]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001 Momchil Velikov | |
3 | * Portions Copyright (C) 2001 Christoph Hellwig | |
cde53535 | 4 | * Copyright (C) 2005 SGI, Christoph Lameter |
7cf9c2c7 | 5 | * Copyright (C) 2006 Nick Piggin |
78c1d784 | 6 | * Copyright (C) 2012 Konstantin Khlebnikov |
6b053b8e MW |
7 | * Copyright (C) 2016 Intel, Matthew Wilcox |
8 | * Copyright (C) 2016 Intel, Ross Zwisler | |
1da177e4 LT |
9 | * |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License as | |
12 | * published by the Free Software Foundation; either version 2, or (at | |
13 | * your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, but | |
16 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with this program; if not, write to the Free Software | |
22 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
e157b555 | 25 | #include <linux/cpu.h> |
1da177e4 LT |
26 | #include <linux/errno.h> |
27 | #include <linux/init.h> | |
28 | #include <linux/kernel.h> | |
8bc3bcc9 | 29 | #include <linux/export.h> |
1da177e4 LT |
30 | #include <linux/radix-tree.h> |
31 | #include <linux/percpu.h> | |
32 | #include <linux/slab.h> | |
ce80b067 | 33 | #include <linux/kmemleak.h> |
1da177e4 LT |
34 | #include <linux/notifier.h> |
35 | #include <linux/cpu.h> | |
1da177e4 LT |
36 | #include <linux/string.h> |
37 | #include <linux/bitops.h> | |
7cf9c2c7 | 38 | #include <linux/rcupdate.h> |
92cf2118 | 39 | #include <linux/preempt.h> /* in_interrupt() */ |
1da177e4 LT |
40 | |
41 | ||
c78c66d1 KS |
42 | /* Number of nodes in fully populated tree of given height */ |
43 | static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly; | |
44 | ||
1da177e4 LT |
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 { | |
2fcd9005 | 67 | unsigned 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 | |
148deab2 MW |
73 | static inline struct radix_tree_node *entry_to_node(void *ptr) |
74 | { | |
75 | return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE); | |
76 | } | |
77 | ||
a4db4dce | 78 | static inline void *node_to_entry(void *ptr) |
27d20fdd | 79 | { |
30ff46cc | 80 | return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE); |
27d20fdd NP |
81 | } |
82 | ||
a4db4dce | 83 | #define RADIX_TREE_RETRY node_to_entry(NULL) |
afe0e395 | 84 | |
db050f29 MW |
85 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
86 | /* Sibling slots point directly to another slot in the same node */ | |
87 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
88 | { | |
89 | void **ptr = node; | |
90 | return (parent->slots <= ptr) && | |
91 | (ptr < parent->slots + RADIX_TREE_MAP_SIZE); | |
92 | } | |
93 | #else | |
94 | static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node) | |
95 | { | |
96 | return false; | |
97 | } | |
98 | #endif | |
99 | ||
100 | static inline unsigned long get_slot_offset(struct radix_tree_node *parent, | |
101 | void **slot) | |
102 | { | |
103 | return slot - parent->slots; | |
104 | } | |
105 | ||
9e85d811 MW |
106 | static unsigned int radix_tree_descend(struct radix_tree_node *parent, |
107 | struct radix_tree_node **nodep, unsigned long index) | |
db050f29 | 108 | { |
9e85d811 | 109 | unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK; |
db050f29 MW |
110 | void **entry = rcu_dereference_raw(parent->slots[offset]); |
111 | ||
112 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
b194d16c | 113 | if (radix_tree_is_internal_node(entry)) { |
8d2c0d36 LT |
114 | if (is_sibling_entry(parent, entry)) { |
115 | void **sibentry = (void **) entry_to_node(entry); | |
116 | offset = get_slot_offset(parent, sibentry); | |
117 | entry = rcu_dereference_raw(*sibentry); | |
db050f29 MW |
118 | } |
119 | } | |
120 | #endif | |
121 | ||
122 | *nodep = (void *)entry; | |
123 | return offset; | |
124 | } | |
125 | ||
612d6c19 NP |
126 | static inline gfp_t root_gfp_mask(struct radix_tree_root *root) |
127 | { | |
128 | return root->gfp_mask & __GFP_BITS_MASK; | |
129 | } | |
130 | ||
643b52b9 NP |
131 | static inline void tag_set(struct radix_tree_node *node, unsigned int tag, |
132 | int offset) | |
133 | { | |
134 | __set_bit(offset, node->tags[tag]); | |
135 | } | |
136 | ||
137 | static inline void tag_clear(struct radix_tree_node *node, unsigned int tag, | |
138 | int offset) | |
139 | { | |
140 | __clear_bit(offset, node->tags[tag]); | |
141 | } | |
142 | ||
143 | static inline int tag_get(struct radix_tree_node *node, unsigned int tag, | |
144 | int offset) | |
145 | { | |
146 | return test_bit(offset, node->tags[tag]); | |
147 | } | |
148 | ||
149 | static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag) | |
150 | { | |
151 | root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT)); | |
152 | } | |
153 | ||
2fcd9005 | 154 | static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag) |
643b52b9 NP |
155 | { |
156 | root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT)); | |
157 | } | |
158 | ||
159 | static inline void root_tag_clear_all(struct radix_tree_root *root) | |
160 | { | |
161 | root->gfp_mask &= __GFP_BITS_MASK; | |
162 | } | |
163 | ||
164 | static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag) | |
165 | { | |
2fcd9005 | 166 | return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT)); |
643b52b9 NP |
167 | } |
168 | ||
7b60e9ad MW |
169 | static inline unsigned root_tags_get(struct radix_tree_root *root) |
170 | { | |
171 | return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT; | |
172 | } | |
173 | ||
643b52b9 NP |
174 | /* |
175 | * Returns 1 if any slot in the node has this tag set. | |
176 | * Otherwise returns 0. | |
177 | */ | |
178 | static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag) | |
179 | { | |
2fcd9005 | 180 | unsigned idx; |
643b52b9 NP |
181 | for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) { |
182 | if (node->tags[tag][idx]) | |
183 | return 1; | |
184 | } | |
185 | return 0; | |
186 | } | |
78c1d784 KK |
187 | |
188 | /** | |
189 | * radix_tree_find_next_bit - find the next set bit in a memory region | |
190 | * | |
191 | * @addr: The address to base the search on | |
192 | * @size: The bitmap size in bits | |
193 | * @offset: The bitnumber to start searching at | |
194 | * | |
195 | * Unrollable variant of find_next_bit() for constant size arrays. | |
196 | * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero. | |
197 | * Returns next bit offset, or size if nothing found. | |
198 | */ | |
199 | static __always_inline unsigned long | |
bc412fca MW |
200 | radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag, |
201 | unsigned long offset) | |
78c1d784 | 202 | { |
bc412fca | 203 | const unsigned long *addr = node->tags[tag]; |
78c1d784 | 204 | |
bc412fca | 205 | if (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
206 | unsigned long tmp; |
207 | ||
208 | addr += offset / BITS_PER_LONG; | |
209 | tmp = *addr >> (offset % BITS_PER_LONG); | |
210 | if (tmp) | |
211 | return __ffs(tmp) + offset; | |
212 | offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1); | |
bc412fca | 213 | while (offset < RADIX_TREE_MAP_SIZE) { |
78c1d784 KK |
214 | tmp = *++addr; |
215 | if (tmp) | |
216 | return __ffs(tmp) + offset; | |
217 | offset += BITS_PER_LONG; | |
218 | } | |
219 | } | |
bc412fca | 220 | return RADIX_TREE_MAP_SIZE; |
78c1d784 KK |
221 | } |
222 | ||
268f42de MW |
223 | static unsigned int iter_offset(const struct radix_tree_iter *iter) |
224 | { | |
225 | return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK; | |
226 | } | |
227 | ||
218ed750 MW |
228 | /* |
229 | * The maximum index which can be stored in a radix tree | |
230 | */ | |
231 | static inline unsigned long shift_maxindex(unsigned int shift) | |
232 | { | |
233 | return (RADIX_TREE_MAP_SIZE << shift) - 1; | |
234 | } | |
235 | ||
236 | static inline unsigned long node_maxindex(struct radix_tree_node *node) | |
237 | { | |
238 | return shift_maxindex(node->shift); | |
239 | } | |
240 | ||
0796c583 | 241 | #ifndef __KERNEL__ |
d0891265 | 242 | static void dump_node(struct radix_tree_node *node, unsigned long index) |
7cf19af4 | 243 | { |
0796c583 | 244 | unsigned long i; |
7cf19af4 | 245 | |
218ed750 MW |
246 | pr_debug("radix node: %p offset %d indices %lu-%lu parent %p tags %lx %lx %lx shift %d count %d exceptional %d\n", |
247 | node, node->offset, index, index | node_maxindex(node), | |
248 | node->parent, | |
0796c583 | 249 | node->tags[0][0], node->tags[1][0], node->tags[2][0], |
218ed750 | 250 | node->shift, node->count, node->exceptional); |
0796c583 RZ |
251 | |
252 | for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) { | |
d0891265 MW |
253 | unsigned long first = index | (i << node->shift); |
254 | unsigned long last = first | ((1UL << node->shift) - 1); | |
0796c583 RZ |
255 | void *entry = node->slots[i]; |
256 | if (!entry) | |
257 | continue; | |
218ed750 MW |
258 | if (entry == RADIX_TREE_RETRY) { |
259 | pr_debug("radix retry offset %ld indices %lu-%lu parent %p\n", | |
260 | i, first, last, node); | |
b194d16c | 261 | } else if (!radix_tree_is_internal_node(entry)) { |
218ed750 MW |
262 | pr_debug("radix entry %p offset %ld indices %lu-%lu parent %p\n", |
263 | entry, i, first, last, node); | |
264 | } else if (is_sibling_entry(node, entry)) { | |
265 | pr_debug("radix sblng %p offset %ld indices %lu-%lu parent %p val %p\n", | |
266 | entry, i, first, last, node, | |
267 | *(void **)entry_to_node(entry)); | |
