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 | ||
e157b555 | 25 | #include <linux/cpu.h> |
1da177e4 LT |
26 | #include <linux/errno.h> |
27 | #include <linux/init.h> | |
28 | #include <linux/kernel.h> | |
8bc3bcc9 | 29 | #include <linux/export.h> |
1da177e4 LT |
30 | #include <linux/radix-tree.h> |
31 | #include <linux/percpu.h> | |
32 | #include <linux/slab.h> | |
ce80b067 | 33 | #include <linux/kmemleak.h> |
1da177e4 | 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 */ | |
35534c86 MW |
86 | static inline |
87 | bool is_sibling_entry(const struct radix_tree_node *parent, void *node) | |
db050f29 MW |
88 | { |
89 | void **ptr = node; | |
90 | return (parent->slots <= ptr) && | |
91 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
92 | } | |
93 | #else | |
35534c86 MW |
94 | static inline |
95 | bool is_sibling_entry(const struct radix_tree_node *parent, void *node) | |
db050f29 MW |
96 | { |
97 | return false; | |
98 | } | |
99 | #endif | |
100 | ||
35534c86 MW |
101 | static inline |
102 | unsigned long get_slot_offset(const struct radix_tree_node *parent, void **slot) | |
db050f29 MW |
103 | { |
104 | return slot - parent->slots; | |
105 | } | |
106 | ||
35534c86 | 107 | static unsigned int radix_tree_descend(const struct radix_tree_node *parent, |
9e85d811 | 108 | struct radix_tree_node **nodep, unsigned long index) |
db050f29 | 109 | { |
9e85d811 | 110 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
db050f29 MW |
111 | void **entry = rcu_dereference_raw(parent->slots[offset]); |
112 | ||
113 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 114 | if (radix_tree_is_internal_node(entry)) { |
8d2c0d36 LT |
115 | if (is_sibling_entry(parent, entry)) { |
116 | void **sibentry = (void **) entry_to_node(entry); | |
117 | offset = get_slot_offset(parent, sibentry); | |
118 | entry = rcu_dereference_raw(*sibentry); | |
db050f29 MW |
119 | } |
120 | } | |
121 | #endif | |
122 | ||
123 | *nodep = (void *)entry; | |
124 | return offset; | |
125 | } | |
126 | ||
35534c86 | 127 | static inline gfp_t root_gfp_mask(const struct radix_tree_root *root) |
612d6c19 NP |
128 | { |
129 | return root->gfp_mask & __GFP_BITS_MASK; | |
130 | } | |
131 | ||
643b52b9 NP |
132 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
133 | int offset) | |
134 | { | |
135 | __set_bit(offset, node->tags[tag]); | |
136 | } | |
137 | ||
138 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
139 | int offset) | |
140 | { | |
141 | __clear_bit(offset, node->tags[tag]); | |
142 | } | |
143 | ||
35534c86 | 144 | static inline int tag_get(const struct radix_tree_node *node, unsigned int tag, |
643b52b9 NP |
145 | int offset) |
146 | { | |
147 | return test_bit(offset, node->tags[tag]); | |
148 | } | |
149 | ||
35534c86 | 150 | static inline void root_tag_set(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
151 | { |
152 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
153 | } | |
154 | ||
2fcd9005 | 155 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
156 | { |
157 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
158 | } | |
159 | ||
160 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
161 | { | |
162 | root->gfp_mask &= __GFP_BITS_MASK; | |
163 | } | |
164 | ||
35534c86 | 165 | static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag) |
643b52b9 | 166 | { |
2fcd9005 | 167 | return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
643b52b9 NP |
168 | } |
169 | ||
35534c86 | 170 | static inline unsigned root_tags_get(const struct radix_tree_root *root) |
7b60e9ad MW |
171 | { |
172 | return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT; | |
173 | } | |
174 | ||
643b52b9 NP |
175 | /* |
176 | * Returns 1 if any slot in the node has this tag set. | |
177 | * Otherwise returns 0. | |
178 | */ | |
35534c86 MW |
179 | static inline int any_tag_set(const struct radix_tree_node *node, |
180 | unsigned int tag) | |
643b52b9 | 181 | { |
2fcd9005 | 182 | unsigned idx; |
643b52b9 NP |
183 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
184 | if (node->tags[tag][idx]) | |
185 | return 1; | |
186 | } | |
187 | return 0; | |
188 | } | |
78c1d784 KK |
189 | |
190 | /** | |
191 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
192 | * | |
193 | * @addr: The address to base the search on | |
194 | * @size: The bitmap size in bits | |
195 | * @offset: The bitnumber to start searching at | |
196 | * | |
197 | * Unrollable variant of find_next_bit() for constant size arrays. | |
198 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
199 | * Returns next bit offset, or size if nothing found. | |
200 | */ | |
201 | static __always_inline unsigned long | |
bc412fca MW |
202 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
203 | unsigned long offset) | |
78c1d784 | 204 | { |
bc412fca | 205 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 206 | |
bc412fca | 207 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
208 | unsigned long tmp; |
209 | ||
210 | addr += offset / BITS_PER_LONG; | |
211 | tmp = *addr >> (offset % BITS_PER_LONG); | |
212 | if (tmp) | |
213 | return __ffs(tmp) + offset; | |
214 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 215 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
216 | tmp = *++addr; |
217 | if (tmp) | |
218 | return __ffs(tmp) + offset; | |
219 | offset += BITS_PER_LONG; | |
220 | } | |
221 | } | |
bc412fca | 222 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
223 | } |
224 | ||
268f42de MW |
225 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
226 | { | |
227 | return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK; | |
228 | } | |
229 | ||
218ed750 MW |
230 | /* |
231 | * The maximum index which can be stored in a radix tree | |
232 | */ | |
233 | static inline unsigned long shift_maxindex(unsigned int shift) | |
234 | { | |
235 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
236 | } | |
237 | ||
35534c86 | 238 | static inline unsigned long node_maxindex(const struct radix_tree_node *node) |
218ed750 MW |
239 | { |
240 | return shift_maxindex(node->shift); | |
241 | } | |
242 | ||
0796c583 | 243 | #ifndef __KERNEL__ |
d0891265 | 244 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 245 | { |
0796c583 | 246 | unsigned long i; |
7cf19af4 | 247 | |
218ed750 MW |
248 | pr_debug("radix node: %p offset %d indices %lu-%lu parent %p tags %lx %lx %lx shift %d count %d exceptional %d\n", |
249 | node, node->offset, index, index | node_maxindex(node), | |
250 | node->parent, | |
0796c583 | 251 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
218ed750 | 252 | node->shift, node->count, node->exceptional); |
0796c583 RZ |
253 | |
254 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
255 | unsigned long first = index | (i << node->shift); |
256 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
257 | void *entry = node->slots[i]; |
258 | if (!entry) | |
259 | continue; | |
218ed750 MW |
260 | if (entry == RADIX_TREE_RETRY) { |
261 | pr_debug("radix retry offset %ld indices %lu-%lu parent %p\n", | |
262 | i, first, last, node); | |
b194d16c | 263 | } else if (!radix_tree_is_internal_node(entry)) { |
218ed750 MW |
264 | pr_debug("radix entry %p offset %ld indices %lu-%lu parent %p\n", |
265 | entry, i, first, last, node); | |
266 | } else if (is_sibling_entry(node, entry)) { | |
267 | pr_debug("radix sblng %p offset %ld indices %lu-%lu parent %p val %p\n", | |
268 | entry, i, first, last, node, | |
269 | *(void **)entry_to_node(entry)); | |
0796c583 | 270 | } else { |
4dd6c098 | 271 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
272 | } |
273 | } | |
7cf19af4 MW |
274 | } |
275 | ||
276 | /* For debug */ | |
277 | static void radix_tree_dump(struct radix_tree_root *root) | |
278 | { | |
d0891265 MW |
279 | pr_debug("radix root: %p rnode %p tags %x\n", |
280 | root, root->rnode, | |
7cf19af4 | 281 | root->gfp_mask >> __GFP_BITS_SHIFT); |
b194d16c | 282 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 283 | return; |
4dd6c098 | 284 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 MW |
285 | } |
286 | #endif | |
287 | ||
1da177e4 LT |
288 | /* |
289 | * This assumes that the caller has performed appropriate preallocation, and | |
290 | * that the caller has pinned this thread of control to the current CPU. | |
291 | */ | |
292 | static struct radix_tree_node * | |
e8de4340 MW |
293 | radix_tree_node_alloc(struct radix_tree_root *root, |
294 | struct radix_tree_node *parent, | |
295 | unsigned int shift, unsigned int offset, | |
296 | unsigned int count, unsigned int exceptional) | |
1da177e4 | 297 | { |
e2848a0e | 298 | struct radix_tree_node *ret = NULL; |
612d6c19 | 299 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 300 | |
5e4c0d97 | 301 | /* |
2fcd9005 MW |
302 | * Preload code isn't irq safe and it doesn't make sense to use |
303 | * preloading during an interrupt anyway as all the allocations have | |
304 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 305 | */ |
d0164adc | 306 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
307 | struct radix_tree_preload *rtp; |
308 | ||
58e698af VD |
309 | /* |
310 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
