]>
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 |
1da177e4 LT |
7 | * |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License as | |
10 | * published by the Free Software Foundation; either version 2, or (at | |
11 | * your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, but | |
14 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
21 | */ | |
22 | ||
23 | #include <linux/errno.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/kernel.h> | |
8bc3bcc9 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/radix-tree.h> |
28 | #include <linux/percpu.h> | |
29 | #include <linux/slab.h> | |
ce80b067 | 30 | #include <linux/kmemleak.h> |
1da177e4 LT |
31 | #include <linux/notifier.h> |
32 | #include <linux/cpu.h> | |
1da177e4 LT |
33 | #include <linux/string.h> |
34 | #include <linux/bitops.h> | |
7cf9c2c7 | 35 | #include <linux/rcupdate.h> |
92cf2118 | 36 | #include <linux/preempt.h> /* in_interrupt() */ |
1da177e4 LT |
37 | |
38 | ||
26fb1589 JM |
39 | /* |
40 | * The height_to_maxindex array needs to be one deeper than the maximum | |
41 | * path as height 0 holds only 1 entry. | |
42 | */ | |
43 | static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
1da177e4 LT |
44 | |
45 | /* | |
46 | * Radix tree node cache. | |
47 | */ | |
e18b890b | 48 | static struct kmem_cache *radix_tree_node_cachep; |
1da177e4 | 49 | |
55368052 NP |
50 | /* |
51 | * The radix tree is variable-height, so an insert operation not only has | |
52 | * to build the branch to its corresponding item, it also has to build the | |
53 | * branch to existing items if the size has to be increased (by | |
54 | * radix_tree_extend). | |
55 | * | |
56 | * The worst case is a zero height tree with just a single item at index 0, | |
57 | * and then inserting an item at index ULONG_MAX. This requires 2 new branches | |
58 | * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared. | |
59 | * Hence: | |
60 | */ | |
61 | #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1) | |
62 | ||
1da177e4 LT |
63 | /* |
64 | * Per-cpu pool of preloaded nodes | |
65 | */ | |
66 | struct radix_tree_preload { | |
67 | int nr; | |
9d2a8da0 KS |
68 | /* nodes->private_data points to next preallocated node */ |
69 | struct radix_tree_node *nodes; | |
1da177e4 | 70 | }; |
8cef7d57 | 71 | static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, }; |
1da177e4 | 72 | |
27d20fdd NP |
73 | static inline void *ptr_to_indirect(void *ptr) |
74 | { | |
75 | return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR); | |
76 | } | |
77 | ||
afe0e395 MW |
78 | #define RADIX_TREE_RETRY ptr_to_indirect(NULL) |
79 | ||
db050f29 MW |
80 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
81 | /* Sibling slots point directly to another slot in the same node */ | |
82 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
83 | { | |
84 | void **ptr = node; | |
85 | return (parent->slots <= ptr) && | |
86 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
87 | } | |
88 | #else | |
89 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
90 | { | |
91 | return false; | |
92 | } | |
93 | #endif | |
94 | ||
95 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
96 | void **slot) | |
97 | { | |
98 | return slot - parent->slots; | |
99 | } | |
100 | ||
101 | static unsigned radix_tree_descend(struct radix_tree_node *parent, | |
102 | struct radix_tree_node **nodep, unsigned offset) | |
103 | { | |
104 | void **entry = rcu_dereference_raw(parent->slots[offset]); | |
105 | ||
106 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
107 | if (radix_tree_is_indirect_ptr(entry)) { | |
108 | unsigned long siboff = get_slot_offset(parent, entry); | |
109 | if (siboff < RADIX_TREE_MAP_SIZE) { | |
110 | offset = siboff; | |
111 | entry = rcu_dereference_raw(parent->slots[offset]); | |
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 | ||
148 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
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 | { | |
160 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
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 | { | |
174 | int idx; | |
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 RZ |
218 | #ifndef __KERNEL__ |
219 | static void dump_node(struct radix_tree_node *node, unsigned offset, | |
220 | unsigned shift, unsigned long index) | |
7cf19af4 | 221 | { |
0796c583 | 222 | unsigned long i; |
7cf19af4 | 223 | |
7cf19af4 | 224 | pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n", |
0796c583 RZ |
225 | node, offset, |
226 | node->tags[0][0], node->tags[1][0], node->tags[2][0], | |
227 | node->path, node->count, node->parent); | |
228 | ||
229 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
230 | unsigned long first = index | (i << shift); | |
231 | unsigned long last = first | ((1UL << shift) - 1); | |
232 | void *entry = node->slots[i]; | |
233 | if (!entry) | |
234 | continue; | |
235 | if (is_sibling_entry(node, entry)) { | |
236 | pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n", | |
237 | entry, i, | |
238 | *(void **)indirect_to_ptr(entry), | |
239 | first, last); | |
240 | } else if (!radix_tree_is_indirect_ptr(entry)) { | |
241 | pr_debug("radix entry %p offset %ld indices %ld-%ld\n", | |
242 | entry, i, first, last); | |
243 | } else { | |
244 | dump_node(indirect_to_ptr(entry), i, | |
245 | shift - RADIX_TREE_MAP_SHIFT, first); | |
246 | } | |
247 | } | |
7cf19af4 MW |
248 | } |
249 | ||
250 | /* For debug */ | |
251 | static void radix_tree_dump(struct radix_tree_root *root) | |
252 | { | |
253 | pr_debug("radix root: %p height %d rnode %p tags %x\n", | |
254 | root, root->height, root->rnode, | |
255 | root->gfp_mask >> __GFP_BITS_SHIFT); | |
256 | if (!radix_tree_is_indirect_ptr(root->rnode)) | |
257 | return; | |
0796c583 RZ |
258 | dump_node(indirect_to_ptr(root->rnode), 0, |
259 | (root->height - 1) * RADIX_TREE_MAP_SHIFT, 0); | |
7cf19af4 MW |
260 | } |
261 | #endif | |
262 | ||
1da177e4 LT |
263 | /* |
264 | * This assumes that the caller has performed appropriate preallocation, and | |
265 | * that the caller has pinned this thread of control to the current CPU. | |
266 | */ | |
267 | static struct radix_tree_node * | |
268 | radix_tree_node_alloc(struct radix_tree_root *root) | |
269 | { | |
e2848a0e | 270 | struct radix_tree_node *ret = NULL; |
612d6c19 | 271 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 272 | |
5e4c0d97 JK |
273 | /* |
274 | * Preload code isn't irq safe and it doesn't make sence to use | |
275 | * preloading in the interrupt anyway as all the allocations have to | |
276 | * be atomic. So just do normal allocation when in interrupt. | |
277 | */ | |
d0164adc | 278 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
279 | struct radix_tree_preload *rtp; |
280 | ||
58e698af VD |
281 | /* |
282 | * Even if the caller has preloaded, try to allocate from the | |
283 | * cache first for the new node to get accounted. | |
284 | */ | |
285 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
286 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
287 | if (ret) | |
288 | goto out; | |
289 | ||
e2848a0e NP |
290 | /* |
291 | * Provided the caller has preloaded here, we will always | |
292 | * succeed in getting a node here (and never reach | |
293 | * kmem_cache_alloc) | |
294 | */ | |
7c8e0181 | 295 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 296 | if (rtp->nr) { |
9d2a8da0 KS |
297 | ret = rtp->nodes; |
298 | rtp->nodes = ret->private_data; | |
