Commit | Line | Data |
---|---|---|
7e1e7763 TG |
1 | /* |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
dc0ee268 | 4 | * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au> |
7e1e7763 TG |
5 | * Copyright (c) 2014 Thomas Graf <tgraf@suug.ch> |
6 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> | |
7 | * | |
7e1e7763 | 8 | * Code partially derived from nft_hash |
dc0ee268 HX |
9 | * Rewritten with rehash code from br_multicast plus single list |
10 | * pointer as suggested by Josh Triplett | |
7e1e7763 TG |
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 | ||
f2dba9c6 | 20 | #include <linux/compiler.h> |
6626af69 | 21 | #include <linux/errno.h> |
31ccde2d | 22 | #include <linux/jhash.h> |
f89bd6f8 | 23 | #include <linux/list_nulls.h> |
97defe1e | 24 | #include <linux/workqueue.h> |
86b35b64 | 25 | #include <linux/mutex.h> |
02fd97c3 | 26 | #include <linux/rcupdate.h> |
7e1e7763 | 27 | |
f89bd6f8 TG |
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 | ||
02fd97c3 HX |
47 | /* Base bits plus 1 bit for nulls marker */ |
48 | #define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1) | |
49 | ||
7e1e7763 | 50 | struct rhash_head { |
5300fdcb | 51 | struct rhash_head __rcu *next; |
7e1e7763 TG |
52 | }; |
53 | ||
97defe1e TG |
54 | /** |
55 | * struct bucket_table - Table of hash buckets | |
56 | * @size: Number of hash buckets | |
63d512d0 | 57 | * @rehash: Current bucket being rehashed |
988dfbd7 | 58 | * @hash_rnd: Random seed to fold into hash |
97defe1e TG |
59 | * @locks_mask: Mask to apply before accessing locks[] |
60 | * @locks: Array of spinlocks protecting individual buckets | |
eddee5ba | 61 | * @walkers: List of active walkers |
9d901bc0 | 62 | * @rcu: RCU structure for freeing the table |
c4db8848 | 63 | * @future_tbl: Table under construction during rehashing |
97defe1e TG |
64 | * @buckets: size * hash buckets |
65 | */ | |
7e1e7763 | 66 | struct bucket_table { |
63d512d0 HX |
67 | unsigned int size; |
68 | unsigned int rehash; | |
988dfbd7 | 69 | u32 hash_rnd; |
b9ebafbe ED |
70 | unsigned int locks_mask; |
71 | spinlock_t *locks; | |
eddee5ba | 72 | struct list_head walkers; |
9d901bc0 | 73 | struct rcu_head rcu; |
b9ebafbe | 74 | |
c4db8848 HX |
75 | struct bucket_table __rcu *future_tbl; |
76 | ||
b9ebafbe | 77 | struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp; |
7e1e7763 TG |
78 | }; |
79 | ||
02fd97c3 HX |
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 | ||
7e1e7763 TG |
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); | |
02fd97c3 HX |
92 | typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg, |
93 | const void *obj); | |
7e1e7763 TG |
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 | |
c2e213cf HX |
103 | * @max_size: Maximum size while expanding |
104 | * @min_size: Minimum size while shrinking | |
f89bd6f8 | 105 | * @nulls_base: Base value to generate nulls marker |
ccd57b1b | 106 | * @insecure_elasticity: Set to true to disable chain length checks |
97defe1e | 107 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) |
31ccde2d | 108 | * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash) |
7e1e7763 | 109 | * @obj_hashfn: Function to hash object |
02fd97c3 | 110 | * @obj_cmpfn: Function to compare key with object |
7e1e7763 TG |
111 | */ |
112 | struct rhashtable_params { | |
113 | size_t nelem_hint; | |
114 | size_t key_len; | |
115 | size_t key_offset; | |
116 | size_t head_offset; | |
c2e213cf HX |
117 | unsigned int max_size; |
118 | unsigned int min_size; | |
f89bd6f8 | 119 | u32 nulls_base; |
ccd57b1b | 120 | bool insecure_elasticity; |
97defe1e | 121 | size_t locks_mul; |
7e1e7763 TG |
122 | rht_hashfn_t hashfn; |
123 | rht_obj_hashfn_t obj_hashfn; | |
02fd97c3 | 124 | rht_obj_cmpfn_t obj_cmpfn; |
7e1e7763 TG |
125 | }; |
126 | ||
127 | /** | |
128 | * struct rhashtable - Hash table handle | |
129 | * @tbl: Bucket table | |
130 | * @nelems: Number of elements in table | |
31ccde2d | 131 | * @key_len: Key length for hashfn |
ccd57b1b | 132 | * @elasticity: Maximum chain length before rehash |
7e1e7763 | 133 | * @p: Configuration parameters |
97defe1e TG |
134 | * @run_work: Deferred worker to expand/shrink asynchronously |
135 | * @mutex: Mutex to protect current/future table swapping | |
