]>
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 | ||
25 | #include <linux/errno.h> | |
26 | #include <linux/init.h> | |
27 | #include <linux/kernel.h> | |
8bc3bcc9 | 28 | #include <linux/export.h> |
1da177e4 LT |
29 | #include <linux/radix-tree.h> |
30 | #include <linux/percpu.h> | |
31 | #include <linux/slab.h> | |
ce80b067 | 32 | #include <linux/kmemleak.h> |
1da177e4 LT |
33 | #include <linux/notifier.h> |
34 | #include <linux/cpu.h> | |
1da177e4 LT |
35 | #include <linux/string.h> |
36 | #include <linux/bitops.h> | |
7cf9c2c7 | 37 | #include <linux/rcupdate.h> |
92cf2118 | 38 | #include <linux/preempt.h> /* in_interrupt() */ |
1da177e4 LT |
39 | |
40 | ||
c78c66d1 KS |
41 | /* Number of nodes in fully populated tree of given height */ |
42 | static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
43 | ||
1da177e4 LT |
44 | /* |
45 | * Radix tree node cache. | |
46 | */ | |
e18b890b | 47 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 48 | |
55368052 NP |
49 | /* |
50 | * The radix tree is variable-height, so an insert operation not only has | |
51 | * to build the branch to its corresponding item, it also has to build the | |
52 | * branch to existing items if the size has to be increased (by | |
53 | * radix_tree_extend). | |
54 | * | |
55 | * The worst case is a zero height tree with just a single item at index 0, | |
56 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
57 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
58 | * Hence: | |
59 | */ | |
60 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
61 | ||
1da177e4 LT |
62 | /* |
63 | * Per-cpu pool of preloaded nodes | |
64 | */ | |
65 | struct radix_tree_preload { | |
2fcd9005 | 66 | unsigned nr; |
9d2a8da0 KS |
67 | /* nodes->private_data points to next preallocated node */ |
68 | struct radix_tree_node *nodes; | |
1da177e4 | 69 | }; |
8cef7d57 | 70 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 71 | |
148deab2 MW |
72 | static inline struct radix_tree_node *entry_to_node(void *ptr) |
73 | { | |
74 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE); | |
75 | } | |
76 | ||
a4db4dce | 77 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 78 | { |
30ff46cc | 79 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
80 | } |
81 | ||
a4db4dce | 82 | #define RADIX_TREE_RETRY node_to_entry(NULL) |
afe0e395 | 83 | |
db050f29 MW |
84 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
85 | /* Sibling slots point directly to another slot in the same node */ | |
86 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
87 | { | |
88 | void **ptr = node; | |
89 | return (parent->slots <= ptr) && | |
90 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
91 | } | |
92 | #else | |
93 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
94 | { | |
95 | return false; | |
96 | } | |
97 | #endif | |
98 | ||
99 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
100 | void **slot) | |
101 | { | |
102 | return slot - parent->slots; | |
103 | } | |
104 | ||
9e85d811 MW |
105 | static unsigned int radix_tree_descend(struct radix_tree_node *parent, |
106 | struct radix_tree_node **nodep, unsigned long index) | |
db050f29 | 107 | { |
9e85d811 | 108 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
db050f29 MW |
109 | void **entry = rcu_dereference_raw(parent->slots[offset]); |
110 | ||
111 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 112 | if (radix_tree_is_internal_node(entry)) { |
8d2c0d36 LT |
113 | if (is_sibling_entry(parent, entry)) { |
114 | void **sibentry = (void **) entry_to_node(entry); | |
115 | offset = get_slot_offset(parent, sibentry); | |
116 | entry = rcu_dereference_raw(*sibentry); | |
db050f29 MW |
117 | } |
118 | } | |
119 | #endif | |
120 | ||
121 | *nodep = (void *)entry; | |
122 | return offset; | |
123 | } | |
124 | ||
612d6c19 NP |
125 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
126 | { | |
127 | return root->gfp_mask & __GFP_BITS_MASK; | |
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 | ||
142 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
143 | int offset) | |
144 | { | |
145 | return test_bit(offset, node->tags[tag]); | |
146 | } | |
147 | ||
148 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
149 | { | |
150 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
151 | } | |
152 | ||
2fcd9005 | 153 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
154 | { |
155 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
156 | } | |
157 | ||
158 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
159 | { | |
160 | root->gfp_mask &= __GFP_BITS_MASK; | |
161 | } | |
162 | ||
163 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
164 | { | |
2fcd9005 | 165 | return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
643b52b9 NP |
166 | } |
167 | ||
7b60e9ad MW |
168 | static inline unsigned root_tags_get(struct radix_tree_root *root) |
169 | { | |
170 | return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT; | |
171 | } | |
172 | ||
643b52b9 NP |
173 | /* |
174 | * Returns 1 if any slot in the node has this tag set. | |
175 | * Otherwise returns 0. | |
176 | */ | |
177 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
178 | { | |
2fcd9005 | 179 | unsigned idx; |
643b52b9 NP |
180 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
181 | if (node->tags[tag][idx]) | |
182 | return 1; | |
183 | } | |
184 | return 0; | |
185 | } | |
78c1d784 KK |
186 | |
187 | /** | |
188 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
189 | * | |
190 | * @addr: The address to base the search on | |
191 | * @size: The bitmap size in bits | |
192 | * @offset: The bitnumber to start searching at | |
193 | * | |
194 | * Unrollable variant of find_next_bit() for constant size arrays. | |
195 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
196 | * Returns next bit offset, or size if nothing found. | |
197 | */ | |
198 | static __always_inline unsigned long | |
bc412fca MW |
199 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
200 | unsigned long offset) | |
78c1d784 | 201 | { |
bc412fca | 202 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 203 | |
bc412fca | 204 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
205 | unsigned long tmp; |
206 | ||
207 | addr += offset / BITS_PER_LONG; | |
208 | tmp = *addr >> (offset % BITS_PER_LONG); | |
209 | if (tmp) | |
210 | return __ffs(tmp) + offset; | |
211 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 212 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
213 | tmp = *++addr; |
214 | if (tmp) | |
215 | return __ffs(tmp) + offset; | |
216 | offset += BITS_PER_LONG; | |
217 | } | |
218 | } | |
bc412fca | 219 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
220 | } |
221 | ||
268f42de MW |
222 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
223 | { | |
224 | return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK; | |
225 | } | |
226 | ||
218ed750 MW |
227 | /* |
228 | * The maximum index which can be stored in a radix tree | |
229 | */ | |
230 | static inline unsigned long shift_maxindex(unsigned int shift) | |
231 | { | |
232 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
233 | } | |
234 | ||
235 | static inline unsigned long node_maxindex(struct radix_tree_node *node) | |
236 | { | |
237 | return shift_maxindex(node->shift); | |
238 | } | |
239 | ||
0796c583 | 240 | #ifndef __KERNEL__ |
d0891265 | 241 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 242 | { |
0796c583 | 243 | unsigned long i; |
7cf19af4 | 244 | |
218ed750 MW |
245 | pr_debug("radix node: %p offset %d indices %lu-%lu parent %p tags %lx %lx %lx shift %d count %d exceptional %d\n", |
246 | node, node->offset, index, index | node_maxindex(node), | |
247 | node->parent, | |