299 | ret->private_data = NULL; | |
1da177e4 LT |
300 | rtp->nr--; |
301 | } | |
ce80b067 CM |
302 | /* |
303 | * Update the allocation stack trace as this is more useful | |
304 | * for debugging. | |
305 | */ | |
306 | kmemleak_update_trace(ret); | |
58e698af | 307 | goto out; |
1da177e4 | 308 | } |
58e698af VD |
309 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
310 | gfp_mask | __GFP_ACCOUNT); | |
311 | out: | |
c0bc9875 | 312 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
313 | return ret; |
314 | } | |
315 | ||
7cf9c2c7 NP |
316 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
317 | { | |
318 | struct radix_tree_node *node = | |
319 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 320 | int i; |
643b52b9 NP |
321 | |
322 | /* | |
323 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
324 | * can leave us with a non-NULL entry in the first slot, so clear | |
325 | * that here to make sure. | |
326 | */ | |
b6dd0865 DC |
327 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
328 | tag_clear(node, i, 0); | |
329 | ||
643b52b9 NP |
330 | node->slots[0] = NULL; |
331 | node->count = 0; | |
332 | ||
7cf9c2c7 NP |
333 | kmem_cache_free(radix_tree_node_cachep, node); |
334 | } | |
335 | ||
1da177e4 LT |
336 | static inline void |
337 | radix_tree_node_free(struct radix_tree_node *node) | |
338 | { | |
7cf9c2c7 | 339 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
340 | } |
341 | ||
342 | /* | |
343 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
344 | * ensure that the addition of a single element in the tree cannot fail. On | |
345 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
346 | * with preemption not disabled. | |
b34df792 DH |
347 | * |
348 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 349 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 350 | */ |
5e4c0d97 | 351 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
352 | { |
353 | struct radix_tree_preload *rtp; | |
354 | struct radix_tree_node *node; | |
355 | int ret = -ENOMEM; | |
356 | ||
357 | preempt_disable(); | |
7c8e0181 | 358 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 359 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 360 | preempt_enable(); |
488514d1 | 361 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
362 | if (node == NULL) |
363 | goto out; | |
364 | preempt_disable(); | |
7c8e0181 | 365 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
366 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
367 | node->private_data = rtp->nodes; | |
368 | rtp->nodes = node; | |
369 | rtp->nr++; | |
370 | } else { | |
1da177e4 | 371 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 372 | } |
1da177e4 LT |
373 | } |
374 | ret = 0; | |
375 | out: | |
376 | return ret; | |
377 | } | |
5e4c0d97 JK |
378 | |
379 | /* | |
380 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
381 | * ensure that the addition of a single element in the tree cannot fail. On | |
382 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
383 | * with preemption not disabled. | |
384 | * | |
385 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 386 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
387 | */ |
388 | int radix_tree_preload(gfp_t gfp_mask) | |
389 | { | |
390 | /* Warn on non-sensical use... */ | |
d0164adc | 391 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
392 | return __radix_tree_preload(gfp_mask); |
393 | } | |
d7f0923d | 394 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 395 | |
5e4c0d97 JK |
396 | /* |
397 | * The same as above function, except we don't guarantee preloading happens. | |
398 | * We do it, if we decide it helps. On success, return zero with preemption | |
399 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
400 | */ | |
401 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
402 | { | |
d0164adc | 403 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
404 | return __radix_tree_preload(gfp_mask); |
405 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
406 | preempt_disable(); | |
407 | return 0; | |
408 | } | |
409 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
410 | ||
1da177e4 LT |
411 | /* |
412 | * Return the maximum key which can be store into a | |
413 | * radix tree with height HEIGHT. | |
414 | */ | |
415 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
416 | { | |
417 | return height_to_maxindex[height]; | |
418 | } | |
419 | ||
1456a439 MW |
420 | static inline unsigned long node_maxindex(struct radix_tree_node *node) |
421 | { | |
422 | return radix_tree_maxindex(node->path & RADIX_TREE_HEIGHT_MASK); | |
423 | } | |
424 | ||
425 | static unsigned radix_tree_load_root(struct radix_tree_root *root, | |
426 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
427 | { | |
428 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
429 | ||
430 | *nodep = node; | |
431 | ||
432 | if (likely(radix_tree_is_indirect_ptr(node))) { | |
433 | node = indirect_to_ptr(node); | |
434 | *maxindex = node_maxindex(node); | |
435 | return (node->path & RADIX_TREE_HEIGHT_MASK) * | |
436 | RADIX_TREE_MAP_SHIFT; | |
437 | } | |
438 | ||
439 | *maxindex = 0; | |
440 | return 0; | |
441 | } | |
442 | ||
1da177e4 LT |
443 | /* |
444 | * Extend a radix tree so it can store key @index. | |
445 | */ | |
e6145236 | 446 | static int radix_tree_extend(struct radix_tree_root *root, |
49ea6ebc | 447 | unsigned long index) |
1da177e4 LT |
448 | { |
449 | struct radix_tree_node *node; | |
e2bdb933 | 450 | struct radix_tree_node *slot; |
1da177e4 | 451 | unsigned int height; |
1da177e4 LT |
452 | int tag; |
453 | ||
454 | /* Figure out what the height should be. */ | |
455 | height = root->height + 1; | |
456 | while (index > radix_tree_maxindex(height)) | |
457 | height++; | |
458 | ||
49ea6ebc | 459 | if (root->rnode == NULL) { |
1da177e4 LT |
460 | root->height = height; |
461 | goto out; | |
462 | } | |
463 | ||
1da177e4 | 464 | do { |
7cf9c2c7 | 465 | unsigned int newheight; |
1da177e4 LT |
466 | if (!(node = radix_tree_node_alloc(root))) |
467 | return -ENOMEM; | |
468 | ||
1da177e4 | 469 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 470 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 471 | if (root_tag_get(root, tag)) |
1da177e4 LT |
472 | tag_set(node, tag, 0); |
473 | } | |
474 | ||
e2bdb933 | 475 | /* Increase the height. */ |
7cf9c2c7 | 476 | newheight = root->height+1; |
449dd698 JW |
477 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
478 | node->path = newheight; | |
1da177e4 | 479 | node->count = 1; |
e2bdb933 HD |
480 | node->parent = NULL; |
481 | slot = root->rnode; | |
49ea6ebc | 482 | if (radix_tree_is_indirect_ptr(slot)) { |
e2bdb933 HD |
483 | slot = indirect_to_ptr(slot); |
484 | slot->parent = node; | |
339e6353 | 485 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
486 | } |
487 | node->slots[0] = slot; | |
27d20fdd | 488 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
489 | rcu_assign_pointer(root->rnode, node); |
490 | root->height = newheight; | |
1da177e4 LT |
491 | } while (height > root->height); |
492 | out: | |
49ea6ebc | 493 | return height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
494 | } |
495 | ||
496 | /** | |
139e5616 | 497 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
498 | * @root: radix tree root |
499 | * @index: index key | |
e6145236 | 500 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
501 | * @nodep: returns node |
502 | * @slotp: returns slot | |
