]>
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 | |
a4db4dce | 72 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 73 | { |
30ff46cc | 74 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
75 | } |
76 | ||
a4db4dce | 77 | #define RADIX_TREE_RETRY node_to_entry(NULL) |
afe0e395 | 78 | |
db050f29 MW |
79 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
80 | /* Sibling slots point directly to another slot in the same node */ | |
81 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
82 | { | |
83 | void **ptr = node; | |
84 | return (parent->slots <= ptr) && | |
85 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
86 | } | |
87 | #else | |
88 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
89 | { | |
90 | return false; | |
91 | } | |
92 | #endif | |
93 | ||
94 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
95 | void **slot) | |
96 | { | |
97 | return slot - parent->slots; | |
98 | } | |
99 | ||
9e85d811 MW |
100 | static unsigned int radix_tree_descend(struct radix_tree_node *parent, |
101 | struct radix_tree_node **nodep, unsigned long index) | |
db050f29 | 102 | { |
9e85d811 | 103 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
db050f29 MW |
104 | void **entry = rcu_dereference_raw(parent->slots[offset]); |
105 | ||
106 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 107 | if (radix_tree_is_internal_node(entry)) { |
8d2c0d36 LT |
108 | if (is_sibling_entry(parent, entry)) { |
109 | void **sibentry = (void **) entry_to_node(entry); | |
110 | offset = get_slot_offset(parent, sibentry); | |
111 | entry = rcu_dereference_raw(*sibentry); | |
db050f29 MW |
112 | } |
113 | } | |
114 | #endif | |
115 | ||
116 | *nodep = (void *)entry; | |
117 | return offset; | |
118 | } | |
119 | ||
612d6c19 NP |
120 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
121 | { | |
122 | return root->gfp_mask & __GFP_BITS_MASK; | |
123 | } | |
124 | ||
643b52b9 NP |
125 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
126 | int offset) | |
127 | { | |
128 | __set_bit(offset, node->tags[tag]); | |
129 | } | |
130 | ||
131 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
132 | int offset) | |
133 | { | |
134 | __clear_bit(offset, node->tags[tag]); | |
135 | } | |
136 | ||
137 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
138 | int offset) | |
139 | { | |
140 | return test_bit(offset, node->tags[tag]); | |
141 | } | |
142 | ||
143 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
144 | { | |
145 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
146 | } | |
147 | ||
2fcd9005 | 148 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
149 | { |
150 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
151 | } | |
152 | ||
153 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
154 | { | |
155 | root->gfp_mask &= __GFP_BITS_MASK; | |
156 | } | |
157 | ||
158 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
159 | { | |
2fcd9005 | 160 | return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
643b52b9 NP |
161 | } |
162 | ||
7b60e9ad MW |
163 | static inline unsigned root_tags_get(struct radix_tree_root *root) |
164 | { | |
165 | return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT; | |
166 | } | |
167 | ||
643b52b9 NP |
168 | /* |
169 | * Returns 1 if any slot in the node has this tag set. | |
170 | * Otherwise returns 0. | |
171 | */ | |
172 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
173 | { | |
2fcd9005 | 174 | unsigned idx; |
643b52b9 NP |
175 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
176 | if (node->tags[tag][idx]) | |
177 | return 1; | |
178 | } | |
179 | return 0; | |
180 | } | |
78c1d784 KK |
181 | |
182 | /** | |
183 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
184 | * | |
185 | * @addr: The address to base the search on | |
186 | * @size: The bitmap size in bits | |
187 | * @offset: The bitnumber to start searching at | |
188 | * | |
189 | * Unrollable variant of find_next_bit() for constant size arrays. | |
190 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
191 | * Returns next bit offset, or size if nothing found. | |
192 | */ | |
193 | static __always_inline unsigned long | |
bc412fca MW |
194 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
195 | unsigned long offset) | |
78c1d784 | 196 | { |
bc412fca | 197 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 198 | |
bc412fca | 199 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
200 | unsigned long tmp; |
201 | ||
202 | addr += offset / BITS_PER_LONG; | |
203 | tmp = *addr >> (offset % BITS_PER_LONG); | |
204 | if (tmp) | |
205 | return __ffs(tmp) + offset; | |
206 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 207 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
208 | tmp = *++addr; |
209 | if (tmp) | |
210 | return __ffs(tmp) + offset; | |
211 | offset += BITS_PER_LONG; | |
212 | } | |
213 | } | |
bc412fca | 214 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
215 | } |
216 | ||
0796c583 | 217 | #ifndef __KERNEL__ |
d0891265 | 218 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 219 | { |
0796c583 | 220 | unsigned long i; |
7cf19af4 | 221 | |
f7942430 | 222 | pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d exceptional %d parent %p\n", |
0c7fa0a8 | 223 | node, node->offset, |
0796c583 | 224 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
f7942430 | 225 | node->shift, node->count, node->exceptional, node->parent); |
0796c583 RZ |
226 | |
227 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
228 | unsigned long first = index | (i << node->shift); |
229 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
230 | void *entry = node->slots[i]; |
231 | if (!entry) | |
232 | continue; | |
233 | if (is_sibling_entry(node, entry)) { | |
234 | pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n", | |
235 | entry, i, | |
4dd6c098 | 236 | *(void **)entry_to_node(entry), |
0796c583 | 237 | first, last); |
b194d16c | 238 | } else if (!radix_tree_is_internal_node(entry)) { |
0796c583 RZ |
239 | pr_debug("radix entry %p offset %ld indices %ld-%ld\n", |
240 | entry, i, first, last); | |
241 | } else { | |
4dd6c098 | 242 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
243 | } |
244 | } | |
7cf19af4 MW |
245 | } |
246 | ||
247 | /* For debug */ | |
248 | static void radix_tree_dump(struct radix_tree_root *root) | |
249 | { | |
d0891265 MW |
250 | pr_debug("radix root: %p rnode %p tags %x\n", |
251 | root, root->rnode, | |
7cf19af4 | 252 | root->gfp_mask >> __GFP_BITS_SHIFT); |
b194d16c | 253 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 254 | return; |
4dd6c098 | 255 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 MW |
256 | } |
257 | #endif | |
258 | ||
1da177e4 LT |
259 | /* |
260 | * This assumes that the caller has performed appropriate preallocation, and | |
261 | * that the caller has pinned this thread of control to the current CPU. | |
262 | */ | |
263 | static struct radix_tree_node * | |
264 | radix_tree_node_alloc(struct radix_tree_root *root) | |
265 | { | |
e2848a0e | 266 | struct radix_tree_node *ret = NULL; |
612d6c19 | 267 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 268 | |
5e4c0d97 | 269 | /* |
2fcd9005 MW |
270 | * Preload code isn't irq safe and it doesn't make sense to use |
271 | * preloading during an interrupt anyway as all the allocations have | |
272 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 273 | */ |
d0164adc | 274 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
275 | struct radix_tree_preload *rtp; |
276 | ||
58e698af VD |
277 | /* |
278 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
279 | * cache first for the new node to get accounted to the memory |
280 | * cgroup. | |
58e698af VD |
281 | */ |
282 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 283 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
284 | if (ret) |
285 | goto out; | |
286 | ||
e2848a0e NP |
287 | /* |
288 | * Provided the caller has preloaded here, we will always | |
