]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
4 | * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au> | |
5 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> | |
6 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> | |
7 | * | |
8 | * Code partially derived from nft_hash | |
9 | * Rewritten with rehash code from br_multicast plus single list | |
10 | * pointer as suggested by Josh Triplett | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or modify | |
13 | * it under the terms of the GNU General Public License version 2 as | |
14 | * published by the Free Software Foundation. | |
15 | */ | |
16 | ||
17 | #ifndef _LINUX_RHASHTABLE_H | |
18 | #define _LINUX_RHASHTABLE_H | |
19 | ||
20 | #include <linux/compiler.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/jhash.h> | |
23 | #include <linux/list_nulls.h> | |
24 | #include <linux/workqueue.h> | |
25 | #include <linux/mutex.h> | |
26 | #include <linux/rcupdate.h> | |
27 | ||
28 | /* | |
29 | * The end of the chain is marked with a special nulls marks which has | |
30 | * the following format: | |
31 | * | |
32 | * +-------+-----------------------------------------------------+-+ | |
33 | * | Base | Hash |1| | |
34 | * +-------+-----------------------------------------------------+-+ | |
35 | * | |
36 | * Base (4 bits) : Reserved to distinguish between multiple tables. | |
37 | * Specified via &struct rhashtable_params.nulls_base. | |
38 | * Hash (27 bits): Full hash (unmasked) of first element added to bucket | |
39 | * 1 (1 bit) : Nulls marker (always set) | |
40 | * | |
41 | * The remaining bits of the next pointer remain unused for now. | |
42 | */ | |
43 | #define RHT_BASE_BITS 4 | |
44 | #define RHT_HASH_BITS 27 | |
45 | #define RHT_BASE_SHIFT RHT_HASH_BITS | |
46 | ||
47 | /* Base bits plus 1 bit for nulls marker */ | |
48 | #define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1) | |
49 | ||
50 | struct rhash_head { | |
51 | struct rhash_head __rcu *next; | |
52 | }; | |
53 | ||
54 | /** | |
55 | * struct bucket_table - Table of hash buckets | |
56 | * @size: Number of hash buckets | |
57 | * @rehash: Current bucket being rehashed | |
58 | * @hash_rnd: Random seed to fold into hash | |
59 | * @locks_mask: Mask to apply before accessing locks[] | |
60 | * @locks: Array of spinlocks protecting individual buckets | |
61 | * @walkers: List of active walkers | |
62 | * @rcu: RCU structure for freeing the table | |
63 | * @future_tbl: Table under construction during rehashing | |
64 | * @buckets: size * hash buckets | |
65 | */ | |
66 | struct bucket_table { | |
67 | unsigned int size; | |
68 | unsigned int rehash; | |
69 | u32 hash_rnd; | |
70 | unsigned int locks_mask; | |
71 | spinlock_t *locks; | |
72 | struct list_head walkers; | |
73 | struct rcu_head rcu; | |
74 | ||
75 | struct bucket_table __rcu *future_tbl; | |
76 | ||
77 | struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp; | |
78 | }; | |
79 | ||
80 | /** | |
81 | * struct rhashtable_compare_arg - Key for the function rhashtable_compare | |
82 | * @ht: Hash table | |
83 | * @key: Key to compare against | |
84 | */ | |
85 | struct rhashtable_compare_arg { | |
86 | struct rhashtable *ht; | |
87 | const void *key; | |
88 | }; | |
89 | ||
90 | typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); | |
91 | typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 seed); | |
92 | typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg, | |
93 | const void *obj); | |
94 | ||
95 | struct rhashtable; | |
96 | ||
97 | /** | |
98 | * struct rhashtable_params - Hash table construction parameters | |
99 | * @nelem_hint: Hint on number of elements, should be 75% of desired size | |
100 | * @key_len: Length of key | |
101 | * @key_offset: Offset of key in struct to be hashed | |
102 | * @head_offset: Offset of rhash_head in struct to be hashed | |
103 | * @max_size: Maximum size while expanding | |
104 | * @min_size: Minimum size while shrinking | |
105 | * @nulls_base: Base value to generate nulls marker | |
106 | * @insecure_elasticity: Set to true to disable chain length checks | |
107 | * @automatic_shrinking: Enable automatic shrinking of tables | |
108 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) | |
109 | * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash) | |
110 | * @obj_hashfn: Function to hash object | |