1da177e4 | 503 | * |
139e5616 JW |
504 | * Create, if necessary, and return the node and slot for an item |
505 | * at position @index in the radix tree @root. | |
506 | * | |
507 | * Until there is more than one item in the tree, no nodes are | |
508 | * allocated and @root->rnode is used as a direct slot instead of | |
509 | * pointing to a node, in which case *@nodep will be NULL. | |
510 | * | |
511 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 512 | */ |
139e5616 | 513 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
514 | unsigned order, struct radix_tree_node **nodep, |
515 | void ***slotp) | |
1da177e4 | 516 | { |
201b6264 | 517 | struct radix_tree_node *node = NULL, *slot; |
49ea6ebc | 518 | unsigned long maxindex; |
139e5616 | 519 | unsigned int height, shift, offset; |
49ea6ebc MW |
520 | unsigned long max = index | ((1UL << order) - 1); |
521 | ||
522 | shift = radix_tree_load_root(root, &slot, &maxindex); | |
1da177e4 LT |
523 | |
524 | /* Make sure the tree is high enough. */ | |
49ea6ebc MW |
525 | if (max > maxindex) { |
526 | int error = radix_tree_extend(root, max); | |
527 | if (error < 0) | |
1da177e4 | 528 | return error; |
49ea6ebc MW |
529 | shift = error; |
530 | slot = root->rnode; | |
531 | if (order == shift) { | |
532 | shift += RADIX_TREE_MAP_SHIFT; | |
533 | root->height++; | |
534 | } | |
1da177e4 LT |
535 | } |
536 | ||
1da177e4 | 537 | height = root->height; |
1da177e4 LT |
538 | |
539 | offset = 0; /* uninitialised var warning */ | |
e6145236 | 540 | while (shift > order) { |
201b6264 | 541 | if (slot == NULL) { |
1da177e4 | 542 | /* Have to add a child node. */ |
201b6264 | 543 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 544 | return -ENOMEM; |
449dd698 | 545 | slot->path = height; |
e2bdb933 | 546 | slot->parent = node; |
201b6264 | 547 | if (node) { |
339e6353 MW |
548 | rcu_assign_pointer(node->slots[offset], |
549 | ptr_to_indirect(slot)); | |
1da177e4 | 550 | node->count++; |
449dd698 | 551 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 552 | } else |
339e6353 MW |
553 | rcu_assign_pointer(root->rnode, |
554 | ptr_to_indirect(slot)); | |
e6145236 MW |
555 | } else if (!radix_tree_is_indirect_ptr(slot)) |
556 | break; | |
1da177e4 LT |
557 | |
558 | /* Go a level down */ | |
e6145236 | 559 | height--; |
0070e28d | 560 | shift -= RADIX_TREE_MAP_SHIFT; |
e6145236 | 561 | node = indirect_to_ptr(slot); |
8a14f4d8 MW |
562 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
563 | offset = radix_tree_descend(node, &slot, offset); | |
e6145236 MW |
564 | } |
565 | ||
57578c2e | 566 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 567 | /* Insert pointers to the canonical entry */ |
3b8c00f6 MW |
568 | if (order > shift) { |
569 | int i, n = 1 << (order - shift); | |
e6145236 MW |
570 | offset = offset & ~(n - 1); |
571 | slot = ptr_to_indirect(&node->slots[offset]); | |
572 | for (i = 0; i < n; i++) { | |
573 | if (node->slots[offset + i]) | |
574 | return -EEXIST; | |
575 | } | |
576 | ||
577 | for (i = 1; i < n; i++) { | |
578 | rcu_assign_pointer(node->slots[offset + i], slot); | |
579 | node->count++; | |
580 | } | |
612d6c19 | 581 | } |
57578c2e | 582 | #endif |
1da177e4 | 583 | |
139e5616 JW |
584 | if (nodep) |
585 | *nodep = node; | |
586 | if (slotp) | |
587 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
588 | return 0; | |
589 | } | |
590 | ||
591 | /** | |
e6145236 | 592 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
593 | * @root: radix tree root |
594 | * @index: index key | |
e6145236 | 595 | * @order: key covers the 2^order indices around index |
139e5616 JW |
596 | * @item: item to insert |
597 | * | |
598 | * Insert an item into the radix tree at position @index. | |
599 | */ | |
e6145236 MW |
600 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
601 | unsigned order, void *item) | |
139e5616 JW |
602 | { |
603 | struct radix_tree_node *node; | |
604 | void **slot; | |
605 | int error; | |
606 | ||
607 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
608 | ||
e6145236 | 609 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
610 | if (error) |
611 | return error; | |
612 | if (*slot != NULL) | |
1da177e4 | 613 | return -EEXIST; |
139e5616 | 614 | rcu_assign_pointer(*slot, item); |
201b6264 | 615 | |
612d6c19 | 616 | if (node) { |
7b60e9ad | 617 | unsigned offset = get_slot_offset(node, slot); |
612d6c19 | 618 | node->count++; |
7b60e9ad MW |
619 | BUG_ON(tag_get(node, 0, offset)); |
620 | BUG_ON(tag_get(node, 1, offset)); | |
621 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 622 | } else { |
7b60e9ad | 623 | BUG_ON(root_tags_get(root)); |
612d6c19 | 624 | } |
1da177e4 | 625 | |
1da177e4 LT |
626 | return 0; |
627 | } | |
e6145236 | 628 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 629 | |
139e5616 JW |
630 | /** |
631 | * __radix_tree_lookup - lookup an item in a radix tree | |
632 | * @root: radix tree root | |
633 | * @index: index key | |
634 | * @nodep: returns node | |
635 | * @slotp: returns slot | |
636 | * | |
637 | * Lookup and return the item at position @index in the radix | |
638 | * tree @root. | |
639 | * | |
640 | * Until there is more than one item in the tree, no nodes are | |
641 | * allocated and @root->rnode is used as a direct slot instead of | |
642 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 643 | */ |
139e5616 JW |
644 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
645 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 646 | { |
139e5616 | 647 | struct radix_tree_node *node, *parent; |
85829954 MW |
648 | unsigned long maxindex; |
649 | unsigned int shift; | |
139e5616 | 650 | void **slot; |
612d6c19 | 651 | |
85829954 MW |
652 | restart: |
653 | parent = NULL; | |
654 | slot = (void **)&root->rnode; | |
655 | shift = radix_tree_load_root(root, &node, &maxindex); | |
656 | if (index > maxindex) | |
1da177e4 LT |
657 | return NULL; |
658 | ||
85829954 MW |
659 | while (radix_tree_is_indirect_ptr(node)) { |
660 | unsigned offset; | |
1da177e4 | 661 | |
85829954 MW |
662 | if (node == RADIX_TREE_RETRY) |
663 | goto restart; | |
664 | parent = indirect_to_ptr(node); | |
1da177e4 | 665 | shift -= RADIX_TREE_MAP_SHIFT; |
85829954 MW |
666 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
667 | offset = radix_tree_descend(parent, &node, offset); | |
668 | slot = parent->slots + offset; | |
669 | } | |
1da177e4 | 670 | |
139e5616 JW |
671 | if (nodep) |
672 | *nodep = parent; | |
673 | if (slotp) | |
674 | *slotp = slot; | |
675 | return node; | |
b72b71c6 HS |
676 | } |
677 | ||
678 | /** | |
679 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
680 | * @root: radix tree root | |
681 | * @index: index key | |
682 | * | |
683 | * Returns: the slot corresponding to the position @index in the | |
684 | * radix tree @root. This is useful for update-if-exists operations. | |
685 | * | |
686 | * This function can be called under rcu_read_lock iff the slot is not | |
687 | * modified by radix_tree_replace_slot, otherwise it must be called | |
688 | * exclusive from other writers. Any dereference of the slot must be done | |
689 | * using radix_tree_deref_slot. | |
690 | */ | |
691 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
692 | { | |
139e5616 JW |
693 | void **slot; |
694 | ||
695 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
696 | return NULL; | |
697 | return slot; | |
a4331366 | 698 | } |
a4331366 HR |
