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> |
b5e2c150 | 5 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> |
7e1e7763 TG |
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 | 90 | typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); |
49f7b33e | 91 | typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, 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 |
b5e2c150 | 107 | * @automatic_shrinking: Enable automatic shrinking of tables |
97defe1e | 108 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) |
31ccde2d | 109 | * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash) |
7e1e7763 | 110 | * @obj_hashfn: Function to hash object |
02fd97c3 | 111 | * @obj_cmpfn: Function to compare key with object |
7e1e7763 TG |
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; | |
c2e213cf HX |
118 | unsigned int max_size; |
119 | unsigned int min_size; | |
f89bd6f8 | 120 | u32 nulls_base; |
ccd57b1b | 121 | bool insecure_elasticity; |
b5e2c150 | 122 | bool automatic_shrinking; |
97defe1e | 123 | size_t locks_mul; |
7e1e7763 TG |
124 | rht_hashfn_t hashfn; |
125 | rht_obj_hashfn_t obj_hashfn; | |
02fd97c3 | 126 | rht_obj_cmpfn_t obj_cmpfn; |
7e1e7763 TG |
127 | }; |
128 | ||
129 | /** | |
130 | * struct rhashtable - Hash table handle | |
131 | * @tbl: Bucket table | |
132 | * @nelems: Number of elements in table | |
31ccde2d | 133 | * @key_len: Key length for hashfn |
ccd57b1b | 134 | * @elasticity: Maximum chain length before rehash |
7e1e7763 | 135 | * @p: Configuration parameters |
97defe1e TG |
136 | * @run_work: Deferred worker to expand/shrink asynchronously |
137 | * @mutex: Mutex to protect current/future table swapping | |
ba7c95ea | 138 | * @lock: Spin lock to protect walker list |
7e1e7763 TG |
139 | */ |
140 | struct rhashtable { | |
141 | struct bucket_table __rcu *tbl; | |
97defe1e | 142 | atomic_t nelems; |
31ccde2d | 143 | unsigned int key_len; |
ccd57b1b | 144 | unsigned int elasticity; |
7e1e7763 | 145 | struct rhashtable_params p; |
57699a40 | 146 | struct work_struct run_work; |
97defe1e | 147 | struct mutex mutex; |
ba7c95ea | 148 | spinlock_t lock; |
7e1e7763 TG |
149 | }; |
150 | ||
f2dba9c6 HX |
151 | /** |
152 | * struct rhashtable_walker - Hash table walker | |
153 | * @list: List entry on list of walkers | |
eddee5ba | 154 | * @tbl: The table that we were walking over |
f2dba9c6 HX |
155 | */ |
156 | struct rhashtable_walker { | |
157 | struct list_head list; | |
eddee5ba | 158 | struct bucket_table *tbl; |
f2dba9c6 HX |
159 | }; |
160 | ||
161 | /** | |
162 | * struct rhashtable_iter - Hash table iterator, fits into netlink cb | |
163 | * @ht: Table to iterate through | |
164 | * @p: Current pointer | |
165 | * @walker: Associated rhashtable walker | |
166 | * @slot: Current slot | |
167 | * @skip: Number of entries to skip in slot | |
168 | */ | |
169 | struct rhashtable_iter { | |
170 | struct rhashtable *ht; | |
171 | struct rhash_head *p; | |
172 | struct rhashtable_walker *walker; | |
173 | unsigned int slot; | |
174 | unsigned int skip; | |
175 | }; | |
176 | ||
f89bd6f8 TG |
177 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) |
178 | { | |
179 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
180 | } | |
181 | ||
182 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
183 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
184 | ||
185 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
186 | { | |
187 | return ((unsigned long) ptr & 1); | |
188 | } | |
189 | ||
190 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
191 | { | |
192 | return ((unsigned long) ptr) >> 1; | |
193 | } | |
194 | ||
02fd97c3 HX |
195 | static inline void *rht_obj(const struct rhashtable *ht, |
196 | const struct rhash_head *he) | |
197 | { | |
198 | return (char *)he - ht->p.head_offset; | |
199 | } | |
200 | ||
201 | static inline unsigned int rht_bucket_index(const struct bucket_table *tbl, | |
202 | unsigned int hash) | |
203 | { | |
204 | return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1); | |
205 | } | |
