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