]>
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> |
2b41226b | 33 | #include <linux/notifier.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 | |
a4db4dce | 72 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 73 | { |
30ff46cc | 74 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
75 | } |
76 | ||
a4db4dce | 77 | #define RADIX_TREE_RETRY node_to_entry(NULL) |
afe0e395 | 78 | |
db050f29 MW |
79 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
80 | /* Sibling slots point directly to another slot in the same node */ | |
81 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
82 | { | |
83 | void **ptr = node; | |
84 | return (parent->slots <= ptr) && | |
85 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
86 | } | |
87 | #else | |
88 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
89 | { | |
90 | return false; | |
91 | } | |
92 | #endif | |
93 | ||
94 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
95 | void **slot) | |
96 | { | |
97 | return slot - parent->slots; | |
98 | } | |
99 | ||
9e85d811 MW |
100 | static unsigned int radix_tree_descend(struct radix_tree_node *parent, |
101 | struct radix_tree_node **nodep, unsigned long index) | |
db050f29 | 102 | { |
9e85d811 | 103 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
db050f29 MW |
104 | void **entry = rcu_dereference_raw(parent->slots[offset]); |
105 | ||
106 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 107 | if (radix_tree_is_internal_node(entry)) { |
8d2c0d36 LT |
108 | if (is_sibling_entry(parent, entry)) { |
109 | void **sibentry = (void **) entry_to_node(entry); | |
110 | offset = get_slot_offset(parent, sibentry); | |
111 | entry = rcu_dereference_raw(*sibentry); | |
db050f29 MW |
112 | } |
113 | } | |
114 | #endif | |
115 | ||
116 | *nodep = (void *)entry; | |
117 | return offset; | |
118 | } | |
119 | ||
612d6c19 NP |
120 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
121 | { | |
122 | return root->gfp_mask & __GFP_BITS_MASK; | |
123 | } | |
124 | ||
643b52b9 NP |
125 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
126 | int offset) | |
127 | { | |
128 | __set_bit(offset, node->tags[tag]); | |
129 | } | |
130 | ||
131 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
132 | int offset) | |
133 | { | |
134 | __clear_bit(offset, node->tags[tag]); | |
135 | } | |
136 | ||
137 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
138 | int offset) | |
139 | { | |
140 | return test_bit(offset, node->tags[tag]); | |
141 | } | |
142 | ||
143 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
144 | { | |
145 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
146 | } | |
147 | ||
2fcd9005 | 148 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
149 | { |
150 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
151 | } | |
152 | ||
153 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
154 | { | |
155 | root->gfp_mask &= __GFP_BITS_MASK; | |
156 | } | |
157 | ||
158 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
159 | { | |
2fcd9005 | 160 | return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
643b52b9 NP |
161 | } |
162 | ||
7b60e9ad MW |
163 | static inline unsigned root_tags_get(struct radix_tree_root *root) |
164 | { | |
165 | return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT; | |
166 | } | |
167 | ||
643b52b9 NP |
168 | /* |
169 | * Returns 1 if any slot in the node has this tag set. | |
170 | * Otherwise returns 0. | |
171 | */ | |
172 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
173 | { | |
2fcd9005 | 174 | unsigned idx; |
643b52b9 NP |
175 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
176 | if (node->tags[tag][idx]) | |
177 | return 1; | |
178 | } | |
179 | return 0; | |
180 | } | |
78c1d784 KK |
181 | |
182 | /** | |
183 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
184 | * | |
185 | * @addr: The address to base the search on | |
186 | * @size: The bitmap size in bits | |
187 | * @offset: The bitnumber to start searching at | |
188 | * | |
189 | * Unrollable variant of find_next_bit() for constant size arrays. | |
190 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
191 | * Returns next bit offset, or size if nothing found. | |
192 | */ | |
193 | static __always_inline unsigned long | |
194 | radix_tree_find_next_bit(const unsigned long *addr, | |
195 | unsigned long size, unsigned long offset) | |
196 | { | |
197 | if (!__builtin_constant_p(size)) | |
198 | return find_next_bit(addr, size, offset); | |
199 | ||
200 | if (offset < size) { | |
201 | unsigned long tmp; | |
202 | ||
203 | addr += offset / BITS_PER_LONG; | |
204 | tmp = *addr >> (offset % BITS_PER_LONG); | |
205 | if (tmp) | |
206 | return __ffs(tmp) + offset; | |
207 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
208 | while (offset < size) { | |
209 | tmp = *++addr; | |
210 | if (tmp) | |
211 | return __ffs(tmp) + offset; | |
212 | offset += BITS_PER_LONG; | |
213 | } | |
214 | } | |
215 | return size; | |
216 | } | |
217 | ||
0796c583 | 218 | #ifndef __KERNEL__ |
d0891265 | 219 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 220 | { |
0796c583 | 221 | unsigned long i; |
7cf19af4 | 222 | |
f7942430 | 223 | pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d exceptional %d parent %p\n", |
0c7fa0a8 | 224 | node, node->offset, |
0796c583 | 225 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
f7942430 | 226 | node->shift, node->count, node->exceptional, node->parent); |
0796c583 RZ |
227 | |
228 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
229 | unsigned long first = index | (i << node->shift); |
230 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
231 | void *entry = node->slots[i]; |
232 | if (!entry) | |
233 | continue; | |
234 | if (is_sibling_entry(node, entry)) { | |
235 | pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n", | |
236 | entry, i, | |
4dd6c098 | 237 | *(void **)entry_to_node(entry), |
0796c583 | 238 | first, last); |
b194d16c | 239 | } else if (!radix_tree_is_internal_node(entry)) { |
0796c583 RZ |
240 | pr_debug("radix entry %p offset %ld indices %ld-%ld\n", |
241 | entry, i, first, last); | |
242 | } else { | |
4dd6c098 | 243 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
244 | } |
245 | } | |
7cf19af4 MW |
246 | } |
247 | ||
248 | /* For debug */ | |
249 | static void radix_tree_dump(struct radix_tree_root *root) | |
250 | { | |
d0891265 MW |
251 | pr_debug("radix root: %p rnode %p tags %x\n", |
252 | root, root->rnode, | |
7cf19af4 | 253 | root->gfp_mask >> __GFP_BITS_SHIFT); |
b194d16c | 254 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 255 | return; |
4dd6c098 | 256 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 MW |
257 | } |
258 | #endif | |
259 | ||
1da177e4 LT |
260 | /* |
261 | * This assumes that the caller has performed appropriate preallocation, and | |
262 | * that the caller has pinned this thread of control to the current CPU. | |
263 | */ | |
264 | static struct radix_tree_node * | |
265 | radix_tree_node_alloc(struct radix_tree_root *root) | |
266 | { | |
e2848a0e | 267 | struct radix_tree_node *ret = NULL; |
612d6c19 | 268 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 269 | |
5e4c0d97 | 270 | /* |
2fcd9005 MW |
271 | * Preload code isn't irq safe and it doesn't make sense to use |
272 | * preloading during an interrupt anyway as all the allocations have | |
273 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 274 | */ |
d0164adc | 275 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
276 | struct radix_tree_preload *rtp; |
277 | ||
58e698af VD |
278 | /* |
279 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
280 | * cache first for the new node to get accounted to the memory |
281 | * cgroup. | |
58e698af VD |
282 | */ |