311 | * cache first for the new node to get accounted to the memory |
312 | * cgroup. | |
58e698af VD |
313 | */ |
314 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 315 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
316 | if (ret) |
317 | goto out; | |
318 | ||
e2848a0e NP |
319 | /* |
320 | * Provided the caller has preloaded here, we will always | |
321 | * succeed in getting a node here (and never reach | |
322 | * kmem_cache_alloc) | |
323 | */ | |
7c8e0181 | 324 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 325 | if (rtp->nr) { |
9d2a8da0 KS |
326 | ret = rtp->nodes; |
327 | rtp->nodes = ret->private_data; | |
328 | ret->private_data = NULL; | |
1da177e4 LT |
329 | rtp->nr--; |
330 | } | |
ce80b067 CM |
331 | /* |
332 | * Update the allocation stack trace as this is more useful | |
333 | * for debugging. | |
334 | */ | |
335 | kmemleak_update_trace(ret); | |
58e698af | 336 | goto out; |
1da177e4 | 337 | } |
05eb6e72 | 338 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 339 | out: |
b194d16c | 340 | BUG_ON(radix_tree_is_internal_node(ret)); |
e8de4340 MW |
341 | if (ret) { |
342 | ret->parent = parent; | |
343 | ret->shift = shift; | |
344 | ret->offset = offset; | |
345 | ret->count = count; | |
346 | ret->exceptional = exceptional; | |
347 | } | |
1da177e4 LT |
348 | return ret; |
349 | } | |
350 | ||
7cf9c2c7 NP |
351 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
352 | { | |
353 | struct radix_tree_node *node = | |
354 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
355 | |
356 | /* | |
175542f5 MW |
357 | * Must only free zeroed nodes into the slab. We can be left with |
358 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
359 | * and tags here. | |
643b52b9 | 360 | */ |
175542f5 MW |
361 | memset(node->slots, 0, sizeof(node->slots)); |
362 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 363 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 364 | |
7cf9c2c7 NP |
365 | kmem_cache_free(radix_tree_node_cachep, node); |
366 | } | |
367 | ||
1da177e4 LT |
368 | static inline void |
369 | radix_tree_node_free(struct radix_tree_node *node) | |
370 | { | |
7cf9c2c7 | 371 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
372 | } |
373 | ||
374 | /* | |
375 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
376 | * ensure that the addition of a single element in the tree cannot fail. On | |
377 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
378 | * with preemption not disabled. | |
b34df792 DH |
379 | * |
380 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 381 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 382 | */ |
2791653a | 383 | static int __radix_tree_preload(gfp_t gfp_mask, unsigned nr) |
1da177e4 LT |
384 | { |
385 | struct radix_tree_preload *rtp; | |
386 | struct radix_tree_node *node; | |
387 | int ret = -ENOMEM; | |
388 | ||
05eb6e72 VD |
389 | /* |
390 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
391 | * they should never be accounted to any particular memory cgroup. | |
392 | */ | |
393 | gfp_mask &= ~__GFP_ACCOUNT; | |
394 | ||
1da177e4 | 395 | preempt_disable(); |
7c8e0181 | 396 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 397 | while (rtp->nr < nr) { |
1da177e4 | 398 | preempt_enable(); |
488514d1 | 399 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
400 | if (node == NULL) |
401 | goto out; | |
402 | preempt_disable(); | |
7c8e0181 | 403 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 404 | if (rtp->nr < nr) { |
9d2a8da0 KS |
405 | node->private_data = rtp->nodes; |
406 | rtp->nodes = node; | |
407 | rtp->nr++; | |
408 | } else { | |
1da177e4 | 409 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 410 | } |
1da177e4 LT |
411 | } |
412 | ret = 0; | |
413 | out: | |
414 | return ret; | |
415 | } | |
5e4c0d97 JK |
416 | |
417 | /* | |
418 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
419 | * ensure that the addition of a single element in the tree cannot fail. On | |
420 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
421 | * with preemption not disabled. | |
422 | * | |
423 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 424 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
425 | */ |
426 | int radix_tree_preload(gfp_t gfp_mask) | |
427 | { | |
428 | /* Warn on non-sensical use... */ | |
d0164adc | 429 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 430 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 431 | } |
d7f0923d | 432 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 433 | |
5e4c0d97 JK |
434 | /* |
435 | * The same as above function, except we don't guarantee preloading happens. | |
436 | * We do it, if we decide it helps. On success, return zero with preemption | |
437 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
438 | */ | |
439 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
440 | { | |
d0164adc | 441 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 442 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
443 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
444 | preempt_disable(); | |
445 | return 0; | |
446 | } | |
447 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
448 | ||
2791653a MW |
449 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
450 | /* | |
451 | * Preload with enough objects to ensure that we can split a single entry | |
452 | * of order @old_order into many entries of size @new_order | |
453 | */ | |
454 | int radix_tree_split_preload(unsigned int old_order, unsigned int new_order, | |
455 | gfp_t gfp_mask) | |
456 | { | |
457 | unsigned top = 1 << (old_order % RADIX_TREE_MAP_SHIFT); | |
458 | unsigned layers = (old_order / RADIX_TREE_MAP_SHIFT) - | |
459 | (new_order / RADIX_TREE_MAP_SHIFT); | |
460 | unsigned nr = 0; | |
461 | ||
462 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); | |
463 | BUG_ON(new_order >= old_order); | |
464 | ||
465 | while (layers--) | |
466 | nr = nr * RADIX_TREE_MAP_SIZE + 1; | |
467 | return __radix_tree_preload(gfp_mask, top * nr); | |
468 | } | |
469 | #endif | |
470 | ||
c78c66d1 KS |
471 | /* |
472 | * The same as function above, but preload number of nodes required to insert | |
473 | * (1 << order) continuous naturally-aligned elements. | |
474 | */ | |
475 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
476 | { | |
477 | unsigned long nr_subtrees; | |
478 | int nr_nodes, subtree_height; | |
479 | ||
480 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
481 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
482 | preempt_disable(); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | /* | |
487 | * Calculate number and height of fully populated subtrees it takes to | |
488 | * store (1 << order) elements. | |
489 | */ | |
490 | nr_subtrees = 1 << order; | |
491 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
492 | subtree_height++) | |
493 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
494 | ||
495 | /* | |
496 | * The worst case is zero height tree with a single item at index 0 and | |
497 | * then inserting items starting at ULONG_MAX - (1 << order). | |
498 | * | |
499 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
500 | * 0-index item. | |
501 | */ | |
502 | nr_nodes = RADIX_TREE_MAX_PATH; | |
503 | ||
504 | /* Plus branch to fully populated subtrees. */ | |
505 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
506 | ||
507 | /* Root node is shared. */ | |
508 | nr_nodes--; | |
509 | ||
510 | /* Plus nodes required to build subtrees. */ | |
511 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
512 | ||
513 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
514 | } | |
515 | ||
35534c86 | 516 | static unsigned radix_tree_load_root(const struct radix_tree_root *root, |
1456a439 MW |
517 | struct radix_tree_node **nodep, unsigned long *maxindex) |
518 | { | |
519 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
520 | ||
521 | *nodep = node; | |
522 | ||
b194d16c | 523 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 524 | node = entry_to_node(node); |
1456a439 | 525 | *maxindex = node_maxindex(node); |
c12e51b0 | 526 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
527 | } |
528 | ||
529 | *maxindex = 0; | |
530 | return 0; | |
531 | } | |
532 | ||
1da177e4 LT |
533 | /* |
534 | * Extend a radix tree so it can store key @index. | |
535 | */ | |
e6145236 | 536 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 537 | unsigned long index, unsigned int shift) |
1da177e4 | 538 | { |
e2bdb933 | 539 | struct radix_tree_node *slot; |
d0891265 | 540 | unsigned int maxshift; |
1da177e4 LT |
541 | int tag; |
542 | ||
d0891265 MW |
543 | /* Figure out what the shift should be. */ |
544 | maxshift = shift; | |
545 | while (index > shift_maxindex(maxshift)) | |
546 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 547 | |
d0891265 MW |
548 | slot = root->rnode; |
549 | if (!slot) | |
1da177e4 | 550 | goto out; |
1da177e4 | 551 | |
1da177e4 | 552 | do { |
e8de4340 MW |
553 | struct radix_tree_node *node = radix_tree_node_alloc(root, |
554 | NULL, shift, 0, 1, 0); | |
2fcd9005 | 555 | if (!node) |
1da177e4 LT |
556 | return -ENOMEM; |
557 | ||
