]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001 Momchil Velikov | |
3 | * Portions Copyright (C) 2001 Christoph Hellwig | |
cde53535 | 4 | * Copyright (C) 2005 SGI, Christoph Lameter |
7cf9c2c7 | 5 | * Copyright (C) 2006 Nick Piggin |
78c1d784 | 6 | * Copyright (C) 2012 Konstantin Khlebnikov |
6b053b8e MW |
7 | * Copyright (C) 2016 Intel, Matthew Wilcox |
8 | * Copyright (C) 2016 Intel, Ross Zwisler | |
1da177e4 LT |
9 | * |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License as | |
12 | * published by the Free Software Foundation; either version 2, or (at | |
13 | * your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, but | |
16 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with this program; if not, write to the Free Software | |
22 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
0a835c4f MW |
25 | #include <linux/bitmap.h> |
26 | #include <linux/bitops.h> | |
460488c5 | 27 | #include <linux/bug.h> |
e157b555 | 28 | #include <linux/cpu.h> |
1da177e4 | 29 | #include <linux/errno.h> |
0a835c4f MW |
30 | #include <linux/export.h> |
31 | #include <linux/idr.h> | |
1da177e4 LT |
32 | #include <linux/init.h> |
33 | #include <linux/kernel.h> | |
0a835c4f | 34 | #include <linux/kmemleak.h> |
1da177e4 | 35 | #include <linux/percpu.h> |
0a835c4f MW |
36 | #include <linux/preempt.h> /* in_interrupt() */ |
37 | #include <linux/radix-tree.h> | |
38 | #include <linux/rcupdate.h> | |
1da177e4 | 39 | #include <linux/slab.h> |
1da177e4 | 40 | #include <linux/string.h> |
02c02bf1 | 41 | #include <linux/xarray.h> |
1da177e4 LT |
42 | |
43 | ||
c78c66d1 KS |
44 | /* Number of nodes in fully populated tree of given height */ |
45 | static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
46 | ||
1da177e4 LT |
47 | /* |
48 | * Radix tree node cache. | |
49 | */ | |
e18b890b | 50 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 51 | |
55368052 NP |
52 | /* |
53 | * The radix tree is variable-height, so an insert operation not only has | |
54 | * to build the branch to its corresponding item, it also has to build the | |
55 | * branch to existing items if the size has to be increased (by | |
56 | * radix_tree_extend). | |
57 | * | |
58 | * The worst case is a zero height tree with just a single item at index 0, | |
59 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
60 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
61 | * Hence: | |
62 | */ | |
63 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
64 | ||
0a835c4f MW |
65 | /* |
66 | * The IDR does not have to be as high as the radix tree since it uses | |
67 | * signed integers, not unsigned longs. | |
68 | */ | |
69 | #define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1) | |
70 | #define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \ | |
71 | RADIX_TREE_MAP_SHIFT)) | |
72 | #define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1) | |
73 | ||
7ad3d4d8 MW |
74 | /* |
75 | * The IDA is even shorter since it uses a bitmap at the last level. | |
76 | */ | |
77 | #define IDA_INDEX_BITS (8 * sizeof(int) - 1 - ilog2(IDA_BITMAP_BITS)) | |
78 | #define IDA_MAX_PATH (DIV_ROUND_UP(IDA_INDEX_BITS, \ | |
79 | RADIX_TREE_MAP_SHIFT)) | |
80 | #define IDA_PRELOAD_SIZE (IDA_MAX_PATH * 2 - 1) | |
81 | ||
1da177e4 LT |
82 | /* |
83 | * Per-cpu pool of preloaded nodes | |
84 | */ | |
85 | struct radix_tree_preload { | |
2fcd9005 | 86 | unsigned nr; |
1293d5c5 | 87 | /* nodes->parent points to next preallocated node */ |
9d2a8da0 | 88 | struct radix_tree_node *nodes; |
1da177e4 | 89 | }; |
8cef7d57 | 90 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 91 | |
148deab2 MW |
92 | static inline struct radix_tree_node *entry_to_node(void *ptr) |
93 | { | |
94 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE); | |
95 | } | |
96 | ||
a4db4dce | 97 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 98 | { |
30ff46cc | 99 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
100 | } |
101 | ||
02c02bf1 | 102 | #define RADIX_TREE_RETRY XA_RETRY_ENTRY |
db050f29 | 103 | |
d7b62727 MW |
104 | static inline unsigned long |
105 | get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot) | |
db050f29 | 106 | { |
76f070b4 | 107 | return parent ? slot - parent->slots : 0; |
db050f29 MW |
108 | } |
109 | ||
35534c86 | 110 | static unsigned int radix_tree_descend(const struct radix_tree_node *parent, |
9e85d811 | 111 | struct radix_tree_node **nodep, unsigned long index) |
db050f29 | 112 | { |
9e85d811 | 113 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
d7b62727 | 114 | void __rcu **entry = rcu_dereference_raw(parent->slots[offset]); |
db050f29 | 115 | |
02c02bf1 MW |
116 | if (xa_is_sibling(entry)) { |
117 | offset = xa_to_sibling(entry); | |
118 | entry = rcu_dereference_raw(parent->slots[offset]); | |
db050f29 | 119 | } |
db050f29 MW |
120 | |
121 | *nodep = (void *)entry; | |
122 | return offset; | |
123 | } | |
124 | ||
35534c86 | 125 | static inline gfp_t root_gfp_mask(const struct radix_tree_root *root) |
612d6c19 | 126 | { |
f8d5d0cc | 127 | return root->xa_flags & (__GFP_BITS_MASK & ~GFP_ZONEMASK); |
612d6c19 NP |
128 | } |
129 | ||
643b52b9 NP |
130 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
131 | int offset) | |
132 | { | |
133 | __set_bit(offset, node->tags[tag]); | |
134 | } | |
135 | ||
136 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
137 | int offset) | |
138 | { | |
139 | __clear_bit(offset, node->tags[tag]); | |
140 | } | |
141 | ||
35534c86 | 142 | static inline int tag_get(const struct radix_tree_node *node, unsigned int tag, |
643b52b9 NP |
143 | int offset) |
144 | { | |
145 | return test_bit(offset, node->tags[tag]); | |
146 | } | |
147 | ||
35534c86 | 148 | static inline void root_tag_set(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 149 | { |
f8d5d0cc | 150 | root->xa_flags |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
151 | } |
152 | ||
2fcd9005 | 153 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 | 154 | { |
f8d5d0cc | 155 | root->xa_flags &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
156 | } |
157 | ||
158 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
159 | { | |
f8d5d0cc | 160 | root->xa_flags &= (__force gfp_t)((1 << ROOT_TAG_SHIFT) - 1); |
643b52b9 NP |
161 | } |
162 | ||
35534c86 | 163 | static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag) |
643b52b9 | 164 | { |
f8d5d0cc | 165 | return (__force int)root->xa_flags & (1 << (tag + ROOT_TAG_SHIFT)); |
643b52b9 NP |
166 | } |
167 | ||
35534c86 | 168 | static inline unsigned root_tags_get(const struct radix_tree_root *root) |
643b52b9 | 169 | { |
f8d5d0cc | 170 | return (__force unsigned)root->xa_flags >> ROOT_TAG_SHIFT; |
643b52b9 NP |
171 | } |
172 | ||
0a835c4f | 173 | static inline bool is_idr(const struct radix_tree_root *root) |
7b60e9ad | 174 | { |
f8d5d0cc | 175 | return !!(root->xa_flags & ROOT_IS_IDR); |
7b60e9ad MW |
176 | } |
177 | ||
643b52b9 NP |
178 | /* |
179 | * Returns 1 if any slot in the node has this tag set. | |
180 | * Otherwise returns 0. | |
181 | */ | |
35534c86 MW |
182 | static inline int any_tag_set(const struct radix_tree_node *node, |
183 | unsigned int tag) | |
643b52b9 | 184 | { |
2fcd9005 | 185 | unsigned idx; |
643b52b9 NP |
186 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
187 | if (node->tags[tag][idx]) | |
188 | return 1; | |
189 | } | |
190 | return 0; | |
191 | } | |
78c1d784 | 192 | |
0a835c4f MW |
193 | static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag) |
194 | { | |
195 | bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE); | |
196 | } | |
197 | ||
78c1d784 KK |
198 | /** |
199 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
200 | * | |
201 | * @addr: The address to base the search on | |
202 | * @size: The bitmap size in bits | |
203 | * @offset: The bitnumber to start searching at | |
204 | * | |
205 | * Unrollable variant of find_next_bit() for constant size arrays. | |
206 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
207 | * Returns next bit offset, or size if nothing found. | |
208 | */ | |
209 | static __always_inline unsigned long | |
bc412fca MW |
210 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
211 | unsigned long offset) | |
78c1d784 | 212 | { |
bc412fca | 213 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 214 | |
bc412fca | 215 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
216 | unsigned long tmp; |
217 | ||
218 | addr += offset / BITS_PER_LONG; | |
219 | tmp = *addr >> (offset % BITS_PER_LONG); | |
220 | if (tmp) | |
221 | return __ffs(tmp) + offset; | |
222 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 223 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
224 | tmp = *++addr; |
225 | if (tmp) | |
226 | return __ffs(tmp) + offset; | |
227 | offset += BITS_PER_LONG; | |
228 | } | |
229 | } | |
bc412fca | 230 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
231 | } |
232 | ||
268f42de MW |
233 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
234 | { | |
235 | return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK; | |
236 | } | |
237 | ||
218ed750 MW |
238 | /* |
239 | * The maximum index which can be stored in a radix tree | |
240 | */ | |
241 | static inline unsigned long shift_maxindex(unsigned int shift) | |
242 | { | |
243 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
244 | } | |
245 | ||
35534c86 | 246 | static inline unsigned long node_maxindex(const struct radix_tree_node *node) |
218ed750 MW |
247 | { |
248 | return shift_maxindex(node->shift); | |
249 | } | |
250 | ||
0a835c4f MW |
251 | static unsigned long next_index(unsigned long index, |
252 | const struct radix_tree_node *node, | |
253 | unsigned long offset) | |
254 | { | |
255 | return (index & ~node_maxindex(node)) + (offset << node->shift); | |
256 | } | |
257 | ||
0796c583 | 258 | #ifndef __KERNEL__ |
0a835c4f MW |
259 | static void dump_ida_node(void *entry, unsigned long index) |
260 | { | |
261 | unsigned long i; | |
262 | ||
263 | if (!entry) | |
264 | return; | |
265 | ||
266 | if (radix_tree_is_internal_node(entry)) { | |
267 | struct radix_tree_node *node = entry_to_node(entry); | |
268 | ||
269 | pr_debug("ida node: %p offset %d indices %lu-%lu parent %p free %lx shift %d count %d\n", | |
270 | node, node->offset, index * IDA_BITMAP_BITS, | |
271 | ((index | node_maxindex(node)) + 1) * | |
272 | IDA_BITMAP_BITS - 1, | |
273 | node->parent, node->tags[0][0], node->shift, | |
274 | node->count); | |
275 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) | |
276 | dump_ida_node(node->slots[i], | |
277 | index | (i << node->shift)); | |
3159f943 | 278 | } else if (xa_is_value(entry)) { |
d37cacc5 MW |
279 | pr_debug("ida excp: %p offset %d indices %lu-%lu data %lx\n", |
280 | entry, (int)(index & RADIX_TREE_MAP_MASK), | |
281 | index * IDA_BITMAP_BITS, | |
3159f943 MW |
282 | index * IDA_BITMAP_BITS + BITS_PER_XA_VALUE, |