0796c583 | 248 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
218ed750 | 249 | node->shift, node->count, node->exceptional); |
0796c583 RZ |
250 | |
251 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
252 | unsigned long first = index | (i << node->shift); |
253 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
254 | void *entry = node->slots[i]; |
255 | if (!entry) | |
256 | continue; | |
218ed750 MW |
257 | if (entry == RADIX_TREE_RETRY) { |
258 | pr_debug("radix retry offset %ld indices %lu-%lu parent %p\n", | |
259 | i, first, last, node); | |
b194d16c | 260 | } else if (!radix_tree_is_internal_node(entry)) { |
218ed750 MW |
261 | pr_debug("radix entry %p offset %ld indices %lu-%lu parent %p\n", |
262 | entry, i, first, last, node); | |
263 | } else if (is_sibling_entry(node, entry)) { | |
264 | pr_debug("radix sblng %p offset %ld indices %lu-%lu parent %p val %p\n", | |
265 | entry, i, first, last, node, | |
266 | *(void **)entry_to_node(entry)); | |
0796c583 | 267 | } else { |
4dd6c098 | 268 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
269 | } |
270 | } | |
7cf19af4 MW |
271 | } |
272 | ||
273 | /* For debug */ | |
274 | static void radix_tree_dump(struct radix_tree_root *root) | |
275 | { | |
d0891265 MW |
276 | pr_debug("radix root: %p rnode %p tags %x\n", |
277 | root, root->rnode, | |
7cf19af4 | 278 | root->gfp_mask >> __GFP_BITS_SHIFT); |
b194d16c | 279 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 280 | return; |
4dd6c098 | 281 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 MW |
282 | } |
283 | #endif | |
284 | ||
1da177e4 LT |
285 | /* |
286 | * This assumes that the caller has performed appropriate preallocation, and | |
287 | * that the caller has pinned this thread of control to the current CPU. | |
288 | */ | |
289 | static struct radix_tree_node * | |
290 | radix_tree_node_alloc(struct radix_tree_root *root) | |
291 | { | |
e2848a0e | 292 | struct radix_tree_node *ret = NULL; |
612d6c19 | 293 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 294 | |
5e4c0d97 | 295 | /* |
2fcd9005 MW |
296 | * Preload code isn't irq safe and it doesn't make sense to use |
297 | * preloading during an interrupt anyway as all the allocations have | |
298 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 299 | */ |
d0164adc | 300 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
301 | struct radix_tree_preload *rtp; |
302 | ||
58e698af VD |
303 | /* |
304 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
305 | * cache first for the new node to get accounted to the memory |
306 | * cgroup. | |
58e698af VD |
307 | */ |
308 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 309 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
310 | if (ret) |
311 | goto out; | |
312 | ||
e2848a0e NP |
313 | /* |
314 | * Provided the caller has preloaded here, we will always | |
315 | * succeed in getting a node here (and never reach | |
316 | * kmem_cache_alloc) | |
317 | */ | |
7c8e0181 | 318 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 319 | if (rtp->nr) { |
9d2a8da0 KS |
320 | ret = rtp->nodes; |
321 | rtp->nodes = ret->private_data; | |
322 | ret->private_data = NULL; | |
1da177e4 LT |
323 | rtp->nr--; |
324 | } | |
ce80b067 CM |
325 | /* |
326 | * Update the allocation stack trace as this is more useful | |
327 | * for debugging. | |
328 | */ | |
329 | kmemleak_update_trace(ret); | |
58e698af | 330 | goto out; |
1da177e4 | 331 | } |
05eb6e72 | 332 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 333 | out: |
b194d16c | 334 | BUG_ON(radix_tree_is_internal_node(ret)); |
1da177e4 LT |
335 | return ret; |
336 | } | |
337 | ||
7cf9c2c7 NP |
338 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
339 | { | |
340 | struct radix_tree_node *node = | |
341 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 342 | int i; |
643b52b9 NP |
343 | |
344 | /* | |
345 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
346 | * can leave us with a non-NULL entry in the first slot, so clear | |
347 | * that here to make sure. | |
348 | */ | |
b6dd0865 DC |
349 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
350 | tag_clear(node, i, 0); | |
351 | ||
643b52b9 | 352 | node->slots[0] = NULL; |
91d9c05a | 353 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 354 | |
7cf9c2c7 NP |
355 | kmem_cache_free(radix_tree_node_cachep, node); |
356 | } | |
357 | ||
1da177e4 LT |
358 | static inline void |
359 | radix_tree_node_free(struct radix_tree_node *node) | |
360 | { | |
7cf9c2c7 | 361 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
362 | } |
363 | ||
364 | /* | |
365 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
366 | * ensure that the addition of a single element in the tree cannot fail. On | |
367 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
368 | * with preemption not disabled. | |
b34df792 DH |
369 | * |
370 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 371 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 372 | */ |
c78c66d1 | 373 | static int __radix_tree_preload(gfp_t gfp_mask, int nr) |
1da177e4 LT |
374 | { |
375 | struct radix_tree_preload *rtp; | |
376 | struct radix_tree_node *node; | |
377 | int ret = -ENOMEM; | |
378 | ||
05eb6e72 VD |
379 | /* |
380 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
381 | * they should never be accounted to any particular memory cgroup. | |
382 | */ | |
383 | gfp_mask &= ~__GFP_ACCOUNT; | |
384 | ||
1da177e4 | 385 | preempt_disable(); |
7c8e0181 | 386 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 387 | while (rtp->nr < nr) { |
1da177e4 | 388 | preempt_enable(); |
488514d1 | 389 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
390 | if (node == NULL) |
391 | goto out; | |
392 | preempt_disable(); | |
7c8e0181 | 393 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 394 | if (rtp->nr < nr) { |
9d2a8da0 KS |
395 | node->private_data = rtp->nodes; |
396 | rtp->nodes = node; | |
397 | rtp->nr++; | |
398 | } else { | |
1da177e4 | 399 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 400 | } |
1da177e4 LT |
401 | } |
402 | ret = 0; | |
403 | out: | |
404 | return ret; | |
405 | } | |
5e4c0d97 JK |
406 | |
407 | /* | |
408 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
409 | * ensure that the addition of a single element in the tree cannot fail. On | |
410 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
411 | * with preemption not disabled. | |
412 | * | |
413 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 414 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
415 | */ |
416 | int radix_tree_preload(gfp_t gfp_mask) | |
417 | { | |
418 | /* Warn on non-sensical use... */ | |
d0164adc | 419 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 420 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 421 | } |
d7f0923d | 422 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 423 | |
5e4c0d97 JK |
424 | /* |
425 | * The same as above function, except we don't guarantee preloading happens. | |
426 | * We do it, if we decide it helps. On success, return zero with preemption | |
427 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
428 | */ | |
429 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
430 | { | |
d0164adc | 431 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 432 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
433 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
434 | preempt_disable(); | |
435 | return 0; | |
436 | } | |
437 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
438 | ||
c78c66d1 KS |
439 | /* |
440 | * The same as function above, but preload number of nodes required to insert | |
441 | * (1 << order) continuous naturally-aligned elements. | |
442 | */ | |
443 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