136 | * @being_destroyed: True if table is set up for destruction | |
7e1e7763 TG |
137 | */ |
138 | struct rhashtable { | |
139 | struct bucket_table __rcu *tbl; | |
97defe1e | 140 | atomic_t nelems; |
a5b6846f | 141 | bool being_destroyed; |
31ccde2d | 142 | unsigned int key_len; |
ccd57b1b | 143 | unsigned int elasticity; |
7e1e7763 | 144 | struct rhashtable_params p; |
57699a40 | 145 | struct work_struct run_work; |
97defe1e | 146 | struct mutex mutex; |
7e1e7763 TG |
147 | }; |
148 | ||
f2dba9c6 HX |
149 | /** |
150 | * struct rhashtable_walker - Hash table walker | |
151 | * @list: List entry on list of walkers | |
eddee5ba | 152 | * @tbl: The table that we were walking over |
f2dba9c6 HX |
153 | */ |
154 | struct rhashtable_walker { | |
155 | struct list_head list; | |
eddee5ba | 156 | struct bucket_table *tbl; |
f2dba9c6 HX |
157 | }; |
158 | ||
159 | /** | |
160 | * struct rhashtable_iter - Hash table iterator, fits into netlink cb | |
161 | * @ht: Table to iterate through | |
162 | * @p: Current pointer | |
163 | * @walker: Associated rhashtable walker | |
164 | * @slot: Current slot | |
165 | * @skip: Number of entries to skip in slot | |
166 | */ | |
167 | struct rhashtable_iter { | |
168 | struct rhashtable *ht; | |
169 | struct rhash_head *p; | |
170 | struct rhashtable_walker *walker; | |
171 | unsigned int slot; | |
172 | unsigned int skip; | |
173 | }; | |
174 | ||
f89bd6f8 TG |
175 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) |
176 | { | |
177 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
178 | } | |
179 | ||
180 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
181 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
182 | ||
183 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
184 | { | |
185 | return ((unsigned long) ptr & 1); | |
186 | } | |
187 | ||
188 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
189 | { | |
190 | return ((unsigned long) ptr) >> 1; | |
191 | } | |
192 | ||
02fd97c3 HX |
193 | static inline void *rht_obj(const struct rhashtable *ht, |
194 | const struct rhash_head *he) | |
195 | { | |
196 | return (char *)he - ht->p.head_offset; | |
197 | } | |
198 | ||
199 | static inline unsigned int rht_bucket_index(const struct bucket_table *tbl, | |
200 | unsigned int hash) | |
201 | { | |
202 | return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1); | |
203 | } | |
204 | ||
205 | static inline unsigned int rht_key_hashfn( | |
206 | struct rhashtable *ht, const struct bucket_table *tbl, | |
207 | const void *key, const struct rhashtable_params params) | |
208 | { | |
31ccde2d | 209 | unsigned hash; |
de91b25c | 210 | |
31ccde2d HX |
211 | /* params must be equal to ht->p if it isn't constant. */ |
212 | if (!__builtin_constant_p(params.key_len)) | |
213 | hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd); | |
214 | else if (params.key_len) { | |
215 | unsigned key_len = params.key_len; | |
216 | ||
217 | if (params.hashfn) | |
218 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
219 | else if (key_len & (sizeof(u32) - 1)) | |
220 | hash = jhash(key, key_len, tbl->hash_rnd); | |
221 | else | |
222 | hash = jhash2(key, key_len / sizeof(u32), | |
223 | tbl->hash_rnd); | |
224 | } else { | |
225 | unsigned key_len = ht->p.key_len; | |
226 | ||
227 | if (params.hashfn) | |
228 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
229 | else | |
230 | hash = jhash(key, key_len, tbl->hash_rnd); | |
231 | } | |
232 | ||
233 | return rht_bucket_index(tbl, hash); | |
02fd97c3 HX |
234 | } |
235 | ||
236 | static inline unsigned int rht_head_hashfn( | |
237 | struct rhashtable *ht, const struct bucket_table *tbl, | |
238 | const struct rhash_head *he, const struct rhashtable_params params) | |
239 | { | |
240 | const char *ptr = rht_obj(ht, he); | |
241 | ||
242 | return likely(params.obj_hashfn) ? | |
243 | rht_bucket_index(tbl, params.obj_hashfn(ptr, tbl->hash_rnd)) : | |
244 | rht_key_hashfn(ht, tbl, ptr + params.key_offset, params); | |
245 | } | |
246 | ||
247 | /** | |
248 | * rht_grow_above_75 - returns true if nelems > 0.75 * table-size | |
249 | * @ht: hash table | |
250 | * @tbl: current table | |
251 | */ | |
252 | static inline bool rht_grow_above_75(const struct rhashtable *ht, | |