289 | * succeed in getting a node here (and never reach | |
290 | * kmem_cache_alloc) | |
291 | */ | |
7c8e0181 | 292 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 293 | if (rtp->nr) { |
9d2a8da0 KS |
294 | ret = rtp->nodes; |
295 | rtp->nodes = ret->private_data; | |
296 | ret->private_data = NULL; | |
1da177e4 LT |
297 | rtp->nr--; |
298 | } | |
ce80b067 CM |
299 | /* |
300 | * Update the allocation stack trace as this is more useful | |
301 | * for debugging. | |
302 | */ | |
303 | kmemleak_update_trace(ret); | |
58e698af | 304 | goto out; |
1da177e4 | 305 | } |
05eb6e72 | 306 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 307 | out: |
b194d16c | 308 | BUG_ON(radix_tree_is_internal_node(ret)); |
1da177e4 LT |
309 | return ret; |
310 | } | |
311 | ||
7cf9c2c7 NP |
312 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
313 | { | |
314 | struct radix_tree_node *node = | |
315 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 316 | int i; |
643b52b9 NP |
317 | |
318 | /* | |
319 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
320 | * can leave us with a non-NULL entry in the first slot, so clear | |
321 | * that here to make sure. | |
322 | */ | |
b6dd0865 DC |
323 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
324 | tag_clear(node, i, 0); | |
325 | ||
643b52b9 | 326 | node->slots[0] = NULL; |
91d9c05a | 327 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 328 | |
7cf9c2c7 NP |
329 | kmem_cache_free(radix_tree_node_cachep, node); |
330 | } | |
331 | ||
1da177e4 LT |
332 | static inline void |
333 | radix_tree_node_free(struct radix_tree_node *node) | |
334 | { | |
7cf9c2c7 | 335 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
336 | } |
337 | ||
338 | /* | |
339 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
340 | * ensure that the addition of a single element in the tree cannot fail. On | |
341 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
342 | * with preemption not disabled. | |
b34df792 DH |
343 | * |
344 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 345 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 346 | */ |
c78c66d1 | 347 | static int __radix_tree_preload(gfp_t gfp_mask, int nr) |
1da177e4 LT |
348 | { |
349 | struct radix_tree_preload *rtp; | |
350 | struct radix_tree_node *node; | |
351 | int ret = -ENOMEM; | |
352 | ||
05eb6e72 VD |
353 | /* |
354 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
355 | * they should never be accounted to any particular memory cgroup. | |
356 | */ | |
357 | gfp_mask &= ~__GFP_ACCOUNT; | |
358 | ||
1da177e4 | 359 | preempt_disable(); |
7c8e0181 | 360 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 361 | while (rtp->nr < nr) { |
1da177e4 | 362 | preempt_enable(); |
488514d1 | 363 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
364 | if (node == NULL) |
365 | goto out; | |
366 | preempt_disable(); | |
7c8e0181 | 367 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 368 | if (rtp->nr < nr) { |
9d2a8da0 KS |
369 | node->private_data = rtp->nodes; |
370 | rtp->nodes = node; | |
371 | rtp->nr++; | |
372 | } else { | |
1da177e4 | 373 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 374 | } |
1da177e4 LT |
375 | } |
376 | ret = 0; | |
377 | out: | |
378 | return ret; | |
379 | } | |
5e4c0d97 JK |
380 | |
381 | /* | |
382 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
383 | * ensure that the addition of a single element in the tree cannot fail. On | |
384 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
385 | * with preemption not disabled. | |
386 | * | |
387 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 388 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
389 | */ |
390 | int radix_tree_preload(gfp_t gfp_mask) | |
391 | { | |
392 | /* Warn on non-sensical use... */ | |
d0164adc | 393 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 394 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 395 | } |
d7f0923d | 396 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 397 | |
5e4c0d97 JK |
398 | /* |
399 | * The same as above function, except we don't guarantee preloading happens. | |
400 | * We do it, if we decide it helps. On success, return zero with preemption | |
401 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
402 | */ | |
403 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
404 | { | |
d0164adc | 405 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 406 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
407 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
408 | preempt_disable(); | |
409 | return 0; | |
410 | } | |
411 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
412 | ||
c78c66d1 KS |
413 | /* |
414 | * The same as function above, but preload number of nodes required to insert | |
415 | * (1 << order) continuous naturally-aligned elements. | |
416 | */ | |
417 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
418 | { | |
419 | unsigned long nr_subtrees; | |
420 | int nr_nodes, subtree_height; | |
421 | ||
422 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
423 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
424 | preempt_disable(); | |
425 | return 0; | |
426 | } | |
427 | ||
428 | /* | |
429 | * Calculate number and height of fully populated subtrees it takes to | |
430 | * store (1 << order) elements. | |
431 | */ | |
432 | nr_subtrees = 1 << order; | |
433 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
434 | subtree_height++) | |
435 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
436 | ||
437 | /* | |
438 | * The worst case is zero height tree with a single item at index 0 and | |
439 | * then inserting items starting at ULONG_MAX - (1 << order). | |
440 | * | |
441 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
442 | * 0-index item. | |
443 | */ | |
444 | nr_nodes = RADIX_TREE_MAX_PATH; | |
445 | ||
446 | /* Plus branch to fully populated subtrees. */ | |
447 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
448 | ||
449 | /* Root node is shared. */ | |
450 | nr_nodes--; | |
451 | ||
452 | /* Plus nodes required to build subtrees. */ | |
453 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
454 | ||
455 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
456 | } | |
457 | ||
1da177e4 | 458 | /* |
d0891265 | 459 | * The maximum index which can be stored in a radix tree |
1da177e4 | 460 | */ |
c12e51b0 MW |
461 | static inline unsigned long shift_maxindex(unsigned int shift) |
462 | { | |
463 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
464 | } | |
465 | ||
1456a439 MW |
466 | static inline unsigned long node_maxindex(struct radix_tree_node *node) |
467 | { | |
c12e51b0 | 468 | return shift_maxindex(node->shift); |
1456a439 MW |
469 | } |
470 | ||
471 | static unsigned radix_tree_load_root(struct radix_tree_root *root, | |
472 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
473 | { | |
474 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
475 | ||
476 | *nodep = node; | |
477 | ||
b194d16c | 478 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 479 | node = entry_to_node(node); |
1456a439 | 480 | *maxindex = node_maxindex(node); |
c12e51b0 | 481 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
482 | } |
483 | ||
484 | *maxindex = 0; | |
485 | return 0; | |
486 | } | |
487 | ||
1da177e4 LT |
488 | /* |
489 | * Extend a radix tree so it can store key @index. | |
490 | */ | |
e6145236 | 491 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 492 | unsigned long index, unsigned int shift) |
1da177e4 | 493 | { |
e2bdb933 | 494 | struct radix_tree_node *slot; |
d0891265 | 495 | unsigned int maxshift; |
1da177e4 LT |
496 | int tag; |
497 | ||
d0891265 MW |
498 | /* Figure out what the shift should be. */ |
499 | maxshift = shift; | |
500 | while (index > shift_maxindex(maxshift)) | |
501 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 502 | |
d0891265 MW |
503 | slot = root->rnode; |
504 | if (!slot) | |
1da177e4 | 505 | goto out; |
1da177e4 | 506 | |