111 | * @obj_cmpfn: Function to compare key with object | |
112 | */ | |
113 | struct rhashtable_params { | |
114 | size_t nelem_hint; | |
115 | size_t key_len; | |
116 | size_t key_offset; | |
117 | size_t head_offset; | |
118 | unsigned int max_size; | |
119 | unsigned int min_size; | |
120 | u32 nulls_base; | |
121 | bool insecure_elasticity; | |
122 | bool automatic_shrinking; | |
123 | size_t locks_mul; | |
124 | rht_hashfn_t hashfn; | |
125 | rht_obj_hashfn_t obj_hashfn; | |
126 | rht_obj_cmpfn_t obj_cmpfn; | |
127 | }; | |
128 | ||
129 | /** | |
130 | * struct rhashtable - Hash table handle | |
131 | * @tbl: Bucket table | |
132 | * @nelems: Number of elements in table | |
133 | * @key_len: Key length for hashfn | |
134 | * @elasticity: Maximum chain length before rehash | |
135 | * @p: Configuration parameters | |
136 | * @run_work: Deferred worker to expand/shrink asynchronously | |
137 | * @mutex: Mutex to protect current/future table swapping | |
138 | * @lock: Spin lock to protect walker list | |
139 | * @being_destroyed: True if table is set up for destruction | |
140 | */ | |
141 | struct rhashtable { | |
142 | struct bucket_table __rcu *tbl; | |
143 | atomic_t nelems; | |
144 | bool being_destroyed; | |
145 | unsigned int key_len; | |
146 | unsigned int elasticity; | |
147 | struct rhashtable_params p; | |
148 | struct work_struct run_work; | |
149 | struct mutex mutex; | |
150 | spinlock_t lock; | |
151 | }; | |
152 | ||
153 | /** | |
154 | * struct rhashtable_walker - Hash table walker | |
155 | * @list: List entry on list of walkers | |
156 | * @tbl: The table that we were walking over | |
157 | */ | |
158 | struct rhashtable_walker { | |
159 | struct list_head list; | |
160 | struct bucket_table *tbl; | |
161 | }; | |
162 | ||
163 | /** | |
164 | * struct rhashtable_iter - Hash table iterator, fits into netlink cb | |
165 | * @ht: Table to iterate through | |
166 | * @p: Current pointer | |
167 | * @walker: Associated rhashtable walker | |
168 | * @slot: Current slot | |
169 | * @skip: Number of entries to skip in slot | |
170 | */ | |
171 | struct rhashtable_iter { | |
172 | struct rhashtable *ht; | |
173 | struct rhash_head *p; | |
174 | struct rhashtable_walker *walker; | |
175 | unsigned int slot; | |
176 | unsigned int skip; | |
177 | }; | |
178 | ||
179 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) | |
180 | { | |
181 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
182 | } | |
183 | ||
184 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
185 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
186 | ||
187 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
188 | { | |
189 | return ((unsigned long) ptr & 1); | |
190 | } | |
191 | ||
192 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
193 | { | |
194 | return ((unsigned long) ptr) >> 1; | |
195 | } | |
196 | ||
197 | static inline void *rht_obj(const struct rhashtable *ht, | |
198 | const struct rhash_head *he) | |
199 | { | |
200 | return (char *)he - ht->p.head_offset; | |
201 | } | |
202 | ||
203 | static inline unsigned int rht_bucket_index(const struct bucket_table *tbl, | |
204 | unsigned int hash) | |
205 | { | |
206 | return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1); | |
207 | } | |
208 | ||
209 | static inline unsigned int rht_key_hashfn( | |
210 | struct rhashtable *ht, const struct bucket_table *tbl, | |
211 | const void *key, const struct rhashtable_params params) | |
212 | { | |
213 | unsigned int hash; | |
214 | ||
215 | /* params must be equal to ht->p if it isn't constant. */ | |
216 | if (!__builtin_constant_p(params.key_len)) | |
217 | hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd); | |
218 | else if (params.key_len) { | |
219 | unsigned int key_len = params.key_len; | |
220 | ||
221 | if (params.hashfn) | |
222 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
223 | else if (key_len & (sizeof(u32) - 1)) | |
224 | hash = jhash(key, key_len, tbl->hash_rnd); | |
225 | else | |
226 | hash = jhash2(key, key_len / sizeof(u32), | |
227 | tbl->hash_rnd); | |
228 | } else { | |
229 | unsigned int key_len = ht->p.key_len; | |
230 | ||
231 | if (params.hashfn) | |
232 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
233 | else | |
234 | hash = jhash(key, key_len, tbl->hash_rnd); | |
235 | } | |
236 | ||
