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 | ||
78 | static inline void *indirect_to_ptr(void *ptr) | |
79 | { | |
80 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR); | |
81 | } | |
82 | ||
db050f29 MW |
83 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
84 | /* Sibling slots point directly to another slot in the same node */ | |
85 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
86 | { | |
87 | void **ptr = node; | |
88 | return (parent->slots <= ptr) && | |
89 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
90 | } | |
91 | #else | |
92 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
93 | { | |
94 | return false; | |
95 | } | |
96 | #endif | |
97 | ||
98 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
99 | void **slot) | |
100 | { | |
101 | return slot - parent->slots; | |
102 | } | |
103 | ||
104 | static unsigned radix_tree_descend(struct radix_tree_node *parent, | |
105 | struct radix_tree_node **nodep, unsigned offset) | |
106 | { | |
107 | void **entry = rcu_dereference_raw(parent->slots[offset]); | |
108 | ||
109 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
110 | if (radix_tree_is_indirect_ptr(entry)) { | |
111 | unsigned long siboff = get_slot_offset(parent, entry); | |
112 | if (siboff < RADIX_TREE_MAP_SIZE) { | |
113 | offset = siboff; | |
114 | entry = rcu_dereference_raw(parent->slots[offset]); | |
115 | } | |
116 | } | |
117 | #endif | |
118 | ||
119 | *nodep = (void *)entry; | |
120 | return offset; | |
121 | } | |
122 | ||
612d6c19 NP |
123 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
124 | { | |
125 | return root->gfp_mask & __GFP_BITS_MASK; | |
126 | } | |
127 | ||
643b52b9 NP |
128 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
129 | int offset) | |
130 | { | |
131 | __set_bit(offset, node->tags[tag]); | |
132 | } | |
133 | ||
134 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
135 | int offset) | |
136 | { | |
137 | __clear_bit(offset, node->tags[tag]); | |
138 | } | |
139 | ||
140 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
141 | int offset) | |
142 | { | |
143 | return test_bit(offset, node->tags[tag]); | |
144 | } | |
145 | ||
146 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
147 | { | |
148 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
149 | } | |
150 | ||
151 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag) | |
152 | { | |
153 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
154 | } | |
155 | ||
156 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
157 | { | |
158 | root->gfp_mask &= __GFP_BITS_MASK; | |
159 | } | |
160 | ||
161 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
162 | { | |
163 | return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); | |
164 | } | |
165 | ||
166 | /* | |
167 | * Returns 1 if any slot in the node has this tag set. | |
168 | * Otherwise returns 0. | |
169 | */ | |
170 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
171 | { | |
172 | int idx; | |
173 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { | |
174 | if (node->tags[tag][idx]) | |
175 | return 1; | |
176 | } | |
177 | return 0; | |
178 | } | |
78c1d784 KK |
179 | |
180 | /** | |
181 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
182 | * | |
183 | * @addr: The address to base the search on | |
184 | * @size: The bitmap size in bits | |
185 | * @offset: The bitnumber to start searching at | |
186 | * | |
187 | * Unrollable variant of find_next_bit() for constant size arrays. | |
188 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
189 | * Returns next bit offset, or size if nothing found. | |
190 | */ | |
191 | static __always_inline unsigned long | |
192 | radix_tree_find_next_bit(const unsigned long *addr, | |
193 | unsigned long size, unsigned long offset) | |
194 | { | |
195 | if (!__builtin_constant_p(size)) | |
196 | return find_next_bit(addr, size, offset); | |
197 | ||
198 | if (offset < size) { | |
199 | unsigned long tmp; | |
200 | ||
201 | addr += offset / BITS_PER_LONG; | |
202 | tmp = *addr >> (offset % BITS_PER_LONG); | |
203 | if (tmp) | |
204 | return __ffs(tmp) + offset; | |
205 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
206 | while (offset < size) { | |
207 | tmp = *++addr; | |
208 | if (tmp) | |
209 | return __ffs(tmp) + offset; | |
210 | offset += BITS_PER_LONG; | |
211 | } | |
212 | } | |
213 | return size; | |
214 | } | |
215 | ||
7cf19af4 MW |
216 | #if 0 |
217 | static void dump_node(void *slot, int height, int offset) | |
218 | { | |
219 | struct radix_tree_node *node; | |
220 | int i; | |
221 | ||
222 | if (!slot) | |
223 | return; | |
224 | ||
225 | if (height == 0) { | |
226 | pr_debug("radix entry %p offset %d\n", slot, offset); | |
227 | return; | |
228 | } | |
229 | ||
230 | node = indirect_to_ptr(slot); | |
231 | pr_debug("radix node: %p offset %d tags %lx %lx %lx path %x count %d parent %p\n", | |
232 | slot, offset, node->tags[0][0], node->tags[1][0], | |
233 | node->tags[2][0], node->path, node->count, node->parent); | |
234 | ||
235 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) | |
236 | dump_node(node->slots[i], height - 1, i); | |
237 | } | |
238 | ||
239 | /* For debug */ | |
240 | static void radix_tree_dump(struct radix_tree_root *root) | |
241 | { | |
242 | pr_debug("radix root: %p height %d rnode %p tags %x\n", | |
243 | root, root->height, root->rnode, | |
244 | root->gfp_mask >> __GFP_BITS_SHIFT); | |
245 | if (!radix_tree_is_indirect_ptr(root->rnode)) | |
246 | return; | |
247 | dump_node(root->rnode, root->height, 0); | |
248 | } | |
249 | #endif | |
250 | ||
1da177e4 LT |
251 | /* |
252 | * This assumes that the caller has performed appropriate preallocation, and | |
253 | * that the caller has pinned this thread of control to the current CPU. | |
254 | */ | |
255 | static struct radix_tree_node * | |
256 | radix_tree_node_alloc(struct radix_tree_root *root) | |
257 | { | |
e2848a0e | 258 | struct radix_tree_node *ret = NULL; |
612d6c19 | 259 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 260 | |
5e4c0d97 JK |
261 | /* |
262 | * Preload code isn't irq safe and it doesn't make sence to use | |
263 | * preloading in the interrupt anyway as all the allocations have to | |
264 | * be atomic. So just do normal allocation when in interrupt. | |
265 | */ | |
d0164adc | 266 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
267 | struct radix_tree_preload *rtp; |
268 | ||
58e698af VD |
269 | /* |
270 | * Even if the caller has preloaded, try to allocate from the | |
271 | * cache first for the new node to get accounted. | |
272 | */ | |
273 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
274 | gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN); | |
275 | if (ret) | |
276 | goto out; | |
277 | ||
e2848a0e NP |
278 | /* |
279 | * Provided the caller has preloaded here, we will always | |
280 | * succeed in getting a node here (and never reach | |
281 | * kmem_cache_alloc) | |
282 | */ | |
7c8e0181 | 283 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 284 | if (rtp->nr) { |
9d2a8da0 KS |
285 | ret = rtp->nodes; |
286 | rtp->nodes = ret->private_data; | |
287 | ret->private_data = NULL; | |
1da177e4 LT |
288 | rtp->nr--; |
289 | } | |
ce80b067 CM |
290 | /* |
291 | * Update the allocation stack trace as this is more useful | |
292 | * for debugging. | |
293 | */ | |
294 | kmemleak_update_trace(ret); | |
58e698af | 295 | goto out; |