1da177e4 | 507 | do { |
2fcd9005 MW |
508 | struct radix_tree_node *node = radix_tree_node_alloc(root); |
509 | ||
510 | if (!node) | |
1da177e4 LT |
511 | return -ENOMEM; |
512 | ||
1da177e4 | 513 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 514 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 515 | if (root_tag_get(root, tag)) |
1da177e4 LT |
516 | tag_set(node, tag, 0); |
517 | } | |
518 | ||
d0891265 MW |
519 | BUG_ON(shift > BITS_PER_LONG); |
520 | node->shift = shift; | |
0c7fa0a8 | 521 | node->offset = 0; |
1da177e4 | 522 | node->count = 1; |
e2bdb933 | 523 | node->parent = NULL; |
f7942430 | 524 | if (radix_tree_is_internal_node(slot)) { |
4dd6c098 | 525 | entry_to_node(slot)->parent = node; |
f7942430 JW |
526 | } else { |
527 | /* Moving an exceptional root->rnode to a node */ | |
528 | if (radix_tree_exceptional_entry(slot)) | |
529 | node->exceptional = 1; | |
530 | } | |
e2bdb933 | 531 | node->slots[0] = slot; |
a4db4dce MW |
532 | slot = node_to_entry(node); |
533 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 534 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 535 | } while (shift <= maxshift); |
1da177e4 | 536 | out: |
d0891265 | 537 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
538 | } |
539 | ||
f4b109c6 JW |
540 | /** |
541 | * radix_tree_shrink - shrink radix tree to minimum height | |
542 | * @root radix tree root | |
543 | */ | |
14b46879 | 544 | static inline void radix_tree_shrink(struct radix_tree_root *root, |
4d693d08 JW |
545 | radix_tree_update_node_t update_node, |
546 | void *private) | |
f4b109c6 | 547 | { |
f4b109c6 JW |
548 | for (;;) { |
549 | struct radix_tree_node *node = root->rnode; | |
550 | struct radix_tree_node *child; | |
551 | ||
552 | if (!radix_tree_is_internal_node(node)) | |
553 | break; | |
554 | node = entry_to_node(node); | |
555 | ||
556 | /* | |
557 | * The candidate node has more than one child, or its child | |
558 | * is not at the leftmost slot, or the child is a multiorder | |
559 | * entry, we cannot shrink. | |
560 | */ | |
561 | if (node->count != 1) | |
562 | break; | |
563 | child = node->slots[0]; | |
564 | if (!child) | |
565 | break; | |
566 | if (!radix_tree_is_internal_node(child) && node->shift) | |
567 | break; | |
568 | ||
569 | if (radix_tree_is_internal_node(child)) | |
570 | entry_to_node(child)->parent = NULL; | |
571 | ||
572 | /* | |
573 | * We don't need rcu_assign_pointer(), since we are simply | |
574 | * moving the node from one part of the tree to another: if it | |
575 | * was safe to dereference the old pointer to it | |
576 | * (node->slots[0]), it will be safe to dereference the new | |
577 | * one (root->rnode) as far as dependent read barriers go. | |
578 | */ | |
579 | root->rnode = child; | |
580 | ||
581 | /* | |
582 | * We have a dilemma here. The node's slot[0] must not be | |
583 | * NULLed in case there are concurrent lookups expecting to | |
584 | * find the item. However if this was a bottom-level node, | |
585 | * then it may be subject to the slot pointer being visible | |
586 | * to callers dereferencing it. If item corresponding to | |
587 | * slot[0] is subsequently deleted, these callers would expect | |
588 | * their slot to become empty sooner or later. | |
589 | * | |
590 | * For example, lockless pagecache will look up a slot, deref | |
591 | * the page pointer, and if the page has 0 refcount it means it | |
592 | * was concurrently deleted from pagecache so try the deref | |
593 | * again. Fortunately there is already a requirement for logic | |
594 | * to retry the entire slot lookup -- the indirect pointer | |
595 | * problem (replacing direct root node with an indirect pointer | |
596 | * also results in a stale slot). So tag the slot as indirect | |
597 | * to force callers to retry. | |
598 | */ | |
4d693d08 JW |
599 | node->count = 0; |
600 | if (!radix_tree_is_internal_node(child)) { | |
f4b109c6 | 601 | node->slots[0] = RADIX_TREE_RETRY; |
4d693d08 JW |
602 | if (update_node) |
603 | update_node(node, private); | |
604 | } | |
f4b109c6 JW |
605 | |
606 | radix_tree_node_free(node); | |
f4b109c6 | 607 | } |
f4b109c6 JW |
608 | } |
609 | ||
14b46879 | 610 | static void delete_node(struct radix_tree_root *root, |
4d693d08 JW |
611 | struct radix_tree_node *node, |
612 | radix_tree_update_node_t update_node, void *private) | |
f4b109c6 | 613 | { |
f4b109c6 JW |
614 | do { |
615 | struct radix_tree_node *parent; | |
616 | ||
617 | if (node->count) { | |
618 | if (node == entry_to_node(root->rnode)) | |
14b46879 JW |
619 | radix_tree_shrink(root, update_node, private); |
620 | return; | |
f4b109c6 JW |
621 | } |
622 | ||
623 | parent = node->parent; | |
624 | if (parent) { | |
625 | parent->slots[node->offset] = NULL; | |
626 | parent->count--; | |
627 | } else { | |
628 | root_tag_clear_all(root); | |
629 | root->rnode = NULL; | |
630 | } | |
631 | ||
632 | radix_tree_node_free(node); | |
f4b109c6 JW |
633 | |
634 | node = parent; | |
635 | } while (node); | |
f4b109c6 JW |
636 | } |
637 | ||
1da177e4 | 638 | /** |
139e5616 | 639 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
640 | * @root: radix tree root |
641 | * @index: index key | |
e6145236 | 642 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
643 | * @nodep: returns node |
644 | * @slotp: returns slot | |
1da177e4 | 645 | * |
139e5616 JW |
646 | * Create, if necessary, and return the node and slot for an item |
647 | * at position @index in the radix tree @root. | |
648 | * | |
649 | * Until there is more than one item in the tree, no nodes are | |
650 | * allocated and @root->rnode is used as a direct slot instead of | |
651 | * pointing to a node, in which case *@nodep will be NULL. | |
652 | * | |
653 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 654 | */ |
139e5616 | 655 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
656 | unsigned order, struct radix_tree_node **nodep, |
657 | void ***slotp) | |
1da177e4 | 658 | { |
89148aa4 MW |
659 | struct radix_tree_node *node = NULL, *child; |
660 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 661 | unsigned long maxindex; |
89148aa4 | 662 | unsigned int shift, offset = 0; |
49ea6ebc MW |
663 | unsigned long max = index | ((1UL << order) - 1); |
664 | ||
89148aa4 | 665 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
666 | |
667 | /* Make sure the tree is high enough. */ | |
49ea6ebc | 668 | if (max > maxindex) { |
d0891265 | 669 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 670 | if (error < 0) |
1da177e4 | 671 | return error; |
49ea6ebc | 672 | shift = error; |
89148aa4 | 673 | child = root->rnode; |
d0891265 | 674 | if (order == shift) |
49ea6ebc | 675 | shift += RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
676 | } |
677 | ||
e6145236 | 678 | while (shift > order) { |
c12e51b0 | 679 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 680 | if (child == NULL) { |
1da177e4 | 681 | /* Have to add a child node. */ |
89148aa4 MW |
682 | child = radix_tree_node_alloc(root); |
683 | if (!child) | |
1da177e4 | 684 | return -ENOMEM; |
89148aa4 MW |
685 | child->shift = shift; |
686 | child->offset = offset; | |
687 | child->parent = node; | |
688 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
689 | if (node) | |
1da177e4 | 690 | node->count++; |
89148aa4 | 691 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 692 | break; |
1da177e4 LT |
693 | |
694 | /* Go a level down */ | |
89148aa4 | 695 | node = entry_to_node(child); |
9e85d811 | 696 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 697 | slot = &node->slots[offset]; |
e6145236 MW |
698 | } |
699 | ||
57578c2e | 700 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 701 | /* Insert pointers to the canonical entry */ |
3b8c00f6 | 702 | if (order > shift) { |
89148aa4 | 703 | unsigned i, n = 1 << (order - shift); |
e6145236 | 704 | offset = offset & ~(n - 1); |
89148aa4 MW |
705 | slot = &node->slots[offset]; |
706 | child = node_to_entry(slot); | |
