | if (parent) { | |
778 | parent->slots[node->offset] = NULL; | |
779 | parent->count--; | |
780 | } else { | |
0a835c4f MW |
781 | /* |
782 | * Shouldn't the tags already have all been cleared | |
783 | * by the caller? | |
784 | */ | |
785 | if (!is_idr(root)) | |
786 | root_tag_clear_all(root); | |
f4b109c6 JW |
787 | root->rnode = NULL; |
788 | } | |
789 | ||
ea07b862 | 790 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 791 | radix_tree_node_free(node); |
0ac398ef | 792 | deleted = true; |
f4b109c6 JW |
793 | |
794 | node = parent; | |
795 | } while (node); | |
0ac398ef MW |
796 | |
797 | return deleted; | |
f4b109c6 JW |
798 | } |
799 | ||
1da177e4 | 800 | /** |
139e5616 | 801 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
802 | * @root: radix tree root |
803 | * @index: index key | |
e6145236 | 804 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
805 | * @nodep: returns node |
806 | * @slotp: returns slot | |
1da177e4 | 807 | * |
139e5616 JW |
808 | * Create, if necessary, and return the node and slot for an item |
809 | * at position @index in the radix tree @root. | |
810 | * | |
811 | * Until there is more than one item in the tree, no nodes are | |
812 | * allocated and @root->rnode is used as a direct slot instead of | |
813 | * pointing to a node, in which case *@nodep will be NULL. | |
814 | * | |
815 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 816 | */ |
139e5616 | 817 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 | 818 | unsigned order, struct radix_tree_node **nodep, |
d7b62727 | 819 | void __rcu ***slotp) |
1da177e4 | 820 | { |
89148aa4 | 821 | struct radix_tree_node *node = NULL, *child; |
d7b62727 | 822 | void __rcu **slot = (void __rcu **)&root->rnode; |
49ea6ebc | 823 | unsigned long maxindex; |
89148aa4 | 824 | unsigned int shift, offset = 0; |
49ea6ebc | 825 | unsigned long max = index | ((1UL << order) - 1); |
0a835c4f | 826 | gfp_t gfp = root_gfp_mask(root); |
49ea6ebc | 827 | |
89148aa4 | 828 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
829 | |
830 | /* Make sure the tree is high enough. */ | |
175542f5 MW |
831 | if (order > 0 && max == ((1UL << order) - 1)) |
832 | max++; | |
49ea6ebc | 833 | if (max > maxindex) { |
0a835c4f | 834 | int error = radix_tree_extend(root, gfp, max, shift); |
49ea6ebc | 835 | if (error < 0) |
1da177e4 | 836 | return error; |
49ea6ebc | 837 | shift = error; |
12320d0f | 838 | child = rcu_dereference_raw(root->rnode); |
1da177e4 LT |
839 | } |
840 | ||
e6145236 | 841 | while (shift > order) { |
c12e51b0 | 842 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 843 | if (child == NULL) { |
1da177e4 | 844 | /* Have to add a child node. */ |
d58275bc | 845 | child = radix_tree_node_alloc(gfp, node, root, shift, |
e8de4340 | 846 | offset, 0, 0); |
89148aa4 | 847 | if (!child) |
1da177e4 | 848 | return -ENOMEM; |
89148aa4 MW |
849 | rcu_assign_pointer(*slot, node_to_entry(child)); |
850 | if (node) | |
1da177e4 | 851 | node->count++; |
89148aa4 | 852 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 853 | break; |
1da177e4 LT |
854 | |
855 | /* Go a level down */ | |
89148aa4 | 856 | node = entry_to_node(child); |
9e85d811 | 857 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 858 | slot = &node->slots[offset]; |
e6145236 MW |
859 | } |
860 | ||
175542f5 MW |
861 | if (nodep) |
862 | *nodep = node; | |
863 | if (slotp) | |
864 | *slotp = slot; | |
865 | return 0; | |
866 | } | |
867 | ||
175542f5 MW |
868 | /* |
869 | * Free any nodes below this node. The tree is presumed to not need | |
870 | * shrinking, and any user data in the tree is presumed to not need a | |
871 | * destructor called on it. If we need to add a destructor, we can | |
872 | * add that functionality later. Note that we may not clear tags or | |
873 | * slots from the tree as an RCU walker may still have a pointer into | |
874 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
875 | * but we'll still have to clear those in rcu_free. | |
876 | */ | |
877 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
878 | { | |
879 | unsigned offset = 0; | |
880 | struct radix_tree_node *child = entry_to_node(node); | |
881 | ||
882 | for (;;) { | |
12320d0f | 883 | void *entry = rcu_dereference_raw(child->slots[offset]); |
66ee620f MW |
884 | if (radix_tree_is_internal_node(entry) && child->shift && |
885 | !is_sibling_entry(child, entry)) { | |
175542f5 MW |
886 | child = entry_to_node(entry); |
887 | offset = 0; | |
888 | continue; | |
889 | } | |
890 | offset++; | |
891 | while (offset == RADIX_TREE_MAP_SIZE) { | |
892 | struct radix_tree_node *old = child; | |
893 | offset = child->offset + 1; | |
894 | child = child->parent; | |
dd040b6f | 895 | WARN_ON_ONCE(!list_empty(&old->private_list)); |
175542f5 MW |
896 | radix_tree_node_free(old); |
897 | if (old == entry_to_node(node)) | |
898 | return; | |
899 | } | |
900 | } | |
901 | } | |
902 | ||
0a835c4f | 903 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
d7b62727 MW |
904 | static inline int insert_entries(struct radix_tree_node *node, |
905 | void __rcu **slot, void *item, unsigned order, bool replace) | |
175542f5 MW |
906 | { |
907 | struct radix_tree_node *child; | |
908 | unsigned i, n, tag, offset, tags = 0; | |
909 | ||
910 | if (node) { | |
e157b555 MW |
911 | if (order > node->shift) |
912 | n = 1 << (order - node->shift); | |
913 | else | |
914 | n = 1; | |
175542f5 MW |
915 | offset = get_slot_offset(node, slot); |
916 | } else { | |
917 | n = 1; | |
918 | offset = 0; | |
919 | } | |
920 | ||
921 | if (n > 1) { | |
e6145236 | 922 | offset = offset & ~(n - 1); |
89148aa4 | 923 | slot = &node->slots[offset]; |
175542f5 MW |
924 | } |
925 | child = node_to_entry(slot); | |
926 | ||
927 | for (i = 0; i < n; i++) { | |
928 | if (slot[i]) { | |
929 | if (replace) { | |
930 | node->count--; | |
931 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
932 | if (tag_get(node, tag, offset + i)) | |
933 | tags |= 1 << tag; | |
934 | } else | |
e6145236 MW |
935 | return -EEXIST; |
936 | } | |
175542f5 | 937 | } |
e6145236 | 938 | |
175542f5 | 939 | for (i = 0; i < n; i++) { |
12320d0f | 940 | struct radix_tree_node *old = rcu_dereference_raw(slot[i]); |
175542f5 | 941 | if (i) { |
89148aa4 | 942 | rcu_assign_pointer(slot[i], child); |
175542f5 MW |
943 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
944 | if (tags & (1 << tag)) | |
945 | tag_clear(node, tag, offset + i); | |
946 | } else { | |
947 | rcu_assign_pointer(slot[i], item); | |
948 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
949 | if (tags & (1 << tag)) | |
950 | tag_set(node, tag, offset); | |
e6145236 | 951 | } |
175542f5 | 952 | if (radix_tree_is_internal_node(old) && |
e157b555 MW |
953 | !is_sibling_entry(node, old) && |
954 | (old != RADIX_TREE_RETRY)) | |
175542f5 | 955 | radix_tree_free_nodes(old); |
3159f943 | 956 | if (xa_is_value(old)) |
175542f5 | 957 | node->exceptional--; |
612d6c19 | 958 | } |
175542f5 MW |
959 | if (node) { |
960 | node->count += n; | |
3159f943 | 961 | if (xa_is_value(item)) |
175542f5 MW |
962 | node->exceptional += n; |
963 | } | |
964 | return n; | |
139e5616 | 965 | } |
175542f5 | 966 | #else |
d7b62727 MW |
967 | static inline int insert_entries(struct radix_tree_node *node, |
968 | void __rcu **slot, void *item, unsigned order, bool replace) | |
175542f5 MW |
969 | { |
970 | if (*slot) | |
971 | return -EEXIST; | |
972 | rcu_assign_pointer(*slot, item); | |