1da177e4 | 296 | } |
58e698af VD |
297 | ret = kmem_cache_alloc(radix_tree_node_cachep, |
298 | gfp_mask | __GFP_ACCOUNT); | |
299 | out: | |
c0bc9875 | 300 | BUG_ON(radix_tree_is_indirect_ptr(ret)); |
1da177e4 LT |
301 | return ret; |
302 | } | |
303 | ||
7cf9c2c7 NP |
304 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
305 | { | |
306 | struct radix_tree_node *node = | |
307 | container_of(head, struct radix_tree_node, rcu_head); | |
b6dd0865 | 308 | int i; |
643b52b9 NP |
309 | |
310 | /* | |
311 | * must only free zeroed nodes into the slab. radix_tree_shrink | |
312 | * can leave us with a non-NULL entry in the first slot, so clear | |
313 | * that here to make sure. | |
314 | */ | |
b6dd0865 DC |
315 | for (i = 0; i < RADIX_TREE_MAX_TAGS; i++) |
316 | tag_clear(node, i, 0); | |
317 | ||
643b52b9 NP |
318 | node->slots[0] = NULL; |
319 | node->count = 0; | |
320 | ||
7cf9c2c7 NP |
321 | kmem_cache_free(radix_tree_node_cachep, node); |
322 | } | |
323 | ||
1da177e4 LT |
324 | static inline void |
325 | radix_tree_node_free(struct radix_tree_node *node) | |
326 | { | |
7cf9c2c7 | 327 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
328 | } |
329 | ||
330 | /* | |
331 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
332 | * ensure that the addition of a single element in the tree cannot fail. On | |
333 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
334 | * with preemption not disabled. | |
b34df792 DH |
335 | * |
336 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 337 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 338 | */ |
5e4c0d97 | 339 | static int __radix_tree_preload(gfp_t gfp_mask) |
1da177e4 LT |
340 | { |
341 | struct radix_tree_preload *rtp; | |
342 | struct radix_tree_node *node; | |
343 | int ret = -ENOMEM; | |
344 | ||
345 | preempt_disable(); | |
7c8e0181 | 346 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 | 347 | while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
1da177e4 | 348 | preempt_enable(); |
488514d1 | 349 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
350 | if (node == NULL) |
351 | goto out; | |
352 | preempt_disable(); | |
7c8e0181 | 353 | rtp = this_cpu_ptr(&radix_tree_preloads); |
9d2a8da0 KS |
354 | if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) { |
355 | node->private_data = rtp->nodes; | |
356 | rtp->nodes = node; | |
357 | rtp->nr++; | |
358 | } else { | |
1da177e4 | 359 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 360 | } |
1da177e4 LT |
361 | } |
362 | ret = 0; | |
363 | out: | |
364 | return ret; | |
365 | } | |
5e4c0d97 JK |
366 | |
367 | /* | |
368 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
369 | * ensure that the addition of a single element in the tree cannot fail. On | |
370 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
371 | * with preemption not disabled. | |
372 | * | |
373 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 374 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
375 | */ |
376 | int radix_tree_preload(gfp_t gfp_mask) | |
377 | { | |
378 | /* Warn on non-sensical use... */ | |
d0164adc | 379 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
5e4c0d97 JK |
380 | return __radix_tree_preload(gfp_mask); |
381 | } | |
d7f0923d | 382 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 383 | |
5e4c0d97 JK |
384 | /* |
385 | * The same as above function, except we don't guarantee preloading happens. | |
386 | * We do it, if we decide it helps. On success, return zero with preemption | |
387 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
388 | */ | |
389 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
390 | { | |
d0164adc | 391 | if (gfpflags_allow_blocking(gfp_mask)) |
5e4c0d97 JK |
392 | return __radix_tree_preload(gfp_mask); |
393 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
394 | preempt_disable(); | |
395 | return 0; | |
396 | } | |
397 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
398 | ||
1da177e4 LT |
399 | /* |
400 | * Return the maximum key which can be store into a | |
401 | * radix tree with height HEIGHT. | |
402 | */ | |
403 | static inline unsigned long radix_tree_maxindex(unsigned int height) | |
404 | { | |
405 | return height_to_maxindex[height]; | |
406 | } | |
407 | ||
408 | /* | |
409 | * Extend a radix tree so it can store key @index. | |
410 | */ | |
e6145236 MW |
411 | static int radix_tree_extend(struct radix_tree_root *root, |
412 | unsigned long index, unsigned order) | |
1da177e4 LT |
413 | { |
414 | struct radix_tree_node *node; | |
e2bdb933 | 415 | struct radix_tree_node *slot; |
1da177e4 | 416 | unsigned int height; |
1da177e4 LT |
417 | int tag; |
418 | ||
419 | /* Figure out what the height should be. */ | |
420 | height = root->height + 1; | |
421 | while (index > radix_tree_maxindex(height)) | |
422 | height++; | |
423 | ||
e6145236 | 424 | if ((root->rnode == NULL) && (order == 0)) { |
1da177e4 LT |
425 | root->height = height; |
426 | goto out; | |
427 | } | |
428 | ||
1da177e4 | 429 | do { |
7cf9c2c7 | 430 | unsigned int newheight; |
1da177e4 LT |
431 | if (!(node = radix_tree_node_alloc(root))) |
432 | return -ENOMEM; | |
433 | ||
1da177e4 | 434 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 435 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 436 | if (root_tag_get(root, tag)) |
1da177e4 LT |
437 | tag_set(node, tag, 0); |
438 | } | |
439 | ||
e2bdb933 | 440 | /* Increase the height. */ |
7cf9c2c7 | 441 | newheight = root->height+1; |
449dd698 JW |
442 | BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK); |
443 | node->path = newheight; | |
1da177e4 | 444 | node->count = 1; |
e2bdb933 HD |
445 | node->parent = NULL; |
446 | slot = root->rnode; | |
339e6353 | 447 | if (radix_tree_is_indirect_ptr(slot) && newheight > 1) { |
e2bdb933 HD |
448 | slot = indirect_to_ptr(slot); |
449 | slot->parent = node; | |
339e6353 | 450 | slot = ptr_to_indirect(slot); |
e2bdb933 HD |
451 | } |
452 | node->slots[0] = slot; | |
27d20fdd | 453 | node = ptr_to_indirect(node); |
7cf9c2c7 NP |
454 | rcu_assign_pointer(root->rnode, node); |
455 | root->height = newheight; | |
1da177e4 LT |
456 | } while (height > root->height); |
457 | out: | |
458 | return 0; | |
459 | } | |
460 | ||
461 | /** | |
139e5616 | 462 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
463 | * @root: radix tree root |
464 | * @index: index key | |
e6145236 | 465 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
466 | * @nodep: returns node |
467 | * @slotp: returns slot | |
1da177e4 | 468 | * |
139e5616 JW |
469 | * Create, if necessary, and return the node and slot for an item |
470 | * at position @index in the radix tree @root. | |
471 | * | |
472 | * Until there is more than one item in the tree, no nodes are | |
473 | * allocated and @root->rnode is used as a direct slot instead of | |
474 | * pointing to a node, in which case *@nodep will be NULL. | |
475 | * | |
476 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 477 | */ |
139e5616 | 478 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
479 | unsigned order, struct radix_tree_node **nodep, |
480 | void ***slotp) | |
1da177e4 | 481 | { |
201b6264 | 482 | struct radix_tree_node *node = NULL, *slot; |
139e5616 | 483 | unsigned int height, shift, offset; |
1da177e4 LT |
484 | int error; |
485 | ||
486 | /* Make sure the tree is high enough. */ | |
612d6c19 | 487 | if (index > radix_tree_maxindex(root->height)) { |