206 | ||
207 | static inline unsigned int rht_key_hashfn( | |
208 | struct rhashtable *ht, const struct bucket_table *tbl, | |
209 | const void *key, const struct rhashtable_params params) | |
210 | { | |
299e5c32 | 211 | unsigned int hash; |
de91b25c | 212 | |
31ccde2d HX |
213 | /* params must be equal to ht->p if it isn't constant. */ |
214 | if (!__builtin_constant_p(params.key_len)) | |
215 | hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd); | |
216 | else if (params.key_len) { | |
299e5c32 | 217 | unsigned int key_len = params.key_len; |
31ccde2d HX |
218 | |
219 | if (params.hashfn) | |
220 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
221 | else if (key_len & (sizeof(u32) - 1)) | |
222 | hash = jhash(key, key_len, tbl->hash_rnd); | |
223 | else | |
224 | hash = jhash2(key, key_len / sizeof(u32), | |
225 | tbl->hash_rnd); | |
226 | } else { | |
299e5c32 | 227 | unsigned int key_len = ht->p.key_len; |
31ccde2d HX |
228 | |
229 | if (params.hashfn) | |
230 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
231 | else | |
232 | hash = jhash(key, key_len, tbl->hash_rnd); | |
233 | } | |
234 | ||
235 | return rht_bucket_index(tbl, hash); | |
02fd97c3 HX |
236 | } |
237 | ||
238 | static inline unsigned int rht_head_hashfn( | |
239 | struct rhashtable *ht, const struct bucket_table *tbl, | |
240 | const struct rhash_head *he, const struct rhashtable_params params) | |
241 | { | |
242 | const char *ptr = rht_obj(ht, he); | |
243 | ||
244 | return likely(params.obj_hashfn) ? | |
49f7b33e PM |
245 | rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?: |
246 | ht->p.key_len, | |
247 | tbl->hash_rnd)) : | |
02fd97c3 HX |
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 | ||
ccd57b1b HX |
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 | ||
02fd97c3 HX |
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 | ||
7e1e7763 | 307 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 308 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 309 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 310 | #else |
97defe1e | 311 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
312 | { |
313 | return 1; | |
314 | } | |
88d6ed15 TG |
315 | |
316 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
317 | u32 hash) | |
318 | { | |
319 | return 1; | |
320 | } | |
7e1e7763 TG |
321 | #endif /* CONFIG_PROVE_LOCKING */ |
322 | ||
488fb86e HX |
323 | int rhashtable_init(struct rhashtable *ht, |
324 | const struct rhashtable_params *params); | |
7e1e7763 | 325 | |
02fd97c3 HX |
326 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, |
327 | struct rhash_head *obj, | |
328 | struct bucket_table *old_tbl); | |
ccd57b1b | 329 | int rhashtable_insert_rehash(struct rhashtable *ht); |
7e1e7763 | 330 | |
f2dba9c6 HX |
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 | ||
6b6f302c TG |
337 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
338 | void (*free_fn)(void *ptr, void *arg), | |
339 | void *arg); | |
97defe1e | 340 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
341 | |
342 | #define rht_dereference(p, ht) \ | |
343 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
344 | ||
345 | #define rht_dereference_rcu(p, ht) \ | |
346 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
347 | ||
88d6ed15 TG |
348 | #define rht_dereference_bucket(p, tbl, hash) \ |
349 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 350 | |
88d6ed15 TG |
351 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
352 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
353 | ||
354 | #define rht_entry(tpos, pos, member) \ | |
355 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
356 | |
357 | /** | |
88d6ed15 TG |
358 | * rht_for_each_continue - continue iterating over hash chain |
359 | * @pos: the &struct rhash_head to use as a loop cursor. | |
360 | * @head: the previous &struct rhash_head to continue from | |
361 | * @tbl: the &struct bucket_table | |
362 | * @hash: the hash value / bucket index | |
7e1e7763 | 363 | */ |
88d6ed15 TG |