444 | { | |
445 | unsigned long nr_subtrees; | |
446 | int nr_nodes, subtree_height; | |
447 | ||
448 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
449 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
450 | preempt_disable(); | |
451 | return 0; | |
452 | } | |
453 | ||
454 | /* | |
455 | * Calculate number and height of fully populated subtrees it takes to | |
456 | * store (1 << order) elements. | |
457 | */ | |
458 | nr_subtrees = 1 << order; | |
459 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
460 | subtree_height++) | |
461 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
462 | ||
463 | /* | |
464 | * The worst case is zero height tree with a single item at index 0 and | |
465 | * then inserting items starting at ULONG_MAX - (1 << order). | |
466 | * | |
467 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
468 | * 0-index item. | |
469 | */ | |
470 | nr_nodes = RADIX_TREE_MAX_PATH; | |
471 | ||
472 | /* Plus branch to fully populated subtrees. */ | |
473 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
474 | ||
475 | /* Root node is shared. */ | |
476 | nr_nodes--; | |
477 | ||
478 | /* Plus nodes required to build subtrees. */ | |
479 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
480 | ||
481 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
482 | } | |
483 | ||
1456a439 MW |
484 | static unsigned radix_tree_load_root(struct radix_tree_root *root, |
485 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
486 | { | |
487 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
488 | ||
489 | *nodep = node; | |
490 | ||
b194d16c | 491 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 492 | node = entry_to_node(node); |
1456a439 | 493 | *maxindex = node_maxindex(node); |
c12e51b0 | 494 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
495 | } |
496 | ||
497 | *maxindex = 0; | |
498 | return 0; | |
499 | } | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * Extend a radix tree so it can store key @index. | |
503 | */ | |
e6145236 | 504 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 505 | unsigned long index, unsigned int shift) |
1da177e4 | 506 | { |
e2bdb933 | 507 | struct radix_tree_node *slot; |
d0891265 | 508 | unsigned int maxshift; |
1da177e4 LT |
509 | int tag; |
510 | ||
d0891265 MW |
511 | /* Figure out what the shift should be. */ |
512 | maxshift = shift; | |
513 | while (index > shift_maxindex(maxshift)) | |
514 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 515 | |
d0891265 MW |
516 | slot = root->rnode; |
517 | if (!slot) | |
1da177e4 | 518 | goto out; |
1da177e4 | 519 | |
1da177e4 | 520 | do { |
2fcd9005 MW |
521 | struct radix_tree_node *node = radix_tree_node_alloc(root); |
522 | ||
523 | if (!node) | |
1da177e4 LT |
524 | return -ENOMEM; |
525 | ||
1da177e4 | 526 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 527 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 528 | if (root_tag_get(root, tag)) |
1da177e4 LT |
529 | tag_set(node, tag, 0); |
530 | } | |
531 | ||
d0891265 MW |
532 | BUG_ON(shift > BITS_PER_LONG); |
533 | node->shift = shift; | |
0c7fa0a8 | 534 | node->offset = 0; |
1da177e4 | 535 | node->count = 1; |
e2bdb933 | 536 | node->parent = NULL; |
f7942430 | 537 | if (radix_tree_is_internal_node(slot)) { |
4dd6c098 | 538 | entry_to_node(slot)->parent = node; |
f7942430 JW |
539 | } else { |
540 | /* Moving an exceptional root->rnode to a node */ | |
541 | if (radix_tree_exceptional_entry(slot)) | |
542 | node->exceptional = 1; | |
543 | } | |
e2bdb933 | 544 | node->slots[0] = slot; |
a4db4dce MW |
545 | slot = node_to_entry(node); |
546 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 547 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 548 | } while (shift <= maxshift); |
1da177e4 | 549 | out: |
d0891265 | 550 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
551 | } |
552 | ||
f4b109c6 JW |
553 | /** |
554 | * radix_tree_shrink - shrink radix tree to minimum height | |
555 | * @root radix tree root | |
556 | */ | |
14b46879 | 557 | static inline void radix_tree_shrink(struct radix_tree_root *root, |
4d693d08 JW |
558 | radix_tree_update_node_t update_node, |
559 | void *private) | |
f4b109c6 | 560 | { |
f4b109c6 JW |
561 | for (;;) { |
562 | struct radix_tree_node *node = root->rnode; | |
563 | struct radix_tree_node *child; | |
564 | ||
565 | if (!radix_tree_is_internal_node(node)) | |
566 | break; | |
567 | node = entry_to_node(node); | |
568 | ||
569 | /* | |
570 | * The candidate node has more than one child, or its child | |
571 | * is not at the leftmost slot, or the child is a multiorder | |
572 | * entry, we cannot shrink. | |
573 | */ | |
574 | if (node->count != 1) | |
575 | break; | |
576 | child = node->slots[0]; | |
577 | if (!child) | |
578 | break; | |
579 | if (!radix_tree_is_internal_node(child) && node->shift) | |
580 | break; | |
581 | ||
582 | if (radix_tree_is_internal_node(child)) | |
583 | entry_to_node(child)->parent = NULL; | |
584 | ||
585 | /* | |
586 | * We don't need rcu_assign_pointer(), since we are simply | |
587 | * moving the node from one part of the tree to another: if it | |
588 | * was safe to dereference the old pointer to it | |
589 | * (node->slots[0]), it will be safe to dereference the new | |
590 | * one (root->rnode) as far as dependent read barriers go. | |
591 | */ | |
592 | root->rnode = child; | |
593 | ||
594 | /* | |
595 | * We have a dilemma here. The node's slot[0] must not be | |
596 | * NULLed in case there are concurrent lookups expecting to | |
597 | * find the item. However if this was a bottom-level node, | |
598 | * then it may be subject to the slot pointer being visible | |
599 | * to callers dereferencing it. If item corresponding to | |
600 | * slot[0] is subsequently deleted, these callers would expect | |
601 | * their slot to become empty sooner or later. | |
602 | * | |
603 | * For example, lockless pagecache will look up a slot, deref | |
604 | * the page pointer, and if the page has 0 refcount it means it | |
605 | * was concurrently deleted from pagecache so try the deref | |
606 | * again. Fortunately there is already a requirement for logic | |
607 | * to retry the entire slot lookup -- the indirect pointer | |
608 | * problem (replacing direct root node with an indirect pointer | |
609 | * also results in a stale slot). So tag the slot as indirect | |
610 | * to force callers to retry. | |
611 | */ | |
4d693d08 JW |
612 | node->count = 0; |
613 | if (!radix_tree_is_internal_node(child)) { | |
f4b109c6 | 614 | node->slots[0] = RADIX_TREE_RETRY; |
4d693d08 JW |
615 | if (update_node) |
616 | update_node(node, private); | |
617 | } | |
f4b109c6 JW |
618 | |
619 | radix_tree_node_free(node); | |
f4b109c6 | 620 | } |
f4b109c6 JW |
621 | } |
622 | ||
14b46879 | 623 | static void delete_node(struct radix_tree_root *root, |
4d693d08 JW |
624 | struct radix_tree_node *node, |
625 | radix_tree_update_node_t update_node, void *private) | |
f4b109c6 | 626 | { |
f4b109c6 JW |
627 | do { |
628 | struct radix_tree_node *parent; | |
629 | ||
630 | if (node->count) { | |
631 | if (node == entry_to_node(root->rnode)) | |
14b46879 JW |
632 | radix_tree_shrink(root, update_node, private); |
633 | return; | |
f4b109c6 JW |
634 | } |
635 | ||
636 | parent = node->parent; | |
637 | if (parent) { | |
638 | parent->slots[node->offset] = NULL; | |
639 | parent->count--; | |
640 | } else { | |
641 | root_tag_clear_all(root); | |
642 | root->rnode = NULL; | |
643 | } | |
644 | ||
645 | radix_tree_node_free(node); | |
f4b109c6 JW |
646 | |
647 | node = parent; | |
648 | } while (node); | |
f4b109c6 JW |
649 | } |
650 | ||
1da177e4 | 651 | /** |
139e5616 | 652 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
653 | * @root: radix tree root |