283 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 284 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
285 | if (ret) |
286 | goto out; | |
287 | ||
e2848a0e NP |
288 | /* |
289 | * Provided the caller has preloaded here, we will always | |
290 | * succeed in getting a node here (and never reach | |
291 | * kmem_cache_alloc) | |
292 | */ | |
7c8e0181 | 293 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 294 | if (rtp->nr) { |
9d2a8da0 KS |
295 | ret = rtp->nodes; |
296 | rtp->nodes = ret->private_data; | |
297 | ret->private_data = NULL; | |
1da177e4 LT |
298 | rtp->nr--; |
299 | } | |
ce80b067 CM |
300 | /* |
301 | * Update the allocation stack trace as this is more useful | |
302 | * for debugging. | |
303 | */ | |
304 | kmemleak_update_trace(ret); | |
58e698af | 305 | goto out; |
1da177e4 | 306 | } |
05eb6e72 | 307 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 308 | out: |
b194d16c | 309 | BUG_ON(radix_tree_is_internal_node(ret)); |
1da177e4 LT |
310 | return ret; |
311 | } | |
312 | ||
7cf9c2c7 NP |
313 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
314 | { | |
315 | struct radix_tree_node *node = | |
316 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 317 | int i; |
643b52b9 NP |
318 | |
319 | /* | |
320 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
321 | * can leave us with a non-NULL entry in the first slot, so clear | |
322 | * that here to make sure. | |
323 | */ | |
b6dd0865 DC |
324 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
325 | tag_clear(node, i, 0); | |
326 | ||
643b52b9 NP |
327 | node->slots[0] = NULL; |
328 | node->count = 0; | |
329 | ||
7cf9c2c7 NP |
330 | kmem_cache_free(radix_tree_node_cachep, node); |
331 | } | |
332 | ||
1da177e4 LT |
333 | static inline void |
334 | radix_tree_node_free(struct radix_tree_node *node) | |
335 | { | |
7cf9c2c7 | 336 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
337 | } |
338 | ||
339 | /* | |
340 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
341 | * ensure that the addition of a single element in the tree cannot fail. On | |
342 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
343 | * with preemption not disabled. | |
b34df792 DH |
344 | * |
345 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 346 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 347 | */ |
c78c66d1 | 348 | static int __radix_tree_preload(gfp_t gfp_mask, int nr) |
1da177e4 LT |
349 | { |
350 | struct radix_tree_preload *rtp; | |
351 | struct radix_tree_node *node; | |
352 | int ret = -ENOMEM; | |
353 | ||
05eb6e72 VD |
354 | /* |
355 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
356 | * they should never be accounted to any particular memory cgroup. | |
357 | */ | |
358 | gfp_mask &= ~__GFP_ACCOUNT; | |
359 | ||
1da177e4 | 360 | preempt_disable(); |
7c8e0181 | 361 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 362 | while (rtp->nr < nr) { |
1da177e4 | 363 | preempt_enable(); |
488514d1 | 364 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
365 | if (node == NULL) |
366 | goto out; | |
367 | preempt_disable(); | |
7c8e0181 | 368 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 369 | if (rtp->nr < nr) { |
9d2a8da0 KS |
370 | node->private_data = rtp->nodes; |
371 | rtp->nodes = node; | |
372 | rtp->nr++; | |
373 | } else { | |
1da177e4 | 374 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 375 | } |
1da177e4 LT |
376 | } |
377 | ret = 0; | |
378 | out: | |
379 | return ret; | |
380 | } | |
5e4c0d97 JK |
381 | |
382 | /* | |
383 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
384 | * ensure that the addition of a single element in the tree cannot fail. On | |
385 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
386 | * with preemption not disabled. | |
387 | * | |
388 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 389 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
390 | */ |
391 | int radix_tree_preload(gfp_t gfp_mask) | |
392 | { | |
393 | /* Warn on non-sensical use... */ | |
d0164adc | 394 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 395 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 396 | } |
d7f0923d | 397 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 398 | |
5e4c0d97 JK |
399 | /* |
400 | * The same as above function, except we don't guarantee preloading happens. | |
401 | * We do it, if we decide it helps. On success, return zero with preemption | |
402 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
403 | */ | |
404 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
405 | { | |
d0164adc | 406 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 407 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
408 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
409 | preempt_disable(); | |
410 | return 0; | |
411 | } | |
412 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
413 | ||
c78c66d1 KS |
414 | /* |
415 | * The same as function above, but preload number of nodes required to insert | |
416 | * (1 << order) continuous naturally-aligned elements. | |
417 | */ | |
418 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
419 | { | |
420 | unsigned long nr_subtrees; | |
421 | int nr_nodes, subtree_height; | |
422 | ||
423 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
424 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
425 | preempt_disable(); | |
426 | return 0; | |
427 | } | |
428 | ||
429 | /* | |
430 | * Calculate number and height of fully populated subtrees it takes to | |
431 | * store (1 << order) elements. | |
432 | */ | |
433 | nr_subtrees = 1 << order; | |
434 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
435 | subtree_height++) | |
436 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
437 | ||
438 | /* | |
439 | * The worst case is zero height tree with a single item at index 0 and | |
440 | * then inserting items starting at ULONG_MAX - (1 << order). | |
441 | * | |
442 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
443 | * 0-index item. | |
444 | */ | |
445 | nr_nodes = RADIX_TREE_MAX_PATH; | |
446 | ||
447 | /* Plus branch to fully populated subtrees. */ | |
448 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
449 | ||
450 | /* Root node is shared. */ | |
451 | nr_nodes--; | |
452 | ||
453 | /* Plus nodes required to build subtrees. */ | |
454 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
455 | ||
456 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
457 | } | |
458 | ||
1da177e4 | 459 | /* |
d0891265 | 460 | * The maximum index which can be stored in a radix tree |
1da177e4 | 461 | */ |
c12e51b0 MW |
462 | static inline unsigned long shift_maxindex(unsigned int shift) |
463 | { | |
464 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
465 | } | |
466 | ||
1456a439 MW |
467 | static inline unsigned long node_maxindex(struct radix_tree_node *node) |
468 | { | |
c12e51b0 | 469 | return shift_maxindex(node->shift); |
1456a439 MW |
470 | } |
471 | ||
472 | static unsigned radix_tree_load_root(struct radix_tree_root *root, | |
473 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
474 | { | |
475 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
476 | ||
477 | *nodep = node; | |
478 | ||
b194d16c | 479 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 480 | node = entry_to_node(node); |
1456a439 | 481 | *maxindex = node_maxindex(node); |
c12e51b0 | 482 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
483 | } |
484 | ||
485 | *maxindex = 0; | |
486 | return 0; | |
487 | } | |
488 | ||
1da177e4 LT |
489 | /* |
490 | * Extend a radix tree so it can store key @index. | |
491 | */ | |
e6145236 | 492 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 493 | unsigned long index, unsigned int shift) |
1da177e4 | 494 | { |
e2bdb933 | 495 | struct radix_tree_node *slot; |
d0891265 | 496 | unsigned int maxshift; |
