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