e6145236 | 488 | error = radix_tree_extend(root, index, order); |
1da177e4 LT |
489 | if (error) |
490 | return error; | |
491 | } | |
492 | ||
e6145236 | 493 | slot = root->rnode; |
c0bc9875 | 494 | |
1da177e4 | 495 | height = root->height; |
0070e28d | 496 | shift = height * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
497 | |
498 | offset = 0; /* uninitialised var warning */ | |
e6145236 | 499 | while (shift > order) { |
201b6264 | 500 | if (slot == NULL) { |
1da177e4 | 501 | /* Have to add a child node. */ |
201b6264 | 502 | if (!(slot = radix_tree_node_alloc(root))) |
1da177e4 | 503 | return -ENOMEM; |
449dd698 | 504 | slot->path = height; |
e2bdb933 | 505 | slot->parent = node; |
201b6264 | 506 | if (node) { |
339e6353 MW |
507 | rcu_assign_pointer(node->slots[offset], |
508 | ptr_to_indirect(slot)); | |
1da177e4 | 509 | node->count++; |
449dd698 | 510 | slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT; |
201b6264 | 511 | } else |
339e6353 MW |
512 | rcu_assign_pointer(root->rnode, |
513 | ptr_to_indirect(slot)); | |
e6145236 MW |
514 | } else if (!radix_tree_is_indirect_ptr(slot)) |
515 | break; | |
1da177e4 LT |
516 | |
517 | /* Go a level down */ | |
e6145236 | 518 | height--; |
0070e28d | 519 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 520 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e6145236 | 521 | node = indirect_to_ptr(slot); |
201b6264 | 522 | slot = node->slots[offset]; |
e6145236 MW |
523 | } |
524 | ||
57578c2e | 525 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
e6145236 | 526 | /* Insert pointers to the canonical entry */ |
3b8c00f6 MW |
527 | if (order > shift) { |
528 | int i, n = 1 << (order - shift); | |
e6145236 MW |
529 | offset = offset & ~(n - 1); |
530 | slot = ptr_to_indirect(&node->slots[offset]); | |
531 | for (i = 0; i < n; i++) { | |
532 | if (node->slots[offset + i]) | |
533 | return -EEXIST; | |
534 | } | |
535 | ||
536 | for (i = 1; i < n; i++) { | |
537 | rcu_assign_pointer(node->slots[offset + i], slot); | |
538 | node->count++; | |
539 | } | |
612d6c19 | 540 | } |
57578c2e | 541 | #endif |
1da177e4 | 542 | |
139e5616 JW |
543 | if (nodep) |
544 | *nodep = node; | |
545 | if (slotp) | |
546 | *slotp = node ? node->slots + offset : (void **)&root->rnode; | |
547 | return 0; | |
548 | } | |
549 | ||
550 | /** | |
e6145236 | 551 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
552 | * @root: radix tree root |
553 | * @index: index key | |
e6145236 | 554 | * @order: key covers the 2^order indices around index |
139e5616 JW |
555 | * @item: item to insert |
556 | * | |
557 | * Insert an item into the radix tree at position @index. | |
558 | */ | |
e6145236 MW |
559 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
560 | unsigned order, void *item) | |
139e5616 JW |
561 | { |
562 | struct radix_tree_node *node; | |
563 | void **slot; | |
564 | int error; | |
565 | ||
566 | BUG_ON(radix_tree_is_indirect_ptr(item)); | |
567 | ||
e6145236 | 568 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
569 | if (error) |
570 | return error; | |
571 | if (*slot != NULL) | |
1da177e4 | 572 | return -EEXIST; |
139e5616 | 573 | rcu_assign_pointer(*slot, item); |
201b6264 | 574 | |
612d6c19 NP |
575 | if (node) { |
576 | node->count++; | |
139e5616 JW |
577 | BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK)); |
578 | BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK)); | |
612d6c19 | 579 | } else { |
612d6c19 NP |
580 | BUG_ON(root_tag_get(root, 0)); |
581 | BUG_ON(root_tag_get(root, 1)); | |
582 | } | |
1da177e4 | 583 | |
1da177e4 LT |
584 | return 0; |
585 | } | |
e6145236 | 586 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 587 | |
139e5616 JW |
588 | /** |
589 | * __radix_tree_lookup - lookup an item in a radix tree | |
590 | * @root: radix tree root | |
591 | * @index: index key | |
592 | * @nodep: returns node | |
593 | * @slotp: returns slot | |
594 | * | |
595 | * Lookup and return the item at position @index in the radix | |
596 | * tree @root. | |
597 | * | |
598 | * Until there is more than one item in the tree, no nodes are | |
599 | * allocated and @root->rnode is used as a direct slot instead of | |
600 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 601 | */ |
139e5616 JW |
602 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
603 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 604 | { |
139e5616 | 605 | struct radix_tree_node *node, *parent; |
1da177e4 | 606 | unsigned int height, shift; |
139e5616 | 607 | void **slot; |
612d6c19 | 608 | |
2676a58c | 609 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 | 610 | if (node == NULL) |
1da177e4 LT |
611 | return NULL; |
612 | ||
c0bc9875 | 613 | if (!radix_tree_is_indirect_ptr(node)) { |
7cf9c2c7 NP |
614 | if (index > 0) |
615 | return NULL; | |
139e5616 JW |
616 | |
617 | if (nodep) | |
618 | *nodep = NULL; | |
619 | if (slotp) | |
620 | *slotp = (void **)&root->rnode; | |
621 | return node; | |
7cf9c2c7 | 622 | } |
27d20fdd | 623 | node = indirect_to_ptr(node); |
7cf9c2c7 | 624 | |
449dd698 | 625 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
626 | if (index > radix_tree_maxindex(height)) |
627 | return NULL; | |
612d6c19 | 628 | |
1da177e4 | 629 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 | 630 | |
7cf9c2c7 | 631 | do { |
139e5616 JW |
632 | parent = node; |
633 | slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK); | |
2676a58c | 634 | node = rcu_dereference_raw(*slot); |
7cf9c2c7 | 635 | if (node == NULL) |
1da177e4 | 636 | return NULL; |
e6145236 MW |
637 | if (!radix_tree_is_indirect_ptr(node)) |
638 | break; | |
339e6353 | 639 | node = indirect_to_ptr(node); |
1da177e4 | 640 | |
1da177e4 LT |
641 | shift -= RADIX_TREE_MAP_SHIFT; |
642 | height--; | |
7cf9c2c7 | 643 | } while (height > 0); |
1da177e4 | 644 | |
139e5616 JW |
645 | if (nodep) |
646 | *nodep = parent; | |
647 | if (slotp) | |
648 | *slotp = slot; | |
649 | return node; | |
b72b71c6 HS |
650 | } |
651 | ||
652 | /** | |
653 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
654 | * @root: radix tree root | |
655 | * @index: index key | |
656 | * | |
657 | * Returns: the slot corresponding to the position @index in the | |
658 | * radix tree @root. This is useful for update-if-exists operations. | |
659 | * | |
660 | * This function can be called under rcu_read_lock iff the slot is not | |
661 | * modified by radix_tree_replace_slot, otherwise it must be called | |
662 | * exclusive from other writers. Any dereference of the slot must be done | |
663 | * using radix_tree_deref_slot. | |
664 | */ | |
665 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
666 | { | |
139e5616 JW |
667 | void **slot; |
668 | ||
669 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
670 | return NULL; | |
671 | return slot; | |
a4331366 | 672 | } |
a4331366 HR |
673 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
674 | ||
675 | /** | |
676 | * radix_tree_lookup - perform lookup operation on a radix tree | |
677 | * @root: radix tree root | |
678 | * @index: index key | |
679 | * | |
680 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
681 | * |
682 | * This function can be called under rcu_read_lock, however the caller | |
683 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
684 | * them safely). No RCU barriers are required to access or modify the | |
685 | * returned item, however. | |