253 | const struct bucket_table *tbl) | |
254 | { | |
255 | /* Expand table when exceeding 75% load */ | |
256 | return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) && | |
257 | (!ht->p.max_size || tbl->size < ht->p.max_size); | |
258 | } | |
259 | ||
260 | /** | |
261 | * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size | |
262 | * @ht: hash table | |
263 | * @tbl: current table | |
264 | */ | |
265 | static inline bool rht_shrink_below_30(const struct rhashtable *ht, | |
266 | const struct bucket_table *tbl) | |
267 | { | |
268 | /* Shrink table beneath 30% load */ | |
269 | return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) && | |
270 | tbl->size > ht->p.min_size; | |
271 | } | |
272 | ||
ccd57b1b HX |
273 | /** |
274 | * rht_grow_above_100 - returns true if nelems > table-size | |
275 | * @ht: hash table | |
276 | * @tbl: current table | |
277 | */ | |
278 | static inline bool rht_grow_above_100(const struct rhashtable *ht, | |
279 | const struct bucket_table *tbl) | |
280 | { | |
281 | return atomic_read(&ht->nelems) > tbl->size; | |
282 | } | |
283 | ||
02fd97c3 HX |
284 | /* The bucket lock is selected based on the hash and protects mutations |
285 | * on a group of hash buckets. | |
286 | * | |
287 | * A maximum of tbl->size/2 bucket locks is allocated. This ensures that | |
288 | * a single lock always covers both buckets which may both contains | |
289 | * entries which link to the same bucket of the old table during resizing. | |
290 | * This allows to simplify the locking as locking the bucket in both | |
291 | * tables during resize always guarantee protection. | |
292 | * | |
293 | * IMPORTANT: When holding the bucket lock of both the old and new table | |
294 | * during expansions and shrinking, the old bucket lock must always be | |
295 | * acquired first. | |
296 | */ | |
297 | static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl, | |
298 | unsigned int hash) | |
299 | { | |
300 | return &tbl->locks[hash & tbl->locks_mask]; | |
301 | } | |
302 | ||
7e1e7763 | 303 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 304 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 305 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 306 | #else |
97defe1e | 307 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
308 | { |
309 | return 1; | |
310 | } | |
88d6ed15 TG |
311 | |
312 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
313 | u32 hash) | |
314 | { | |
315 | return 1; | |
316 | } | |
7e1e7763 TG |
317 | #endif /* CONFIG_PROVE_LOCKING */ |
318 | ||
488fb86e HX |
319 | int rhashtable_init(struct rhashtable *ht, |
320 | const struct rhashtable_params *params); | |
7e1e7763 | 321 | |
02fd97c3 HX |
322 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, |
323 | struct rhash_head *obj, | |
324 | struct bucket_table *old_tbl); | |
ccd57b1b | 325 | int rhashtable_insert_rehash(struct rhashtable *ht); |
7e1e7763 | 326 | |
f2dba9c6 HX |
327 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter); |
328 | void rhashtable_walk_exit(struct rhashtable_iter *iter); | |
329 | int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); | |
330 | void *rhashtable_walk_next(struct rhashtable_iter *iter); | |
331 | void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); | |
332 | ||
97defe1e | 333 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
334 | |
335 | #define rht_dereference(p, ht) \ | |
336 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
337 | ||
338 | #define rht_dereference_rcu(p, ht) \ | |
339 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
340 | ||
88d6ed15 TG |
341 | #define rht_dereference_bucket(p, tbl, hash) \ |
342 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 343 | |
88d6ed15 TG |
344 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
345 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
346 | ||
347 | #define rht_entry(tpos, pos, member) \ | |
348 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
349 | |
350 | /** | |
88d6ed15 TG |
351 | * rht_for_each_continue - continue iterating over hash chain |
352 | * @pos: the &struct rhash_head to use as a loop cursor. | |
353 | * @head: the previous &struct rhash_head to continue from | |