973 | if (node) { | |
974 | node->count++; | |
3159f943 | 975 | if (xa_is_value(item)) |
175542f5 MW |
976 | node->exceptional++; |
977 | } | |
978 | return 1; | |
979 | } | |
980 | #endif | |
139e5616 JW |
981 | |
982 | /** | |
e6145236 | 983 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
984 | * @root: radix tree root |
985 | * @index: index key | |
e6145236 | 986 | * @order: key covers the 2^order indices around index |
139e5616 JW |
987 | * @item: item to insert |
988 | * | |
989 | * Insert an item into the radix tree at position @index. | |
990 | */ | |
e6145236 MW |
991 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
992 | unsigned order, void *item) | |
139e5616 JW |
993 | { |
994 | struct radix_tree_node *node; | |
d7b62727 | 995 | void __rcu **slot; |
139e5616 JW |
996 | int error; |
997 | ||
b194d16c | 998 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 999 | |
e6145236 | 1000 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
1001 | if (error) |
1002 | return error; | |
175542f5 MW |
1003 | |
1004 | error = insert_entries(node, slot, item, order, false); | |
1005 | if (error < 0) | |
1006 | return error; | |
201b6264 | 1007 | |
612d6c19 | 1008 | if (node) { |
7b60e9ad | 1009 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
1010 | BUG_ON(tag_get(node, 0, offset)); |
1011 | BUG_ON(tag_get(node, 1, offset)); | |
1012 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 1013 | } else { |
7b60e9ad | 1014 | BUG_ON(root_tags_get(root)); |
612d6c19 | 1015 | } |
1da177e4 | 1016 | |
1da177e4 LT |
1017 | return 0; |
1018 | } | |
e6145236 | 1019 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 1020 | |
139e5616 JW |
1021 | /** |
1022 | * __radix_tree_lookup - lookup an item in a radix tree | |
1023 | * @root: radix tree root | |
1024 | * @index: index key | |
1025 | * @nodep: returns node | |
1026 | * @slotp: returns slot | |
1027 | * | |
1028 | * Lookup and return the item at position @index in the radix | |
1029 | * tree @root. | |
1030 | * | |
1031 | * Until there is more than one item in the tree, no nodes are | |
1032 | * allocated and @root->rnode is used as a direct slot instead of | |
1033 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 1034 | */ |
35534c86 MW |
1035 | void *__radix_tree_lookup(const struct radix_tree_root *root, |
1036 | unsigned long index, struct radix_tree_node **nodep, | |
d7b62727 | 1037 | void __rcu ***slotp) |
1da177e4 | 1038 | { |
139e5616 | 1039 | struct radix_tree_node *node, *parent; |
85829954 | 1040 | unsigned long maxindex; |
d7b62727 | 1041 | void __rcu **slot; |
612d6c19 | 1042 | |
85829954 MW |
1043 | restart: |
1044 | parent = NULL; | |
d7b62727 | 1045 | slot = (void __rcu **)&root->rnode; |
9e85d811 | 1046 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 1047 | if (index > maxindex) |
1da177e4 LT |
1048 | return NULL; |
1049 | ||
b194d16c | 1050 | while (radix_tree_is_internal_node(node)) { |
85829954 | 1051 | unsigned offset; |
1da177e4 | 1052 | |
85829954 MW |
1053 | if (node == RADIX_TREE_RETRY) |
1054 | goto restart; | |
4dd6c098 | 1055 | parent = entry_to_node(node); |
9e85d811 | 1056 | offset = radix_tree_descend(parent, &node, index); |
85829954 | 1057 | slot = parent->slots + offset; |
66ee620f MW |
1058 | if (parent->shift == 0) |
1059 | break; | |
85829954 | 1060 | } |
1da177e4 | 1061 | |
139e5616 JW |
1062 | if (nodep) |
1063 | *nodep = parent; | |
1064 | if (slotp) | |
1065 | *slotp = slot; | |
1066 | return node; | |
b72b71c6 HS |
1067 | } |
1068 | ||
1069 | /** | |
1070 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
1071 | * @root: radix tree root | |
1072 | * @index: index key | |
1073 | * | |
1074 | * Returns: the slot corresponding to the position @index in the | |
1075 | * radix tree @root. This is useful for update-if-exists operations. | |
1076 | * | |
1077 | * This function can be called under rcu_read_lock iff the slot is not | |
1078 | * modified by radix_tree_replace_slot, otherwise it must be called | |
1079 | * exclusive from other writers. Any dereference of the slot must be done | |
1080 | * using radix_tree_deref_slot. | |
1081 | */ | |
d7b62727 | 1082 | void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root, |
35534c86 | 1083 | unsigned long index) |
b72b71c6 | 1084 | { |
d7b62727 | 1085 | void __rcu **slot; |
139e5616 JW |
1086 | |
1087 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
1088 | return NULL; | |
1089 | return slot; | |
a4331366 | 1090 | } |
a4331366 HR |
1091 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
1092 | ||
1093 | /** | |
1094 | * radix_tree_lookup - perform lookup operation on a radix tree | |
1095 | * @root: radix tree root | |
1096 | * @index: index key | |
1097 | * | |
1098 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
1099 | * |
1100 | * This function can be called under rcu_read_lock, however the caller | |
1101 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
1102 | * them safely). No RCU barriers are required to access or modify the | |
1103 | * returned item, however. | |
a4331366 | 1104 | */ |
35534c86 | 1105 | void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index) |
a4331366 | 1106 | { |
139e5616 | 1107 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
1108 | } |
1109 | EXPORT_SYMBOL(radix_tree_lookup); | |
1110 | ||
0a835c4f | 1111 | static inline void replace_sibling_entries(struct radix_tree_node *node, |
d7b62727 | 1112 | void __rcu **slot, int count, int exceptional) |
a90eb3a2 | 1113 | { |
a90eb3a2 MW |
1114 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
1115 | void *ptr = node_to_entry(slot); | |
0a835c4f | 1116 | unsigned offset = get_slot_offset(node, slot) + 1; |
a90eb3a2 | 1117 | |
0a835c4f | 1118 | while (offset < RADIX_TREE_MAP_SIZE) { |
12320d0f | 1119 | if (rcu_dereference_raw(node->slots[offset]) != ptr) |
a90eb3a2 | 1120 | break; |
0a835c4f MW |
1121 | if (count < 0) { |
1122 | node->slots[offset] = NULL; | |
1123 | node->count--; | |
1124 | } | |
1125 | node->exceptional += exceptional; | |
1126 | offset++; | |
a90eb3a2 MW |
1127 | } |
1128 | #endif | |
a90eb3a2 MW |
1129 | } |
1130 | ||
d7b62727 MW |
1131 | static void replace_slot(void __rcu **slot, void *item, |
1132 | struct radix_tree_node *node, int count, int exceptional) | |
f7942430 | 1133 | { |
0a835c4f | 1134 | if (node && (count || exceptional)) { |
f4b109c6 | 1135 | node->count += count; |
0a835c4f MW |
1136 | node->exceptional += exceptional; |
1137 | replace_sibling_entries(node, slot, count, exceptional); | |
f4b109c6 | 1138 | } |
f7942430 JW |
1139 | |
1140 | rcu_assign_pointer(*slot, item); | |
1141 | } | |
1142 | ||
0a835c4f MW |
1143 | static bool node_tag_get(const struct radix_tree_root *root, |
1144 | const struct radix_tree_node *node, | |
1145 | unsigned int tag, unsigned int offset) | |
a90eb3a2 | 1146 | { |
0a835c4f MW |
1147 | if (node) |
1148 | return tag_get(node, tag, offset); | |
1149 | return root_tag_get(root, tag); | |
1150 | } | |
a90eb3a2 | 1151 | |
0a835c4f MW |
1152 | /* |
1153 | * IDR users want to be able to store NULL in the tree, so if the slot isn't | |
1154 | * free, don't adjust the count, even if it's transitioning between NULL and | |
1155 | * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still | |
1156 | * have empty bits, but it only stores NULL in slots when they're being | |
1157 | * deleted. | |
1158 | */ | |
1159 | static int calculate_count(struct radix_tree_root *root, | |