283 | xa_to_value(entry)); | |
0a835c4f MW |
284 | } else { |
285 | struct ida_bitmap *bitmap = entry; | |
286 | ||
287 | pr_debug("ida btmp: %p offset %d indices %lu-%lu data", bitmap, | |
288 | (int)(index & RADIX_TREE_MAP_MASK), | |
289 | index * IDA_BITMAP_BITS, | |
290 | (index + 1) * IDA_BITMAP_BITS - 1); | |
291 | for (i = 0; i < IDA_BITMAP_LONGS; i++) | |
292 | pr_cont(" %lx", bitmap->bitmap[i]); | |
293 | pr_cont("\n"); | |
294 | } | |
295 | } | |
296 | ||
297 | static void ida_dump(struct ida *ida) | |
298 | { | |
299 | struct radix_tree_root *root = &ida->ida_rt; | |
f8d5d0cc MW |
300 | pr_debug("ida: %p node %p free %d\n", ida, root->xa_head, |
301 | root->xa_flags >> ROOT_TAG_SHIFT); | |
302 | dump_ida_node(root->xa_head, 0); | |
0a835c4f | 303 | } |
7cf19af4 MW |
304 | #endif |
305 | ||
1da177e4 LT |
306 | /* |
307 | * This assumes that the caller has performed appropriate preallocation, and | |
308 | * that the caller has pinned this thread of control to the current CPU. | |
309 | */ | |
310 | static struct radix_tree_node * | |
0a835c4f | 311 | radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent, |
d58275bc | 312 | struct radix_tree_root *root, |
e8de4340 | 313 | unsigned int shift, unsigned int offset, |
01959dfe | 314 | unsigned int count, unsigned int nr_values) |
1da177e4 | 315 | { |
e2848a0e | 316 | struct radix_tree_node *ret = NULL; |
1da177e4 | 317 | |
5e4c0d97 | 318 | /* |
2fcd9005 MW |
319 | * Preload code isn't irq safe and it doesn't make sense to use |
320 | * preloading during an interrupt anyway as all the allocations have | |
321 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 322 | */ |
d0164adc | 323 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
324 | struct radix_tree_preload *rtp; |
325 | ||
58e698af VD |
326 | /* |
327 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
328 | * cache first for the new node to get accounted to the memory |
329 | * cgroup. | |
58e698af VD |
330 | */ |
331 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 332 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
333 | if (ret) |
334 | goto out; | |
335 | ||
e2848a0e NP |
336 | /* |
337 | * Provided the caller has preloaded here, we will always | |
338 | * succeed in getting a node here (and never reach | |
339 | * kmem_cache_alloc) | |
340 | */ | |
7c8e0181 | 341 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 342 | if (rtp->nr) { |
9d2a8da0 | 343 | ret = rtp->nodes; |
1293d5c5 | 344 | rtp->nodes = ret->parent; |
1da177e4 LT |
345 | rtp->nr--; |
346 | } | |
ce80b067 CM |
347 | /* |
348 | * Update the allocation stack trace as this is more useful | |
349 | * for debugging. | |
350 | */ | |
351 | kmemleak_update_trace(ret); | |
58e698af | 352 | goto out; |
1da177e4 | 353 | } |
05eb6e72 | 354 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 355 | out: |
b194d16c | 356 | BUG_ON(radix_tree_is_internal_node(ret)); |
e8de4340 | 357 | if (ret) { |
e8de4340 MW |
358 | ret->shift = shift; |
359 | ret->offset = offset; | |
360 | ret->count = count; | |
01959dfe | 361 | ret->nr_values = nr_values; |
d58275bc | 362 | ret->parent = parent; |
01959dfe | 363 | ret->array = root; |
e8de4340 | 364 | } |
1da177e4 LT |
365 | return ret; |
366 | } | |
367 | ||
7cf9c2c7 NP |
368 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
369 | { | |
370 | struct radix_tree_node *node = | |
371 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
372 | |
373 | /* | |
175542f5 MW |
374 | * Must only free zeroed nodes into the slab. We can be left with |
375 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
376 | * and tags here. | |
643b52b9 | 377 | */ |
175542f5 MW |
378 | memset(node->slots, 0, sizeof(node->slots)); |
379 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 380 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 381 | |
7cf9c2c7 NP |
382 | kmem_cache_free(radix_tree_node_cachep, node); |
383 | } | |
384 | ||
1da177e4 LT |
385 | static inline void |
386 | radix_tree_node_free(struct radix_tree_node *node) | |
387 | { | |
7cf9c2c7 | 388 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
389 | } |
390 | ||
391 | /* | |
392 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
393 | * ensure that the addition of a single element in the tree cannot fail. On | |
394 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
395 | * with preemption not disabled. | |
b34df792 DH |
396 | * |
397 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 398 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 399 | */ |
bc9ae224 | 400 | static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr) |
1da177e4 LT |
401 | { |
402 | struct radix_tree_preload *rtp; | |
403 | struct radix_tree_node *node; | |
404 | int ret = -ENOMEM; | |
405 | ||
05eb6e72 VD |
406 | /* |
407 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
408 | * they should never be accounted to any particular memory cgroup. | |
409 | */ | |
410 | gfp_mask &= ~__GFP_ACCOUNT; | |
411 | ||
1da177e4 | 412 | preempt_disable(); |
7c8e0181 | 413 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 414 | while (rtp->nr < nr) { |
1da177e4 | 415 | preempt_enable(); |
488514d1 | 416 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
417 | if (node == NULL) |
418 | goto out; | |
419 | preempt_disable(); | |
7c8e0181 | 420 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 421 | if (rtp->nr < nr) { |
1293d5c5 | 422 | node->parent = rtp->nodes; |
9d2a8da0 KS |
423 | rtp->nodes = node; |
424 | rtp->nr++; | |
425 | } else { | |
1da177e4 | 426 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 427 | } |
1da177e4 LT |
428 | } |
429 | ret = 0; | |
430 | out: | |
431 | return ret; | |
432 | } | |
5e4c0d97 JK |
433 | |
434 | /* | |
435 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
436 | * ensure that the addition of a single element in the tree cannot fail. On | |
437 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
438 | * with preemption not disabled. | |
439 | * | |
440 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 441 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
442 | */ |
443 | int radix_tree_preload(gfp_t gfp_mask) | |
444 | { | |
445 | /* Warn on non-sensical use... */ | |
d0164adc | 446 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 447 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 448 | } |
d7f0923d | 449 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 450 | |
5e4c0d97 JK |
451 | /* |
452 | * The same as above function, except we don't guarantee preloading happens. | |
453 | * We do it, if we decide it helps. On success, return zero with preemption | |
454 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
455 | */ | |
456 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
457 | { | |
d0164adc | 458 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 459 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
460 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
461 | preempt_disable(); | |
462 | return 0; | |
463 | } | |
464 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
465 | ||
2791653a MW |
466 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
467 | /* | |
468 | * Preload with enough objects to ensure that we can split a single entry | |
469 | * of order @old_order into many entries of size @new_order | |
470 | */ | |
471 | int radix_tree_split_preload(unsigned int old_order, unsigned int new_order, | |
472 | gfp_t gfp_mask) | |
473 | { | |
474 | unsigned top = 1 << (old_order % RADIX_TREE_MAP_SHIFT); | |
475 | unsigned layers = (old_order / RADIX_TREE_MAP_SHIFT) - | |
476 | (new_order / RADIX_TREE_MAP_SHIFT); | |
477 | unsigned nr = 0; | |
478 | ||
479 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); | |
480 | BUG_ON(new_order >= old_order); | |
481 | ||
482 | while (layers--) | |
483 | nr = nr * RADIX_TREE_MAP_SIZE + 1; | |
484 | return __radix_tree_preload(gfp_mask, top * nr); | |
485 | } | |
486 | #endif | |
487 | ||
c78c66d1 KS |
488 | /* |
489 | * The same as function above, but preload number of nodes required to insert | |
490 | * (1 << order) continuous naturally-aligned elements. | |
491 | */ | |
492 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
493 | { | |
494 | unsigned long nr_subtrees; | |
495 | int nr_nodes, subtree_height; | |
496 | ||
497 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
498 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
499 | preempt_disable(); | |
500 | return 0; | |
501 | } | |
502 | ||
503 | /* | |
504 | * Calculate number and height of fully populated subtrees it takes to | |
505 | * store (1 << order) elements. | |
506 | */ | |
507 | nr_subtrees = 1 << order; | |
508 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
509 | subtree_height++) | |
510 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
511 | ||
512 | /* | |
513 | * The worst case is zero height tree with a single item at index 0 and | |
514 | * then inserting items starting at ULONG_MAX - (1 << order). | |
515 | * | |
516 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
517 | * 0-index item. | |
518 | */ | |
519 | nr_nodes = RADIX_TREE_MAX_PATH; | |
520 | ||
521 | /* Plus branch to fully populated subtrees. */ | |
522 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
523 | ||
524 | /* Root node is shared. */ | |
525 | nr_nodes--; | |
526 | ||
527 | /* Plus nodes required to build subtrees. */ | |
528 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
529 | ||
530 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
531 | } | |
532 | ||
35534c86 | 533 | static unsigned radix_tree_load_root(const struct radix_tree_root *root, |
1456a439 MW |
534 | struct radix_tree_node **nodep, unsigned long *maxindex) |
535 | { | |
f8d5d0cc | 536 | struct radix_tree_node *node = rcu_dereference_raw(root->xa_head); |
1456a439 MW |
537 | |
538 | *nodep = node; | |
539 | ||
b194d16c | 540 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 541 | node = entry_to_node(node); |
1456a439 | 542 | *maxindex = node_maxindex(node); |
c12e51b0 | 543 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
544 | } |
545 | ||
546 | *maxindex = 0; | |
547 | return 0; | |
548 | } | |
549 | ||
1da177e4 LT |
550 | /* |
551 | * Extend a radix tree so it can store key @index. | |
552 | */ | |
0a835c4f | 553 | static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp, |
d0891265 | 554 | unsigned long index, unsigned int shift) |
1da177e4 | 555 | { |
d7b62727 | 556 | void *entry; |
d0891265 | 557 | unsigned int maxshift; |
1da177e4 LT |
558 | int tag; |
559 | ||
d0891265 MW |
560 | /* Figure out what the shift should be. */ |
561 | maxshift = shift; | |
562 | while (index > shift_maxindex(maxshift)) | |
563 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 564 | |
f8d5d0cc | 565 | entry = rcu_dereference_raw(root->xa_head); |
d7b62727 | 566 | if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE))) |