0796c583 | 268 | } else { |
4dd6c098 | 269 | dump_node(entry_to_node(entry), first); |
0796c583 RZ |
270 | } |
271 | } | |
7cf19af4 MW |
272 | } |
273 | ||
274 | /* For debug */ | |
275 | static void radix_tree_dump(struct radix_tree_root *root) | |
276 | { | |
d0891265 MW |
277 | pr_debug("radix root: %p rnode %p tags %x\n", |
278 | root, root->rnode, | |
7cf19af4 | 279 | root->gfp_mask >> __GFP_BITS_SHIFT); |
b194d16c | 280 | if (!radix_tree_is_internal_node(root->rnode)) |
7cf19af4 | 281 | return; |
4dd6c098 | 282 | dump_node(entry_to_node(root->rnode), 0); |
7cf19af4 MW |
283 | } |
284 | #endif | |
285 | ||
1da177e4 LT |
286 | /* |
287 | * This assumes that the caller has performed appropriate preallocation, and | |
288 | * that the caller has pinned this thread of control to the current CPU. | |
289 | */ | |
290 | static struct radix_tree_node * | |
291 | radix_tree_node_alloc(struct radix_tree_root *root) | |
292 | { | |
e2848a0e | 293 | struct radix_tree_node *ret = NULL; |
612d6c19 | 294 | gfp_t gfp_mask = root_gfp_mask(root); |
1da177e4 | 295 | |
5e4c0d97 | 296 | /* |
2fcd9005 MW |
297 | * Preload code isn't irq safe and it doesn't make sense to use |
298 | * preloading during an interrupt anyway as all the allocations have | |
299 | * to be atomic. So just do normal allocation when in interrupt. | |
5e4c0d97 | 300 | */ |
d0164adc | 301 | if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) { |
1da177e4 LT |
302 | struct radix_tree_preload *rtp; |
303 | ||
58e698af VD |
304 | /* |
305 | * Even if the caller has preloaded, try to allocate from the | |
05eb6e72 VD |
306 | * cache first for the new node to get accounted to the memory |
307 | * cgroup. | |
58e698af VD |
308 | */ |
309 | ret = kmem_cache_alloc(radix_tree_node_cachep, | |
05eb6e72 | 310 | gfp_mask | __GFP_NOWARN); |
58e698af VD |
311 | if (ret) |
312 | goto out; | |
313 | ||
e2848a0e NP |
314 | /* |
315 | * Provided the caller has preloaded here, we will always | |
316 | * succeed in getting a node here (and never reach | |
317 | * kmem_cache_alloc) | |
318 | */ | |
7c8e0181 | 319 | rtp = this_cpu_ptr(&radix_tree_preloads); |
1da177e4 | 320 | if (rtp->nr) { |
9d2a8da0 KS |
321 | ret = rtp->nodes; |
322 | rtp->nodes = ret->private_data; | |
323 | ret->private_data = NULL; | |
1da177e4 LT |
324 | rtp->nr--; |
325 | } | |
ce80b067 CM |
326 | /* |
327 | * Update the allocation stack trace as this is more useful | |
328 | * for debugging. | |
329 | */ | |
330 | kmemleak_update_trace(ret); | |
58e698af | 331 | goto out; |
1da177e4 | 332 | } |
05eb6e72 | 333 | ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
58e698af | 334 | out: |
b194d16c | 335 | BUG_ON(radix_tree_is_internal_node(ret)); |
1da177e4 LT |
336 | return ret; |
337 | } | |
338 | ||
7cf9c2c7 NP |
339 | static void radix_tree_node_rcu_free(struct rcu_head *head) |
340 | { | |
341 | struct radix_tree_node *node = | |
342 | container_of(head, struct radix_tree_node, rcu_head); | |
643b52b9 NP |
343 | |
344 | /* | |
175542f5 MW |
345 | * Must only free zeroed nodes into the slab. We can be left with |
346 | * non-NULL entries by radix_tree_free_nodes, so clear the entries | |
347 | * and tags here. | |
643b52b9 | 348 | */ |
175542f5 MW |
349 | memset(node->slots, 0, sizeof(node->slots)); |
350 | memset(node->tags, 0, sizeof(node->tags)); | |
91d9c05a | 351 | INIT_LIST_HEAD(&node->private_list); |
643b52b9 | 352 | |
7cf9c2c7 NP |
353 | kmem_cache_free(radix_tree_node_cachep, node); |
354 | } | |
355 | ||
1da177e4 LT |
356 | static inline void |
357 | radix_tree_node_free(struct radix_tree_node *node) | |
358 | { | |
7cf9c2c7 | 359 | call_rcu(&node->rcu_head, radix_tree_node_rcu_free); |
1da177e4 LT |
360 | } |
361 | ||
362 | /* | |
363 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
364 | * ensure that the addition of a single element in the tree cannot fail. On | |
365 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
366 | * with preemption not disabled. | |
b34df792 DH |
367 | * |
368 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 369 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
1da177e4 | 370 | */ |
2791653a | 371 | static int __radix_tree_preload(gfp_t gfp_mask, unsigned nr) |
1da177e4 LT |
372 | { |
373 | struct radix_tree_preload *rtp; | |
374 | struct radix_tree_node *node; | |
375 | int ret = -ENOMEM; | |
376 | ||
05eb6e72 VD |
377 | /* |
378 | * Nodes preloaded by one cgroup can be be used by another cgroup, so | |
379 | * they should never be accounted to any particular memory cgroup. | |
380 | */ | |
381 | gfp_mask &= ~__GFP_ACCOUNT; | |
382 | ||
1da177e4 | 383 | preempt_disable(); |
7c8e0181 | 384 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 385 | while (rtp->nr < nr) { |
1da177e4 | 386 | preempt_enable(); |
488514d1 | 387 | node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask); |
1da177e4 LT |
388 | if (node == NULL) |
389 | goto out; | |
390 | preempt_disable(); | |
7c8e0181 | 391 | rtp = this_cpu_ptr(&radix_tree_preloads); |
c78c66d1 | 392 | if (rtp->nr < nr) { |
9d2a8da0 KS |
393 | node->private_data = rtp->nodes; |
394 | rtp->nodes = node; | |
395 | rtp->nr++; | |
396 | } else { | |
1da177e4 | 397 | kmem_cache_free(radix_tree_node_cachep, node); |
9d2a8da0 | 398 | } |
1da177e4 LT |
399 | } |
400 | ret = 0; | |
401 | out: | |
402 | return ret; | |
403 | } | |
5e4c0d97 JK |
404 | |
405 | /* | |
406 | * Load up this CPU's radix_tree_node buffer with sufficient objects to | |
407 | * ensure that the addition of a single element in the tree cannot fail. On | |
408 | * success, return zero, with preemption disabled. On error, return -ENOMEM | |
409 | * with preemption not disabled. | |
410 | * | |
411 | * To make use of this facility, the radix tree must be initialised without | |
d0164adc | 412 | * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE(). |
5e4c0d97 JK |
413 | */ |
414 | int radix_tree_preload(gfp_t gfp_mask) | |
415 | { | |
416 | /* Warn on non-sensical use... */ | |
d0164adc | 417 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); |
c78c66d1 | 418 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 | 419 | } |
d7f0923d | 420 | EXPORT_SYMBOL(radix_tree_preload); |
1da177e4 | 421 | |
5e4c0d97 JK |
422 | /* |
423 | * The same as above function, except we don't guarantee preloading happens. | |
424 | * We do it, if we decide it helps. On success, return zero with preemption | |
425 | * disabled. On error, return -ENOMEM with preemption not disabled. | |
426 | */ | |
427 | int radix_tree_maybe_preload(gfp_t gfp_mask) | |
428 | { | |
d0164adc | 429 | if (gfpflags_allow_blocking(gfp_mask)) |
c78c66d1 | 430 | return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE); |
5e4c0d97 JK |
431 | /* Preloading doesn't help anything with this gfp mask, skip it */ |
432 | preempt_disable(); | |
433 | return 0; | |
434 | } | |
435 | EXPORT_SYMBOL(radix_tree_maybe_preload); | |
436 | ||
2791653a MW |
437 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
438 | /* | |
439 | * Preload with enough objects to ensure that we can split a single entry | |
440 | * of order @old_order into many entries of size @new_order | |
441 | */ | |
442 | int radix_tree_split_preload(unsigned int old_order, unsigned int new_order, | |
443 | gfp_t gfp_mask) | |
444 | { | |
445 | unsigned top = 1 << (old_order % RADIX_TREE_MAP_SHIFT); | |
446 | unsigned layers = (old_order / RADIX_TREE_MAP_SHIFT) - | |
447 | (new_order / RADIX_TREE_MAP_SHIFT); | |
448 | unsigned nr = 0; | |
449 | ||
450 | WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask)); | |
451 | BUG_ON(new_order >= old_order); | |
452 | ||
453 | while (layers--) | |
454 | nr = nr * RADIX_TREE_MAP_SIZE + 1; | |
455 | return __radix_tree_preload(gfp_mask, top * nr); | |
456 | } | |
457 | #endif | |
458 | ||
c78c66d1 KS |
459 | /* |
460 | * The same as function above, but preload number of nodes required to insert | |
461 | * (1 << order) continuous naturally-aligned elements. | |
462 | */ | |
463 | int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order) | |
464 | { | |
465 | unsigned long nr_subtrees; | |
466 | int nr_nodes, subtree_height; | |
467 | ||
468 | /* Preloading doesn't help anything with this gfp mask, skip it */ | |
469 | if (!gfpflags_allow_blocking(gfp_mask)) { | |
470 | preempt_disable(); | |
471 | return 0; | |
472 | } | |
473 | ||
474 | /* | |
475 | * Calculate number and height of fully populated subtrees it takes to | |
476 | * store (1 << order) elements. | |
477 | */ | |
478 | nr_subtrees = 1 << order; | |
479 | for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE; | |
480 | subtree_height++) | |
481 | nr_subtrees >>= RADIX_TREE_MAP_SHIFT; | |
482 | ||
483 | /* | |
484 | * The worst case is zero height tree with a single item at index 0 and | |
485 | * then inserting items starting at ULONG_MAX - (1 << order). | |
486 | * | |
487 | * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to | |
488 | * 0-index item. | |
489 | */ | |
490 | nr_nodes = RADIX_TREE_MAX_PATH; | |
491 | ||
492 | /* Plus branch to fully populated subtrees. */ | |
493 | nr_nodes += RADIX_TREE_MAX_PATH - subtree_height; | |
494 | ||
495 | /* Root node is shared. */ | |
496 | nr_nodes--; | |
497 | ||
498 | /* Plus nodes required to build subtrees. */ | |
499 | nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height]; | |
500 | ||
501 | return __radix_tree_preload(gfp_mask, nr_nodes); | |
502 | } | |
503 | ||