e6145236 | 707 | for (i = 0; i < n; i++) { |
89148aa4 | 708 | if (slot[i]) |
e6145236 MW |
709 | return -EEXIST; |
710 | } | |
711 | ||
712 | for (i = 1; i < n; i++) { | |
89148aa4 | 713 | rcu_assign_pointer(slot[i], child); |
e6145236 MW |
714 | node->count++; |
715 | } | |
612d6c19 | 716 | } |
57578c2e | 717 | #endif |
1da177e4 | 718 | |
139e5616 JW |
719 | if (nodep) |
720 | *nodep = node; | |
721 | if (slotp) | |
89148aa4 | 722 | *slotp = slot; |
139e5616 JW |
723 | return 0; |
724 | } | |
725 | ||
726 | /** | |
e6145236 | 727 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
728 | * @root: radix tree root |
729 | * @index: index key | |
e6145236 | 730 | * @order: key covers the 2^order indices around index |
139e5616 JW |
731 | * @item: item to insert |
732 | * | |
733 | * Insert an item into the radix tree at position @index. | |
734 | */ | |
e6145236 MW |
735 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
736 | unsigned order, void *item) | |
139e5616 JW |
737 | { |
738 | struct radix_tree_node *node; | |
739 | void **slot; | |
740 | int error; | |
741 | ||
b194d16c | 742 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 743 | |
e6145236 | 744 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
745 | if (error) |
746 | return error; | |
747 | if (*slot != NULL) | |
1da177e4 | 748 | return -EEXIST; |
139e5616 | 749 | rcu_assign_pointer(*slot, item); |
201b6264 | 750 | |
612d6c19 | 751 | if (node) { |
7b60e9ad | 752 | unsigned offset = get_slot_offset(node, slot); |
612d6c19 | 753 | node->count++; |
f7942430 JW |
754 | if (radix_tree_exceptional_entry(item)) |
755 | node->exceptional++; | |
7b60e9ad MW |
756 | BUG_ON(tag_get(node, 0, offset)); |
757 | BUG_ON(tag_get(node, 1, offset)); | |
758 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 759 | } else { |
7b60e9ad | 760 | BUG_ON(root_tags_get(root)); |
612d6c19 | 761 | } |
1da177e4 | 762 | |
1da177e4 LT |
763 | return 0; |
764 | } | |
e6145236 | 765 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 766 | |
139e5616 JW |
767 | /** |
768 | * __radix_tree_lookup - lookup an item in a radix tree | |
769 | * @root: radix tree root | |
770 | * @index: index key | |
771 | * @nodep: returns node | |
772 | * @slotp: returns slot | |
773 | * | |
774 | * Lookup and return the item at position @index in the radix | |
775 | * tree @root. | |
776 | * | |
777 | * Until there is more than one item in the tree, no nodes are | |
778 | * allocated and @root->rnode is used as a direct slot instead of | |
779 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 780 | */ |
139e5616 JW |
781 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
782 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 783 | { |
139e5616 | 784 | struct radix_tree_node *node, *parent; |
85829954 | 785 | unsigned long maxindex; |
139e5616 | 786 | void **slot; |
612d6c19 | 787 | |
85829954 MW |
788 | restart: |
789 | parent = NULL; | |
790 | slot = (void **)&root->rnode; | |
9e85d811 | 791 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 792 | if (index > maxindex) |
1da177e4 LT |
793 | return NULL; |
794 | ||
b194d16c | 795 | while (radix_tree_is_internal_node(node)) { |
85829954 | 796 | unsigned offset; |
1da177e4 | 797 | |
85829954 MW |
798 | if (node == RADIX_TREE_RETRY) |
799 | goto restart; | |
4dd6c098 | 800 | parent = entry_to_node(node); |
9e85d811 | 801 | offset = radix_tree_descend(parent, &node, index); |
85829954 MW |
802 | slot = parent->slots + offset; |
803 | } | |
1da177e4 | 804 | |
139e5616 JW |
805 | if (nodep) |
806 | *nodep = parent; | |
807 | if (slotp) | |
808 | *slotp = slot; | |
809 | return node; | |
b72b71c6 HS |
810 | } |
811 | ||
812 | /** | |
813 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
814 | * @root: radix tree root | |
815 | * @index: index key | |
816 | * | |
817 | * Returns: the slot corresponding to the position @index in the | |
818 | * radix tree @root. This is useful for update-if-exists operations. | |
819 | * | |
820 | * This function can be called under rcu_read_lock iff the slot is not | |
821 | * modified by radix_tree_replace_slot, otherwise it must be called | |
822 | * exclusive from other writers. Any dereference of the slot must be done | |
823 | * using radix_tree_deref_slot. | |
824 | */ | |
825 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
826 | { | |
139e5616 JW |
827 | void **slot; |
828 | ||
829 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
830 | return NULL; | |
831 | return slot; | |
a4331366 | 832 | } |
a4331366 HR |
833 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
834 | ||
835 | /** | |
836 | * radix_tree_lookup - perform lookup operation on a radix tree | |
837 | * @root: radix tree root | |
838 | * @index: index key | |
839 | * | |
840 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
841 | * |
842 | * This function can be called under rcu_read_lock, however the caller | |
843 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
844 | * them safely). No RCU barriers are required to access or modify the | |
845 | * returned item, however. | |
a4331366 HR |
846 | */ |
847 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
848 | { | |
139e5616 | 849 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
850 | } |
851 | EXPORT_SYMBOL(radix_tree_lookup); | |
852 | ||
6d75f366 JW |
853 | static void replace_slot(struct radix_tree_root *root, |
854 | struct radix_tree_node *node, | |
855 | void **slot, void *item, | |
856 | bool warn_typeswitch) | |
f7942430 JW |
857 | { |
858 | void *old = rcu_dereference_raw(*slot); | |
f4b109c6 | 859 | int count, exceptional; |
f7942430 JW |
860 | |
861 | WARN_ON_ONCE(radix_tree_is_internal_node(item)); | |
f7942430 | 862 | |
f4b109c6 | 863 | count = !!item - !!old; |
f7942430 JW |
864 | exceptional = !!radix_tree_exceptional_entry(item) - |
865 | !!radix_tree_exceptional_entry(old); | |
866 | ||
f4b109c6 | 867 | WARN_ON_ONCE(warn_typeswitch && (count || exceptional)); |
f7942430 | 868 | |
f4b109c6 JW |
869 | if (node) { |
870 | node->count += count; | |
f7942430 | 871 | node->exceptional += exceptional; |
f4b109c6 | 872 | } |
f7942430 JW |
873 | |
874 | rcu_assign_pointer(*slot, item); | |
875 | } | |
876 | ||
6d75f366 JW |
877 | /** |
878 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
879 | * @root: radix tree root |
880 | * @node: pointer to tree node | |
881 | * @slot: pointer to slot in @node | |
882 | * @item: new item to store in the slot. | |
883 | * @update_node: callback for changing leaf nodes | |
884 | * @private: private data to pass to @update_node | |
6d75f366 JW |
885 | * |
886 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
887 | * across slot lookup and replacement. | |
888 | */ | |
889 | void __radix_tree_replace(struct radix_tree_root *root, | |
890 | struct radix_tree_node *node, | |
4d693d08 JW |
891 | void **slot, void *item, |
892 | radix_tree_update_node_t update_node, void *private) | |
6d75f366 JW |
893 | { |
894 | /* | |
f4b109c6 JW |
895 | * This function supports replacing exceptional entries and |
896 | * deleting entries, but that needs accounting against the | |
897 | * node unless the slot is root->rnode. | |
6d75f366 JW |
898 | */ |
899 | replace_slot(root, node, slot, item, | |
900 | !node && slot != (void **)&root->rnode); | |
f4b109c6 | 901 | |
4d693d08 JW |
902 | if (!node) |
903 | return; | |
904 | ||
905 | if (update_node) | |
906 | update_node(node, private); | |
907 | ||
908 | delete_node(root, node, update_node, private); | |
6d75f366 JW |
909 | } |
910 | ||
911 | /** | |
912 | * radix_tree_replace_slot - replace item in a slot | |
913 | * @root: radix tree root | |
914 | * @slot: pointer to slot | |
915 | * @item: new item to store in the slot. | |
916 | * | |
917 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
918 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
919 | * across slot lookup and replacement. | |
920 | * | |
921 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