1da177e4 | 567 | goto out; |
1da177e4 | 568 | |
1da177e4 | 569 | do { |
0a835c4f | 570 | struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL, |
d58275bc | 571 | root, shift, 0, 1, 0); |
2fcd9005 | 572 | if (!node) |
1da177e4 LT |
573 | return -ENOMEM; |
574 | ||
0a835c4f MW |
575 | if (is_idr(root)) { |
576 | all_tag_set(node, IDR_FREE); | |
577 | if (!root_tag_get(root, IDR_FREE)) { | |
578 | tag_clear(node, IDR_FREE, 0); | |
579 | root_tag_set(root, IDR_FREE); | |
580 | } | |
581 | } else { | |
582 | /* Propagate the aggregated tag info to the new child */ | |
583 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
584 | if (root_tag_get(root, tag)) | |
585 | tag_set(node, tag, 0); | |
586 | } | |
1da177e4 LT |
587 | } |
588 | ||
d0891265 | 589 | BUG_ON(shift > BITS_PER_LONG); |
d7b62727 MW |
590 | if (radix_tree_is_internal_node(entry)) { |
591 | entry_to_node(entry)->parent = node; | |
3159f943 | 592 | } else if (xa_is_value(entry)) { |
01959dfe MW |
593 | /* Moving a value entry root->xa_head to a node */ |
594 | node->nr_values = 1; | |
f7942430 | 595 | } |
d7b62727 MW |
596 | /* |
597 | * entry was already in the radix tree, so we do not need | |
598 | * rcu_assign_pointer here | |
599 | */ | |
600 | node->slots[0] = (void __rcu *)entry; | |
601 | entry = node_to_entry(node); | |
f8d5d0cc | 602 | rcu_assign_pointer(root->xa_head, entry); |
d0891265 | 603 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 604 | } while (shift <= maxshift); |
1da177e4 | 605 | out: |
d0891265 | 606 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
607 | } |
608 | ||
f4b109c6 JW |
609 | /** |
610 | * radix_tree_shrink - shrink radix tree to minimum height | |
611 | * @root radix tree root | |
612 | */ | |
0ac398ef | 613 | static inline bool radix_tree_shrink(struct radix_tree_root *root, |
c7df8ad2 | 614 | radix_tree_update_node_t update_node) |
f4b109c6 | 615 | { |
0ac398ef MW |
616 | bool shrunk = false; |
617 | ||
f4b109c6 | 618 | for (;;) { |
f8d5d0cc | 619 | struct radix_tree_node *node = rcu_dereference_raw(root->xa_head); |
f4b109c6 JW |
620 | struct radix_tree_node *child; |
621 | ||
622 | if (!radix_tree_is_internal_node(node)) | |
623 | break; | |
624 | node = entry_to_node(node); | |
625 | ||
626 | /* | |
627 | * The candidate node has more than one child, or its child | |
628 | * is not at the leftmost slot, or the child is a multiorder | |
629 | * entry, we cannot shrink. | |
630 | */ | |
631 | if (node->count != 1) | |
632 | break; | |
12320d0f | 633 | child = rcu_dereference_raw(node->slots[0]); |
f4b109c6 JW |
634 | if (!child) |
635 | break; | |
636 | if (!radix_tree_is_internal_node(child) && node->shift) | |
637 | break; | |
638 | ||
66ee620f MW |
639 | /* |
640 | * For an IDR, we must not shrink entry 0 into the root in | |
641 | * case somebody calls idr_replace() with a pointer that | |
642 | * appears to be an internal entry | |
643 | */ | |
644 | if (!node->shift && is_idr(root)) | |
645 | break; | |
646 | ||
f4b109c6 JW |
647 | if (radix_tree_is_internal_node(child)) |
648 | entry_to_node(child)->parent = NULL; | |
649 | ||
650 | /* | |
651 | * We don't need rcu_assign_pointer(), since we are simply | |
652 | * moving the node from one part of the tree to another: if it | |
653 | * was safe to dereference the old pointer to it | |
654 | * (node->slots[0]), it will be safe to dereference the new | |
f8d5d0cc | 655 | * one (root->xa_head) as far as dependent read barriers go. |
f4b109c6 | 656 | */ |
f8d5d0cc | 657 | root->xa_head = (void __rcu *)child; |
0a835c4f MW |
658 | if (is_idr(root) && !tag_get(node, IDR_FREE, 0)) |
659 | root_tag_clear(root, IDR_FREE); | |
f4b109c6 JW |
660 | |
661 | /* | |
662 | * We have a dilemma here. The node's slot[0] must not be | |
663 | * NULLed in case there are concurrent lookups expecting to | |
664 | * find the item. However if this was a bottom-level node, | |
665 | * then it may be subject to the slot pointer being visible | |
666 | * to callers dereferencing it. If item corresponding to | |
667 | * slot[0] is subsequently deleted, these callers would expect | |
668 | * their slot to become empty sooner or later. | |
669 | * | |
670 | * For example, lockless pagecache will look up a slot, deref | |
671 | * the page pointer, and if the page has 0 refcount it means it | |
672 | * was concurrently deleted from pagecache so try the deref | |
673 | * again. Fortunately there is already a requirement for logic | |
674 | * to retry the entire slot lookup -- the indirect pointer | |
675 | * problem (replacing direct root node with an indirect pointer | |
676 | * also results in a stale slot). So tag the slot as indirect | |
677 | * to force callers to retry. | |
678 | */ | |
4d693d08 JW |
679 | node->count = 0; |
680 | if (!radix_tree_is_internal_node(child)) { | |
d7b62727 | 681 | node->slots[0] = (void __rcu *)RADIX_TREE_RETRY; |
4d693d08 | 682 | if (update_node) |
c7df8ad2 | 683 | update_node(node); |
4d693d08 | 684 | } |
f4b109c6 | 685 | |
ea07b862 | 686 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 687 | radix_tree_node_free(node); |
0ac398ef | 688 | shrunk = true; |
f4b109c6 | 689 | } |
0ac398ef MW |
690 | |
691 | return shrunk; | |
f4b109c6 JW |
692 | } |
693 | ||
0ac398ef | 694 | static bool delete_node(struct radix_tree_root *root, |
4d693d08 | 695 | struct radix_tree_node *node, |
c7df8ad2 | 696 | radix_tree_update_node_t update_node) |
f4b109c6 | 697 | { |
0ac398ef MW |
698 | bool deleted = false; |
699 | ||
f4b109c6 JW |
700 | do { |
701 | struct radix_tree_node *parent; | |
702 | ||
703 | if (node->count) { | |
12320d0f | 704 | if (node_to_entry(node) == |
f8d5d0cc MW |
705 | rcu_dereference_raw(root->xa_head)) |
706 | deleted |= radix_tree_shrink(root, update_node); | |
0ac398ef | 707 | return deleted; |
f4b109c6 JW |
708 | } |
709 | ||
710 | parent = node->parent; | |
711 | if (parent) { | |
712 | parent->slots[node->offset] = NULL; | |
713 | parent->count--; | |
714 | } else { | |
0a835c4f MW |
715 | /* |
716 | * Shouldn't the tags already have all been cleared | |
717 | * by the caller? | |
718 | */ | |
719 | if (!is_idr(root)) | |
720 | root_tag_clear_all(root); | |
f8d5d0cc | 721 | root->xa_head = NULL; |
f4b109c6 JW |
722 | } |
723 | ||
ea07b862 | 724 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 725 | radix_tree_node_free(node); |
0ac398ef | 726 | deleted = true; |
f4b109c6 JW |
727 | |
728 | node = parent; | |
729 | } while (node); | |
0ac398ef MW |
730 | |
731 | return deleted; | |
f4b109c6 JW |
732 | } |
733 | ||
1da177e4 | 734 | /** |
139e5616 | 735 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
736 | * @root: radix tree root |
737 | * @index: index key | |
e6145236 | 738 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
739 | * @nodep: returns node |
740 | * @slotp: returns slot | |
1da177e4 | 741 | * |
139e5616 JW |
742 | * Create, if necessary, and return the node and slot for an item |
743 | * at position @index in the radix tree @root. | |
744 | * | |
745 | * Until there is more than one item in the tree, no nodes are | |
f8d5d0cc | 746 | * allocated and @root->xa_head is used as a direct slot instead of |
139e5616 JW |
747 | * pointing to a node, in which case *@nodep will be NULL. |
748 | * | |
749 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 750 | */ |
139e5616 | 751 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 | 752 | unsigned order, struct radix_tree_node **nodep, |
d7b62727 | 753 | void __rcu ***slotp) |
1da177e4 | 754 | { |
89148aa4 | 755 | struct radix_tree_node *node = NULL, *child; |
f8d5d0cc | 756 | void __rcu **slot = (void __rcu **)&root->xa_head; |
49ea6ebc | 757 | unsigned long maxindex; |
89148aa4 | 758 | unsigned int shift, offset = 0; |
49ea6ebc | 759 | unsigned long max = index | ((1UL << order) - 1); |
0a835c4f | 760 | gfp_t gfp = root_gfp_mask(root); |
49ea6ebc | 761 | |
89148aa4 | 762 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
763 | |
764 | /* Make sure the tree is high enough. */ | |
175542f5 MW |
765 | if (order > 0 && max == ((1UL << order) - 1)) |
766 | max++; | |
49ea6ebc | 767 | if (max > maxindex) { |
0a835c4f | 768 | int error = radix_tree_extend(root, gfp, max, shift); |
49ea6ebc | 769 | if (error < 0) |
1da177e4 | 770 | return error; |
49ea6ebc | 771 | shift = error; |
f8d5d0cc | 772 | child = rcu_dereference_raw(root->xa_head); |
1da177e4 LT |
773 | } |
774 | ||
e6145236 | 775 | while (shift > order) { |
c12e51b0 | 776 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 777 | if (child == NULL) { |
1da177e4 | 778 | /* Have to add a child node. */ |
d58275bc | 779 | child = radix_tree_node_alloc(gfp, node, root, shift, |
e8de4340 | 780 | offset, 0, 0); |
89148aa4 | 781 | if (!child) |
1da177e4 | 782 | return -ENOMEM; |
89148aa4 MW |
783 | rcu_assign_pointer(*slot, node_to_entry(child)); |
784 | if (node) | |
1da177e4 | 785 | node->count++; |
89148aa4 | 786 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 787 | break; |
1da177e4 LT |
788 | |
789 | /* Go a level down */ | |
89148aa4 | 790 | node = entry_to_node(child); |
9e85d811 | 791 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 792 | slot = &node->slots[offset]; |
e6145236 MW |
793 | } |
794 | ||
175542f5 MW |
795 | if (nodep) |
796 | *nodep = node; | |
797 | if (slotp) | |
798 | *slotp = slot; | |
799 | return 0; | |
800 | } | |
801 | ||
175542f5 MW |
802 | /* |
803 | * Free any nodes below this node. The tree is presumed to not need | |
804 | * shrinking, and any user data in the tree is presumed to not need a | |
805 | * destructor called on it. If we need to add a destructor, we can | |
806 | * add that functionality later. Note that we may not clear tags or | |
807 | * slots from the tree as an RCU walker may still have a pointer into | |
808 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
809 | * but we'll still have to clear those in rcu_free. | |
810 | */ | |
811 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
812 | { | |
813 | unsigned offset = 0; | |
814 | struct radix_tree_node *child = entry_to_node(node); | |
815 | ||
816 | for (;;) { | |
12320d0f | 817 | void *entry = rcu_dereference_raw(child->slots[offset]); |
02c02bf1 | 818 | if (xa_is_node(entry) && child->shift) { |
175542f5 MW |
819 | child = entry_to_node(entry); |
820 | offset = 0; | |
821 | continue; | |
822 | } | |
823 | offset++; | |
824 | while (offset == RADIX_TREE_MAP_SIZE) { | |
825 | struct radix_tree_node *old = child; | |
826 | offset = child->offset + 1; | |
827 | child = child->parent; | |
dd040b6f | 828 | WARN_ON_ONCE(!list_empty(&old->private_list)); |
175542f5 MW |
829 | radix_tree_node_free(old); |
830 | if (old == entry_to_node(node)) | |
831 | return; | |
832 | } | |
833 | } | |
834 | } | |
835 | ||
0a835c4f | 836 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
d7b62727 MW |