a4331366 HR |
686 | */ |
687 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
688 | { | |
139e5616 | 689 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
690 | } |
691 | EXPORT_SYMBOL(radix_tree_lookup); | |
692 | ||
693 | /** | |
694 | * radix_tree_tag_set - set a tag on a radix tree node | |
695 | * @root: radix tree root | |
696 | * @index: index key | |
697 | * @tag: tag index | |
698 | * | |
daff89f3 JC |
699 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
700 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
701 | * the root all the way down to the leaf node. |
702 | * | |
703 | * Returns the address of the tagged item. Setting a tag on a not-present | |
704 | * item is a bug. | |
705 | */ | |
706 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 707 | unsigned long index, unsigned int tag) |
1da177e4 LT |
708 | { |
709 | unsigned int height, shift; | |
201b6264 | 710 | struct radix_tree_node *slot; |
1da177e4 LT |
711 | |
712 | height = root->height; | |
4c91c364 | 713 | BUG_ON(index > radix_tree_maxindex(height)); |
1da177e4 | 714 | |
27d20fdd | 715 | slot = indirect_to_ptr(root->rnode); |
612d6c19 | 716 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
717 | |
718 | while (height > 0) { | |
719 | int offset; | |
720 | ||
721 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
d5274261 NP |
722 | if (!tag_get(slot, tag, offset)) |
723 | tag_set(slot, tag, offset); | |
201b6264 CL |
724 | slot = slot->slots[offset]; |
725 | BUG_ON(slot == NULL); | |
e6145236 MW |
726 | if (!radix_tree_is_indirect_ptr(slot)) |
727 | break; | |
339e6353 | 728 | slot = indirect_to_ptr(slot); |
1da177e4 LT |
729 | shift -= RADIX_TREE_MAP_SHIFT; |
730 | height--; | |
731 | } | |
732 | ||
612d6c19 NP |
733 | /* set the root's tag bit */ |
734 | if (slot && !root_tag_get(root, tag)) | |
735 | root_tag_set(root, tag); | |
736 | ||
201b6264 | 737 | return slot; |
1da177e4 LT |
738 | } |
739 | EXPORT_SYMBOL(radix_tree_tag_set); | |
740 | ||
741 | /** | |
742 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
743 | * @root: radix tree root | |
744 | * @index: index key | |
745 | * @tag: tag index | |
746 | * | |
daff89f3 JC |
747 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
748 | * corresponding to @index in the radix tree. If | |
1da177e4 LT |
749 | * this causes the leaf node to have no tags set then clear the tag in the |
750 | * next-to-leaf node, etc. | |
751 | * | |
752 | * Returns the address of the tagged item on success, else NULL. ie: | |
753 | * has the same return value and semantics as radix_tree_lookup(). | |
754 | */ | |
755 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 756 | unsigned long index, unsigned int tag) |
1da177e4 | 757 | { |
e2bdb933 | 758 | struct radix_tree_node *node = NULL; |
612d6c19 | 759 | struct radix_tree_node *slot = NULL; |
1da177e4 | 760 | unsigned int height, shift; |
e2bdb933 | 761 | int uninitialized_var(offset); |
1da177e4 LT |
762 | |
763 | height = root->height; | |
764 | if (index > radix_tree_maxindex(height)) | |
765 | goto out; | |
766 | ||
e2bdb933 | 767 | shift = height * RADIX_TREE_MAP_SHIFT; |
339e6353 | 768 | slot = root->rnode; |
1da177e4 | 769 | |
e2bdb933 | 770 | while (shift) { |
201b6264 | 771 | if (slot == NULL) |
1da177e4 | 772 | goto out; |
e6145236 MW |
773 | if (!radix_tree_is_indirect_ptr(slot)) |
774 | break; | |
339e6353 | 775 | slot = indirect_to_ptr(slot); |
1da177e4 | 776 | |
e2bdb933 | 777 | shift -= RADIX_TREE_MAP_SHIFT; |
1da177e4 | 778 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; |
e2bdb933 | 779 | node = slot; |
201b6264 | 780 | slot = slot->slots[offset]; |
1da177e4 LT |
781 | } |
782 | ||
612d6c19 | 783 | if (slot == NULL) |
1da177e4 LT |
784 | goto out; |
785 | ||
e2bdb933 HD |
786 | while (node) { |
787 | if (!tag_get(node, tag, offset)) | |
d5274261 | 788 | goto out; |
e2bdb933 HD |
789 | tag_clear(node, tag, offset); |
790 | if (any_tag_set(node, tag)) | |
6e954b9e | 791 | goto out; |
e2bdb933 HD |
792 | |
793 | index >>= RADIX_TREE_MAP_SHIFT; | |
794 | offset = index & RADIX_TREE_MAP_MASK; | |
795 | node = node->parent; | |
612d6c19 NP |
796 | } |
797 | ||
798 | /* clear the root's tag bit */ | |
799 | if (root_tag_get(root, tag)) | |
800 | root_tag_clear(root, tag); | |
801 | ||
1da177e4 | 802 | out: |
612d6c19 | 803 | return slot; |
1da177e4 LT |
804 | } |
805 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
806 | ||
1da177e4 | 807 | /** |
32605a18 MT |
808 | * radix_tree_tag_get - get a tag on a radix tree node |
809 | * @root: radix tree root | |
810 | * @index: index key | |
daff89f3 | 811 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 812 | * |
32605a18 | 813 | * Return values: |
1da177e4 | 814 | * |
612d6c19 NP |
815 | * 0: tag not present or not set |
816 | * 1: tag set | |
ce82653d DH |
817 | * |
818 | * Note that the return value of this function may not be relied on, even if | |
819 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
820 | * from concurrency. | |
1da177e4 LT |
821 | */ |
822 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 823 | unsigned long index, unsigned int tag) |
1da177e4 LT |
824 | { |
825 | unsigned int height, shift; | |
7cf9c2c7 | 826 | struct radix_tree_node *node; |
1da177e4 | 827 | |
612d6c19 NP |
828 | /* check the root's tag bit */ |
829 | if (!root_tag_get(root, tag)) | |
830 | return 0; | |
831 | ||
2676a58c | 832 | node = rcu_dereference_raw(root->rnode); |
7cf9c2c7 NP |
833 | if (node == NULL) |
834 | return 0; | |
835 | ||
c0bc9875 | 836 | if (!radix_tree_is_indirect_ptr(node)) |
7cf9c2c7 | 837 | return (index == 0); |
27d20fdd | 838 | node = indirect_to_ptr(node); |
7cf9c2c7 | 839 | |
449dd698 | 840 | height = node->path & RADIX_TREE_HEIGHT_MASK; |
7cf9c2c7 NP |
841 | if (index > radix_tree_maxindex(height)) |
842 | return 0; | |
612d6c19 | 843 | |
1da177e4 | 844 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
845 | |
846 | for ( ; ; ) { | |
847 | int offset; | |
848 | ||
7cf9c2c7 | 849 | if (node == NULL) |
1da177e4 | 850 | return 0; |
339e6353 | 851 | node = indirect_to_ptr(node); |
1da177e4 LT |
852 | |
853 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
7cf9c2c7 | 854 | if (!tag_get(node, tag, offset)) |
3fa36acb | 855 | return 0; |
ce82653d | 856 | if (height == 1) |
3fa36acb | 857 | return 1; |
2676a58c | 858 | node = rcu_dereference_raw(node->slots[offset]); |
e6145236 MW |
859 | if (!radix_tree_is_indirect_ptr(node)) |
860 | return 1; | |
1da177e4 LT |
861 | shift -= RADIX_TREE_MAP_SHIFT; |
862 | height--; | |
863 | } | |
864 | } | |
865 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 866 | |
78c1d784 KK |
867 | /** |
868 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
869 | * | |
870 | * @root: radix tree root | |
871 | * @iter: iterator state | |
872 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
873 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
874 | */ | |
875 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
876 | struct radix_tree_iter *iter, unsigned flags) | |
877 | { | |
878 | unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
879 | struct radix_tree_node *rnode, *node; | |
449dd698 | 880 | unsigned long index, offset, height; |
78c1d784 KK |
881 | |
882 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
883 | return NULL; | |
884 | ||
885 | /* | |
886 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