699 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
700 | ||
701 | /** | |
702 | * radix_tree_lookup - perform lookup operation on a radix tree | |
703 | * @root: radix tree root | |
704 | * @index: index key | |
705 | * | |
706 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
707 | * |
708 | * This function can be called under rcu_read_lock, however the caller | |
709 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
710 | * them safely). No RCU barriers are required to access or modify the | |
711 | * returned item, however. | |
a4331366 HR |
712 | */ |
713 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
714 | { | |
139e5616 | 715 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
716 | } |
717 | EXPORT_SYMBOL(radix_tree_lookup); | |
718 | ||
719 | /** | |
720 | * radix_tree_tag_set - set a tag on a radix tree node | |
721 | * @root: radix tree root | |
722 | * @index: index key | |
723 | * @tag: tag index | |
724 | * | |
daff89f3 JC |
725 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
726 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
727 | * the root all the way down to the leaf node. |
728 | * | |
729 | * Returns the address of the tagged item. Setting a tag on a not-present | |
730 | * item is a bug. | |
731 | */ | |
732 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 733 | unsigned long index, unsigned int tag) |
1da177e4 | 734 | { |
fb969909 RZ |
735 | struct radix_tree_node *node, *parent; |
736 | unsigned long maxindex; | |
737 | unsigned int shift; | |
1da177e4 | 738 | |
fb969909 RZ |
739 | shift = radix_tree_load_root(root, &node, &maxindex); |
740 | BUG_ON(index > maxindex); | |
1da177e4 | 741 | |
fb969909 RZ |
742 | while (radix_tree_is_indirect_ptr(node)) { |
743 | unsigned offset; | |
1da177e4 | 744 | |
1da177e4 | 745 | shift -= RADIX_TREE_MAP_SHIFT; |
fb969909 RZ |
746 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
747 | ||
748 | parent = indirect_to_ptr(node); | |
749 | offset = radix_tree_descend(parent, &node, offset); | |
750 | BUG_ON(!node); | |
751 | ||
752 | if (!tag_get(parent, tag, offset)) | |
753 | tag_set(parent, tag, offset); | |
1da177e4 LT |
754 | } |
755 | ||
612d6c19 | 756 | /* set the root's tag bit */ |
fb969909 | 757 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
758 | root_tag_set(root, tag); |
759 | ||
fb969909 | 760 | return node; |
1da177e4 LT |
761 | } |
762 | EXPORT_SYMBOL(radix_tree_tag_set); | |
763 | ||
764 | /** | |
765 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
766 | * @root: radix tree root | |
767 | * @index: index key | |
768 | * @tag: tag index | |
769 | * | |
daff89f3 JC |
770 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
771 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
772 | * this causes the leaf node to have no tags set then clear the tag in the |
773 | * next-to-leaf node, etc. | |
774 | * | |
775 | * Returns the address of the tagged item on success, else NULL. ie: | |
776 | * has the same return value and semantics as radix_tree_lookup(). | |
777 | */ | |
778 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 779 | unsigned long index, unsigned int tag) |
1da177e4 | 780 | { |
00f47b58 RZ |
781 | struct radix_tree_node *node, *parent; |
782 | unsigned long maxindex; | |
783 | unsigned int shift; | |
e2bdb933 | 784 | int uninitialized_var(offset); |
1da177e4 | 785 | |
00f47b58 RZ |
786 | shift = radix_tree_load_root(root, &node, &maxindex); |
787 | if (index > maxindex) | |
788 | return NULL; | |
1da177e4 | 789 | |
00f47b58 | 790 | parent = NULL; |
1da177e4 | 791 | |
00f47b58 | 792 | while (radix_tree_is_indirect_ptr(node)) { |
e2bdb933 | 793 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 794 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
00f47b58 RZ |
795 | |
796 | parent = indirect_to_ptr(node); | |
797 | offset = radix_tree_descend(parent, &node, offset); | |
1da177e4 LT |
798 | } |
799 | ||
00f47b58 | 800 | if (node == NULL) |
1da177e4 LT |
801 | goto out; |
802 | ||
00f47b58 RZ |
803 | index >>= shift; |
804 | ||
805 | while (parent) { | |
806 | if (!tag_get(parent, tag, offset)) | |
d5274261 | 807 | goto out; |
00f47b58 RZ |
808 | tag_clear(parent, tag, offset); |
809 | if (any_tag_set(parent, tag)) | |
6e954b9e | 810 | goto out; |
e2bdb933 HD |
811 | |
812 | index >>= RADIX_TREE_MAP_SHIFT; | |
813 | offset = index & RADIX_TREE_MAP_MASK; | |
00f47b58 | 814 | parent = parent->parent; |
612d6c19 NP |
815 | } |
816 | ||
817 | /* clear the root's tag bit */ | |
818 | if (root_tag_get(root, tag)) | |
819 | root_tag_clear(root, tag); | |
820 | ||
1da177e4 | 821 | out: |
00f47b58 | 822 | return node; |
1da177e4 LT |
823 | } |
824 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
825 | ||
1da177e4 | 826 | /** |
32605a18 MT |
827 | * radix_tree_tag_get - get a tag on a radix tree node |
828 | * @root: radix tree root | |
829 | * @index: index key | |
daff89f3 | 830 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 831 | * |
32605a18 | 832 | * Return values: |
1da177e4 | 833 | * |
612d6c19 NP |
834 | * 0: tag not present or not set |
835 | * 1: tag set | |
ce82653d DH |
836 | * |
837 | * Note that the return value of this function may not be relied on, even if | |
838 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
839 | * from concurrency. | |
1da177e4 LT |
840 | */ |
841 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 842 | unsigned long index, unsigned int tag) |
1da177e4 | 843 | { |
4589ba6d RZ |
844 | struct radix_tree_node *node, *parent; |
845 | unsigned long maxindex; | |
846 | unsigned int shift; | |
1da177e4 | 847 | |
612d6c19 NP |
848 | if (!root_tag_get(root, tag)) |
849 | return 0; | |
850 | ||
4589ba6d RZ |
851 | shift = radix_tree_load_root(root, &node, &maxindex); |
852 | if (index > maxindex) | |
853 | return 0; | |
7cf9c2c7 NP |
854 | if (node == NULL) |
855 | return 0; | |
856 | ||
4589ba6d RZ |
857 | while (radix_tree_is_indirect_ptr(node)) { |
858 | int offset; | |
612d6c19 | 859 | |
4589ba6d RZ |
860 | shift -= RADIX_TREE_MAP_SHIFT; |
861 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1da177e4 | 862 | |
4589ba6d RZ |
863 | parent = indirect_to_ptr(node); |
864 | offset = radix_tree_descend(parent, &node, offset); | |
1da177e4 | 865 | |
4589ba6d | 866 | if (!node) |
1da177e4 | 867 | return 0; |
4589ba6d | 868 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 869 | return 0; |
4589ba6d RZ |
870 | if (node == RADIX_TREE_RETRY) |
871 | break; | |
1da177e4 | 872 | } |
4589ba6d RZ |
873 | |
874 | return 1; | |
1da177e4 LT |
875 | } |
876 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 877 | |
21ef5339 RZ |
878 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
879 | unsigned int shift) | |
880 | { | |
881 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
882 | iter->shift = shift; | |
883 | #endif | |
884 | } | |
885 | ||
78c1d784 KK |
886 | /** |
887 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
888 | * | |
889 | * @root: radix tree root | |
890 | * @iter: iterator state | |
891 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
892 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
893 | */ | |
894 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
895 | struct radix_tree_iter *iter, unsigned flags) | |
896 | { | |
897 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
898 | struct radix_tree_node *rnode, *node; | |
21ef5339 | 899 | unsigned long index, offset, maxindex; |
78c1d784 KK |
900 | |