1da177e4 LT |
497 | int tag; |
498 | ||
d0891265 MW |
499 | /* Figure out what the shift should be. */ |
500 | maxshift = shift; | |
501 | while (index > shift_maxindex(maxshift)) | |
502 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 503 | |
d0891265 MW |
504 | slot = root->rnode; |
505 | if (!slot) | |
1da177e4 | 506 | goto out; |
1da177e4 | 507 | |
1da177e4 | 508 | do { |
2fcd9005 MW |
509 | struct radix_tree_node *node = radix_tree_node_alloc(root); |
510 | ||
511 | if (!node) | |
1da177e4 LT |
512 | return -ENOMEM; |
513 | ||
1da177e4 | 514 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 515 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 516 | if (root_tag_get(root, tag)) |
1da177e4 LT |
517 | tag_set(node, tag, 0); |
518 | } | |
519 | ||
d0891265 MW |
520 | BUG_ON(shift > BITS_PER_LONG); |
521 | node->shift = shift; | |
0c7fa0a8 | 522 | node->offset = 0; |
1da177e4 | 523 | node->count = 1; |
e2bdb933 | 524 | node->parent = NULL; |
f7942430 | 525 | if (radix_tree_is_internal_node(slot)) { |
4dd6c098 | 526 | entry_to_node(slot)->parent = node; |
f7942430 JW |
527 | } else { |
528 | /* Moving an exceptional root->rnode to a node */ | |
529 | if (radix_tree_exceptional_entry(slot)) | |
530 | node->exceptional = 1; | |
531 | } | |
e2bdb933 | 532 | node->slots[0] = slot; |
a4db4dce MW |
533 | slot = node_to_entry(node); |
534 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 535 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 536 | } while (shift <= maxshift); |
1da177e4 | 537 | out: |
d0891265 | 538 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
539 | } |
540 | ||
541 | /** | |
139e5616 | 542 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
543 | * @root: radix tree root |
544 | * @index: index key | |
e6145236 | 545 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
546 | * @nodep: returns node |
547 | * @slotp: returns slot | |
1da177e4 | 548 | * |
139e5616 JW |
549 | * Create, if necessary, and return the node and slot for an item |
550 | * at position @index in the radix tree @root. | |
551 | * | |
552 | * Until there is more than one item in the tree, no nodes are | |
553 | * allocated and @root->rnode is used as a direct slot instead of | |
554 | * pointing to a node, in which case *@nodep will be NULL. | |
555 | * | |
556 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 557 | */ |
139e5616 | 558 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
559 | unsigned order, struct radix_tree_node **nodep, |
560 | void ***slotp) | |
1da177e4 | 561 | { |
89148aa4 MW |
562 | struct radix_tree_node *node = NULL, *child; |
563 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 564 | unsigned long maxindex; |
89148aa4 | 565 | unsigned int shift, offset = 0; |
49ea6ebc MW |
566 | unsigned long max = index | ((1UL << order) - 1); |
567 | ||
89148aa4 | 568 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
569 | |
570 | /* Make sure the tree is high enough. */ | |
49ea6ebc | 571 | if (max > maxindex) { |
d0891265 | 572 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 573 | if (error < 0) |
1da177e4 | 574 | return error; |
49ea6ebc | 575 | shift = error; |
89148aa4 | 576 | child = root->rnode; |
d0891265 | 577 | if (order == shift) |
49ea6ebc | 578 | shift += RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
579 | } |
580 | ||
e6145236 | 581 | while (shift > order) { |
c12e51b0 | 582 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 583 | if (child == NULL) { |
1da177e4 | 584 | /* Have to add a child node. */ |
89148aa4 MW |
585 | child = radix_tree_node_alloc(root); |
586 | if (!child) | |
1da177e4 | 587 | return -ENOMEM; |
89148aa4 MW |
588 | child->shift = shift; |
589 | child->offset = offset; | |
590 | child->parent = node; | |
591 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
592 | if (node) | |
1da177e4 | 593 | node->count++; |
89148aa4 | 594 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 595 | break; |
1da177e4 LT |
596 | |
597 | /* Go a level down */ | |
89148aa4 | 598 | node = entry_to_node(child); |
9e85d811 | 599 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 600 | slot = &node->slots[offset]; |
e6145236 MW |
601 | } |
602 | ||
57578c2e | 603 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 604 | /* Insert pointers to the canonical entry */ |
3b8c00f6 | 605 | if (order > shift) { |
89148aa4 | 606 | unsigned i, n = 1 << (order - shift); |
e6145236 | 607 | offset = offset & ~(n - 1); |
89148aa4 MW |
608 | slot = &node->slots[offset]; |
609 | child = node_to_entry(slot); | |
e6145236 | 610 | for (i = 0; i < n; i++) { |
89148aa4 | 611 | if (slot[i]) |
e6145236 MW |
612 | return -EEXIST; |
613 | } | |
614 | ||
615 | for (i = 1; i < n; i++) { | |
89148aa4 | 616 | rcu_assign_pointer(slot[i], child); |
e6145236 MW |
617 | node->count++; |
618 | } | |
612d6c19 | 619 | } |
57578c2e | 620 | #endif |
1da177e4 | 621 | |
139e5616 JW |
622 | if (nodep) |
623 | *nodep = node; | |
624 | if (slotp) | |
89148aa4 | 625 | *slotp = slot; |
139e5616 JW |
626 | return 0; |
627 | } | |
628 | ||
629 | /** | |
e6145236 | 630 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
631 | * @root: radix tree root |
632 | * @index: index key | |
e6145236 | 633 | * @order: key covers the 2^order indices around index |
139e5616 JW |
634 | * @item: item to insert |
635 | * | |
636 | * Insert an item into the radix tree at position @index. | |
637 | */ | |
e6145236 MW |
638 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
639 | unsigned order, void *item) | |
139e5616 JW |
640 | { |
641 | struct radix_tree_node *node; | |
642 | void **slot; | |
643 | int error; | |
644 | ||
b194d16c | 645 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 646 | |
e6145236 | 647 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
648 | if (error) |
649 | return error; | |
650 | if (*slot != NULL) | |
1da177e4 | 651 | return -EEXIST; |
139e5616 | 652 | rcu_assign_pointer(*slot, item); |
201b6264 | 653 | |
612d6c19 | 654 | if (node) { |
7b60e9ad | 655 | unsigned offset = get_slot_offset(node, slot); |
612d6c19 | 656 | node->count++; |
f7942430 JW |
657 | if (radix_tree_exceptional_entry(item)) |
658 | node->exceptional++; | |
7b60e9ad MW |
659 | BUG_ON(tag_get(node, 0, offset)); |
660 | BUG_ON(tag_get(node, 1, offset)); | |
661 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 662 | } else { |
7b60e9ad | 663 | BUG_ON(root_tags_get(root)); |
612d6c19 | 664 | } |
1da177e4 | 665 | |
1da177e4 LT |
666 | return 0; |
667 | } | |
e6145236 | 668 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 669 | |
139e5616 JW |
670 | /** |
671 | * __radix_tree_lookup - lookup an item in a radix tree | |
672 | * @root: radix tree root | |
673 | * @index: index key | |
674 | * @nodep: returns node | |
675 | * @slotp: returns slot | |
676 | * | |
677 | * Lookup and return the item at position @index in the radix | |
678 | * tree @root. | |
679 | * | |
680 | * Until there is more than one item in the tree, no nodes are | |
681 | * allocated and @root->rnode is used as a direct slot instead of | |
682 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 683 | */ |
139e5616 JW |
684 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
685 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 686 | { |
139e5616 | 687 | struct radix_tree_node *node, *parent; |
85829954 | 688 | unsigned long maxindex; |
139e5616 | 689 | void **slot; |
612d6c19 | 690 | |
85829954 MW |
691 | restart: |
692 | parent = NULL; | |
693 | slot = (void **)&root->rnode; | |
9e85d811 | 694 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 695 | if (index > maxindex) |
1da177e4 LT |
696 | return NULL; |
697 | ||
b194d16c | 698 | while (radix_tree_is_internal_node(node)) { |
85829954 | 699 | unsigned offset; |
1da177e4 | 700 | |