837 | static inline int insert_entries(struct radix_tree_node *node, |
838 | void __rcu **slot, void *item, unsigned order, bool replace) | |
175542f5 | 839 | { |
02c02bf1 | 840 | void *sibling; |
175542f5 MW |
841 | unsigned i, n, tag, offset, tags = 0; |
842 | ||
843 | if (node) { | |
e157b555 MW |
844 | if (order > node->shift) |
845 | n = 1 << (order - node->shift); | |
846 | else | |
847 | n = 1; | |
175542f5 MW |
848 | offset = get_slot_offset(node, slot); |
849 | } else { | |
850 | n = 1; | |
851 | offset = 0; | |
852 | } | |
853 | ||
854 | if (n > 1) { | |
e6145236 | 855 | offset = offset & ~(n - 1); |
89148aa4 | 856 | slot = &node->slots[offset]; |
175542f5 | 857 | } |
02c02bf1 | 858 | sibling = xa_mk_sibling(offset); |
175542f5 MW |
859 | |
860 | for (i = 0; i < n; i++) { | |
861 | if (slot[i]) { | |
862 | if (replace) { | |
863 | node->count--; | |
864 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
865 | if (tag_get(node, tag, offset + i)) | |
866 | tags |= 1 << tag; | |
867 | } else | |
e6145236 MW |
868 | return -EEXIST; |
869 | } | |
175542f5 | 870 | } |
e6145236 | 871 | |
175542f5 | 872 | for (i = 0; i < n; i++) { |
12320d0f | 873 | struct radix_tree_node *old = rcu_dereference_raw(slot[i]); |
175542f5 | 874 | if (i) { |
02c02bf1 | 875 | rcu_assign_pointer(slot[i], sibling); |
175542f5 MW |
876 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
877 | if (tags & (1 << tag)) | |
878 | tag_clear(node, tag, offset + i); | |
879 | } else { | |
880 | rcu_assign_pointer(slot[i], item); | |
881 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
882 | if (tags & (1 << tag)) | |
883 | tag_set(node, tag, offset); | |
e6145236 | 884 | } |
02c02bf1 | 885 | if (xa_is_node(old)) |
175542f5 | 886 | radix_tree_free_nodes(old); |
3159f943 | 887 | if (xa_is_value(old)) |
01959dfe | 888 | node->nr_values--; |
612d6c19 | 889 | } |
175542f5 MW |
890 | if (node) { |
891 | node->count += n; | |
3159f943 | 892 | if (xa_is_value(item)) |
01959dfe | 893 | node->nr_values += n; |
175542f5 MW |
894 | } |
895 | return n; | |
139e5616 | 896 | } |
175542f5 | 897 | #else |
d7b62727 MW |
898 | static inline int insert_entries(struct radix_tree_node *node, |
899 | void __rcu **slot, void *item, unsigned order, bool replace) | |
175542f5 MW |
900 | { |
901 | if (*slot) | |
902 | return -EEXIST; | |
903 | rcu_assign_pointer(*slot, item); | |
904 | if (node) { | |
905 | node->count++; | |
3159f943 | 906 | if (xa_is_value(item)) |
01959dfe | 907 | node->nr_values++; |
175542f5 MW |
908 | } |
909 | return 1; | |
910 | } | |
911 | #endif | |
139e5616 JW |
912 | |
913 | /** | |
e6145236 | 914 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
915 | * @root: radix tree root |
916 | * @index: index key | |
e6145236 | 917 | * @order: key covers the 2^order indices around index |
139e5616 JW |
918 | * @item: item to insert |
919 | * | |
920 | * Insert an item into the radix tree at position @index. | |
921 | */ | |
e6145236 MW |
922 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
923 | unsigned order, void *item) | |
139e5616 JW |
924 | { |
925 | struct radix_tree_node *node; | |
d7b62727 | 926 | void __rcu **slot; |
139e5616 JW |
927 | int error; |
928 | ||
b194d16c | 929 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 930 | |
e6145236 | 931 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
932 | if (error) |
933 | return error; | |
175542f5 MW |
934 | |
935 | error = insert_entries(node, slot, item, order, false); | |
936 | if (error < 0) | |
937 | return error; | |
201b6264 | 938 | |
612d6c19 | 939 | if (node) { |
7b60e9ad | 940 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
941 | BUG_ON(tag_get(node, 0, offset)); |
942 | BUG_ON(tag_get(node, 1, offset)); | |
943 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 944 | } else { |
7b60e9ad | 945 | BUG_ON(root_tags_get(root)); |
612d6c19 | 946 | } |
1da177e4 | 947 | |
1da177e4 LT |
948 | return 0; |
949 | } | |
e6145236 | 950 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 951 | |
139e5616 JW |
952 | /** |
953 | * __radix_tree_lookup - lookup an item in a radix tree | |
954 | * @root: radix tree root | |
955 | * @index: index key | |
956 | * @nodep: returns node | |
957 | * @slotp: returns slot | |
958 | * | |
959 | * Lookup and return the item at position @index in the radix | |
960 | * tree @root. | |
961 | * | |
962 | * Until there is more than one item in the tree, no nodes are | |
f8d5d0cc | 963 | * allocated and @root->xa_head is used as a direct slot instead of |
139e5616 | 964 | * pointing to a node, in which case *@nodep will be NULL. |
7cf9c2c7 | 965 | */ |
35534c86 MW |
966 | void *__radix_tree_lookup(const struct radix_tree_root *root, |
967 | unsigned long index, struct radix_tree_node **nodep, | |
d7b62727 | 968 | void __rcu ***slotp) |
1da177e4 | 969 | { |
139e5616 | 970 | struct radix_tree_node *node, *parent; |
85829954 | 971 | unsigned long maxindex; |
d7b62727 | 972 | void __rcu **slot; |
612d6c19 | 973 | |
85829954 MW |
974 | restart: |
975 | parent = NULL; | |
f8d5d0cc | 976 | slot = (void __rcu **)&root->xa_head; |
9e85d811 | 977 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 978 | if (index > maxindex) |
1da177e4 LT |
979 | return NULL; |
980 | ||
b194d16c | 981 | while (radix_tree_is_internal_node(node)) { |
85829954 | 982 | unsigned offset; |
1da177e4 | 983 | |
85829954 MW |
984 | if (node == RADIX_TREE_RETRY) |
985 | goto restart; | |
4dd6c098 | 986 | parent = entry_to_node(node); |
9e85d811 | 987 | offset = radix_tree_descend(parent, &node, index); |
85829954 | 988 | slot = parent->slots + offset; |
66ee620f MW |
989 | if (parent->shift == 0) |
990 | break; | |
85829954 | 991 | } |
1da177e4 | 992 | |
139e5616 JW |
993 | if (nodep) |
994 | *nodep = parent; | |
995 | if (slotp) | |
996 | *slotp = slot; | |
997 | return node; | |
b72b71c6 HS |
998 | } |
999 | ||
1000 | /** | |
1001 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
1002 | * @root: radix tree root | |
1003 | * @index: index key | |
1004 | * | |
1005 | * Returns: the slot corresponding to the position @index in the | |
1006 | * radix tree @root. This is useful for update-if-exists operations. | |
1007 | * | |
1008 | * This function can be called under rcu_read_lock iff the slot is not | |
1009 | * modified by radix_tree_replace_slot, otherwise it must be called | |
1010 | * exclusive from other writers. Any dereference of the slot must be done | |
1011 | * using radix_tree_deref_slot. | |
1012 | */ | |
d7b62727 | 1013 | void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root, |
35534c86 | 1014 | unsigned long index) |
b72b71c6 | 1015 | { |
d7b62727 | 1016 | void __rcu **slot; |
139e5616 JW |
1017 | |
1018 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
1019 | return NULL; | |
1020 | return slot; | |
a4331366 | 1021 | } |
a4331366 HR |
1022 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
1023 | ||
1024 | /** | |
1025 | * radix_tree_lookup - perform lookup operation on a radix tree | |
1026 | * @root: radix tree root | |
1027 | * @index: index key | |
1028 | * | |
1029 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
1030 | * |
1031 | * This function can be called under rcu_read_lock, however the caller | |
1032 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
1033 | * them safely). No RCU barriers are required to access or modify the | |
1034 | * returned item, however. | |
a4331366 | 1035 | */ |
35534c86 | 1036 | void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index) |
a4331366 | 1037 | { |
139e5616 | 1038 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
1039 | } |
1040 | EXPORT_SYMBOL(radix_tree_lookup); | |
1041 | ||
0a835c4f | 1042 | static inline void replace_sibling_entries(struct radix_tree_node *node, |
01959dfe | 1043 | void __rcu **slot, int count, int values) |
a90eb3a2 | 1044 | { |
a90eb3a2 | 1045 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
02c02bf1 MW |
1046 | unsigned offset = get_slot_offset(node, slot); |
1047 | void *ptr = xa_mk_sibling(offset); | |
a90eb3a2 | 1048 | |
02c02bf1 | 1049 | while (++offset < RADIX_TREE_MAP_SIZE) { |
12320d0f | 1050 | if (rcu_dereference_raw(node->slots[offset]) != ptr) |
a90eb3a2 | 1051 | break; |
0a835c4f MW |
1052 | if (count < 0) { |
1053 | node->slots[offset] = NULL; | |
1054 | node->count--; | |
1055 | } | |
01959dfe | 1056 | node->nr_values += values; |
a90eb3a2 MW |
1057 | } |
1058 | #endif | |
a90eb3a2 MW |
1059 | } |
1060 | ||
d7b62727 | 1061 | static void replace_slot(void __rcu **slot, void *item, |
01959dfe | 1062 | struct radix_tree_node *node, int count, int values) |
f7942430 | 1063 | { |
01959dfe | 1064 | if (node && (count || values)) { |
f4b109c6 | 1065 | node->count += count; |
01959dfe MW |
1066 | node->nr_values += values; |
1067 | replace_sibling_entries(node, slot, count, values); | |
f4b109c6 | 1068 | } |
f7942430 JW |
1069 | |
1070 | rcu_assign_pointer(*slot, item); | |
1071 | } | |
1072 | ||
0a835c4f MW |
1073 | static bool node_tag_get(const struct radix_tree_root *root, |
1074 | const struct radix_tree_node *node, | |
1075 | unsigned int tag, unsigned int offset) | |
a90eb3a2 | 1076 | { |
0a835c4f MW |
1077 | if (node) |
1078 | return tag_get(node, tag, offset); | |
1079 | return root_tag_get(root, tag); | |
1080 | } | |
a90eb3a2 | 1081 | |
0a835c4f MW |
1082 | /* |
1083 | * IDR users want to be able to store NULL in the tree, so if the slot isn't | |
1084 | * free, don't adjust the count, even if it's transitioning between NULL and | |
1085 | * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still | |
1086 | * have empty bits, but it only stores NULL in slots when they're being | |
1087 | * deleted. | |
1088 | */ | |
1089 | static int calculate_count(struct radix_tree_root *root, | |
d7b62727 | 1090 | struct radix_tree_node *node, void __rcu **slot, |
0a835c4f MW |
1091 | void *item, void *old) |
1092 | { | |
1093 | if (is_idr(root)) { | |
1094 | unsigned offset = get_slot_offset(node, slot); | |
1095 | bool free = node_tag_get(root, node, IDR_FREE, offset); | |
1096 | if (!free) | |
1097 | return 0; | |
1098 | if (!old) | |
1099 | return 1; | |
a90eb3a2 | 1100 | } |
0a835c4f | 1101 | return !!item - !!old; |
a90eb3a2 MW |
1102 | } |
1103 | ||
6d75f366 JW |
1104 | /** |
1105 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
1106 | * @root: radix tree root |
1107 | * @node: pointer to tree node | |
1108 | * @slot: pointer to slot in @node | |
1109 | * @item: new item to store in the slot. | |
1110 | * @update_node: callback for changing leaf nodes | |
6d75f366 JW |
1111 | * |
1112 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
1113 | * across slot lookup and replacement. | |
1114 | */ | |
1115 | void __radix_tree_replace(struct radix_tree_root *root, | |
1116 | struct radix_tree_node *node, | |
d7b62727 | 1117 | void __rcu **slot, void *item, |
c7df8ad2 | 1118 | radix_tree_update_node_t update_node) |