901 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
902 | return NULL; | |
903 | ||
904 | /* | |
905 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
906 | * during iterating; it can be zero only at the beginning. | |
907 | * And we cannot overflow iter->next_index in a single step, | |
908 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
909 | * |
910 | * This condition also used by radix_tree_next_slot() to stop | |
911 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
912 | */ |
913 | index = iter->next_index; | |
914 | if (!index && iter->index) | |
915 | return NULL; | |
916 | ||
21ef5339 RZ |
917 | restart: |
918 | shift = radix_tree_load_root(root, &rnode, &maxindex); | |
919 | if (index > maxindex) | |
920 | return NULL; | |
921 | ||
78c1d784 KK |
922 | if (radix_tree_is_indirect_ptr(rnode)) { |
923 | rnode = indirect_to_ptr(rnode); | |
21ef5339 | 924 | } else if (rnode) { |
78c1d784 | 925 | /* Single-slot tree */ |
21ef5339 RZ |
926 | iter->index = index; |
927 | iter->next_index = maxindex + 1; | |
78c1d784 | 928 | iter->tags = 1; |
21ef5339 | 929 | __set_iter_shift(iter, shift); |
78c1d784 KK |
930 | return (void **)&root->rnode; |
931 | } else | |
932 | return NULL; | |
933 | ||
21ef5339 | 934 | shift -= RADIX_TREE_MAP_SHIFT; |
78c1d784 KK |
935 | offset = index >> shift; |
936 | ||
78c1d784 KK |
937 | node = rnode; |
938 | while (1) { | |
e6145236 | 939 | struct radix_tree_node *slot; |
21ef5339 RZ |
940 | unsigned new_off = radix_tree_descend(node, &slot, offset); |
941 | ||
942 | if (new_off < offset) { | |
943 | offset = new_off; | |
944 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
945 | index |= offset << shift; | |
946 | } | |
947 | ||
78c1d784 | 948 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
21ef5339 | 949 | !tag_get(node, tag, offset) : !slot) { |
78c1d784 KK |
950 | /* Hole detected */ |
951 | if (flags & RADIX_TREE_ITER_CONTIG) | |
952 | return NULL; | |
953 | ||
954 | if (flags & RADIX_TREE_ITER_TAGGED) | |
955 | offset = radix_tree_find_next_bit( | |
956 | node->tags[tag], | |
957 | RADIX_TREE_MAP_SIZE, | |
958 | offset + 1); | |
959 | else | |
960 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
961 | void *slot = node->slots[offset]; |
962 | if (is_sibling_entry(node, slot)) | |
963 | continue; | |
964 | if (slot) | |
78c1d784 KK |
965 | break; |
966 | } | |
967 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
968 | index += offset << shift; | |
969 | /* Overflow after ~0UL */ | |
970 | if (!index) | |
971 | return NULL; | |
972 | if (offset == RADIX_TREE_MAP_SIZE) | |
973 | goto restart; | |
21ef5339 | 974 | slot = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
975 | } |
976 | ||
21ef5339 | 977 | if ((slot == NULL) || (slot == RADIX_TREE_RETRY)) |
78c1d784 | 978 | goto restart; |
e6145236 MW |
979 | if (!radix_tree_is_indirect_ptr(slot)) |
980 | break; | |
21ef5339 | 981 | |
e6145236 | 982 | node = indirect_to_ptr(slot); |
78c1d784 KK |
983 | shift -= RADIX_TREE_MAP_SHIFT; |
984 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
985 | } | |
986 | ||
987 | /* Update the iterator state */ | |
21ef5339 RZ |
988 | iter->index = index & ~((1 << shift) - 1); |
989 | iter->next_index = (index | ((RADIX_TREE_MAP_SIZE << shift) - 1)) + 1; | |
990 | __set_iter_shift(iter, shift); | |
78c1d784 KK |
991 | |
992 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
993 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
994 | unsigned tag_long, tag_bit; | |
995 | ||
996 | tag_long = offset / BITS_PER_LONG; | |
997 | tag_bit = offset % BITS_PER_LONG; | |
998 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
999 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1000 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1001 | /* Pick tags from next element */ | |
1002 | if (tag_bit) | |
1003 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1004 | (BITS_PER_LONG - tag_bit); | |
1005 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1006 | iter->next_index = index + BITS_PER_LONG; | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | return node->slots + offset; | |
1011 | } | |
1012 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1013 | ||
ebf8aa44 JK |
1014 | /** |
1015 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
1016 | * tag if item has another tag set | |
1017 | * @root: radix tree root | |
1018 | * @first_indexp: pointer to a starting index of a range to scan | |
1019 | * @last_index: last index of a range to scan | |
1020 | * @nr_to_tag: maximum number items to tag | |
1021 | * @iftag: tag index to test | |
1022 | * @settag: tag index to set if tested tag is set | |
1023 | * | |
1024 | * This function scans range of radix tree from first_index to last_index | |
1025 | * (inclusive). For each item in the range if iftag is set, the function sets | |
1026 | * also settag. The function stops either after tagging nr_to_tag items or | |
1027 | * after reaching last_index. | |
1028 | * | |
144dcfc0 DC |
1029 | * The tags must be set from the leaf level only and propagated back up the |
1030 | * path to the root. We must do this so that we resolve the full path before | |
1031 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
1032 | * we can get to the leaf node and find that the index that has the iftag | |
1033 | * set is outside the range we are scanning. This reults in dangling tags and | |
1034 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1035 | * | |
ebf8aa44 JK |
1036 | * The function returns number of leaves where the tag was set and sets |
1037 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
1038 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1039 | * be prepared to handle that. | |
ebf8aa44 JK |
1040 | */ |
1041 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1042 | unsigned long *first_indexp, unsigned long last_index, | |
1043 | unsigned long nr_to_tag, | |
1044 | unsigned int iftag, unsigned int settag) | |
1045 | { | |
070c5ac2 MW |
1046 | struct radix_tree_node *slot, *node = NULL; |
1047 | unsigned long maxindex; | |
1048 | unsigned int shift = radix_tree_load_root(root, &slot, &maxindex); | |
144dcfc0 DC |
1049 | unsigned long tagged = 0; |
1050 | unsigned long index = *first_indexp; | |
ebf8aa44 | 1051 | |
070c5ac2 | 1052 | last_index = min(last_index, maxindex); |
ebf8aa44 JK |
1053 | if (index > last_index) |
1054 | return 0; | |
1055 | if (!nr_to_tag) | |
1056 | return 0; | |
1057 | if (!root_tag_get(root, iftag)) { | |
1058 | *first_indexp = last_index + 1; | |
1059 | return 0; | |
1060 | } | |
070c5ac2 | 1061 | if (!radix_tree_is_indirect_ptr(slot)) { |
ebf8aa44 JK |
1062 | *first_indexp = last_index + 1; |
1063 | root_tag_set(root, settag); | |
1064 | return 1; | |
1065 | } | |
1066 | ||
070c5ac2 MW |
1067 | node = indirect_to_ptr(slot); |
1068 | shift -= RADIX_TREE_MAP_SHIFT; | |
ebf8aa44 JK |
1069 | |
1070 | for (;;) { | |
e2bdb933 | 1071 | unsigned long upindex; |
070c5ac2 | 1072 | unsigned offset; |
ebf8aa44 JK |
1073 | |
1074 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
070c5ac2 MW |
1075 | offset = radix_tree_descend(node, &slot, offset); |
1076 | if (!slot) | |
ebf8aa44 | 1077 | goto next; |
070c5ac2 | 1078 | if (!tag_get(node, iftag, offset)) |
ebf8aa44 | 1079 | goto next; |
070c5ac2 MW |
1080 | /* Sibling slots never have tags set on them */ |
1081 | if (radix_tree_is_indirect_ptr(slot)) { | |
1082 | node = indirect_to_ptr(slot); | |
1083 | shift -= RADIX_TREE_MAP_SHIFT; | |