85829954 MW |
701 | if (node == RADIX_TREE_RETRY) |
702 | goto restart; | |
4dd6c098 | 703 | parent = entry_to_node(node); |
9e85d811 | 704 | offset = radix_tree_descend(parent, &node, index); |
85829954 MW |
705 | slot = parent->slots + offset; |
706 | } | |
1da177e4 | 707 | |
139e5616 JW |
708 | if (nodep) |
709 | *nodep = parent; | |
710 | if (slotp) | |
711 | *slotp = slot; | |
712 | return node; | |
b72b71c6 HS |
713 | } |
714 | ||
715 | /** | |
716 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
717 | * @root: radix tree root | |
718 | * @index: index key | |
719 | * | |
720 | * Returns: the slot corresponding to the position @index in the | |
721 | * radix tree @root. This is useful for update-if-exists operations. | |
722 | * | |
723 | * This function can be called under rcu_read_lock iff the slot is not | |
724 | * modified by radix_tree_replace_slot, otherwise it must be called | |
725 | * exclusive from other writers. Any dereference of the slot must be done | |
726 | * using radix_tree_deref_slot. | |
727 | */ | |
728 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
729 | { | |
139e5616 JW |
730 | void **slot; |
731 | ||
732 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
733 | return NULL; | |
734 | return slot; | |
a4331366 | 735 | } |
a4331366 HR |
736 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
737 | ||
738 | /** | |
739 | * radix_tree_lookup - perform lookup operation on a radix tree | |
740 | * @root: radix tree root | |
741 | * @index: index key | |
742 | * | |
743 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
744 | * |
745 | * This function can be called under rcu_read_lock, however the caller | |
746 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
747 | * them safely). No RCU barriers are required to access or modify the | |
748 | * returned item, however. | |
a4331366 HR |
749 | */ |
750 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
751 | { | |
139e5616 | 752 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
753 | } |
754 | EXPORT_SYMBOL(radix_tree_lookup); | |
755 | ||
f7942430 JW |
756 | /** |
757 | * __radix_tree_replace - replace item in a slot | |
758 | * @root: radix tree root | |
759 | * @node: pointer to tree node | |
760 | * @slot: pointer to slot in @node | |
761 | * @item: new item to store in the slot. | |
762 | * | |
763 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
764 | * across slot lookup and replacement. | |
765 | */ | |
766 | void __radix_tree_replace(struct radix_tree_root *root, | |
767 | struct radix_tree_node *node, | |
768 | void **slot, void *item) | |
769 | { | |
770 | void *old = rcu_dereference_raw(*slot); | |
771 | int exceptional; | |
772 | ||
773 | WARN_ON_ONCE(radix_tree_is_internal_node(item)); | |
774 | WARN_ON_ONCE(!!item - !!old); | |
775 | ||
776 | exceptional = !!radix_tree_exceptional_entry(item) - | |
777 | !!radix_tree_exceptional_entry(old); | |
778 | ||
779 | WARN_ON_ONCE(exceptional && !node && slot != (void **)&root->rnode); | |
780 | ||
781 | if (node) | |
782 | node->exceptional += exceptional; | |
783 | ||
784 | rcu_assign_pointer(*slot, item); | |
785 | } | |
786 | ||
1da177e4 LT |
787 | /** |
788 | * radix_tree_tag_set - set a tag on a radix tree node | |
789 | * @root: radix tree root | |
790 | * @index: index key | |
2fcd9005 | 791 | * @tag: tag index |
1da177e4 | 792 | * |
daff89f3 JC |
793 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
794 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
795 | * the root all the way down to the leaf node. |
796 | * | |
2fcd9005 | 797 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
798 | * item is a bug. |
799 | */ | |
800 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 801 | unsigned long index, unsigned int tag) |
1da177e4 | 802 | { |
fb969909 RZ |
803 | struct radix_tree_node *node, *parent; |
804 | unsigned long maxindex; | |
1da177e4 | 805 | |
9e85d811 | 806 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 807 | BUG_ON(index > maxindex); |
1da177e4 | 808 | |
b194d16c | 809 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 810 | unsigned offset; |
1da177e4 | 811 | |
4dd6c098 | 812 | parent = entry_to_node(node); |
9e85d811 | 813 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
814 | BUG_ON(!node); |
815 | ||
816 | if (!tag_get(parent, tag, offset)) | |
817 | tag_set(parent, tag, offset); | |
1da177e4 LT |
818 | } |
819 | ||
612d6c19 | 820 | /* set the root's tag bit */ |
fb969909 | 821 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
822 | root_tag_set(root, tag); |
823 | ||
fb969909 | 824 | return node; |
1da177e4 LT |
825 | } |
826 | EXPORT_SYMBOL(radix_tree_tag_set); | |
827 | ||
d604c324 MW |
828 | static void node_tag_clear(struct radix_tree_root *root, |
829 | struct radix_tree_node *node, | |
830 | unsigned int tag, unsigned int offset) | |
831 | { | |
832 | while (node) { | |
833 | if (!tag_get(node, tag, offset)) | |
834 | return; | |
835 | tag_clear(node, tag, offset); | |
836 | if (any_tag_set(node, tag)) | |
837 | return; | |
838 | ||
839 | offset = node->offset; | |
840 | node = node->parent; | |
841 | } | |
842 | ||
843 | /* clear the root's tag bit */ | |
844 | if (root_tag_get(root, tag)) | |
845 | root_tag_clear(root, tag); | |
846 | } | |
847 | ||
1da177e4 LT |
848 | /** |
849 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
850 | * @root: radix tree root | |
851 | * @index: index key | |
2fcd9005 | 852 | * @tag: tag index |
1da177e4 | 853 | * |
daff89f3 | 854 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
855 | * corresponding to @index in the radix tree. If this causes |
856 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
857 | * next-to-leaf node, etc. |
858 | * | |
859 | * Returns the address of the tagged item on success, else NULL. ie: | |
860 | * has the same return value and semantics as radix_tree_lookup(). | |
861 | */ | |
862 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 863 | unsigned long index, unsigned int tag) |
1da177e4 | 864 | { |
00f47b58 RZ |
865 | struct radix_tree_node *node, *parent; |
866 | unsigned long maxindex; | |
e2bdb933 | 867 | int uninitialized_var(offset); |
1da177e4 | 868 | |
9e85d811 | 869 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
870 | if (index > maxindex) |
871 | return NULL; | |
1da177e4 | 872 | |
00f47b58 | 873 | parent = NULL; |
1da177e4 | 874 | |
b194d16c | 875 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 876 | parent = entry_to_node(node); |
9e85d811 | 877 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
878 | } |
879 | ||
d604c324 MW |
880 | if (node) |
881 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 882 | |
00f47b58 | 883 | return node; |
1da177e4 LT |
884 | } |
885 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
886 | ||
1da177e4 | 887 | /** |
32605a18 MT |
888 | * radix_tree_tag_get - get a tag on a radix tree node |
889 | * @root: radix tree root | |
890 | * @index: index key | |
2fcd9005 | 891 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 892 | * |
32605a18 | 893 | * Return values: |
1da177e4 | 894 | * |
612d6c19 NP |
895 | * 0: tag not present or not set |
896 | * 1: tag set | |
ce82653d DH |
897 | * |
898 | * Note that the return value of this function may not be relied on, even if | |
899 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
900 | * from concurrency. | |
1da177e4 LT |
901 | */ |
902 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 903 | unsigned long index, unsigned int tag) |
1da177e4 | 904 | { |
4589ba6d RZ |
905 | struct radix_tree_node *node, *parent; |
906 | unsigned long maxindex; | |
1da177e4 | 907 | |
612d6c19 NP |
908 | if (!root_tag_get(root, tag)) |
909 | return 0; | |
910 | ||
9e85d811 | 911 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