364 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
365 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 366 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
367 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
368 | ||
369 | /** | |
370 | * rht_for_each - iterate over hash chain | |
371 | * @pos: the &struct rhash_head to use as a loop cursor. | |
372 | * @tbl: the &struct bucket_table | |
373 | * @hash: the hash value / bucket index | |
374 | */ | |
375 | #define rht_for_each(pos, tbl, hash) \ | |
376 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
377 | ||
378 | /** | |
379 | * rht_for_each_entry_continue - continue iterating over hash chain | |
380 | * @tpos: the type * to use as a loop cursor. | |
381 | * @pos: the &struct rhash_head to use as a loop cursor. | |
382 | * @head: the previous &struct rhash_head to continue from | |
383 | * @tbl: the &struct bucket_table | |
384 | * @hash: the hash value / bucket index | |
385 | * @member: name of the &struct rhash_head within the hashable struct. | |
386 | */ | |
387 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
388 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 389 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 390 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
391 | |
392 | /** | |
393 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
394 | * @tpos: the type * to use as a loop cursor. |
395 | * @pos: the &struct rhash_head to use as a loop cursor. | |
396 | * @tbl: the &struct bucket_table | |
397 | * @hash: the hash value / bucket index | |
398 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 399 | */ |
88d6ed15 TG |
400 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
401 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
402 | tbl, hash, member) | |
7e1e7763 TG |
403 | |
404 | /** | |
405 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
406 | * @tpos: the type * to use as a loop cursor. |
407 | * @pos: the &struct rhash_head to use as a loop cursor. | |
408 | * @next: the &struct rhash_head to use as next in loop cursor. | |
409 | * @tbl: the &struct bucket_table | |
410 | * @hash: the hash value / bucket index | |
411 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
412 | * |
413 | * This hash chain list-traversal primitive allows for the looped code to | |
414 | * remove the loop cursor from the list. | |
415 | */ | |
88d6ed15 TG |
416 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
417 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
418 | next = !rht_is_a_nulls(pos) ? \ |
419 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
420 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
607954b0 PM |
421 | pos = next, \ |
422 | next = !rht_is_a_nulls(pos) ? \ | |
423 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
88d6ed15 TG |
424 | |
425 | /** | |
426 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
427 | * @pos: the &struct rhash_head to use as a loop cursor. | |
428 | * @head: the previous &struct rhash_head to continue from | |
429 | * @tbl: the &struct bucket_table | |
430 | * @hash: the hash value / bucket index | |
431 | * | |
432 | * This hash chain list-traversal primitive may safely run concurrently with | |
433 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
434 | * traversal is guarded by rcu_read_lock(). | |
435 | */ | |
436 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
437 | for (({barrier(); }), \ | |
438 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 439 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 440 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
441 | |
442 | /** | |
443 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
444 | * @pos: the &struct rhash_head to use as a loop cursor. |
445 | * @tbl: the &struct bucket_table | |
446 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
447 | * |
448 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 449 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
450 | * traversal is guarded by rcu_read_lock(). |
451 | */ | |