1456a439 MW |
504 | static unsigned radix_tree_load_root(struct radix_tree_root *root, |
505 | struct radix_tree_node **nodep, unsigned long *maxindex) | |
506 | { | |
507 | struct radix_tree_node *node = rcu_dereference_raw(root->rnode); | |
508 | ||
509 | *nodep = node; | |
510 | ||
b194d16c | 511 | if (likely(radix_tree_is_internal_node(node))) { |
4dd6c098 | 512 | node = entry_to_node(node); |
1456a439 | 513 | *maxindex = node_maxindex(node); |
c12e51b0 | 514 | return node->shift + RADIX_TREE_MAP_SHIFT; |
1456a439 MW |
515 | } |
516 | ||
517 | *maxindex = 0; | |
518 | return 0; | |
519 | } | |
520 | ||
1da177e4 LT |
521 | /* |
522 | * Extend a radix tree so it can store key @index. | |
523 | */ | |
e6145236 | 524 | static int radix_tree_extend(struct radix_tree_root *root, |
d0891265 | 525 | unsigned long index, unsigned int shift) |
1da177e4 | 526 | { |
e2bdb933 | 527 | struct radix_tree_node *slot; |
d0891265 | 528 | unsigned int maxshift; |
1da177e4 LT |
529 | int tag; |
530 | ||
d0891265 MW |
531 | /* Figure out what the shift should be. */ |
532 | maxshift = shift; | |
533 | while (index > shift_maxindex(maxshift)) | |
534 | maxshift += RADIX_TREE_MAP_SHIFT; | |
1da177e4 | 535 | |
d0891265 MW |
536 | slot = root->rnode; |
537 | if (!slot) | |
1da177e4 | 538 | goto out; |
1da177e4 | 539 | |
1da177e4 | 540 | do { |
2fcd9005 MW |
541 | struct radix_tree_node *node = radix_tree_node_alloc(root); |
542 | ||
543 | if (!node) | |
1da177e4 LT |
544 | return -ENOMEM; |
545 | ||
1da177e4 | 546 | /* Propagate the aggregated tag info into the new root */ |
daff89f3 | 547 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
612d6c19 | 548 | if (root_tag_get(root, tag)) |
1da177e4 LT |
549 | tag_set(node, tag, 0); |
550 | } | |
551 | ||
d0891265 MW |
552 | BUG_ON(shift > BITS_PER_LONG); |
553 | node->shift = shift; | |
0c7fa0a8 | 554 | node->offset = 0; |
1da177e4 | 555 | node->count = 1; |
e2bdb933 | 556 | node->parent = NULL; |
f7942430 | 557 | if (radix_tree_is_internal_node(slot)) { |
4dd6c098 | 558 | entry_to_node(slot)->parent = node; |
f7942430 JW |
559 | } else { |
560 | /* Moving an exceptional root->rnode to a node */ | |
561 | if (radix_tree_exceptional_entry(slot)) | |
562 | node->exceptional = 1; | |
563 | } | |
e2bdb933 | 564 | node->slots[0] = slot; |
a4db4dce MW |
565 | slot = node_to_entry(node); |
566 | rcu_assign_pointer(root->rnode, slot); | |
d0891265 | 567 | shift += RADIX_TREE_MAP_SHIFT; |
d0891265 | 568 | } while (shift <= maxshift); |
1da177e4 | 569 | out: |
d0891265 | 570 | return maxshift + RADIX_TREE_MAP_SHIFT; |
1da177e4 LT |
571 | } |
572 | ||
f4b109c6 JW |
573 | /** |
574 | * radix_tree_shrink - shrink radix tree to minimum height | |
575 | * @root radix tree root | |
576 | */ | |
14b46879 | 577 | static inline void radix_tree_shrink(struct radix_tree_root *root, |
4d693d08 JW |
578 | radix_tree_update_node_t update_node, |
579 | void *private) | |
f4b109c6 | 580 | { |
f4b109c6 JW |
581 | for (;;) { |
582 | struct radix_tree_node *node = root->rnode; | |
583 | struct radix_tree_node *child; | |
584 | ||
585 | if (!radix_tree_is_internal_node(node)) | |
586 | break; | |
587 | node = entry_to_node(node); | |
588 | ||
589 | /* | |
590 | * The candidate node has more than one child, or its child | |
591 | * is not at the leftmost slot, or the child is a multiorder | |
592 | * entry, we cannot shrink. | |
593 | */ | |
594 | if (node->count != 1) | |
595 | break; | |
596 | child = node->slots[0]; | |
597 | if (!child) | |
598 | break; | |
599 | if (!radix_tree_is_internal_node(child) && node->shift) | |
600 | break; | |
601 | ||
602 | if (radix_tree_is_internal_node(child)) | |
603 | entry_to_node(child)->parent = NULL; | |
604 | ||
605 | /* | |
606 | * We don't need rcu_assign_pointer(), since we are simply | |
607 | * moving the node from one part of the tree to another: if it | |
608 | * was safe to dereference the old pointer to it | |
609 | * (node->slots[0]), it will be safe to dereference the new | |
610 | * one (root->rnode) as far as dependent read barriers go. | |
611 | */ | |
612 | root->rnode = child; | |
613 | ||
614 | /* | |
615 | * We have a dilemma here. The node's slot[0] must not be | |
616 | * NULLed in case there are concurrent lookups expecting to | |
617 | * find the item. However if this was a bottom-level node, | |
618 | * then it may be subject to the slot pointer being visible | |
619 | * to callers dereferencing it. If item corresponding to | |
620 | * slot[0] is subsequently deleted, these callers would expect | |
621 | * their slot to become empty sooner or later. | |
622 | * | |
623 | * For example, lockless pagecache will look up a slot, deref | |
624 | * the page pointer, and if the page has 0 refcount it means it | |
625 | * was concurrently deleted from pagecache so try the deref | |
626 | * again. Fortunately there is already a requirement for logic | |
627 | * to retry the entire slot lookup -- the indirect pointer | |
628 | * problem (replacing direct root node with an indirect pointer | |
629 | * also results in a stale slot). So tag the slot as indirect | |
630 | * to force callers to retry. | |
631 | */ | |
4d693d08 JW |
632 | node->count = 0; |
633 | if (!radix_tree_is_internal_node(child)) { | |
f4b109c6 | 634 | node->slots[0] = RADIX_TREE_RETRY; |
4d693d08 JW |
635 | if (update_node) |
636 | update_node(node, private); | |
637 | } | |
f4b109c6 JW |
638 | |
639 | radix_tree_node_free(node); | |
f4b109c6 | 640 | } |
f4b109c6 JW |
641 | } |
642 | ||
14b46879 | 643 | static void delete_node(struct radix_tree_root *root, |
4d693d08 JW |
644 | struct radix_tree_node *node, |
645 | radix_tree_update_node_t update_node, void *private) | |
f4b109c6 | 646 | { |
f4b109c6 JW |
647 | do { |
648 | struct radix_tree_node *parent; | |
649 | ||
650 | if (node->count) { | |
651 | if (node == entry_to_node(root->rnode)) | |
14b46879 JW |
652 | radix_tree_shrink(root, update_node, private); |
653 | return; | |
f4b109c6 JW |
654 | } |
655 | ||
656 | parent = node->parent; | |
657 | if (parent) { | |
658 | parent->slots[node->offset] = NULL; | |
659 | parent->count--; | |
660 | } else { | |
661 | root_tag_clear_all(root); | |
662 | root->rnode = NULL; | |
663 | } | |
664 | ||
665 | radix_tree_node_free(node); | |
f4b109c6 JW |
666 | |
667 | node = parent; | |
668 | } while (node); | |
f4b109c6 JW |
669 | } |
670 | ||
1da177e4 | 671 | /** |
139e5616 | 672 | * __radix_tree_create - create a slot in a radix tree |
1da177e4 LT |
673 | * @root: radix tree root |
674 | * @index: index key | |
e6145236 | 675 | * @order: index occupies 2^order aligned slots |
139e5616 JW |
676 | * @nodep: returns node |
677 | * @slotp: returns slot | |
1da177e4 | 678 | * |
139e5616 JW |
679 | * Create, if necessary, and return the node and slot for an item |
680 | * at position @index in the radix tree @root. | |
681 | * | |
682 | * Until there is more than one item in the tree, no nodes are | |
683 | * allocated and @root->rnode is used as a direct slot instead of | |
684 | * pointing to a node, in which case *@nodep will be NULL. | |
685 | * | |
686 | * Returns -ENOMEM, or 0 for success. | |
1da177e4 | 687 | */ |
139e5616 | 688 | int __radix_tree_create(struct radix_tree_root *root, unsigned long index, |
e6145236 MW |
689 | unsigned order, struct radix_tree_node **nodep, |
690 | void ***slotp) | |
1da177e4 | 691 | { |
89148aa4 MW |
692 | struct radix_tree_node *node = NULL, *child; |
693 | void **slot = (void **)&root->rnode; | |
49ea6ebc | 694 | unsigned long maxindex; |
89148aa4 | 695 | unsigned int shift, offset = 0; |
49ea6ebc MW |
696 | unsigned long max = index | ((1UL << order) - 1); |
697 | ||
89148aa4 | 698 | shift = radix_tree_load_root(root, &child, &maxindex); |
1da177e4 LT |
699 | |
700 | /* Make sure the tree is high enough. */ | |
175542f5 MW |
701 | if (order > 0 && max == ((1UL << order) - 1)) |
702 | max++; | |
49ea6ebc | 703 | if (max > maxindex) { |
d0891265 | 704 | int error = radix_tree_extend(root, max, shift); |
49ea6ebc | 705 | if (error < 0) |
1da177e4 | 706 | return error; |
49ea6ebc | 707 | shift = error; |
89148aa4 | 708 | child = root->rnode; |
1da177e4 LT |
709 | } |
710 | ||
e6145236 | 711 | while (shift > order) { |
c12e51b0 | 712 | shift -= RADIX_TREE_MAP_SHIFT; |
89148aa4 | 713 | if (child == NULL) { |
1da177e4 | 714 | /* Have to add a child node. */ |
89148aa4 MW |
715 | child = radix_tree_node_alloc(root); |
716 | if (!child) | |
1da177e4 | 717 | return -ENOMEM; |
89148aa4 MW |
718 | child->shift = shift; |
719 | child->offset = offset; | |
175542f5 MW |
720 | child->count = 0; |
721 | child->exceptional = 0; | |
89148aa4 MW |
722 | child->parent = node; |
723 | rcu_assign_pointer(*slot, node_to_entry(child)); | |
724 | if (node) | |
1da177e4 | 725 | node->count++; |
89148aa4 | 726 | } else if (!radix_tree_is_internal_node(child)) |
e6145236 | 727 | break; |
1da177e4 LT |
728 | |
729 | /* Go a level down */ | |
89148aa4 | 730 | node = entry_to_node(child); |
9e85d811 | 731 | offset = radix_tree_descend(node, &child, index); |
89148aa4 | 732 | slot = &node->slots[offset]; |
e6145236 MW |
733 | } |
734 | ||
175542f5 MW |
735 | if (nodep) |
736 | *nodep = node; | |
737 | if (slotp) | |
738 | *slotp = slot; | |
739 | return 0; | |
740 | } | |
741 | ||
57578c2e | 742 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
175542f5 MW |
743 | /* |
744 | * Free any nodes below this node. The tree is presumed to not need | |