912 | if (index > maxindex) |
913 | return 0; | |
7cf9c2c7 NP |
914 | if (node == NULL) |
915 | return 0; | |
916 | ||
b194d16c | 917 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 918 | unsigned offset; |
1da177e4 | 919 | |
4dd6c098 | 920 | parent = entry_to_node(node); |
9e85d811 | 921 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 922 | |
4589ba6d | 923 | if (!node) |
1da177e4 | 924 | return 0; |
4589ba6d | 925 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 926 | return 0; |
4589ba6d RZ |
927 | if (node == RADIX_TREE_RETRY) |
928 | break; | |
1da177e4 | 929 | } |
4589ba6d RZ |
930 | |
931 | return 1; | |
1da177e4 LT |
932 | } |
933 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 934 | |
21ef5339 RZ |
935 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
936 | unsigned int shift) | |
937 | { | |
938 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
939 | iter->shift = shift; | |
940 | #endif | |
941 | } | |
942 | ||
78c1d784 KK |
943 | /** |
944 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
945 | * | |
946 | * @root: radix tree root | |
947 | * @iter: iterator state | |
948 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
949 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
950 | */ | |
951 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
952 | struct radix_tree_iter *iter, unsigned flags) | |
953 | { | |
9e85d811 | 954 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 955 | struct radix_tree_node *node, *child; |
21ef5339 | 956 | unsigned long index, offset, maxindex; |
78c1d784 KK |
957 | |
958 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
959 | return NULL; | |
960 | ||
961 | /* | |
962 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
963 | * during iterating; it can be zero only at the beginning. | |
964 | * And we cannot overflow iter->next_index in a single step, | |
965 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
966 | * |
967 | * This condition also used by radix_tree_next_slot() to stop | |
968 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
969 | */ |
970 | index = iter->next_index; | |
971 | if (!index && iter->index) | |
972 | return NULL; | |
973 | ||
21ef5339 | 974 | restart: |
9e85d811 | 975 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
976 | if (index > maxindex) |
977 | return NULL; | |
8c1244de MW |
978 | if (!child) |
979 | return NULL; | |
21ef5339 | 980 | |
8c1244de | 981 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 982 | /* Single-slot tree */ |
21ef5339 RZ |
983 | iter->index = index; |
984 | iter->next_index = maxindex + 1; | |
78c1d784 | 985 | iter->tags = 1; |
8c1244de | 986 | __set_iter_shift(iter, 0); |
78c1d784 | 987 | return (void **)&root->rnode; |
8c1244de | 988 | } |
21ef5339 | 989 | |
8c1244de MW |
990 | do { |
991 | node = entry_to_node(child); | |
9e85d811 | 992 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 993 | |
78c1d784 | 994 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 995 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
996 | /* Hole detected */ |
997 | if (flags & RADIX_TREE_ITER_CONTIG) | |
998 | return NULL; | |
999 | ||
1000 | if (flags & RADIX_TREE_ITER_TAGGED) | |
1001 | offset = radix_tree_find_next_bit( | |
1002 | node->tags[tag], | |
1003 | RADIX_TREE_MAP_SIZE, | |
1004 | offset + 1); | |
1005 | else | |
1006 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
1007 | void *slot = node->slots[offset]; |
1008 | if (is_sibling_entry(node, slot)) | |
1009 | continue; | |
1010 | if (slot) | |
78c1d784 KK |
1011 | break; |
1012 | } | |
8c1244de | 1013 | index &= ~node_maxindex(node); |
9e85d811 | 1014 | index += offset << node->shift; |
78c1d784 KK |
1015 | /* Overflow after ~0UL */ |
1016 | if (!index) | |
1017 | return NULL; | |
1018 | if (offset == RADIX_TREE_MAP_SIZE) | |
1019 | goto restart; | |
8c1244de | 1020 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1021 | } |
1022 | ||
8c1244de | 1023 | if ((child == NULL) || (child == RADIX_TREE_RETRY)) |
78c1d784 | 1024 | goto restart; |
8c1244de | 1025 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
1026 | |
1027 | /* Update the iterator state */ | |
8c1244de MW |
1028 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1029 | iter->next_index = (index | node_maxindex(node)) + 1; | |
9e85d811 | 1030 | __set_iter_shift(iter, node->shift); |
78c1d784 KK |
1031 | |
1032 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
1033 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1034 | unsigned tag_long, tag_bit; | |
1035 | ||
1036 | tag_long = offset / BITS_PER_LONG; | |
1037 | tag_bit = offset % BITS_PER_LONG; | |
1038 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1039 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1040 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1041 | /* Pick tags from next element */ | |
1042 | if (tag_bit) | |
1043 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1044 | (BITS_PER_LONG - tag_bit); | |
1045 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1046 | iter->next_index = index + BITS_PER_LONG; | |
1047 | } | |
1048 | } | |
1049 | ||
1050 | return node->slots + offset; | |
1051 | } | |
1052 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1053 | ||
ebf8aa44 JK |
1054 | /** |
1055 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
1056 | * tag if item has another tag set | |
1057 | * @root: radix tree root | |
1058 | * @first_indexp: pointer to a starting index of a range to scan | |
1059 | * @last_index: last index of a range to scan | |
1060 | * @nr_to_tag: maximum number items to tag | |
1061 | * @iftag: tag index to test | |
1062 | * @settag: tag index to set if tested tag is set | |
1063 | * | |
1064 | * This function scans range of radix tree from first_index to last_index | |
1065 | * (inclusive). For each item in the range if iftag is set, the function sets | |
1066 | * also settag. The function stops either after tagging nr_to_tag items or | |
1067 | * after reaching last_index. | |
1068 | * | |
144dcfc0 DC |
1069 | * The tags must be set from the leaf level only and propagated back up the |
1070 | * path to the root. We must do this so that we resolve the full path before | |
1071 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
1072 | * we can get to the leaf node and find that the index that has the iftag | |
1073 | * set is outside the range we are scanning. This reults in dangling tags and | |
1074 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1075 | * | |
2fcd9005 | 1076 | * The function returns the number of leaves where the tag was set and sets |
ebf8aa44 | 1077 | * *first_indexp to the first unscanned index. |
d5ed3a4a JK |
1078 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1079 | * be prepared to handle that. | |
ebf8aa44 JK |
1080 | */ |
1081 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1082 | unsigned long *first_indexp, unsigned long last_index, | |
1083 | unsigned long nr_to_tag, | |
1084 | unsigned int iftag, unsigned int settag) | |
1085 | { | |
a8e4da25 | 1086 | struct radix_tree_node *parent, *node, *child; |
070c5ac2 | 1087 | unsigned long maxindex; |
144dcfc0 DC |
1088 | unsigned long tagged = 0; |
1089 | unsigned long index = *first_indexp; | |
ebf8aa44 | 1090 | |
9e85d811 | 1091 | radix_tree_load_root(root, &child, &maxindex); |
070c5ac2 | 1092 | last_index = min(last_index, maxindex); |
ebf8aa44 JK |
1093 | if (index > last_index) |
1094 | return 0; | |
1095 | if (!nr_to_tag) | |
1096 | return 0; | |
1097 | if (!root_tag_get(root, iftag)) { | |
1098 | *first_indexp = last_index + 1; | |
1099 | return 0; | |
1100 | } | |
a8e4da25 | 1101 | if (!radix_tree_is_internal_node(child)) { |
ebf8aa44 JK |
1102 | *first_indexp = last_index + 1; |
1103 | root_tag_set(root, settag); | |
1104 | return 1; | |