6d75f366 | 1119 | { |
0a835c4f | 1120 | void *old = rcu_dereference_raw(*slot); |
01959dfe | 1121 | int values = !!xa_is_value(item) - !!xa_is_value(old); |
0a835c4f MW |
1122 | int count = calculate_count(root, node, slot, item, old); |
1123 | ||
6d75f366 | 1124 | /* |
01959dfe | 1125 | * This function supports replacing value entries and |
f4b109c6 | 1126 | * deleting entries, but that needs accounting against the |
f8d5d0cc | 1127 | * node unless the slot is root->xa_head. |
6d75f366 | 1128 | */ |
f8d5d0cc | 1129 | WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->xa_head) && |
01959dfe MW |
1130 | (count || values)); |
1131 | replace_slot(slot, item, node, count, values); | |
f4b109c6 | 1132 | |
4d693d08 JW |
1133 | if (!node) |
1134 | return; | |
1135 | ||
1136 | if (update_node) | |
c7df8ad2 | 1137 | update_node(node); |
4d693d08 | 1138 | |
c7df8ad2 | 1139 | delete_node(root, node, update_node); |
6d75f366 JW |
1140 | } |
1141 | ||
1142 | /** | |
1143 | * radix_tree_replace_slot - replace item in a slot | |
1144 | * @root: radix tree root | |
1145 | * @slot: pointer to slot | |
1146 | * @item: new item to store in the slot. | |
1147 | * | |
1148 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
1149 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
1150 | * across slot lookup and replacement. | |
1151 | * | |
1152 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
01959dfe | 1153 | * regular entries, and value entries, as that requires accounting |
f4b109c6 | 1154 | * inside the radix tree node. When switching from one type of entry or |
e157b555 MW |
1155 | * deleting, use __radix_tree_lookup() and __radix_tree_replace() or |
1156 | * radix_tree_iter_replace(). | |
6d75f366 JW |
1157 | */ |
1158 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
d7b62727 | 1159 | void __rcu **slot, void *item) |
6d75f366 | 1160 | { |
c7df8ad2 | 1161 | __radix_tree_replace(root, NULL, slot, item, NULL); |
6d75f366 | 1162 | } |
10257d71 | 1163 | EXPORT_SYMBOL(radix_tree_replace_slot); |
6d75f366 | 1164 | |
e157b555 MW |
1165 | /** |
1166 | * radix_tree_iter_replace - replace item in a slot | |
1167 | * @root: radix tree root | |
1168 | * @slot: pointer to slot | |
1169 | * @item: new item to store in the slot. | |
1170 | * | |
1171 | * For use with radix_tree_split() and radix_tree_for_each_slot(). | |
1172 | * Caller must hold tree write locked across split and replacement. | |
1173 | */ | |
1174 | void radix_tree_iter_replace(struct radix_tree_root *root, | |
d7b62727 MW |
1175 | const struct radix_tree_iter *iter, |
1176 | void __rcu **slot, void *item) | |
e157b555 | 1177 | { |
c7df8ad2 | 1178 | __radix_tree_replace(root, iter->node, slot, item, NULL); |
e157b555 MW |
1179 | } |
1180 | ||
175542f5 MW |
1181 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
1182 | /** | |
1183 | * radix_tree_join - replace multiple entries with one multiorder entry | |
1184 | * @root: radix tree root | |
1185 | * @index: an index inside the new entry | |
1186 | * @order: order of the new entry | |
1187 | * @item: new entry | |
1188 | * | |
1189 | * Call this function to replace several entries with one larger entry. | |
1190 | * The existing entries are presumed to not need freeing as a result of | |
1191 | * this call. | |
1192 | * | |
1193 | * The replacement entry will have all the tags set on it that were set | |
1194 | * on any of the entries it is replacing. | |
1195 | */ | |
1196 | int radix_tree_join(struct radix_tree_root *root, unsigned long index, | |
1197 | unsigned order, void *item) | |
1198 | { | |
1199 | struct radix_tree_node *node; | |
d7b62727 | 1200 | void __rcu **slot; |
175542f5 MW |
1201 | int error; |
1202 | ||
1203 | BUG_ON(radix_tree_is_internal_node(item)); | |
1204 | ||
1205 | error = __radix_tree_create(root, index, order, &node, &slot); | |
1206 | if (!error) | |
1207 | error = insert_entries(node, slot, item, order, true); | |
1208 | if (error > 0) | |
1209 | error = 0; | |
1210 | ||
1211 | return error; | |
1212 | } | |
e157b555 MW |
1213 | |
1214 | /** | |
1215 | * radix_tree_split - Split an entry into smaller entries | |
1216 | * @root: radix tree root | |
1217 | * @index: An index within the large entry | |
1218 | * @order: Order of new entries | |
1219 | * | |
1220 | * Call this function as the first step in replacing a multiorder entry | |
1221 | * with several entries of lower order. After this function returns, | |
1222 | * loop over the relevant portion of the tree using radix_tree_for_each_slot() | |
1223 | * and call radix_tree_iter_replace() to set up each new entry. | |
1224 | * | |
1225 | * The tags from this entry are replicated to all the new entries. | |
1226 | * | |
1227 | * The radix tree should be locked against modification during the entire | |
1228 | * replacement operation. Lock-free lookups will see RADIX_TREE_RETRY which | |
1229 | * should prompt RCU walkers to restart the lookup from the root. | |
1230 | */ | |
1231 | int radix_tree_split(struct radix_tree_root *root, unsigned long index, | |
1232 | unsigned order) | |
1233 | { | |
1234 | struct radix_tree_node *parent, *node, *child; | |
d7b62727 | 1235 | void __rcu **slot; |
e157b555 MW |
1236 | unsigned int offset, end; |
1237 | unsigned n, tag, tags = 0; | |
0a835c4f | 1238 | gfp_t gfp = root_gfp_mask(root); |
e157b555 MW |
1239 | |
1240 | if (!__radix_tree_lookup(root, index, &parent, &slot)) | |
1241 | return -ENOENT; | |
1242 | if (!parent) | |
1243 | return -ENOENT; | |
1244 | ||
1245 | offset = get_slot_offset(parent, slot); | |
1246 | ||
1247 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1248 | if (tag_get(parent, tag, offset)) | |
1249 | tags |= 1 << tag; | |
1250 | ||
1251 | for (end = offset + 1; end < RADIX_TREE_MAP_SIZE; end++) { | |
02c02bf1 | 1252 | if (!xa_is_sibling(rcu_dereference_raw(parent->slots[end]))) |
e157b555 MW |
1253 | break; |
1254 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1255 | if (tags & (1 << tag)) | |
1256 | tag_set(parent, tag, end); | |
1257 | /* rcu_assign_pointer ensures tags are set before RETRY */ | |
1258 | rcu_assign_pointer(parent->slots[end], RADIX_TREE_RETRY); | |
1259 | } | |
1260 | rcu_assign_pointer(parent->slots[offset], RADIX_TREE_RETRY); | |
01959dfe | 1261 | parent->nr_values -= (end - offset); |
e157b555 MW |
1262 | |
1263 | if (order == parent->shift) | |
1264 | return 0; | |
1265 | if (order > parent->shift) { | |
1266 | while (offset < end) | |
1267 | offset += insert_entries(parent, &parent->slots[offset], | |
1268 | RADIX_TREE_RETRY, order, true); | |
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | node = parent; | |
1273 | ||
1274 | for (;;) { | |
1275 | if (node->shift > order) { | |
d58275bc | 1276 | child = radix_tree_node_alloc(gfp, node, root, |
e8de4340 MW |
1277 | node->shift - RADIX_TREE_MAP_SHIFT, |
1278 | offset, 0, 0); | |
e157b555 MW |
1279 | if (!child) |
1280 | goto nomem; | |
e157b555 MW |
1281 | if (node != parent) { |
1282 | node->count++; | |
12320d0f MW |
1283 | rcu_assign_pointer(node->slots[offset], |
1284 | node_to_entry(child)); | |
e157b555 MW |
1285 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
1286 | if (tags & (1 << tag)) | |
1287 | tag_set(node, tag, offset); | |
1288 | } | |
1289 | ||
1290 | node = child; | |
1291 | offset = 0; | |
1292 | continue; | |
1293 | } | |
1294 | ||
1295 | n = insert_entries(node, &node->slots[offset], | |
1296 | RADIX_TREE_RETRY, order, false); | |
1297 | BUG_ON(n > RADIX_TREE_MAP_SIZE); | |
1298 | ||
1299 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1300 | if (tags & (1 << tag)) | |
1301 | tag_set(node, tag, offset); | |
1302 | offset += n; | |
1303 | ||
1304 | while (offset == RADIX_TREE_MAP_SIZE) { | |
1305 | if (node == parent) | |
1306 | break; | |
1307 | offset = node->offset; | |
1308 | child = node; | |
1309 | node = node->parent; | |
1310 | rcu_assign_pointer(node->slots[offset], | |
1311 | node_to_entry(child)); | |
1312 | offset++; | |
1313 | } | |
1314 | if ((node == parent) && (offset == end)) | |
1315 | return 0; | |
1316 | } | |
1317 | ||
1318 | nomem: | |
1319 | /* Shouldn't happen; did user forget to preload? */ | |
1320 | /* TODO: free all the allocated nodes */ | |
1321 | WARN_ON(1); | |
1322 | return -ENOMEM; | |
1323 | } | |
175542f5 MW |
1324 | #endif |
1325 | ||
30b888ba MW |
1326 | static void node_tag_set(struct radix_tree_root *root, |
1327 | struct radix_tree_node *node, | |
1328 | unsigned int tag, unsigned int offset) | |
1329 | { | |
1330 | while (node) { | |
1331 | if (tag_get(node, tag, offset)) | |
1332 | return; | |
1333 | tag_set(node, tag, offset); | |
1334 | offset = node->offset; | |
1335 | node = node->parent; | |
1336 | } | |
1337 | ||
1338 | if (!root_tag_get(root, tag)) | |
1339 | root_tag_set(root, tag); | |
1340 | } | |
1341 | ||
1da177e4 LT |
1342 | /** |
1343 | * radix_tree_tag_set - set a tag on a radix tree node | |
1344 | * @root: radix tree root | |
1345 | * @index: index key | |
2fcd9005 | 1346 | * @tag: tag index |
1da177e4 | 1347 | * |
daff89f3 JC |
1348 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
1349 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
1350 | * the root all the way down to the leaf node. |
1351 | * | |
2fcd9005 | 1352 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
1353 | * item is a bug. |
1354 | */ | |
1355 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 1356 | unsigned long index, unsigned int tag) |
1da177e4 | 1357 | { |
fb969909 RZ |
1358 | struct radix_tree_node *node, *parent; |
1359 | unsigned long maxindex; | |
1da177e4 | 1360 | |
9e85d811 | 1361 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 1362 | BUG_ON(index > maxindex); |
1da177e4 | 1363 | |
b194d16c | 1364 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 1365 | unsigned offset; |
1da177e4 | 1366 | |
4dd6c098 | 1367 | parent = entry_to_node(node); |
9e85d811 | 1368 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
1369 | BUG_ON(!node); |
1370 | ||
1371 | if (!tag_get(parent, tag, offset)) | |
1372 | tag_set(parent, tag, offset); | |
1da177e4 LT |
1373 | } |
1374 | ||
612d6c19 | 1375 | /* set the root's tag bit */ |
fb969909 | 1376 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
1377 | root_tag_set(root, tag); |
1378 | ||
fb969909 | 1379 | return node; |
1da177e4 LT |
1380 | } |
1381 | EXPORT_SYMBOL(radix_tree_tag_set); | |
1382 | ||
30b888ba MW |
1383 | /** |
1384 | * radix_tree_iter_tag_set - set a tag on the current iterator entry | |
1385 | * @root: radix tree root | |
1386 | * @iter: iterator state | |
1387 | * @tag: tag to set | |
1388 | */ | |
1389 | void radix_tree_iter_tag_set(struct radix_tree_root *root, | |
1390 | const struct radix_tree_iter *iter, unsigned int tag) | |
1391 | { | |
1392 | node_tag_set(root, iter->node, tag, iter_offset(iter)); | |
1393 | } | |
1394 | ||
d604c324 MW |
1395 | static void node_tag_clear(struct radix_tree_root *root, |