745 | * shrinking, and any user data in the tree is presumed to not need a | |
746 | * destructor called on it. If we need to add a destructor, we can | |
747 | * add that functionality later. Note that we may not clear tags or | |
748 | * slots from the tree as an RCU walker may still have a pointer into | |
749 | * this subtree. We could replace the entries with RADIX_TREE_RETRY, | |
750 | * but we'll still have to clear those in rcu_free. | |
751 | */ | |
752 | static void radix_tree_free_nodes(struct radix_tree_node *node) | |
753 | { | |
754 | unsigned offset = 0; | |
755 | struct radix_tree_node *child = entry_to_node(node); | |
756 | ||
757 | for (;;) { | |
758 | void *entry = child->slots[offset]; | |
759 | if (radix_tree_is_internal_node(entry) && | |
760 | !is_sibling_entry(child, entry)) { | |
761 | child = entry_to_node(entry); | |
762 | offset = 0; | |
763 | continue; | |
764 | } | |
765 | offset++; | |
766 | while (offset == RADIX_TREE_MAP_SIZE) { | |
767 | struct radix_tree_node *old = child; | |
768 | offset = child->offset + 1; | |
769 | child = child->parent; | |
770 | radix_tree_node_free(old); | |
771 | if (old == entry_to_node(node)) | |
772 | return; | |
773 | } | |
774 | } | |
775 | } | |
776 | ||
777 | static inline int insert_entries(struct radix_tree_node *node, void **slot, | |
778 | void *item, unsigned order, bool replace) | |
779 | { | |
780 | struct radix_tree_node *child; | |
781 | unsigned i, n, tag, offset, tags = 0; | |
782 | ||
783 | if (node) { | |
e157b555 MW |
784 | if (order > node->shift) |
785 | n = 1 << (order - node->shift); | |
786 | else | |
787 | n = 1; | |
175542f5 MW |
788 | offset = get_slot_offset(node, slot); |
789 | } else { | |
790 | n = 1; | |
791 | offset = 0; | |
792 | } | |
793 | ||
794 | if (n > 1) { | |
e6145236 | 795 | offset = offset & ~(n - 1); |
89148aa4 | 796 | slot = &node->slots[offset]; |
175542f5 MW |
797 | } |
798 | child = node_to_entry(slot); | |
799 | ||
800 | for (i = 0; i < n; i++) { | |
801 | if (slot[i]) { | |
802 | if (replace) { | |
803 | node->count--; | |
804 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
805 | if (tag_get(node, tag, offset + i)) | |
806 | tags |= 1 << tag; | |
807 | } else | |
e6145236 MW |
808 | return -EEXIST; |
809 | } | |
175542f5 | 810 | } |
e6145236 | 811 | |
175542f5 MW |
812 | for (i = 0; i < n; i++) { |
813 | struct radix_tree_node *old = slot[i]; | |
814 | if (i) { | |
89148aa4 | 815 | rcu_assign_pointer(slot[i], child); |
175542f5 MW |
816 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) |
817 | if (tags & (1 << tag)) | |
818 | tag_clear(node, tag, offset + i); | |
819 | } else { | |
820 | rcu_assign_pointer(slot[i], item); | |
821 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
822 | if (tags & (1 << tag)) | |
823 | tag_set(node, tag, offset); | |
e6145236 | 824 | } |
175542f5 | 825 | if (radix_tree_is_internal_node(old) && |
e157b555 MW |
826 | !is_sibling_entry(node, old) && |
827 | (old != RADIX_TREE_RETRY)) | |
175542f5 MW |
828 | radix_tree_free_nodes(old); |
829 | if (radix_tree_exceptional_entry(old)) | |
830 | node->exceptional--; | |
612d6c19 | 831 | } |
175542f5 MW |
832 | if (node) { |
833 | node->count += n; | |
834 | if (radix_tree_exceptional_entry(item)) | |
835 | node->exceptional += n; | |
836 | } | |
837 | return n; | |
139e5616 | 838 | } |
175542f5 MW |
839 | #else |
840 | static inline int insert_entries(struct radix_tree_node *node, void **slot, | |
841 | void *item, unsigned order, bool replace) | |
842 | { | |
843 | if (*slot) | |
844 | return -EEXIST; | |
845 | rcu_assign_pointer(*slot, item); | |
846 | if (node) { | |
847 | node->count++; | |
848 | if (radix_tree_exceptional_entry(item)) | |
849 | node->exceptional++; | |
850 | } | |
851 | return 1; | |
852 | } | |
853 | #endif | |
139e5616 JW |
854 | |
855 | /** | |
e6145236 | 856 | * __radix_tree_insert - insert into a radix tree |
139e5616 JW |
857 | * @root: radix tree root |
858 | * @index: index key | |
e6145236 | 859 | * @order: key covers the 2^order indices around index |
139e5616 JW |
860 | * @item: item to insert |
861 | * | |
862 | * Insert an item into the radix tree at position @index. | |
863 | */ | |
e6145236 MW |
864 | int __radix_tree_insert(struct radix_tree_root *root, unsigned long index, |
865 | unsigned order, void *item) | |
139e5616 JW |
866 | { |
867 | struct radix_tree_node *node; | |
868 | void **slot; | |
869 | int error; | |
870 | ||
b194d16c | 871 | BUG_ON(radix_tree_is_internal_node(item)); |
139e5616 | 872 | |
e6145236 | 873 | error = __radix_tree_create(root, index, order, &node, &slot); |
139e5616 JW |
874 | if (error) |
875 | return error; | |
175542f5 MW |
876 | |
877 | error = insert_entries(node, slot, item, order, false); | |
878 | if (error < 0) | |
879 | return error; | |
201b6264 | 880 | |
612d6c19 | 881 | if (node) { |
7b60e9ad | 882 | unsigned offset = get_slot_offset(node, slot); |
7b60e9ad MW |
883 | BUG_ON(tag_get(node, 0, offset)); |
884 | BUG_ON(tag_get(node, 1, offset)); | |
885 | BUG_ON(tag_get(node, 2, offset)); | |
612d6c19 | 886 | } else { |
7b60e9ad | 887 | BUG_ON(root_tags_get(root)); |
612d6c19 | 888 | } |
1da177e4 | 889 | |
1da177e4 LT |
890 | return 0; |
891 | } | |
e6145236 | 892 | EXPORT_SYMBOL(__radix_tree_insert); |
1da177e4 | 893 | |
139e5616 JW |
894 | /** |
895 | * __radix_tree_lookup - lookup an item in a radix tree | |
896 | * @root: radix tree root | |
897 | * @index: index key | |
898 | * @nodep: returns node | |
899 | * @slotp: returns slot | |
900 | * | |
901 | * Lookup and return the item at position @index in the radix | |
902 | * tree @root. | |
903 | * | |
904 | * Until there is more than one item in the tree, no nodes are | |
905 | * allocated and @root->rnode is used as a direct slot instead of | |
906 | * pointing to a node, in which case *@nodep will be NULL. | |
7cf9c2c7 | 907 | */ |
139e5616 JW |
908 | void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index, |
909 | struct radix_tree_node **nodep, void ***slotp) | |
1da177e4 | 910 | { |
139e5616 | 911 | struct radix_tree_node *node, *parent; |
85829954 | 912 | unsigned long maxindex; |
139e5616 | 913 | void **slot; |
612d6c19 | 914 | |
85829954 MW |
915 | restart: |
916 | parent = NULL; | |
917 | slot = (void **)&root->rnode; | |
9e85d811 | 918 | radix_tree_load_root(root, &node, &maxindex); |
85829954 | 919 | if (index > maxindex) |
1da177e4 LT |
920 | return NULL; |
921 | ||
b194d16c | 922 | while (radix_tree_is_internal_node(node)) { |
85829954 | 923 | unsigned offset; |
1da177e4 | 924 | |
85829954 MW |
925 | if (node == RADIX_TREE_RETRY) |
926 | goto restart; | |
4dd6c098 | 927 | parent = entry_to_node(node); |
9e85d811 | 928 | offset = radix_tree_descend(parent, &node, index); |
85829954 MW |
929 | slot = parent->slots + offset; |
930 | } | |
1da177e4 | 931 | |
139e5616 JW |
932 | if (nodep) |
933 | *nodep = parent; | |
934 | if (slotp) | |
935 | *slotp = slot; | |
936 | return node; | |
b72b71c6 HS |
937 | } |
938 | ||
939 | /** | |
940 | * radix_tree_lookup_slot - lookup a slot in a radix tree | |
941 | * @root: radix tree root | |
942 | * @index: index key | |
943 | * | |
944 | * Returns: the slot corresponding to the position @index in the | |
945 | * radix tree @root. This is useful for update-if-exists operations. | |
946 | * | |
947 | * This function can be called under rcu_read_lock iff the slot is not | |
948 | * modified by radix_tree_replace_slot, otherwise it must be called | |
949 | * exclusive from other writers. Any dereference of the slot must be done | |
950 | * using radix_tree_deref_slot. | |
951 | */ | |
952 | void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index) | |
953 | { | |
139e5616 JW |
954 | void **slot; |
955 | ||
956 | if (!__radix_tree_lookup(root, index, NULL, &slot)) | |
957 | return NULL; | |
958 | return slot; | |
a4331366 | 959 | } |
a4331366 HR |
960 | EXPORT_SYMBOL(radix_tree_lookup_slot); |
961 | ||
962 | /** | |
963 | * radix_tree_lookup - perform lookup operation on a radix tree | |
964 | * @root: radix tree root | |
965 | * @index: index key | |
966 | * | |
967 | * Lookup the item at the position @index in the radix tree @root. | |
7cf9c2c7 NP |
968 | * |
969 | * This function can be called under rcu_read_lock, however the caller | |
970 | * must manage lifetimes of leaf nodes (eg. RCU may also be used to free | |
971 | * them safely). No RCU barriers are required to access or modify the | |
972 | * returned item, however. | |
a4331366 HR |
973 | */ |
974 | void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index) | |
975 | { | |
139e5616 | 976 | return __radix_tree_lookup(root, index, NULL, NULL); |
1da177e4 LT |
977 | } |
978 | EXPORT_SYMBOL(radix_tree_lookup); | |
979 | ||
6d75f366 JW |
980 | static void replace_slot(struct radix_tree_root *root, |
981 | struct radix_tree_node *node, | |
982 | void **slot, void *item, | |
983 | bool warn_typeswitch) | |
f7942430 JW |
984 | { |
985 | void *old = rcu_dereference_raw(*slot); | |
f4b109c6 | 986 | int count, exceptional; |
f7942430 JW |
987 | |
988 | WARN_ON_ONCE(radix_tree_is_internal_node(item)); | |
f7942430 | 989 | |
f4b109c6 | 990 | count = !!item - !!old; |
f7942430 JW |
991 | exceptional = !!radix_tree_exceptional_entry(item) - |
992 | !!radix_tree_exceptional_entry(old); | |
993 | ||
f4b109c6 | 994 | WARN_ON_ONCE(warn_typeswitch && (count || exceptional)); |