d7b62727 | 1160 | struct radix_tree_node *node, void __rcu **slot, |
0a835c4f MW |
1161 | void *item, void *old) |
1162 | { | |
1163 | if (is_idr(root)) { | |
1164 | unsigned offset = get_slot_offset(node, slot); | |
1165 | bool free = node_tag_get(root, node, IDR_FREE, offset); | |
1166 | if (!free) | |
1167 | return 0; | |
1168 | if (!old) | |
1169 | return 1; | |
a90eb3a2 | 1170 | } |
0a835c4f | 1171 | return !!item - !!old; |
a90eb3a2 MW |
1172 | } |
1173 | ||
6d75f366 JW |
1174 | /** |
1175 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
1176 | * @root: radix tree root |
1177 | * @node: pointer to tree node | |
1178 | * @slot: pointer to slot in @node | |
1179 | * @item: new item to store in the slot. | |
1180 | * @update_node: callback for changing leaf nodes | |
6d75f366 JW |
1181 | * |
1182 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
1183 | * across slot lookup and replacement. | |
1184 | */ | |
1185 | void __radix_tree_replace(struct radix_tree_root *root, | |
1186 | struct radix_tree_node *node, | |
d7b62727 | 1187 | void __rcu **slot, void *item, |
c7df8ad2 | 1188 | radix_tree_update_node_t update_node) |
6d75f366 | 1189 | { |
0a835c4f | 1190 | void *old = rcu_dereference_raw(*slot); |
3159f943 | 1191 | int exceptional = !!xa_is_value(item) - !!xa_is_value(old); |
0a835c4f MW |
1192 | int count = calculate_count(root, node, slot, item, old); |
1193 | ||
6d75f366 | 1194 | /* |
f4b109c6 JW |
1195 | * This function supports replacing exceptional entries and |
1196 | * deleting entries, but that needs accounting against the | |
1197 | * node unless the slot is root->rnode. | |
6d75f366 | 1198 | */ |
d7b62727 | 1199 | WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->rnode) && |
0a835c4f MW |
1200 | (count || exceptional)); |
1201 | replace_slot(slot, item, node, count, exceptional); | |
f4b109c6 | 1202 | |
4d693d08 JW |
1203 | if (!node) |
1204 | return; | |
1205 | ||
1206 | if (update_node) | |
c7df8ad2 | 1207 | update_node(node); |
4d693d08 | 1208 | |
c7df8ad2 | 1209 | delete_node(root, node, update_node); |
6d75f366 JW |
1210 | } |
1211 | ||
1212 | /** | |
1213 | * radix_tree_replace_slot - replace item in a slot | |
1214 | * @root: radix tree root | |
1215 | * @slot: pointer to slot | |
1216 | * @item: new item to store in the slot. | |
1217 | * | |
1218 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
1219 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
1220 | * across slot lookup and replacement. | |
1221 | * | |
1222 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
1223 | * regular entries, and exceptional entries, as that requires accounting | |
f4b109c6 | 1224 | * inside the radix tree node. When switching from one type of entry or |
e157b555 MW |
1225 | * deleting, use __radix_tree_lookup() and __radix_tree_replace() or |
1226 | * radix_tree_iter_replace(). | |
6d75f366 JW |
1227 | */ |
1228 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
d7b62727 | 1229 | void __rcu **slot, void *item) |
6d75f366 | 1230 | { |
c7df8ad2 | 1231 | __radix_tree_replace(root, NULL, slot, item, NULL); |
6d75f366 | 1232 | } |
10257d71 | 1233 | EXPORT_SYMBOL(radix_tree_replace_slot); |
6d75f366 | 1234 | |
e157b555 MW |
1235 | /** |
1236 | * radix_tree_iter_replace - replace item in a slot | |
1237 | * @root: radix tree root | |
1238 | * @slot: pointer to slot | |
1239 | * @item: new item to store in the slot. | |
1240 | * | |
1241 | * For use with radix_tree_split() and radix_tree_for_each_slot(). | |
1242 | * Caller must hold tree write locked across split and replacement. | |
1243 | */ | |
1244 | void radix_tree_iter_replace(struct radix_tree_root *root, | |
d7b62727 MW |
1245 | const struct radix_tree_iter *iter, |
1246 | void __rcu **slot, void *item) | |
e157b555 | 1247 | { |
c7df8ad2 | 1248 | __radix_tree_replace(root, iter->node, slot, item, NULL); |
e157b555 MW |
1249 | } |
1250 | ||
175542f5 MW |
1251 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
1252 | /** | |
1253 | * radix_tree_join - replace multiple entries with one multiorder entry | |
1254 | * @root: radix tree root | |
1255 | * @index: an index inside the new entry | |
1256 | * @order: order of the new entry | |
1257 | * @item: new entry | |
1258 | * | |
1259 | * Call this function to replace several entries with one larger entry. | |
1260 | * The existing entries are presumed to not need freeing as a result of | |
1261 | * this call. | |
1262 | * | |
1263 | * The replacement entry will have all the tags set on it that were set | |
1264 | * on any of the entries it is replacing. | |
1265 | */ | |
1266 | int radix_tree_join(struct radix_tree_root *root, unsigned long index, | |
1267 | unsigned order, void *item) | |
1268 | { | |
1269 | struct radix_tree_node *node; | |
d7b62727 | 1270 | void __rcu **slot; |
175542f5 MW |
1271 | int error; |
1272 | ||
1273 | BUG_ON(radix_tree_is_internal_node(item)); | |
1274 | ||
1275 | error = __radix_tree_create(root, index, order, &node, &slot); | |
1276 | if (!error) | |
1277 | error = insert_entries(node, slot, item, order, true); | |
1278 | if (error > 0) | |
1279 | error = 0; | |
1280 | ||
1281 | return error; | |
1282 | } | |
e157b555 MW |
1283 | |
1284 | /** | |
1285 | * radix_tree_split - Split an entry into smaller entries | |
1286 | * @root: radix tree root | |
1287 | * @index: An index within the large entry | |
1288 | * @order: Order of new entries | |
1289 | * | |
1290 | * Call this function as the first step in replacing a multiorder entry | |
1291 | * with several entries of lower order. After this function returns, | |
1292 | * loop over the relevant portion of the tree using radix_tree_for_each_slot() | |
1293 | * and call radix_tree_iter_replace() to set up each new entry. | |
1294 | * | |
1295 | * The tags from this entry are replicated to all the new entries. | |
1296 | * | |
1297 | * The radix tree should be locked against modification during the entire | |
1298 | * replacement operation. Lock-free lookups will see RADIX_TREE_RETRY which | |
1299 | * should prompt RCU walkers to restart the lookup from the root. | |
1300 | */ | |
1301 | int radix_tree_split(struct radix_tree_root *root, unsigned long index, | |
1302 | unsigned order) | |
1303 | { | |
1304 | struct radix_tree_node *parent, *node, *child; | |
d7b62727 | 1305 | void __rcu **slot; |
e157b555 MW |
1306 | unsigned int offset, end; |
1307 | unsigned n, tag, tags = 0; | |
0a835c4f | 1308 | gfp_t gfp = root_gfp_mask(root); |
e157b555 MW |
1309 | |
1310 | if (!__radix_tree_lookup(root, index, &parent, &slot)) | |
1311 | return -ENOENT; | |
1312 | if (!parent) | |
1313 | return -ENOENT; | |
1314 | ||
1315 | offset = get_slot_offset(parent, slot); | |
1316 | ||
1317 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1318 | if (tag_get(parent, tag, offset)) | |
1319 | tags |= 1 << tag; | |
1320 | ||
1321 | for (end = offset + 1; end < RADIX_TREE_MAP_SIZE; end++) { | |
12320d0f MW |
1322 | if (!is_sibling_entry(parent, |
1323 | rcu_dereference_raw(parent->slots[end]))) | |
e157b555 MW |
1324 | break; |
1325 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1326 | if (tags & (1 << tag)) | |
1327 | tag_set(parent, tag, end); | |
1328 | /* rcu_assign_pointer ensures tags are set before RETRY */ | |
1329 | rcu_assign_pointer(parent->slots[end], RADIX_TREE_RETRY); | |
1330 | } | |
1331 | rcu_assign_pointer(parent->slots[offset], RADIX_TREE_RETRY); | |