237 | return rht_bucket_index(tbl, hash); | |
238 | } | |
239 | ||
240 | static inline unsigned int rht_head_hashfn( | |
241 | struct rhashtable *ht, const struct bucket_table *tbl, | |
242 | const struct rhash_head *he, const struct rhashtable_params params) | |
243 | { | |
244 | const char *ptr = rht_obj(ht, he); | |
245 | ||
246 | return likely(params.obj_hashfn) ? | |
247 | rht_bucket_index(tbl, params.obj_hashfn(ptr, tbl->hash_rnd)) : | |
248 | rht_key_hashfn(ht, tbl, ptr + params.key_offset, params); | |
249 | } | |
250 | ||
251 | /** | |
252 | * rht_grow_above_75 - returns true if nelems > 0.75 * table-size | |
253 | * @ht: hash table | |
254 | * @tbl: current table | |
255 | */ | |
256 | static inline bool rht_grow_above_75(const struct rhashtable *ht, | |
257 | const struct bucket_table *tbl) | |
258 | { | |
259 | /* Expand table when exceeding 75% load */ | |
260 | return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) && | |
261 | (!ht->p.max_size || tbl->size < ht->p.max_size); | |
262 | } | |
263 | ||
264 | /** | |
265 | * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size | |
266 | * @ht: hash table | |
267 | * @tbl: current table | |
268 | */ | |
269 | static inline bool rht_shrink_below_30(const struct rhashtable *ht, | |
270 | const struct bucket_table *tbl) | |
271 | { | |
272 | /* Shrink table beneath 30% load */ | |
273 | return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) && | |
274 | tbl->size > ht->p.min_size; | |
275 | } | |
276 | ||
277 | /** | |
278 | * rht_grow_above_100 - returns true if nelems > table-size | |
279 | * @ht: hash table | |
280 | * @tbl: current table | |
281 | */ | |
282 | static inline bool rht_grow_above_100(const struct rhashtable *ht, | |
283 | const struct bucket_table *tbl) | |
284 | { | |
285 | return atomic_read(&ht->nelems) > tbl->size; | |
286 | } | |
287 | ||
288 | /* The bucket lock is selected based on the hash and protects mutations | |
289 | * on a group of hash buckets. | |
290 | * | |
291 | * A maximum of tbl->size/2 bucket locks is allocated. This ensures that | |
292 | * a single lock always covers both buckets which may both contains | |
293 | * entries which link to the same bucket of the old table during resizing. | |
294 | * This allows to simplify the locking as locking the bucket in both | |
295 | * tables during resize always guarantee protection. | |
296 | * | |
297 | * IMPORTANT: When holding the bucket lock of both the old and new table | |
298 | * during expansions and shrinking, the old bucket lock must always be | |
299 | * acquired first. | |
300 | */ | |
301 | static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl, | |
302 | unsigned int hash) | |
303 | { | |
304 | return &tbl->locks[hash & tbl->locks_mask]; | |
305 | } | |
306 | ||
307 | #ifdef CONFIG_PROVE_LOCKING | |
308 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); | |
309 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); | |
310 | #else | |
311 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
312 | { | |
313 | return 1; | |
314 | } | |
315 | ||
316 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
317 | u32 hash) | |
318 | { | |
319 | return 1; | |
320 | } | |
321 | #endif /* CONFIG_PROVE_LOCKING */ | |
322 | ||
323 | int rhashtable_init(struct rhashtable *ht, | |
324 | const struct rhashtable_params *params); | |
325 | ||
326 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
327 | struct rhash_head *obj, | |
328 | struct bucket_table *old_tbl); | |
329 | int rhashtable_insert_rehash(struct rhashtable *ht); | |
330 | ||
331 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter); | |
332 | void rhashtable_walk_exit(struct rhashtable_iter *iter); | |
333 | int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); | |
334 | void *rhashtable_walk_next(struct rhashtable_iter *iter); | |
335 | void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); | |
336 | ||
337 | void rhashtable_destroy(struct rhashtable *ht); | |
338 | ||
339 | #define rht_dereference(p, ht) \ | |
340 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
341 | ||
342 | #define rht_dereference_rcu(p, ht) \ | |
343 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
344 | ||
345 | #define rht_dereference_bucket(p, tbl, hash) \ | |
346 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
347 | ||