1105 | } | |
1106 | ||
a8e4da25 | 1107 | node = entry_to_node(child); |
ebf8aa44 JK |
1108 | |
1109 | for (;;) { | |
9e85d811 | 1110 | unsigned offset = radix_tree_descend(node, &child, index); |
a8e4da25 | 1111 | if (!child) |
ebf8aa44 | 1112 | goto next; |
070c5ac2 | 1113 | if (!tag_get(node, iftag, offset)) |
ebf8aa44 | 1114 | goto next; |
070c5ac2 | 1115 | /* Sibling slots never have tags set on them */ |
a8e4da25 MW |
1116 | if (radix_tree_is_internal_node(child)) { |
1117 | node = entry_to_node(child); | |
070c5ac2 | 1118 | continue; |
144dcfc0 DC |
1119 | } |
1120 | ||
1121 | /* tag the leaf */ | |
070c5ac2 MW |
1122 | tagged++; |
1123 | tag_set(node, settag, offset); | |
144dcfc0 DC |
1124 | |
1125 | /* walk back up the path tagging interior nodes */ | |
a8e4da25 MW |
1126 | parent = node; |
1127 | for (;;) { | |
1128 | offset = parent->offset; | |
1129 | parent = parent->parent; | |
1130 | if (!parent) | |
1131 | break; | |
144dcfc0 | 1132 | /* stop if we find a node with the tag already set */ |
a8e4da25 | 1133 | if (tag_get(parent, settag, offset)) |
144dcfc0 | 1134 | break; |
a8e4da25 | 1135 | tag_set(parent, settag, offset); |
ebf8aa44 | 1136 | } |
070c5ac2 | 1137 | next: |
9e85d811 MW |
1138 | /* Go to next entry in node */ |
1139 | index = ((index >> node->shift) + 1) << node->shift; | |
d5ed3a4a JK |
1140 | /* Overflow can happen when last_index is ~0UL... */ |
1141 | if (index > last_index || !index) | |
ebf8aa44 | 1142 | break; |
9e85d811 | 1143 | offset = (index >> node->shift) & RADIX_TREE_MAP_MASK; |
070c5ac2 | 1144 | while (offset == 0) { |
ebf8aa44 JK |
1145 | /* |
1146 | * We've fully scanned this node. Go up. Because | |
1147 | * last_index is guaranteed to be in the tree, what | |
1148 | * we do below cannot wander astray. | |
1149 | */ | |
070c5ac2 | 1150 | node = node->parent; |
9e85d811 | 1151 | offset = (index >> node->shift) & RADIX_TREE_MAP_MASK; |
ebf8aa44 | 1152 | } |
070c5ac2 MW |
1153 | if (is_sibling_entry(node, node->slots[offset])) |
1154 | goto next; | |
1155 | if (tagged >= nr_to_tag) | |
1156 | break; | |
ebf8aa44 JK |
1157 | } |
1158 | /* | |
ac15ee69 TO |
1159 | * We need not to tag the root tag if there is no tag which is set with |
1160 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1161 | */ |
ac15ee69 TO |
1162 | if (tagged > 0) |
1163 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1164 | *first_indexp = index; |
1165 | ||
1166 | return tagged; | |
1167 | } | |
1168 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1169 | ||
1da177e4 LT |
1170 | /** |
1171 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1172 | * @root: radix tree root | |
1173 | * @results: where the results of the lookup are placed | |
1174 | * @first_index: start the lookup from this key | |
1175 | * @max_items: place up to this many items at *results | |
1176 | * | |
1177 | * Performs an index-ascending scan of the tree for present items. Places | |
1178 | * them at *@results and returns the number of items which were placed at | |
1179 | * *@results. | |
1180 | * | |
1181 | * The implementation is naive. | |
7cf9c2c7 NP |
1182 | * |
1183 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1184 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1185 | * an atomic snapshot of the tree at a single point in time, the |
1186 | * semantics of an RCU protected gang lookup are as though multiple | |
1187 | * radix_tree_lookups have been issued in individual locks, and results | |
1188 | * stored in 'results'. | |
1da177e4 LT |
1189 | */ |
1190 | unsigned int | |
1191 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1192 | unsigned long first_index, unsigned int max_items) | |
1193 | { | |
cebbd29e KK |
1194 | struct radix_tree_iter iter; |
1195 | void **slot; | |
1196 | unsigned int ret = 0; | |
7cf9c2c7 | 1197 | |
cebbd29e | 1198 | if (unlikely(!max_items)) |
7cf9c2c7 | 1199 | return 0; |
1da177e4 | 1200 | |
cebbd29e | 1201 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1202 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1203 | if (!results[ret]) |
1204 | continue; | |
b194d16c | 1205 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1206 | slot = radix_tree_iter_retry(&iter); |
1207 | continue; | |
1208 | } | |
cebbd29e | 1209 | if (++ret == max_items) |
1da177e4 | 1210 | break; |
1da177e4 | 1211 | } |
7cf9c2c7 | 1212 | |
1da177e4 LT |
1213 | return ret; |
1214 | } | |
1215 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1216 | ||
47feff2c NP |
1217 | /** |
1218 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1219 | * @root: radix tree root | |
1220 | * @results: where the results of the lookup are placed | |
6328650b | 1221 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1222 | * @first_index: start the lookup from this key |
1223 | * @max_items: place up to this many items at *results | |
1224 | * | |
1225 | * Performs an index-ascending scan of the tree for present items. Places | |
1226 | * their slots at *@results and returns the number of items which were | |
1227 | * placed at *@results. | |
1228 | * | |
1229 | * The implementation is naive. | |
1230 | * | |
1231 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1232 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1233 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1234 | */ | |
1235 | unsigned int | |
6328650b HD |
1236 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1237 | void ***results, unsigned long *indices, | |
47feff2c NP |
1238 | unsigned long first_index, unsigned int max_items) |
1239 | { | |
cebbd29e KK |
1240 | struct radix_tree_iter iter; |
1241 | void **slot; | |
1242 | unsigned int ret = 0; | |
47feff2c | 1243 | |
cebbd29e | 1244 | if (unlikely(!max_items)) |
47feff2c NP |
1245 | return 0; |
1246 | ||
cebbd29e KK |
1247 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1248 | results[ret] = slot; | |
6328650b | 1249 | if (indices) |
cebbd29e KK |
1250 | indices[ret] = iter.index; |
1251 | if (++ret == max_items) | |
47feff2c | 1252 | break; |
47feff2c NP |
1253 | } |
1254 | ||
1255 | return ret; | |
1256 | } | |
1257 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1258 | ||
1da177e4 LT |
1259 | /** |
1260 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1261 | * based on a tag | |
1262 | * @root: radix tree root | |
1263 | * @results: where the results of the lookup are placed | |
1264 | * @first_index: start the lookup from this key | |
1265 | * @max_items: place up to this many items at *results | |
daff89f3 | 1266 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1267 | * |
1268 | * Performs an index-ascending scan of the tree for present items which | |
1269 | * have the tag indexed by @tag set. Places the items at *@results and | |
1270 | * returns the number of items which were placed at *@results. | |
1271 | */ | |
1272 | unsigned int | |
1273 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1274 | unsigned long first_index, unsigned int max_items, |
1275 | unsigned int tag) | |
1da177e4 | 1276 | { |
cebbd29e KK |
1277 | struct radix_tree_iter iter; |
1278 | void **slot; | |
1279 | unsigned int ret = 0; | |
612d6c19 | 1280 | |
cebbd29e | 1281 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1282 | return 0; |
1283 | ||
cebbd29e | 1284 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1285 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1286 | if (!results[ret]) |
1287 | continue; | |
b194d16c | 1288 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1289 | slot = radix_tree_iter_retry(&iter); |
1290 | continue; | |
1291 | } | |
cebbd29e | 1292 | if (++ret == max_items) |
1da177e4 | 1293 | break; |
1da177e4 | 1294 | } |
7cf9c2c7 | 1295 | |
1da177e4 LT |
1296 | return ret; |
1297 | } | |
1298 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1299 | ||