887 | * during iterating; it can be zero only at the beginning. | |
888 | * And we cannot overflow iter->next_index in a single step, | |
889 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
890 | * |
891 | * This condition also used by radix_tree_next_slot() to stop | |
892 | * contiguous iterating, and forbid swithing to the next chunk. | |
78c1d784 KK |
893 | */ |
894 | index = iter->next_index; | |
895 | if (!index && iter->index) | |
896 | return NULL; | |
897 | ||
898 | rnode = rcu_dereference_raw(root->rnode); | |
899 | if (radix_tree_is_indirect_ptr(rnode)) { | |
900 | rnode = indirect_to_ptr(rnode); | |
901 | } else if (rnode && !index) { | |
902 | /* Single-slot tree */ | |
903 | iter->index = 0; | |
904 | iter->next_index = 1; | |
905 | iter->tags = 1; | |
906 | return (void **)&root->rnode; | |
907 | } else | |
908 | return NULL; | |
909 | ||
910 | restart: | |
449dd698 JW |
911 | height = rnode->path & RADIX_TREE_HEIGHT_MASK; |
912 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
78c1d784 KK |
913 | offset = index >> shift; |
914 | ||
915 | /* Index outside of the tree */ | |
916 | if (offset >= RADIX_TREE_MAP_SIZE) | |
917 | return NULL; | |
918 | ||
919 | node = rnode; | |
920 | while (1) { | |
e6145236 | 921 | struct radix_tree_node *slot; |
78c1d784 KK |
922 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
923 | !test_bit(offset, node->tags[tag]) : | |
924 | !node->slots[offset]) { | |
925 | /* Hole detected */ | |
926 | if (flags & RADIX_TREE_ITER_CONTIG) | |
927 | return NULL; | |
928 | ||
929 | if (flags & RADIX_TREE_ITER_TAGGED) | |
930 | offset = radix_tree_find_next_bit( | |
931 | node->tags[tag], | |
932 | RADIX_TREE_MAP_SIZE, | |
933 | offset + 1); | |
934 | else | |
935 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
936 | if (node->slots[offset]) | |
937 | break; | |
938 | } | |
939 | index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1); | |
940 | index += offset << shift; | |
941 | /* Overflow after ~0UL */ | |
942 | if (!index) | |
943 | return NULL; | |
944 | if (offset == RADIX_TREE_MAP_SIZE) | |
945 | goto restart; | |
946 | } | |
947 | ||
948 | /* This is leaf-node */ | |
949 | if (!shift) | |
950 | break; | |
951 | ||
e6145236 MW |
952 | slot = rcu_dereference_raw(node->slots[offset]); |
953 | if (slot == NULL) | |
78c1d784 | 954 | goto restart; |
e6145236 MW |
955 | if (!radix_tree_is_indirect_ptr(slot)) |
956 | break; | |
957 | node = indirect_to_ptr(slot); | |
78c1d784 KK |
958 | shift -= RADIX_TREE_MAP_SHIFT; |
959 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
960 | } | |
961 | ||
962 | /* Update the iterator state */ | |
963 | iter->index = index; | |
964 | iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1; | |
965 | ||
966 | /* Construct iter->tags bit-mask from node->tags[tag] array */ | |
967 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
968 | unsigned tag_long, tag_bit; | |
969 | ||
970 | tag_long = offset / BITS_PER_LONG; | |
971 | tag_bit = offset % BITS_PER_LONG; | |
972 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
973 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
974 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
975 | /* Pick tags from next element */ | |
976 | if (tag_bit) | |
977 | iter->tags |= node->tags[tag][tag_long + 1] << | |
978 | (BITS_PER_LONG - tag_bit); | |
979 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
980 | iter->next_index = index + BITS_PER_LONG; | |
981 | } | |
982 | } | |
983 | ||
984 | return node->slots + offset; | |
985 | } | |
986 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
987 | ||
ebf8aa44 JK |
988 | /** |
989 | * radix_tree_range_tag_if_tagged - for each item in given range set given | |
990 | * tag if item has another tag set | |
991 | * @root: radix tree root | |
992 | * @first_indexp: pointer to a starting index of a range to scan | |
993 | * @last_index: last index of a range to scan | |
994 | * @nr_to_tag: maximum number items to tag | |
995 | * @iftag: tag index to test | |
996 | * @settag: tag index to set if tested tag is set | |
997 | * | |
998 | * This function scans range of radix tree from first_index to last_index | |
999 | * (inclusive). For each item in the range if iftag is set, the function sets | |
1000 | * also settag. The function stops either after tagging nr_to_tag items or | |
1001 | * after reaching last_index. | |
1002 | * | |
144dcfc0 DC |
1003 | * The tags must be set from the leaf level only and propagated back up the |
1004 | * path to the root. We must do this so that we resolve the full path before | |
1005 | * setting any tags on intermediate nodes. If we set tags as we descend, then | |
1006 | * we can get to the leaf node and find that the index that has the iftag | |
1007 | * set is outside the range we are scanning. This reults in dangling tags and | |
1008 | * can lead to problems with later tag operations (e.g. livelocks on lookups). | |
1009 | * | |
ebf8aa44 JK |
1010 | * The function returns number of leaves where the tag was set and sets |
1011 | * *first_indexp to the first unscanned index. | |
d5ed3a4a JK |
1012 | * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must |
1013 | * be prepared to handle that. | |
ebf8aa44 JK |
1014 | */ |
1015 | unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root, | |
1016 | unsigned long *first_indexp, unsigned long last_index, | |
1017 | unsigned long nr_to_tag, | |
1018 | unsigned int iftag, unsigned int settag) | |
1019 | { | |
144dcfc0 | 1020 | unsigned int height = root->height; |
e2bdb933 | 1021 | struct radix_tree_node *node = NULL; |
144dcfc0 DC |
1022 | struct radix_tree_node *slot; |
1023 | unsigned int shift; | |
1024 | unsigned long tagged = 0; | |
1025 | unsigned long index = *first_indexp; | |
ebf8aa44 JK |
1026 | |
1027 | last_index = min(last_index, radix_tree_maxindex(height)); | |
1028 | if (index > last_index) | |
1029 | return 0; | |
1030 | if (!nr_to_tag) | |
1031 | return 0; | |
1032 | if (!root_tag_get(root, iftag)) { | |
1033 | *first_indexp = last_index + 1; | |
1034 | return 0; | |
1035 | } | |
1036 | if (height == 0) { | |
1037 | *first_indexp = last_index + 1; | |
1038 | root_tag_set(root, settag); | |
1039 | return 1; | |
1040 | } | |
1041 | ||
1042 | shift = (height - 1) * RADIX_TREE_MAP_SHIFT; | |
27d20fdd | 1043 | slot = indirect_to_ptr(root->rnode); |
ebf8aa44 JK |
1044 | |
1045 | for (;;) { | |
e2bdb933 | 1046 | unsigned long upindex; |
ebf8aa44 JK |
1047 | int offset; |
1048 | ||
1049 | offset = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1050 | if (!slot->slots[offset]) | |
1051 | goto next; | |
1052 | if (!tag_get(slot, iftag, offset)) | |
1053 | goto next; | |
e2bdb933 | 1054 | if (shift) { |
e2bdb933 | 1055 | node = slot; |
144dcfc0 | 1056 | slot = slot->slots[offset]; |
e6145236 MW |
1057 | if (radix_tree_is_indirect_ptr(slot)) { |
1058 | slot = indirect_to_ptr(slot); | |
1059 | shift -= RADIX_TREE_MAP_SHIFT; | |
1060 | continue; | |
1061 | } else { | |
1062 | slot = node; | |
1063 | node = node->parent; | |
1064 | } | |
144dcfc0 DC |
1065 | } |
1066 | ||
1067 | /* tag the leaf */ | |
e6145236 | 1068 | tagged += 1 << shift; |
ebf8aa44 | 1069 | tag_set(slot, settag, offset); |
144dcfc0 DC |
1070 | |
1071 | /* walk back up the path tagging interior nodes */ | |
e2bdb933 HD |
1072 | upindex = index; |
1073 | while (node) { | |
1074 | upindex >>= RADIX_TREE_MAP_SHIFT; | |
