]>
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); | |
643b52b9 NP |
342 | |
343 | /* | |
175542f5 MW |
344 | * Must only free zeroed nodes into the slab. We can be left with |
345 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
346 | * and tags here. | |
643b52b9 | 347 | */ |
175542f5 MW |
348 | memset(node->slots, 0, sizeof(node->slots)); |
349 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 350 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 351 | |
7cf9c2c7 NP |
352 | kmem_cache_free(radix_tree_node_cachep, node); |
353 | } | |
354 | ||
1da177e4 LT |
355 | static inline void |
356 | radix_tree_node_free(struct radix_tree_node *node) | |
357 | { | |
7cf9c2c7 | 358 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
359 | } |
360 | ||
361 | /* | |
362 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
363 | * ensure that the addition of a single element in the tree cannot fail. On | |
364 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
365 | * with preemption not disabled. | |
b34df792 DH |
366 | * |
367 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 368 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 369 | */ |
c78c66d1 | 370 | static int __radix_tree_preload(gfp_t gfp_mask, int nr) |
1da177e4 LT |
371 | { |
372 | struct radix_tree_preload *rtp; | |
373 | struct radix_tree_node *node; | |
374 | int ret = -ENOMEM; | |
375 | ||
05eb6e72 VD |
376 | /* |
377 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
378 | * they should never be accounted to any particular memory cgroup. | |
379 | */ | |
380 | gfp_mask &= ~__GFP_ACCOUNT; | |
381 | ||
1da177e4 | 382 | preempt_disable(); |
7c8e0181 | 383 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 384 | while (rtp->nr < nr) { |
1da177e4 | 385 | preempt_enable(); |
488514d1 | 386 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
387 | if (node == NULL) |
388 | goto out; | |
389 | preempt_disable(); | |
7c8e0181 | 390 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 391 | if (rtp->nr < nr) { |
9d2a8da0 KS |
392 | node->private_data = rtp->nodes; |
393 | rtp->nodes = node; | |
394 | rtp->nr++; | |
395 | } else { | |
1da177e4 | 396 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 397 | } |
1da177e4 LT |
398 | } |
399 | ret = 0; | |
400 | out: | |
401 | return ret; | |
402 | } | |
5e4c0d97 JK |
403 | |
404 | /* | |
405 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
406 | * ensure that the addition of a single element in the tree cannot fail. On | |
407 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
408 | * with preemption not disabled. | |
409 | * | |
410 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 411 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
412 | */ |
413 | int radix_tree_preload(gfp_t gfp_mask) | |
414 | { | |
415 | /* Warn on non-sensical use... */ | |
d0164adc | 416 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 417 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 418 | } |
d7f0923d | 419 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 420 | |
5e4c0d97 JK |
421 | /* |
422 | * The same as above function, except we don't guarantee preloading happens. | |
423 | * We do it, if we decide it helps. On success, return zero with preemption | |
424 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
425 | */ | |
426 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
427 | { | |
d0164adc | 428 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 429 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
430 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
431 | preempt_disable(); | |
432 | return 0; | |
433 | } | |
434 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
435 | ||
c78c66d1 KS |
436 | /* |
437 | * The same as function above, but preload number of nodes required to insert | |
438 | * (1 << order) continuous naturally-aligned elements. | |
439 | */ | |
440 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
441 | { | |
442 | unsigned long nr_subtrees; | |
443 | int nr_nodes, subtree_height; | |
444 | ||
445 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
446 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
447 | preempt_disable(); | |
448 | return 0; | |
449 | } | |
450 | ||
451 | /* | |
452 | * Calculate number and height of fully populated subtrees it takes to | |
453 | * store (1 << order) elements. | |
454 | */ | |
455 | nr_subtrees = 1 << order; | |
456 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
457 | subtree_height++) | |
458 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
459 | ||
460 | /* | |
461 | * The worst case is zero height tree with a single item at index 0 and | |
462 | * then inserting items starting at ULONG_MAX - (1 << order). | |
463 | * | |
464 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
465 | * 0-index item. | |
466 | */ | |
467 | nr_nodes = RADIX_TREE_MAX_PATH; | |
468 | ||
469 | /* Plus branch to fully populated subtrees. */ | |
470 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
471 | ||
472 | /* Root node is shared. */ | |
473 | nr_nodes--; | |
474 | ||
475 | /* Plus nodes required to build subtrees. */ | |
476 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
477 | ||
478 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
479 | } | |
480 | ||
1456a439 MW |
481 | static unsigned radix_tree_load_root(struct radix_tree_root *root, |
482 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
483 | { | |
484 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
485 | ||
486 | *nodep = node; | |
487 | ||
b194d16c | 488 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 489 | node = entry_to_node(node); |
1456a439 | 490 | *maxindex = node_maxindex(node); |
c12e51b0 | 491 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
492 | } |
493 | ||
494 | *maxindex = 0; | |
495 | return 0; | |
496 | } | |
497 | ||
1da177e4 LT |
498 | /* |
499 | * Extend a radix tree so it can store key @index. | |
500 | */ | |
e6145236 | 501 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 502 | unsigned long index, unsigned int shift) |
1da177e4 | 503 | { |
e2bdb933 | 504 | struct radix_tree_node *slot; |
d0891265 | 505 | unsigned int maxshift; |
1da177e4 LT |
506 | int tag; |
507 | ||
d0891265 MW |
508 | /* Figure out what the shift should be. */ |
509 | maxshift = shift; | |
510 | while (index > shift_maxindex(maxshift)) | |
511 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 512 | |
d0891265 MW |
513 | slot = root->rnode; |
514 | if (!slot) | |
1da177e4 | 515 | goto out; |
1da177e4 | 516 | |
1da177e4 | 517 | do { |
2fcd9005 MW |
518 | struct radix_tree_node *node = radix_tree_node_alloc(root); |
519 | ||
520 | if (!node) | |
1da177e4 LT |
521 | return -ENOMEM; |
522 | ||
1da177e4 | 523 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 524 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 525 | if (root_tag_get(root, tag)) |