922 | * regular entries, and exceptional entries, as that requires accounting | |
f4b109c6 JW |
923 | * inside the radix tree node. When switching from one type of entry or |
924 | * deleting, use __radix_tree_lookup() and __radix_tree_replace(). | |
6d75f366 JW |
925 | */ |
926 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
927 | void **slot, void *item) | |
928 | { | |
929 | replace_slot(root, NULL, slot, item, true); | |
930 | } | |
931 | ||
1da177e4 LT |
932 | /** |
933 | * radix_tree_tag_set - set a tag on a radix tree node | |
934 | * @root: radix tree root | |
935 | * @index: index key | |
2fcd9005 | 936 | * @tag: tag index |
1da177e4 | 937 | * |
daff89f3 JC |
938 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
939 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
940 | * the root all the way down to the leaf node. |
941 | * | |
2fcd9005 | 942 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
943 | * item is a bug. |
944 | */ | |
945 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 946 | unsigned long index, unsigned int tag) |
1da177e4 | 947 | { |
fb969909 RZ |
948 | struct radix_tree_node *node, *parent; |
949 | unsigned long maxindex; | |
1da177e4 | 950 | |
9e85d811 | 951 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 952 | BUG_ON(index > maxindex); |
1da177e4 | 953 | |
b194d16c | 954 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 955 | unsigned offset; |
1da177e4 | 956 | |
4dd6c098 | 957 | parent = entry_to_node(node); |
9e85d811 | 958 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
959 | BUG_ON(!node); |
960 | ||
961 | if (!tag_get(parent, tag, offset)) | |
962 | tag_set(parent, tag, offset); | |
1da177e4 LT |
963 | } |
964 | ||
612d6c19 | 965 | /* set the root's tag bit */ |
fb969909 | 966 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
967 | root_tag_set(root, tag); |
968 | ||
fb969909 | 969 | return node; |
1da177e4 LT |
970 | } |
971 | EXPORT_SYMBOL(radix_tree_tag_set); | |
972 | ||
d604c324 MW |
973 | static void node_tag_clear(struct radix_tree_root *root, |
974 | struct radix_tree_node *node, | |
975 | unsigned int tag, unsigned int offset) | |
976 | { | |
977 | while (node) { | |
978 | if (!tag_get(node, tag, offset)) | |
979 | return; | |
980 | tag_clear(node, tag, offset); | |
981 | if (any_tag_set(node, tag)) | |
982 | return; | |
983 | ||
984 | offset = node->offset; | |
985 | node = node->parent; | |
986 | } | |
987 | ||
988 | /* clear the root's tag bit */ | |
989 | if (root_tag_get(root, tag)) | |
990 | root_tag_clear(root, tag); | |
991 | } | |
992 | ||
9498d2bb MW |
993 | static void node_tag_set(struct radix_tree_root *root, |
994 | struct radix_tree_node *node, | |
995 | unsigned int tag, unsigned int offset) | |
996 | { | |
997 | while (node) { | |
998 | if (tag_get(node, tag, offset)) | |
999 | return; | |
1000 | tag_set(node, tag, offset); | |
1001 | offset = node->offset; | |
1002 | node = node->parent; | |
1003 | } | |
1004 | ||
1005 | if (!root_tag_get(root, tag)) | |
1006 | root_tag_set(root, tag); | |
1007 | } | |
1008 | ||
1da177e4 LT |
1009 | /** |
1010 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1011 | * @root: radix tree root | |
1012 | * @index: index key | |
2fcd9005 | 1013 | * @tag: tag index |
1da177e4 | 1014 | * |
daff89f3 | 1015 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1016 | * corresponding to @index in the radix tree. If this causes |
1017 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1018 | * next-to-leaf node, etc. |
1019 | * | |
1020 | * Returns the address of the tagged item on success, else NULL. ie: | |
1021 | * has the same return value and semantics as radix_tree_lookup(). | |
1022 | */ | |
1023 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1024 | unsigned long index, unsigned int tag) |
1da177e4 | 1025 | { |
00f47b58 RZ |
1026 | struct radix_tree_node *node, *parent; |
1027 | unsigned long maxindex; | |
e2bdb933 | 1028 | int uninitialized_var(offset); |
1da177e4 | 1029 | |
9e85d811 | 1030 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1031 | if (index > maxindex) |
1032 | return NULL; | |
1da177e4 | 1033 | |
00f47b58 | 1034 | parent = NULL; |
1da177e4 | 1035 | |
b194d16c | 1036 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1037 | parent = entry_to_node(node); |
9e85d811 | 1038 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1039 | } |
1040 | ||
d604c324 MW |
1041 | if (node) |
1042 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1043 | |
00f47b58 | 1044 | return node; |
1da177e4 LT |
1045 | } |
1046 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1047 | ||
1da177e4 | 1048 | /** |
32605a18 MT |
1049 | * radix_tree_tag_get - get a tag on a radix tree node |
1050 | * @root: radix tree root | |
1051 | * @index: index key | |
2fcd9005 | 1052 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1053 | * |
32605a18 | 1054 | * Return values: |
1da177e4 | 1055 | * |
612d6c19 NP |
1056 | * 0: tag not present or not set |
1057 | * 1: tag set | |
ce82653d DH |
1058 | * |
1059 | * Note that the return value of this function may not be relied on, even if | |
1060 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1061 | * from concurrency. | |
1da177e4 LT |
1062 | */ |
1063 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 1064 | unsigned long index, unsigned int tag) |
1da177e4 | 1065 | { |
4589ba6d RZ |
1066 | struct radix_tree_node *node, *parent; |
1067 | unsigned long maxindex; | |
1da177e4 | 1068 | |
612d6c19 NP |
1069 | if (!root_tag_get(root, tag)) |
1070 | return 0; | |
1071 | ||
9e85d811 | 1072 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1073 | if (index > maxindex) |
1074 | return 0; | |
7cf9c2c7 NP |
1075 | if (node == NULL) |
1076 | return 0; | |
1077 | ||
b194d16c | 1078 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1079 | unsigned offset; |
1da177e4 | 1080 | |
4dd6c098 | 1081 | parent = entry_to_node(node); |
9e85d811 | 1082 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1083 | |
4589ba6d | 1084 | if (!node) |
1da177e4 | 1085 | return 0; |
4589ba6d | 1086 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1087 | return 0; |
4589ba6d RZ |
1088 | if (node == RADIX_TREE_RETRY) |
1089 | break; | |
1da177e4 | 1090 | } |
4589ba6d RZ |
1091 | |
1092 | return 1; | |
1da177e4 LT |
1093 | } |
1094 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1095 | |
21ef5339 RZ |
1096 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1097 | unsigned int shift) | |
1098 | { | |
1099 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1100 | iter->shift = shift; | |
1101 | #endif | |
1102 | } | |
1103 | ||
78c1d784 KK |
1104 | /** |
1105 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1106 | * | |
1107 | * @root: radix tree root | |
1108 | * @iter: iterator state | |
1109 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1110 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1111 | */ | |
1112 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
1113 | struct radix_tree_iter *iter, unsigned flags) | |
1114 | { | |
9e85d811 | 1115 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1116 | struct radix_tree_node *node, *child; |
21ef5339 | 1117 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1118 | |
1119 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1120 | return NULL; | |
1121 | ||
1122 | /* | |
1123 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1124 | * during iterating; it can be zero only at the beginning. | |
1125 | * And we cannot overflow iter->next_index in a single step, | |
1126 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1127 | * |
1128 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1129 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1130 | */ |
1131 | index = iter->next_index; | |
1132 | if (!index && iter->index) | |
1133 | return NULL; | |
1134 | ||
21ef5339 | 1135 | restart: |
9e85d811 | 1136 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1137 | if (index > maxindex) |
1138 | return NULL; | |