1da177e4 | 558 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 559 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 560 | if (root_tag_get(root, tag)) |
1da177e4 LT |
561 | tag_set(node, tag, 0); |
562 | } | |
563 | ||
d0891265 | 564 | BUG_ON(shift > BITS_PER_LONG); |
f7942430 | 565 | if (radix_tree_is_internal_node(slot)) { |
4dd6c098 | 566 | entry_to_node(slot)->parent = node; |
e8de4340 | 567 | } else if (radix_tree_exceptional_entry(slot)) { |
f7942430 | 568 | /* Moving an exceptional root->rnode to a node */ |
e8de4340 | 569 | node->exceptional = 1; |
f7942430 | 570 | } |
e2bdb933 | 571 | node->slots[0] = slot; |
a4db4dce MW |
572 | slot = node_to_entry(node); |
573 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 574 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 575 | } while (shift <= maxshift); |
1da177e4 | 576 | out: |
d0891265 | 577 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
578 | } |
579 | ||
f4b109c6 JW |
580 | /** |
581 | * radix_tree_shrink - shrink radix tree to minimum height | |
582 | * @root radix tree root | |
583 | */ | |
14b46879 | 584 | static inline void radix_tree_shrink(struct radix_tree_root *root, |
4d693d08 JW |
585 | radix_tree_update_node_t update_node, |
586 | void *private) | |
f4b109c6 | 587 | { |
f4b109c6 JW |
588 | for (;;) { |
589 | struct radix_tree_node *node = root->rnode; | |
590 | struct radix_tree_node *child; | |
591 | ||
592 | if (!radix_tree_is_internal_node(node)) | |
593 | break; | |
594 | node = entry_to_node(node); | |
595 | ||
596 | /* | |
597 | * The candidate node has more than one child, or its child | |
598 | * is not at the leftmost slot, or the child is a multiorder | |
599 | * entry, we cannot shrink. | |
600 | */ | |
601 | if (node->count != 1) | |
602 | break; | |
603 | child = node->slots[0]; | |
604 | if (!child) | |
605 | break; | |
606 | if (!radix_tree_is_internal_node(child) && node->shift) | |
607 | break; | |
608 | ||
609 | if (radix_tree_is_internal_node(child)) | |
610 | entry_to_node(child)->parent = NULL; | |
611 | ||
612 | /* | |
613 | * We don't need rcu_assign_pointer(), since we are simply | |
614 | * moving the node from one part of the tree to another: if it | |
615 | * was safe to dereference the old pointer to it | |
616 | * (node->slots[0]), it will be safe to dereference the new | |
617 | * one (root->rnode) as far as dependent read barriers go. | |
618 | */ | |
619 | root->rnode = child; | |
620 | ||
621 | /* | |
622 | * We have a dilemma here. The node's slot[0] must not be | |
623 | * NULLed in case there are concurrent lookups expecting to | |
624 | * find the item. However if this was a bottom-level node, | |
625 | * then it may be subject to the slot pointer being visible | |
626 | * to callers dereferencing it. If item corresponding to | |
627 | * slot[0] is subsequently deleted, these callers would expect | |
628 | * their slot to become empty sooner or later. | |
629 | * | |
630 | * For example, lockless pagecache will look up a slot, deref | |
631 | * the page pointer, and if the page has 0 refcount it means it | |
632 | * was concurrently deleted from pagecache so try the deref | |
633 | * again. Fortunately there is already a requirement for logic | |
634 | * to retry the entire slot lookup -- the indirect pointer | |
635 | * problem (replacing direct root node with an indirect pointer | |
636 | * also results in a stale slot). So tag the slot as indirect | |
637 | * to force callers to retry. | |
638 | */ | |
4d693d08 JW |
639 | node->count = 0; |
640 | if (!radix_tree_is_internal_node(child)) { | |
f4b109c6 | 641 | node->slots[0] = RADIX_TREE_RETRY; |
4d693d08 JW |
642 | if (update_node) |
643 | update_node(node, private); | |
644 | } | |
f4b109c6 | 645 | |
ea07b862 | 646 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 647 | radix_tree_node_free(node); |
f4b109c6 | 648 | } |
f4b109c6 JW |
649 | } |
650 | ||
14b46879 | 651 | static void delete_node(struct radix_tree_root *root, |
4d693d08 JW |
652 | struct radix_tree_node *node, |
653 | radix_tree_update_node_t update_node, void *private) | |
f4b109c6 | 654 | { |
f4b109c6 JW |
655 | do { |
656 | struct radix_tree_node *parent; | |
657 | ||
658 | if (node->count) { | |
659 | if (node == entry_to_node(root->rnode)) | |
14b46879 JW |
660 | radix_tree_shrink(root, update_node, private); |
661 | return; | |
f4b109c6 JW |
662 | } |
663 | ||
664 | parent = node->parent; | |
665 | if (parent) { | |
666 | parent->slots[node->offset] = NULL; | |
667 | parent->count--; | |
668 | } else { | |
669 | root_tag_clear_all(root); | |
670 | root->rnode = NULL; | |
671 | } | |
672 | ||
ea07b862 | 673 | WARN_ON_ONCE(!list_empty(&node->private_list)); |
f4b109c6 | 674 | radix_tree_node_free(node); |
f4b109c6 JW |
675 | |
676 | node = parent; | |
677 | } while (node); | |
f4b109c6 JW |
678 | } |
679 | ||
1da177e4 | 680 | /** |
139e5616 | 681 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
682 | * @root: radix tree root |
683 | * @index: index key | |
e6145236 | 684 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
685 | * @nodep: returns node |
686 | * @slotp: returns slot | |
1da177e4 | 687 | * |
139e5616 JW |
688 | * Create, if necessary, and return the node and slot for an item |
689 | * at position @index in the radix tree @root. | |
690 | * | |
691 | * Until there is more than one item in the tree, no nodes are | |
692 | * allocated and @root->rnode is used as a direct slot instead of | |
693 | * pointing to a node, in which case *@nodep will be NULL. | |
694 | * | |
695 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 696 | */ |
139e5616 | 697 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
698 | unsigned order, struct radix_tree_node **nodep, |
699 | void ***slotp) | |
1da177e4 | 700 | { |
89148aa4 MW |
701 | struct radix_tree_node *node = NULL, *child; |
702 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 703 | unsigned long maxindex; |
89148aa4 | 704 | unsigned int shift, offset = 0; |
49ea6ebc MW |
705 | unsigned long max = index | ((1UL << order) - 1); |
706 | ||
89148aa4 | 707 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
708 | |
709 | /* Make sure the tree is high enough. */ | |
175542f5 MW |
710 | if (order > 0 && max == ((1UL << order) - 1)) |
711 | max++; | |
49ea6ebc | 712 | if (max > maxindex) { |
d0891265 | 713 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 714 | if (error < 0) |
1da177e4 | 715 | return error; |
49ea6ebc | 716 | shift = error; |
89148aa4 | 717 | child = root->rnode; |
1da177e4 LT |
718 | } |
719 | ||
e6145236 | 720 | while (shift > order) { |
c12e51b0 | 721 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 722 | if (child == NULL) { |
1da177e4 | 723 | /* Have to add a child node. */ |
e8de4340 MW |
724 | child = radix_tree_node_alloc(root, node, shift, |
725 | offset, 0, 0); | |
89148aa4 | 726 | if (!child) |
1da177e4 | 727 | return -ENOMEM; |
89148aa4 MW |
728 | rcu_assign_pointer(*slot, node_to_entry(child)); |
729 | if (node) | |
1da177e4 | 730 | node->count++; |
89148aa4 | 731 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 732 | break; |
1da177e4 LT |
733 | |
734 | /* Go a level down */ | |
89148aa4 | 735 | node = entry_to_node(child); |
9e85d811 | 736 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 737 | slot = &node->slots[offset]; |
e6145236 MW |
738 | } |
739 | ||
175542f5 MW |
740 | if (nodep) |
741 | *nodep = node; | |
742 | if (slotp) | |
743 | *slotp = slot; | |
744 | return 0; | |
745 | } | |
746 | ||
57578c2e | 747 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
175542f5 MW |
748 | /* |
749 | * Free any nodes below this node. The tree is presumed to not need | |
750 | * shrinking, and any user data in the tree is presumed to not need a | |
751 | * destructor called on it. If we need to add a destructor, we can | |
752 | * add that functionality later. Note that we may not clear tags or | |
753 | * slots from the tree as an RCU walker may still have a pointer into | |
754 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
755 | * but we'll still have to clear those in rcu_free. | |
756 | */ | |
757 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
758 | { | |
759 | unsigned offset = 0; | |
760 | struct radix_tree_node *child = entry_to_node(node); | |
761 | ||
762 | for (;;) { | |
763 | void *entry = child->slots[offset]; | |
764 | if (radix_tree_is_internal_node(entry) && | |
765 | !is_sibling_entry(child, entry)) { | |
766 | child = entry_to_node(entry); | |
767 | offset = 0; | |
768 | continue; | |
769 | } | |
770 | offset++; | |
771 | while (offset == RADIX_TREE_MAP_SIZE) { | |
772 | struct radix_tree_node *old = child; | |
773 | offset = child->offset + 1; | |
774 | child = child->parent; | |
dd040b6f | 775 | WARN_ON_ONCE(!list_empty(&old->private_list)); |
175542f5 MW |
776 | radix_tree_node_free(old); |
777 | if (old == entry_to_node(node)) | |
778 | return; | |
779 | } | |
780 | } | |
781 | } | |
782 | ||
783 | static inline int insert_entries(struct radix_tree_node *node, void **slot, | |
784 | void *item, unsigned order, bool replace) | |
785 | { | |
786 | struct radix_tree_node *child; | |
787 | unsigned i, n, tag, offset, tags = 0; | |
788 | ||