654 | * @index: index key | |
e6145236 | 655 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
656 | * @nodep: returns node |
657 | * @slotp: returns slot | |
1da177e4 | 658 | * |
139e5616 JW |
659 | * Create, if necessary, and return the node and slot for an item |
660 | * at position @index in the radix tree @root. | |
661 | * | |
662 | * Until there is more than one item in the tree, no nodes are | |
663 | * allocated and @root->rnode is used as a direct slot instead of | |
664 | * pointing to a node, in which case *@nodep will be NULL. | |
665 | * | |
666 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 667 | */ |
139e5616 | 668 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
669 | unsigned order, struct radix_tree_node **nodep, |
670 | void ***slotp) | |
1da177e4 | 671 | { |
89148aa4 MW |
672 | struct radix_tree_node *node = NULL, *child; |
673 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 674 | unsigned long maxindex; |
89148aa4 | 675 | unsigned int shift, offset = 0; |
49ea6ebc MW |
676 | unsigned long max = index | ((1UL << order) - 1); |
677 | ||
89148aa4 | 678 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
679 | |
680 | /* Make sure the tree is high enough. */ | |
49ea6ebc | 681 | if (max > maxindex) { |
d0891265 | 682 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 683 | if (error < 0) |
1da177e4 | 684 | return error; |
49ea6ebc | 685 | shift = error; |
89148aa4 | 686 | child = root->rnode; |
d0891265 | 687 | if (order == shift) |
49ea6ebc | 688 | shift += RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
689 | } |
690 | ||
e6145236 | 691 | while (shift > order) { |
c12e51b0 | 692 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 693 | if (child == NULL) { |
1da177e4 | 694 | /* Have to add a child node. */ |
89148aa4 MW |
695 | child = radix_tree_node_alloc(root); |
696 | if (!child) | |
1da177e4 | 697 | return -ENOMEM; |
89148aa4 MW |
698 | child->shift = shift; |
699 | child->offset = offset; | |
700 | child->parent = node; | |
701 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
702 | if (node) | |
1da177e4 | 703 | node->count++; |
89148aa4 | 704 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 705 | break; |
1da177e4 LT |
706 | |
707 | /* Go a level down */ | |
89148aa4 | 708 | node = entry_to_node(child); |
9e85d811 | 709 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 710 | slot = &node->slots[offset]; |
e6145236 MW |
711 | } |
712 | ||
57578c2e | 713 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 714 | /* Insert pointers to the canonical entry */ |
3b8c00f6 | 715 | if (order > shift) { |
89148aa4 | 716 | unsigned i, n = 1 << (order - shift); |
e6145236 | 717 | offset = offset & ~(n - 1); |
89148aa4 MW |
718 | slot = &node->slots[offset]; |
719 | child = node_to_entry(slot); | |
e6145236 | 720 | for (i = 0; i < n; i++) { |
89148aa4 | 721 | if (slot[i]) |
e6145236 MW |
722 | return -EEXIST; |
723 | } | |
724 | ||
725 | for (i = 1; i < n; i++) { | |
89148aa4 | 726 | rcu_assign_pointer(slot[i], child); |
e6145236 MW |
727 | node->count++; |
728 | } | |
612d6c19 | 729 | } |
57578c2e | 730 | #endif |
1da177e4 | 731 | |
139e5616 JW |
732 | if (nodep) |
733 | *nodep = node; | |
734 | if (slotp) | |
89148aa4 | 735 | *slotp = slot; |
139e5616 JW |
736 | return 0; |
737 | } | |
738 | ||
739 | /** | |
e6145236 | 740 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
741 | * @root: radix tree root |
742 | * @index: index key | |
e6145236 | 743 | * @order: key covers the 2^order indices around index |
139e5616 JW |
744 | * @item: item to insert |
745 | * | |
746 | * Insert an item into the radix tree at position @index. | |
747 | */ | |
e6145236 MW |
748 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
749 | unsigned order, void *item) | |
139e5616 JW |
750 | { |
751 | struct radix_tree_node *node; | |
752 | void **slot; | |
753 | int error; | |
754 | ||
b194d16c | 755 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 756 | |
e6145236 | 757 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
758 | if (error) |
759 | return error; | |
760 | if (*slot != NULL) | |
1da177e4 | 761 | return -EEXIST; |
139e5616 | 762 | rcu_assign_pointer(*slot, item); |
201b6264 | 763 | |
612d6c19 | 764 | if (node) { |
7b60e9ad | 765 | unsigned offset = get_slot_offset(node, slot); |
612d6c19 | 766 | node->count++; |
f7942430 JW |
767 | if (radix_tree_exceptional_entry(item)) |
768 | node->exceptional++; | |
7b60e9ad MW |
769 | BUG_ON(tag_get(node, 0, offset)); |
770 | BUG_ON(tag_get(node, 1, offset)); | |
771 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 772 | } else { |
7b60e9ad | 773 | BUG_ON(root_tags_get(root)); |
612d6c19 | 774 | } |
1da177e4 | 775 | |
1da177e4 LT |
776 | return 0; |
777 | } | |
e6145236 | 778 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 779 | |
139e5616 JW |
780 | /** |
781 | * __radix_tree_lookup - lookup an item in a radix tree | |
782 | * @root: radix tree root | |
783 | * @index: index key | |
784 | * @nodep: returns node | |
785 | * @slotp: returns slot | |
786 | * | |
787 | * Lookup and return the item at position @index in the radix | |
788 | * tree @root. | |
789 | * | |
790 | * Until there is more than one item in the tree, no nodes are | |
791 | * allocated and @root->rnode is used as a direct slot instead of | |
792 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 793 | */ |
139e5616 JW |
794 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
795 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 796 | { |
139e5616 | 797 | struct radix_tree_node *node, *parent; |
85829954 | 798 | unsigned long maxindex; |
139e5616 | 799 | void **slot; |
612d6c19 | 800 | |
85829954 MW |
801 | restart: |
802 | parent = NULL; | |
803 | slot = (void **)&root->rnode; | |
9e85d811 | 804 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 805 | if (index > maxindex) |
1da177e4 LT |
806 | return NULL; |
807 | ||
b194d16c | 808 | while (radix_tree_is_internal_node(node)) { |
85829954 | 809 | unsigned offset; |
1da177e4 | 810 | |
85829954 MW |
811 | if (node == RADIX_TREE_RETRY) |
812 | goto restart; | |
4dd6c098 | 813 | parent = entry_to_node(node); |
9e85d811 | 814 | offset = radix_tree_descend(parent, &node, index); |
85829954 MW |
815 | slot = parent->slots + offset; |
816 | } | |
1da177e4 | 817 | |
139e5616 JW |
818 | if (nodep) |
819 | *nodep = parent; | |
820 | if (slotp) | |
821 | *slotp = slot; | |
822 | return node; | |
b72b71c6 HS |
823 | } |
824 | ||
825 | /** | |
826 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
827 | * @root: radix tree root | |
828 | * @index: index key | |
829 | * | |
830 | * Returns: the slot corresponding to the position @index in the | |
831 | * radix tree @root. This is useful for update-if-exists operations. | |
832 | * | |
833 | * This function can be called under rcu_read_lock iff the slot is not | |
834 | * modified by radix_tree_replace_slot, otherwise it must be called | |
835 | * exclusive from other writers. Any dereference of the slot must be done | |
836 | * using radix_tree_deref_slot. | |
837 | */ | |
838 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
839 | { | |
139e5616 JW |
840 | void **slot; |
841 | ||
842 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
843 | return NULL; | |
844 | return slot; | |
a4331366 | 845 | } |
a4331366 HR |
846 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
847 | ||
848 | /** | |
849 | * radix_tree_lookup - perform lookup operation on a radix tree | |
850 | * @root: radix tree root | |
851 | * @index: index key | |
852 | * | |
853 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
854 | * |
855 | * This function can be called under rcu_read_lock, however the caller | |