f7942430 | 995 | |
f4b109c6 JW |
996 | if (node) { |
997 | node->count += count; | |
f7942430 | 998 | node->exceptional += exceptional; |
f4b109c6 | 999 | } |
f7942430 JW |
1000 | |
1001 | rcu_assign_pointer(*slot, item); | |
1002 | } | |
1003 | ||
6d75f366 JW |
1004 | /** |
1005 | * __radix_tree_replace - replace item in a slot | |
4d693d08 JW |
1006 | * @root: radix tree root |
1007 | * @node: pointer to tree node | |
1008 | * @slot: pointer to slot in @node | |
1009 | * @item: new item to store in the slot. | |
1010 | * @update_node: callback for changing leaf nodes | |
1011 | * @private: private data to pass to @update_node | |
6d75f366 JW |
1012 | * |
1013 | * For use with __radix_tree_lookup(). Caller must hold tree write locked | |
1014 | * across slot lookup and replacement. | |
1015 | */ | |
1016 | void __radix_tree_replace(struct radix_tree_root *root, | |
1017 | struct radix_tree_node *node, | |
4d693d08 JW |
1018 | void **slot, void *item, |
1019 | radix_tree_update_node_t update_node, void *private) | |
6d75f366 JW |
1020 | { |
1021 | /* | |
f4b109c6 JW |
1022 | * This function supports replacing exceptional entries and |
1023 | * deleting entries, but that needs accounting against the | |
1024 | * node unless the slot is root->rnode. | |
6d75f366 JW |
1025 | */ |
1026 | replace_slot(root, node, slot, item, | |
1027 | !node && slot != (void **)&root->rnode); | |
f4b109c6 | 1028 | |
4d693d08 JW |
1029 | if (!node) |
1030 | return; | |
1031 | ||
1032 | if (update_node) | |
1033 | update_node(node, private); | |
1034 | ||
1035 | delete_node(root, node, update_node, private); | |
6d75f366 JW |
1036 | } |
1037 | ||
1038 | /** | |
1039 | * radix_tree_replace_slot - replace item in a slot | |
1040 | * @root: radix tree root | |
1041 | * @slot: pointer to slot | |
1042 | * @item: new item to store in the slot. | |
1043 | * | |
1044 | * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(), | |
1045 | * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked | |
1046 | * across slot lookup and replacement. | |
1047 | * | |
1048 | * NOTE: This cannot be used to switch between non-entries (empty slots), | |
1049 | * regular entries, and exceptional entries, as that requires accounting | |
f4b109c6 | 1050 | * inside the radix tree node. When switching from one type of entry or |
e157b555 MW |
1051 | * deleting, use __radix_tree_lookup() and __radix_tree_replace() or |
1052 | * radix_tree_iter_replace(). | |
6d75f366 JW |
1053 | */ |
1054 | void radix_tree_replace_slot(struct radix_tree_root *root, | |
1055 | void **slot, void *item) | |
1056 | { | |
1057 | replace_slot(root, NULL, slot, item, true); | |
1058 | } | |
1059 | ||
e157b555 MW |
1060 | /** |
1061 | * radix_tree_iter_replace - replace item in a slot | |
1062 | * @root: radix tree root | |
1063 | * @slot: pointer to slot | |
1064 | * @item: new item to store in the slot. | |
1065 | * | |
1066 | * For use with radix_tree_split() and radix_tree_for_each_slot(). | |
1067 | * Caller must hold tree write locked across split and replacement. | |
1068 | */ | |
1069 | void radix_tree_iter_replace(struct radix_tree_root *root, | |
1070 | const struct radix_tree_iter *iter, void **slot, void *item) | |
1071 | { | |
1072 | __radix_tree_replace(root, iter->node, slot, item, NULL, NULL); | |
1073 | } | |
1074 | ||
175542f5 MW |
1075 | #ifdef CONFIG_RADIX_TREE_MULTIORDER |
1076 | /** | |
1077 | * radix_tree_join - replace multiple entries with one multiorder entry | |
1078 | * @root: radix tree root | |
1079 | * @index: an index inside the new entry | |
1080 | * @order: order of the new entry | |
1081 | * @item: new entry | |
1082 | * | |
1083 | * Call this function to replace several entries with one larger entry. | |
1084 | * The existing entries are presumed to not need freeing as a result of | |
1085 | * this call. | |
1086 | * | |
1087 | * The replacement entry will have all the tags set on it that were set | |
1088 | * on any of the entries it is replacing. | |
1089 | */ | |
1090 | int radix_tree_join(struct radix_tree_root *root, unsigned long index, | |
1091 | unsigned order, void *item) | |
1092 | { | |
1093 | struct radix_tree_node *node; | |
1094 | void **slot; | |
1095 | int error; | |
1096 | ||
1097 | BUG_ON(radix_tree_is_internal_node(item)); | |
1098 | ||
1099 | error = __radix_tree_create(root, index, order, &node, &slot); | |
1100 | if (!error) | |
1101 | error = insert_entries(node, slot, item, order, true); | |
1102 | if (error > 0) | |
1103 | error = 0; | |
1104 | ||
1105 | return error; | |
1106 | } | |
e157b555 MW |
1107 | |
1108 | /** | |
1109 | * radix_tree_split - Split an entry into smaller entries | |
1110 | * @root: radix tree root | |
1111 | * @index: An index within the large entry | |
1112 | * @order: Order of new entries | |
1113 | * | |
1114 | * Call this function as the first step in replacing a multiorder entry | |
1115 | * with several entries of lower order. After this function returns, | |
1116 | * loop over the relevant portion of the tree using radix_tree_for_each_slot() | |
1117 | * and call radix_tree_iter_replace() to set up each new entry. | |
1118 | * | |
1119 | * The tags from this entry are replicated to all the new entries. | |
1120 | * | |
1121 | * The radix tree should be locked against modification during the entire | |
1122 | * replacement operation. Lock-free lookups will see RADIX_TREE_RETRY which | |
1123 | * should prompt RCU walkers to restart the lookup from the root. | |
1124 | */ | |
1125 | int radix_tree_split(struct radix_tree_root *root, unsigned long index, | |
1126 | unsigned order) | |
1127 | { | |
1128 | struct radix_tree_node *parent, *node, *child; | |
1129 | void **slot; | |
1130 | unsigned int offset, end; | |
1131 | unsigned n, tag, tags = 0; | |
1132 | ||
1133 | if (!__radix_tree_lookup(root, index, &parent, &slot)) | |
1134 | return -ENOENT; | |
1135 | if (!parent) | |
1136 | return -ENOENT; | |
1137 | ||
1138 | offset = get_slot_offset(parent, slot); | |
1139 | ||
1140 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1141 | if (tag_get(parent, tag, offset)) | |
1142 | tags |= 1 << tag; | |
1143 | ||
1144 | for (end = offset + 1; end < RADIX_TREE_MAP_SIZE; end++) { | |
1145 | if (!is_sibling_entry(parent, parent->slots[end])) | |
1146 | break; | |
1147 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1148 | if (tags & (1 << tag)) | |
1149 | tag_set(parent, tag, end); | |
1150 | /* rcu_assign_pointer ensures tags are set before RETRY */ | |
1151 | rcu_assign_pointer(parent->slots[end], RADIX_TREE_RETRY); | |
1152 | } | |
1153 | rcu_assign_pointer(parent->slots[offset], RADIX_TREE_RETRY); | |
1154 | parent->exceptional -= (end - offset); | |
1155 | ||
1156 | if (order == parent->shift) | |
1157 | return 0; | |
1158 | if (order > parent->shift) { | |
1159 | while (offset < end) | |
1160 | offset += insert_entries(parent, &parent->slots[offset], | |
1161 | RADIX_TREE_RETRY, order, true); | |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | node = parent; | |
1166 | ||
1167 | for (;;) { | |
1168 | if (node->shift > order) { | |
1169 | child = radix_tree_node_alloc(root); | |
1170 | if (!child) | |
1171 | goto nomem; | |
1172 | child->shift = node->shift - RADIX_TREE_MAP_SHIFT; | |
1173 | child->offset = offset; | |
1174 | child->count = 0; | |
1175 | child->parent = node; | |
1176 | if (node != parent) { | |
1177 | node->count++; | |
1178 | node->slots[offset] = node_to_entry(child); | |
1179 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1180 | if (tags & (1 << tag)) | |
1181 | tag_set(node, tag, offset); | |
1182 | } | |
1183 | ||
1184 | node = child; | |
1185 | offset = 0; | |
1186 | continue; | |
1187 | } | |
1188 | ||
1189 | n = insert_entries(node, &node->slots[offset], | |
1190 | RADIX_TREE_RETRY, order, false); | |
1191 | BUG_ON(n > RADIX_TREE_MAP_SIZE); | |
1192 | ||
1193 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1194 | if (tags & (1 << tag)) | |
1195 | tag_set(node, tag, offset); | |
1196 | offset += n; | |
1197 | ||
1198 | while (offset == RADIX_TREE_MAP_SIZE) { | |
1199 | if (node == parent) | |
1200 | break; | |
1201 | offset = node->offset; | |
1202 | child = node; | |
1203 | node = node->parent; | |
1204 | rcu_assign_pointer(node->slots[offset], | |
1205 | node_to_entry(child)); | |
1206 | offset++; | |
1207 | } | |
1208 | if ((node == parent) && (offset == end)) | |
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | nomem: | |
1213 | /* Shouldn't happen; did user forget to preload? */ | |
1214 | /* TODO: free all the allocated nodes */ | |
1215 | WARN_ON(1); | |
1216 | return -ENOMEM; | |
1217 | } | |
175542f5 MW |
1218 | #endif |
1219 | ||
1da177e4 LT |
1220 | /** |
1221 | * radix_tree_tag_set - set a tag on a radix tree node | |
1222 | * @root: radix tree root | |
1223 | * @index: index key | |
2fcd9005 | 1224 | * @tag: tag index |
1da177e4 | 1225 | * |
daff89f3 JC |
1226 | * Set the search tag (which must be < RADIX_TREE_MAX_TAGS) |
1227 | * corresponding to @index in the radix tree. From | |
1da177e4 LT |
1228 | * the root all the way down to the leaf node. |
1229 | * | |
2fcd9005 | 1230 | * Returns the address of the tagged item. Setting a tag on a not-present |
1da177e4 LT |
1231 | * item is a bug. |
1232 | */ | |
1233 | void *radix_tree_tag_set(struct radix_tree_root *root, | |
daff89f3 | 1234 | unsigned long index, unsigned int tag) |
1da177e4 | 1235 | { |