1396 | struct radix_tree_node *node, | |
1397 | unsigned int tag, unsigned int offset) | |
1398 | { | |
1399 | while (node) { | |
1400 | if (!tag_get(node, tag, offset)) | |
1401 | return; | |
1402 | tag_clear(node, tag, offset); | |
1403 | if (any_tag_set(node, tag)) | |
1404 | return; | |
1405 | ||
1406 | offset = node->offset; | |
1407 | node = node->parent; | |
1408 | } | |
1409 | ||
1410 | /* clear the root's tag bit */ | |
1411 | if (root_tag_get(root, tag)) | |
1412 | root_tag_clear(root, tag); | |
1413 | } | |
1414 | ||
1da177e4 LT |
1415 | /** |
1416 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1417 | * @root: radix tree root | |
1418 | * @index: index key | |
2fcd9005 | 1419 | * @tag: tag index |
1da177e4 | 1420 | * |
daff89f3 | 1421 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1422 | * corresponding to @index in the radix tree. If this causes |
1423 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1424 | * next-to-leaf node, etc. |
1425 | * | |
1426 | * Returns the address of the tagged item on success, else NULL. ie: | |
1427 | * has the same return value and semantics as radix_tree_lookup(). | |
1428 | */ | |
1429 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1430 | unsigned long index, unsigned int tag) |
1da177e4 | 1431 | { |
00f47b58 RZ |
1432 | struct radix_tree_node *node, *parent; |
1433 | unsigned long maxindex; | |
e2bdb933 | 1434 | int uninitialized_var(offset); |
1da177e4 | 1435 | |
9e85d811 | 1436 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1437 | if (index > maxindex) |
1438 | return NULL; | |
1da177e4 | 1439 | |
00f47b58 | 1440 | parent = NULL; |
1da177e4 | 1441 | |
b194d16c | 1442 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1443 | parent = entry_to_node(node); |
9e85d811 | 1444 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1445 | } |
1446 | ||
d604c324 MW |
1447 | if (node) |
1448 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1449 | |
00f47b58 | 1450 | return node; |
1da177e4 LT |
1451 | } |
1452 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1453 | ||
30b888ba MW |
1454 | /** |
1455 | * radix_tree_iter_tag_clear - clear a tag on the current iterator entry | |
1456 | * @root: radix tree root | |
1457 | * @iter: iterator state | |
1458 | * @tag: tag to clear | |
1459 | */ | |
1460 | void radix_tree_iter_tag_clear(struct radix_tree_root *root, | |
1461 | const struct radix_tree_iter *iter, unsigned int tag) | |
1462 | { | |
1463 | node_tag_clear(root, iter->node, tag, iter_offset(iter)); | |
1464 | } | |
1465 | ||
1da177e4 | 1466 | /** |
32605a18 MT |
1467 | * radix_tree_tag_get - get a tag on a radix tree node |
1468 | * @root: radix tree root | |
1469 | * @index: index key | |
2fcd9005 | 1470 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1471 | * |
32605a18 | 1472 | * Return values: |
1da177e4 | 1473 | * |
612d6c19 NP |
1474 | * 0: tag not present or not set |
1475 | * 1: tag set | |
ce82653d DH |
1476 | * |
1477 | * Note that the return value of this function may not be relied on, even if | |
1478 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1479 | * from concurrency. | |
1da177e4 | 1480 | */ |
35534c86 | 1481 | int radix_tree_tag_get(const struct radix_tree_root *root, |
daff89f3 | 1482 | unsigned long index, unsigned int tag) |
1da177e4 | 1483 | { |
4589ba6d RZ |
1484 | struct radix_tree_node *node, *parent; |
1485 | unsigned long maxindex; | |
1da177e4 | 1486 | |
612d6c19 NP |
1487 | if (!root_tag_get(root, tag)) |
1488 | return 0; | |
1489 | ||
9e85d811 | 1490 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1491 | if (index > maxindex) |
1492 | return 0; | |
7cf9c2c7 | 1493 | |
b194d16c | 1494 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1495 | unsigned offset; |
1da177e4 | 1496 | |
4dd6c098 | 1497 | parent = entry_to_node(node); |
9e85d811 | 1498 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1499 | |
4589ba6d | 1500 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1501 | return 0; |
4589ba6d RZ |
1502 | if (node == RADIX_TREE_RETRY) |
1503 | break; | |
1da177e4 | 1504 | } |
4589ba6d RZ |
1505 | |
1506 | return 1; | |
1da177e4 LT |
1507 | } |
1508 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1509 | |
21ef5339 RZ |
1510 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1511 | unsigned int shift) | |
1512 | { | |
1513 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1514 | iter->shift = shift; | |
1515 | #endif | |
1516 | } | |
1517 | ||
148deab2 MW |
1518 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1519 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1520 | struct radix_tree_node *node, unsigned offset, | |
1521 | unsigned tag) | |
1522 | { | |
1523 | unsigned tag_long = offset / BITS_PER_LONG; | |
1524 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1525 | ||
0a835c4f MW |
1526 | if (!node) { |
1527 | iter->tags = 1; | |
1528 | return; | |
1529 | } | |
1530 | ||
148deab2 MW |
1531 | iter->tags = node->tags[tag][tag_long] >> tag_bit; |
1532 | ||
1533 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1534 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1535 | /* Pick tags from next element */ | |
1536 | if (tag_bit) | |
1537 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1538 | (BITS_PER_LONG - tag_bit); | |
1539 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1540 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1541 | } | |
1542 | } | |
1543 | ||
1544 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
d7b62727 MW |
1545 | static void __rcu **skip_siblings(struct radix_tree_node **nodep, |
1546 | void __rcu **slot, struct radix_tree_iter *iter) | |
148deab2 | 1547 | { |
148deab2 MW |
1548 | while (iter->index < iter->next_index) { |
1549 | *nodep = rcu_dereference_raw(*slot); | |
02c02bf1 | 1550 | if (*nodep && !xa_is_sibling(*nodep)) |
148deab2 MW |
1551 | return slot; |
1552 | slot++; | |
1553 | iter->index = __radix_tree_iter_add(iter, 1); | |
1554 | iter->tags >>= 1; | |
1555 | } | |
1556 | ||
1557 | *nodep = NULL; | |
1558 | return NULL; | |
1559 | } | |
1560 | ||
d7b62727 MW |
1561 | void __rcu **__radix_tree_next_slot(void __rcu **slot, |
1562 | struct radix_tree_iter *iter, unsigned flags) | |
148deab2 MW |
1563 | { |
1564 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
9f418224 | 1565 | struct radix_tree_node *node; |
148deab2 MW |
1566 | |
1567 | slot = skip_siblings(&node, slot, iter); | |
1568 | ||
1569 | while (radix_tree_is_internal_node(node)) { | |
1570 | unsigned offset; | |
1571 | unsigned long next_index; | |
1572 | ||
1573 | if (node == RADIX_TREE_RETRY) | |
1574 | return slot; | |
1575 | node = entry_to_node(node); | |
268f42de | 1576 | iter->node = node; |
148deab2 MW |
1577 | iter->shift = node->shift; |
1578 | ||
1579 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1580 | offset = radix_tree_find_next_bit(node, tag, 0); | |
1581 | if (offset == RADIX_TREE_MAP_SIZE) | |
1582 | return NULL; | |
1583 | slot = &node->slots[offset]; | |
1584 | iter->index = __radix_tree_iter_add(iter, offset); | |
1585 | set_iter_tags(iter, node, offset, tag); | |
1586 | node = rcu_dereference_raw(*slot); | |
1587 | } else { | |
1588 | offset = 0; | |
1589 | slot = &node->slots[0]; | |
1590 | for (;;) { | |
1591 | node = rcu_dereference_raw(*slot); | |
1592 | if (node) | |
1593 | break; | |
1594 | slot++; | |
1595 | offset++; | |
1596 | if (offset == RADIX_TREE_MAP_SIZE) | |
1597 | return NULL; | |
1598 | } | |
1599 | iter->index = __radix_tree_iter_add(iter, offset); | |
1600 | } | |
1601 | if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0)) | |
1602 | goto none; | |
1603 | next_index = (iter->index | shift_maxindex(iter->shift)) + 1; | |
1604 | if (next_index < iter->next_index) | |
1605 | iter->next_index = next_index; | |
1606 | } | |
1607 | ||
1608 | return slot; | |
1609 | none: | |
1610 | iter->next_index = 0; | |
1611 | return NULL; | |
1612 | } | |
1613 | EXPORT_SYMBOL(__radix_tree_next_slot); | |
1614 | #else | |
d7b62727 MW |
1615 | static void __rcu **skip_siblings(struct radix_tree_node **nodep, |
1616 | void __rcu **slot, struct radix_tree_iter *iter) | |
148deab2 MW |
1617 | { |
1618 | return slot; | |
1619 | } | |
1620 | #endif | |
1621 | ||
d7b62727 MW |
1622 | void __rcu **radix_tree_iter_resume(void __rcu **slot, |
1623 | struct radix_tree_iter *iter) | |
148deab2 MW |
1624 | { |
1625 | struct radix_tree_node *node; | |
1626 | ||
1627 | slot++; | |
1628 | iter->index = __radix_tree_iter_add(iter, 1); | |
148deab2 MW |
1629 | skip_siblings(&node, slot, iter); |
1630 | iter->next_index = iter->index; | |
1631 | iter->tags = 0; | |
1632 | return NULL; | |
1633 | } | |
1634 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1635 | ||
78c1d784 KK |
1636 | /** |
1637 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1638 | * | |
1639 | * @root: radix tree root | |
1640 | * @iter: iterator state | |
1641 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1642 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1643 | */ | |
d7b62727 | 1644 | void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root, |
78c1d784 KK |
1645 | struct radix_tree_iter *iter, unsigned flags) |
1646 | { | |
9e85d811 | 1647 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1648 | struct radix_tree_node *node, *child; |
21ef5339 | 1649 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1650 | |
1651 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1652 | return NULL; | |
1653 | ||
1654 | /* | |
1655 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1656 | * during iterating; it can be zero only at the beginning. | |
1657 | * And we cannot overflow iter->next_index in a single step, | |
1658 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1659 | * |
1660 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1661 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1662 | */ |
1663 | index = iter->next_index; | |
1664 | if (!index && iter->index) | |
1665 | return NULL; | |
1666 | ||
21ef5339 | 1667 | restart: |
9e85d811 | 1668 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1669 | if (index > maxindex) |
1670 | return NULL; | |
8c1244de MW |
1671 | if (!child) |
1672 | return NULL; | |
21ef5339 | 1673 | |
8c1244de | 1674 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1675 | /* Single-slot tree */ |
21ef5339 RZ |
1676 | iter->index = index; |
1677 | iter->next_index = maxindex + 1; | |
78c1d784 | 1678 | iter->tags = 1; |
268f42de | 1679 | iter->node = NULL; |
8c1244de | 1680 | __set_iter_shift(iter, 0); |
f8d5d0cc | 1681 | return (void __rcu **)&root->xa_head; |
8c1244de | 1682 | } |
21ef5339 | 1683 | |
8c1244de MW |
1684 | do { |
1685 | node = entry_to_node(child); | |
9e85d811 | 1686 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1687 | |