1084 | continue; | |
144dcfc0 DC |
1085 | } |
1086 | ||
1087 | /* tag the leaf */ | |
070c5ac2 MW |
1088 | tagged++; |
1089 | tag_set(node, settag, offset); | |
144dcfc0 | 1090 | |
070c5ac2 | 1091 | slot = node->parent; |
144dcfc0 | 1092 | /* walk back up the path tagging interior nodes */ |
070c5ac2 MW |
1093 | upindex = index >> shift; |
1094 | while (slot) { | |
e2bdb933 HD |
1095 | upindex >>= RADIX_TREE_MAP_SHIFT; |
1096 | offset = upindex & RADIX_TREE_MAP_MASK; | |
1097 | ||
144dcfc0 | 1098 | /* stop if we find a node with the tag already set */ |
070c5ac2 | 1099 | if (tag_get(slot, settag, offset)) |
144dcfc0 | 1100 | break; |
070c5ac2 MW |
1101 | tag_set(slot, settag, offset); |
1102 | slot = slot->parent; | |
ebf8aa44 | 1103 | } |
144dcfc0 | 1104 | |
070c5ac2 | 1105 | next: |
ebf8aa44 JK |
1106 | /* Go to next item at level determined by 'shift' */ |
1107 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1108 | /* Overflow can happen when last_index is ~0UL... */ |
1109 | if (index > last_index || !index) | |
ebf8aa44 | 1110 | break; |
070c5ac2 MW |
1111 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
1112 | while (offset == 0) { | |
ebf8aa44 JK |
1113 | /* |
1114 | * We've fully scanned this node. Go up. Because | |
1115 | * last_index is guaranteed to be in the tree, what | |
1116 | * we do below cannot wander astray. | |
1117 | */ | |
070c5ac2 | 1118 | node = node->parent; |
ebf8aa44 | 1119 | shift += RADIX_TREE_MAP_SHIFT; |
070c5ac2 | 1120 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
ebf8aa44 | 1121 | } |
070c5ac2 MW |
1122 | if (is_sibling_entry(node, node->slots[offset])) |
1123 | goto next; | |
1124 | if (tagged >= nr_to_tag) | |
1125 | break; | |
ebf8aa44 JK |
1126 | } |
1127 | /* | |
ac15ee69 TO |
1128 | * We need not to tag the root tag if there is no tag which is set with |
1129 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1130 | */ |
ac15ee69 TO |
1131 | if (tagged > 0) |
1132 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1133 | *first_indexp = index; |
1134 | ||
1135 | return tagged; | |
1136 | } | |
1137 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1138 | ||
1da177e4 LT |
1139 | /** |
1140 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1141 | * @root: radix tree root | |
1142 | * @results: where the results of the lookup are placed | |
1143 | * @first_index: start the lookup from this key | |
1144 | * @max_items: place up to this many items at *results | |
1145 | * | |
1146 | * Performs an index-ascending scan of the tree for present items. Places | |
1147 | * them at *@results and returns the number of items which were placed at | |
1148 | * *@results. | |
1149 | * | |
1150 | * The implementation is naive. | |
7cf9c2c7 NP |
1151 | * |
1152 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1153 | * rcu_read_lock. In this case, rather than the returned results being | |
1154 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1155 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1156 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1157 | */ |
1158 | unsigned int | |
1159 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1160 | unsigned long first_index, unsigned int max_items) | |
1161 | { | |
cebbd29e KK |
1162 | struct radix_tree_iter iter; |
1163 | void **slot; | |
1164 | unsigned int ret = 0; | |
7cf9c2c7 | 1165 | |
cebbd29e | 1166 | if (unlikely(!max_items)) |
7cf9c2c7 | 1167 | return 0; |
1da177e4 | 1168 | |
cebbd29e | 1169 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1170 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1171 | if (!results[ret]) |
1172 | continue; | |
46437f9a MW |
1173 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1174 | slot = radix_tree_iter_retry(&iter); | |
1175 | continue; | |
1176 | } | |
cebbd29e | 1177 | if (++ret == max_items) |
1da177e4 | 1178 | break; |
1da177e4 | 1179 | } |
7cf9c2c7 | 1180 | |
1da177e4 LT |
1181 | return ret; |
1182 | } | |
1183 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1184 | ||
47feff2c NP |
1185 | /** |
1186 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1187 | * @root: radix tree root | |
1188 | * @results: where the results of the lookup are placed | |
6328650b | 1189 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1190 | * @first_index: start the lookup from this key |
1191 | * @max_items: place up to this many items at *results | |
1192 | * | |
1193 | * Performs an index-ascending scan of the tree for present items. Places | |
1194 | * their slots at *@results and returns the number of items which were | |
1195 | * placed at *@results. | |
1196 | * | |
1197 | * The implementation is naive. | |
1198 | * | |
1199 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1200 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1201 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1202 | */ | |
1203 | unsigned int | |
6328650b HD |
1204 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1205 | void ***results, unsigned long *indices, | |
47feff2c NP |
1206 | unsigned long first_index, unsigned int max_items) |
1207 | { | |
cebbd29e KK |
1208 | struct radix_tree_iter iter; |
1209 | void **slot; | |
1210 | unsigned int ret = 0; | |
47feff2c | 1211 | |
cebbd29e | 1212 | if (unlikely(!max_items)) |
47feff2c NP |
1213 | return 0; |
1214 | ||
cebbd29e KK |
1215 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1216 | results[ret] = slot; | |
6328650b | 1217 | if (indices) |
cebbd29e KK |
1218 | indices[ret] = iter.index; |
1219 | if (++ret == max_items) | |
47feff2c | 1220 | break; |
47feff2c NP |
1221 | } |
1222 | ||
1223 | return ret; | |
1224 | } | |
1225 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1226 | ||
1da177e4 LT |
1227 | /** |
1228 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1229 | * based on a tag | |
1230 | * @root: radix tree root | |
1231 | * @results: where the results of the lookup are placed | |
1232 | * @first_index: start the lookup from this key | |
1233 | * @max_items: place up to this many items at *results | |
daff89f3 | 1234 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1235 | * |
1236 | * Performs an index-ascending scan of the tree for present items which | |
1237 | * have the tag indexed by @tag set. Places the items at *@results and | |
1238 | * returns the number of items which were placed at *@results. | |
1239 | */ | |
1240 | unsigned int | |
1241 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1242 | unsigned long first_index, unsigned int max_items, |
1243 | unsigned int tag) | |
1da177e4 | 1244 | { |
cebbd29e KK |
1245 | struct radix_tree_iter iter; |
1246 | void **slot; | |
1247 | unsigned int ret = 0; | |
612d6c19 | 1248 | |
cebbd29e | 1249 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1250 | return 0; |
1251 | ||
cebbd29e | 1252 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1253 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1254 | if (!results[ret]) |
1255 | continue; | |
46437f9a MW |
1256 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1257 | slot = radix_tree_iter_retry(&iter); | |
1258 | continue; | |
1259 | } | |
cebbd29e | 1260 | if (++ret == max_items) |
1da177e4 | 1261 | break; |
1da177e4 | 1262 | } |
7cf9c2c7 | 1263 | |
1da177e4 LT |
1264 | return ret; |
1265 | } | |
1266 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1267 | ||
47feff2c NP |
1268 | /** |
1269 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1270 | * radix tree based on a tag | |