1da177e4 LT |
526 | tag_set(node, tag, 0); |
527 | } | |
528 | ||
d0891265 MW |
529 | BUG_ON(shift > BITS_PER_LONG); |
530 | node->shift = shift; | |
0c7fa0a8 | 531 | node->offset = 0; |
1da177e4 | 532 | node->count = 1; |
e2bdb933 | 533 | node->parent = NULL; |
f7942430 | 534 | if (radix_tree_is_internal_node(slot)) { |
4dd6c098 | 535 | entry_to_node(slot)->parent = node; |
f7942430 JW |
536 | } else { |
537 | /* Moving an exceptional root->rnode to a node */ | |
538 | if (radix_tree_exceptional_entry(slot)) | |
539 | node->exceptional = 1; | |
540 | } | |
e2bdb933 | 541 | node->slots[0] = slot; |
a4db4dce MW |
542 | slot = node_to_entry(node); |
543 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 544 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 545 | } while (shift <= maxshift); |
1da177e4 | 546 | out: |
d0891265 | 547 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
548 | } |
549 | ||
f4b109c6 JW |
550 | /** |
551 | * radix_tree_shrink - shrink radix tree to minimum height | |
552 | * @root radix tree root | |
553 | */ | |
14b46879 | 554 | static inline void radix_tree_shrink(struct radix_tree_root *root, |
4d693d08 JW |
555 | radix_tree_update_node_t update_node, |
556 | void *private) | |
f4b109c6 | 557 | { |
f4b109c6 JW |
558 | for (;;) { |
559 | struct radix_tree_node *node = root->rnode; | |
560 | struct radix_tree_node *child; | |
561 | ||
562 | if (!radix_tree_is_internal_node(node)) | |
563 | break; | |
564 | node = entry_to_node(node); | |
565 | ||
566 | /* | |
567 | * The candidate node has more than one child, or its child | |
568 | * is not at the leftmost slot, or the child is a multiorder | |
569 | * entry, we cannot shrink. | |
570 | */ | |
571 | if (node->count != 1) | |
572 | break; | |
573 | child = node->slots[0]; | |
574 | if (!child) | |
575 | break; | |
576 | if (!radix_tree_is_internal_node(child) && node->shift) | |
577 | break; | |
578 | ||
579 | if (radix_tree_is_internal_node(child)) | |
580 | entry_to_node(child)->parent = NULL; | |
581 | ||
582 | /* | |
583 | * We don't need rcu_assign_pointer(), since we are simply | |
584 | * moving the node from one part of the tree to another: if it | |
585 | * was safe to dereference the old pointer to it | |
586 | * (node->slots[0]), it will be safe to dereference the new | |
587 | * one (root->rnode) as far as dependent read barriers go. | |
588 | */ | |
589 | root->rnode = child; | |
590 | ||
591 | /* | |
592 | * We have a dilemma here. The node's slot[0] must not be | |
593 | * NULLed in case there are concurrent lookups expecting to | |
594 | * find the item. However if this was a bottom-level node, | |
595 | * then it may be subject to the slot pointer being visible | |
596 | * to callers dereferencing it. If item corresponding to | |
597 | * slot[0] is subsequently deleted, these callers would expect | |
598 | * their slot to become empty sooner or later. | |
599 | * | |
600 | * For example, lockless pagecache will look up a slot, deref | |
601 | * the page pointer, and if the page has 0 refcount it means it | |
602 | * was concurrently deleted from pagecache so try the deref | |
603 | * again. Fortunately there is already a requirement for logic | |
604 | * to retry the entire slot lookup -- the indirect pointer | |
605 | * problem (replacing direct root node with an indirect pointer | |
606 | * also results in a stale slot). So tag the slot as indirect | |
607 | * to force callers to retry. | |
608 | */ | |
4d693d08 JW |
609 | node->count = 0; |
610 | if (!radix_tree_is_internal_node(child)) { | |
f4b109c6 | 611 | node->slots[0] = RADIX_TREE_RETRY; |
4d693d08 JW |
612 | if (update_node) |
613 | update_node(node, private); | |
614 | } | |
f4b109c6 JW |
615 | |
616 | radix_tree_node_free(node); | |
f4b109c6 | 617 | } |
f4b109c6 JW |
618 | } |
619 | ||
14b46879 | 620 | static void delete_node(struct radix_tree_root *root, |
4d693d08 JW |
621 | struct radix_tree_node *node, |
622 | radix_tree_update_node_t update_node, void *private) | |
f4b109c6 | 623 | { |
f4b109c6 JW |
624 | do { |
625 | struct radix_tree_node *parent; | |
626 | ||
627 | if (node->count) { | |
628 | if (node == entry_to_node(root->rnode)) | |
14b46879 JW |
629 | radix_tree_shrink(root, update_node, private); |
630 | return; | |
f4b109c6 JW |
631 | } |
632 | ||
633 | parent = node->parent; | |
634 | if (parent) { | |
635 | parent->slots[node->offset] = NULL; | |
636 | parent->count--; | |
637 | } else { | |
638 | root_tag_clear_all(root); | |
639 | root->rnode = NULL; | |
640 | } | |
641 | ||
642 | radix_tree_node_free(node); | |
f4b109c6 JW |
643 | |
644 | node = parent; | |
645 | } while (node); | |
f4b109c6 JW |
646 | } |
647 | ||
1da177e4 | 648 | /** |
139e5616 | 649 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
650 | * @root: radix tree root |
651 | * @index: index key | |
e6145236 | 652 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
653 | * @nodep: returns node |
654 | * @slotp: returns slot | |
1da177e4 | 655 | * |
139e5616 JW |
656 | * Create, if necessary, and return the node and slot for an item |
657 | * at position @index in the radix tree @root. | |
658 | * | |
659 | * Until there is more than one item in the tree, no nodes are | |
660 | * allocated and @root->rnode is used as a direct slot instead of | |
661 | * pointing to a node, in which case *@nodep will be NULL. | |
662 | * | |
663 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 664 | */ |
139e5616 | 665 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
666 | unsigned order, struct radix_tree_node **nodep, |
667 | void ***slotp) | |
1da177e4 | 668 | { |
89148aa4 MW |
669 | struct radix_tree_node *node = NULL, *child; |
670 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 671 | unsigned long maxindex; |
89148aa4 | 672 | unsigned int shift, offset = 0; |
49ea6ebc MW |
673 | unsigned long max = index | ((1UL << order) - 1); |
674 | ||
89148aa4 | 675 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
676 | |
677 | /* Make sure the tree is high enough. */ | |
175542f5 MW |
678 | if (order > 0 && max == ((1UL << order) - 1)) |
679 | max++; | |
49ea6ebc | 680 | if (max > maxindex) { |
d0891265 | 681 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 682 | if (error < 0) |
1da177e4 | 683 | return error; |
49ea6ebc | 684 | shift = error; |
89148aa4 | 685 | child = root->rnode; |
1da177e4 LT |
686 | } |
687 | ||
e6145236 | 688 | while (shift > order) { |
c12e51b0 | 689 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 690 | if (child == NULL) { |
1da177e4 | 691 | /* Have to add a child node. */ |
89148aa4 MW |
692 | child = radix_tree_node_alloc(root); |
693 | if (!child) | |
1da177e4 | 694 | return -ENOMEM; |
89148aa4 MW |
695 | child->shift = shift; |
696 | child->offset = offset; | |
175542f5 MW |
697 | child->count = 0; |