348 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ | |
349 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
350 | ||
351 | #define rht_entry(tpos, pos, member) \ | |
352 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
353 | ||
354 | /** | |
355 | * rht_for_each_continue - continue iterating over hash chain | |
356 | * @pos: the &struct rhash_head to use as a loop cursor. | |
357 | * @head: the previous &struct rhash_head to continue from | |
358 | * @tbl: the &struct bucket_table | |
359 | * @hash: the hash value / bucket index | |
360 | */ | |
361 | #define rht_for_each_continue(pos, head, tbl, hash) \ | |
362 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
363 | !rht_is_a_nulls(pos); \ | |
364 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) | |
365 | ||
366 | /** | |
367 | * rht_for_each - iterate over hash chain | |
368 | * @pos: the &struct rhash_head to use as a loop cursor. | |
369 | * @tbl: the &struct bucket_table | |
370 | * @hash: the hash value / bucket index | |
371 | */ | |
372 | #define rht_for_each(pos, tbl, hash) \ | |
373 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
374 | ||
375 | /** | |
376 | * rht_for_each_entry_continue - continue iterating over hash chain | |
377 | * @tpos: the type * to use as a loop cursor. | |
378 | * @pos: the &struct rhash_head to use as a loop cursor. | |
379 | * @head: the previous &struct rhash_head to continue from | |
380 | * @tbl: the &struct bucket_table | |
381 | * @hash: the hash value / bucket index | |
382 | * @member: name of the &struct rhash_head within the hashable struct. | |
383 | */ | |
384 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
385 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
386 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
387 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) | |
388 | ||
389 | /** | |
390 | * rht_for_each_entry - iterate over hash chain of given type | |
391 | * @tpos: the type * to use as a loop cursor. | |
392 | * @pos: the &struct rhash_head to use as a loop cursor. | |
393 | * @tbl: the &struct bucket_table | |
394 | * @hash: the hash value / bucket index | |
395 | * @member: name of the &struct rhash_head within the hashable struct. | |
396 | */ | |
397 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ | |
398 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
399 | tbl, hash, member) | |
400 | ||
401 | /** | |
402 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
403 | * @tpos: the type * to use as a loop cursor. | |
404 | * @pos: the &struct rhash_head to use as a loop cursor. | |
405 | * @next: the &struct rhash_head to use as next in loop cursor. | |
406 | * @tbl: the &struct bucket_table | |
407 | * @hash: the hash value / bucket index | |
408 | * @member: name of the &struct rhash_head within the hashable struct. | |
409 | * | |
410 | * This hash chain list-traversal primitive allows for the looped code to | |
411 | * remove the loop cursor from the list. | |
412 | */ | |
413 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ | |
414 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
415 | next = !rht_is_a_nulls(pos) ? \ | |
416 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
417 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
418 | pos = next, \ | |
419 | next = !rht_is_a_nulls(pos) ? \ | |
420 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
421 | ||
422 | /** | |
423 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
424 | * @pos: the &struct rhash_head to use as a loop cursor. | |
425 | * @head: the previous &struct rhash_head to continue from | |
426 | * @tbl: the &struct bucket_table | |
427 | * @hash: the hash value / bucket index | |
428 | * | |
429 | * This hash chain list-traversal primitive may safely run concurrently with | |
430 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
431 | * traversal is guarded by rcu_read_lock(). | |
432 | */ | |
433 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
434 | for (({barrier(); }), \ | |
435 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
436 | !rht_is_a_nulls(pos); \ | |
437 | pos = rcu_dereference_raw(pos->next)) | |
438 | ||
439 | /** | |
440 | * rht_for_each_rcu - iterate over rcu hash chain | |
441 | * @pos: the &struct rhash_head to use as a loop cursor. | |
442 | * @tbl: the &struct bucket_table | |