856 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
857 | * them safely). No RCU barriers are required to access or modify the | |
858 | * returned item, however. | |
a4331366 HR |
859 | */ |
860 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
861 | { | |
139e5616 | 862 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
863 | } |
864 | EXPORT_SYMBOL(radix_tree_lookup); | |
865 | ||
6d75f366 JW |
866 | static void replace_slot(struct radix_tree_root *root, |
867 | struct radix_tree_node *node, | |
868 | void **slot, void *item, | |
869 | bool warn_typeswitch) | |
f7942430 JW |
870 | { |
871 | void *old = rcu_dereference_raw(*slot); | |
f4b109c6 | 872 | int count, exceptional; |
f7942430 JW |
873 | |
874 | WARN_ON_ONCE(radix_tree_is_internal_node(item)); | |
f7942430 | 875 | |
f4b109c6 | 876 | count = !!item - !!old; |
f7942430 JW |
877 | exceptional = !!radix_tree_exceptional_entry(item) - |
878 | !!radix_tree_exceptional_entry(old); | |
879 | ||
f4b109c6 | 880 | WARN_ON_ONCE(warn_typeswitch && (count || exceptional)); |
f7942430 | 881 | |
f4b109c6 JW |
882 | if (node) { |
883 | node->count += count; | |
f7942430 | 884 | node->exceptional += exceptional; |
f4b109c6 | 885 | } |
f7942430 JW |
886 | |
887 | rcu_assign_pointer(*slot, item); | |
888 | } | |
889 | ||
6d75f366 JW |
890 | /** |
891 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
892 | * @root: radix tree root |
893 | * @node: pointer to tree node | |
894 | * @slot: pointer to slot in @node | |
895 | * @item: new item to store in the slot. | |
896 | * @update_node: callback for changing leaf nodes | |
897 | * @private: private data to pass to @update_node | |
6d75f366 JW |
898 | * |
899 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
900 | * across slot lookup and replacement. | |
901 | */ | |
902 | void __radix_tree_replace(struct radix_tree_root *root, | |
903 | struct radix_tree_node *node, | |
4d693d08 JW |
904 | void **slot, void *item, |
905 | radix_tree_update_node_t update_node, void *private) | |
6d75f366 JW |
906 | { |
907 | /* | |
f4b109c6 JW |
908 | * This function supports replacing exceptional entries and |
909 | * deleting entries, but that needs accounting against the | |
910 | * node unless the slot is root->rnode. | |
6d75f366 JW |
911 | */ |
912 | replace_slot(root, node, slot, item, | |
913 | !node && slot != (void **)&root->rnode); | |
f4b109c6 | 914 | |
4d693d08 JW |
915 | if (!node) |
916 | return; | |
917 | ||
918 | if (update_node) | |
919 | update_node(node, private); | |
920 | ||
921 | delete_node(root, node, update_node, private); | |
6d75f366 JW |
922 | } |
923 | ||
924 | /** | |
925 | * radix_tree_replace_slot - replace item in a slot | |
926 | * @root: radix tree root | |
927 | * @slot: pointer to slot | |
928 | * @item: new item to store in the slot. | |
929 | * | |
930 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
931 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
932 | * across slot lookup and replacement. | |
933 | * | |
934 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
935 | * regular entries, and exceptional entries, as that requires accounting | |
f4b109c6 JW |
936 | * inside the radix tree node. When switching from one type of entry or |
937 | * deleting, use __radix_tree_lookup() and __radix_tree_replace(). | |
6d75f366 JW |
938 | */ |
939 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
940 | void **slot, void *item) | |
941 | { | |
942 | replace_slot(root, NULL, slot, item, true); | |
943 | } | |
944 | ||
1da177e4 LT |
945 | /** |
946 | * radix_tree_tag_set - set a tag on a radix tree node | |
947 | * @root: radix tree root | |
948 | * @index: index key | |
2fcd9005 | 949 | * @tag: tag index |
1da177e4 | 950 | * |
daff89f3 JC |
951 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
952 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
953 | * the root all the way down to the leaf node. |
954 | * | |
2fcd9005 | 955 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
956 | * item is a bug. |
957 | */ | |
958 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 959 | unsigned long index, unsigned int tag) |
1da177e4 | 960 | { |
fb969909 RZ |
961 | struct radix_tree_node *node, *parent; |
962 | unsigned long maxindex; | |
1da177e4 | 963 | |
9e85d811 | 964 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 965 | BUG_ON(index > maxindex); |
1da177e4 | 966 | |
b194d16c | 967 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 968 | unsigned offset; |
1da177e4 | 969 | |
4dd6c098 | 970 | parent = entry_to_node(node); |
9e85d811 | 971 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
972 | BUG_ON(!node); |
973 | ||
974 | if (!tag_get(parent, tag, offset)) | |
975 | tag_set(parent, tag, offset); | |
1da177e4 LT |
976 | } |
977 | ||
612d6c19 | 978 | /* set the root's tag bit */ |
fb969909 | 979 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
980 | root_tag_set(root, tag); |
981 | ||
fb969909 | 982 | return node; |
1da177e4 LT |
983 | } |
984 | EXPORT_SYMBOL(radix_tree_tag_set); | |
985 | ||
d604c324 MW |
986 | static void node_tag_clear(struct radix_tree_root *root, |
987 | struct radix_tree_node *node, | |
988 | unsigned int tag, unsigned int offset) | |
989 | { | |
990 | while (node) { | |
991 | if (!tag_get(node, tag, offset)) | |
992 | return; | |
993 | tag_clear(node, tag, offset); | |
994 | if (any_tag_set(node, tag)) | |
995 | return; | |
996 | ||
997 | offset = node->offset; | |
998 | node = node->parent; | |
999 | } | |
1000 | ||
1001 | /* clear the root's tag bit */ | |
1002 | if (root_tag_get(root, tag)) | |
1003 | root_tag_clear(root, tag); | |
1004 | } | |
1005 | ||
9498d2bb MW |
1006 | static void node_tag_set(struct radix_tree_root *root, |
1007 | struct radix_tree_node *node, | |
1008 | unsigned int tag, unsigned int offset) | |
1009 | { | |
1010 | while (node) { | |
1011 | if (tag_get(node, tag, offset)) | |
1012 | return; | |
1013 | tag_set(node, tag, offset); | |
1014 | offset = node->offset; | |
1015 | node = node->parent; | |
1016 | } | |
1017 | ||
1018 | if (!root_tag_get(root, tag)) | |
1019 | root_tag_set(root, tag); | |
1020 | } | |
1021 | ||
268f42de MW |
1022 | /** |
1023 | * radix_tree_iter_tag_set - set a tag on the current iterator entry | |
1024 | * @root: radix tree root | |
1025 | * @iter: iterator state | |
1026 | * @tag: tag to set | |
1027 | */ | |
1028 | void radix_tree_iter_tag_set(struct radix_tree_root *root, | |
1029 | const struct radix_tree_iter *iter, unsigned int tag) | |
1030 | { | |
1031 | node_tag_set(root, iter->node, tag, iter_offset(iter)); | |
1032 | } | |
1033 | ||
1da177e4 LT |
1034 | /** |
1035 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1036 | * @root: radix tree root | |
1037 | * @index: index key | |
2fcd9005 | 1038 | * @tag: tag index |
1da177e4 | 1039 | * |
daff89f3 | 1040 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1041 | * corresponding to @index in the radix tree. If this causes |
1042 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1043 | * next-to-leaf node, etc. |
1044 | * | |
1045 | * Returns the address of the tagged item on success, else NULL. ie: | |
1046 | * has the same return value and semantics as radix_tree_lookup(). | |
1047 | */ | |
1048 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1049 | unsigned long index, unsigned int tag) |
1da177e4 | 1050 | { |
00f47b58 RZ |
1051 | struct radix_tree_node *node, *parent; |
1052 | unsigned long maxindex; | |
e2bdb933 | 1053 | int uninitialized_var(offset); |
1da177e4 | 1054 | |
9e85d811 | 1055 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1056 | if (index > maxindex) |
1057 | return NULL; | |
1da177e4 | 1058 | |
00f47b58 | 1059 | parent = NULL; |
1da177e4 | 1060 | |