698 | child->exceptional = 0; | |
89148aa4 MW |
699 | child->parent = node; |
700 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
701 | if (node) | |
1da177e4 | 702 | node->count++; |
89148aa4 | 703 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 704 | break; |
1da177e4 LT |
705 | |
706 | /* Go a level down */ | |
89148aa4 | 707 | node = entry_to_node(child); |
9e85d811 | 708 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 709 | slot = &node->slots[offset]; |
e6145236 MW |
710 | } |
711 | ||
175542f5 MW |
712 | if (nodep) |
713 | *nodep = node; | |
714 | if (slotp) | |
715 | *slotp = slot; | |
716 | return 0; | |
717 | } | |
718 | ||
57578c2e | 719 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
175542f5 MW |
720 | /* |
721 | * Free any nodes below this node. The tree is presumed to not need | |
722 | * shrinking, and any user data in the tree is presumed to not need a | |
723 | * destructor called on it. If we need to add a destructor, we can | |
724 | * add that functionality later. Note that we may not clear tags or | |
725 | * slots from the tree as an RCU walker may still have a pointer into | |
726 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
727 | * but we'll still have to clear those in rcu_free. | |
728 | */ | |
729 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
730 | { | |
731 | unsigned offset = 0; | |
732 | struct radix_tree_node *child = entry_to_node(node); | |
733 | ||
734 | for (;;) { | |
735 | void *entry = child->slots[offset]; | |
736 | if (radix_tree_is_internal_node(entry) && | |
737 | !is_sibling_entry(child, entry)) { | |
738 | child = entry_to_node(entry); | |
739 | offset = 0; | |
740 | continue; | |
741 | } | |
742 | offset++; | |
743 | while (offset == RADIX_TREE_MAP_SIZE) { | |
744 | struct radix_tree_node *old = child; | |
745 | offset = child->offset + 1; | |
746 | child = child->parent; | |
747 | radix_tree_node_free(old); | |
748 | if (old == entry_to_node(node)) | |
749 | return; | |
750 | } | |
751 | } | |
752 | } | |
753 | ||
754 | static inline int insert_entries(struct radix_tree_node *node, void **slot, | |
755 | void *item, unsigned order, bool replace) | |
756 | { | |
757 | struct radix_tree_node *child; | |
758 | unsigned i, n, tag, offset, tags = 0; | |
759 | ||
760 | if (node) { | |
761 | n = 1 << (order - node->shift); | |
762 | offset = get_slot_offset(node, slot); | |
763 | } else { | |
764 | n = 1; | |
765 | offset = 0; | |
766 | } | |
767 | ||
768 | if (n > 1) { | |
e6145236 | 769 | offset = offset & ~(n - 1); |
89148aa4 | 770 | slot = &node->slots[offset]; |
175542f5 MW |
771 | } |
772 | child = node_to_entry(slot); | |
773 | ||
774 | for (i = 0; i < n; i++) { | |
775 | if (slot[i]) { | |
776 | if (replace) { | |
777 | node->count--; | |
778 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
779 | if (tag_get(node, tag, offset + i)) | |
780 | tags |= 1 << tag; | |
781 | } else | |
e6145236 MW |
782 | return -EEXIST; |
783 | } | |
175542f5 | 784 | } |
e6145236 | 785 | |
175542f5 MW |
786 | for (i = 0; i < n; i++) { |
787 | struct radix_tree_node *old = slot[i]; | |
788 | if (i) { | |
89148aa4 | 789 | rcu_assign_pointer(slot[i], child); |
175542f5 MW |
790 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
791 | if (tags & (1 << tag)) | |
792 | tag_clear(node, tag, offset + i); | |
793 | } else { | |
794 | rcu_assign_pointer(slot[i], item); | |
795 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
796 | if (tags & (1 << tag)) | |
797 | tag_set(node, tag, offset); | |
e6145236 | 798 | } |
175542f5 MW |
799 | if (radix_tree_is_internal_node(old) && |
800 | !is_sibling_entry(node, old)) | |
801 | radix_tree_free_nodes(old); | |
802 | if (radix_tree_exceptional_entry(old)) | |
803 | node->exceptional--; | |
612d6c19 | 804 | } |
175542f5 MW |
805 | if (node) { |
806 | node->count += n; | |
807 | if (radix_tree_exceptional_entry(item)) | |
808 | node->exceptional += n; | |
809 | } | |
810 | return n; | |
139e5616 | 811 | } |
175542f5 MW |
812 | #else |
813 | static inline int insert_entries(struct radix_tree_node *node, void **slot, | |
814 | void *item, unsigned order, bool replace) | |
815 | { | |
816 | if (*slot) | |
817 | return -EEXIST; | |
818 | rcu_assign_pointer(*slot, item); | |
819 | if (node) { | |
820 | node->count++; | |
821 | if (radix_tree_exceptional_entry(item)) | |
822 | node->exceptional++; | |
823 | } | |
824 | return 1; | |
825 | } | |
826 | #endif | |
139e5616 JW |
827 | |
828 | /** | |
e6145236 | 829 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
830 | * @root: radix tree root |
831 | * @index: index key | |
e6145236 | 832 | * @order: key covers the 2^order indices around index |
139e5616 JW |
833 | * @item: item to insert |
834 | * | |
835 | * Insert an item into the radix tree at position @index. | |
836 | */ | |
e6145236 MW |
837 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
838 | unsigned order, void *item) | |
139e5616 JW |
839 | { |
840 | struct radix_tree_node *node; | |
841 | void **slot; | |
842 | int error; | |
843 | ||
b194d16c | 844 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 845 | |
e6145236 | 846 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
847 | if (error) |
848 | return error; | |
175542f5 MW |
849 | |
850 | error = insert_entries(node, slot, item, order, false); | |
851 | if (error < 0) | |
852 | return error; | |
201b6264 | 853 | |
612d6c19 | 854 | if (node) { |
7b60e9ad | 855 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
856 | BUG_ON(tag_get(node, 0, offset)); |
857 | BUG_ON(tag_get(node, 1, offset)); | |
858 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 859 | } else { |
7b60e9ad | 860 | BUG_ON(root_tags_get(root)); |
612d6c19 | 861 | } |
1da177e4 | 862 | |
1da177e4 LT |
863 | return 0; |
864 | } | |
e6145236 | 865 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 866 | |
139e5616 JW |
867 | /** |
868 | * __radix_tree_lookup - lookup an item in a radix tree | |
869 | * @root: radix tree root | |
870 | * @index: index key | |
871 | * @nodep: returns node | |
872 | * @slotp: returns slot | |
873 | * | |
874 | * Lookup and return the item at position @index in the radix | |
875 | * tree @root. | |
876 | * | |
877 | * Until there is more than one item in the tree, no nodes are | |
878 | * allocated and @root->rnode is used as a direct slot instead of | |
879 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 880 | */ |
139e5616 JW |
881 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
882 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 883 | { |
139e5616 | 884 | struct radix_tree_node *node, *parent; |
85829954 | 885 | unsigned long maxindex; |
139e5616 | 886 | void **slot; |
612d6c19 | 887 | |
85829954 MW |
888 | restart: |
889 | parent = NULL; | |
890 | slot = (void **)&root->rnode; | |