8c1244de MW |
1139 | if (!child) |
1140 | return NULL; | |
21ef5339 | 1141 | |
8c1244de | 1142 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1143 | /* Single-slot tree */ |
21ef5339 RZ |
1144 | iter->index = index; |
1145 | iter->next_index = maxindex + 1; | |
78c1d784 | 1146 | iter->tags = 1; |
8c1244de | 1147 | __set_iter_shift(iter, 0); |
78c1d784 | 1148 | return (void **)&root->rnode; |
8c1244de | 1149 | } |
21ef5339 | 1150 | |
8c1244de MW |
1151 | do { |
1152 | node = entry_to_node(child); | |
9e85d811 | 1153 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1154 | |
78c1d784 | 1155 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1156 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1157 | /* Hole detected */ |
1158 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1159 | return NULL; | |
1160 | ||
1161 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1162 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1163 | offset + 1); |
1164 | else | |
1165 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
1166 | void *slot = node->slots[offset]; |
1167 | if (is_sibling_entry(node, slot)) | |
1168 | continue; | |
1169 | if (slot) | |
78c1d784 KK |
1170 | break; |
1171 | } | |
8c1244de | 1172 | index &= ~node_maxindex(node); |
9e85d811 | 1173 | index += offset << node->shift; |
78c1d784 KK |
1174 | /* Overflow after ~0UL */ |
1175 | if (!index) | |
1176 | return NULL; | |
1177 | if (offset == RADIX_TREE_MAP_SIZE) | |
1178 | goto restart; | |
8c1244de | 1179 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1180 | } |
1181 | ||
8c1244de | 1182 | if ((child == NULL) || (child == RADIX_TREE_RETRY)) |
78c1d784 | 1183 | goto restart; |
8c1244de | 1184 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
1185 | |
1186 | /* Update the iterator state */ | |
8c1244de MW |
1187 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1188 | iter->next_index = (index | node_maxindex(node)) + 1; | |
9e85d811 | 1189 | __set_iter_shift(iter, node->shift); |
78c1d784 KK |
1190 | |
1191 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
1192 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1193 | unsigned tag_long, tag_bit; | |
1194 | ||
1195 | tag_long = offset / BITS_PER_LONG; | |
1196 | tag_bit = offset % BITS_PER_LONG; | |
1197 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1198 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1199 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1200 | /* Pick tags from next element */ | |
1201 | if (tag_bit) | |
1202 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1203 | (BITS_PER_LONG - tag_bit); | |
1204 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1205 | iter->next_index = index + BITS_PER_LONG; | |
1206 | } | |
1207 | } | |
1208 | ||
1209 | return node->slots + offset; | |
1210 | } | |
1211 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1212 | ||
ebf8aa44 JK |
1213 | /** |
1214 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
1215 | * tag if item has another tag set | |
1216 | * @root: radix tree root | |
1217 | * @first_indexp: pointer to a starting index of a range to scan | |
1218 | * @last_index: last index of a range to scan | |
1219 | * @nr_to_tag: maximum number items to tag | |
1220 | * @iftag: tag index to test | |
1221 | * @settag: tag index to set if tested tag is set | |
1222 | * | |
1223 | * This function scans range of radix tree from first_index to last_index | |
1224 | * (inclusive). For each item in the range if iftag is set, the function sets | |
1225 | * also settag. The function stops either after tagging nr_to_tag items or | |
1226 | * after reaching last_index. | |
1227 | * | |
144dcfc0 DC |
1228 | * The tags must be set from the leaf level only and propagated back up the |
1229 | * path to the root. We must do this so that we resolve the full path before | |
1230 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
1231 | * we can get to the leaf node and find that the index that has the iftag | |
1232 | * set is outside the range we are scanning. This reults in dangling tags and | |
1233 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1234 | * | |
2fcd9005 | 1235 | * The function returns the number of leaves where the tag was set and sets |
ebf8aa44 | 1236 | * *first_indexp to the first unscanned index. |
d5ed3a4a JK |
1237 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1238 | * be prepared to handle that. | |
ebf8aa44 JK |
1239 | */ |
1240 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1241 | unsigned long *first_indexp, unsigned long last_index, | |
1242 | unsigned long nr_to_tag, | |
1243 | unsigned int iftag, unsigned int settag) | |
1244 | { | |
9498d2bb | 1245 | struct radix_tree_node *node, *child; |
070c5ac2 | 1246 | unsigned long maxindex; |
144dcfc0 DC |
1247 | unsigned long tagged = 0; |
1248 | unsigned long index = *first_indexp; | |
ebf8aa44 | 1249 | |
9e85d811 | 1250 | radix_tree_load_root(root, &child, &maxindex); |
070c5ac2 | 1251 | last_index = min(last_index, maxindex); |
ebf8aa44 JK |
1252 | if (index > last_index) |
1253 | return 0; | |
1254 | if (!nr_to_tag) | |
1255 | return 0; | |
1256 | if (!root_tag_get(root, iftag)) { | |
1257 | *first_indexp = last_index + 1; | |
1258 | return 0; | |
1259 | } | |
a8e4da25 | 1260 | if (!radix_tree_is_internal_node(child)) { |
ebf8aa44 JK |
1261 | *first_indexp = last_index + 1; |
1262 | root_tag_set(root, settag); | |
1263 | return 1; | |
1264 | } | |
1265 | ||
a8e4da25 | 1266 | node = entry_to_node(child); |
ebf8aa44 JK |
1267 | |
1268 | for (;;) { | |
9e85d811 | 1269 | unsigned offset = radix_tree_descend(node, &child, index); |
a8e4da25 | 1270 | if (!child) |
ebf8aa44 | 1271 | goto next; |
070c5ac2 | 1272 | if (!tag_get(node, iftag, offset)) |
ebf8aa44 | 1273 | goto next; |
070c5ac2 | 1274 | /* Sibling slots never have tags set on them */ |
a8e4da25 MW |
1275 | if (radix_tree_is_internal_node(child)) { |
1276 | node = entry_to_node(child); | |
070c5ac2 | 1277 | continue; |
144dcfc0 DC |
1278 | } |
1279 | ||
070c5ac2 | 1280 | tagged++; |
9498d2bb | 1281 | node_tag_set(root, node, settag, offset); |
070c5ac2 | 1282 | next: |
9e85d811 MW |
1283 | /* Go to next entry in node */ |
1284 | index = ((index >> node->shift) + 1) << node->shift; | |
d5ed3a4a JK |
1285 | /* Overflow can happen when last_index is ~0UL... */ |
1286 | if (index > last_index || !index) | |
ebf8aa44 | 1287 | break; |
9e85d811 | 1288 | offset = (index >> node->shift) & RADIX_TREE_MAP_MASK; |
070c5ac2 | 1289 | while (offset == 0) { |
ebf8aa44 JK |
1290 | /* |
1291 | * We've fully scanned this node. Go up. Because | |
1292 | * last_index is guaranteed to be in the tree, what | |
1293 | * we do below cannot wander astray. | |
1294 | */ | |
070c5ac2 | 1295 | node = node->parent; |
9e85d811 | 1296 | offset = (index >> node->shift) & RADIX_TREE_MAP_MASK; |
ebf8aa44 | 1297 | } |
070c5ac2 MW |
1298 | if (is_sibling_entry(node, node->slots[offset])) |
1299 | goto next; | |
1300 | if (tagged >= nr_to_tag) | |
1301 | break; | |
ebf8aa44 | 1302 | } |
9498d2bb | 1303 | |
ebf8aa44 JK |
1304 | *first_indexp = index; |
1305 | ||
1306 | return tagged; | |
1307 | } | |
1308 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1309 | ||
1da177e4 LT |
1310 | /** |
1311 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1312 | * @root: radix tree root | |
1313 | * @results: where the results of the lookup are placed | |
1314 | * @first_index: start the lookup from this key | |
1315 | * @max_items: place up to this many items at *results | |
1316 | * | |
1317 | * Performs an index-ascending scan of the tree for present items. Places | |
1318 | * them at *@results and returns the number of items which were placed at | |
1319 | * *@results. | |
1320 | * | |
1321 | * The implementation is naive. | |
7cf9c2c7 NP |
1322 | * |
1323 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1324 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1325 | * an atomic snapshot of the tree at a single point in time, the |