b194d16c | 1061 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1062 | parent = entry_to_node(node); |
9e85d811 | 1063 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1064 | } |
1065 | ||
d604c324 MW |
1066 | if (node) |
1067 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1068 | |
00f47b58 | 1069 | return node; |
1da177e4 LT |
1070 | } |
1071 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1072 | ||
1da177e4 | 1073 | /** |
32605a18 MT |
1074 | * radix_tree_tag_get - get a tag on a radix tree node |
1075 | * @root: radix tree root | |
1076 | * @index: index key | |
2fcd9005 | 1077 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1078 | * |
32605a18 | 1079 | * Return values: |
1da177e4 | 1080 | * |
612d6c19 NP |
1081 | * 0: tag not present or not set |
1082 | * 1: tag set | |
ce82653d DH |
1083 | * |
1084 | * Note that the return value of this function may not be relied on, even if | |
1085 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1086 | * from concurrency. | |
1da177e4 LT |
1087 | */ |
1088 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 1089 | unsigned long index, unsigned int tag) |
1da177e4 | 1090 | { |
4589ba6d RZ |
1091 | struct radix_tree_node *node, *parent; |
1092 | unsigned long maxindex; | |
1da177e4 | 1093 | |
612d6c19 NP |
1094 | if (!root_tag_get(root, tag)) |
1095 | return 0; | |
1096 | ||
9e85d811 | 1097 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1098 | if (index > maxindex) |
1099 | return 0; | |
7cf9c2c7 NP |
1100 | if (node == NULL) |
1101 | return 0; | |
1102 | ||
b194d16c | 1103 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1104 | unsigned offset; |
1da177e4 | 1105 | |
4dd6c098 | 1106 | parent = entry_to_node(node); |
9e85d811 | 1107 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1108 | |
4589ba6d | 1109 | if (!node) |
1da177e4 | 1110 | return 0; |
4589ba6d | 1111 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1112 | return 0; |
4589ba6d RZ |
1113 | if (node == RADIX_TREE_RETRY) |
1114 | break; | |
1da177e4 | 1115 | } |
4589ba6d RZ |
1116 | |
1117 | return 1; | |
1da177e4 LT |
1118 | } |
1119 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1120 | |
21ef5339 RZ |
1121 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1122 | unsigned int shift) | |
1123 | { | |
1124 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1125 | iter->shift = shift; | |
1126 | #endif | |
1127 | } | |
1128 | ||
148deab2 MW |
1129 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1130 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1131 | struct radix_tree_node *node, unsigned offset, | |
1132 | unsigned tag) | |
1133 | { | |
1134 | unsigned tag_long = offset / BITS_PER_LONG; | |
1135 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1136 | ||
1137 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1138 | ||
1139 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1140 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1141 | /* Pick tags from next element */ | |
1142 | if (tag_bit) | |
1143 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1144 | (BITS_PER_LONG - tag_bit); | |
1145 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1146 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1151 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1152 | void **slot, struct radix_tree_iter *iter) | |
1153 | { | |
1154 | void *sib = node_to_entry(slot - 1); | |
1155 | ||
1156 | while (iter->index < iter->next_index) { | |
1157 | *nodep = rcu_dereference_raw(*slot); | |
1158 | if (*nodep && *nodep != sib) | |
1159 | return slot; | |
1160 | slot++; | |
1161 | iter->index = __radix_tree_iter_add(iter, 1); | |
1162 | iter->tags >>= 1; | |
1163 | } | |
1164 | ||
1165 | *nodep = NULL; | |
1166 | return NULL; | |
1167 | } | |
1168 | ||
1169 | void ** __radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, | |
1170 | unsigned flags) | |
1171 | { | |
1172 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
1173 | struct radix_tree_node *node = rcu_dereference_raw(*slot); | |
1174 | ||
1175 | slot = skip_siblings(&node, slot, iter); | |
1176 | ||
1177 | while (radix_tree_is_internal_node(node)) { | |
1178 | unsigned offset; | |
1179 | unsigned long next_index; | |
1180 | ||
1181 | if (node == RADIX_TREE_RETRY) | |
1182 | return slot; | |
1183 | node = entry_to_node(node); | |
268f42de | 1184 | iter->node = node; |
148deab2 MW |
1185 | iter->shift = node->shift; |
1186 | ||
1187 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1188 | offset = radix_tree_find_next_bit(node, tag, 0); | |
1189 | if (offset == RADIX_TREE_MAP_SIZE) | |
1190 | return NULL; | |
1191 | slot = &node->slots[offset]; | |
1192 | iter->index = __radix_tree_iter_add(iter, offset); | |
1193 | set_iter_tags(iter, node, offset, tag); | |
1194 | node = rcu_dereference_raw(*slot); | |
1195 | } else { | |
1196 | offset = 0; | |
1197 | slot = &node->slots[0]; | |
1198 | for (;;) { | |
1199 | node = rcu_dereference_raw(*slot); | |
1200 | if (node) | |
1201 | break; | |
1202 | slot++; | |
1203 | offset++; | |
1204 | if (offset == RADIX_TREE_MAP_SIZE) | |
1205 | return NULL; | |
1206 | } | |
1207 | iter->index = __radix_tree_iter_add(iter, offset); | |
1208 | } | |
1209 | if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0)) | |
1210 | goto none; | |
1211 | next_index = (iter->index | shift_maxindex(iter->shift)) + 1; | |
1212 | if (next_index < iter->next_index) | |
1213 | iter->next_index = next_index; | |
1214 | } | |
1215 | ||
1216 | return slot; | |
1217 | none: | |
1218 | iter->next_index = 0; | |
1219 | return NULL; | |
1220 | } | |
1221 | EXPORT_SYMBOL(__radix_tree_next_slot); | |
1222 | #else | |
1223 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1224 | void **slot, struct radix_tree_iter *iter) | |
1225 | { | |
1226 | return slot; | |
1227 | } | |
1228 | #endif | |
1229 | ||
1230 | void **radix_tree_iter_resume(void **slot, struct radix_tree_iter *iter) | |
1231 | { | |
1232 | struct radix_tree_node *node; | |
1233 | ||
1234 | slot++; | |
1235 | iter->index = __radix_tree_iter_add(iter, 1); | |
1236 | node = rcu_dereference_raw(*slot); | |
1237 | skip_siblings(&node, slot, iter); | |
1238 | iter->next_index = iter->index; | |
1239 | iter->tags = 0; | |
1240 | return NULL; | |
1241 | } | |
1242 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1243 | ||
78c1d784 KK |
1244 | /** |
1245 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1246 | * | |
1247 | * @root: radix tree root | |
1248 | * @iter: iterator state | |
1249 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1250 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1251 | */ | |
1252 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
1253 | struct radix_tree_iter *iter, unsigned flags) | |
1254 | { | |
9e85d811 | 1255 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1256 | struct radix_tree_node *node, *child; |
21ef5339 | 1257 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1258 | |
1259 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1260 | return NULL; | |
1261 | ||
1262 | /* | |
1263 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1264 | * during iterating; it can be zero only at the beginning. | |
1265 | * And we cannot overflow iter->next_index in a single step, | |
1266 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1267 | * |
1268 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1269 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1270 | */ |
1271 | index = iter->next_index; | |
1272 | if (!index && iter->index) | |
1273 | return NULL; | |
1274 | ||
21ef5339 | 1275 | restart: |
9e85d811 | 1276 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1277 | if (index > maxindex) |
1278 | return NULL; | |
8c1244de MW |
1279 | if (!child) |
1280 | return NULL; | |
21ef5339 | 1281 | |