9e85d811 | 891 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 892 | if (index > maxindex) |
1da177e4 LT |
893 | return NULL; |
894 | ||
b194d16c | 895 | while (radix_tree_is_internal_node(node)) { |
85829954 | 896 | unsigned offset; |
1da177e4 | 897 | |
85829954 MW |
898 | if (node == RADIX_TREE_RETRY) |
899 | goto restart; | |
4dd6c098 | 900 | parent = entry_to_node(node); |
9e85d811 | 901 | offset = radix_tree_descend(parent, &node, index); |
85829954 MW |
902 | slot = parent->slots + offset; |
903 | } | |
1da177e4 | 904 | |
139e5616 JW |
905 | if (nodep) |
906 | *nodep = parent; | |
907 | if (slotp) | |
908 | *slotp = slot; | |
909 | return node; | |
b72b71c6 HS |
910 | } |
911 | ||
912 | /** | |
913 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
914 | * @root: radix tree root | |
915 | * @index: index key | |
916 | * | |
917 | * Returns: the slot corresponding to the position @index in the | |
918 | * radix tree @root. This is useful for update-if-exists operations. | |
919 | * | |
920 | * This function can be called under rcu_read_lock iff the slot is not | |
921 | * modified by radix_tree_replace_slot, otherwise it must be called | |
922 | * exclusive from other writers. Any dereference of the slot must be done | |
923 | * using radix_tree_deref_slot. | |
924 | */ | |
925 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
926 | { | |
139e5616 JW |
927 | void **slot; |
928 | ||
929 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
930 | return NULL; | |
931 | return slot; | |
a4331366 | 932 | } |
a4331366 HR |
933 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
934 | ||
935 | /** | |
936 | * radix_tree_lookup - perform lookup operation on a radix tree | |
937 | * @root: radix tree root | |
938 | * @index: index key | |
939 | * | |
940 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
941 | * |
942 | * This function can be called under rcu_read_lock, however the caller | |
943 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
944 | * them safely). No RCU barriers are required to access or modify the | |
945 | * returned item, however. | |
a4331366 HR |
946 | */ |
947 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
948 | { | |
139e5616 | 949 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
950 | } |
951 | EXPORT_SYMBOL(radix_tree_lookup); | |
952 | ||
6d75f366 JW |
953 | static void replace_slot(struct radix_tree_root *root, |
954 | struct radix_tree_node *node, | |
955 | void **slot, void *item, | |
956 | bool warn_typeswitch) | |
f7942430 JW |
957 | { |
958 | void *old = rcu_dereference_raw(*slot); | |
f4b109c6 | 959 | int count, exceptional; |
f7942430 JW |
960 | |
961 | WARN_ON_ONCE(radix_tree_is_internal_node(item)); | |
f7942430 | 962 | |
f4b109c6 | 963 | count = !!item - !!old; |
f7942430 JW |
964 | exceptional = !!radix_tree_exceptional_entry(item) - |
965 | !!radix_tree_exceptional_entry(old); | |
966 | ||
f4b109c6 | 967 | WARN_ON_ONCE(warn_typeswitch && (count || exceptional)); |
f7942430 | 968 | |
f4b109c6 JW |
969 | if (node) { |
970 | node->count += count; | |
f7942430 | 971 | node->exceptional += exceptional; |
f4b109c6 | 972 | } |
f7942430 JW |
973 | |
974 | rcu_assign_pointer(*slot, item); | |
975 | } | |
976 | ||
6d75f366 JW |
977 | /** |
978 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
979 | * @root: radix tree root |
980 | * @node: pointer to tree node | |
981 | * @slot: pointer to slot in @node | |
982 | * @item: new item to store in the slot. | |
983 | * @update_node: callback for changing leaf nodes | |
984 | * @private: private data to pass to @update_node | |
6d75f366 JW |
985 | * |
986 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
987 | * across slot lookup and replacement. | |
988 | */ | |
989 | void __radix_tree_replace(struct radix_tree_root *root, | |
990 | struct radix_tree_node *node, | |
4d693d08 JW |
991 | void **slot, void *item, |
992 | radix_tree_update_node_t update_node, void *private) | |
6d75f366 JW |
993 | { |
994 | /* | |
f4b109c6 JW |
995 | * This function supports replacing exceptional entries and |
996 | * deleting entries, but that needs accounting against the | |
997 | * node unless the slot is root->rnode. | |
6d75f366 JW |
998 | */ |
999 | replace_slot(root, node, slot, item, | |
1000 | !node && slot != (void **)&root->rnode); | |
f4b109c6 | 1001 | |
4d693d08 JW |
1002 | if (!node) |
1003 | return; | |
1004 | ||
1005 | if (update_node) | |
1006 | update_node(node, private); | |
1007 | ||
1008 | delete_node(root, node, update_node, private); | |
6d75f366 JW |
1009 | } |
1010 | ||
1011 | /** | |
1012 | * radix_tree_replace_slot - replace item in a slot | |
1013 | * @root: radix tree root | |
1014 | * @slot: pointer to slot | |
1015 | * @item: new item to store in the slot. | |
1016 | * | |
1017 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
1018 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
1019 | * across slot lookup and replacement. | |
1020 | * | |
1021 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
1022 | * regular entries, and exceptional entries, as that requires accounting | |
f4b109c6 JW |
1023 | * inside the radix tree node. When switching from one type of entry or |
1024 | * deleting, use __radix_tree_lookup() and __radix_tree_replace(). | |
6d75f366 JW |
1025 | */ |
1026 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
1027 | void **slot, void *item) | |
1028 | { | |
1029 | replace_slot(root, NULL, slot, item, true); | |
1030 | } | |
1031 | ||
175542f5 MW |
1032 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
1033 | /** | |
1034 | * radix_tree_join - replace multiple entries with one multiorder entry | |
1035 | * @root: radix tree root | |
1036 | * @index: an index inside the new entry | |
1037 | * @order: order of the new entry | |
1038 | * @item: new entry | |
1039 | * | |
1040 | * Call this function to replace several entries with one larger entry. | |
1041 | * The existing entries are presumed to not need freeing as a result of | |
1042 | * this call. | |
1043 | * | |
1044 | * The replacement entry will have all the tags set on it that were set | |
1045 | * on any of the entries it is replacing. | |
1046 | */ | |
1047 | int radix_tree_join(struct radix_tree_root *root, unsigned long index, | |
1048 | unsigned order, void *item) | |
1049 | { | |
1050 | struct radix_tree_node *node; | |
1051 | void **slot; | |
1052 | int error; | |
1053 | ||
1054 | BUG_ON(radix_tree_is_internal_node(item)); | |
1055 | ||
1056 | error = __radix_tree_create(root, index, order, &node, &slot); | |
1057 | if (!error) | |
1058 | error = insert_entries(node, slot, item, order, true); | |
1059 | if (error > 0) | |
1060 | error = 0; | |
1061 | ||
1062 | return error; | |
1063 | } | |
1064 | #endif | |
1065 | ||
1da177e4 LT |
1066 | /** |
1067 | * radix_tree_tag_set - set a tag on a radix tree node | |
1068 | * @root: radix tree root | |