354 | * @tbl: the &struct bucket_table | |
355 | * @hash: the hash value / bucket index | |
7e1e7763 | 356 | */ |
88d6ed15 TG |
357 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
358 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 359 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
360 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
361 | ||
362 | /** | |
363 | * rht_for_each - iterate over hash chain | |
364 | * @pos: the &struct rhash_head to use as a loop cursor. | |
365 | * @tbl: the &struct bucket_table | |
366 | * @hash: the hash value / bucket index | |
367 | */ | |
368 | #define rht_for_each(pos, tbl, hash) \ | |
369 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
370 | ||
371 | /** | |
372 | * rht_for_each_entry_continue - continue iterating over hash chain | |
373 | * @tpos: the type * to use as a loop cursor. | |
374 | * @pos: the &struct rhash_head to use as a loop cursor. | |
375 | * @head: the previous &struct rhash_head to continue from | |
376 | * @tbl: the &struct bucket_table | |
377 | * @hash: the hash value / bucket index | |
378 | * @member: name of the &struct rhash_head within the hashable struct. | |
379 | */ | |
380 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
381 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 382 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 383 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
384 | |
385 | /** | |
386 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
387 | * @tpos: the type * to use as a loop cursor. |
388 | * @pos: the &struct rhash_head to use as a loop cursor. | |
389 | * @tbl: the &struct bucket_table | |
390 | * @hash: the hash value / bucket index | |
391 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 392 | */ |
88d6ed15 TG |
393 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
394 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
395 | tbl, hash, member) | |
7e1e7763 TG |
396 | |
397 | /** | |
398 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
399 | * @tpos: the type * to use as a loop cursor. |
400 | * @pos: the &struct rhash_head to use as a loop cursor. | |
401 | * @next: the &struct rhash_head to use as next in loop cursor. | |
402 | * @tbl: the &struct bucket_table | |
403 | * @hash: the hash value / bucket index | |
404 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
405 | * |
406 | * This hash chain list-traversal primitive allows for the looped code to | |
407 | * remove the loop cursor from the list. | |
408 | */ | |
88d6ed15 TG |
409 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
410 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
411 | next = !rht_is_a_nulls(pos) ? \ |
412 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
413 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
607954b0 PM |
414 | pos = next, \ |
415 | next = !rht_is_a_nulls(pos) ? \ | |
416 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
88d6ed15 TG |
417 | |
418 | /** | |
419 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
420 | * @pos: the &struct rhash_head to use as a loop cursor. | |
421 | * @head: the previous &struct rhash_head to continue from | |
422 | * @tbl: the &struct bucket_table | |
423 | * @hash: the hash value / bucket index | |
424 | * | |
425 | * This hash chain list-traversal primitive may safely run concurrently with | |
426 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
427 | * traversal is guarded by rcu_read_lock(). | |
428 | */ | |
429 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
430 | for (({barrier(); }), \ | |
431 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 432 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 433 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
434 | |
435 | /** | |
436 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
437 | * @pos: the &struct rhash_head to use as a loop cursor. |
438 | * @tbl: the &struct bucket_table | |
439 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
440 | * |
441 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 442 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
443 | * traversal is guarded by rcu_read_lock(). |
444 | */ | |
88d6ed15 TG |
445 | #define rht_for_each_rcu(pos, tbl, hash) \ |