78c1d784 | 1688 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1689 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1690 | /* Hole detected */ |
1691 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1692 | return NULL; | |
1693 | ||
1694 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1695 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1696 | offset + 1); |
1697 | else | |
1698 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
12320d0f MW |
1699 | void *slot = rcu_dereference_raw( |
1700 | node->slots[offset]); | |
02c02bf1 | 1701 | if (xa_is_sibling(slot)) |
21ef5339 RZ |
1702 | continue; |
1703 | if (slot) | |
78c1d784 KK |
1704 | break; |
1705 | } | |
8c1244de | 1706 | index &= ~node_maxindex(node); |
9e85d811 | 1707 | index += offset << node->shift; |
78c1d784 KK |
1708 | /* Overflow after ~0UL */ |
1709 | if (!index) | |
1710 | return NULL; | |
1711 | if (offset == RADIX_TREE_MAP_SIZE) | |
1712 | goto restart; | |
8c1244de | 1713 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1714 | } |
1715 | ||
e157b555 | 1716 | if (!child) |
78c1d784 | 1717 | goto restart; |
e157b555 MW |
1718 | if (child == RADIX_TREE_RETRY) |
1719 | break; | |
66ee620f | 1720 | } while (node->shift && radix_tree_is_internal_node(child)); |
78c1d784 KK |
1721 | |
1722 | /* Update the iterator state */ | |
8c1244de MW |
1723 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1724 | iter->next_index = (index | node_maxindex(node)) + 1; | |
268f42de | 1725 | iter->node = node; |
9e85d811 | 1726 | __set_iter_shift(iter, node->shift); |
78c1d784 | 1727 | |
148deab2 MW |
1728 | if (flags & RADIX_TREE_ITER_TAGGED) |
1729 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1730 | |
1731 | return node->slots + offset; | |
1732 | } | |
1733 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1734 | ||
1da177e4 LT |
1735 | /** |
1736 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1737 | * @root: radix tree root | |
1738 | * @results: where the results of the lookup are placed | |
1739 | * @first_index: start the lookup from this key | |
1740 | * @max_items: place up to this many items at *results | |
1741 | * | |
1742 | * Performs an index-ascending scan of the tree for present items. Places | |
1743 | * them at *@results and returns the number of items which were placed at | |
1744 | * *@results. | |
1745 | * | |
1746 | * The implementation is naive. | |
7cf9c2c7 NP |
1747 | * |
1748 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1749 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1750 | * an atomic snapshot of the tree at a single point in time, the |
1751 | * semantics of an RCU protected gang lookup are as though multiple | |
1752 | * radix_tree_lookups have been issued in individual locks, and results | |
1753 | * stored in 'results'. | |
1da177e4 LT |
1754 | */ |
1755 | unsigned int | |
35534c86 | 1756 | radix_tree_gang_lookup(const struct radix_tree_root *root, void **results, |
1da177e4 LT |
1757 | unsigned long first_index, unsigned int max_items) |
1758 | { | |
cebbd29e | 1759 | struct radix_tree_iter iter; |
d7b62727 | 1760 | void __rcu **slot; |
cebbd29e | 1761 | unsigned int ret = 0; |
7cf9c2c7 | 1762 | |
cebbd29e | 1763 | if (unlikely(!max_items)) |
7cf9c2c7 | 1764 | return 0; |
1da177e4 | 1765 | |
cebbd29e | 1766 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1767 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1768 | if (!results[ret]) |
1769 | continue; | |
b194d16c | 1770 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1771 | slot = radix_tree_iter_retry(&iter); |
1772 | continue; | |
1773 | } | |
cebbd29e | 1774 | if (++ret == max_items) |
1da177e4 | 1775 | break; |
1da177e4 | 1776 | } |
7cf9c2c7 | 1777 | |
1da177e4 LT |
1778 | return ret; |
1779 | } | |
1780 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1781 | ||
47feff2c NP |
1782 | /** |
1783 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1784 | * @root: radix tree root | |
1785 | * @results: where the results of the lookup are placed | |
6328650b | 1786 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1787 | * @first_index: start the lookup from this key |
1788 | * @max_items: place up to this many items at *results | |
1789 | * | |
1790 | * Performs an index-ascending scan of the tree for present items. Places | |
1791 | * their slots at *@results and returns the number of items which were | |
1792 | * placed at *@results. | |
1793 | * | |
1794 | * The implementation is naive. | |
1795 | * | |
1796 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1797 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1798 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1799 | */ | |
1800 | unsigned int | |
35534c86 | 1801 | radix_tree_gang_lookup_slot(const struct radix_tree_root *root, |
d7b62727 | 1802 | void __rcu ***results, unsigned long *indices, |
47feff2c NP |
1803 | unsigned long first_index, unsigned int max_items) |
1804 | { | |
cebbd29e | 1805 | struct radix_tree_iter iter; |
d7b62727 | 1806 | void __rcu **slot; |
cebbd29e | 1807 | unsigned int ret = 0; |
47feff2c | 1808 | |
cebbd29e | 1809 | if (unlikely(!max_items)) |
47feff2c NP |
1810 | return 0; |
1811 | ||
cebbd29e KK |
1812 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1813 | results[ret] = slot; | |
6328650b | 1814 | if (indices) |
cebbd29e KK |
1815 | indices[ret] = iter.index; |
1816 | if (++ret == max_items) | |
47feff2c | 1817 | break; |
47feff2c NP |
1818 | } |
1819 | ||
1820 | return ret; | |
1821 | } | |
1822 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1823 | ||
1da177e4 LT |
1824 | /** |
1825 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1826 | * based on a tag | |
1827 | * @root: radix tree root | |
1828 | * @results: where the results of the lookup are placed | |
1829 | * @first_index: start the lookup from this key | |
1830 | * @max_items: place up to this many items at *results | |
daff89f3 | 1831 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1832 | * |
1833 | * Performs an index-ascending scan of the tree for present items which | |
1834 | * have the tag indexed by @tag set. Places the items at *@results and | |
1835 | * returns the number of items which were placed at *@results. | |
1836 | */ | |
1837 | unsigned int | |
35534c86 | 1838 | radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results, |
daff89f3 JC |
1839 | unsigned long first_index, unsigned int max_items, |
1840 | unsigned int tag) | |
1da177e4 | 1841 | { |
cebbd29e | 1842 | struct radix_tree_iter iter; |
d7b62727 | 1843 | void __rcu **slot; |
cebbd29e | 1844 | unsigned int ret = 0; |
612d6c19 | 1845 | |
cebbd29e | 1846 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1847 | return 0; |
1848 | ||
cebbd29e | 1849 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1850 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1851 | if (!results[ret]) |
1852 | continue; | |
b194d16c | 1853 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1854 | slot = radix_tree_iter_retry(&iter); |
1855 | continue; | |
1856 | } | |
cebbd29e | 1857 | if (++ret == max_items) |
1da177e4 | 1858 | break; |
1da177e4 | 1859 | } |
7cf9c2c7 | 1860 | |
1da177e4 LT |
1861 | return ret; |
1862 | } | |
1863 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1864 | ||
47feff2c NP |
1865 | /** |
1866 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1867 | * radix tree based on a tag | |
1868 | * @root: radix tree root | |
1869 | * @results: where the results of the lookup are placed | |
1870 | * @first_index: start the lookup from this key | |
1871 | * @max_items: place up to this many items at *results | |
1872 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1873 | * | |
1874 | * Performs an index-ascending scan of the tree for present items which | |
1875 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1876 | * returns the number of slots which were placed at *@results. | |
1877 | */ | |
1878 | unsigned int | |
35534c86 | 1879 | radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root, |
d7b62727 | 1880 | void __rcu ***results, unsigned long first_index, |
35534c86 | 1881 | unsigned int max_items, unsigned int tag) |
47feff2c | 1882 | { |
cebbd29e | 1883 | struct radix_tree_iter iter; |
d7b62727 | 1884 | void __rcu **slot; |
cebbd29e | 1885 | unsigned int ret = 0; |
47feff2c | 1886 | |
cebbd29e | 1887 | if (unlikely(!max_items)) |
47feff2c NP |
1888 | return 0; |
1889 | ||
cebbd29e KK |
1890 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1891 | results[ret] = slot; | |
1892 | if (++ret == max_items) | |
47feff2c | 1893 | break; |
47feff2c NP |
1894 | } |
1895 | ||
1896 | return ret; | |
1897 | } | |
1898 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1899 | ||
139e5616 JW |
1900 | /** |
1901 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1902 | * @root: radix tree root | |
139e5616 | 1903 | * @node: node containing @index |
ea07b862 | 1904 | * @update_node: callback for changing leaf nodes |
139e5616 JW |
1905 | * |
1906 | * After clearing the slot at @index in @node from radix tree | |
1907 | * rooted at @root, call this function to attempt freeing the | |
1908 | * node and shrinking the tree. | |
139e5616 | 1909 | */ |
14b46879 | 1910 | void __radix_tree_delete_node(struct radix_tree_root *root, |
ea07b862 | 1911 | struct radix_tree_node *node, |
c7df8ad2 | 1912 | radix_tree_update_node_t update_node) |
139e5616 | 1913 | { |
c7df8ad2 | 1914 | delete_node(root, node, update_node); |
139e5616 JW |
1915 | } |
1916 | ||
0ac398ef | 1917 | static bool __radix_tree_delete(struct radix_tree_root *root, |
d7b62727 | 1918 | struct radix_tree_node *node, void __rcu **slot) |
0ac398ef | 1919 | { |
0a835c4f | 1920 | void *old = rcu_dereference_raw(*slot); |
01959dfe | 1921 | int values = xa_is_value(old) ? -1 : 0; |
0ac398ef MW |
1922 | unsigned offset = get_slot_offset(node, slot); |
1923 | int tag; | |
1924 | ||
0a835c4f MW |
1925 | if (is_idr(root)) |
1926 | node_tag_set(root, node, IDR_FREE, offset); | |
1927 | else | |
1928 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1929 | node_tag_clear(root, node, tag, offset); | |
0ac398ef | 1930 | |
01959dfe | 1931 | replace_slot(slot, NULL, node, -1, values); |
c7df8ad2 | 1932 | return node && delete_node(root, node, NULL); |
0ac398ef MW |
1933 | } |
1934 | ||
1da177e4 | 1935 | /** |
0ac398ef MW |
1936 | * radix_tree_iter_delete - delete the entry at this iterator position |
1937 | * @root: radix tree root | |
1938 | * @iter: iterator state | |
1939 | * @slot: pointer to slot | |
1da177e4 | 1940 | * |
0ac398ef MW |
1941 | * Delete the entry at the position currently pointed to by the iterator. |
1942 | * This may result in the current node being freed; if it is, the iterator | |
1943 | * is advanced so that it will not reference the freed memory. This | |
1944 | * function may be called without any locking if there are no other threads | |
1945 | * which can access this tree. | |
1946 | */ | |