789 | if (node) { | |
e157b555 MW |
790 | if (order > node->shift) |
791 | n = 1 << (order - node->shift); | |
792 | else | |
793 | n = 1; | |
175542f5 MW |
794 | offset = get_slot_offset(node, slot); |
795 | } else { | |
796 | n = 1; | |
797 | offset = 0; | |
798 | } | |
799 | ||
800 | if (n > 1) { | |
e6145236 | 801 | offset = offset & ~(n - 1); |
89148aa4 | 802 | slot = &node->slots[offset]; |
175542f5 MW |
803 | } |
804 | child = node_to_entry(slot); | |
805 | ||
806 | for (i = 0; i < n; i++) { | |
807 | if (slot[i]) { | |
808 | if (replace) { | |
809 | node->count--; | |
810 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
811 | if (tag_get(node, tag, offset + i)) | |
812 | tags |= 1 << tag; | |
813 | } else | |
e6145236 MW |
814 | return -EEXIST; |
815 | } | |
175542f5 | 816 | } |
e6145236 | 817 | |
175542f5 MW |
818 | for (i = 0; i < n; i++) { |
819 | struct radix_tree_node *old = slot[i]; | |
820 | if (i) { | |
89148aa4 | 821 | rcu_assign_pointer(slot[i], child); |
175542f5 MW |
822 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
823 | if (tags & (1 << tag)) | |
824 | tag_clear(node, tag, offset + i); | |
825 | } else { | |
826 | rcu_assign_pointer(slot[i], item); | |
827 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
828 | if (tags & (1 << tag)) | |
829 | tag_set(node, tag, offset); | |
e6145236 | 830 | } |
175542f5 | 831 | if (radix_tree_is_internal_node(old) && |
e157b555 MW |
832 | !is_sibling_entry(node, old) && |
833 | (old != RADIX_TREE_RETRY)) | |
175542f5 MW |
834 | radix_tree_free_nodes(old); |
835 | if (radix_tree_exceptional_entry(old)) | |
836 | node->exceptional--; | |
612d6c19 | 837 | } |
175542f5 MW |
838 | if (node) { |
839 | node->count += n; | |
840 | if (radix_tree_exceptional_entry(item)) | |
841 | node->exceptional += n; | |
842 | } | |
843 | return n; | |
139e5616 | 844 | } |
175542f5 MW |
845 | #else |
846 | static inline int insert_entries(struct radix_tree_node *node, void **slot, | |
847 | void *item, unsigned order, bool replace) | |
848 | { | |
849 | if (*slot) | |
850 | return -EEXIST; | |
851 | rcu_assign_pointer(*slot, item); | |
852 | if (node) { | |
853 | node->count++; | |
854 | if (radix_tree_exceptional_entry(item)) | |
855 | node->exceptional++; | |
856 | } | |
857 | return 1; | |
858 | } | |
859 | #endif | |
139e5616 JW |
860 | |
861 | /** | |
e6145236 | 862 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
863 | * @root: radix tree root |
864 | * @index: index key | |
e6145236 | 865 | * @order: key covers the 2^order indices around index |
139e5616 JW |
866 | * @item: item to insert |
867 | * | |
868 | * Insert an item into the radix tree at position @index. | |
869 | */ | |
e6145236 MW |
870 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
871 | unsigned order, void *item) | |
139e5616 JW |
872 | { |
873 | struct radix_tree_node *node; | |
874 | void **slot; | |
875 | int error; | |
876 | ||
b194d16c | 877 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 878 | |
e6145236 | 879 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
880 | if (error) |
881 | return error; | |
175542f5 MW |
882 | |
883 | error = insert_entries(node, slot, item, order, false); | |
884 | if (error < 0) | |
885 | return error; | |
201b6264 | 886 | |
612d6c19 | 887 | if (node) { |
7b60e9ad | 888 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
889 | BUG_ON(tag_get(node, 0, offset)); |
890 | BUG_ON(tag_get(node, 1, offset)); | |
891 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 892 | } else { |
7b60e9ad | 893 | BUG_ON(root_tags_get(root)); |
612d6c19 | 894 | } |
1da177e4 | 895 | |
1da177e4 LT |
896 | return 0; |
897 | } | |
e6145236 | 898 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 899 | |
139e5616 JW |
900 | /** |
901 | * __radix_tree_lookup - lookup an item in a radix tree | |
902 | * @root: radix tree root | |
903 | * @index: index key | |
904 | * @nodep: returns node | |
905 | * @slotp: returns slot | |
906 | * | |
907 | * Lookup and return the item at position @index in the radix | |
908 | * tree @root. | |
909 | * | |
910 | * Until there is more than one item in the tree, no nodes are | |
911 | * allocated and @root->rnode is used as a direct slot instead of | |
912 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 913 | */ |
35534c86 MW |
914 | void *__radix_tree_lookup(const struct radix_tree_root *root, |
915 | unsigned long index, struct radix_tree_node **nodep, | |
916 | void ***slotp) | |
1da177e4 | 917 | { |
139e5616 | 918 | struct radix_tree_node *node, *parent; |
85829954 | 919 | unsigned long maxindex; |
139e5616 | 920 | void **slot; |
612d6c19 | 921 | |
85829954 MW |
922 | restart: |
923 | parent = NULL; | |
924 | slot = (void **)&root->rnode; | |
9e85d811 | 925 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 926 | if (index > maxindex) |
1da177e4 LT |
927 | return NULL; |
928 | ||
b194d16c | 929 | while (radix_tree_is_internal_node(node)) { |
85829954 | 930 | unsigned offset; |
1da177e4 | 931 | |
85829954 MW |
932 | if (node == RADIX_TREE_RETRY) |
933 | goto restart; | |
4dd6c098 | 934 | parent = entry_to_node(node); |
9e85d811 | 935 | offset = radix_tree_descend(parent, &node, index); |
85829954 MW |
936 | slot = parent->slots + offset; |
937 | } | |
1da177e4 | 938 | |
139e5616 JW |
939 | if (nodep) |
940 | *nodep = parent; | |
941 | if (slotp) | |
942 | *slotp = slot; | |
943 | return node; | |
b72b71c6 HS |
944 | } |
945 | ||
946 | /** | |
947 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
948 | * @root: radix tree root | |
949 | * @index: index key | |
950 | * | |
951 | * Returns: the slot corresponding to the position @index in the | |
952 | * radix tree @root. This is useful for update-if-exists operations. | |
953 | * | |
954 | * This function can be called under rcu_read_lock iff the slot is not | |
955 | * modified by radix_tree_replace_slot, otherwise it must be called | |
956 | * exclusive from other writers. Any dereference of the slot must be done | |
957 | * using radix_tree_deref_slot. | |
958 | */ | |
35534c86 MW |
959 | void **radix_tree_lookup_slot(const struct radix_tree_root *root, |
960 | unsigned long index) | |
b72b71c6 | 961 | { |
139e5616 JW |
962 | void **slot; |
963 | ||
964 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
965 | return NULL; | |
966 | return slot; | |
a4331366 | 967 | } |
a4331366 HR |
968 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
969 | ||
970 | /** | |
971 | * radix_tree_lookup - perform lookup operation on a radix tree | |
972 | * @root: radix tree root | |
973 | * @index: index key | |
974 | * | |
975 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
976 | * |
977 | * This function can be called under rcu_read_lock, however the caller | |
978 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
979 | * them safely). No RCU barriers are required to access or modify the | |
980 | * returned item, however. | |
a4331366 | 981 | */ |
35534c86 | 982 | void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index) |
a4331366 | 983 | { |
139e5616 | 984 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
985 | } |
986 | EXPORT_SYMBOL(radix_tree_lookup); | |
987 | ||
a90eb3a2 MW |
988 | static inline int slot_count(struct radix_tree_node *node, |
989 | void **slot) | |
990 | { | |
991 | int n = 1; | |
992 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
993 | void *ptr = node_to_entry(slot); | |
994 | unsigned offset = get_slot_offset(node, slot); | |
995 | int i; | |
996 | ||
997 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
998 | if (node->slots[offset + i] != ptr) | |
999 | break; | |
1000 | n++; | |
1001 | } | |
1002 | #endif | |
1003 | return n; | |
1004 | } | |
1005 | ||
6d75f366 JW |
1006 | static void replace_slot(struct radix_tree_root *root, |
1007 | struct radix_tree_node *node, | |
1008 | void **slot, void *item, | |
1009 | bool warn_typeswitch) | |
f7942430 JW |
1010 | { |
1011 | void *old = rcu_dereference_raw(*slot); | |
f4b109c6 | 1012 | int count, exceptional; |
f7942430 JW |
1013 | |
1014 | WARN_ON_ONCE(radix_tree_is_internal_node(item)); | |
f7942430 | 1015 | |
f4b109c6 | 1016 | count = !!item - !!old; |
f7942430 JW |
1017 | exceptional = !!radix_tree_exceptional_entry(item) - |
1018 | !!radix_tree_exceptional_entry(old); | |
1019 | ||
f4b109c6 | 1020 | WARN_ON_ONCE(warn_typeswitch && (count || exceptional)); |
f7942430 | 1021 | |
f4b109c6 JW |
1022 | if (node) { |
1023 | node->count += count; | |
a90eb3a2 MW |
1024 | if (exceptional) { |
1025 | exceptional *= slot_count(node, slot); | |
1026 | node->exceptional += exceptional; | |
1027 | } | |
f4b109c6 | 1028 | } |
f7942430 JW |
1029 | |
1030 | rcu_assign_pointer(*slot, item); | |
1031 | } | |
1032 | ||
a90eb3a2 MW |
1033 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1034 | void **slot) | |
1035 | { | |
1036 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1037 | bool exceptional = radix_tree_exceptional_entry(*slot); | |
1038 | void *ptr = node_to_entry(slot); | |
1039 | unsigned offset = get_slot_offset(node, slot); | |
1040 | int i; | |
1041 | ||
1042 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1043 | if (node->slots[offset + i] != ptr) | |