1332 | parent->exceptional -= (end - offset); | |
1333 | ||
1334 | if (order == parent->shift) | |
1335 | return 0; | |
1336 | if (order > parent->shift) { | |
1337 | while (offset < end) | |
1338 | offset += insert_entries(parent, &parent->slots[offset], | |
1339 | RADIX_TREE_RETRY, order, true); | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | node = parent; | |
1344 | ||
1345 | for (;;) { | |
1346 | if (node->shift > order) { | |
d58275bc | 1347 | child = radix_tree_node_alloc(gfp, node, root, |
e8de4340 MW |
1348 | node->shift - RADIX_TREE_MAP_SHIFT, |
1349 | offset, 0, 0); | |
e157b555 MW |
1350 | if (!child) |
1351 | goto nomem; | |
e157b555 MW |
1352 | if (node != parent) { |
1353 | node->count++; | |
12320d0f MW |
1354 | rcu_assign_pointer(node->slots[offset], |
1355 | node_to_entry(child)); | |
e157b555 MW |
1356 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
1357 | if (tags & (1 << tag)) | |
1358 | tag_set(node, tag, offset); | |
1359 | } | |
1360 | ||
1361 | node = child; | |
1362 | offset = 0; | |
1363 | continue; | |
1364 | } | |
1365 | ||
1366 | n = insert_entries(node, &node->slots[offset], | |
1367 | RADIX_TREE_RETRY, order, false); | |
1368 | BUG_ON(n > RADIX_TREE_MAP_SIZE); | |
1369 | ||
1370 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1371 | if (tags & (1 << tag)) | |
1372 | tag_set(node, tag, offset); | |
1373 | offset += n; | |
1374 | ||
1375 | while (offset == RADIX_TREE_MAP_SIZE) { | |
1376 | if (node == parent) | |
1377 | break; | |
1378 | offset = node->offset; | |
1379 | child = node; | |
1380 | node = node->parent; | |
1381 | rcu_assign_pointer(node->slots[offset], | |
1382 | node_to_entry(child)); | |
1383 | offset++; | |
1384 | } | |
1385 | if ((node == parent) && (offset == end)) | |
1386 | return 0; | |
1387 | } | |
1388 | ||
1389 | nomem: | |
1390 | /* Shouldn't happen; did user forget to preload? */ | |
1391 | /* TODO: free all the allocated nodes */ | |
1392 | WARN_ON(1); | |
1393 | return -ENOMEM; | |
1394 | } | |
175542f5 MW |
1395 | #endif |
1396 | ||
30b888ba MW |
1397 | static void node_tag_set(struct radix_tree_root *root, |
1398 | struct radix_tree_node *node, | |
1399 | unsigned int tag, unsigned int offset) | |
1400 | { | |
1401 | while (node) { | |
1402 | if (tag_get(node, tag, offset)) | |
1403 | return; | |
1404 | tag_set(node, tag, offset); | |
1405 | offset = node->offset; | |
1406 | node = node->parent; | |
1407 | } | |
1408 | ||
1409 | if (!root_tag_get(root, tag)) | |
1410 | root_tag_set(root, tag); | |
1411 | } | |
1412 | ||
1da177e4 LT |
1413 | /** |
1414 | * radix_tree_tag_set - set a tag on a radix tree node | |
1415 | * @root: radix tree root | |
1416 | * @index: index key | |
2fcd9005 | 1417 | * @tag: tag index |
1da177e4 | 1418 | * |
daff89f3 JC |
1419 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
1420 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
1421 | * the root all the way down to the leaf node. |
1422 | * | |
2fcd9005 | 1423 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
1424 | * item is a bug. |
1425 | */ | |
1426 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 1427 | unsigned long index, unsigned int tag) |
1da177e4 | 1428 | { |
fb969909 RZ |
1429 | struct radix_tree_node *node, *parent; |
1430 | unsigned long maxindex; | |
1da177e4 | 1431 | |
9e85d811 | 1432 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 1433 | BUG_ON(index > maxindex); |
1da177e4 | 1434 | |
b194d16c | 1435 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 1436 | unsigned offset; |
1da177e4 | 1437 | |
4dd6c098 | 1438 | parent = entry_to_node(node); |
9e85d811 | 1439 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
1440 | BUG_ON(!node); |
1441 | ||
1442 | if (!tag_get(parent, tag, offset)) | |
1443 | tag_set(parent, tag, offset); | |
1da177e4 LT |
1444 | } |
1445 | ||
612d6c19 | 1446 | /* set the root's tag bit */ |
fb969909 | 1447 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
1448 | root_tag_set(root, tag); |
1449 | ||
fb969909 | 1450 | return node; |
1da177e4 LT |
1451 | } |
1452 | EXPORT_SYMBOL(radix_tree_tag_set); | |
1453 | ||
30b888ba MW |
1454 | /** |
1455 | * radix_tree_iter_tag_set - set a tag on the current iterator entry | |
1456 | * @root: radix tree root | |
1457 | * @iter: iterator state | |
1458 | * @tag: tag to set | |
1459 | */ | |
1460 | void radix_tree_iter_tag_set(struct radix_tree_root *root, | |
1461 | const struct radix_tree_iter *iter, unsigned int tag) | |
1462 | { | |
1463 | node_tag_set(root, iter->node, tag, iter_offset(iter)); | |
1464 | } | |
1465 | ||
d604c324 MW |
1466 | static void node_tag_clear(struct radix_tree_root *root, |
1467 | struct radix_tree_node *node, | |
1468 | unsigned int tag, unsigned int offset) | |
1469 | { | |
1470 | while (node) { | |
1471 | if (!tag_get(node, tag, offset)) | |
1472 | return; | |
1473 | tag_clear(node, tag, offset); | |
1474 | if (any_tag_set(node, tag)) | |
1475 | return; | |
1476 | ||
1477 | offset = node->offset; | |
1478 | node = node->parent; | |
1479 | } | |
1480 | ||
1481 | /* clear the root's tag bit */ | |
1482 | if (root_tag_get(root, tag)) | |
1483 | root_tag_clear(root, tag); | |
1484 | } | |
1485 | ||
1da177e4 LT |
1486 | /** |
1487 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1488 | * @root: radix tree root | |
1489 | * @index: index key | |
2fcd9005 | 1490 | * @tag: tag index |
1da177e4 | 1491 | * |
daff89f3 | 1492 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1493 | * corresponding to @index in the radix tree. If this causes |
1494 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1495 | * next-to-leaf node, etc. |
1496 | * | |
1497 | * Returns the address of the tagged item on success, else NULL. ie: | |
1498 | * has the same return value and semantics as radix_tree_lookup(). | |
1499 | */ | |
1500 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1501 | unsigned long index, unsigned int tag) |
1da177e4 | 1502 | { |
00f47b58 RZ |
1503 | struct radix_tree_node *node, *parent; |
1504 | unsigned long maxindex; | |
e2bdb933 | 1505 | int uninitialized_var(offset); |
1da177e4 | 1506 | |
9e85d811 | 1507 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1508 | if (index > maxindex) |
1509 | return NULL; | |
1da177e4 | 1510 | |
00f47b58 | 1511 | parent = NULL; |
1da177e4 | 1512 | |
b194d16c | 1513 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1514 | parent = entry_to_node(node); |
9e85d811 | 1515 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1516 | } |
1517 | ||
d604c324 MW |
1518 | if (node) |
1519 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1520 | |
00f47b58 | 1521 | return node; |
1da177e4 LT |
1522 | } |
1523 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1524 | ||
30b888ba MW |
1525 | /** |
1526 | * radix_tree_iter_tag_clear - clear a tag on the current iterator entry | |
1527 | * @root: radix tree root | |
1528 | * @iter: iterator state | |
1529 | * @tag: tag to clear | |
1530 | */ | |
1531 | void radix_tree_iter_tag_clear(struct radix_tree_root *root, | |
1532 | const struct radix_tree_iter *iter, unsigned int tag) | |
1533 | { | |
1534 | node_tag_clear(root, iter->node, tag, iter_offset(iter)); | |
1535 | } | |
1536 | ||
1da177e4 | 1537 | /** |
32605a18 MT |
1538 | * radix_tree_tag_get - get a tag on a radix tree node |
1539 | * @root: radix tree root | |
1540 | * @index: index key | |
2fcd9005 | 1541 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1542 | * |