47feff2c NP |
1300 | /** |
1301 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1302 | * radix tree based on a tag | |
1303 | * @root: radix tree root | |
1304 | * @results: where the results of the lookup are placed | |
1305 | * @first_index: start the lookup from this key | |
1306 | * @max_items: place up to this many items at *results | |
1307 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1308 | * | |
1309 | * Performs an index-ascending scan of the tree for present items which | |
1310 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1311 | * returns the number of slots which were placed at *@results. | |
1312 | */ | |
1313 | unsigned int | |
1314 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1315 | unsigned long first_index, unsigned int max_items, | |
1316 | unsigned int tag) | |
1317 | { | |
cebbd29e KK |
1318 | struct radix_tree_iter iter; |
1319 | void **slot; | |
1320 | unsigned int ret = 0; | |
47feff2c | 1321 | |
cebbd29e | 1322 | if (unlikely(!max_items)) |
47feff2c NP |
1323 | return 0; |
1324 | ||
cebbd29e KK |
1325 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1326 | results[ret] = slot; | |
1327 | if (++ret == max_items) | |
47feff2c | 1328 | break; |
47feff2c NP |
1329 | } |
1330 | ||
1331 | return ret; | |
1332 | } | |
1333 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1334 | ||
e504f3fd HD |
1335 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1336 | #include <linux/sched.h> /* for cond_resched() */ | |
1337 | ||
0a2efc6c MW |
1338 | struct locate_info { |
1339 | unsigned long found_index; | |
1340 | bool stop; | |
1341 | }; | |
1342 | ||
e504f3fd HD |
1343 | /* |
1344 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1345 | */ | |
1346 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
0a2efc6c | 1347 | unsigned long index, struct locate_info *info) |
e504f3fd | 1348 | { |
e504f3fd HD |
1349 | unsigned long i; |
1350 | ||
0a2efc6c | 1351 | do { |
9e85d811 | 1352 | unsigned int shift = slot->shift; |
e504f3fd | 1353 | |
0a2efc6c MW |
1354 | for (i = (index >> shift) & RADIX_TREE_MAP_MASK; |
1355 | i < RADIX_TREE_MAP_SIZE; | |
1356 | i++, index += (1UL << shift)) { | |
1357 | struct radix_tree_node *node = | |
1358 | rcu_dereference_raw(slot->slots[i]); | |
1359 | if (node == RADIX_TREE_RETRY) | |
1360 | goto out; | |
b194d16c | 1361 | if (!radix_tree_is_internal_node(node)) { |
0a2efc6c MW |
1362 | if (node == item) { |
1363 | info->found_index = index; | |
1364 | info->stop = true; | |
1365 | goto out; | |
1366 | } | |
1367 | continue; | |
e6145236 | 1368 | } |
4dd6c098 | 1369 | node = entry_to_node(node); |
0a2efc6c MW |
1370 | if (is_sibling_entry(slot, node)) |
1371 | continue; | |
1372 | slot = node; | |
1373 | break; | |
e6145236 | 1374 | } |
9e85d811 | 1375 | } while (i < RADIX_TREE_MAP_SIZE); |
e504f3fd | 1376 | |
e504f3fd | 1377 | out: |
0a2efc6c MW |
1378 | if ((index == 0) && (i == RADIX_TREE_MAP_SIZE)) |
1379 | info->stop = true; | |
e504f3fd HD |
1380 | return index; |
1381 | } | |
1382 | ||
1383 | /** | |
1384 | * radix_tree_locate_item - search through radix tree for item | |
1385 | * @root: radix tree root | |
1386 | * @item: item to be found | |
1387 | * | |
1388 | * Returns index where item was found, or -1 if not found. | |
1389 | * Caller must hold no lock (since this time-consuming function needs | |
1390 | * to be preemptible), and must check afterwards if item is still there. | |
1391 | */ | |
1392 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1393 | { | |
1394 | struct radix_tree_node *node; | |
1395 | unsigned long max_index; | |
1396 | unsigned long cur_index = 0; | |
0a2efc6c MW |
1397 | struct locate_info info = { |
1398 | .found_index = -1, | |
1399 | .stop = false, | |
1400 | }; | |
e504f3fd HD |
1401 | |
1402 | do { | |
1403 | rcu_read_lock(); | |
1404 | node = rcu_dereference_raw(root->rnode); | |
b194d16c | 1405 | if (!radix_tree_is_internal_node(node)) { |
e504f3fd HD |
1406 | rcu_read_unlock(); |
1407 | if (node == item) | |
0a2efc6c | 1408 | info.found_index = 0; |
e504f3fd HD |
1409 | break; |
1410 | } | |
1411 | ||
4dd6c098 | 1412 | node = entry_to_node(node); |
0a2efc6c MW |
1413 | |
1414 | max_index = node_maxindex(node); | |
5f30fc94 HD |
1415 | if (cur_index > max_index) { |
1416 | rcu_read_unlock(); | |
e504f3fd | 1417 | break; |
5f30fc94 | 1418 | } |
e504f3fd | 1419 | |
0a2efc6c | 1420 | cur_index = __locate(node, item, cur_index, &info); |
e504f3fd HD |
1421 | rcu_read_unlock(); |
1422 | cond_resched(); | |
0a2efc6c | 1423 | } while (!info.stop && cur_index <= max_index); |
e504f3fd | 1424 | |
0a2efc6c | 1425 | return info.found_index; |
e504f3fd HD |
1426 | } |
1427 | #else | |
1428 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1429 | { | |
1430 | return -1; | |
1431 | } | |
1432 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1433 | |
a5f51c96 | 1434 | /** |
d0891265 | 1435 | * radix_tree_shrink - shrink radix tree to minimum height |
a5f51c96 NP |
1436 | * @root radix tree root |
1437 | */ | |
fb209019 | 1438 | static inline bool radix_tree_shrink(struct radix_tree_root *root) |
a5f51c96 | 1439 | { |
fb209019 MW |
1440 | bool shrunk = false; |
1441 | ||
d0891265 | 1442 | for (;;) { |
af49a63e MW |
1443 | struct radix_tree_node *node = root->rnode; |
1444 | struct radix_tree_node *child; | |
a5f51c96 | 1445 | |
af49a63e | 1446 | if (!radix_tree_is_internal_node(node)) |
d0891265 | 1447 | break; |
af49a63e | 1448 | node = entry_to_node(node); |
c0bc9875 NP |
1449 | |
1450 | /* | |
1451 | * The candidate node has more than one child, or its child | |
d0891265 MW |
1452 | * is not at the leftmost slot, or the child is a multiorder |
1453 | * entry, we cannot shrink. | |
c0bc9875 | 1454 | */ |
af49a63e | 1455 | if (node->count != 1) |
c0bc9875 | 1456 | break; |
af49a63e MW |
1457 | child = node->slots[0]; |
1458 | if (!child) | |
c0bc9875 | 1459 | break; |
af49a63e | 1460 | if (!radix_tree_is_internal_node(child) && node->shift) |
afe0e395 MW |
1461 | break; |
1462 | ||
af49a63e MW |
1463 | if (radix_tree_is_internal_node(child)) |
1464 | entry_to_node(child)->parent = NULL; | |
c0bc9875 | 1465 | |
7cf9c2c7 NP |
1466 | /* |
1467 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1468 | * moving the node from one part of the tree to another: if it |
1469 | * was safe to dereference the old pointer to it | |
af49a63e | 1470 | * (node->slots[0]), it will be safe to dereference the new |
27d20fdd | 1471 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1472 | */ |
af49a63e | 1473 | root->rnode = child; |
27d20fdd NP |
1474 | |
1475 | /* | |
1476 | * We have a dilemma here. The node's slot[0] must not be | |
1477 | * NULLed in case there are concurrent lookups expecting to | |
1478 | * find the item. However if this was a bottom-level node, | |
1479 | * then it may be subject to the slot pointer being visible | |
1480 | * to callers dereferencing it. If item corresponding to | |
1481 | * slot[0] is subsequently deleted, these callers would expect | |
1482 | * their slot to become empty sooner or later. | |
1483 | * | |
1484 | * For example, lockless pagecache will look up a slot, deref | |
2fcd9005 | 1485 | * the page pointer, and if the page has 0 refcount it means it |
27d20fdd NP |
1486 | * was concurrently deleted from pagecache so try the deref |
1487 | * again. Fortunately there is already a requirement for logic | |
1488 | * to retry the entire slot lookup -- the indirect pointer | |
1489 | * problem (replacing direct root node with an indirect pointer | |
1490 | * also results in a stale slot). So tag the slot as indirect | |
1491 | * to force callers to retry. | |
1492 | */ | |
af49a63e MW |
1493 | if (!radix_tree_is_internal_node(child)) |
1494 | node->slots[0] = RADIX_TREE_RETRY; | |