fb969909 RZ |
1236 | struct radix_tree_node *node, *parent; |
1237 | unsigned long maxindex; | |
1da177e4 | 1238 | |
9e85d811 | 1239 | radix_tree_load_root(root, &node, &maxindex); |
fb969909 | 1240 | BUG_ON(index > maxindex); |
1da177e4 | 1241 | |
b194d16c | 1242 | while (radix_tree_is_internal_node(node)) { |
fb969909 | 1243 | unsigned offset; |
1da177e4 | 1244 | |
4dd6c098 | 1245 | parent = entry_to_node(node); |
9e85d811 | 1246 | offset = radix_tree_descend(parent, &node, index); |
fb969909 RZ |
1247 | BUG_ON(!node); |
1248 | ||
1249 | if (!tag_get(parent, tag, offset)) | |
1250 | tag_set(parent, tag, offset); | |
1da177e4 LT |
1251 | } |
1252 | ||
612d6c19 | 1253 | /* set the root's tag bit */ |
fb969909 | 1254 | if (!root_tag_get(root, tag)) |
612d6c19 NP |
1255 | root_tag_set(root, tag); |
1256 | ||
fb969909 | 1257 | return node; |
1da177e4 LT |
1258 | } |
1259 | EXPORT_SYMBOL(radix_tree_tag_set); | |
1260 | ||
d604c324 MW |
1261 | static void node_tag_clear(struct radix_tree_root *root, |
1262 | struct radix_tree_node *node, | |
1263 | unsigned int tag, unsigned int offset) | |
1264 | { | |
1265 | while (node) { | |
1266 | if (!tag_get(node, tag, offset)) | |
1267 | return; | |
1268 | tag_clear(node, tag, offset); | |
1269 | if (any_tag_set(node, tag)) | |
1270 | return; | |
1271 | ||
1272 | offset = node->offset; | |
1273 | node = node->parent; | |
1274 | } | |
1275 | ||
1276 | /* clear the root's tag bit */ | |
1277 | if (root_tag_get(root, tag)) | |
1278 | root_tag_clear(root, tag); | |
1279 | } | |
1280 | ||
9498d2bb MW |
1281 | static void node_tag_set(struct radix_tree_root *root, |
1282 | struct radix_tree_node *node, | |
1283 | unsigned int tag, unsigned int offset) | |
1284 | { | |
1285 | while (node) { | |
1286 | if (tag_get(node, tag, offset)) | |
1287 | return; | |
1288 | tag_set(node, tag, offset); | |
1289 | offset = node->offset; | |
1290 | node = node->parent; | |
1291 | } | |
1292 | ||
1293 | if (!root_tag_get(root, tag)) | |
1294 | root_tag_set(root, tag); | |
1295 | } | |
1296 | ||
268f42de MW |
1297 | /** |
1298 | * radix_tree_iter_tag_set - set a tag on the current iterator entry | |
1299 | * @root: radix tree root | |
1300 | * @iter: iterator state | |
1301 | * @tag: tag to set | |
1302 | */ | |
1303 | void radix_tree_iter_tag_set(struct radix_tree_root *root, | |
1304 | const struct radix_tree_iter *iter, unsigned int tag) | |
1305 | { | |
1306 | node_tag_set(root, iter->node, tag, iter_offset(iter)); | |
1307 | } | |
1308 | ||
1da177e4 LT |
1309 | /** |
1310 | * radix_tree_tag_clear - clear a tag on a radix tree node | |
1311 | * @root: radix tree root | |
1312 | * @index: index key | |
2fcd9005 | 1313 | * @tag: tag index |
1da177e4 | 1314 | * |
daff89f3 | 1315 | * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS) |
2fcd9005 MW |
1316 | * corresponding to @index in the radix tree. If this causes |
1317 | * the leaf node to have no tags set then clear the tag in the | |
1da177e4 LT |
1318 | * next-to-leaf node, etc. |
1319 | * | |
1320 | * Returns the address of the tagged item on success, else NULL. ie: | |
1321 | * has the same return value and semantics as radix_tree_lookup(). | |
1322 | */ | |
1323 | void *radix_tree_tag_clear(struct radix_tree_root *root, | |
daff89f3 | 1324 | unsigned long index, unsigned int tag) |
1da177e4 | 1325 | { |
00f47b58 RZ |
1326 | struct radix_tree_node *node, *parent; |
1327 | unsigned long maxindex; | |
e2bdb933 | 1328 | int uninitialized_var(offset); |
1da177e4 | 1329 | |
9e85d811 | 1330 | radix_tree_load_root(root, &node, &maxindex); |
00f47b58 RZ |
1331 | if (index > maxindex) |
1332 | return NULL; | |
1da177e4 | 1333 | |
00f47b58 | 1334 | parent = NULL; |
1da177e4 | 1335 | |
b194d16c | 1336 | while (radix_tree_is_internal_node(node)) { |
4dd6c098 | 1337 | parent = entry_to_node(node); |
9e85d811 | 1338 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 LT |
1339 | } |
1340 | ||
d604c324 MW |
1341 | if (node) |
1342 | node_tag_clear(root, parent, tag, offset); | |
1da177e4 | 1343 | |
00f47b58 | 1344 | return node; |
1da177e4 LT |
1345 | } |
1346 | EXPORT_SYMBOL(radix_tree_tag_clear); | |
1347 | ||
1da177e4 | 1348 | /** |
32605a18 MT |
1349 | * radix_tree_tag_get - get a tag on a radix tree node |
1350 | * @root: radix tree root | |
1351 | * @index: index key | |
2fcd9005 | 1352 | * @tag: tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 | 1353 | * |
32605a18 | 1354 | * Return values: |
1da177e4 | 1355 | * |
612d6c19 NP |
1356 | * 0: tag not present or not set |
1357 | * 1: tag set | |
ce82653d DH |
1358 | * |
1359 | * Note that the return value of this function may not be relied on, even if | |
1360 | * the RCU lock is held, unless tag modification and node deletion are excluded | |
1361 | * from concurrency. | |
1da177e4 LT |
1362 | */ |
1363 | int radix_tree_tag_get(struct radix_tree_root *root, | |
daff89f3 | 1364 | unsigned long index, unsigned int tag) |
1da177e4 | 1365 | { |
4589ba6d RZ |
1366 | struct radix_tree_node *node, *parent; |
1367 | unsigned long maxindex; | |
1da177e4 | 1368 | |
612d6c19 NP |
1369 | if (!root_tag_get(root, tag)) |
1370 | return 0; | |
1371 | ||
9e85d811 | 1372 | radix_tree_load_root(root, &node, &maxindex); |
4589ba6d RZ |
1373 | if (index > maxindex) |
1374 | return 0; | |
7cf9c2c7 NP |
1375 | if (node == NULL) |
1376 | return 0; | |
1377 | ||
b194d16c | 1378 | while (radix_tree_is_internal_node(node)) { |
9e85d811 | 1379 | unsigned offset; |
1da177e4 | 1380 | |
4dd6c098 | 1381 | parent = entry_to_node(node); |
9e85d811 | 1382 | offset = radix_tree_descend(parent, &node, index); |
1da177e4 | 1383 | |
4589ba6d | 1384 | if (!node) |
1da177e4 | 1385 | return 0; |
4589ba6d | 1386 | if (!tag_get(parent, tag, offset)) |
3fa36acb | 1387 | return 0; |
4589ba6d RZ |
1388 | if (node == RADIX_TREE_RETRY) |
1389 | break; | |
1da177e4 | 1390 | } |
4589ba6d RZ |
1391 | |
1392 | return 1; | |
1da177e4 LT |
1393 | } |
1394 | EXPORT_SYMBOL(radix_tree_tag_get); | |
1da177e4 | 1395 | |
21ef5339 RZ |
1396 | static inline void __set_iter_shift(struct radix_tree_iter *iter, |
1397 | unsigned int shift) | |
1398 | { | |
1399 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1400 | iter->shift = shift; | |
1401 | #endif | |
1402 | } | |
1403 | ||
148deab2 MW |
1404 | /* Construct iter->tags bit-mask from node->tags[tag] array */ |
1405 | static void set_iter_tags(struct radix_tree_iter *iter, | |
1406 | struct radix_tree_node *node, unsigned offset, | |
1407 | unsigned tag) | |
1408 | { | |
1409 | unsigned tag_long = offset / BITS_PER_LONG; | |
1410 | unsigned tag_bit = offset % BITS_PER_LONG; | |
1411 | ||
1412 | iter->tags = node->tags[tag][tag_long] >> tag_bit; | |
1413 | ||
1414 | /* This never happens if RADIX_TREE_TAG_LONGS == 1 */ | |
1415 | if (tag_long < RADIX_TREE_TAG_LONGS - 1) { | |
1416 | /* Pick tags from next element */ | |
1417 | if (tag_bit) | |
1418 | iter->tags |= node->tags[tag][tag_long + 1] << | |
1419 | (BITS_PER_LONG - tag_bit); | |
1420 | /* Clip chunk size, here only BITS_PER_LONG tags */ | |
1421 | iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG); | |
1422 | } | |
1423 | } | |
1424 | ||
1425 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1426 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1427 | void **slot, struct radix_tree_iter *iter) | |
1428 | { | |
1429 | void *sib = node_to_entry(slot - 1); | |
1430 | ||
1431 | while (iter->index < iter->next_index) { | |
1432 | *nodep = rcu_dereference_raw(*slot); | |
1433 | if (*nodep && *nodep != sib) | |
1434 | return slot; | |
1435 | slot++; | |
1436 | iter->index = __radix_tree_iter_add(iter, 1); | |
1437 | iter->tags >>= 1; | |
1438 | } | |
1439 | ||
1440 | *nodep = NULL; | |
1441 | return NULL; | |
1442 | } | |
1443 | ||
1444 | void ** __radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, | |
1445 | unsigned flags) | |
1446 | { | |
1447 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; | |
1448 | struct radix_tree_node *node = rcu_dereference_raw(*slot); | |
1449 | ||
1450 | slot = skip_siblings(&node, slot, iter); | |
1451 | ||
1452 | while (radix_tree_is_internal_node(node)) { | |
1453 | unsigned offset; | |
1454 | unsigned long next_index; | |
1455 | ||
1456 | if (node == RADIX_TREE_RETRY) | |
1457 | return slot; | |
1458 | node = entry_to_node(node); | |
268f42de | 1459 | iter->node = node; |
148deab2 MW |
1460 | iter->shift = node->shift; |
1461 | ||
1462 | if (flags & RADIX_TREE_ITER_TAGGED) { | |
1463 | offset = radix_tree_find_next_bit(node, tag, 0); | |
1464 | if (offset == RADIX_TREE_MAP_SIZE) | |
1465 | return NULL; | |
1466 | slot = &node->slots[offset]; | |
1467 | iter->index = __radix_tree_iter_add(iter, offset); | |
1468 | set_iter_tags(iter, node, offset, tag); | |
1469 | node = rcu_dereference_raw(*slot); | |
1470 | } else { | |
1471 | offset = 0; | |
1472 | slot = &node->slots[0]; | |
1473 | for (;;) { | |
1474 | node = rcu_dereference_raw(*slot); | |
1475 | if (node) | |
1476 | break; | |
1477 | slot++; | |
1478 | offset++; | |
1479 | if (offset == RADIX_TREE_MAP_SIZE) | |
1480 | return NULL; | |
1481 | } | |
1482 | iter->index = __radix_tree_iter_add(iter, offset); | |
1483 | } | |
1484 | if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0)) | |
1485 | goto none; | |
1486 | next_index = (iter->index | shift_maxindex(iter->shift)) + 1; | |
1487 | if (next_index < iter->next_index) | |
1488 | iter->next_index = next_index; | |