1271 | * @root: radix tree root | |
1272 | * @results: where the results of the lookup are placed | |
1273 | * @first_index: start the lookup from this key | |
1274 | * @max_items: place up to this many items at *results | |
1275 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1276 | * | |
1277 | * Performs an index-ascending scan of the tree for present items which | |
1278 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1279 | * returns the number of slots which were placed at *@results. | |
1280 | */ | |
1281 | unsigned int | |
1282 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1283 | unsigned long first_index, unsigned int max_items, | |
1284 | unsigned int tag) | |
1285 | { | |
cebbd29e KK |
1286 | struct radix_tree_iter iter; |
1287 | void **slot; | |
1288 | unsigned int ret = 0; | |
47feff2c | 1289 | |
cebbd29e | 1290 | if (unlikely(!max_items)) |
47feff2c NP |
1291 | return 0; |
1292 | ||
cebbd29e KK |
1293 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1294 | results[ret] = slot; | |
1295 | if (++ret == max_items) | |
47feff2c | 1296 | break; |
47feff2c NP |
1297 | } |
1298 | ||
1299 | return ret; | |
1300 | } | |
1301 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1302 | ||
e504f3fd HD |
1303 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1304 | #include <linux/sched.h> /* for cond_resched() */ | |
1305 | ||
0a2efc6c MW |
1306 | struct locate_info { |
1307 | unsigned long found_index; | |
1308 | bool stop; | |
1309 | }; | |
1310 | ||
e504f3fd HD |
1311 | /* |
1312 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1313 | */ | |
1314 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
0a2efc6c | 1315 | unsigned long index, struct locate_info *info) |
e504f3fd HD |
1316 | { |
1317 | unsigned int shift, height; | |
1318 | unsigned long i; | |
1319 | ||
449dd698 | 1320 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
0a2efc6c | 1321 | shift = height * RADIX_TREE_MAP_SHIFT; |
e504f3fd | 1322 | |
0a2efc6c MW |
1323 | do { |
1324 | shift -= RADIX_TREE_MAP_SHIFT; | |
e504f3fd | 1325 | |
0a2efc6c MW |
1326 | for (i = (index >> shift) & RADIX_TREE_MAP_MASK; |
1327 | i < RADIX_TREE_MAP_SIZE; | |
1328 | i++, index += (1UL << shift)) { | |
1329 | struct radix_tree_node *node = | |
1330 | rcu_dereference_raw(slot->slots[i]); | |
1331 | if (node == RADIX_TREE_RETRY) | |
1332 | goto out; | |
1333 | if (!radix_tree_is_indirect_ptr(node)) { | |
1334 | if (node == item) { | |
1335 | info->found_index = index; | |
1336 | info->stop = true; | |
1337 | goto out; | |
1338 | } | |
1339 | continue; | |
e6145236 | 1340 | } |
0a2efc6c MW |
1341 | node = indirect_to_ptr(node); |
1342 | if (is_sibling_entry(slot, node)) | |
1343 | continue; | |
1344 | slot = node; | |
1345 | break; | |
e6145236 | 1346 | } |
0a2efc6c MW |
1347 | if (i == RADIX_TREE_MAP_SIZE) |
1348 | break; | |
1349 | } while (shift); | |
e504f3fd | 1350 | |
e504f3fd | 1351 | out: |
0a2efc6c MW |
1352 | if ((index == 0) && (i == RADIX_TREE_MAP_SIZE)) |
1353 | info->stop = true; | |
e504f3fd HD |
1354 | return index; |
1355 | } | |
1356 | ||
1357 | /** | |
1358 | * radix_tree_locate_item - search through radix tree for item | |
1359 | * @root: radix tree root | |
1360 | * @item: item to be found | |
1361 | * | |
1362 | * Returns index where item was found, or -1 if not found. | |
1363 | * Caller must hold no lock (since this time-consuming function needs | |
1364 | * to be preemptible), and must check afterwards if item is still there. | |
1365 | */ | |
1366 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1367 | { | |
1368 | struct radix_tree_node *node; | |
1369 | unsigned long max_index; | |
1370 | unsigned long cur_index = 0; | |
0a2efc6c MW |
1371 | struct locate_info info = { |
1372 | .found_index = -1, | |
1373 | .stop = false, | |
1374 | }; | |
e504f3fd HD |
1375 | |
1376 | do { | |
1377 | rcu_read_lock(); | |
1378 | node = rcu_dereference_raw(root->rnode); | |
1379 | if (!radix_tree_is_indirect_ptr(node)) { | |
1380 | rcu_read_unlock(); | |
1381 | if (node == item) | |
0a2efc6c | 1382 | info.found_index = 0; |
e504f3fd HD |
1383 | break; |
1384 | } | |
1385 | ||
1386 | node = indirect_to_ptr(node); | |
0a2efc6c MW |
1387 | |
1388 | max_index = node_maxindex(node); | |
5f30fc94 HD |
1389 | if (cur_index > max_index) { |
1390 | rcu_read_unlock(); | |
e504f3fd | 1391 | break; |
5f30fc94 | 1392 | } |
e504f3fd | 1393 | |
0a2efc6c | 1394 | cur_index = __locate(node, item, cur_index, &info); |
e504f3fd HD |
1395 | rcu_read_unlock(); |
1396 | cond_resched(); | |
0a2efc6c | 1397 | } while (!info.stop && cur_index <= max_index); |
e504f3fd | 1398 | |
0a2efc6c | 1399 | return info.found_index; |
e504f3fd HD |
1400 | } |
1401 | #else | |
1402 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1403 | { | |
1404 | return -1; | |
1405 | } | |
1406 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1407 | |
a5f51c96 NP |
1408 | /** |
1409 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1410 | * @root radix tree root | |
1411 | */ | |
1412 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1413 | { | |
1414 | /* try to shrink tree height */ | |
c0bc9875 | 1415 | while (root->height > 0) { |
a5f51c96 | 1416 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1417 | struct radix_tree_node *slot; |
a5f51c96 | 1418 | |
c0bc9875 | 1419 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1420 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1421 | |
1422 | /* | |
1423 | * The candidate node has more than one child, or its child | |
e6145236 MW |
1424 | * is not at the leftmost slot, or it is a multiorder entry, |
1425 | * we cannot shrink. | |
c0bc9875 NP |
1426 | */ |
1427 | if (to_free->count != 1) | |
1428 | break; | |
339e6353 MW |
1429 | slot = to_free->slots[0]; |
1430 | if (!slot) | |
c0bc9875 | 1431 | break; |
afe0e395 MW |
1432 | if (!radix_tree_is_indirect_ptr(slot) && (root->height > 1)) |
1433 | break; | |
1434 | ||
1435 | if (radix_tree_is_indirect_ptr(slot)) { | |
1436 | slot = indirect_to_ptr(slot); | |
1437 | slot->parent = NULL; | |
1438 | slot = ptr_to_indirect(slot); | |
1439 | } | |
c0bc9875 | 1440 | |
7cf9c2c7 NP |
1441 | /* |
1442 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1443 | * moving the node from one part of the tree to another: if it |
1444 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1445 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1446 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1447 | */ |
e2bdb933 | 1448 | root->rnode = slot; |
a5f51c96 | 1449 | root->height--; |
27d20fdd NP |
1450 | |
1451 | /* | |
1452 | * We have a dilemma here. The node's slot[0] must not be | |
1453 | * NULLed in case there are concurrent lookups expecting to | |
1454 | * find the item. However if this was a bottom-level node, | |
1455 | * then it may be subject to the slot pointer being visible | |
1456 | * to callers dereferencing it. If item corresponding to | |
1457 | * slot[0] is subsequently deleted, these callers would expect | |
1458 | * their slot to become empty sooner or later. | |
1459 | * | |
1460 | * For example, lockless pagecache will look up a slot, deref | |
1461 | * the page pointer, and if the page is 0 refcount it means it | |
1462 | * was concurrently deleted from pagecache so try the deref | |
1463 | * again. Fortunately there is already a requirement for logic | |
1464 | * to retry the entire slot lookup -- the indirect pointer | |