27d20fdd | 1495 | |
af49a63e | 1496 | radix_tree_node_free(node); |
fb209019 | 1497 | shrunk = true; |
a5f51c96 | 1498 | } |
fb209019 MW |
1499 | |
1500 | return shrunk; | |
a5f51c96 NP |
1501 | } |
1502 | ||
139e5616 JW |
1503 | /** |
1504 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1505 | * @root: radix tree root | |
139e5616 JW |
1506 | * @node: node containing @index |
1507 | * | |
1508 | * After clearing the slot at @index in @node from radix tree | |
1509 | * rooted at @root, call this function to attempt freeing the | |
1510 | * node and shrinking the tree. | |
1511 | * | |
1512 | * Returns %true if @node was freed, %false otherwise. | |
1513 | */ | |
449dd698 | 1514 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1515 | struct radix_tree_node *node) |
1516 | { | |
1517 | bool deleted = false; | |
1518 | ||
1519 | do { | |
1520 | struct radix_tree_node *parent; | |
1521 | ||
1522 | if (node->count) { | |
4dd6c098 | 1523 | if (node == entry_to_node(root->rnode)) |
fb209019 | 1524 | deleted |= radix_tree_shrink(root); |
139e5616 JW |
1525 | return deleted; |
1526 | } | |
1527 | ||
1528 | parent = node->parent; | |
1529 | if (parent) { | |
0c7fa0a8 | 1530 | parent->slots[node->offset] = NULL; |
139e5616 JW |
1531 | parent->count--; |
1532 | } else { | |
1533 | root_tag_clear_all(root); | |
139e5616 JW |
1534 | root->rnode = NULL; |
1535 | } | |
1536 | ||
1537 | radix_tree_node_free(node); | |
1538 | deleted = true; | |
1539 | ||
1540 | node = parent; | |
1541 | } while (node); | |
1542 | ||
1543 | return deleted; | |
1544 | } | |
1545 | ||
57578c2e MW |
1546 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1547 | void *ptr, unsigned offset) | |
1548 | { | |
1549 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1550 | int i; | |
1551 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1552 | if (node->slots[offset + i] != ptr) | |
1553 | break; | |
1554 | node->slots[offset + i] = NULL; | |
1555 | node->count--; | |
1556 | } | |
1557 | #endif | |
1558 | } | |
1559 | ||
1da177e4 | 1560 | /** |
53c59f26 | 1561 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1562 | * @root: radix tree root |
1563 | * @index: index key | |
53c59f26 | 1564 | * @item: expected item |
1da177e4 | 1565 | * |
53c59f26 | 1566 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1567 | * |
53c59f26 JW |
1568 | * Returns the address of the deleted item, or NULL if it was not present |
1569 | * or the entry at the given @index was not @item. | |
1da177e4 | 1570 | */ |
53c59f26 JW |
1571 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1572 | unsigned long index, void *item) | |
1da177e4 | 1573 | { |
139e5616 | 1574 | struct radix_tree_node *node; |
57578c2e | 1575 | unsigned int offset; |
139e5616 JW |
1576 | void **slot; |
1577 | void *entry; | |
d5274261 | 1578 | int tag; |
1da177e4 | 1579 | |
139e5616 JW |
1580 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1581 | if (!entry) | |
1582 | return NULL; | |
1da177e4 | 1583 | |
139e5616 JW |
1584 | if (item && entry != item) |
1585 | return NULL; | |
1586 | ||
1587 | if (!node) { | |
612d6c19 NP |
1588 | root_tag_clear_all(root); |
1589 | root->rnode = NULL; | |
139e5616 | 1590 | return entry; |
612d6c19 | 1591 | } |
1da177e4 | 1592 | |
29e0967c | 1593 | offset = get_slot_offset(node, slot); |
53c59f26 | 1594 | |
d604c324 MW |
1595 | /* Clear all tags associated with the item to be deleted. */ |
1596 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1597 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1598 | |
a4db4dce | 1599 | delete_sibling_entries(node, node_to_entry(slot), offset); |
139e5616 JW |
1600 | node->slots[offset] = NULL; |
1601 | node->count--; | |
f7942430 JW |
1602 | if (radix_tree_exceptional_entry(entry)) |
1603 | node->exceptional--; | |
e2bdb933 | 1604 | |
449dd698 | 1605 | __radix_tree_delete_node(root, node); |
612d6c19 | 1606 | |
139e5616 | 1607 | return entry; |
1da177e4 | 1608 | } |
53c59f26 JW |
1609 | EXPORT_SYMBOL(radix_tree_delete_item); |
1610 | ||
1611 | /** | |
1612 | * radix_tree_delete - delete an item from a radix tree | |
1613 | * @root: radix tree root | |
1614 | * @index: index key | |
1615 | * | |
1616 | * Remove the item at @index from the radix tree rooted at @root. | |
1617 | * | |
1618 | * Returns the address of the deleted item, or NULL if it was not present. | |
1619 | */ | |
1620 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1621 | { | |
1622 | return radix_tree_delete_item(root, index, NULL); | |
1623 | } | |
1da177e4 LT |
1624 | EXPORT_SYMBOL(radix_tree_delete); |
1625 | ||
d3798ae8 JW |
1626 | void radix_tree_clear_tags(struct radix_tree_root *root, |
1627 | struct radix_tree_node *node, | |
1628 | void **slot) | |
d604c324 | 1629 | { |
d604c324 MW |
1630 | if (node) { |
1631 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1632 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1633 | node_tag_clear(root, node, tag, offset); | |
1634 | } else { | |
1635 | /* Clear root node tags */ | |
1636 | root->gfp_mask &= __GFP_BITS_MASK; | |
1637 | } | |
d604c324 MW |
1638 | } |
1639 | ||
1da177e4 LT |
1640 | /** |
1641 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1642 | * @root: radix tree root | |
1643 | * @tag: tag to test | |
1644 | */ | |
daff89f3 | 1645 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1646 | { |
612d6c19 | 1647 | return root_tag_get(root, tag); |
1da177e4 LT |
1648 | } |
1649 | EXPORT_SYMBOL(radix_tree_tagged); | |
1650 | ||
1651 | static void | |
449dd698 | 1652 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1653 | { |
449dd698 JW |
1654 | struct radix_tree_node *node = arg; |
1655 | ||
1656 | memset(node, 0, sizeof(*node)); | |
1657 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1658 | } |
1659 | ||
c78c66d1 KS |
1660 | static __init unsigned long __maxindex(unsigned int height) |
1661 | { | |
1662 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
1663 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1664 | ||
1665 | if (shift < 0) | |
1666 | return ~0UL; | |
1667 | if (shift >= BITS_PER_LONG) | |
1668 | return 0UL; | |
1669 | return ~0UL >> shift; | |
1670 | } | |
1671 | ||
1672 | static __init void radix_tree_init_maxnodes(void) | |
1673 | { | |
1674 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
1675 | unsigned int i, j; | |
1676 | ||
1677 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1678 | height_to_maxindex[i] = __maxindex(i); | |
1679 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
1680 | for (j = i; j > 0; j--) | |
1681 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
1682 | } | |
1683 | } | |
1684 | ||
1da177e4 | 1685 | static int radix_tree_callback(struct notifier_block *nfb, |
2fcd9005 | 1686 | unsigned long action, void *hcpu) |
1da177e4 | 1687 | { |
2fcd9005 MW |
1688 | int cpu = (long)hcpu; |
1689 | struct radix_tree_preload *rtp; | |
1690 | struct radix_tree_node *node; | |
1691 | ||
1692 | /* Free per-cpu pool of preloaded nodes */ | |
1693 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
1694 | rtp = &per_cpu(radix_tree_preloads, cpu); | |
1695 | while (rtp->nr) { | |
9d2a8da0 KS |
1696 | node = rtp->nodes; |
1697 | rtp->nodes = node->private_data; | |
1698 | kmem_cache_free(radix_tree_node_cachep, node); | |
1699 | rtp->nr--; | |
2fcd9005 MW |
1700 | } |
1701 | } | |
1702 | return NOTIFY_OK; | |
1da177e4 | 1703 | } |
1da177e4 LT |
1704 | |
1705 | void __init radix_tree_init(void) | |
1706 | { | |
1707 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1708 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1709 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1710 | radix_tree_node_ctor); | |
c78c66d1 | 1711 | radix_tree_init_maxnodes(); |
1da177e4 LT |
1712 | hotcpu_notifier(radix_tree_callback, 0); |
1713 | } |