446 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
447 | ||
448 | /** | |
449 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
450 | * @tpos: the type * to use as a loop cursor. | |
451 | * @pos: the &struct rhash_head to use as a loop cursor. | |
452 | * @head: the previous &struct rhash_head to continue from | |
453 | * @tbl: the &struct bucket_table | |
454 | * @hash: the hash value / bucket index | |
455 | * @member: name of the &struct rhash_head within the hashable struct. | |
456 | * | |
457 | * This hash chain list-traversal primitive may safely run concurrently with | |
458 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
459 | * traversal is guarded by rcu_read_lock(). | |
460 | */ | |
461 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
462 | for (({barrier(); }), \ | |
463 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 464 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 465 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
466 | |
467 | /** | |
468 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
469 | * @tpos: the type * to use as a loop cursor. |
470 | * @pos: the &struct rhash_head to use as a loop cursor. | |
471 | * @tbl: the &struct bucket_table | |
472 | * @hash: the hash value / bucket index | |
473 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
474 | * |
475 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 476 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
477 | * traversal is guarded by rcu_read_lock(). |
478 | */ | |
88d6ed15 TG |
479 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
480 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
481 | tbl, hash, member) | |
7e1e7763 | 482 | |
02fd97c3 HX |
483 | static inline int rhashtable_compare(struct rhashtable_compare_arg *arg, |
484 | const void *obj) | |
485 | { | |
486 | struct rhashtable *ht = arg->ht; | |
487 | const char *ptr = obj; | |
488 | ||
489 | return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len); | |
490 | } | |
491 | ||
492 | /** | |
493 | * rhashtable_lookup_fast - search hash table, inlined version | |
494 | * @ht: hash table | |
495 | * @key: the pointer to the key | |
496 | * @params: hash table parameters | |
497 | * | |
498 | * Computes the hash value for the key and traverses the bucket chain looking | |
499 | * for a entry with an identical key. The first matching entry is returned. | |
500 | * | |
501 | * Returns the first entry on which the compare function returned true. | |
502 | */ | |
503 | static inline void *rhashtable_lookup_fast( | |
504 | struct rhashtable *ht, const void *key, | |
505 | const struct rhashtable_params params) | |
506 | { | |
507 | struct rhashtable_compare_arg arg = { | |
508 | .ht = ht, | |
509 | .key = key, | |
510 | }; | |
511 | const struct bucket_table *tbl; | |
512 | struct rhash_head *he; | |
513 | unsigned hash; | |
514 | ||
515 | rcu_read_lock(); | |
516 | ||
517 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
518 | restart: | |
519 | hash = rht_key_hashfn(ht, tbl, key, params); | |
520 | rht_for_each_rcu(he, tbl, hash) { | |
521 | if (params.obj_cmpfn ? | |
522 | params.obj_cmpfn(&arg, rht_obj(ht, he)) : | |
523 | rhashtable_compare(&arg, rht_obj(ht, he))) | |
524 | continue; | |
525 | rcu_read_unlock(); | |
526 | return rht_obj(ht, he); | |
527 | } | |
528 | ||
529 | /* Ensure we see any new tables. */ | |
530 | smp_rmb(); | |
531 | ||
532 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
533 | if (unlikely(tbl)) | |
534 | goto restart; | |
535 | rcu_read_unlock(); | |
536 | ||
537 | return NULL; | |
538 | } | |
539 | ||
540 | static inline int __rhashtable_insert_fast( | |
541 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
542 | const struct rhashtable_params params) | |
543 | { | |
544 | struct rhashtable_compare_arg arg = { | |
545 | .ht = ht, | |
546 | .key = key, | |
547 | }; | |
02fd97c3 HX |
548 | struct bucket_table *tbl, *new_tbl; |
549 | struct rhash_head *head; | |
550 | spinlock_t *lock; | |
ccd57b1b | 551 | unsigned elasticity; |
02fd97c3 | 552 | unsigned hash; |
ccd57b1b | 553 | int err; |
02fd97c3 | 554 | |
ccd57b1b | 555 | restart: |
02fd97c3 HX |
556 | rcu_read_lock(); |
557 | ||
558 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