1489 | } | |
1490 | ||
1491 | return slot; | |
1492 | none: | |
1493 | iter->next_index = 0; | |
1494 | return NULL; | |
1495 | } | |
1496 | EXPORT_SYMBOL(__radix_tree_next_slot); | |
1497 | #else | |
1498 | static void **skip_siblings(struct radix_tree_node **nodep, | |
1499 | void **slot, struct radix_tree_iter *iter) | |
1500 | { | |
1501 | return slot; | |
1502 | } | |
1503 | #endif | |
1504 | ||
1505 | void **radix_tree_iter_resume(void **slot, struct radix_tree_iter *iter) | |
1506 | { | |
1507 | struct radix_tree_node *node; | |
1508 | ||
1509 | slot++; | |
1510 | iter->index = __radix_tree_iter_add(iter, 1); | |
1511 | node = rcu_dereference_raw(*slot); | |
1512 | skip_siblings(&node, slot, iter); | |
1513 | iter->next_index = iter->index; | |
1514 | iter->tags = 0; | |
1515 | return NULL; | |
1516 | } | |
1517 | EXPORT_SYMBOL(radix_tree_iter_resume); | |
1518 | ||
78c1d784 KK |
1519 | /** |
1520 | * radix_tree_next_chunk - find next chunk of slots for iteration | |
1521 | * | |
1522 | * @root: radix tree root | |
1523 | * @iter: iterator state | |
1524 | * @flags: RADIX_TREE_ITER_* flags and tag index | |
1525 | * Returns: pointer to chunk first slot, or NULL if iteration is over | |
1526 | */ | |
1527 | void **radix_tree_next_chunk(struct radix_tree_root *root, | |
1528 | struct radix_tree_iter *iter, unsigned flags) | |
1529 | { | |
9e85d811 | 1530 | unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK; |
8c1244de | 1531 | struct radix_tree_node *node, *child; |
21ef5339 | 1532 | unsigned long index, offset, maxindex; |
78c1d784 KK |
1533 | |
1534 | if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag)) | |
1535 | return NULL; | |
1536 | ||
1537 | /* | |
1538 | * Catch next_index overflow after ~0UL. iter->index never overflows | |
1539 | * during iterating; it can be zero only at the beginning. | |
1540 | * And we cannot overflow iter->next_index in a single step, | |
1541 | * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG. | |
fffaee36 KK |
1542 | * |
1543 | * This condition also used by radix_tree_next_slot() to stop | |
91b9677c | 1544 | * contiguous iterating, and forbid switching to the next chunk. |
78c1d784 KK |
1545 | */ |
1546 | index = iter->next_index; | |
1547 | if (!index && iter->index) | |
1548 | return NULL; | |
1549 | ||
21ef5339 | 1550 | restart: |
9e85d811 | 1551 | radix_tree_load_root(root, &child, &maxindex); |
21ef5339 RZ |
1552 | if (index > maxindex) |
1553 | return NULL; | |
8c1244de MW |
1554 | if (!child) |
1555 | return NULL; | |
21ef5339 | 1556 | |
8c1244de | 1557 | if (!radix_tree_is_internal_node(child)) { |
78c1d784 | 1558 | /* Single-slot tree */ |
21ef5339 RZ |
1559 | iter->index = index; |
1560 | iter->next_index = maxindex + 1; | |
78c1d784 | 1561 | iter->tags = 1; |
268f42de | 1562 | iter->node = NULL; |
8c1244de | 1563 | __set_iter_shift(iter, 0); |
78c1d784 | 1564 | return (void **)&root->rnode; |
8c1244de | 1565 | } |
21ef5339 | 1566 | |
8c1244de MW |
1567 | do { |
1568 | node = entry_to_node(child); | |
9e85d811 | 1569 | offset = radix_tree_descend(node, &child, index); |
21ef5339 | 1570 | |
78c1d784 | 1571 | if ((flags & RADIX_TREE_ITER_TAGGED) ? |
8c1244de | 1572 | !tag_get(node, tag, offset) : !child) { |
78c1d784 KK |
1573 | /* Hole detected */ |
1574 | if (flags & RADIX_TREE_ITER_CONTIG) | |
1575 | return NULL; | |
1576 | ||
1577 | if (flags & RADIX_TREE_ITER_TAGGED) | |
bc412fca | 1578 | offset = radix_tree_find_next_bit(node, tag, |
78c1d784 KK |
1579 | offset + 1); |
1580 | else | |
1581 | while (++offset < RADIX_TREE_MAP_SIZE) { | |
21ef5339 RZ |
1582 | void *slot = node->slots[offset]; |
1583 | if (is_sibling_entry(node, slot)) | |
1584 | continue; | |
1585 | if (slot) | |
78c1d784 KK |
1586 | break; |
1587 | } | |
8c1244de | 1588 | index &= ~node_maxindex(node); |
9e85d811 | 1589 | index += offset << node->shift; |
78c1d784 KK |
1590 | /* Overflow after ~0UL */ |
1591 | if (!index) | |
1592 | return NULL; | |
1593 | if (offset == RADIX_TREE_MAP_SIZE) | |
1594 | goto restart; | |
8c1244de | 1595 | child = rcu_dereference_raw(node->slots[offset]); |
78c1d784 KK |
1596 | } |
1597 | ||
e157b555 | 1598 | if (!child) |
78c1d784 | 1599 | goto restart; |
e157b555 MW |
1600 | if (child == RADIX_TREE_RETRY) |
1601 | break; | |
8c1244de | 1602 | } while (radix_tree_is_internal_node(child)); |
78c1d784 KK |
1603 | |
1604 | /* Update the iterator state */ | |
8c1244de MW |
1605 | iter->index = (index &~ node_maxindex(node)) | (offset << node->shift); |
1606 | iter->next_index = (index | node_maxindex(node)) + 1; | |
268f42de | 1607 | iter->node = node; |
9e85d811 | 1608 | __set_iter_shift(iter, node->shift); |
78c1d784 | 1609 | |
148deab2 MW |
1610 | if (flags & RADIX_TREE_ITER_TAGGED) |
1611 | set_iter_tags(iter, node, offset, tag); | |
78c1d784 KK |
1612 | |
1613 | return node->slots + offset; | |
1614 | } | |
1615 | EXPORT_SYMBOL(radix_tree_next_chunk); | |
1616 | ||
1da177e4 LT |
1617 | /** |
1618 | * radix_tree_gang_lookup - perform multiple lookup on a radix tree | |
1619 | * @root: radix tree root | |
1620 | * @results: where the results of the lookup are placed | |
1621 | * @first_index: start the lookup from this key | |
1622 | * @max_items: place up to this many items at *results | |
1623 | * | |
1624 | * Performs an index-ascending scan of the tree for present items. Places | |
1625 | * them at *@results and returns the number of items which were placed at | |
1626 | * *@results. | |
1627 | * | |
1628 | * The implementation is naive. | |
7cf9c2c7 NP |
1629 | * |
1630 | * Like radix_tree_lookup, radix_tree_gang_lookup may be called under | |
1631 | * rcu_read_lock. In this case, rather than the returned results being | |
2fcd9005 MW |
1632 | * an atomic snapshot of the tree at a single point in time, the |
1633 | * semantics of an RCU protected gang lookup are as though multiple | |
1634 | * radix_tree_lookups have been issued in individual locks, and results | |
1635 | * stored in 'results'. | |
1da177e4 LT |
1636 | */ |
1637 | unsigned int | |
1638 | radix_tree_gang_lookup(struct radix_tree_root *root, void **results, | |
1639 | unsigned long first_index, unsigned int max_items) | |
1640 | { | |
cebbd29e KK |
1641 | struct radix_tree_iter iter; |
1642 | void **slot; | |
1643 | unsigned int ret = 0; | |
7cf9c2c7 | 1644 | |
cebbd29e | 1645 | if (unlikely(!max_items)) |
7cf9c2c7 | 1646 | return 0; |
1da177e4 | 1647 | |
cebbd29e | 1648 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
46437f9a | 1649 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1650 | if (!results[ret]) |
1651 | continue; | |
b194d16c | 1652 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1653 | slot = radix_tree_iter_retry(&iter); |
1654 | continue; | |
1655 | } | |
cebbd29e | 1656 | if (++ret == max_items) |
1da177e4 | 1657 | break; |
1da177e4 | 1658 | } |
7cf9c2c7 | 1659 | |
1da177e4 LT |
1660 | return ret; |
1661 | } | |
1662 | EXPORT_SYMBOL(radix_tree_gang_lookup); | |
1663 | ||
47feff2c NP |
1664 | /** |
1665 | * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree | |
1666 | * @root: radix tree root | |
1667 | * @results: where the results of the lookup are placed | |
6328650b | 1668 | * @indices: where their indices should be placed (but usually NULL) |
47feff2c NP |
1669 | * @first_index: start the lookup from this key |
1670 | * @max_items: place up to this many items at *results | |
1671 | * | |
1672 | * Performs an index-ascending scan of the tree for present items. Places | |
1673 | * their slots at *@results and returns the number of items which were | |
1674 | * placed at *@results. | |
1675 | * | |
1676 | * The implementation is naive. | |
1677 | * | |
1678 | * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must | |
1679 | * be dereferenced with radix_tree_deref_slot, and if using only RCU | |
1680 | * protection, radix_tree_deref_slot may fail requiring a retry. | |
1681 | */ | |
1682 | unsigned int | |
6328650b HD |
1683 | radix_tree_gang_lookup_slot(struct radix_tree_root *root, |
1684 | void ***results, unsigned long *indices, | |
47feff2c NP |
1685 | unsigned long first_index, unsigned int max_items) |
1686 | { | |
cebbd29e KK |
1687 | struct radix_tree_iter iter; |
1688 | void **slot; | |
1689 | unsigned int ret = 0; | |
47feff2c | 1690 | |
cebbd29e | 1691 | if (unlikely(!max_items)) |
47feff2c NP |
1692 | return 0; |
1693 | ||
cebbd29e KK |
1694 | radix_tree_for_each_slot(slot, root, &iter, first_index) { |
1695 | results[ret] = slot; | |
6328650b | 1696 | if (indices) |
cebbd29e KK |
1697 | indices[ret] = iter.index; |
1698 | if (++ret == max_items) | |
47feff2c | 1699 | break; |
47feff2c NP |
1700 | } |
1701 | ||
1702 | return ret; | |
1703 | } | |
1704 | EXPORT_SYMBOL(radix_tree_gang_lookup_slot); | |
1705 | ||
1da177e4 LT |
1706 | /** |
1707 | * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree | |
1708 | * based on a tag | |
1709 | * @root: radix tree root | |
1710 | * @results: where the results of the lookup are placed | |
1711 | * @first_index: start the lookup from this key | |
1712 | * @max_items: place up to this many items at *results | |
daff89f3 | 1713 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) |
1da177e4 LT |
1714 | * |
1715 | * Performs an index-ascending scan of the tree for present items which | |
1716 | * have the tag indexed by @tag set. Places the items at *@results and | |