1069 | * @index: index key | |
2fcd9005 | 1070 | * @tag: tag index |
1da177e4 | 1071 | * |
daff89f3 JC |
1072 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
1073 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
1074 | * the root all the way down to the leaf node. |
1075 | * | |
2fcd9005 | 1076 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
1077 | * item is a bug. |
1078 | */ | |
1079 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 1080 | unsigned long index, unsigned int tag) |
1da177e4 | 1081 | { |
fb969909 RZ |
1082 | struct radix_tree_node *node, *parent; |
1083 | unsigned long maxindex; | |
1da177e4 | 1084 | |
9e85d811 | 1085 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 1086 | BUG_ON(index > maxindex); |
1da177e4 | 1087 | |
b194d16c | 1088 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 1089 | unsigned offset; |
1da177e4 | 1090 | |
4dd6c098 | 1091 | parent = entry_to_node(node); |
9e85d811 | 1092 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
1093 | BUG_ON(!node); |
1094 | ||
1095 | if (!tag_get(parent, tag, offset)) | |
1096 | tag_set(parent, tag, offset); | |
1da177e4 LT |
1097 | } |
1098 | ||
612d6c19 | 1099 | /* set the root's tag bit */ |
fb969909 | 1100 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
1101 | root_tag_set(root, tag); |
1102 | ||
fb969909 | 1103 | return node; |
1da177e4 LT |
1104 | } |
1105 | EXPORT_SYMBOL(radix_tree_tag_set); | |
1106 | ||
d604c324 MW |
1107 | static void node_tag_clear(struct radix_tree_root *root, |
1108 | struct radix_tree_node *node, | |
1109 | unsigned int tag, unsigned int offset) | |
1110 | { | |
1111 | while (node) { | |
1112 | if (!tag_get(node, tag, offset)) | |
1113 | return; | |
1114 | tag_clear(node, tag, offset); | |
1115 | if (any_tag_set(node, tag)) | |
1116 | return; | |
1117 | ||
1118 | offset = node->offset; | |
1119 | node = node->parent; | |
1120 | } | |
1121 | ||
1122 | /* clear the root's tag bit */ | |
1123 | if (root_tag_get(root, tag)) | |
1124 | root_tag_clear(root, tag); | |
1125 | } | |
1126 | ||
9498d2bb MW |
1127 | static void node_tag_set(struct radix_tree_root *root, |
1128 | struct radix_tree_node *node, | |
1129 | unsigned int tag, unsigned int offset) | |
1130 | { | |
1131 | while (node) { | |
1132 | if (tag_get(node, tag, offset)) | |
1133 | return; | |
1134 | tag_set(node, tag, offset); | |
1135 | offset = node->offset; | |
1136 | node = node->parent; | |
1137 | } | |
1138 | ||
1139 | if (!root_tag_get(root, tag)) | |
1140 | root_tag_set(root, tag); | |
1141 | } | |
1142 | ||
268f42de MW |
1143 | /** |
1144 | * radix_tree_iter_tag_set - set a tag on the current iterator entry | |
1145 | * @root: radix tree root | |
1146 | * @iter: iterator state | |
1147 | * @tag: tag to set | |
1148 | */ | |
1149 | void radix_tree_iter_tag_set(struct radix_tree_root *root, | |
1150 | const struct radix_tree_iter *iter, unsigned int tag) | |
1151 | { | |
1152 | node_tag_set(root, iter->node, tag, iter_offset(iter)); | |
1153 | } | |
1154 | ||
1da177e4 LT |
1155 | /** |
1156 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1157 | * @root: radix tree root | |
1158 | * @index: index key | |
2fcd9005 | 1159 | * @tag: tag index |
1da177e4 | 1160 | * |
daff89f3 | 1161 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1162 | * corresponding to @index in the radix tree. If this causes |
1163 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1164 | * next-to-leaf node, etc. |
1165 | * | |
1166 | * Returns the address of the tagged item on success, else NULL. ie: | |
1167 | * has the same return value and semantics as radix_tree_lookup(). | |
1168 | */ | |
1169 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1170 | unsigned long index, unsigned int tag) |
1da177e4 | 1171 | { |
00f47b58 RZ |
1172 | struct radix_tree_node *node, *parent; |
1173 | unsigned long maxindex; | |
e2bdb933 | 1174 | int uninitialized_var(offset); |
1da177e4 | 1175 | |
9e85d811 | 1176 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1177 | if (index > maxindex) |
1178 | return NULL; | |
1da177e4 | 1179 | |
00f47b58 | 1180 | parent = NULL; |
1da177e4 | 1181 | |
b194d16c | 1182 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1183 | parent = entry_to_node(node); |
9e85d811 | 1184 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1185 | } |
1186 | ||
d604c324 MW |
1187 | if (node) |
1188 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1189 | |
00f47b58 | 1190 | return node; |
1da177e4 LT |
1191 | } |
1192 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1193 | ||
1da177e4 | 1194 | /** |
32605a18 MT |
1195 | * radix_tree_tag_get - get a tag on a radix tree node |
1196 | * @root: radix tree root | |
1197 | * @index: index key | |
2fcd9005 | 1198 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1199 | * |
32605a18 | 1200 | * Return values: |
1da177e4 | 1201 | * |
612d6c19 NP |
1202 | * 0: tag not present or not set |
1203 | * 1: tag set | |
ce82653d DH |
1204 | * |
1205 | * Note that the return value of this function may not be relied on, even if | |
1206 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1207 | * from concurrency. | |
1da177e4 LT |
1208 | */ |
1209 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 1210 | unsigned long index, unsigned int tag) |
1da177e4 | 1211 | { |
4589ba6d RZ |
1212 | struct radix_tree_node *node, *parent; |
1213 | unsigned long maxindex; | |
1da177e4 | 1214 | |
612d6c19 NP |
1215 | if (!root_tag_get(root, tag)) |
1216 | return 0; | |
1217 | ||
9e85d811 | 1218 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1219 | if (index > maxindex) |
1220 | return 0; | |
7cf9c2c7 NP |
1221 | if (node == NULL) |
1222 | return 0; | |
1223 | ||
b194d16c | 1224 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1225 | unsigned offset; |
1da177e4 | 1226 | |
4dd6c098 | 1227 | parent = entry_to_node(node); |
9e85d811 | 1228 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1229 | |
4589ba6d | 1230 | if (!node) |
1da177e4 | 1231 | return 0; |
4589ba6d | 1232 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1233 | return 0; |
4589ba6d RZ |
1234 | if (node == RADIX_TREE_RETRY) |
1235 | break; | |
1da177e4 | 1236 | } |
4589ba6d RZ |
1237 | |
1238 | return 1; | |
1da177e4 LT |
1239 | } |
1240 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1241 | |
21ef5339 RZ |
1242 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1243 | unsigned int shift) | |
1244 | { | |
1245 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1246 | iter->shift = shift; | |
1247 | #endif | |
1248 | } | |
1249 | ||
148deab2 MW |
1250 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1251 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1252 | struct radix_tree_node *node, unsigned offset, | |
1253 | unsigned tag) | |
1254 | { | |
1255 | unsigned tag_long = offset / BITS_PER_LONG; | |