1075 | offset = upindex & RADIX_TREE_MAP_MASK; | |
1076 | ||
144dcfc0 | 1077 | /* stop if we find a node with the tag already set */ |
e2bdb933 | 1078 | if (tag_get(node, settag, offset)) |
144dcfc0 | 1079 | break; |
e2bdb933 HD |
1080 | tag_set(node, settag, offset); |
1081 | node = node->parent; | |
ebf8aa44 | 1082 | } |
144dcfc0 | 1083 | |
e2bdb933 HD |
1084 | /* |
1085 | * Small optimization: now clear that node pointer. | |
1086 | * Since all of this slot's ancestors now have the tag set | |
1087 | * from setting it above, we have no further need to walk | |
1088 | * back up the tree setting tags, until we update slot to | |
1089 | * point to another radix_tree_node. | |
1090 | */ | |
1091 | node = NULL; | |
1092 | ||
ebf8aa44 JK |
1093 | next: |
1094 | /* Go to next item at level determined by 'shift' */ | |
1095 | index = ((index >> shift) + 1) << shift; | |
d5ed3a4a JK |
1096 | /* Overflow can happen when last_index is ~0UL... */ |
1097 | if (index > last_index || !index) | |
ebf8aa44 JK |
1098 | break; |
1099 | if (tagged >= nr_to_tag) | |
1100 | break; | |
1101 | while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) { | |
1102 | /* | |
1103 | * We've fully scanned this node. Go up. Because | |
1104 | * last_index is guaranteed to be in the tree, what | |
1105 | * we do below cannot wander astray. | |
1106 | */ | |
e2bdb933 | 1107 | slot = slot->parent; |
ebf8aa44 JK |
1108 | shift += RADIX_TREE_MAP_SHIFT; |
1109 | } | |
1110 | } | |
1111 | /* | |
ac15ee69 TO |
1112 | * We need not to tag the root tag if there is no tag which is set with |
1113 | * settag within the range from *first_indexp to last_index. | |
ebf8aa44 | 1114 | */ |
ac15ee69 TO |
1115 | if (tagged > 0) |
1116 | root_tag_set(root, settag); | |
ebf8aa44 JK |
1117 | *first_indexp = index; |
1118 | ||
1119 | return tagged; | |
1120 | } | |
1121 | EXPORT_SYMBOL(radix_tree_range_tag_if_tagged); | |
1122 | ||
1da177e4 LT |
1123 | /** |
1124 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1125 | * @root: radix tree root | |
1126 | * @results: where the results of the lookup are placed | |
1127 | * @first_index: start the lookup from this key | |
1128 | * @max_items: place up to this many items at *results | |
1129 | * | |
1130 | * Performs an index-ascending scan of the tree for present items. Places | |
1131 | * them at *@results and returns the number of items which were placed at | |
1132 | * *@results. | |
1133 | * | |
1134 | * The implementation is naive. | |
7cf9c2c7 NP |
1135 | * |
1136 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1137 | * rcu_read_lock. In this case, rather than the returned results being | |
1138 | * an atomic snapshot of the tree at a single point in time, the semantics | |
1139 | * of an RCU protected gang lookup are as though multiple radix_tree_lookups | |
1140 | * have been issued in individual locks, and results stored in 'results'. | |
1da177e4 LT |
1141 | */ |
1142 | unsigned int | |
1143 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1144 | unsigned long first_index, unsigned int max_items) | |
1145 | { | |
cebbd29e KK |
1146 | struct radix_tree_iter iter; |
1147 | void **slot; | |
1148 | unsigned int ret = 0; | |
7cf9c2c7 | 1149 | |
cebbd29e | 1150 | if (unlikely(!max_items)) |
7cf9c2c7 | 1151 | return 0; |
1da177e4 | 1152 | |
cebbd29e | 1153 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1154 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1155 | if (!results[ret]) |
1156 | continue; | |
46437f9a MW |
1157 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1158 | slot = radix_tree_iter_retry(&iter); | |
1159 | continue; | |
1160 | } | |
cebbd29e | 1161 | if (++ret == max_items) |
1da177e4 | 1162 | break; |
1da177e4 | 1163 | } |
7cf9c2c7 | 1164 | |
1da177e4 LT |
1165 | return ret; |
1166 | } | |
1167 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1168 | ||
47feff2c NP |
1169 | /** |
1170 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1171 | * @root: radix tree root | |
1172 | * @results: where the results of the lookup are placed | |
6328650b | 1173 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
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 | * their slots at *@results and returns the number of items which were | |
1179 | * placed at *@results. | |
1180 | * | |
1181 | * The implementation is naive. | |
1182 | * | |
1183 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1184 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1185 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1186 | */ | |
1187 | unsigned int | |
6328650b HD |
1188 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1189 | void ***results, unsigned long *indices, | |
47feff2c NP |
1190 | unsigned long first_index, unsigned int max_items) |
1191 | { | |
cebbd29e KK |
1192 | struct radix_tree_iter iter; |
1193 | void **slot; | |
1194 | unsigned int ret = 0; | |
47feff2c | 1195 | |
cebbd29e | 1196 | if (unlikely(!max_items)) |
47feff2c NP |
1197 | return 0; |
1198 | ||
cebbd29e KK |
1199 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1200 | results[ret] = slot; | |
6328650b | 1201 | if (indices) |
cebbd29e KK |
1202 | indices[ret] = iter.index; |
1203 | if (++ret == max_items) | |
47feff2c | 1204 | break; |
47feff2c NP |
1205 | } |
1206 | ||
1207 | return ret; | |
1208 | } | |
1209 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1210 | ||
1da177e4 LT |
1211 | /** |
1212 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1213 | * based on a tag | |
1214 | * @root: radix tree root | |
1215 | * @results: where the results of the lookup are placed | |
1216 | * @first_index: start the lookup from this key | |
1217 | * @max_items: place up to this many items at *results | |
daff89f3 | 1218 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1219 | * |
1220 | * Performs an index-ascending scan of the tree for present items which | |
1221 | * have the tag indexed by @tag set. Places the items at *@results and | |
1222 | * returns the number of items which were placed at *@results. | |
1223 | */ | |
1224 | unsigned int | |
1225 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1226 | unsigned long first_index, unsigned int max_items, |
1227 | unsigned int tag) | |
1da177e4 | 1228 | { |
cebbd29e KK |
1229 | struct radix_tree_iter iter; |
1230 | void **slot; | |
1231 | unsigned int ret = 0; | |
612d6c19 | 1232 | |
cebbd29e | 1233 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1234 | return 0; |
1235 | ||
cebbd29e | 1236 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1237 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1238 | if (!results[ret]) |
1239 | continue; | |
46437f9a MW |
1240 | if (radix_tree_is_indirect_ptr(results[ret])) { |
1241 | slot = radix_tree_iter_retry(&iter); | |
1242 | continue; | |
1243 | } | |
cebbd29e | 1244 | if (++ret == max_items) |
1da177e4 | 1245 | break; |
1da177e4 | 1246 | } |
7cf9c2c7 | 1247 | |
1da177e4 LT |
1248 | return ret; |
1249 | } | |
1250 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1251 | ||
47feff2c NP |
1252 | /** |
1253 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1254 | * radix tree based on a tag | |
1255 | * @root: radix tree root | |
1256 | * @results: where the results of the lookup are placed | |
1257 | * @first_index: start the lookup from this key | |
1258 | * @max_items: place up to this many items at *results | |