8c1244de | 1282 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1283 | /* Single-slot tree */ |
21ef5339 RZ |
1284 | iter->index = index; |
1285 | iter->next_index = maxindex + 1; | |
78c1d784 | 1286 | iter->tags = 1; |
268f42de | 1287 | iter->node = NULL; |
8c1244de | 1288 | __set_iter_shift(iter, 0); |
78c1d784 | 1289 | return (void **)&root->rnode; |
8c1244de | 1290 | } |
21ef5339 | 1291 | |
8c1244de MW |
1292 | do { |
1293 | node = entry_to_node(child); | |
9e85d811 | 1294 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1295 | |
78c1d784 | 1296 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1297 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1298 | /* Hole detected */ |
1299 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1300 | return NULL; | |
1301 | ||
1302 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1303 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1304 | offset + 1); |
1305 | else | |
1306 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
1307 | void *slot = node->slots[offset]; |
1308 | if (is_sibling_entry(node, slot)) | |
1309 | continue; | |
1310 | if (slot) | |
78c1d784 KK |
1311 | break; |
1312 | } | |
8c1244de | 1313 | index &= ~node_maxindex(node); |
9e85d811 | 1314 | index += offset << node->shift; |
78c1d784 KK |
1315 | /* Overflow after ~0UL */ |
1316 | if (!index) | |
1317 | return NULL; | |
1318 | if (offset == RADIX_TREE_MAP_SIZE) | |
1319 | goto restart; | |
8c1244de | 1320 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1321 | } |
1322 | ||
8c1244de | 1323 | if ((child == NULL) || (child == RADIX_TREE_RETRY)) |
78c1d784 | 1324 | goto restart; |
8c1244de | 1325 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
1326 | |
1327 | /* Update the iterator state */ | |
8c1244de MW |
1328 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1329 | iter->next_index = (index | node_maxindex(node)) + 1; | |
268f42de | 1330 | iter->node = node; |
9e85d811 | 1331 | __set_iter_shift(iter, node->shift); |
78c1d784 | 1332 | |
148deab2 MW |
1333 | if (flags & RADIX_TREE_ITER_TAGGED) |
1334 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1335 | |
1336 | return node->slots + offset; | |
1337 | } | |
1338 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1339 | ||
1da177e4 LT |
1340 | /** |
1341 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1342 | * @root: radix tree root | |
1343 | * @results: where the results of the lookup are placed | |
1344 | * @first_index: start the lookup from this key | |
1345 | * @max_items: place up to this many items at *results | |
1346 | * | |
1347 | * Performs an index-ascending scan of the tree for present items. Places | |
1348 | * them at *@results and returns the number of items which were placed at | |
1349 | * *@results. | |
1350 | * | |
1351 | * The implementation is naive. | |
7cf9c2c7 NP |
1352 | * |
1353 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1354 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1355 | * an atomic snapshot of the tree at a single point in time, the |
1356 | * semantics of an RCU protected gang lookup are as though multiple | |
1357 | * radix_tree_lookups have been issued in individual locks, and results | |
1358 | * stored in 'results'. | |
1da177e4 LT |
1359 | */ |
1360 | unsigned int | |
1361 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1362 | unsigned long first_index, unsigned int max_items) | |
1363 | { | |
cebbd29e KK |
1364 | struct radix_tree_iter iter; |
1365 | void **slot; | |
1366 | unsigned int ret = 0; | |
7cf9c2c7 | 1367 | |
cebbd29e | 1368 | if (unlikely(!max_items)) |
7cf9c2c7 | 1369 | return 0; |
1da177e4 | 1370 | |
cebbd29e | 1371 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1372 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1373 | if (!results[ret]) |
1374 | continue; | |
b194d16c | 1375 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1376 | slot = radix_tree_iter_retry(&iter); |
1377 | continue; | |
1378 | } | |
cebbd29e | 1379 | if (++ret == max_items) |
1da177e4 | 1380 | break; |
1da177e4 | 1381 | } |
7cf9c2c7 | 1382 | |
1da177e4 LT |
1383 | return ret; |
1384 | } | |
1385 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1386 | ||
47feff2c NP |
1387 | /** |
1388 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1389 | * @root: radix tree root | |
1390 | * @results: where the results of the lookup are placed | |
6328650b | 1391 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1392 | * @first_index: start the lookup from this key |
1393 | * @max_items: place up to this many items at *results | |
1394 | * | |
1395 | * Performs an index-ascending scan of the tree for present items. Places | |
1396 | * their slots at *@results and returns the number of items which were | |
1397 | * placed at *@results. | |
1398 | * | |
1399 | * The implementation is naive. | |
1400 | * | |
1401 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1402 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1403 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1404 | */ | |
1405 | unsigned int | |
6328650b HD |
1406 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1407 | void ***results, unsigned long *indices, | |
47feff2c NP |
1408 | unsigned long first_index, unsigned int max_items) |
1409 | { | |
cebbd29e KK |
1410 | struct radix_tree_iter iter; |
1411 | void **slot; | |
1412 | unsigned int ret = 0; | |
47feff2c | 1413 | |
cebbd29e | 1414 | if (unlikely(!max_items)) |
47feff2c NP |
1415 | return 0; |
1416 | ||
cebbd29e KK |
1417 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1418 | results[ret] = slot; | |
6328650b | 1419 | if (indices) |
cebbd29e KK |
1420 | indices[ret] = iter.index; |
1421 | if (++ret == max_items) | |
47feff2c | 1422 | break; |
47feff2c NP |
1423 | } |
1424 | ||
1425 | return ret; | |
1426 | } | |
1427 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1428 | ||
1da177e4 LT |
1429 | /** |
1430 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1431 | * based on a tag | |
1432 | * @root: radix tree root | |
1433 | * @results: where the results of the lookup are placed | |
1434 | * @first_index: start the lookup from this key | |
1435 | * @max_items: place up to this many items at *results | |
daff89f3 | 1436 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1437 | * |
1438 | * Performs an index-ascending scan of the tree for present items which | |
1439 | * have the tag indexed by @tag set. Places the items at *@results and | |
1440 | * returns the number of items which were placed at *@results. | |
1441 | */ | |
1442 | unsigned int | |
1443 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1444 | unsigned long first_index, unsigned int max_items, |
1445 | unsigned int tag) | |
1da177e4 | 1446 | { |
cebbd29e KK |
1447 | struct radix_tree_iter iter; |
1448 | void **slot; | |
1449 | unsigned int ret = 0; | |
612d6c19 | 1450 | |
cebbd29e | 1451 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1452 | return 0; |
1453 | ||
cebbd29e | 1454 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1455 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1456 | if (!results[ret]) |
1457 | continue; | |
b194d16c | 1458 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1459 | slot = radix_tree_iter_retry(&iter); |
1460 | continue; | |
1461 | } | |
cebbd29e | 1462 | if (++ret == max_items) |
1da177e4 | 1463 | break; |
1da177e4 | 1464 | } |
7cf9c2c7 | 1465 | |
1da177e4 LT |
1466 | return ret; |
1467 | } | |
1468 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1469 | ||
47feff2c NP |
1470 | /** |
1471 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1472 | * radix tree based on a tag | |
1473 | * @root: radix tree root | |
1474 | * @results: where the results of the lookup are placed | |