88d6ed15 TG |
452 | #define rht_for_each_rcu(pos, tbl, hash) \ |
453 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
454 | ||
455 | /** | |
456 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
457 | * @tpos: the type * to use as a loop cursor. | |
458 | * @pos: the &struct rhash_head to use as a loop cursor. | |
459 | * @head: the previous &struct rhash_head to continue from | |
460 | * @tbl: the &struct bucket_table | |
461 | * @hash: the hash value / bucket index | |
462 | * @member: name of the &struct rhash_head within the hashable struct. | |
463 | * | |
464 | * This hash chain list-traversal primitive may safely run concurrently with | |
465 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
466 | * traversal is guarded by rcu_read_lock(). | |
467 | */ | |
468 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
469 | for (({barrier(); }), \ | |
470 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 471 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 472 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
473 | |
474 | /** | |
475 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
476 | * @tpos: the type * to use as a loop cursor. |
477 | * @pos: the &struct rhash_head to use as a loop cursor. | |
478 | * @tbl: the &struct bucket_table | |
479 | * @hash: the hash value / bucket index | |
480 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
481 | * |
482 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 483 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
484 | * traversal is guarded by rcu_read_lock(). |
485 | */ | |
88d6ed15 TG |
486 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
487 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
488 | tbl, hash, member) | |
7e1e7763 | 489 | |
02fd97c3 HX |
490 | static inline int rhashtable_compare(struct rhashtable_compare_arg *arg, |
491 | const void *obj) | |
492 | { | |
493 | struct rhashtable *ht = arg->ht; | |
494 | const char *ptr = obj; | |
495 | ||
496 | return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len); | |
497 | } | |
498 | ||
499 | /** | |
500 | * rhashtable_lookup_fast - search hash table, inlined version | |
501 | * @ht: hash table | |
502 | * @key: the pointer to the key | |
503 | * @params: hash table parameters | |
504 | * | |
505 | * Computes the hash value for the key and traverses the bucket chain looking | |
506 | * for a entry with an identical key. The first matching entry is returned. | |
507 | * | |
508 | * Returns the first entry on which the compare function returned true. | |
509 | */ | |
510 | static inline void *rhashtable_lookup_fast( | |
511 | struct rhashtable *ht, const void *key, | |
512 | const struct rhashtable_params params) | |
513 | { | |
514 | struct rhashtable_compare_arg arg = { | |
515 | .ht = ht, | |
516 | .key = key, | |
517 | }; | |
518 | const struct bucket_table *tbl; | |
519 | struct rhash_head *he; | |
299e5c32 | 520 | unsigned int hash; |
02fd97c3 HX |
521 | |
522 | rcu_read_lock(); | |
523 | ||
524 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
525 | restart: | |
526 | hash = rht_key_hashfn(ht, tbl, key, params); | |
527 | rht_for_each_rcu(he, tbl, hash) { | |
528 | if (params.obj_cmpfn ? | |
529 | params.obj_cmpfn(&arg, rht_obj(ht, he)) : | |
530 | rhashtable_compare(&arg, rht_obj(ht, he))) | |
531 | continue; | |
532 | rcu_read_unlock(); | |
533 | return rht_obj(ht, he); | |
534 | } | |
535 | ||
536 | /* Ensure we see any new tables. */ | |
537 | smp_rmb(); | |
538 | ||
539 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
540 | if (unlikely(tbl)) | |
541 | goto restart; | |
542 | rcu_read_unlock(); | |
543 | ||
544 | return NULL; | |
545 | } | |
546 | ||
ac833bdd | 547 | /* Internal function, please use rhashtable_insert_fast() instead */ |
02fd97c3 HX |
548 | static inline int __rhashtable_insert_fast( |
549 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
550 | const struct rhashtable_params params) | |
551 | { | |
552 | struct rhashtable_compare_arg arg = { | |
553 | .ht = ht, | |