1259 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1260 | * | |
1261 | * Performs an index-ascending scan of the tree for present items which | |
1262 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1263 | * returns the number of slots which were placed at *@results. | |
1264 | */ | |
1265 | unsigned int | |
1266 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1267 | unsigned long first_index, unsigned int max_items, | |
1268 | unsigned int tag) | |
1269 | { | |
cebbd29e KK |
1270 | struct radix_tree_iter iter; |
1271 | void **slot; | |
1272 | unsigned int ret = 0; | |
47feff2c | 1273 | |
cebbd29e | 1274 | if (unlikely(!max_items)) |
47feff2c NP |
1275 | return 0; |
1276 | ||
cebbd29e KK |
1277 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1278 | results[ret] = slot; | |
1279 | if (++ret == max_items) | |
47feff2c | 1280 | break; |
47feff2c NP |
1281 | } |
1282 | ||
1283 | return ret; | |
1284 | } | |
1285 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1286 | ||
e504f3fd HD |
1287 | #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP) |
1288 | #include <linux/sched.h> /* for cond_resched() */ | |
1289 | ||
1290 | /* | |
1291 | * This linear search is at present only useful to shmem_unuse_inode(). | |
1292 | */ | |
1293 | static unsigned long __locate(struct radix_tree_node *slot, void *item, | |
1294 | unsigned long index, unsigned long *found_index) | |
1295 | { | |
1296 | unsigned int shift, height; | |
1297 | unsigned long i; | |
1298 | ||
449dd698 | 1299 | height = slot->path & RADIX_TREE_HEIGHT_MASK; |
e504f3fd HD |
1300 | shift = (height-1) * RADIX_TREE_MAP_SHIFT; |
1301 | ||
1302 | for ( ; height > 1; height--) { | |
1303 | i = (index >> shift) & RADIX_TREE_MAP_MASK; | |
1304 | for (;;) { | |
1305 | if (slot->slots[i] != NULL) | |
1306 | break; | |
1307 | index &= ~((1UL << shift) - 1); | |
1308 | index += 1UL << shift; | |
1309 | if (index == 0) | |
1310 | goto out; /* 32-bit wraparound */ | |
1311 | i++; | |
1312 | if (i == RADIX_TREE_MAP_SIZE) | |
1313 | goto out; | |
1314 | } | |
1315 | ||
e504f3fd HD |
1316 | slot = rcu_dereference_raw(slot->slots[i]); |
1317 | if (slot == NULL) | |
1318 | goto out; | |
e6145236 MW |
1319 | if (!radix_tree_is_indirect_ptr(slot)) { |
1320 | if (slot == item) { | |
1321 | *found_index = index + i; | |
1322 | index = 0; | |
1323 | } else { | |
1324 | index += shift; | |
1325 | } | |
1326 | goto out; | |
1327 | } | |
339e6353 | 1328 | slot = indirect_to_ptr(slot); |
e6145236 | 1329 | shift -= RADIX_TREE_MAP_SHIFT; |
e504f3fd HD |
1330 | } |
1331 | ||
1332 | /* Bottom level: check items */ | |
1333 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
1334 | if (slot->slots[i] == item) { | |
1335 | *found_index = index + i; | |
1336 | index = 0; | |
1337 | goto out; | |
1338 | } | |
1339 | } | |
1340 | index += RADIX_TREE_MAP_SIZE; | |
1341 | out: | |
1342 | return index; | |
1343 | } | |
1344 | ||
1345 | /** | |
1346 | * radix_tree_locate_item - search through radix tree for item | |
1347 | * @root: radix tree root | |
1348 | * @item: item to be found | |
1349 | * | |
1350 | * Returns index where item was found, or -1 if not found. | |
1351 | * Caller must hold no lock (since this time-consuming function needs | |
1352 | * to be preemptible), and must check afterwards if item is still there. | |
1353 | */ | |
1354 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1355 | { | |
1356 | struct radix_tree_node *node; | |
1357 | unsigned long max_index; | |
1358 | unsigned long cur_index = 0; | |
1359 | unsigned long found_index = -1; | |
1360 | ||
1361 | do { | |
1362 | rcu_read_lock(); | |
1363 | node = rcu_dereference_raw(root->rnode); | |
1364 | if (!radix_tree_is_indirect_ptr(node)) { | |
1365 | rcu_read_unlock(); | |
1366 | if (node == item) | |
1367 | found_index = 0; | |
1368 | break; | |
1369 | } | |
1370 | ||
1371 | node = indirect_to_ptr(node); | |
449dd698 JW |
1372 | max_index = radix_tree_maxindex(node->path & |
1373 | RADIX_TREE_HEIGHT_MASK); | |
5f30fc94 HD |
1374 | if (cur_index > max_index) { |
1375 | rcu_read_unlock(); | |
e504f3fd | 1376 | break; |
5f30fc94 | 1377 | } |
e504f3fd HD |
1378 | |
1379 | cur_index = __locate(node, item, cur_index, &found_index); | |
1380 | rcu_read_unlock(); | |
1381 | cond_resched(); | |
1382 | } while (cur_index != 0 && cur_index <= max_index); | |
1383 | ||
1384 | return found_index; | |
1385 | } | |
1386 | #else | |
1387 | unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item) | |
1388 | { | |
1389 | return -1; | |
1390 | } | |
1391 | #endif /* CONFIG_SHMEM && CONFIG_SWAP */ | |
47feff2c | 1392 | |
a5f51c96 NP |
1393 | /** |
1394 | * radix_tree_shrink - shrink height of a radix tree to minimal | |
1395 | * @root radix tree root | |
1396 | */ | |
1397 | static inline void radix_tree_shrink(struct radix_tree_root *root) | |
1398 | { | |
1399 | /* try to shrink tree height */ | |
c0bc9875 | 1400 | while (root->height > 0) { |
a5f51c96 | 1401 | struct radix_tree_node *to_free = root->rnode; |
e2bdb933 | 1402 | struct radix_tree_node *slot; |
a5f51c96 | 1403 | |
c0bc9875 | 1404 | BUG_ON(!radix_tree_is_indirect_ptr(to_free)); |
27d20fdd | 1405 | to_free = indirect_to_ptr(to_free); |
c0bc9875 NP |
1406 | |
1407 | /* | |
1408 | * The candidate node has more than one child, or its child | |
e6145236 MW |
1409 | * is not at the leftmost slot, or it is a multiorder entry, |
1410 | * we cannot shrink. | |
c0bc9875 NP |
1411 | */ |
1412 | if (to_free->count != 1) | |
1413 | break; | |
339e6353 MW |
1414 | slot = to_free->slots[0]; |
1415 | if (!slot) | |
c0bc9875 NP |
1416 | break; |
1417 | ||
7cf9c2c7 NP |
1418 | /* |
1419 | * We don't need rcu_assign_pointer(), since we are simply | |
27d20fdd NP |
1420 | * moving the node from one part of the tree to another: if it |
1421 | * was safe to dereference the old pointer to it | |
7cf9c2c7 | 1422 | * (to_free->slots[0]), it will be safe to dereference the new |
27d20fdd | 1423 | * one (root->rnode) as far as dependent read barriers go. |
7cf9c2c7 | 1424 | */ |
e2bdb933 | 1425 | if (root->height > 1) { |
e6145236 MW |
1426 | if (!radix_tree_is_indirect_ptr(slot)) |
1427 | break; | |
1428 | ||
339e6353 | 1429 | slot = indirect_to_ptr(slot); |
e2bdb933 HD |
1430 | slot->parent = NULL; |
1431 | slot = ptr_to_indirect(slot); | |
1432 | } | |
1433 | root->rnode = slot; | |
a5f51c96 | 1434 | root->height--; |
27d20fdd NP |
1435 | |
1436 | /* | |
1437 | * We have a dilemma here. The node's slot[0] must not be | |
1438 | * NULLed in case there are concurrent lookups expecting to | |
1439 | * find the item. However if this was a bottom-level node, | |
1440 | * then it may be subject to the slot pointer being visible | |
1441 | * to callers dereferencing it. If item corresponding to | |
1442 | * slot[0] is subsequently deleted, these callers would expect | |
1443 | * their slot to become empty sooner or later. | |
1444 | * | |
1445 | * For example, lockless pagecache will look up a slot, deref | |
1446 | * the page pointer, and if the page is 0 refcount it means it | |
1447 | * was concurrently deleted from pagecache so try the deref | |
1448 | * again. Fortunately there is already a requirement for logic | |
1449 | * to retry the entire slot lookup -- the indirect pointer | |
1450 | * problem (replacing direct root node with an indirect pointer | |
1451 | * also results in a stale slot). So tag the slot as indirect | |
1452 | * to force callers to retry. | |