1044 | break; | |
1045 | node->slots[offset + i] = NULL; | |
1046 | node->count--; | |
1047 | if (exceptional) | |
1048 | node->exceptional--; | |
1049 | } | |
1050 | #endif | |
1051 | } | |
1052 | ||
6d75f366 JW |
1053 | /** |
1054 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
1055 | * @root: radix tree root |
1056 | * @node: pointer to tree node | |
1057 | * @slot: pointer to slot in @node | |
1058 | * @item: new item to store in the slot. | |
1059 | * @update_node: callback for changing leaf nodes | |
1060 | * @private: private data to pass to @update_node | |
6d75f366 JW |
1061 | * |
1062 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
1063 | * across slot lookup and replacement. | |
1064 | */ | |
1065 | void __radix_tree_replace(struct radix_tree_root *root, | |
1066 | struct radix_tree_node *node, | |
4d693d08 JW |
1067 | void **slot, void *item, |
1068 | radix_tree_update_node_t update_node, void *private) | |
6d75f366 | 1069 | { |
a90eb3a2 MW |
1070 | if (!item) |
1071 | delete_sibling_entries(node, slot); | |
6d75f366 | 1072 | /* |
f4b109c6 JW |
1073 | * This function supports replacing exceptional entries and |
1074 | * deleting entries, but that needs accounting against the | |
1075 | * node unless the slot is root->rnode. | |
6d75f366 JW |
1076 | */ |
1077 | replace_slot(root, node, slot, item, | |
1078 | !node && slot != (void **)&root->rnode); | |
f4b109c6 | 1079 | |
4d693d08 JW |
1080 | if (!node) |
1081 | return; | |
1082 | ||
1083 | if (update_node) | |
1084 | update_node(node, private); | |
1085 | ||
1086 | delete_node(root, node, update_node, private); | |
6d75f366 JW |
1087 | } |
1088 | ||
1089 | /** | |
1090 | * radix_tree_replace_slot - replace item in a slot | |
1091 | * @root: radix tree root | |
1092 | * @slot: pointer to slot | |
1093 | * @item: new item to store in the slot. | |
1094 | * | |
1095 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
1096 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
1097 | * across slot lookup and replacement. | |
1098 | * | |
1099 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
1100 | * regular entries, and exceptional entries, as that requires accounting | |
f4b109c6 | 1101 | * inside the radix tree node. When switching from one type of entry or |
e157b555 MW |
1102 | * deleting, use __radix_tree_lookup() and __radix_tree_replace() or |
1103 | * radix_tree_iter_replace(). | |
6d75f366 JW |
1104 | */ |
1105 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
1106 | void **slot, void *item) | |
1107 | { | |
1108 | replace_slot(root, NULL, slot, item, true); | |
1109 | } | |
1110 | ||
e157b555 MW |
1111 | /** |
1112 | * radix_tree_iter_replace - replace item in a slot | |
1113 | * @root: radix tree root | |
1114 | * @slot: pointer to slot | |
1115 | * @item: new item to store in the slot. | |
1116 | * | |
1117 | * For use with radix_tree_split() and radix_tree_for_each_slot(). | |
1118 | * Caller must hold tree write locked across split and replacement. | |
1119 | */ | |
1120 | void radix_tree_iter_replace(struct radix_tree_root *root, | |
1121 | const struct radix_tree_iter *iter, void **slot, void *item) | |
1122 | { | |
1123 | __radix_tree_replace(root, iter->node, slot, item, NULL, NULL); | |
1124 | } | |
1125 | ||
175542f5 MW |
1126 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
1127 | /** | |
1128 | * radix_tree_join - replace multiple entries with one multiorder entry | |
1129 | * @root: radix tree root | |
1130 | * @index: an index inside the new entry | |
1131 | * @order: order of the new entry | |
1132 | * @item: new entry | |
1133 | * | |
1134 | * Call this function to replace several entries with one larger entry. | |
1135 | * The existing entries are presumed to not need freeing as a result of | |
1136 | * this call. | |
1137 | * | |
1138 | * The replacement entry will have all the tags set on it that were set | |
1139 | * on any of the entries it is replacing. | |
1140 | */ | |
1141 | int radix_tree_join(struct radix_tree_root *root, unsigned long index, | |
1142 | unsigned order, void *item) | |
1143 | { | |
1144 | struct radix_tree_node *node; | |
1145 | void **slot; | |
1146 | int error; | |
1147 | ||
1148 | BUG_ON(radix_tree_is_internal_node(item)); | |
1149 | ||
1150 | error = __radix_tree_create(root, index, order, &node, &slot); | |
1151 | if (!error) | |
1152 | error = insert_entries(node, slot, item, order, true); | |
1153 | if (error > 0) | |
1154 | error = 0; | |
1155 | ||
1156 | return error; | |
1157 | } | |
e157b555 MW |
1158 | |
1159 | /** | |
1160 | * radix_tree_split - Split an entry into smaller entries | |
1161 | * @root: radix tree root | |
1162 | * @index: An index within the large entry | |
1163 | * @order: Order of new entries | |
1164 | * | |
1165 | * Call this function as the first step in replacing a multiorder entry | |
1166 | * with several entries of lower order. After this function returns, | |
1167 | * loop over the relevant portion of the tree using radix_tree_for_each_slot() | |
1168 | * and call radix_tree_iter_replace() to set up each new entry. | |
1169 | * | |
1170 | * The tags from this entry are replicated to all the new entries. | |
1171 | * | |
1172 | * The radix tree should be locked against modification during the entire | |
1173 | * replacement operation. Lock-free lookups will see RADIX_TREE_RETRY which | |
1174 | * should prompt RCU walkers to restart the lookup from the root. | |
1175 | */ | |
1176 | int radix_tree_split(struct radix_tree_root *root, unsigned long index, | |
1177 | unsigned order) | |
1178 | { | |
1179 | struct radix_tree_node *parent, *node, *child; | |
1180 | void **slot; | |
1181 | unsigned int offset, end; | |
1182 | unsigned n, tag, tags = 0; | |
1183 | ||
1184 | if (!__radix_tree_lookup(root, index, &parent, &slot)) | |
1185 | return -ENOENT; | |
1186 | if (!parent) | |
1187 | return -ENOENT; | |
1188 | ||
1189 | offset = get_slot_offset(parent, slot); | |
1190 | ||
1191 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1192 | if (tag_get(parent, tag, offset)) | |
1193 | tags |= 1 << tag; | |
1194 | ||
1195 | for (end = offset + 1; end < RADIX_TREE_MAP_SIZE; end++) { | |
1196 | if (!is_sibling_entry(parent, parent->slots[end])) | |
1197 | break; | |
1198 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1199 | if (tags & (1 << tag)) | |
1200 | tag_set(parent, tag, end); | |
1201 | /* rcu_assign_pointer ensures tags are set before RETRY */ | |
1202 | rcu_assign_pointer(parent->slots[end], RADIX_TREE_RETRY); | |
1203 | } | |
1204 | rcu_assign_pointer(parent->slots[offset], RADIX_TREE_RETRY); | |
1205 | parent->exceptional -= (end - offset); | |
1206 | ||
1207 | if (order == parent->shift) | |
1208 | return 0; | |
1209 | if (order > parent->shift) { | |
1210 | while (offset < end) | |
1211 | offset += insert_entries(parent, &parent->slots[offset], | |
1212 | RADIX_TREE_RETRY, order, true); | |
1213 | return 0; | |
1214 | } | |
1215 | ||
1216 | node = parent; | |
1217 | ||
1218 | for (;;) { | |
1219 | if (node->shift > order) { | |
e8de4340 MW |
1220 | child = radix_tree_node_alloc(root, node, |
1221 | node->shift - RADIX_TREE_MAP_SHIFT, | |
1222 | offset, 0, 0); | |
e157b555 MW |
1223 | if (!child) |
1224 | goto nomem; | |
e157b555 MW |
1225 | if (node != parent) { |
1226 | node->count++; | |
1227 | node->slots[offset] = node_to_entry(child); | |
1228 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1229 | if (tags & (1 << tag)) | |
1230 | tag_set(node, tag, offset); | |
1231 | } | |
1232 | ||
1233 | node = child; | |
1234 | offset = 0; | |
1235 | continue; | |
1236 | } | |
1237 | ||
1238 | n = insert_entries(node, &node->slots[offset], | |
1239 | RADIX_TREE_RETRY, order, false); | |
1240 | BUG_ON(n > RADIX_TREE_MAP_SIZE); | |
1241 | ||
1242 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1243 | if (tags & (1 << tag)) | |
1244 | tag_set(node, tag, offset); | |
1245 | offset += n; | |
1246 | ||
1247 | while (offset == RADIX_TREE_MAP_SIZE) { | |
1248 | if (node == parent) | |
1249 | break; | |
1250 | offset = node->offset; | |
1251 | child = node; | |
1252 | node = node->parent; | |
1253 | rcu_assign_pointer(node->slots[offset], | |
1254 | node_to_entry(child)); | |
1255 | offset++; | |
1256 | } | |
1257 | if ((node == parent) && (offset == end)) | |
1258 | return 0; | |
1259 | } | |
1260 | ||
1261 | nomem: | |
1262 | /* Shouldn't happen; did user forget to preload? */ | |
1263 | /* TODO: free all the allocated nodes */ | |
1264 | WARN_ON(1); | |
1265 | return -ENOMEM; | |
1266 | } | |
175542f5 MW |
1267 | #endif |
1268 | ||
30b888ba MW |
1269 | static void node_tag_set(struct radix_tree_root *root, |
1270 | struct radix_tree_node *node, | |
1271 | unsigned int tag, unsigned int offset) | |
1272 | { | |
1273 | while (node) { | |
1274 | if (tag_get(node, tag, offset)) | |
1275 | return; | |
1276 | tag_set(node, tag, offset); | |
1277 | offset = node->offset; | |
1278 | node = node->parent; | |
1279 | } | |
1280 | ||
1281 | if (!root_tag_get(root, tag)) | |
1282 | root_tag_set(root, tag); | |
1283 | } | |
1284 | ||
1da177e4 LT |
1285 | /** |
1286 | * radix_tree_tag_set - set a tag on a radix tree node | |