1256 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1257 | ||
1258 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1259 | ||
1260 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1261 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1262 | /* Pick tags from next element */ | |
1263 | if (tag_bit) | |
1264 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1265 | (BITS_PER_LONG - tag_bit); | |
1266 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1267 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1272 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1273 | void **slot, struct radix_tree_iter *iter) | |
1274 | { | |
1275 | void *sib = node_to_entry(slot - 1); | |
1276 | ||
1277 | while (iter->index < iter->next_index) { | |
1278 | *nodep = rcu_dereference_raw(*slot); | |
1279 | if (*nodep && *nodep != sib) | |
1280 | return slot; | |
1281 | slot++; | |
1282 | iter->index = __radix_tree_iter_add(iter, 1); | |
1283 | iter->tags >>= 1; | |
1284 | } | |
1285 | ||
1286 | *nodep = NULL; | |
1287 | return NULL; | |
1288 | } | |
1289 | ||
1290 | void ** __radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, | |
1291 | unsigned flags) | |
1292 | { | |
1293 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
1294 | struct radix_tree_node *node = rcu_dereference_raw(*slot); | |
1295 | ||
1296 | slot = skip_siblings(&node, slot, iter); | |
1297 | ||
1298 | while (radix_tree_is_internal_node(node)) { | |
1299 | unsigned offset; | |
1300 | unsigned long next_index; | |
1301 | ||
1302 | if (node == RADIX_TREE_RETRY) | |
1303 | return slot; | |
1304 | node = entry_to_node(node); | |
268f42de | 1305 | iter->node = node; |
148deab2 MW |
1306 | iter->shift = node->shift; |
1307 | ||
1308 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1309 | offset = radix_tree_find_next_bit(node, tag, 0); | |
1310 | if (offset == RADIX_TREE_MAP_SIZE) | |
1311 | return NULL; | |
1312 | slot = &node->slots[offset]; | |
1313 | iter->index = __radix_tree_iter_add(iter, offset); | |
1314 | set_iter_tags(iter, node, offset, tag); | |
1315 | node = rcu_dereference_raw(*slot); | |
1316 | } else { | |
1317 | offset = 0; | |
1318 | slot = &node->slots[0]; | |
1319 | for (;;) { | |
1320 | node = rcu_dereference_raw(*slot); | |
1321 | if (node) | |
1322 | break; | |
1323 | slot++; | |
1324 | offset++; | |
1325 | if (offset == RADIX_TREE_MAP_SIZE) | |
1326 | return NULL; | |
1327 | } | |
1328 | iter->index = __radix_tree_iter_add(iter, offset); | |
1329 | } | |
1330 | if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0)) | |
1331 | goto none; | |
1332 | next_index = (iter->index | shift_maxindex(iter->shift)) + 1; | |
1333 | if (next_index < iter->next_index) | |
1334 | iter->next_index = next_index; | |
1335 | } | |
1336 | ||
1337 | return slot; | |
1338 | none: | |
1339 | iter->next_index = 0; | |
1340 | return NULL; | |
1341 | } | |
1342 | EXPORT_SYMBOL(__radix_tree_next_slot); | |
1343 | #else | |
1344 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1345 | void **slot, struct radix_tree_iter *iter) | |
1346 | { | |
1347 | return slot; | |
1348 | } | |
1349 | #endif | |
1350 | ||
1351 | void **radix_tree_iter_resume(void **slot, struct radix_tree_iter *iter) | |
1352 | { | |
1353 | struct radix_tree_node *node; | |
1354 | ||
1355 | slot++; | |
1356 | iter->index = __radix_tree_iter_add(iter, 1); | |
1357 | node = rcu_dereference_raw(*slot); | |
1358 | skip_siblings(&node, slot, iter); | |
1359 | iter->next_index = iter->index; | |
1360 | iter->tags = 0; | |
1361 | return NULL; | |
1362 | } | |
1363 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1364 | ||
78c1d784 KK |
1365 | /** |
1366 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1367 | * | |
1368 | * @root: radix tree root | |
1369 | * @iter: iterator state | |
1370 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1371 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1372 | */ | |
1373 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
1374 | struct radix_tree_iter *iter, unsigned flags) | |
1375 | { | |
9e85d811 | 1376 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1377 | struct radix_tree_node *node, *child; |
21ef5339 | 1378 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1379 | |
1380 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1381 | return NULL; | |
1382 | ||
1383 | /* | |
1384 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1385 | * during iterating; it can be zero only at the beginning. | |
1386 | * And we cannot overflow iter->next_index in a single step, | |
1387 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1388 | * |
1389 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1390 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1391 | */ |
1392 | index = iter->next_index; | |
1393 | if (!index && iter->index) | |
1394 | return NULL; | |
1395 | ||
21ef5339 | 1396 | restart: |
9e85d811 | 1397 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1398 | if (index > maxindex) |
1399 | return NULL; | |
8c1244de MW |
1400 | if (!child) |
1401 | return NULL; | |
21ef5339 | 1402 | |
8c1244de | 1403 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1404 | /* Single-slot tree */ |
21ef5339 RZ |
1405 | iter->index = index; |
1406 | iter->next_index = maxindex + 1; | |
78c1d784 | 1407 | iter->tags = 1; |
268f42de | 1408 | iter->node = NULL; |
8c1244de | 1409 | __set_iter_shift(iter, 0); |
78c1d784 | 1410 | return (void **)&root->rnode; |
8c1244de | 1411 | } |
21ef5339 | 1412 | |
8c1244de MW |
1413 | do { |
1414 | node = entry_to_node(child); | |
9e85d811 | 1415 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1416 | |
78c1d784 | 1417 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1418 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1419 | /* Hole detected */ |
1420 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1421 | return NULL; | |
1422 | ||
1423 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1424 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1425 | offset + 1); |
1426 | else | |
1427 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
1428 | void *slot = node->slots[offset]; |
1429 | if (is_sibling_entry(node, slot)) | |
1430 | continue; | |
1431 | if (slot) | |
78c1d784 KK |
1432 | break; |
1433 | } | |
8c1244de | 1434 | index &= ~node_maxindex(node); |
9e85d811 | 1435 | index += offset << node->shift; |
78c1d784 KK |
1436 | /* Overflow after ~0UL */ |
1437 | if (!index) | |
1438 | return NULL; | |
1439 | if (offset == RADIX_TREE_MAP_SIZE) | |
1440 | goto restart; | |
8c1244de | 1441 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1442 | } |
1443 | ||
8c1244de | 1444 | if ((child == NULL) || (child == RADIX_TREE_RETRY)) |
78c1d784 | 1445 | goto restart; |
8c1244de | 1446 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
1447 | |
1448 | /* Update the iterator state */ | |
8c1244de MW |