1326 | * semantics of an RCU protected gang lookup are as though multiple | |
1327 | * radix_tree_lookups have been issued in individual locks, and results | |
1328 | * stored in 'results'. | |
1da177e4 LT |
1329 | */ |
1330 | unsigned int | |
1331 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1332 | unsigned long first_index, unsigned int max_items) | |
1333 | { | |
cebbd29e KK |
1334 | struct radix_tree_iter iter; |
1335 | void **slot; | |
1336 | unsigned int ret = 0; | |
7cf9c2c7 | 1337 | |
cebbd29e | 1338 | if (unlikely(!max_items)) |
7cf9c2c7 | 1339 | return 0; |
1da177e4 | 1340 | |
cebbd29e | 1341 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1342 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1343 | if (!results[ret]) |
1344 | continue; | |
b194d16c | 1345 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1346 | slot = radix_tree_iter_retry(&iter); |
1347 | continue; | |
1348 | } | |
cebbd29e | 1349 | if (++ret == max_items) |
1da177e4 | 1350 | break; |
1da177e4 | 1351 | } |
7cf9c2c7 | 1352 | |
1da177e4 LT |
1353 | return ret; |
1354 | } | |
1355 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1356 | ||
47feff2c NP |
1357 | /** |
1358 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1359 | * @root: radix tree root | |
1360 | * @results: where the results of the lookup are placed | |
6328650b | 1361 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1362 | * @first_index: start the lookup from this key |
1363 | * @max_items: place up to this many items at *results | |
1364 | * | |
1365 | * Performs an index-ascending scan of the tree for present items. Places | |
1366 | * their slots at *@results and returns the number of items which were | |
1367 | * placed at *@results. | |
1368 | * | |
1369 | * The implementation is naive. | |
1370 | * | |
1371 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1372 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1373 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1374 | */ | |
1375 | unsigned int | |
6328650b HD |
1376 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1377 | void ***results, unsigned long *indices, | |
47feff2c NP |
1378 | unsigned long first_index, unsigned int max_items) |
1379 | { | |
cebbd29e KK |
1380 | struct radix_tree_iter iter; |
1381 | void **slot; | |
1382 | unsigned int ret = 0; | |
47feff2c | 1383 | |
cebbd29e | 1384 | if (unlikely(!max_items)) |
47feff2c NP |
1385 | return 0; |
1386 | ||
cebbd29e KK |
1387 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1388 | results[ret] = slot; | |
6328650b | 1389 | if (indices) |
cebbd29e KK |
1390 | indices[ret] = iter.index; |
1391 | if (++ret == max_items) | |
47feff2c | 1392 | break; |
47feff2c NP |
1393 | } |
1394 | ||
1395 | return ret; | |
1396 | } | |
1397 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1398 | ||
1da177e4 LT |
1399 | /** |
1400 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1401 | * based on a tag | |
1402 | * @root: radix tree root | |
1403 | * @results: where the results of the lookup are placed | |
1404 | * @first_index: start the lookup from this key | |
1405 | * @max_items: place up to this many items at *results | |
daff89f3 | 1406 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1407 | * |
1408 | * Performs an index-ascending scan of the tree for present items which | |
1409 | * have the tag indexed by @tag set. Places the items at *@results and | |
1410 | * returns the number of items which were placed at *@results. | |
1411 | */ | |
1412 | unsigned int | |
1413 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1414 | unsigned long first_index, unsigned int max_items, |
1415 | unsigned int tag) | |
1da177e4 | 1416 | { |
cebbd29e KK |
1417 | struct radix_tree_iter iter; |
1418 | void **slot; | |
1419 | unsigned int ret = 0; | |
612d6c19 | 1420 | |
cebbd29e | 1421 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1422 | return 0; |
1423 | ||
cebbd29e | 1424 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1425 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1426 | if (!results[ret]) |
1427 | continue; | |
b194d16c | 1428 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1429 | slot = radix_tree_iter_retry(&iter); |
1430 | continue; | |
1431 | } | |
cebbd29e | 1432 | if (++ret == max_items) |
1da177e4 | 1433 | break; |
1da177e4 | 1434 | } |
7cf9c2c7 | 1435 | |
1da177e4 LT |
1436 | return ret; |
1437 | } | |
1438 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1439 | ||
47feff2c NP |
1440 | /** |
1441 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1442 | * radix tree based on a tag | |
1443 | * @root: radix tree root | |
1444 | * @results: where the results of the lookup are placed | |
1445 | * @first_index: start the lookup from this key | |
1446 | * @max_items: place up to this many items at *results | |
1447 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1448 | * | |
1449 | * Performs an index-ascending scan of the tree for present items which | |
1450 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1451 | * returns the number of slots which were placed at *@results. | |
1452 | */ | |
1453 | unsigned int | |
1454 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1455 | unsigned long first_index, unsigned int max_items, | |
1456 | unsigned int tag) | |
1457 | { | |
cebbd29e KK |
1458 | struct radix_tree_iter iter; |
1459 | void **slot; | |
1460 | unsigned int ret = 0; | |
47feff2c | 1461 | |
cebbd29e | 1462 | if (unlikely(!max_items)) |
47feff2c NP |
1463 | return 0; |
1464 | ||
cebbd29e KK |
1465 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1466 | results[ret] = slot; | |
1467 | if (++ret == max_items) | |
47feff2c | 1468 | break; |
47feff2c NP |
1469 | } |
1470 | ||
1471 | return ret; | |
1472 | } | |
1473 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1474 | ||
e504f3fd HD |
1475 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1476 | #include <linux/sched.h> /* for cond_resched() */ | |
1477 | ||
0a2efc6c MW |
1478 | struct locate_info { |
1479 | unsigned long found_index; | |
1480 | bool stop; | |
1481 | }; | |
1482 | ||
e504f3fd HD |
1483 | /* |
1484 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1485 | */ | |
1486 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
0a2efc6c | 1487 | unsigned long index, struct locate_info *info) |
e504f3fd | 1488 | { |
e504f3fd HD |
1489 | unsigned long i; |
1490 | ||
0a2efc6c | 1491 | do { |
9e85d811 | 1492 | unsigned int shift = slot->shift; |
e504f3fd | 1493 | |
0a2efc6c MW |
1494 | for (i = (index >> shift) & RADIX_TREE_MAP_MASK; |
1495 | i < RADIX_TREE_MAP_SIZE; | |
1496 | i++, index += (1UL << shift)) { | |
1497 | struct radix_tree_node *node = | |
1498 | rcu_dereference_raw(slot->slots[i]); | |
1499 | if (node == RADIX_TREE_RETRY) | |
1500 | goto out; | |
b194d16c | 1501 | if (!radix_tree_is_internal_node(node)) { |
0a2efc6c MW |
1502 | if (node == item) { |
1503 | info->found_index = index; | |
1504 | info->stop = true; | |
1505 | goto out; | |
1506 | } | |
1507 | continue; | |
e6145236 | 1508 | } |
4dd6c098 | 1509 | node = entry_to_node(node); |
0a2efc6c MW |
1510 | if (is_sibling_entry(slot, node)) |
1511 | continue; | |
1512 | slot = node; | |
1513 | break; | |
e6145236 | 1514 | } |
9e85d811 | 1515 | } while (i < RADIX_TREE_MAP_SIZE); |
e504f3fd | 1516 | |
e504f3fd | 1517 | out: |
0a2efc6c MW |
1518 | if ((index == 0) && (i == RADIX_TREE_MAP_SIZE)) |
1519 | info->stop = true; | |
e504f3fd HD |
1520 | return index; |
1521 | } | |
1522 | ||
1523 | /** | |
1524 | * radix_tree_locate_item - search through radix tree for item | |
1525 | * @root: radix tree root | |
1526 | * @item: item to be found | |
1527 | * | |
1528 | * Returns index where item was found, or -1 if not found. | |
1529 | * Caller must hold no lock (since this time-consuming function needs | |
1530 | * to be preemptible), and must check afterwards if item is still there. | |
1531 | */ | |
1532 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1533 | { | |