1287 | * @root: radix tree root | |
1288 | * @index: index key | |
2fcd9005 | 1289 | * @tag: tag index |
1da177e4 | 1290 | * |
daff89f3 JC |
1291 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
1292 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
1293 | * the root all the way down to the leaf node. |
1294 | * | |
2fcd9005 | 1295 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
1296 | * item is a bug. |
1297 | */ | |
1298 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 1299 | unsigned long index, unsigned int tag) |
1da177e4 | 1300 | { |
fb969909 RZ |
1301 | struct radix_tree_node *node, *parent; |
1302 | unsigned long maxindex; | |
1da177e4 | 1303 | |
9e85d811 | 1304 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 1305 | BUG_ON(index > maxindex); |
1da177e4 | 1306 | |
b194d16c | 1307 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 1308 | unsigned offset; |
1da177e4 | 1309 | |
4dd6c098 | 1310 | parent = entry_to_node(node); |
9e85d811 | 1311 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
1312 | BUG_ON(!node); |
1313 | ||
1314 | if (!tag_get(parent, tag, offset)) | |
1315 | tag_set(parent, tag, offset); | |
1da177e4 LT |
1316 | } |
1317 | ||
612d6c19 | 1318 | /* set the root's tag bit */ |
fb969909 | 1319 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
1320 | root_tag_set(root, tag); |
1321 | ||
fb969909 | 1322 | return node; |
1da177e4 LT |
1323 | } |
1324 | EXPORT_SYMBOL(radix_tree_tag_set); | |
1325 | ||
30b888ba MW |
1326 | /** |
1327 | * radix_tree_iter_tag_set - set a tag on the current iterator entry | |
1328 | * @root: radix tree root | |
1329 | * @iter: iterator state | |
1330 | * @tag: tag to set | |
1331 | */ | |
1332 | void radix_tree_iter_tag_set(struct radix_tree_root *root, | |
1333 | const struct radix_tree_iter *iter, unsigned int tag) | |
1334 | { | |
1335 | node_tag_set(root, iter->node, tag, iter_offset(iter)); | |
1336 | } | |
1337 | ||
d604c324 MW |
1338 | static void node_tag_clear(struct radix_tree_root *root, |
1339 | struct radix_tree_node *node, | |
1340 | unsigned int tag, unsigned int offset) | |
1341 | { | |
1342 | while (node) { | |
1343 | if (!tag_get(node, tag, offset)) | |
1344 | return; | |
1345 | tag_clear(node, tag, offset); | |
1346 | if (any_tag_set(node, tag)) | |
1347 | return; | |
1348 | ||
1349 | offset = node->offset; | |
1350 | node = node->parent; | |
1351 | } | |
1352 | ||
1353 | /* clear the root's tag bit */ | |
1354 | if (root_tag_get(root, tag)) | |
1355 | root_tag_clear(root, tag); | |
1356 | } | |
1357 | ||
1da177e4 LT |
1358 | /** |
1359 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1360 | * @root: radix tree root | |
1361 | * @index: index key | |
2fcd9005 | 1362 | * @tag: tag index |
1da177e4 | 1363 | * |
daff89f3 | 1364 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1365 | * corresponding to @index in the radix tree. If this causes |
1366 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1367 | * next-to-leaf node, etc. |
1368 | * | |
1369 | * Returns the address of the tagged item on success, else NULL. ie: | |
1370 | * has the same return value and semantics as radix_tree_lookup(). | |
1371 | */ | |
1372 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1373 | unsigned long index, unsigned int tag) |
1da177e4 | 1374 | { |
00f47b58 RZ |
1375 | struct radix_tree_node *node, *parent; |
1376 | unsigned long maxindex; | |
e2bdb933 | 1377 | int uninitialized_var(offset); |
1da177e4 | 1378 | |
9e85d811 | 1379 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1380 | if (index > maxindex) |
1381 | return NULL; | |
1da177e4 | 1382 | |
00f47b58 | 1383 | parent = NULL; |
1da177e4 | 1384 | |
b194d16c | 1385 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1386 | parent = entry_to_node(node); |
9e85d811 | 1387 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1388 | } |
1389 | ||
d604c324 MW |
1390 | if (node) |
1391 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1392 | |
00f47b58 | 1393 | return node; |
1da177e4 LT |
1394 | } |
1395 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1396 | ||
30b888ba MW |
1397 | /** |
1398 | * radix_tree_iter_tag_clear - clear a tag on the current iterator entry | |
1399 | * @root: radix tree root | |
1400 | * @iter: iterator state | |
1401 | * @tag: tag to clear | |
1402 | */ | |
1403 | void radix_tree_iter_tag_clear(struct radix_tree_root *root, | |
1404 | const struct radix_tree_iter *iter, unsigned int tag) | |
1405 | { | |
1406 | node_tag_clear(root, iter->node, tag, iter_offset(iter)); | |
1407 | } | |
1408 | ||
1da177e4 | 1409 | /** |
32605a18 MT |
1410 | * radix_tree_tag_get - get a tag on a radix tree node |
1411 | * @root: radix tree root | |
1412 | * @index: index key | |
2fcd9005 | 1413 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1414 | * |
32605a18 | 1415 | * Return values: |
1da177e4 | 1416 | * |
612d6c19 NP |
1417 | * 0: tag not present or not set |
1418 | * 1: tag set | |
ce82653d DH |
1419 | * |
1420 | * Note that the return value of this function may not be relied on, even if | |
1421 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1422 | * from concurrency. | |
1da177e4 | 1423 | */ |
35534c86 | 1424 | int radix_tree_tag_get(const struct radix_tree_root *root, |
daff89f3 | 1425 | unsigned long index, unsigned int tag) |
1da177e4 | 1426 | { |
4589ba6d RZ |
1427 | struct radix_tree_node *node, *parent; |
1428 | unsigned long maxindex; | |
1da177e4 | 1429 | |
612d6c19 NP |
1430 | if (!root_tag_get(root, tag)) |
1431 | return 0; | |
1432 | ||
9e85d811 | 1433 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1434 | if (index > maxindex) |
1435 | return 0; | |
7cf9c2c7 NP |
1436 | if (node == NULL) |
1437 | return 0; | |
1438 | ||
b194d16c | 1439 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1440 | unsigned offset; |
1da177e4 | 1441 | |
4dd6c098 | 1442 | parent = entry_to_node(node); |
9e85d811 | 1443 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1444 | |
4589ba6d | 1445 | if (!node) |
1da177e4 | 1446 | return 0; |
4589ba6d | 1447 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1448 | return 0; |
4589ba6d RZ |
1449 | if (node == RADIX_TREE_RETRY) |
1450 | break; | |
1da177e4 | 1451 | } |
4589ba6d RZ |
1452 | |
1453 | return 1; | |
1da177e4 LT |
1454 | } |
1455 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1456 | |
21ef5339 RZ |
1457 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1458 | unsigned int shift) | |
1459 | { | |
1460 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1461 | iter->shift = shift; | |
1462 | #endif | |
1463 | } | |
1464 | ||
148deab2 MW |
1465 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1466 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1467 | struct radix_tree_node *node, unsigned offset, | |
1468 | unsigned tag) | |
1469 | { | |
1470 | unsigned tag_long = offset / BITS_PER_LONG; | |
1471 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1472 | ||
1473 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1474 | ||
1475 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1476 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1477 | /* Pick tags from next element */ | |
1478 | if (tag_bit) | |
1479 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1480 | (BITS_PER_LONG - tag_bit); | |
1481 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1482 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1483 | } | |
1484 | } | |
1485 | ||
1486 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1487 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1488 | void **slot, struct radix_tree_iter *iter) | |
1489 | { | |
1490 | void *sib = node_to_entry(slot - 1); | |
1491 | ||
1492 | while (iter->index < iter->next_index) { | |
1493 | *nodep = rcu_dereference_raw(*slot); | |
1494 | if (*nodep && *nodep != sib) | |
1495 | return slot; | |
1496 | slot++; | |
1497 | iter->index = __radix_tree_iter_add(iter, 1); | |
1498 | iter->tags >>= 1; | |
1499 | } | |
1500 | ||
1501 | *nodep = NULL; | |
1502 | return NULL; | |
1503 | } | |
1504 | ||
1505 | void ** __radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, | |
1506 | unsigned flags) | |
1507 | { | |
1508 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
1509 | struct radix_tree_node *node = rcu_dereference_raw(*slot); | |
1510 | ||
1511 | slot = skip_siblings(&node, slot, iter); | |
1512 | ||
1513 | while (radix_tree_is_internal_node(node)) { | |
1514 | unsigned offset; | |
1515 | unsigned long next_index; | |
1516 | ||
1517 | if (node == RADIX_TREE_RETRY) | |
1518 | return slot; | |
1519 | node = entry_to_node(node); | |
268f42de | 1520 | iter->node = node; |
148deab2 MW |
1521 | iter->shift = node->shift; |
1522 | ||
1523 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1524 | offset = radix_tree_find_next_bit(node, tag, 0); | |
1525 | if (offset == RADIX_TREE_MAP_SIZE) | |
1526 | return NULL; | |
1527 | slot = &node->slots[offset]; | |
1528 | iter->index = __radix_tree_iter_add(iter, offset); | |
1529 | set_iter_tags(iter, node, offset, tag); | |
1530 | node = rcu_dereference_raw(*slot); | |
1531 | } else { | |
1532 | offset = 0; | |
1533 | slot = &node->slots[0]; | |
1534 | for (;;) { | |