02fd97c3 | 559 | |
b824478b HX |
560 | /* All insertions must grab the oldest table containing |
561 | * the hashed bucket that is yet to be rehashed. | |
02fd97c3 | 562 | */ |
b824478b HX |
563 | for (;;) { |
564 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
565 | lock = rht_bucket_lock(tbl, hash); | |
566 | spin_lock_bh(lock); | |
567 | ||
568 | if (tbl->rehash <= hash) | |
569 | break; | |
570 | ||
571 | spin_unlock_bh(lock); | |
572 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
573 | } | |
574 | ||
02fd97c3 HX |
575 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
576 | if (unlikely(new_tbl)) { | |
577 | err = rhashtable_insert_slow(ht, key, obj, new_tbl); | |
ccd57b1b HX |
578 | if (err == -EAGAIN) |
579 | goto slow_path; | |
02fd97c3 HX |
580 | goto out; |
581 | } | |
582 | ||
ccd57b1b HX |
583 | if (unlikely(rht_grow_above_100(ht, tbl))) { |
584 | slow_path: | |
585 | spin_unlock_bh(lock); | |
586 | rcu_read_unlock(); | |
587 | err = rhashtable_insert_rehash(ht); | |
588 | if (err) | |
589 | return err; | |
590 | ||
591 | goto restart; | |
592 | } | |
02fd97c3 | 593 | |
ccd57b1b HX |
594 | err = -EEXIST; |
595 | elasticity = ht->elasticity; | |
02fd97c3 | 596 | rht_for_each(head, tbl, hash) { |
ccd57b1b HX |
597 | if (key && |
598 | unlikely(!(params.obj_cmpfn ? | |
02fd97c3 HX |
599 | params.obj_cmpfn(&arg, rht_obj(ht, head)) : |
600 | rhashtable_compare(&arg, rht_obj(ht, head))))) | |
601 | goto out; | |
ccd57b1b HX |
602 | if (!--elasticity) |
603 | goto slow_path; | |
02fd97c3 HX |
604 | } |
605 | ||
02fd97c3 HX |
606 | err = 0; |
607 | ||
608 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
609 | ||
610 | RCU_INIT_POINTER(obj->next, head); | |
611 | ||
612 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
613 | ||
614 | atomic_inc(&ht->nelems); | |
615 | if (rht_grow_above_75(ht, tbl)) | |
616 | schedule_work(&ht->run_work); | |
617 | ||
618 | out: | |
619 | spin_unlock_bh(lock); | |
620 | rcu_read_unlock(); | |
621 | ||
622 | return err; | |
623 | } | |
624 | ||
625 | /** | |
626 | * rhashtable_insert_fast - insert object into hash table | |
627 | * @ht: hash table | |
628 | * @obj: pointer to hash head inside object | |
629 | * @params: hash table parameters | |
630 | * | |
631 | * Will take a per bucket spinlock to protect against mutual mutations | |
632 | * on the same bucket. Multiple insertions may occur in parallel unless | |
633 | * they map to the same bucket lock. | |
634 | * | |
635 | * It is safe to call this function from atomic context. | |
636 | * | |
637 | * Will trigger an automatic deferred table resizing if the size grows | |
638 | * beyond the watermark indicated by grow_decision() which can be passed | |
639 | * to rhashtable_init(). | |
640 | */ | |
641 | static inline int rhashtable_insert_fast( | |
642 | struct rhashtable *ht, struct rhash_head *obj, | |
643 | const struct rhashtable_params params) | |
644 | { | |
645 | return __rhashtable_insert_fast(ht, NULL, obj, params); | |
646 | } | |
647 | ||
648 | /** | |
649 | * rhashtable_lookup_insert_fast - lookup and insert object into hash table | |
650 | * @ht: hash table | |
651 | * @obj: pointer to hash head inside object | |
652 | * @params: hash table parameters | |
653 | * | |
654 | * Locks down the bucket chain in both the old and new table if a resize | |
655 | * is in progress to ensure that writers can't remove from the old table | |
656 | * and can't insert to the new table during the atomic operation of search | |
657 | * and insertion. Searches for duplicates in both the old and new table if | |
658 | * a resize is in progress. | |
659 | * | |
660 | * This lookup function may only be used for fixed key hash table (key_len | |
661 | * parameter set). It will BUG() if used inappropriately. | |
662 | * | |
663 | * It is safe to call this function from atomic context. | |
664 | * | |
665 | * Will trigger an automatic deferred table resizing if the size grows | |
666 | * beyond the watermark indicated by grow_decision() which can be passed | |
667 | * to rhashtable_init(). | |
668 | */ | |
669 | static inline int rhashtable_lookup_insert_fast( | |
670 | struct rhashtable *ht, struct rhash_head *obj, | |