32605a18 | 1543 | * Return values: |
1da177e4 | 1544 | * |
612d6c19 NP |
1545 | * 0: tag not present or not set |
1546 | * 1: tag set | |
ce82653d DH |
1547 | * |
1548 | * Note that the return value of this function may not be relied on, even if | |
1549 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1550 | * from concurrency. | |
1da177e4 | 1551 | */ |
35534c86 | 1552 | int radix_tree_tag_get(const struct radix_tree_root *root, |
daff89f3 | 1553 | unsigned long index, unsigned int tag) |
1da177e4 | 1554 | { |
4589ba6d RZ |
1555 | struct radix_tree_node *node, *parent; |
1556 | unsigned long maxindex; | |
1da177e4 | 1557 | |
612d6c19 NP |
1558 | if (!root_tag_get(root, tag)) |
1559 | return 0; | |
1560 | ||
9e85d811 | 1561 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1562 | if (index > maxindex) |
1563 | return 0; | |
7cf9c2c7 | 1564 | |
b194d16c | 1565 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1566 | unsigned offset; |
1da177e4 | 1567 | |
4dd6c098 | 1568 | parent = entry_to_node(node); |
9e85d811 | 1569 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1570 | |
4589ba6d | 1571 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1572 | return 0; |
4589ba6d RZ |
1573 | if (node == RADIX_TREE_RETRY) |
1574 | break; | |
1da177e4 | 1575 | } |
4589ba6d RZ |
1576 | |
1577 | return 1; | |
1da177e4 LT |
1578 | } |
1579 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1580 | |
21ef5339 RZ |
1581 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1582 | unsigned int shift) | |
1583 | { | |
1584 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1585 | iter->shift = shift; | |
1586 | #endif | |
1587 | } | |
1588 | ||
148deab2 MW |
1589 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1590 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1591 | struct radix_tree_node *node, unsigned offset, | |
1592 | unsigned tag) | |
1593 | { | |
1594 | unsigned tag_long = offset / BITS_PER_LONG; | |
1595 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1596 | ||
0a835c4f MW |
1597 | if (!node) { |
1598 | iter->tags = 1; | |
1599 | return; | |
1600 | } | |
1601 | ||
148deab2 MW |
1602 | iter->tags = node->tags[tag][tag_long] >> tag_bit; |
1603 | ||
1604 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1605 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1606 | /* Pick tags from next element */ | |
1607 | if (tag_bit) | |
1608 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1609 | (BITS_PER_LONG - tag_bit); | |
1610 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1611 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1612 | } | |
1613 | } | |
1614 | ||
1615 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
d7b62727 MW |
1616 | static void __rcu **skip_siblings(struct radix_tree_node **nodep, |
1617 | void __rcu **slot, struct radix_tree_iter *iter) | |
148deab2 | 1618 | { |
148deab2 MW |
1619 | while (iter->index < iter->next_index) { |
1620 | *nodep = rcu_dereference_raw(*slot); | |
9f418224 | 1621 | if (*nodep && !is_sibling_entry(iter->node, *nodep)) |
148deab2 MW |
1622 | return slot; |
1623 | slot++; | |
1624 | iter->index = __radix_tree_iter_add(iter, 1); | |
1625 | iter->tags >>= 1; | |
1626 | } | |
1627 | ||
1628 | *nodep = NULL; | |
1629 | return NULL; | |
1630 | } | |
1631 | ||
d7b62727 MW |
1632 | void __rcu **__radix_tree_next_slot(void __rcu **slot, |
1633 | struct radix_tree_iter *iter, unsigned flags) | |
148deab2 MW |
1634 | { |
1635 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
9f418224 | 1636 | struct radix_tree_node *node; |
148deab2 MW |
1637 | |
1638 | slot = skip_siblings(&node, slot, iter); | |
1639 | ||
1640 | while (radix_tree_is_internal_node(node)) { | |
1641 | unsigned offset; | |
1642 | unsigned long next_index; | |
1643 | ||
1644 | if (node == RADIX_TREE_RETRY) | |
1645 | return slot; | |
1646 | node = entry_to_node(node); | |
268f42de | 1647 | iter->node = node; |
148deab2 MW |
1648 | iter->shift = node->shift; |
1649 | ||
1650 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1651 | offset = radix_tree_find_next_bit(node, tag, 0); | |
1652 | if (offset == RADIX_TREE_MAP_SIZE) | |
1653 | return NULL; | |
1654 | slot = &node->slots[offset]; | |
1655 | iter->index = __radix_tree_iter_add(iter, offset); | |
1656 | set_iter_tags(iter, node, offset, tag); | |
1657 | node = rcu_dereference_raw(*slot); | |
1658 | } else { | |
1659 | offset = 0; | |
1660 | slot = &node->slots[0]; | |
1661 | for (;;) { | |
1662 | node = rcu_dereference_raw(*slot); | |
1663 | if (node) | |
1664 | break; | |
1665 | slot++; | |
1666 | offset++; | |
1667 | if (offset == RADIX_TREE_MAP_SIZE) | |
1668 | return NULL; | |
1669 | } | |
1670 | iter->index = __radix_tree_iter_add(iter, offset); | |
1671 | } | |
1672 | if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0)) | |
1673 | goto none; | |
1674 | next_index = (iter->index | shift_maxindex(iter->shift)) + 1; | |
1675 | if (next_index < iter->next_index) | |
1676 | iter->next_index = next_index; | |
1677 | } | |
1678 | ||
1679 | return slot; | |
1680 | none: | |
1681 | iter->next_index = 0; | |
1682 | return NULL; | |
1683 | } | |
1684 | EXPORT_SYMBOL(__radix_tree_next_slot); | |
1685 | #else | |
d7b62727 MW |
1686 | static void __rcu **skip_siblings(struct radix_tree_node **nodep, |
1687 | void __rcu **slot, struct radix_tree_iter *iter) | |
148deab2 MW |
1688 | { |
1689 | return slot; | |
1690 | } | |
1691 | #endif | |
1692 | ||
d7b62727 MW |
1693 | void __rcu **radix_tree_iter_resume(void __rcu **slot, |
1694 | struct radix_tree_iter *iter) | |
148deab2 MW |
1695 | { |
1696 | struct radix_tree_node *node; | |
1697 | ||
1698 | slot++; | |
1699 | iter->index = __radix_tree_iter_add(iter, 1); | |
148deab2 MW |
1700 | skip_siblings(&node, slot, iter); |
1701 | iter->next_index = iter->index; | |
1702 | iter->tags = 0; | |
1703 | return NULL; | |
1704 | } | |
1705 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1706 | ||
78c1d784 KK |
1707 | /** |
1708 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1709 | * | |
1710 | * @root: radix tree root | |
1711 | * @iter: iterator state | |
1712 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1713 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1714 | */ | |
d7b62727 | 1715 | void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root, |
78c1d784 KK |
1716 | struct radix_tree_iter *iter, unsigned flags) |
1717 | { | |
9e85d811 | 1718 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1719 | struct radix_tree_node *node, *child; |
21ef5339 | 1720 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1721 | |
1722 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1723 | return NULL; | |
1724 | ||
1725 | /* | |
1726 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1727 | * during iterating; it can be zero only at the beginning. | |
1728 | * And we cannot overflow iter->next_index in a single step, | |
1729 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1730 | * |
1731 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1732 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1733 | */ |
1734 | index = iter->next_index; | |
1735 | if (!index && iter->index) | |
1736 | return NULL; | |
1737 | ||
21ef5339 | 1738 | restart: |
9e85d811 | 1739 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1740 | if (index > maxindex) |
1741 | return NULL; | |
8c1244de MW |
1742 | if (!child) |
1743 | return NULL; | |