1475 | * @first_index: start the lookup from this key | |
1476 | * @max_items: place up to this many items at *results | |
1477 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1478 | * | |
1479 | * Performs an index-ascending scan of the tree for present items which | |
1480 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1481 | * returns the number of slots which were placed at *@results. | |
1482 | */ | |
1483 | unsigned int | |
1484 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1485 | unsigned long first_index, unsigned int max_items, | |
1486 | unsigned int tag) | |
1487 | { | |
cebbd29e KK |
1488 | struct radix_tree_iter iter; |
1489 | void **slot; | |
1490 | unsigned int ret = 0; | |
47feff2c | 1491 | |
cebbd29e | 1492 | if (unlikely(!max_items)) |
47feff2c NP |
1493 | return 0; |
1494 | ||
cebbd29e KK |
1495 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1496 | results[ret] = slot; | |
1497 | if (++ret == max_items) | |
47feff2c | 1498 | break; |
47feff2c NP |
1499 | } |
1500 | ||
1501 | return ret; | |
1502 | } | |
1503 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1504 | ||
139e5616 JW |
1505 | /** |
1506 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1507 | * @root: radix tree root | |
139e5616 JW |
1508 | * @node: node containing @index |
1509 | * | |
1510 | * After clearing the slot at @index in @node from radix tree | |
1511 | * rooted at @root, call this function to attempt freeing the | |
1512 | * node and shrinking the tree. | |
139e5616 | 1513 | */ |
14b46879 | 1514 | void __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1515 | struct radix_tree_node *node) |
1516 | { | |
14b46879 | 1517 | delete_node(root, node, NULL, NULL); |
139e5616 JW |
1518 | } |
1519 | ||
57578c2e MW |
1520 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1521 | void *ptr, unsigned offset) | |
1522 | { | |
1523 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1524 | int i; | |
1525 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1526 | if (node->slots[offset + i] != ptr) | |
1527 | break; | |
1528 | node->slots[offset + i] = NULL; | |
1529 | node->count--; | |
1530 | } | |
1531 | #endif | |
1532 | } | |
1533 | ||
1da177e4 | 1534 | /** |
53c59f26 | 1535 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1536 | * @root: radix tree root |
1537 | * @index: index key | |
53c59f26 | 1538 | * @item: expected item |
1da177e4 | 1539 | * |
53c59f26 | 1540 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1541 | * |
53c59f26 JW |
1542 | * Returns the address of the deleted item, or NULL if it was not present |
1543 | * or the entry at the given @index was not @item. | |
1da177e4 | 1544 | */ |
53c59f26 JW |
1545 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1546 | unsigned long index, void *item) | |
1da177e4 | 1547 | { |
139e5616 | 1548 | struct radix_tree_node *node; |
57578c2e | 1549 | unsigned int offset; |
139e5616 JW |
1550 | void **slot; |
1551 | void *entry; | |
d5274261 | 1552 | int tag; |
1da177e4 | 1553 | |
139e5616 JW |
1554 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1555 | if (!entry) | |
1556 | return NULL; | |
1da177e4 | 1557 | |
139e5616 JW |
1558 | if (item && entry != item) |
1559 | return NULL; | |
1560 | ||
1561 | if (!node) { | |
612d6c19 NP |
1562 | root_tag_clear_all(root); |
1563 | root->rnode = NULL; | |
139e5616 | 1564 | return entry; |
612d6c19 | 1565 | } |
1da177e4 | 1566 | |
29e0967c | 1567 | offset = get_slot_offset(node, slot); |
53c59f26 | 1568 | |
d604c324 MW |
1569 | /* Clear all tags associated with the item to be deleted. */ |
1570 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1571 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1572 | |
a4db4dce | 1573 | delete_sibling_entries(node, node_to_entry(slot), offset); |
4d693d08 | 1574 | __radix_tree_replace(root, node, slot, NULL, NULL, NULL); |
612d6c19 | 1575 | |
139e5616 | 1576 | return entry; |
1da177e4 | 1577 | } |
53c59f26 JW |
1578 | EXPORT_SYMBOL(radix_tree_delete_item); |
1579 | ||
1580 | /** | |
1581 | * radix_tree_delete - delete an item from a radix tree | |
1582 | * @root: radix tree root | |
1583 | * @index: index key | |
1584 | * | |
1585 | * Remove the item at @index from the radix tree rooted at @root. | |
1586 | * | |
1587 | * Returns the address of the deleted item, or NULL if it was not present. | |
1588 | */ | |
1589 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1590 | { | |
1591 | return radix_tree_delete_item(root, index, NULL); | |
1592 | } | |
1da177e4 LT |
1593 | EXPORT_SYMBOL(radix_tree_delete); |
1594 | ||
d3798ae8 JW |
1595 | void radix_tree_clear_tags(struct radix_tree_root *root, |
1596 | struct radix_tree_node *node, | |
1597 | void **slot) | |
d604c324 | 1598 | { |
d604c324 MW |
1599 | if (node) { |
1600 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1601 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1602 | node_tag_clear(root, node, tag, offset); | |
1603 | } else { | |
1604 | /* Clear root node tags */ | |
1605 | root->gfp_mask &= __GFP_BITS_MASK; | |
1606 | } | |
d604c324 MW |
1607 | } |
1608 | ||
1da177e4 LT |
1609 | /** |
1610 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1611 | * @root: radix tree root | |
1612 | * @tag: tag to test | |
1613 | */ | |
daff89f3 | 1614 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1615 | { |
612d6c19 | 1616 | return root_tag_get(root, tag); |
1da177e4 LT |
1617 | } |
1618 | EXPORT_SYMBOL(radix_tree_tagged); | |
1619 | ||
1620 | static void | |
449dd698 | 1621 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1622 | { |
449dd698 JW |
1623 | struct radix_tree_node *node = arg; |
1624 | ||
1625 | memset(node, 0, sizeof(*node)); | |
1626 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1627 | } |
1628 | ||
c78c66d1 KS |
1629 | static __init unsigned long __maxindex(unsigned int height) |
1630 | { | |
1631 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
1632 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1633 | ||
1634 | if (shift < 0) | |
1635 | return ~0UL; | |
1636 | if (shift >= BITS_PER_LONG) | |
1637 | return 0UL; | |
1638 | return ~0UL >> shift; | |
1639 | } | |
1640 | ||
1641 | static __init void radix_tree_init_maxnodes(void) | |
1642 | { | |
1643 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
1644 | unsigned int i, j; | |
1645 | ||
1646 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1647 | height_to_maxindex[i] = __maxindex(i); | |
1648 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
1649 | for (j = i; j > 0; j--) | |
1650 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
1651 | } | |
1652 | } | |
1653 | ||
d544abd5 | 1654 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1655 | { |
2fcd9005 MW |
1656 | struct radix_tree_preload *rtp; |
1657 | struct radix_tree_node *node; | |
1658 | ||
1659 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1660 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1661 | while (rtp->nr) { | |
1662 | node = rtp->nodes; | |
1663 | rtp->nodes = node->private_data; | |
1664 | kmem_cache_free(radix_tree_node_cachep, node); | |
1665 | rtp->nr--; | |
2fcd9005 | 1666 | } |
d544abd5 | 1667 | return 0; |
1da177e4 | 1668 | } |
1da177e4 LT |
1669 | |
1670 | void __init radix_tree_init(void) | |
1671 | { | |
d544abd5 | 1672 | int ret; |
1da177e4 LT |
1673 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
1674 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1675 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1676 | radix_tree_node_ctor); | |
c78c66d1 | 1677 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
1678 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
1679 | NULL, radix_tree_cpu_dead); | |
1680 | WARN_ON(ret < 0); | |
1da177e4 | 1681 | } |