671 | const struct rhashtable_params params) | |
672 | { | |
673 | const char *key = rht_obj(ht, obj); | |
674 | ||
675 | BUG_ON(ht->p.obj_hashfn); | |
676 | ||
677 | return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, | |
678 | params); | |
679 | } | |
680 | ||
681 | /** | |
682 | * rhashtable_lookup_insert_key - search and insert object to hash table | |
683 | * with explicit key | |
684 | * @ht: hash table | |
685 | * @key: key | |
686 | * @obj: pointer to hash head inside object | |
687 | * @params: hash table parameters | |
688 | * | |
689 | * Locks down the bucket chain in both the old and new table if a resize | |
690 | * is in progress to ensure that writers can't remove from the old table | |
691 | * and can't insert to the new table during the atomic operation of search | |
692 | * and insertion. Searches for duplicates in both the old and new table if | |
693 | * a resize is in progress. | |
694 | * | |
695 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
696 | * | |
697 | * Will trigger an automatic deferred table resizing if the size grows | |
698 | * beyond the watermark indicated by grow_decision() which can be passed | |
699 | * to rhashtable_init(). | |
700 | * | |
701 | * Returns zero on success. | |
702 | */ | |
703 | static inline int rhashtable_lookup_insert_key( | |
704 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
705 | const struct rhashtable_params params) | |
706 | { | |
707 | BUG_ON(!ht->p.obj_hashfn || !key); | |
708 | ||
709 | return __rhashtable_insert_fast(ht, key, obj, params); | |
710 | } | |
711 | ||
712 | static inline int __rhashtable_remove_fast( | |
713 | struct rhashtable *ht, struct bucket_table *tbl, | |
714 | struct rhash_head *obj, const struct rhashtable_params params) | |
715 | { | |
716 | struct rhash_head __rcu **pprev; | |
717 | struct rhash_head *he; | |
718 | spinlock_t * lock; | |
719 | unsigned hash; | |
720 | int err = -ENOENT; | |
721 | ||
722 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
723 | lock = rht_bucket_lock(tbl, hash); | |
724 | ||
725 | spin_lock_bh(lock); | |
726 | ||
727 | pprev = &tbl->buckets[hash]; | |
728 | rht_for_each(he, tbl, hash) { | |
729 | if (he != obj) { | |
730 | pprev = &he->next; | |
731 | continue; | |
732 | } | |
733 | ||
734 | rcu_assign_pointer(*pprev, obj->next); | |
735 | err = 0; | |
736 | break; | |
737 | } | |
738 | ||
739 | spin_unlock_bh(lock); | |
740 | ||
741 | return err; | |
742 | } | |
743 | ||
744 | /** | |
745 | * rhashtable_remove_fast - remove object from hash table | |
746 | * @ht: hash table | |
747 | * @obj: pointer to hash head inside object | |
748 | * @params: hash table parameters | |
749 | * | |
750 | * Since the hash chain is single linked, the removal operation needs to | |
751 | * walk the bucket chain upon removal. The removal operation is thus | |
752 | * considerable slow if the hash table is not correctly sized. | |
753 | * | |
754 | * Will automatically shrink the table via rhashtable_expand() if the | |
755 | * shrink_decision function specified at rhashtable_init() returns true. | |
756 | * | |
757 | * Returns zero on success, -ENOENT if the entry could not be found. | |
758 | */ | |
759 | static inline int rhashtable_remove_fast( | |
760 | struct rhashtable *ht, struct rhash_head *obj, | |
761 | const struct rhashtable_params params) | |
762 | { | |
763 | struct bucket_table *tbl; | |
764 | int err; | |
765 | ||
766 | rcu_read_lock(); | |
767 | ||
768 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
769 | ||
770 | /* Because we have already taken (and released) the bucket | |
771 | * lock in old_tbl, if we find that future_tbl is not yet | |
772 | * visible then that guarantees the entry to still be in | |
773 | * the old tbl if it exists. | |
774 | */ | |
775 | while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) && | |
776 | (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) | |
777 | ; | |
778 | ||
779 | if (err) | |
780 | goto out; | |
781 | ||
782 | atomic_dec(&ht->nelems); | |
783 | if (rht_shrink_below_30(ht, tbl)) | |
784 | schedule_work(&ht->run_work); | |
785 | ||
786 | out: | |
787 | rcu_read_unlock(); | |
788 | ||
789 | return err; | |
790 | } | |
791 | ||
7e1e7763 | 792 | #endif /* _LINUX_RHASHTABLE_H */ |