1449 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1450 | iter->next_index = (index | node_maxindex(node)) + 1; | |
268f42de | 1451 | iter->node = node; |
9e85d811 | 1452 | __set_iter_shift(iter, node->shift); |
78c1d784 | 1453 | |
148deab2 MW |
1454 | if (flags & RADIX_TREE_ITER_TAGGED) |
1455 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1456 | |
1457 | return node->slots + offset; | |
1458 | } | |
1459 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1460 | ||
1da177e4 LT |
1461 | /** |
1462 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1463 | * @root: radix tree root | |
1464 | * @results: where the results of the lookup are placed | |
1465 | * @first_index: start the lookup from this key | |
1466 | * @max_items: place up to this many items at *results | |
1467 | * | |
1468 | * Performs an index-ascending scan of the tree for present items. Places | |
1469 | * them at *@results and returns the number of items which were placed at | |
1470 | * *@results. | |
1471 | * | |
1472 | * The implementation is naive. | |
7cf9c2c7 NP |
1473 | * |
1474 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1475 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1476 | * an atomic snapshot of the tree at a single point in time, the |
1477 | * semantics of an RCU protected gang lookup are as though multiple | |
1478 | * radix_tree_lookups have been issued in individual locks, and results | |
1479 | * stored in 'results'. | |
1da177e4 LT |
1480 | */ |
1481 | unsigned int | |
1482 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1483 | unsigned long first_index, unsigned int max_items) | |
1484 | { | |
cebbd29e KK |
1485 | struct radix_tree_iter iter; |
1486 | void **slot; | |
1487 | unsigned int ret = 0; | |
7cf9c2c7 | 1488 | |
cebbd29e | 1489 | if (unlikely(!max_items)) |
7cf9c2c7 | 1490 | return 0; |
1da177e4 | 1491 | |
cebbd29e | 1492 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1493 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1494 | if (!results[ret]) |
1495 | continue; | |
b194d16c | 1496 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1497 | slot = radix_tree_iter_retry(&iter); |
1498 | continue; | |
1499 | } | |
cebbd29e | 1500 | if (++ret == max_items) |
1da177e4 | 1501 | break; |
1da177e4 | 1502 | } |
7cf9c2c7 | 1503 | |
1da177e4 LT |
1504 | return ret; |
1505 | } | |
1506 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1507 | ||
47feff2c NP |
1508 | /** |
1509 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1510 | * @root: radix tree root | |
1511 | * @results: where the results of the lookup are placed | |
6328650b | 1512 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1513 | * @first_index: start the lookup from this key |
1514 | * @max_items: place up to this many items at *results | |
1515 | * | |
1516 | * Performs an index-ascending scan of the tree for present items. Places | |
1517 | * their slots at *@results and returns the number of items which were | |
1518 | * placed at *@results. | |
1519 | * | |
1520 | * The implementation is naive. | |
1521 | * | |
1522 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1523 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1524 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1525 | */ | |
1526 | unsigned int | |
6328650b HD |
1527 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1528 | void ***results, unsigned long *indices, | |
47feff2c NP |
1529 | unsigned long first_index, unsigned int max_items) |
1530 | { | |
cebbd29e KK |
1531 | struct radix_tree_iter iter; |
1532 | void **slot; | |
1533 | unsigned int ret = 0; | |
47feff2c | 1534 | |
cebbd29e | 1535 | if (unlikely(!max_items)) |
47feff2c NP |
1536 | return 0; |
1537 | ||
cebbd29e KK |
1538 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1539 | results[ret] = slot; | |
6328650b | 1540 | if (indices) |
cebbd29e KK |
1541 | indices[ret] = iter.index; |
1542 | if (++ret == max_items) | |
47feff2c | 1543 | break; |
47feff2c NP |
1544 | } |
1545 | ||
1546 | return ret; | |
1547 | } | |
1548 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1549 | ||
1da177e4 LT |
1550 | /** |
1551 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1552 | * based on a tag | |
1553 | * @root: radix tree root | |
1554 | * @results: where the results of the lookup are placed | |
1555 | * @first_index: start the lookup from this key | |
1556 | * @max_items: place up to this many items at *results | |
daff89f3 | 1557 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1558 | * |
1559 | * Performs an index-ascending scan of the tree for present items which | |
1560 | * have the tag indexed by @tag set. Places the items at *@results and | |
1561 | * returns the number of items which were placed at *@results. | |
1562 | */ | |
1563 | unsigned int | |
1564 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1565 | unsigned long first_index, unsigned int max_items, |
1566 | unsigned int tag) | |
1da177e4 | 1567 | { |
cebbd29e KK |
1568 | struct radix_tree_iter iter; |
1569 | void **slot; | |
1570 | unsigned int ret = 0; | |
612d6c19 | 1571 | |
cebbd29e | 1572 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1573 | return 0; |
1574 | ||
cebbd29e | 1575 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1576 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1577 | if (!results[ret]) |
1578 | continue; | |
b194d16c | 1579 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1580 | slot = radix_tree_iter_retry(&iter); |
1581 | continue; | |
1582 | } | |
cebbd29e | 1583 | if (++ret == max_items) |
1da177e4 | 1584 | break; |
1da177e4 | 1585 | } |
7cf9c2c7 | 1586 | |
1da177e4 LT |
1587 | return ret; |
1588 | } | |
1589 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1590 | ||
47feff2c NP |
1591 | /** |
1592 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1593 | * radix tree based on a tag | |
1594 | * @root: radix tree root | |
1595 | * @results: where the results of the lookup are placed | |
1596 | * @first_index: start the lookup from this key | |
1597 | * @max_items: place up to this many items at *results | |
1598 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1599 | * | |
1600 | * Performs an index-ascending scan of the tree for present items which | |
1601 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1602 | * returns the number of slots which were placed at *@results. | |
1603 | */ | |
1604 | unsigned int | |
1605 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1606 | unsigned long first_index, unsigned int max_items, | |
1607 | unsigned int tag) | |
1608 | { | |
cebbd29e KK |
1609 | struct radix_tree_iter iter; |
1610 | void **slot; | |
1611 | unsigned int ret = 0; | |
47feff2c | 1612 | |
cebbd29e | 1613 | if (unlikely(!max_items)) |
47feff2c NP |
1614 | return 0; |
1615 | ||
cebbd29e KK |