1535 | node = rcu_dereference_raw(*slot); | |
1536 | if (node) | |
1537 | break; | |
1538 | slot++; | |
1539 | offset++; | |
1540 | if (offset == RADIX_TREE_MAP_SIZE) | |
1541 | return NULL; | |
1542 | } | |
1543 | iter->index = __radix_tree_iter_add(iter, offset); | |
1544 | } | |
1545 | if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0)) | |
1546 | goto none; | |
1547 | next_index = (iter->index | shift_maxindex(iter->shift)) + 1; | |
1548 | if (next_index < iter->next_index) | |
1549 | iter->next_index = next_index; | |
1550 | } | |
1551 | ||
1552 | return slot; | |
1553 | none: | |
1554 | iter->next_index = 0; | |
1555 | return NULL; | |
1556 | } | |
1557 | EXPORT_SYMBOL(__radix_tree_next_slot); | |
1558 | #else | |
1559 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1560 | void **slot, struct radix_tree_iter *iter) | |
1561 | { | |
1562 | return slot; | |
1563 | } | |
1564 | #endif | |
1565 | ||
1566 | void **radix_tree_iter_resume(void **slot, struct radix_tree_iter *iter) | |
1567 | { | |
1568 | struct radix_tree_node *node; | |
1569 | ||
1570 | slot++; | |
1571 | iter->index = __radix_tree_iter_add(iter, 1); | |
1572 | node = rcu_dereference_raw(*slot); | |
1573 | skip_siblings(&node, slot, iter); | |
1574 | iter->next_index = iter->index; | |
1575 | iter->tags = 0; | |
1576 | return NULL; | |
1577 | } | |
1578 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1579 | ||
78c1d784 KK |
1580 | /** |
1581 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1582 | * | |
1583 | * @root: radix tree root | |
1584 | * @iter: iterator state | |
1585 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1586 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1587 | */ | |
35534c86 | 1588 | void **radix_tree_next_chunk(const struct radix_tree_root *root, |
78c1d784 KK |
1589 | struct radix_tree_iter *iter, unsigned flags) |
1590 | { | |
9e85d811 | 1591 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1592 | struct radix_tree_node *node, *child; |
21ef5339 | 1593 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1594 | |
1595 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1596 | return NULL; | |
1597 | ||
1598 | /* | |
1599 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1600 | * during iterating; it can be zero only at the beginning. | |
1601 | * And we cannot overflow iter->next_index in a single step, | |
1602 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1603 | * |
1604 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1605 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1606 | */ |
1607 | index = iter->next_index; | |
1608 | if (!index && iter->index) | |
1609 | return NULL; | |
1610 | ||
21ef5339 | 1611 | restart: |
9e85d811 | 1612 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1613 | if (index > maxindex) |
1614 | return NULL; | |
8c1244de MW |
1615 | if (!child) |
1616 | return NULL; | |
21ef5339 | 1617 | |
8c1244de | 1618 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1619 | /* Single-slot tree */ |
21ef5339 RZ |
1620 | iter->index = index; |
1621 | iter->next_index = maxindex + 1; | |
78c1d784 | 1622 | iter->tags = 1; |
268f42de | 1623 | iter->node = NULL; |
8c1244de | 1624 | __set_iter_shift(iter, 0); |
78c1d784 | 1625 | return (void **)&root->rnode; |
8c1244de | 1626 | } |
21ef5339 | 1627 | |
8c1244de MW |
1628 | do { |
1629 | node = entry_to_node(child); | |
9e85d811 | 1630 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1631 | |
78c1d784 | 1632 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1633 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1634 | /* Hole detected */ |
1635 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1636 | return NULL; | |
1637 | ||
1638 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1639 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1640 | offset + 1); |
1641 | else | |
1642 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
1643 | void *slot = node->slots[offset]; |
1644 | if (is_sibling_entry(node, slot)) | |
1645 | continue; | |
1646 | if (slot) | |
78c1d784 KK |
1647 | break; |
1648 | } | |
8c1244de | 1649 | index &= ~node_maxindex(node); |
9e85d811 | 1650 | index += offset << node->shift; |
78c1d784 KK |
1651 | /* Overflow after ~0UL */ |
1652 | if (!index) | |
1653 | return NULL; | |
1654 | if (offset == RADIX_TREE_MAP_SIZE) | |
1655 | goto restart; | |
8c1244de | 1656 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1657 | } |
1658 | ||
e157b555 | 1659 | if (!child) |
78c1d784 | 1660 | goto restart; |
e157b555 MW |
1661 | if (child == RADIX_TREE_RETRY) |
1662 | break; | |
8c1244de | 1663 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
1664 | |
1665 | /* Update the iterator state */ | |
8c1244de MW |
1666 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1667 | iter->next_index = (index | node_maxindex(node)) + 1; | |
268f42de | 1668 | iter->node = node; |
9e85d811 | 1669 | __set_iter_shift(iter, node->shift); |
78c1d784 | 1670 | |
148deab2 MW |
1671 | if (flags & RADIX_TREE_ITER_TAGGED) |
1672 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1673 | |
1674 | return node->slots + offset; | |
1675 | } | |
1676 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1677 | ||
1da177e4 LT |
1678 | /** |
1679 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1680 | * @root: radix tree root | |
1681 | * @results: where the results of the lookup are placed | |
1682 | * @first_index: start the lookup from this key | |
1683 | * @max_items: place up to this many items at *results | |
1684 | * | |
1685 | * Performs an index-ascending scan of the tree for present items. Places | |
1686 | * them at *@results and returns the number of items which were placed at | |
1687 | * *@results. | |
1688 | * | |
1689 | * The implementation is naive. | |
7cf9c2c7 NP |
1690 | * |
1691 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1692 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1693 | * an atomic snapshot of the tree at a single point in time, the |
1694 | * semantics of an RCU protected gang lookup are as though multiple | |
1695 | * radix_tree_lookups have been issued in individual locks, and results | |
1696 | * stored in 'results'. | |
1da177e4 LT |
1697 | */ |
1698 | unsigned int | |
35534c86 | 1699 | radix_tree_gang_lookup(const struct radix_tree_root *root, void **results, |
1da177e4 LT |
1700 | unsigned long first_index, unsigned int max_items) |
1701 | { | |
cebbd29e KK |
1702 | struct radix_tree_iter iter; |
1703 | void **slot; | |
1704 | unsigned int ret = 0; | |
7cf9c2c7 | 1705 | |
cebbd29e | 1706 | if (unlikely(!max_items)) |
7cf9c2c7 | 1707 | return 0; |
1da177e4 | 1708 | |
cebbd29e | 1709 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1710 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1711 | if (!results[ret]) |
1712 | continue; | |
b194d16c | 1713 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1714 | slot = radix_tree_iter_retry(&iter); |
1715 | continue; | |
1716 | } | |
cebbd29e | 1717 | if (++ret == max_items) |
1da177e4 | 1718 | break; |
1da177e4 | 1719 | } |
7cf9c2c7 | 1720 | |
1da177e4 LT |
1721 | return ret; |
1722 | } | |
1723 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1724 | ||
47feff2c NP |
1725 | /** |
1726 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1727 | * @root: radix tree root | |
1728 | * @results: where the results of the lookup are placed | |
6328650b | 1729 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1730 | * @first_index: start the lookup from this key |
1731 | * @max_items: place up to this many items at *results | |
1732 | * | |
1733 | * Performs an index-ascending scan of the tree for present items. Places | |
1734 | * their slots at *@results and returns the number of items which were | |
1735 | * placed at *@results. | |
1736 | * | |
1737 | * The implementation is naive. | |
1738 | * | |
1739 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1740 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1741 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1742 | */ | |
1743 | unsigned int | |
35534c86 | 1744 | radix_tree_gang_lookup_slot(const struct radix_tree_root *root, |
6328650b | 1745 | void ***results, unsigned long *indices, |
47feff2c NP |
1746 | unsigned long first_index, unsigned int max_items) |
1747 | { | |
cebbd29e KK |
1748 | struct radix_tree_iter iter; |
1749 | void **slot; | |
1750 | unsigned int ret = 0; | |
47feff2c | 1751 | |
cebbd29e | 1752 | if (unlikely(!max_items)) |
47feff2c NP |
1753 | return 0; |
1754 | ||
cebbd29e KK |
1755 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1756 | results[ret] = slot; | |
6328650b | 1757 | if (indices) |
cebbd29e KK |
1758 | indices[ret] = iter.index; |
1759 | if (++ret == max_items) | |
47feff2c | 1760 | break; |
47feff2c NP |
1761 | } |
1762 | ||
1763 | return ret; | |
1764 | } | |
1765 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1766 | ||
1da177e4 LT |
1767 | /** |
1768 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1769 | * based on a tag | |
1770 | * @root: radix tree root | |
1771 | * @results: where the results of the lookup are placed | |
1772 | * @first_index: start the lookup from this key | |