1947 | void radix_tree_iter_delete(struct radix_tree_root *root, | |
d7b62727 | 1948 | struct radix_tree_iter *iter, void __rcu **slot) |
0ac398ef MW |
1949 | { |
1950 | if (__radix_tree_delete(root, iter->node, slot)) | |
1951 | iter->index = iter->next_index; | |
1952 | } | |
d1b48c1e | 1953 | EXPORT_SYMBOL(radix_tree_iter_delete); |
0ac398ef MW |
1954 | |
1955 | /** | |
1956 | * radix_tree_delete_item - delete an item from a radix tree | |
1957 | * @root: radix tree root | |
1958 | * @index: index key | |
1959 | * @item: expected item | |
1da177e4 | 1960 | * |
0ac398ef | 1961 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1962 | * |
0ac398ef MW |
1963 | * Return: the deleted entry, or %NULL if it was not present |
1964 | * or the entry at the given @index was not @item. | |
1da177e4 | 1965 | */ |
53c59f26 JW |
1966 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1967 | unsigned long index, void *item) | |
1da177e4 | 1968 | { |
0a835c4f | 1969 | struct radix_tree_node *node = NULL; |
7a4deea1 | 1970 | void __rcu **slot = NULL; |
139e5616 | 1971 | void *entry; |
1da177e4 | 1972 | |
139e5616 | 1973 | entry = __radix_tree_lookup(root, index, &node, &slot); |
7a4deea1 MW |
1974 | if (!slot) |
1975 | return NULL; | |
0a835c4f MW |
1976 | if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE, |
1977 | get_slot_offset(node, slot)))) | |
139e5616 | 1978 | return NULL; |
1da177e4 | 1979 | |
139e5616 JW |
1980 | if (item && entry != item) |
1981 | return NULL; | |
1982 | ||
0ac398ef | 1983 | __radix_tree_delete(root, node, slot); |
612d6c19 | 1984 | |
139e5616 | 1985 | return entry; |
1da177e4 | 1986 | } |
53c59f26 JW |
1987 | EXPORT_SYMBOL(radix_tree_delete_item); |
1988 | ||
1989 | /** | |
0ac398ef MW |
1990 | * radix_tree_delete - delete an entry from a radix tree |
1991 | * @root: radix tree root | |
1992 | * @index: index key | |
53c59f26 | 1993 | * |
0ac398ef | 1994 | * Remove the entry at @index from the radix tree rooted at @root. |
53c59f26 | 1995 | * |
0ac398ef | 1996 | * Return: The deleted entry, or %NULL if it was not present. |
53c59f26 JW |
1997 | */ |
1998 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1999 | { | |
2000 | return radix_tree_delete_item(root, index, NULL); | |
2001 | } | |
1da177e4 LT |
2002 | EXPORT_SYMBOL(radix_tree_delete); |
2003 | ||
d3798ae8 JW |
2004 | void radix_tree_clear_tags(struct radix_tree_root *root, |
2005 | struct radix_tree_node *node, | |
d7b62727 | 2006 | void __rcu **slot) |
d604c324 | 2007 | { |
d604c324 MW |
2008 | if (node) { |
2009 | unsigned int tag, offset = get_slot_offset(node, slot); | |
2010 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
2011 | node_tag_clear(root, node, tag, offset); | |
2012 | } else { | |
0a835c4f | 2013 | root_tag_clear_all(root); |
d604c324 | 2014 | } |
d604c324 MW |
2015 | } |
2016 | ||
1da177e4 LT |
2017 | /** |
2018 | * radix_tree_tagged - test whether any items in the tree are tagged | |
2019 | * @root: radix tree root | |
2020 | * @tag: tag to test | |
2021 | */ | |
35534c86 | 2022 | int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 2023 | { |
612d6c19 | 2024 | return root_tag_get(root, tag); |
1da177e4 LT |
2025 | } |
2026 | EXPORT_SYMBOL(radix_tree_tagged); | |
2027 | ||
0a835c4f MW |
2028 | /** |
2029 | * idr_preload - preload for idr_alloc() | |
2030 | * @gfp_mask: allocation mask to use for preloading | |
2031 | * | |
2032 | * Preallocate memory to use for the next call to idr_alloc(). This function | |
2033 | * returns with preemption disabled. It will be enabled by idr_preload_end(). | |
2034 | */ | |
2035 | void idr_preload(gfp_t gfp_mask) | |
2036 | { | |
bc9ae224 ED |
2037 | if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE)) |
2038 | preempt_disable(); | |
0a835c4f MW |
2039 | } |
2040 | EXPORT_SYMBOL(idr_preload); | |
2041 | ||
7ad3d4d8 MW |
2042 | int ida_pre_get(struct ida *ida, gfp_t gfp) |
2043 | { | |
7ad3d4d8 MW |
2044 | /* |
2045 | * The IDA API has no preload_end() equivalent. Instead, | |
2046 | * ida_get_new() can return -EAGAIN, prompting the caller | |
2047 | * to return to the ida_pre_get() step. | |
2048 | */ | |
bc9ae224 ED |
2049 | if (!__radix_tree_preload(gfp, IDA_PRELOAD_SIZE)) |
2050 | preempt_enable(); | |
7ad3d4d8 MW |
2051 | |
2052 | if (!this_cpu_read(ida_bitmap)) { | |
b1a8a7a7 | 2053 | struct ida_bitmap *bitmap = kzalloc(sizeof(*bitmap), gfp); |
7ad3d4d8 MW |
2054 | if (!bitmap) |
2055 | return 0; | |
4ecd9542 MW |
2056 | if (this_cpu_cmpxchg(ida_bitmap, NULL, bitmap)) |
2057 | kfree(bitmap); | |
7ad3d4d8 MW |
2058 | } |
2059 | ||
2060 | return 1; | |
2061 | } | |
7ad3d4d8 | 2062 | |
460488c5 | 2063 | void __rcu **idr_get_free(struct radix_tree_root *root, |
388f79fd CM |
2064 | struct radix_tree_iter *iter, gfp_t gfp, |
2065 | unsigned long max) | |
0a835c4f MW |
2066 | { |
2067 | struct radix_tree_node *node = NULL, *child; | |
f8d5d0cc | 2068 | void __rcu **slot = (void __rcu **)&root->xa_head; |
0a835c4f | 2069 | unsigned long maxindex, start = iter->next_index; |
0a835c4f MW |
2070 | unsigned int shift, offset = 0; |
2071 | ||
2072 | grow: | |
2073 | shift = radix_tree_load_root(root, &child, &maxindex); | |
2074 | if (!radix_tree_tagged(root, IDR_FREE)) | |
2075 | start = max(start, maxindex + 1); | |
2076 | if (start > max) | |
2077 | return ERR_PTR(-ENOSPC); | |
2078 | ||
2079 | if (start > maxindex) { | |
2080 | int error = radix_tree_extend(root, gfp, start, shift); | |
2081 | if (error < 0) | |
2082 | return ERR_PTR(error); | |
2083 | shift = error; | |
f8d5d0cc | 2084 | child = rcu_dereference_raw(root->xa_head); |
0a835c4f | 2085 | } |
66ee620f MW |
2086 | if (start == 0 && shift == 0) |
2087 | shift = RADIX_TREE_MAP_SHIFT; | |
0a835c4f MW |
2088 | |
2089 | while (shift) { | |
2090 | shift -= RADIX_TREE_MAP_SHIFT; | |
2091 | if (child == NULL) { | |
2092 | /* Have to add a child node. */ | |
d58275bc MW |
2093 | child = radix_tree_node_alloc(gfp, node, root, shift, |
2094 | offset, 0, 0); | |
0a835c4f MW |
2095 | if (!child) |
2096 | return ERR_PTR(-ENOMEM); | |
2097 | all_tag_set(child, IDR_FREE); | |
2098 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
2099 | if (node) | |
2100 | node->count++; | |
2101 | } else if (!radix_tree_is_internal_node(child)) | |
2102 | break; | |
2103 | ||
2104 | node = entry_to_node(child); | |
2105 | offset = radix_tree_descend(node, &child, start); | |
2106 | if (!tag_get(node, IDR_FREE, offset)) { | |
2107 | offset = radix_tree_find_next_bit(node, IDR_FREE, | |
2108 | offset + 1); | |
2109 | start = next_index(start, node, offset); | |
2110 | if (start > max) | |
2111 | return ERR_PTR(-ENOSPC); | |
2112 | while (offset == RADIX_TREE_MAP_SIZE) { | |
2113 | offset = node->offset + 1; | |
2114 | node = node->parent; | |
2115 | if (!node) | |
2116 | goto grow; | |
2117 | shift = node->shift; | |
2118 | } | |
2119 | child = rcu_dereference_raw(node->slots[offset]); | |
2120 | } | |
2121 | slot = &node->slots[offset]; | |
2122 | } | |
2123 | ||
2124 | iter->index = start; | |
2125 | if (node) | |
2126 | iter->next_index = 1 + min(max, (start | node_maxindex(node))); | |
2127 | else | |
2128 | iter->next_index = 1; | |
2129 | iter->node = node; | |
2130 | __set_iter_shift(iter, shift); | |
2131 | set_iter_tags(iter, node, offset, IDR_FREE); | |
2132 | ||
2133 | return slot; | |
2134 | } | |
2135 | ||
2136 | /** | |
2137 | * idr_destroy - release all internal memory from an IDR | |
2138 | * @idr: idr handle | |
2139 | * | |
2140 | * After this function is called, the IDR is empty, and may be reused or | |
2141 | * the data structure containing it may be freed. | |
2142 | * | |
2143 | * A typical clean-up sequence for objects stored in an idr tree will use | |
2144 | * idr_for_each() to free all objects, if necessary, then idr_destroy() to | |
2145 | * free the memory used to keep track of those objects. | |
2146 | */ | |
2147 | void idr_destroy(struct idr *idr) | |
2148 | { | |
f8d5d0cc | 2149 | struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.xa_head); |
0a835c4f MW |
2150 | if (radix_tree_is_internal_node(node)) |
2151 | radix_tree_free_nodes(node); | |
f8d5d0cc | 2152 | idr->idr_rt.xa_head = NULL; |
0a835c4f MW |
2153 | root_tag_set(&idr->idr_rt, IDR_FREE); |
2154 | } | |
2155 | EXPORT_SYMBOL(idr_destroy); | |
2156 | ||
1da177e4 | 2157 | static void |
449dd698 | 2158 | radix_tree_node_ctor(void *arg) |
1da177e4 | 2159 | { |
449dd698 JW |
2160 | struct radix_tree_node *node = arg; |
2161 | ||
2162 | memset(node, 0, sizeof(*node)); | |
2163 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
2164 | } |
2165 | ||
c78c66d1 KS |
2166 | static __init unsigned long __maxindex(unsigned int height) |
2167 | { | |
2168 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
2169 | int shift = RADIX_TREE_INDEX_BITS - width; | |
2170 | ||
2171 | if (shift < 0) | |
2172 | return ~0UL; | |
2173 | if (shift >= BITS_PER_LONG) | |
2174 | return 0UL; | |
2175 | return ~0UL >> shift; | |
2176 | } | |
2177 | ||
2178 | static __init void radix_tree_init_maxnodes(void) | |
2179 | { | |
2180 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
2181 | unsigned int i, j; | |
2182 | ||
2183 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
2184 | height_to_maxindex[i] = __maxindex(i); | |
2185 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
2186 | for (j = i; j > 0; j--) | |
2187 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
2188 | } | |
2189 | } | |
2190 | ||
d544abd5 | 2191 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 2192 | { |
2fcd9005 MW |
2193 | struct radix_tree_preload *rtp; |
2194 | struct radix_tree_node *node; | |
2195 | ||
2196 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
2197 | rtp = &per_cpu(radix_tree_preloads, cpu); |
2198 | while (rtp->nr) { | |
2199 | node = rtp->nodes; | |
1293d5c5 | 2200 | rtp->nodes = node->parent; |
d544abd5 SAS |
2201 | kmem_cache_free(radix_tree_node_cachep, node); |
2202 | rtp->nr--; | |
2fcd9005 | 2203 | } |
7ad3d4d8 MW |
2204 | kfree(per_cpu(ida_bitmap, cpu)); |
2205 | per_cpu(ida_bitmap, cpu) = NULL; | |
d544abd5 | 2206 | return 0; |
1da177e4 | 2207 | } |
1da177e4 LT |
2208 | |
2209 | void __init radix_tree_init(void) | |
2210 | { | |
d544abd5 | 2211 | int ret; |
7e784422 MH |
2212 | |
2213 | BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32); | |
fa290cda | 2214 | BUILD_BUG_ON(ROOT_IS_IDR & ~GFP_ZONEMASK); |
02c02bf1 | 2215 | BUILD_BUG_ON(XA_CHUNK_SIZE > 255); |
1da177e4 LT |
2216 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
2217 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
2218 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
2219 | radix_tree_node_ctor); | |
c78c66d1 | 2220 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
2221 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
2222 | NULL, radix_tree_cpu_dead); | |
2223 | WARN_ON(ret < 0); | |
1da177e4 | 2224 | } |