21ef5339 | 1744 | |
8c1244de | 1745 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1746 | /* Single-slot tree */ |
21ef5339 RZ |
1747 | iter->index = index; |
1748 | iter->next_index = maxindex + 1; | |
78c1d784 | 1749 | iter->tags = 1; |
268f42de | 1750 | iter->node = NULL; |
8c1244de | 1751 | __set_iter_shift(iter, 0); |
d7b62727 | 1752 | return (void __rcu **)&root->rnode; |
8c1244de | 1753 | } |
21ef5339 | 1754 | |
8c1244de MW |
1755 | do { |
1756 | node = entry_to_node(child); | |
9e85d811 | 1757 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1758 | |
78c1d784 | 1759 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1760 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1761 | /* Hole detected */ |
1762 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1763 | return NULL; | |
1764 | ||
1765 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1766 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1767 | offset + 1); |
1768 | else | |
1769 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
12320d0f MW |
1770 | void *slot = rcu_dereference_raw( |
1771 | node->slots[offset]); | |
21ef5339 RZ |
1772 | if (is_sibling_entry(node, slot)) |
1773 | continue; | |
1774 | if (slot) | |
78c1d784 KK |
1775 | break; |
1776 | } | |
8c1244de | 1777 | index &= ~node_maxindex(node); |
9e85d811 | 1778 | index += offset << node->shift; |
78c1d784 KK |
1779 | /* Overflow after ~0UL */ |
1780 | if (!index) | |
1781 | return NULL; | |
1782 | if (offset == RADIX_TREE_MAP_SIZE) | |
1783 | goto restart; | |
8c1244de | 1784 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1785 | } |
1786 | ||
e157b555 | 1787 | if (!child) |
78c1d784 | 1788 | goto restart; |
e157b555 MW |
1789 | if (child == RADIX_TREE_RETRY) |
1790 | break; | |
66ee620f | 1791 | } while (node->shift && radix_tree_is_internal_node(child)); |
78c1d784 KK |
1792 | |
1793 | /* Update the iterator state */ | |
8c1244de MW |
1794 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1795 | iter->next_index = (index | node_maxindex(node)) + 1; | |
268f42de | 1796 | iter->node = node; |
9e85d811 | 1797 | __set_iter_shift(iter, node->shift); |
78c1d784 | 1798 | |
148deab2 MW |
1799 | if (flags & RADIX_TREE_ITER_TAGGED) |
1800 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1801 | |
1802 | return node->slots + offset; | |
1803 | } | |
1804 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1805 | ||
1da177e4 LT |
1806 | /** |
1807 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1808 | * @root: radix tree root | |
1809 | * @results: where the results of the lookup are placed | |
1810 | * @first_index: start the lookup from this key | |
1811 | * @max_items: place up to this many items at *results | |
1812 | * | |
1813 | * Performs an index-ascending scan of the tree for present items. Places | |
1814 | * them at *@results and returns the number of items which were placed at | |
1815 | * *@results. | |
1816 | * | |
1817 | * The implementation is naive. | |
7cf9c2c7 NP |
1818 | * |
1819 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1820 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1821 | * an atomic snapshot of the tree at a single point in time, the |
1822 | * semantics of an RCU protected gang lookup are as though multiple | |
1823 | * radix_tree_lookups have been issued in individual locks, and results | |
1824 | * stored in 'results'. | |
1da177e4 LT |
1825 | */ |
1826 | unsigned int | |
35534c86 | 1827 | radix_tree_gang_lookup(const struct radix_tree_root *root, void **results, |
1da177e4 LT |
1828 | unsigned long first_index, unsigned int max_items) |
1829 | { | |
cebbd29e | 1830 | struct radix_tree_iter iter; |
d7b62727 | 1831 | void __rcu **slot; |
cebbd29e | 1832 | unsigned int ret = 0; |
7cf9c2c7 | 1833 | |
cebbd29e | 1834 | if (unlikely(!max_items)) |
7cf9c2c7 | 1835 | return 0; |
1da177e4 | 1836 | |
cebbd29e | 1837 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1838 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1839 | if (!results[ret]) |
1840 | continue; | |
b194d16c | 1841 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1842 | slot = radix_tree_iter_retry(&iter); |
1843 | continue; | |
1844 | } | |
cebbd29e | 1845 | if (++ret == max_items) |
1da177e4 | 1846 | break; |
1da177e4 | 1847 | } |
7cf9c2c7 | 1848 | |
1da177e4 LT |
1849 | return ret; |
1850 | } | |
1851 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1852 | ||
47feff2c NP |
1853 | /** |
1854 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1855 | * @root: radix tree root | |
1856 | * @results: where the results of the lookup are placed | |
6328650b | 1857 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1858 | * @first_index: start the lookup from this key |
1859 | * @max_items: place up to this many items at *results | |
1860 | * | |
1861 | * Performs an index-ascending scan of the tree for present items. Places | |
1862 | * their slots at *@results and returns the number of items which were | |
1863 | * placed at *@results. | |
1864 | * | |
1865 | * The implementation is naive. | |
1866 | * | |
1867 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1868 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1869 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1870 | */ | |
1871 | unsigned int | |
35534c86 | 1872 | radix_tree_gang_lookup_slot(const struct radix_tree_root *root, |
d7b62727 | 1873 | void __rcu ***results, unsigned long *indices, |
47feff2c NP |
1874 | unsigned long first_index, unsigned int max_items) |
1875 | { | |
cebbd29e | 1876 | struct radix_tree_iter iter; |
d7b62727 | 1877 | void __rcu **slot; |
cebbd29e | 1878 | unsigned int ret = 0; |
47feff2c | 1879 | |
cebbd29e | 1880 | if (unlikely(!max_items)) |
47feff2c NP |
1881 | return 0; |
1882 | ||
cebbd29e KK |
1883 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1884 | results[ret] = slot; | |
6328650b | 1885 | if (indices) |
cebbd29e KK |
1886 | indices[ret] = iter.index; |
1887 | if (++ret == max_items) | |
47feff2c | 1888 | break; |
47feff2c NP |
1889 | } |
1890 | ||
1891 | return ret; | |
1892 | } | |
1893 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1894 | ||
1da177e4 LT |
1895 | /** |
1896 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1897 | * based on a tag | |
1898 | * @root: radix tree root | |
1899 | * @results: where the results of the lookup are placed | |
1900 | * @first_index: start the lookup from this key | |
1901 | * @max_items: place up to this many items at *results | |
daff89f3 | 1902 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1903 | * |
1904 | * Performs an index-ascending scan of the tree for present items which | |
1905 | * have the tag indexed by @tag set. Places the items at *@results and | |
1906 | * returns the number of items which were placed at *@results. | |
1907 | */ | |
1908 | unsigned int | |
35534c86 | 1909 | radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results, |
daff89f3 JC |
1910 | unsigned long first_index, unsigned int max_items, |
1911 | unsigned int tag) | |
1da177e4 | 1912 | { |
cebbd29e | 1913 | struct radix_tree_iter iter; |
d7b62727 | 1914 | void __rcu **slot; |
cebbd29e | 1915 | unsigned int ret = 0; |
612d6c19 | 1916 | |
cebbd29e | 1917 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1918 | return 0; |
1919 | ||
cebbd29e | 1920 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1921 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1922 | if (!results[ret]) |