1773 | * @max_items: place up to this many items at *results | |
daff89f3 | 1774 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1775 | * |
1776 | * Performs an index-ascending scan of the tree for present items which | |
1777 | * have the tag indexed by @tag set. Places the items at *@results and | |
1778 | * returns the number of items which were placed at *@results. | |
1779 | */ | |
1780 | unsigned int | |
35534c86 | 1781 | radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results, |
daff89f3 JC |
1782 | unsigned long first_index, unsigned int max_items, |
1783 | unsigned int tag) | |
1da177e4 | 1784 | { |
cebbd29e KK |
1785 | struct radix_tree_iter iter; |
1786 | void **slot; | |
1787 | unsigned int ret = 0; | |
612d6c19 | 1788 | |
cebbd29e | 1789 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1790 | return 0; |
1791 | ||
cebbd29e | 1792 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1793 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1794 | if (!results[ret]) |
1795 | continue; | |
b194d16c | 1796 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1797 | slot = radix_tree_iter_retry(&iter); |
1798 | continue; | |
1799 | } | |
cebbd29e | 1800 | if (++ret == max_items) |
1da177e4 | 1801 | break; |
1da177e4 | 1802 | } |
7cf9c2c7 | 1803 | |
1da177e4 LT |
1804 | return ret; |
1805 | } | |
1806 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1807 | ||
47feff2c NP |
1808 | /** |
1809 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1810 | * radix tree based on a tag | |
1811 | * @root: radix tree root | |
1812 | * @results: where the results of the lookup are placed | |
1813 | * @first_index: start the lookup from this key | |
1814 | * @max_items: place up to this many items at *results | |
1815 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1816 | * | |
1817 | * Performs an index-ascending scan of the tree for present items which | |
1818 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1819 | * returns the number of slots which were placed at *@results. | |
1820 | */ | |
1821 | unsigned int | |
35534c86 MW |
1822 | radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root, |
1823 | void ***results, unsigned long first_index, | |
1824 | unsigned int max_items, unsigned int tag) | |
47feff2c | 1825 | { |
cebbd29e KK |
1826 | struct radix_tree_iter iter; |
1827 | void **slot; | |
1828 | unsigned int ret = 0; | |
47feff2c | 1829 | |
cebbd29e | 1830 | if (unlikely(!max_items)) |
47feff2c NP |
1831 | return 0; |
1832 | ||
cebbd29e KK |
1833 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1834 | results[ret] = slot; | |
1835 | if (++ret == max_items) | |
47feff2c | 1836 | break; |
47feff2c NP |
1837 | } |
1838 | ||
1839 | return ret; | |
1840 | } | |
1841 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1842 | ||
139e5616 JW |
1843 | /** |
1844 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1845 | * @root: radix tree root | |
139e5616 | 1846 | * @node: node containing @index |
ea07b862 JW |
1847 | * @update_node: callback for changing leaf nodes |
1848 | * @private: private data to pass to @update_node | |
139e5616 JW |
1849 | * |
1850 | * After clearing the slot at @index in @node from radix tree | |
1851 | * rooted at @root, call this function to attempt freeing the | |
1852 | * node and shrinking the tree. | |
139e5616 | 1853 | */ |
14b46879 | 1854 | void __radix_tree_delete_node(struct radix_tree_root *root, |
ea07b862 JW |
1855 | struct radix_tree_node *node, |
1856 | radix_tree_update_node_t update_node, | |
1857 | void *private) | |
139e5616 | 1858 | { |
ea07b862 | 1859 | delete_node(root, node, update_node, private); |
139e5616 JW |
1860 | } |
1861 | ||
1da177e4 | 1862 | /** |
53c59f26 | 1863 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1864 | * @root: radix tree root |
1865 | * @index: index key | |
53c59f26 | 1866 | * @item: expected item |
1da177e4 | 1867 | * |
53c59f26 | 1868 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1869 | * |
53c59f26 JW |
1870 | * Returns the address of the deleted item, or NULL if it was not present |
1871 | * or the entry at the given @index was not @item. | |
1da177e4 | 1872 | */ |
53c59f26 JW |
1873 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1874 | unsigned long index, void *item) | |
1da177e4 | 1875 | { |
139e5616 | 1876 | struct radix_tree_node *node; |
57578c2e | 1877 | unsigned int offset; |
139e5616 JW |
1878 | void **slot; |
1879 | void *entry; | |
d5274261 | 1880 | int tag; |
1da177e4 | 1881 | |
139e5616 JW |
1882 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1883 | if (!entry) | |
1884 | return NULL; | |
1da177e4 | 1885 | |
139e5616 JW |
1886 | if (item && entry != item) |
1887 | return NULL; | |
1888 | ||
1889 | if (!node) { | |
612d6c19 NP |
1890 | root_tag_clear_all(root); |
1891 | root->rnode = NULL; | |
139e5616 | 1892 | return entry; |
612d6c19 | 1893 | } |
1da177e4 | 1894 | |
29e0967c | 1895 | offset = get_slot_offset(node, slot); |
53c59f26 | 1896 | |
d604c324 MW |
1897 | /* Clear all tags associated with the item to be deleted. */ |
1898 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1899 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1900 | |
4d693d08 | 1901 | __radix_tree_replace(root, node, slot, NULL, NULL, NULL); |
612d6c19 | 1902 | |
139e5616 | 1903 | return entry; |
1da177e4 | 1904 | } |
53c59f26 JW |
1905 | EXPORT_SYMBOL(radix_tree_delete_item); |
1906 | ||
1907 | /** | |
1908 | * radix_tree_delete - delete an item from a radix tree | |
1909 | * @root: radix tree root | |
1910 | * @index: index key | |
1911 | * | |
1912 | * Remove the item at @index from the radix tree rooted at @root. | |
1913 | * | |
1914 | * Returns the address of the deleted item, or NULL if it was not present. | |
1915 | */ | |
1916 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1917 | { | |
1918 | return radix_tree_delete_item(root, index, NULL); | |
1919 | } | |
1da177e4 LT |
1920 | EXPORT_SYMBOL(radix_tree_delete); |
1921 | ||
d3798ae8 JW |
1922 | void radix_tree_clear_tags(struct radix_tree_root *root, |
1923 | struct radix_tree_node *node, | |
1924 | void **slot) | |
d604c324 | 1925 | { |
d604c324 MW |
1926 | if (node) { |
1927 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1928 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1929 | node_tag_clear(root, node, tag, offset); | |
1930 | } else { | |
1931 | /* Clear root node tags */ | |
1932 | root->gfp_mask &= __GFP_BITS_MASK; | |
1933 | } | |
d604c324 MW |
1934 | } |
1935 | ||
1da177e4 LT |
1936 | /** |
1937 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1938 | * @root: radix tree root | |
1939 | * @tag: tag to test | |
1940 | */ | |
35534c86 | 1941 | int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1942 | { |
612d6c19 | 1943 | return root_tag_get(root, tag); |
1da177e4 LT |
1944 | } |
1945 | EXPORT_SYMBOL(radix_tree_tagged); | |
1946 | ||
1947 | static void | |
449dd698 | 1948 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1949 | { |
449dd698 JW |
1950 | struct radix_tree_node *node = arg; |
1951 | ||
1952 | memset(node, 0, sizeof(*node)); | |
1953 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1954 | } |
1955 | ||
c78c66d1 KS |
1956 | static __init unsigned long __maxindex(unsigned int height) |
1957 | { | |
1958 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
1959 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1960 | ||
1961 | if (shift < 0) | |
1962 | return ~0UL; | |
1963 | if (shift >= BITS_PER_LONG) | |
1964 | return 0UL; | |
1965 | return ~0UL >> shift; | |
1966 | } | |
1967 | ||
1968 | static __init void radix_tree_init_maxnodes(void) | |
1969 | { | |
1970 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
1971 | unsigned int i, j; | |
1972 | ||
1973 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1974 | height_to_maxindex[i] = __maxindex(i); | |
1975 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
1976 | for (j = i; j > 0; j--) | |
1977 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
1978 | } | |
1979 | } | |
1980 | ||
d544abd5 | 1981 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1982 | { |
2fcd9005 MW |
1983 | struct radix_tree_preload *rtp; |
1984 | struct radix_tree_node *node; | |
1985 | ||
1986 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1987 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1988 | while (rtp->nr) { | |
1989 | node = rtp->nodes; | |
1990 | rtp->nodes = node->private_data; | |
1991 | kmem_cache_free(radix_tree_node_cachep, node); | |
1992 | rtp->nr--; | |
2fcd9005 | 1993 | } |
d544abd5 | 1994 | return 0; |
1da177e4 | 1995 | } |
1da177e4 LT |
1996 | |
1997 | void __init radix_tree_init(void) | |
1998 | { | |
d544abd5 | 1999 | int ret; |
1da177e4 LT |
2000 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
2001 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
2002 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
2003 | radix_tree_node_ctor); | |
c78c66d1 | 2004 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
2005 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
2006 | NULL, radix_tree_cpu_dead); | |
2007 | WARN_ON(ret < 0); | |
1da177e4 | 2008 | } |