443 | * @hash: the hash value / bucket index | |
444 | * | |
445 | * This hash chain list-traversal primitive may safely run concurrently with | |
446 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
447 | * traversal is guarded by rcu_read_lock(). | |
448 | */ | |
449 | #define rht_for_each_rcu(pos, tbl, hash) \ | |
450 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
451 | ||
452 | /** | |
453 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
454 | * @tpos: the type * to use as a loop cursor. | |
455 | * @pos: the &struct rhash_head to use as a loop cursor. | |
456 | * @head: the previous &struct rhash_head to continue from | |
457 | * @tbl: the &struct bucket_table | |
458 | * @hash: the hash value / bucket index | |
459 | * @member: name of the &struct rhash_head within the hashable struct. | |
460 | * | |
461 | * This hash chain list-traversal primitive may safely run concurrently with | |
462 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
463 | * traversal is guarded by rcu_read_lock(). | |
464 | */ | |
465 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
466 | for (({barrier(); }), \ | |
467 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
468 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
469 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) | |
470 | ||
471 | /** | |
472 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
473 | * @tpos: the type * to use as a loop cursor. | |
474 | * @pos: the &struct rhash_head to use as a loop cursor. | |
475 | * @tbl: the &struct bucket_table | |
476 | * @hash: the hash value / bucket index | |
477 | * @member: name of the &struct rhash_head within the hashable struct. | |
478 | * | |
479 | * This hash chain list-traversal primitive may safely run concurrently with | |
480 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
481 | * traversal is guarded by rcu_read_lock(). | |
482 | */ | |
483 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ | |
484 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
485 | tbl, hash, member) | |
486 | ||
487 | static inline int rhashtable_compare(struct rhashtable_compare_arg *arg, | |
488 | const void *obj) | |
489 | { | |
490 | struct rhashtable *ht = arg->ht; | |
491 | const char *ptr = obj; | |
492 | ||
493 | return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len); | |
494 | } | |
495 | ||
496 | /** | |
497 | * rhashtable_lookup_fast - search hash table, inlined version | |
498 | * @ht: hash table | |
499 | * @key: the pointer to the key | |
500 | * @params: hash table parameters | |
501 | * | |
502 | * Computes the hash value for the key and traverses the bucket chain looking | |
503 | * for a entry with an identical key. The first matching entry is returned. | |
504 | * | |
505 | * Returns the first entry on which the compare function returned true. | |
506 | */ | |
507 | static inline void *rhashtable_lookup_fast( | |
508 | struct rhashtable *ht, const void *key, | |
509 | const struct rhashtable_params params) | |
510 | { | |
511 | struct rhashtable_compare_arg arg = { | |
512 | .ht = ht, | |
513 | .key = key, | |
514 | }; | |
515 | const struct bucket_table *tbl; | |
516 | struct rhash_head *he; | |
517 | unsigned int hash; | |
518 | ||
519 | rcu_read_lock(); | |
520 | ||
521 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
522 | restart: | |
523 | hash = rht_key_hashfn(ht, tbl, key, params); | |
524 | rht_for_each_rcu(he, tbl, hash) { | |
525 | if (params.obj_cmpfn ? | |
526 | params.obj_cmpfn(&arg, rht_obj(ht, he)) : | |
527 | rhashtable_compare(&arg, rht_obj(ht, he))) | |
528 | continue; | |
529 | rcu_read_unlock(); | |
530 | return rht_obj(ht, he); | |
531 | } | |
532 | ||
533 | /* Ensure we see any new tables. */ | |
534 | smp_rmb(); | |
535 | ||
536 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
537 | if (unlikely(tbl)) | |
538 | goto restart; | |
539 | rcu_read_unlock(); | |
540 | ||
541 | return NULL; | |
542 | } | |
543 | ||
544 | /* Internal function, please use rhashtable_insert_fast() instead */ | |
545 | static inline int __rhashtable_insert_fast( | |
546 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
547 | const struct rhashtable_params params) | |
548 | { | |
549 | struct rhashtable_compare_arg arg = { | |
550 | .ht = ht, | |
551 | .key = key, | |
552 | }; | |
553 | struct bucket_table *tbl, *new_tbl; | |