1923 | continue; | |
b194d16c | 1924 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1925 | slot = radix_tree_iter_retry(&iter); |
1926 | continue; | |
1927 | } | |
cebbd29e | 1928 | if (++ret == max_items) |
1da177e4 | 1929 | break; |
1da177e4 | 1930 | } |
7cf9c2c7 | 1931 | |
1da177e4 LT |
1932 | return ret; |
1933 | } | |
1934 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1935 | ||
47feff2c NP |
1936 | /** |
1937 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1938 | * radix tree based on a tag | |
1939 | * @root: radix tree root | |
1940 | * @results: where the results of the lookup are placed | |
1941 | * @first_index: start the lookup from this key | |
1942 | * @max_items: place up to this many items at *results | |
1943 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1944 | * | |
1945 | * Performs an index-ascending scan of the tree for present items which | |
1946 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1947 | * returns the number of slots which were placed at *@results. | |
1948 | */ | |
1949 | unsigned int | |
35534c86 | 1950 | radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root, |
d7b62727 | 1951 | void __rcu ***results, unsigned long first_index, |
35534c86 | 1952 | unsigned int max_items, unsigned int tag) |
47feff2c | 1953 | { |
cebbd29e | 1954 | struct radix_tree_iter iter; |
d7b62727 | 1955 | void __rcu **slot; |
cebbd29e | 1956 | unsigned int ret = 0; |
47feff2c | 1957 | |
cebbd29e | 1958 | if (unlikely(!max_items)) |
47feff2c NP |
1959 | return 0; |
1960 | ||
cebbd29e KK |
1961 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1962 | results[ret] = slot; | |
1963 | if (++ret == max_items) | |
47feff2c | 1964 | break; |
47feff2c NP |
1965 | } |
1966 | ||
1967 | return ret; | |
1968 | } | |
1969 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1970 | ||
139e5616 JW |
1971 | /** |
1972 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1973 | * @root: radix tree root | |
139e5616 | 1974 | * @node: node containing @index |
ea07b862 | 1975 | * @update_node: callback for changing leaf nodes |
139e5616 JW |
1976 | * |
1977 | * After clearing the slot at @index in @node from radix tree | |
1978 | * rooted at @root, call this function to attempt freeing the | |
1979 | * node and shrinking the tree. | |
139e5616 | 1980 | */ |
14b46879 | 1981 | void __radix_tree_delete_node(struct radix_tree_root *root, |
ea07b862 | 1982 | struct radix_tree_node *node, |
c7df8ad2 | 1983 | radix_tree_update_node_t update_node) |
139e5616 | 1984 | { |
c7df8ad2 | 1985 | delete_node(root, node, update_node); |
139e5616 JW |
1986 | } |
1987 | ||
0ac398ef | 1988 | static bool __radix_tree_delete(struct radix_tree_root *root, |
d7b62727 | 1989 | struct radix_tree_node *node, void __rcu **slot) |
0ac398ef | 1990 | { |
0a835c4f | 1991 | void *old = rcu_dereference_raw(*slot); |
3159f943 | 1992 | int exceptional = xa_is_value(old) ? -1 : 0; |
0ac398ef MW |
1993 | unsigned offset = get_slot_offset(node, slot); |
1994 | int tag; | |
1995 | ||
0a835c4f MW |
1996 | if (is_idr(root)) |
1997 | node_tag_set(root, node, IDR_FREE, offset); | |
1998 | else | |
1999 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
2000 | node_tag_clear(root, node, tag, offset); | |
0ac398ef | 2001 | |
0a835c4f | 2002 | replace_slot(slot, NULL, node, -1, exceptional); |
c7df8ad2 | 2003 | return node && delete_node(root, node, NULL); |
0ac398ef MW |
2004 | } |
2005 | ||
1da177e4 | 2006 | /** |
0ac398ef MW |
2007 | * radix_tree_iter_delete - delete the entry at this iterator position |
2008 | * @root: radix tree root | |
2009 | * @iter: iterator state | |
2010 | * @slot: pointer to slot | |
1da177e4 | 2011 | * |
0ac398ef MW |
2012 | * Delete the entry at the position currently pointed to by the iterator. |
2013 | * This may result in the current node being freed; if it is, the iterator | |
2014 | * is advanced so that it will not reference the freed memory. This | |
2015 | * function may be called without any locking if there are no other threads | |
2016 | * which can access this tree. | |
2017 | */ | |
2018 | void radix_tree_iter_delete(struct radix_tree_root *root, | |
d7b62727 | 2019 | struct radix_tree_iter *iter, void __rcu **slot) |
0ac398ef MW |
2020 | { |
2021 | if (__radix_tree_delete(root, iter->node, slot)) | |
2022 | iter->index = iter->next_index; | |
2023 | } | |
d1b48c1e | 2024 | EXPORT_SYMBOL(radix_tree_iter_delete); |
0ac398ef MW |
2025 | |
2026 | /** | |
2027 | * radix_tree_delete_item - delete an item from a radix tree | |
2028 | * @root: radix tree root | |
2029 | * @index: index key | |
2030 | * @item: expected item | |
1da177e4 | 2031 | * |
0ac398ef | 2032 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 2033 | * |
0ac398ef MW |
2034 | * Return: the deleted entry, or %NULL if it was not present |
2035 | * or the entry at the given @index was not @item. | |
1da177e4 | 2036 | */ |
53c59f26 JW |
2037 | void *radix_tree_delete_item(struct radix_tree_root *root, |
2038 | unsigned long index, void *item) | |
1da177e4 | 2039 | { |
0a835c4f | 2040 | struct radix_tree_node *node = NULL; |
7a4deea1 | 2041 | void __rcu **slot = NULL; |
139e5616 | 2042 | void *entry; |
1da177e4 | 2043 | |
139e5616 | 2044 | entry = __radix_tree_lookup(root, index, &node, &slot); |
7a4deea1 MW |
2045 | if (!slot) |
2046 | return NULL; | |
0a835c4f MW |
2047 | if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE, |
2048 | get_slot_offset(node, slot)))) | |
139e5616 | 2049 | return NULL; |
1da177e4 | 2050 | |
139e5616 JW |
2051 | if (item && entry != item) |
2052 | return NULL; | |
2053 | ||
0ac398ef | 2054 | __radix_tree_delete(root, node, slot); |
612d6c19 | 2055 | |
139e5616 | 2056 | return entry; |
1da177e4 | 2057 | } |
53c59f26 JW |
2058 | EXPORT_SYMBOL(radix_tree_delete_item); |
2059 | ||
2060 | /** | |
0ac398ef MW |
2061 | * radix_tree_delete - delete an entry from a radix tree |
2062 | * @root: radix tree root | |
2063 | * @index: index key | |
53c59f26 | 2064 | * |
0ac398ef | 2065 | * Remove the entry at @index from the radix tree rooted at @root. |
53c59f26 | 2066 | * |
0ac398ef | 2067 | * Return: The deleted entry, or %NULL if it was not present. |
53c59f26 JW |
2068 | */ |
2069 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
2070 | { | |
2071 | return radix_tree_delete_item(root, index, NULL); | |
2072 | } | |
1da177e4 LT |
2073 | EXPORT_SYMBOL(radix_tree_delete); |
2074 | ||
d3798ae8 JW |
2075 | void radix_tree_clear_tags(struct radix_tree_root *root, |
2076 | struct radix_tree_node *node, | |
d7b62727 | 2077 | void __rcu **slot) |
d604c324 | 2078 | { |
d604c324 MW |
2079 | if (node) { |
2080 | unsigned int tag, offset = get_slot_offset(node, slot); | |
2081 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
2082 | node_tag_clear(root, node, tag, offset); | |
2083 | } else { | |
0a835c4f | 2084 | root_tag_clear_all(root); |
d604c324 | 2085 | } |
d604c324 MW |
2086 | } |
2087 | ||
1da177e4 LT |
2088 | /** |
2089 | * radix_tree_tagged - test whether any items in the tree are tagged | |
2090 | * @root: radix tree root | |
2091 | * @tag: tag to test | |
2092 | */ | |
35534c86 | 2093 | int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 2094 | { |
612d6c19 | 2095 | return root_tag_get(root, tag); |
1da177e4 LT |
2096 | } |
2097 | EXPORT_SYMBOL(radix_tree_tagged); | |
2098 | ||
0a835c4f MW |
2099 | /** |
2100 | * idr_preload - preload for idr_alloc() | |