1453 | */ | |
1454 | if (root->height == 0) | |
1455 | *((unsigned long *)&to_free->slots[0]) |= | |
1456 | RADIX_TREE_INDIRECT_PTR; | |
1457 | ||
a5f51c96 NP |
1458 | radix_tree_node_free(to_free); |
1459 | } | |
1460 | } | |
1461 | ||
139e5616 JW |
1462 | /** |
1463 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1464 | * @root: radix tree root | |
139e5616 JW |
1465 | * @node: node containing @index |
1466 | * | |
1467 | * After clearing the slot at @index in @node from radix tree | |
1468 | * rooted at @root, call this function to attempt freeing the | |
1469 | * node and shrinking the tree. | |
1470 | * | |
1471 | * Returns %true if @node was freed, %false otherwise. | |
1472 | */ | |
449dd698 | 1473 | bool __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1474 | struct radix_tree_node *node) |
1475 | { | |
1476 | bool deleted = false; | |
1477 | ||
1478 | do { | |
1479 | struct radix_tree_node *parent; | |
1480 | ||
1481 | if (node->count) { | |
1482 | if (node == indirect_to_ptr(root->rnode)) { | |
1483 | radix_tree_shrink(root); | |
1484 | if (root->height == 0) | |
1485 | deleted = true; | |
1486 | } | |
1487 | return deleted; | |
1488 | } | |
1489 | ||
1490 | parent = node->parent; | |
1491 | if (parent) { | |
449dd698 | 1492 | unsigned int offset; |
139e5616 | 1493 | |
449dd698 JW |
1494 | offset = node->path >> RADIX_TREE_HEIGHT_SHIFT; |
1495 | parent->slots[offset] = NULL; | |
139e5616 JW |
1496 | parent->count--; |
1497 | } else { | |
1498 | root_tag_clear_all(root); | |
1499 | root->height = 0; | |
1500 | root->rnode = NULL; | |
1501 | } | |
1502 | ||
1503 | radix_tree_node_free(node); | |
1504 | deleted = true; | |
1505 | ||
1506 | node = parent; | |
1507 | } while (node); | |
1508 | ||
1509 | return deleted; | |
1510 | } | |
1511 | ||
57578c2e MW |
1512 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1513 | void *ptr, unsigned offset) | |
1514 | { | |
1515 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1516 | int i; | |
1517 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1518 | if (node->slots[offset + i] != ptr) | |
1519 | break; | |
1520 | node->slots[offset + i] = NULL; | |
1521 | node->count--; | |
1522 | } | |
1523 | #endif | |
1524 | } | |
1525 | ||
1da177e4 | 1526 | /** |
53c59f26 | 1527 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1528 | * @root: radix tree root |
1529 | * @index: index key | |
53c59f26 | 1530 | * @item: expected item |
1da177e4 | 1531 | * |
53c59f26 | 1532 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1533 | * |
53c59f26 JW |
1534 | * Returns the address of the deleted item, or NULL if it was not present |
1535 | * or the entry at the given @index was not @item. | |
1da177e4 | 1536 | */ |
53c59f26 JW |
1537 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1538 | unsigned long index, void *item) | |
1da177e4 | 1539 | { |
139e5616 | 1540 | struct radix_tree_node *node; |
57578c2e | 1541 | unsigned int offset; |
139e5616 JW |
1542 | void **slot; |
1543 | void *entry; | |
d5274261 | 1544 | int tag; |
1da177e4 | 1545 | |
139e5616 JW |
1546 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1547 | if (!entry) | |
1548 | return NULL; | |
1da177e4 | 1549 | |
139e5616 JW |
1550 | if (item && entry != item) |
1551 | return NULL; | |
1552 | ||
1553 | if (!node) { | |
612d6c19 NP |
1554 | root_tag_clear_all(root); |
1555 | root->rnode = NULL; | |
139e5616 | 1556 | return entry; |
612d6c19 | 1557 | } |
1da177e4 | 1558 | |
29e0967c | 1559 | offset = get_slot_offset(node, slot); |
53c59f26 | 1560 | |
1da177e4 | 1561 | /* |
e2bdb933 HD |
1562 | * Clear all tags associated with the item to be deleted. |
1563 | * This way of doing it would be inefficient, but seldom is any set. | |
1da177e4 | 1564 | */ |
daff89f3 | 1565 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
e2bdb933 | 1566 | if (tag_get(node, tag, offset)) |
612d6c19 | 1567 | radix_tree_tag_clear(root, index, tag); |
d5274261 | 1568 | } |
1da177e4 | 1569 | |
57578c2e | 1570 | delete_sibling_entries(node, ptr_to_indirect(slot), offset); |
139e5616 JW |
1571 | node->slots[offset] = NULL; |
1572 | node->count--; | |
e2bdb933 | 1573 | |
449dd698 | 1574 | __radix_tree_delete_node(root, node); |
612d6c19 | 1575 | |
139e5616 | 1576 | return entry; |
1da177e4 | 1577 | } |
53c59f26 JW |
1578 | EXPORT_SYMBOL(radix_tree_delete_item); |
1579 | ||
1580 | /** | |
1581 | * radix_tree_delete - delete an item from a radix tree | |
1582 | * @root: radix tree root | |
1583 | * @index: index key | |
1584 | * | |
1585 | * Remove the item at @index from the radix tree rooted at @root. | |
1586 | * | |
1587 | * Returns the address of the deleted item, or NULL if it was not present. | |
1588 | */ | |
1589 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1590 | { | |
1591 | return radix_tree_delete_item(root, index, NULL); | |
1592 | } | |
1da177e4 LT |
1593 | EXPORT_SYMBOL(radix_tree_delete); |
1594 | ||
1595 | /** | |
1596 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1597 | * @root: radix tree root | |
1598 | * @tag: tag to test | |
1599 | */ | |
daff89f3 | 1600 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1601 | { |
612d6c19 | 1602 | return root_tag_get(root, tag); |
1da177e4 LT |
1603 | } |
1604 | EXPORT_SYMBOL(radix_tree_tagged); | |
1605 | ||
1606 | static void | |
449dd698 | 1607 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1608 | { |
449dd698 JW |
1609 | struct radix_tree_node *node = arg; |
1610 | ||
1611 | memset(node, 0, sizeof(*node)); | |
1612 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1613 | } |
1614 | ||
1615 | static __init unsigned long __maxindex(unsigned int height) | |
1616 | { | |
430d275a PL |
1617 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; |
1618 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1619 | ||
1620 | if (shift < 0) | |
1621 | return ~0UL; | |
1622 | if (shift >= BITS_PER_LONG) | |
1623 | return 0UL; | |
1624 | return ~0UL >> shift; | |
1da177e4 LT |
1625 | } |
1626 | ||
1627 | static __init void radix_tree_init_maxindex(void) | |
1628 | { | |
1629 | unsigned int i; | |
1630 | ||
1631 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1632 | height_to_maxindex[i] = __maxindex(i); | |
1633 | } | |
1634 | ||
1da177e4 LT |
1635 | static int radix_tree_callback(struct notifier_block *nfb, |
1636 | unsigned long action, | |
1637 | void *hcpu) | |
1638 | { | |
1639 | int cpu = (long)hcpu; | |
1640 | struct radix_tree_preload *rtp; | |
9d2a8da0 | 1641 | struct radix_tree_node *node; |
1da177e4 LT |
1642 | |
1643 | /* Free per-cpu pool of perloaded nodes */ | |
8bb78442 | 1644 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
1645 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1646 | while (rtp->nr) { | |
9d2a8da0 KS |
1647 | node = rtp->nodes; |
1648 | rtp->nodes = node->private_data; | |
1649 | kmem_cache_free(radix_tree_node_cachep, node); | |
1650 | rtp->nr--; | |
1da177e4 LT |
1651 | } |
1652 | } | |
1653 | return NOTIFY_OK; | |
1654 | } | |
1da177e4 LT |
1655 | |
1656 | void __init radix_tree_init(void) | |
1657 | { | |
1658 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", | |
1659 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1660 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1661 | radix_tree_node_ctor); | |
1da177e4 LT |
1662 | radix_tree_init_maxindex(); |
1663 | hotcpu_notifier(radix_tree_callback, 0); | |
1664 | } |