1534 | struct radix_tree_node *node; | |
1535 | unsigned long max_index; | |
1536 | unsigned long cur_index = 0; | |
0a2efc6c MW |
1537 | struct locate_info info = { |
1538 | .found_index = -1, | |
1539 | .stop = false, | |
1540 | }; | |
e504f3fd HD |
1541 | |
1542 | do { | |
1543 | rcu_read_lock(); | |
1544 | node = rcu_dereference_raw(root->rnode); | |
b194d16c | 1545 | if (!radix_tree_is_internal_node(node)) { |
e504f3fd HD |
1546 | rcu_read_unlock(); |
1547 | if (node == item) | |
0a2efc6c | 1548 | info.found_index = 0; |
e504f3fd HD |
1549 | break; |
1550 | } | |
1551 | ||
4dd6c098 | 1552 | node = entry_to_node(node); |
0a2efc6c MW |
1553 | |
1554 | max_index = node_maxindex(node); | |
5f30fc94 HD |
1555 | if (cur_index > max_index) { |
1556 | rcu_read_unlock(); | |
e504f3fd | 1557 | break; |
5f30fc94 | 1558 | } |
e504f3fd | 1559 | |
0a2efc6c | 1560 | cur_index = __locate(node, item, cur_index, &info); |
e504f3fd HD |
1561 | rcu_read_unlock(); |
1562 | cond_resched(); | |
0a2efc6c | 1563 | } while (!info.stop && cur_index <= max_index); |
e504f3fd | 1564 | |
0a2efc6c | 1565 | return info.found_index; |
e504f3fd HD |
1566 | } |
1567 | #else | |
1568 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1569 | { | |
1570 | return -1; | |
1571 | } | |
1572 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1573 | |
139e5616 JW |
1574 | /** |
1575 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1576 | * @root: radix tree root | |
139e5616 JW |
1577 | * @node: node containing @index |
1578 | * | |
1579 | * After clearing the slot at @index in @node from radix tree | |
1580 | * rooted at @root, call this function to attempt freeing the | |
1581 | * node and shrinking the tree. | |
139e5616 | 1582 | */ |
14b46879 | 1583 | void __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1584 | struct radix_tree_node *node) |
1585 | { | |
14b46879 | 1586 | delete_node(root, node, NULL, NULL); |
139e5616 JW |
1587 | } |
1588 | ||
57578c2e MW |
1589 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1590 | void *ptr, unsigned offset) | |
1591 | { | |
1592 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1593 | int i; | |
1594 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1595 | if (node->slots[offset + i] != ptr) | |
1596 | break; | |
1597 | node->slots[offset + i] = NULL; | |
1598 | node->count--; | |
1599 | } | |
1600 | #endif | |
1601 | } | |
1602 | ||
1da177e4 | 1603 | /** |
53c59f26 | 1604 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1605 | * @root: radix tree root |
1606 | * @index: index key | |
53c59f26 | 1607 | * @item: expected item |
1da177e4 | 1608 | * |
53c59f26 | 1609 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1610 | * |
53c59f26 JW |
1611 | * Returns the address of the deleted item, or NULL if it was not present |
1612 | * or the entry at the given @index was not @item. | |
1da177e4 | 1613 | */ |
53c59f26 JW |
1614 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1615 | unsigned long index, void *item) | |
1da177e4 | 1616 | { |
139e5616 | 1617 | struct radix_tree_node *node; |
57578c2e | 1618 | unsigned int offset; |
139e5616 JW |
1619 | void **slot; |
1620 | void *entry; | |
d5274261 | 1621 | int tag; |
1da177e4 | 1622 | |
139e5616 JW |
1623 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1624 | if (!entry) | |
1625 | return NULL; | |
1da177e4 | 1626 | |
139e5616 JW |
1627 | if (item && entry != item) |
1628 | return NULL; | |
1629 | ||
1630 | if (!node) { | |
612d6c19 NP |
1631 | root_tag_clear_all(root); |
1632 | root->rnode = NULL; | |
139e5616 | 1633 | return entry; |
612d6c19 | 1634 | } |
1da177e4 | 1635 | |
29e0967c | 1636 | offset = get_slot_offset(node, slot); |
53c59f26 | 1637 | |
d604c324 MW |
1638 | /* Clear all tags associated with the item to be deleted. */ |
1639 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1640 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1641 | |
a4db4dce | 1642 | delete_sibling_entries(node, node_to_entry(slot), offset); |
4d693d08 | 1643 | __radix_tree_replace(root, node, slot, NULL, NULL, NULL); |
612d6c19 | 1644 | |
139e5616 | 1645 | return entry; |
1da177e4 | 1646 | } |
53c59f26 JW |
1647 | EXPORT_SYMBOL(radix_tree_delete_item); |
1648 | ||
1649 | /** | |
1650 | * radix_tree_delete - delete an item from a radix tree | |
1651 | * @root: radix tree root | |
1652 | * @index: index key | |
1653 | * | |
1654 | * Remove the item at @index from the radix tree rooted at @root. | |
1655 | * | |
1656 | * Returns the address of the deleted item, or NULL if it was not present. | |
1657 | */ | |
1658 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1659 | { | |
1660 | return radix_tree_delete_item(root, index, NULL); | |
1661 | } | |
1da177e4 LT |
1662 | EXPORT_SYMBOL(radix_tree_delete); |
1663 | ||
d3798ae8 JW |
1664 | void radix_tree_clear_tags(struct radix_tree_root *root, |
1665 | struct radix_tree_node *node, | |
1666 | void **slot) | |
d604c324 | 1667 | { |
d604c324 MW |
1668 | if (node) { |
1669 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1670 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1671 | node_tag_clear(root, node, tag, offset); | |
1672 | } else { | |
1673 | /* Clear root node tags */ | |
1674 | root->gfp_mask &= __GFP_BITS_MASK; | |
1675 | } | |
d604c324 MW |
1676 | } |
1677 | ||
1da177e4 LT |
1678 | /** |
1679 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1680 | * @root: radix tree root | |
1681 | * @tag: tag to test | |
1682 | */ | |
daff89f3 | 1683 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1684 | { |
612d6c19 | 1685 | return root_tag_get(root, tag); |
1da177e4 LT |
1686 | } |
1687 | EXPORT_SYMBOL(radix_tree_tagged); | |
1688 | ||
1689 | static void | |
449dd698 | 1690 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1691 | { |
449dd698 JW |
1692 | struct radix_tree_node *node = arg; |
1693 | ||
1694 | memset(node, 0, sizeof(*node)); | |
1695 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1696 | } |
1697 | ||
c78c66d1 KS |
1698 | static __init unsigned long __maxindex(unsigned int height) |
1699 | { | |
1700 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
1701 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1702 | ||
1703 | if (shift < 0) | |
1704 | return ~0UL; | |
1705 | if (shift >= BITS_PER_LONG) | |
1706 | return 0UL; | |
1707 | return ~0UL >> shift; | |
1708 | } | |
1709 | ||
1710 | static __init void radix_tree_init_maxnodes(void) | |
1711 | { | |
1712 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
1713 | unsigned int i, j; | |
1714 | ||
1715 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1716 | height_to_maxindex[i] = __maxindex(i); | |
1717 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
1718 | for (j = i; j > 0; j--) | |
1719 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
1720 | } | |
1721 | } | |
1722 | ||
d544abd5 | 1723 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1724 | { |
2fcd9005 MW |
1725 | struct radix_tree_preload *rtp; |
1726 | struct radix_tree_node *node; | |
1727 | ||
1728 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1729 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1730 | while (rtp->nr) { | |
1731 | node = rtp->nodes; | |
1732 | rtp->nodes = node->private_data; | |
1733 | kmem_cache_free(radix_tree_node_cachep, node); | |
1734 | rtp->nr--; | |
2fcd9005 | 1735 | } |
d544abd5 | 1736 | return 0; |
1da177e4 | 1737 | } |
1da177e4 LT |
1738 | |
1739 | void __init radix_tree_init(void) | |
1740 | { | |
d544abd5 | 1741 | int ret; |
1da177e4 LT |
1742 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
1743 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1744 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1745 | radix_tree_node_ctor); | |
c78c66d1 | 1746 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
1747 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
1748 | NULL, radix_tree_cpu_dead); | |
1749 | WARN_ON(ret < 0); | |
1da177e4 | 1750 | } |