554 | struct rhash_head *head; | |
555 | spinlock_t *lock; | |
556 | unsigned int elasticity; | |
557 | unsigned int hash; | |
558 | int err; | |
559 | ||
560 | restart: | |
561 | rcu_read_lock(); | |
562 | ||
563 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
564 | ||
565 | /* All insertions must grab the oldest table containing | |
566 | * the hashed bucket that is yet to be rehashed. | |
567 | */ | |
568 | for (;;) { | |
569 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
570 | lock = rht_bucket_lock(tbl, hash); | |
571 | spin_lock_bh(lock); | |
572 | ||
573 | if (tbl->rehash <= hash) | |
574 | break; | |
575 | ||
576 | spin_unlock_bh(lock); | |
577 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
578 | } | |
579 | ||
580 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
581 | if (unlikely(new_tbl)) { | |
582 | err = rhashtable_insert_slow(ht, key, obj, new_tbl); | |
583 | if (err == -EAGAIN) | |
584 | goto slow_path; | |
585 | goto out; | |
586 | } | |
587 | ||
588 | if (unlikely(rht_grow_above_100(ht, tbl))) { | |
589 | slow_path: | |
590 | spin_unlock_bh(lock); | |
591 | err = rhashtable_insert_rehash(ht); | |
592 | rcu_read_unlock(); | |
593 | if (err) | |
594 | return err; | |
595 | ||
596 | goto restart; | |
597 | } | |
598 | ||
599 | err = -EEXIST; | |
600 | elasticity = ht->elasticity; | |
601 | rht_for_each(head, tbl, hash) { | |
602 | if (key && | |
603 | unlikely(!(params.obj_cmpfn ? | |
604 | params.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
605 | rhashtable_compare(&arg, rht_obj(ht, head))))) | |
606 | goto out; | |
607 | if (!--elasticity) | |
608 | goto slow_path; | |
609 | } | |
610 | ||
611 | err = 0; | |
612 | ||
613 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
614 | ||
615 | RCU_INIT_POINTER(obj->next, head); | |
616 | ||
617 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
618 | ||
619 | atomic_inc(&ht->nelems); | |
620 | if (rht_grow_above_75(ht, tbl)) | |
621 | schedule_work(&ht->run_work); | |
622 | ||
623 | out: | |
624 | spin_unlock_bh(lock); | |
625 | rcu_read_unlock(); | |
626 | ||
627 | return err; | |
628 | } | |
629 | ||
630 | /** | |
631 | * rhashtable_insert_fast - insert object into hash table | |
632 | * @ht: hash table | |
633 | * @obj: pointer to hash head inside object | |
634 | * @params: hash table parameters | |
635 | * | |
636 | * Will take a per bucket spinlock to protect against mutual mutations | |
637 | * on the same bucket. Multiple insertions may occur in parallel unless | |
638 | * they map to the same bucket lock. | |
639 | * | |
640 | * It is safe to call this function from atomic context. | |
641 | * | |
642 | * Will trigger an automatic deferred table resizing if the size grows | |
643 | * beyond the watermark indicated by grow_decision() which can be passed | |
644 | * to rhashtable_init(). | |
645 | */ | |
646 | static inline int rhashtable_insert_fast( | |
647 | struct rhashtable *ht, struct rhash_head *obj, | |
648 | const struct rhashtable_params params) | |
649 | { | |
650 | return __rhashtable_insert_fast(ht, NULL, obj, params); | |
651 | } | |
652 | ||
653 | /** | |
654 | * rhashtable_lookup_insert_fast - lookup and insert object into hash table | |
655 | * @ht: hash table | |
656 | * @obj: pointer to hash head inside object | |
657 | * @params: hash table parameters | |
658 | * | |
659 | * Locks down the bucket chain in both the old and new table if a resize | |
660 | * is in progress to ensure that writers can't remove from the old table | |
661 | * and can't insert to the new table during the atomic operation of search | |
662 | * and insertion. Searches for duplicates in both the old and new table if | |
663 | * a resize is in progress. | |
664 | * | |
665 | * This lookup function may only be used for fixed key hash table (key_len | |
666 | * parameter set). It will BUG() if used inappropriately. | |
667 | * | |
668 | * It is safe to call this function from atomic context. | |
669 | * | |
670 | * Will trigger an automatic deferred table resizing if the size grows | |
671 | * beyond the watermark indicated by grow_decision() which can be passed | |
672 | * to rhashtable_init(). | |
673 | */ | |
674 | static inline int rhashtable_lookup_insert_fast( | |
675 | struct rhashtable *ht, struct rhash_head *obj, | |
676 | const struct rhashtable_params params) | |