1717 | * returns the number of items which were placed at *@results. | |
1718 | */ | |
1719 | unsigned int | |
1720 | radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results, | |
daff89f3 JC |
1721 | unsigned long first_index, unsigned int max_items, |
1722 | unsigned int tag) | |
1da177e4 | 1723 | { |
cebbd29e KK |
1724 | struct radix_tree_iter iter; |
1725 | void **slot; | |
1726 | unsigned int ret = 0; | |
612d6c19 | 1727 | |
cebbd29e | 1728 | if (unlikely(!max_items)) |
7cf9c2c7 NP |
1729 | return 0; |
1730 | ||
cebbd29e | 1731 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
46437f9a | 1732 | results[ret] = rcu_dereference_raw(*slot); |
cebbd29e KK |
1733 | if (!results[ret]) |
1734 | continue; | |
b194d16c | 1735 | if (radix_tree_is_internal_node(results[ret])) { |
46437f9a MW |
1736 | slot = radix_tree_iter_retry(&iter); |
1737 | continue; | |
1738 | } | |
cebbd29e | 1739 | if (++ret == max_items) |
1da177e4 | 1740 | break; |
1da177e4 | 1741 | } |
7cf9c2c7 | 1742 | |
1da177e4 LT |
1743 | return ret; |
1744 | } | |
1745 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag); | |
1746 | ||
47feff2c NP |
1747 | /** |
1748 | * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a | |
1749 | * radix tree based on a tag | |
1750 | * @root: radix tree root | |
1751 | * @results: where the results of the lookup are placed | |
1752 | * @first_index: start the lookup from this key | |
1753 | * @max_items: place up to this many items at *results | |
1754 | * @tag: the tag index (< RADIX_TREE_MAX_TAGS) | |
1755 | * | |
1756 | * Performs an index-ascending scan of the tree for present items which | |
1757 | * have the tag indexed by @tag set. Places the slots at *@results and | |
1758 | * returns the number of slots which were placed at *@results. | |
1759 | */ | |
1760 | unsigned int | |
1761 | radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results, | |
1762 | unsigned long first_index, unsigned int max_items, | |
1763 | unsigned int tag) | |
1764 | { | |
cebbd29e KK |
1765 | struct radix_tree_iter iter; |
1766 | void **slot; | |
1767 | unsigned int ret = 0; | |
47feff2c | 1768 | |
cebbd29e | 1769 | if (unlikely(!max_items)) |
47feff2c NP |
1770 | return 0; |
1771 | ||
cebbd29e KK |
1772 | radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) { |
1773 | results[ret] = slot; | |
1774 | if (++ret == max_items) | |
47feff2c | 1775 | break; |
47feff2c NP |
1776 | } |
1777 | ||
1778 | return ret; | |
1779 | } | |
1780 | EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot); | |
1781 | ||
139e5616 JW |
1782 | /** |
1783 | * __radix_tree_delete_node - try to free node after clearing a slot | |
1784 | * @root: radix tree root | |
139e5616 JW |
1785 | * @node: node containing @index |
1786 | * | |
1787 | * After clearing the slot at @index in @node from radix tree | |
1788 | * rooted at @root, call this function to attempt freeing the | |
1789 | * node and shrinking the tree. | |
139e5616 | 1790 | */ |
14b46879 | 1791 | void __radix_tree_delete_node(struct radix_tree_root *root, |
139e5616 JW |
1792 | struct radix_tree_node *node) |
1793 | { | |
14b46879 | 1794 | delete_node(root, node, NULL, NULL); |
139e5616 JW |
1795 | } |
1796 | ||
57578c2e MW |
1797 | static inline void delete_sibling_entries(struct radix_tree_node *node, |
1798 | void *ptr, unsigned offset) | |
1799 | { | |
1800 | #ifdef CONFIG_RADIX_TREE_MULTIORDER | |
1801 | int i; | |
1802 | for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) { | |
1803 | if (node->slots[offset + i] != ptr) | |
1804 | break; | |
1805 | node->slots[offset + i] = NULL; | |
1806 | node->count--; | |
1807 | } | |
1808 | #endif | |
1809 | } | |
1810 | ||
1da177e4 | 1811 | /** |
53c59f26 | 1812 | * radix_tree_delete_item - delete an item from a radix tree |
1da177e4 LT |
1813 | * @root: radix tree root |
1814 | * @index: index key | |
53c59f26 | 1815 | * @item: expected item |
1da177e4 | 1816 | * |
53c59f26 | 1817 | * Remove @item at @index from the radix tree rooted at @root. |
1da177e4 | 1818 | * |
53c59f26 JW |
1819 | * Returns the address of the deleted item, or NULL if it was not present |
1820 | * or the entry at the given @index was not @item. | |
1da177e4 | 1821 | */ |
53c59f26 JW |
1822 | void *radix_tree_delete_item(struct radix_tree_root *root, |
1823 | unsigned long index, void *item) | |
1da177e4 | 1824 | { |
139e5616 | 1825 | struct radix_tree_node *node; |
57578c2e | 1826 | unsigned int offset; |
139e5616 JW |
1827 | void **slot; |
1828 | void *entry; | |
d5274261 | 1829 | int tag; |
1da177e4 | 1830 | |
139e5616 JW |
1831 | entry = __radix_tree_lookup(root, index, &node, &slot); |
1832 | if (!entry) | |
1833 | return NULL; | |
1da177e4 | 1834 | |
139e5616 JW |
1835 | if (item && entry != item) |
1836 | return NULL; | |
1837 | ||
1838 | if (!node) { | |
612d6c19 NP |
1839 | root_tag_clear_all(root); |
1840 | root->rnode = NULL; | |
139e5616 | 1841 | return entry; |
612d6c19 | 1842 | } |
1da177e4 | 1843 | |
29e0967c | 1844 | offset = get_slot_offset(node, slot); |
53c59f26 | 1845 | |
d604c324 MW |
1846 | /* Clear all tags associated with the item to be deleted. */ |
1847 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1848 | node_tag_clear(root, node, tag, offset); | |
1da177e4 | 1849 | |
a4db4dce | 1850 | delete_sibling_entries(node, node_to_entry(slot), offset); |
4d693d08 | 1851 | __radix_tree_replace(root, node, slot, NULL, NULL, NULL); |
612d6c19 | 1852 | |
139e5616 | 1853 | return entry; |
1da177e4 | 1854 | } |
53c59f26 JW |
1855 | EXPORT_SYMBOL(radix_tree_delete_item); |
1856 | ||
1857 | /** | |
1858 | * radix_tree_delete - delete an item from a radix tree | |
1859 | * @root: radix tree root | |
1860 | * @index: index key | |
1861 | * | |
1862 | * Remove the item at @index from the radix tree rooted at @root. | |
1863 | * | |
1864 | * Returns the address of the deleted item, or NULL if it was not present. | |
1865 | */ | |
1866 | void *radix_tree_delete(struct radix_tree_root *root, unsigned long index) | |
1867 | { | |
1868 | return radix_tree_delete_item(root, index, NULL); | |
1869 | } | |
1da177e4 LT |
1870 | EXPORT_SYMBOL(radix_tree_delete); |
1871 | ||
d3798ae8 JW |
1872 | void radix_tree_clear_tags(struct radix_tree_root *root, |
1873 | struct radix_tree_node *node, | |
1874 | void **slot) | |
d604c324 | 1875 | { |
d604c324 MW |
1876 | if (node) { |
1877 | unsigned int tag, offset = get_slot_offset(node, slot); | |
1878 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) | |
1879 | node_tag_clear(root, node, tag, offset); | |
1880 | } else { | |
1881 | /* Clear root node tags */ | |
1882 | root->gfp_mask &= __GFP_BITS_MASK; | |
1883 | } | |
d604c324 MW |
1884 | } |
1885 | ||
1da177e4 LT |
1886 | /** |
1887 | * radix_tree_tagged - test whether any items in the tree are tagged | |
1888 | * @root: radix tree root | |
1889 | * @tag: tag to test | |
1890 | */ | |
daff89f3 | 1891 | int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag) |
1da177e4 | 1892 | { |
612d6c19 | 1893 | return root_tag_get(root, tag); |
1da177e4 LT |
1894 | } |
1895 | EXPORT_SYMBOL(radix_tree_tagged); | |
1896 | ||
1897 | static void | |
449dd698 | 1898 | radix_tree_node_ctor(void *arg) |
1da177e4 | 1899 | { |
449dd698 JW |
1900 | struct radix_tree_node *node = arg; |
1901 | ||
1902 | memset(node, 0, sizeof(*node)); | |
1903 | INIT_LIST_HEAD(&node->private_list); | |
1da177e4 LT |
1904 | } |
1905 | ||
c78c66d1 KS |
1906 | static __init unsigned long __maxindex(unsigned int height) |
1907 | { | |
1908 | unsigned int width = height * RADIX_TREE_MAP_SHIFT; | |
1909 | int shift = RADIX_TREE_INDEX_BITS - width; | |
1910 | ||
1911 | if (shift < 0) | |
1912 | return ~0UL; | |
1913 | if (shift >= BITS_PER_LONG) | |
1914 | return 0UL; | |
1915 | return ~0UL >> shift; | |
1916 | } | |
1917 | ||
1918 | static __init void radix_tree_init_maxnodes(void) | |
1919 | { | |
1920 | unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1]; | |
1921 | unsigned int i, j; | |
1922 | ||
1923 | for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++) | |
1924 | height_to_maxindex[i] = __maxindex(i); | |
1925 | for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) { | |
1926 | for (j = i; j > 0; j--) | |
1927 | height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1; | |
1928 | } | |
1929 | } | |
1930 | ||
d544abd5 | 1931 | static int radix_tree_cpu_dead(unsigned int cpu) |
1da177e4 | 1932 | { |
2fcd9005 MW |
1933 | struct radix_tree_preload *rtp; |
1934 | struct radix_tree_node *node; | |
1935 | ||
1936 | /* Free per-cpu pool of preloaded nodes */ | |
d544abd5 SAS |
1937 | rtp = &per_cpu(radix_tree_preloads, cpu); |
1938 | while (rtp->nr) { | |
1939 | node = rtp->nodes; | |
1940 | rtp->nodes = node->private_data; | |
1941 | kmem_cache_free(radix_tree_node_cachep, node); | |
1942 | rtp->nr--; | |
2fcd9005 | 1943 | } |
d544abd5 | 1944 | return 0; |
1da177e4 | 1945 | } |
1da177e4 LT |
1946 | |
1947 | void __init radix_tree_init(void) | |
1948 | { | |
d544abd5 | 1949 | int ret; |
1da177e4 LT |
1950 | radix_tree_node_cachep = kmem_cache_create("radix_tree_node", |
1951 | sizeof(struct radix_tree_node), 0, | |
488514d1 CL |
1952 | SLAB_PANIC | SLAB_RECLAIM_ACCOUNT, |
1953 | radix_tree_node_ctor); | |
c78c66d1 | 1954 | radix_tree_init_maxnodes(); |
d544abd5 SAS |
1955 | ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead", |
1956 | NULL, radix_tree_cpu_dead); | |
1957 | WARN_ON(ret < 0); | |
1da177e4 | 1958 | } |