1616 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1617 | results[ret] = slot; | |
1618 | if (++ret == max_items) | |
47feff2c | 1619 | break; |
47feff2c NP |
1620 | } |
1621 | ||
1622 | return ret; | |
1623 | } | |
1624 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1625 | ||
139e5616 JW |
1626 | /** |
1627 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1628 | * @root: radix tree root | |
139e5616 JW |
1629 | * @node: node containing @index |
1630 | * | |
1631 | * After clearing the slot at @index in @node from radix tree | |
1632 | * rooted at @root, call this function to attempt freeing the | |
1633 | * node and shrinking the tree. | |
139e5616 | 1634 | */ |
14b46879 | 1635 | void __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1636 | struct radix_tree_node *node) |
1637 | { | |
14b46879 | 1638 | delete_node(root, node, NULL, NULL); |
139e5616 JW |
1639 | } |
1640 | ||
57578c2e MW |
1641 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1642 | void *ptr, unsigned offset) | |
1643 | { | |
1644 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1645 | int i; | |
1646 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1647 | if (node->slots[offset + i] != ptr) | |
1648 | break; | |
1649 | node->slots[offset + i] = NULL; | |
1650 | node->count--; | |
1651 | } | |
1652 | #endif | |
1653 | } | |
1654 | ||
1da177e4 | 1655 | /** |
53c59f26 | 1656 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1657 | * @root: radix tree root |
1658 | * @index: index key | |
53c59f26 | 1659 | * @item: expected item |
1da177e4 | 1660 | * |
53c59f26 | 1661 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1662 | * |
53c59f26 JW |
1663 | * Returns the address of the deleted item, or NULL if it was not present |
1664 | * or the entry at the given @index was not @item. | |
1da177e4 | 1665 | */ |
53c59f26 JW |
1666 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1667 | unsigned long index, void *item) | |
1da177e4 | 1668 | { |
139e5616 | 1669 | struct radix_tree_node *node; |
57578c2e | 1670 | unsigned int offset; |
139e5616 JW |
1671 | void **slot; |
1672 | void *entry; | |
d5274261 | 1673 | int tag; |
1da177e4 | 1674 | |
139e5616 JW |
1675 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1676 | if (!entry) | |
1677 | return NULL; | |
1da177e4 | 1678 | |
139e5616 JW |
1679 | if (item && entry != item) |
1680 | return NULL; | |
1681 | ||
1682 | if (!node) { | |
612d6c19 NP |
1683 | root_tag_clear_all(root); |
1684 | root->rnode = NULL; | |
139e5616 | 1685 | return entry; |
612d6c19 | 1686 | } |
1da177e4 | 1687 | |
29e0967c | 1688 | offset = get_slot_offset(node, slot); |
53c59f26 | 1689 | |
d604c324 MW |
1690 | /* Clear all tags associated with the item to be deleted. */ |
1691 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1692 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1693 | |
a4db4dce | 1694 | delete_sibling_entries(node, node_to_entry(slot), offset); |
4d693d08 | 1695 | __radix_tree_replace(root, node, slot, NULL, NULL, NULL); |
612d6c19 | 1696 | |
139e5616 | 1697 | return entry; |
1da177e4 | 1698 | } |
53c59f26 JW |
1699 | EXPORT_SYMBOL(radix_tree_delete_item); |
1700 | ||
1701 | /** | |
1702 | * radix_tree_delete - delete an item from a radix tree | |
1703 | * @root: radix tree root | |
1704 | * @index: index key | |
1705 | * | |
1706 | * Remove the item at @index from the radix tree rooted at @root. | |
1707 | * | |
1708 | * Returns the address of the deleted item, or NULL if it was not present. | |
1709 | */ | |
1710 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1711 | { | |
1712 | return radix_tree_delete_item(root, index, NULL); | |
1713 | } | |
1da177e4 LT |
1714 | EXPORT_SYMBOL(radix_tree_delete); |
1715 | ||
d3798ae8 JW |
1716 | void radix_tree_clear_tags(struct radix_tree_root *root, |
1717 | struct radix_tree_node *node, | |
1718 | void **slot) | |
d604c324 | 1719 | { |
d604c324 MW |
1720 | if (node) { |
1721 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1722 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1723 | node_tag_clear(root, node, tag, offset); | |
1724 | } else { | |
1725 | /* Clear root node tags */ | |
1726 | root->gfp_mask &= __GFP_BITS_MASK; | |
1727 | } | |
d604c324 MW |
1728 | } |
1729 | ||
1da177e4 LT |
1730 | /** |
1731 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1732 | * @root: radix tree root | |
1733 | * @tag: tag to test | |
1734 | */ | |
daff89f3 | 1735 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1736 | { |
612d6c19 | 1737 | return root_tag_get(root, tag); |
1da177e4 LT |
1738 | } |
1739 | EXPORT_SYMBOL(radix_tree_tagged); | |
1740 | ||
1741 | static void | |
449dd698 | 1742 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1743 | { |
449dd698 JW |
1744 | struct radix_tree_node *node = arg; |
1745 | ||
1746 | memset(node, 0, sizeof(*node)); | |
1747 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1748 | } |
1749 | ||
c78c66d1 KS |
1750 | static __init unsigned long __maxindex(unsigned int height) |
1751 | { | |
1752 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
1753 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1754 | ||
1755 | if (shift < 0) | |
1756 | return ~0UL; | |
1757 | if (shift >= BITS_PER_LONG) | |
1758 | return 0UL; | |
1759 | return ~0UL >> shift; | |
1760 | } | |
1761 | ||
1762 | static __init void radix_tree_init_maxnodes(void) | |
1763 | { | |
1764 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
1765 | unsigned int i, j; | |
1766 | ||
1767 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1768 | height_to_maxindex[i] = __maxindex(i); | |
1769 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
1770 | for (j = i; j > 0; j--) | |
1771 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
1772 | } | |
1773 | } | |
1774 | ||
d544abd5 | 1775 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1776 | { |
2fcd9005 MW |
1777 | struct radix_tree_preload *rtp; |
1778 | struct radix_tree_node *node; | |
1779 | ||
1780 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1781 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1782 | while (rtp->nr) { | |
1783 | node = rtp->nodes; | |
1784 | rtp->nodes = node->private_data; | |
1785 | kmem_cache_free(radix_tree_node_cachep, node); | |
1786 | rtp->nr--; | |
2fcd9005 | 1787 | } |
d544abd5 | 1788 | return 0; |
1da177e4 | 1789 | } |
1da177e4 LT |
1790 | |
1791 | void __init radix_tree_init(void) | |
1792 | { | |
d544abd5 | 1793 | int ret; |
1da177e4 LT |
1794 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
1795 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1796 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1797 | radix_tree_node_ctor); | |
c78c66d1 | 1798 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
1799 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
1800 | NULL, radix_tree_cpu_dead); | |
1801 | WARN_ON(ret < 0); | |
1da177e4 | 1802 | } |