1465 | * problem (replacing direct root node with an indirect pointer | |
1466 | * also results in a stale slot). So tag the slot as indirect | |
1467 | * to force callers to retry. | |
1468 | */ | |
afe0e395 MW |
1469 | if (!radix_tree_is_indirect_ptr(slot)) |
1470 | to_free->slots[0] = RADIX_TREE_RETRY; | |
27d20fdd | 1471 | |
a5f51c96 NP |
1472 | radix_tree_node_free(to_free); |
1473 | } | |
1474 | } | |
1475 | ||
139e5616 JW |
1476 | /** |
1477 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1478 | * @root: radix tree root | |
139e5616 JW |
1479 | * @node: node containing @index |
1480 | * | |
1481 | * After clearing the slot at @index in @node from radix tree | |
1482 | * rooted at @root, call this function to attempt freeing the | |
1483 | * node and shrinking the tree. | |
1484 | * | |
1485 | * Returns %true if @node was freed, %false otherwise. | |
1486 | */ | |
449dd698 | 1487 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1488 | struct radix_tree_node *node) |
1489 | { | |
1490 | bool deleted = false; | |
1491 | ||
1492 | do { | |
1493 | struct radix_tree_node *parent; | |
1494 | ||
1495 | if (node->count) { | |
1496 | if (node == indirect_to_ptr(root->rnode)) { | |
1497 | radix_tree_shrink(root); | |
1498 | if (root->height == 0) | |
1499 | deleted = true; | |
1500 | } | |
1501 | return deleted; | |
1502 | } | |
1503 | ||
1504 | parent = node->parent; | |
1505 | if (parent) { | |
449dd698 | 1506 | unsigned int offset; |
139e5616 | 1507 | |
449dd698 JW |
1508 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1509 | parent->slots[offset] = NULL; | |
139e5616 JW |
1510 | parent->count--; |
1511 | } else { | |
1512 | root_tag_clear_all(root); | |
1513 | root->height = 0; | |
1514 | root->rnode = NULL; | |
1515 | } | |
1516 | ||
1517 | radix_tree_node_free(node); | |
1518 | deleted = true; | |
1519 | ||
1520 | node = parent; | |
1521 | } while (node); | |
1522 | ||
1523 | return deleted; | |
1524 | } | |
1525 | ||
57578c2e MW |
1526 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1527 | void *ptr, unsigned offset) | |
1528 | { | |
1529 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1530 | int i; | |
1531 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1532 | if (node->slots[offset + i] != ptr) | |
1533 | break; | |
1534 | node->slots[offset + i] = NULL; | |
1535 | node->count--; | |
1536 | } | |
1537 | #endif | |
1538 | } | |
1539 | ||
1da177e4 | 1540 | /** |
53c59f26 | 1541 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1542 | * @root: radix tree root |
1543 | * @index: index key | |
53c59f26 | 1544 | * @item: expected item |
1da177e4 | 1545 | * |
53c59f26 | 1546 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1547 | * |
53c59f26 JW |
1548 | * Returns the address of the deleted item, or NULL if it was not present |
1549 | * or the entry at the given @index was not @item. | |
1da177e4 | 1550 | */ |
53c59f26 JW |
1551 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1552 | unsigned long index, void *item) | |
1da177e4 | 1553 | { |
139e5616 | 1554 | struct radix_tree_node *node; |
57578c2e | 1555 | unsigned int offset; |
139e5616 JW |
1556 | void **slot; |
1557 | void *entry; | |
d5274261 | 1558 | int tag; |
1da177e4 | 1559 | |
139e5616 JW |
1560 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1561 | if (!entry) | |
1562 | return NULL; | |
1da177e4 | 1563 | |
139e5616 JW |
1564 | if (item && entry != item) |
1565 | return NULL; | |
1566 | ||
1567 | if (!node) { | |
612d6c19 NP |
1568 | root_tag_clear_all(root); |
1569 | root->rnode = NULL; | |
139e5616 | 1570 | return entry; |
612d6c19 | 1571 | } |
1da177e4 | 1572 | |
29e0967c | 1573 | offset = get_slot_offset(node, slot); |
53c59f26 | 1574 | |
1da177e4 | 1575 | /* |
e2bdb933 HD |
1576 | * Clear all tags associated with the item to be deleted. |
1577 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1578 | */ |
daff89f3 | 1579 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1580 | if (tag_get(node, tag, offset)) |
612d6c19 | 1581 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1582 | } |
1da177e4 | 1583 | |
57578c2e | 1584 | delete_sibling_entries(node, ptr_to_indirect(slot), offset); |
139e5616 JW |
1585 | node->slots[offset] = NULL; |
1586 | node->count--; | |
e2bdb933 | 1587 | |
449dd698 | 1588 | __radix_tree_delete_node(root, node); |
612d6c19 | 1589 | |
139e5616 | 1590 | return entry; |
1da177e4 | 1591 | } |
53c59f26 JW |
1592 | EXPORT_SYMBOL(radix_tree_delete_item); |
1593 | ||
1594 | /** | |
1595 | * radix_tree_delete - delete an item from a radix tree | |
1596 | * @root: radix tree root | |
1597 | * @index: index key | |
1598 | * | |
1599 | * Remove the item at @index from the radix tree rooted at @root. | |
1600 | * | |
1601 | * Returns the address of the deleted item, or NULL if it was not present. | |
1602 | */ | |
1603 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1604 | { | |
1605 | return radix_tree_delete_item(root, index, NULL); | |
1606 | } | |
1da177e4 LT |
1607 | EXPORT_SYMBOL(radix_tree_delete); |
1608 | ||
1609 | /** | |
1610 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1611 | * @root: radix tree root | |
1612 | * @tag: tag to test | |
1613 | */ | |
daff89f3 | 1614 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1615 | { |
612d6c19 | 1616 | return root_tag_get(root, tag); |
1da177e4 LT |
1617 | } |
1618 | EXPORT_SYMBOL(radix_tree_tagged); | |
1619 | ||
1620 | static void | |
449dd698 | 1621 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1622 | { |
449dd698 JW |
1623 | struct radix_tree_node *node = arg; |
1624 | ||
1625 | memset(node, 0, sizeof(*node)); | |
1626 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1627 | } |
1628 | ||
1629 | static __init unsigned long __maxindex(unsigned int height) | |
1630 | { | |
430d275a PL |
1631 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1632 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1633 | ||
1634 | if (shift < 0) | |
1635 | return ~0UL; | |
1636 | if (shift >= BITS_PER_LONG) | |
1637 | return 0UL; | |
1638 | return ~0UL >> shift; | |
1da177e4 LT |
1639 | } |
1640 | ||
1641 | static __init void radix_tree_init_maxindex(void) | |
1642 | { | |
1643 | unsigned int i; | |
1644 | ||
1645 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1646 | height_to_maxindex[i] = __maxindex(i); | |
1647 | } | |
1648 | ||
1da177e4 LT |
1649 | static int radix_tree_callback(struct notifier_block *nfb, |
1650 | unsigned long action, | |
1651 | void *hcpu) | |
1652 | { | |
1653 | int cpu = (long)hcpu; | |
1654 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1655 | struct radix_tree_node *node; |
1da177e4 LT |
1656 | |
1657 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1658 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1659 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1660 | while (rtp->nr) { | |
9d2a8da0 KS |
1661 | node = rtp->nodes; |
1662 | rtp->nodes = node->private_data; | |
1663 | kmem_cache_free(radix_tree_node_cachep, node); | |
1664 | rtp->nr--; | |
1da177e4 LT |
1665 | } |
1666 | } | |
1667 | return NOTIFY_OK; | |
1668 | } | |
1da177e4 LT |
1669 | |
1670 | void __init radix_tree_init(void) | |
1671 | { | |
1672 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1673 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1674 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1675 | radix_tree_node_ctor); | |
1da177e4 LT |
1676 | radix_tree_init_maxindex(); |
1677 | hotcpu_notifier(radix_tree_callback, 0); | |
1678 | } |