677 | { | |
678 | const char *key = rht_obj(ht, obj); | |
679 | ||
680 | BUG_ON(ht->p.obj_hashfn); | |
681 | ||
682 | return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, | |
683 | params); | |
684 | } | |
685 | ||
686 | /** | |
687 | * rhashtable_lookup_insert_key - search and insert object to hash table | |
688 | * with explicit key | |
689 | * @ht: hash table | |
690 | * @key: key | |
691 | * @obj: pointer to hash head inside object | |
692 | * @params: hash table parameters | |
693 | * | |
694 | * Locks down the bucket chain in both the old and new table if a resize | |
695 | * is in progress to ensure that writers can't remove from the old table | |
696 | * and can't insert to the new table during the atomic operation of search | |
697 | * and insertion. Searches for duplicates in both the old and new table if | |
698 | * a resize is in progress. | |
699 | * | |
700 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
701 | * | |
702 | * Will trigger an automatic deferred table resizing if the size grows | |
703 | * beyond the watermark indicated by grow_decision() which can be passed | |
704 | * to rhashtable_init(). | |
705 | * | |
706 | * Returns zero on success. | |
707 | */ | |
708 | static inline int rhashtable_lookup_insert_key( | |
709 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
710 | const struct rhashtable_params params) | |
711 | { | |
712 | BUG_ON(!ht->p.obj_hashfn || !key); | |
713 | ||
714 | return __rhashtable_insert_fast(ht, key, obj, params); | |
715 | } | |
716 | ||
717 | /* Internal function, please use rhashtable_remove_fast() instead */ | |
718 | static inline int __rhashtable_remove_fast( | |
719 | struct rhashtable *ht, struct bucket_table *tbl, | |
720 | struct rhash_head *obj, const struct rhashtable_params params) | |
721 | { | |
722 | struct rhash_head __rcu **pprev; | |
723 | struct rhash_head *he; | |
724 | spinlock_t * lock; | |
725 | unsigned int hash; | |
726 | int err = -ENOENT; | |
727 | ||
728 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
729 | lock = rht_bucket_lock(tbl, hash); | |
730 | ||
731 | spin_lock_bh(lock); | |
732 | ||
733 | pprev = &tbl->buckets[hash]; | |
734 | rht_for_each(he, tbl, hash) { | |
735 | if (he != obj) { | |
736 | pprev = &he->next; | |
737 | continue; | |
738 | } | |
739 | ||
740 | rcu_assign_pointer(*pprev, obj->next); | |
741 | err = 0; | |
742 | break; | |
743 | } | |
744 | ||
745 | spin_unlock_bh(lock); | |
746 | ||
747 | return err; | |
748 | } | |
749 | ||
750 | /** | |
751 | * rhashtable_remove_fast - remove object from hash table | |
752 | * @ht: hash table | |
753 | * @obj: pointer to hash head inside object | |
754 | * @params: hash table parameters | |
755 | * | |
756 | * Since the hash chain is single linked, the removal operation needs to | |
757 | * walk the bucket chain upon removal. The removal operation is thus | |
758 | * considerable slow if the hash table is not correctly sized. | |
759 | * | |
760 | * Will automatically shrink the table via rhashtable_expand() if the | |
761 | * shrink_decision function specified at rhashtable_init() returns true. | |
762 | * | |
763 | * Returns zero on success, -ENOENT if the entry could not be found. | |
764 | */ | |
765 | static inline int rhashtable_remove_fast( | |
766 | struct rhashtable *ht, struct rhash_head *obj, | |
767 | const struct rhashtable_params params) | |
768 | { | |
769 | struct bucket_table *tbl; | |
770 | int err; | |
771 | ||
772 | rcu_read_lock(); | |
773 | ||
774 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
775 | ||
776 | /* Because we have already taken (and released) the bucket | |
777 | * lock in old_tbl, if we find that future_tbl is not yet | |
778 | * visible then that guarantees the entry to still be in | |
779 | * the old tbl if it exists. | |
780 | */ | |
781 | while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) && | |
782 | (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) | |
783 | ; | |
784 | ||
785 | if (err) | |
786 | goto out; | |
787 | ||
788 | atomic_dec(&ht->nelems); | |
789 | if (unlikely(ht->p.automatic_shrinking && | |
790 | rht_shrink_below_30(ht, tbl))) | |
791 | schedule_work(&ht->run_work); | |
792 | ||
793 | out: | |
794 | rcu_read_unlock(); | |
795 | ||
796 | return err; | |
797 | } | |
798 | ||
799 | #endif /* _LINUX_RHASHTABLE_H */ |