2101 | * @gfp_mask: allocation mask to use for preloading | |
2102 | * | |
2103 | * Preallocate memory to use for the next call to idr_alloc(). This function | |
2104 | * returns with preemption disabled. It will be enabled by idr_preload_end(). | |
2105 | */ | |
2106 | void idr_preload(gfp_t gfp_mask) | |
2107 | { | |
bc9ae224 ED |
2108 | if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE)) |
2109 | preempt_disable(); | |
0a835c4f MW |
2110 | } |
2111 | EXPORT_SYMBOL(idr_preload); | |
2112 | ||
7ad3d4d8 MW |
2113 | int ida_pre_get(struct ida *ida, gfp_t gfp) |
2114 | { | |
7ad3d4d8 MW |
2115 | /* |
2116 | * The IDA API has no preload_end() equivalent. Instead, | |
2117 | * ida_get_new() can return -EAGAIN, prompting the caller | |
2118 | * to return to the ida_pre_get() step. | |
2119 | */ | |
bc9ae224 ED |
2120 | if (!__radix_tree_preload(gfp, IDA_PRELOAD_SIZE)) |
2121 | preempt_enable(); | |
7ad3d4d8 MW |
2122 | |
2123 | if (!this_cpu_read(ida_bitmap)) { | |
b1a8a7a7 | 2124 | struct ida_bitmap *bitmap = kzalloc(sizeof(*bitmap), gfp); |
7ad3d4d8 MW |
2125 | if (!bitmap) |
2126 | return 0; | |
4ecd9542 MW |
2127 | if (this_cpu_cmpxchg(ida_bitmap, NULL, bitmap)) |
2128 | kfree(bitmap); | |
7ad3d4d8 MW |
2129 | } |
2130 | ||
2131 | return 1; | |
2132 | } | |
7ad3d4d8 | 2133 | |
460488c5 | 2134 | void __rcu **idr_get_free(struct radix_tree_root *root, |
388f79fd CM |
2135 | struct radix_tree_iter *iter, gfp_t gfp, |
2136 | unsigned long max) | |
0a835c4f MW |
2137 | { |
2138 | struct radix_tree_node *node = NULL, *child; | |
d7b62727 | 2139 | void __rcu **slot = (void __rcu **)&root->rnode; |
0a835c4f | 2140 | unsigned long maxindex, start = iter->next_index; |
0a835c4f MW |
2141 | unsigned int shift, offset = 0; |
2142 | ||
2143 | grow: | |
2144 | shift = radix_tree_load_root(root, &child, &maxindex); | |
2145 | if (!radix_tree_tagged(root, IDR_FREE)) | |
2146 | start = max(start, maxindex + 1); | |
2147 | if (start > max) | |
2148 | return ERR_PTR(-ENOSPC); | |
2149 | ||
2150 | if (start > maxindex) { | |
2151 | int error = radix_tree_extend(root, gfp, start, shift); | |
2152 | if (error < 0) | |
2153 | return ERR_PTR(error); | |
2154 | shift = error; | |
2155 | child = rcu_dereference_raw(root->rnode); | |
2156 | } | |
66ee620f MW |
2157 | if (start == 0 && shift == 0) |
2158 | shift = RADIX_TREE_MAP_SHIFT; | |
0a835c4f MW |
2159 | |
2160 | while (shift) { | |
2161 | shift -= RADIX_TREE_MAP_SHIFT; | |
2162 | if (child == NULL) { | |
2163 | /* Have to add a child node. */ | |
d58275bc MW |
2164 | child = radix_tree_node_alloc(gfp, node, root, shift, |
2165 | offset, 0, 0); | |
0a835c4f MW |
2166 | if (!child) |
2167 | return ERR_PTR(-ENOMEM); | |
2168 | all_tag_set(child, IDR_FREE); | |
2169 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
2170 | if (node) | |
2171 | node->count++; | |
2172 | } else if (!radix_tree_is_internal_node(child)) | |
2173 | break; | |
2174 | ||
2175 | node = entry_to_node(child); | |
2176 | offset = radix_tree_descend(node, &child, start); | |
2177 | if (!tag_get(node, IDR_FREE, offset)) { | |
2178 | offset = radix_tree_find_next_bit(node, IDR_FREE, | |
2179 | offset + 1); | |
2180 | start = next_index(start, node, offset); | |
2181 | if (start > max) | |
2182 | return ERR_PTR(-ENOSPC); | |
2183 | while (offset == RADIX_TREE_MAP_SIZE) { | |
2184 | offset = node->offset + 1; | |
2185 | node = node->parent; | |
2186 | if (!node) | |
2187 | goto grow; | |
2188 | shift = node->shift; | |
2189 | } | |
2190 | child = rcu_dereference_raw(node->slots[offset]); | |
2191 | } | |
2192 | slot = &node->slots[offset]; | |
2193 | } | |
2194 | ||
2195 | iter->index = start; | |
2196 | if (node) | |
2197 | iter->next_index = 1 + min(max, (start | node_maxindex(node))); | |
2198 | else | |
2199 | iter->next_index = 1; | |
2200 | iter->node = node; | |
2201 | __set_iter_shift(iter, shift); | |
2202 | set_iter_tags(iter, node, offset, IDR_FREE); | |
2203 | ||
2204 | return slot; | |
2205 | } | |
2206 | ||
2207 | /** | |
2208 | * idr_destroy - release all internal memory from an IDR | |
2209 | * @idr: idr handle | |
2210 | * | |
2211 | * After this function is called, the IDR is empty, and may be reused or | |
2212 | * the data structure containing it may be freed. | |
2213 | * | |
2214 | * A typical clean-up sequence for objects stored in an idr tree will use | |
2215 | * idr_for_each() to free all objects, if necessary, then idr_destroy() to | |
2216 | * free the memory used to keep track of those objects. | |
2217 | */ | |
2218 | void idr_destroy(struct idr *idr) | |
2219 | { | |
2220 | struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.rnode); | |
2221 | if (radix_tree_is_internal_node(node)) | |
2222 | radix_tree_free_nodes(node); | |
2223 | idr->idr_rt.rnode = NULL; | |
2224 | root_tag_set(&idr->idr_rt, IDR_FREE); | |
2225 | } | |
2226 | EXPORT_SYMBOL(idr_destroy); | |
2227 | ||
1da177e4 | 2228 | static void |
449dd698 | 2229 | radix_tree_node_ctor(void *arg) |
1da177e4 | 2230 | { |
449dd698 JW |
2231 | struct radix_tree_node *node = arg; |
2232 | ||
2233 | memset(node, 0, sizeof(*node)); | |
2234 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
2235 | } |
2236 | ||
c78c66d1 KS |
2237 | static __init unsigned long __maxindex(unsigned int height) |
2238 | { | |
2239 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
2240 | int shift = RADIX_TREE_INDEX_BITS - width; | |
2241 | ||
2242 | if (shift < 0) | |
2243 | return ~0UL; | |
2244 | if (shift >= BITS_PER_LONG) | |
2245 | return 0UL; | |
2246 | return ~0UL >> shift; | |
2247 | } | |
2248 | ||
2249 | static __init void radix_tree_init_maxnodes(void) | |
2250 | { | |
2251 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
2252 | unsigned int i, j; | |
2253 | ||
2254 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
2255 | height_to_maxindex[i] = __maxindex(i); | |
2256 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
2257 | for (j = i; j > 0; j--) | |
2258 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
2259 | } | |
2260 | } | |
2261 | ||
d544abd5 | 2262 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 2263 | { |
2fcd9005 MW |
2264 | struct radix_tree_preload *rtp; |
2265 | struct radix_tree_node *node; | |
2266 | ||
2267 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
2268 | rtp = &per_cpu(radix_tree_preloads, cpu); |
2269 | while (rtp->nr) { | |
2270 | node = rtp->nodes; | |
1293d5c5 | 2271 | rtp->nodes = node->parent; |
d544abd5 SAS |
2272 | kmem_cache_free(radix_tree_node_cachep, node); |
2273 | rtp->nr--; | |
2fcd9005 | 2274 | } |
7ad3d4d8 MW |
2275 | kfree(per_cpu(ida_bitmap, cpu)); |
2276 | per_cpu(ida_bitmap, cpu) = NULL; | |
d544abd5 | 2277 | return 0; |
1da177e4 | 2278 | } |
1da177e4 LT |
2279 | |
2280 | void __init radix_tree_init(void) | |
2281 | { | |
d544abd5 | 2282 | int ret; |
7e784422 MH |
2283 | |
2284 | BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32); | |
fa290cda | 2285 | BUILD_BUG_ON(ROOT_IS_IDR & ~GFP_ZONEMASK); |
1da177e4 LT |
2286 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
2287 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
2288 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
2289 | radix_tree_node_ctor); | |
c78c66d1 | 2290 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
2291 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
2292 | NULL, radix_tree_cpu_dead); | |
2293 | WARN_ON(ret < 0); | |
1da177e4 | 2294 | } |