554 | .key = key, | |
555 | }; | |
02fd97c3 HX |
556 | struct bucket_table *tbl, *new_tbl; |
557 | struct rhash_head *head; | |
558 | spinlock_t *lock; | |
299e5c32 TG |
559 | unsigned int elasticity; |
560 | unsigned int hash; | |
ccd57b1b | 561 | int err; |
02fd97c3 | 562 | |
ccd57b1b | 563 | restart: |
02fd97c3 HX |
564 | rcu_read_lock(); |
565 | ||
566 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
02fd97c3 | 567 | |
b824478b HX |
568 | /* All insertions must grab the oldest table containing |
569 | * the hashed bucket that is yet to be rehashed. | |
02fd97c3 | 570 | */ |
b824478b HX |
571 | for (;;) { |
572 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
573 | lock = rht_bucket_lock(tbl, hash); | |
574 | spin_lock_bh(lock); | |
575 | ||
576 | if (tbl->rehash <= hash) | |
577 | break; | |
578 | ||
579 | spin_unlock_bh(lock); | |
580 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
581 | } | |
582 | ||
02fd97c3 HX |
583 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
584 | if (unlikely(new_tbl)) { | |
585 | err = rhashtable_insert_slow(ht, key, obj, new_tbl); | |
ccd57b1b HX |
586 | if (err == -EAGAIN) |
587 | goto slow_path; | |
02fd97c3 HX |
588 | goto out; |
589 | } | |
590 | ||
ccd57b1b HX |
591 | if (unlikely(rht_grow_above_100(ht, tbl))) { |
592 | slow_path: | |
593 | spin_unlock_bh(lock); | |
ccd57b1b | 594 | err = rhashtable_insert_rehash(ht); |
58be8a58 | 595 | rcu_read_unlock(); |
ccd57b1b HX |
596 | if (err) |
597 | return err; | |
598 | ||
599 | goto restart; | |
600 | } | |
02fd97c3 | 601 | |
ccd57b1b HX |
602 | err = -EEXIST; |
603 | elasticity = ht->elasticity; | |
02fd97c3 | 604 | rht_for_each(head, tbl, hash) { |
ccd57b1b HX |
605 | if (key && |
606 | unlikely(!(params.obj_cmpfn ? | |
02fd97c3 HX |
607 | params.obj_cmpfn(&arg, rht_obj(ht, head)) : |
608 | rhashtable_compare(&arg, rht_obj(ht, head))))) | |
609 | goto out; | |
ccd57b1b HX |
610 | if (!--elasticity) |
611 | goto slow_path; | |
02fd97c3 HX |
612 | } |
613 | ||
02fd97c3 HX |
614 | err = 0; |
615 | ||
616 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
617 | ||
618 | RCU_INIT_POINTER(obj->next, head); | |
619 | ||
620 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
621 | ||
622 | atomic_inc(&ht->nelems); | |
623 | if (rht_grow_above_75(ht, tbl)) | |
624 | schedule_work(&ht->run_work); | |
625 | ||
626 | out: | |
627 | spin_unlock_bh(lock); | |
628 | rcu_read_unlock(); | |
629 | ||
630 | return err; | |
631 | } | |
632 | ||
633 | /** | |
634 | * rhashtable_insert_fast - insert object into hash table | |
635 | * @ht: hash table | |
636 | * @obj: pointer to hash head inside object | |
637 | * @params: hash table parameters | |
638 | * | |
639 | * Will take a per bucket spinlock to protect against mutual mutations | |
640 | * on the same bucket. Multiple insertions may occur in parallel unless | |
641 | * they map to the same bucket lock. | |
642 | * | |
643 | * It is safe to call this function from atomic context. | |
644 | * | |
645 | * Will trigger an automatic deferred table resizing if the size grows | |
646 | * beyond the watermark indicated by grow_decision() which can be passed | |
647 | * to rhashtable_init(). | |
648 | */ | |
649 | static inline int rhashtable_insert_fast( | |
650 | struct rhashtable *ht, struct rhash_head *obj, | |
651 | const struct rhashtable_params params) | |
652 | { | |
653 | return __rhashtable_insert_fast(ht, NULL, obj, params); | |
654 | } | |
655 | ||
656 | /** | |
657 | * rhashtable_lookup_insert_fast - lookup and insert object into hash table | |
658 | * @ht: hash table | |
659 | * @obj: pointer to hash head inside object | |
660 | * @params: hash table parameters | |
661 | * | |
662 | * Locks down the bucket chain in both the old and new table if a resize | |
663 | * is in progress to ensure that writers can't remove from the old table | |
664 | * and can't insert to the new table during the atomic operation of search | |
665 | * and insertion. Searches for duplicates in both the old and new table if | |
666 | * a resize is in progress. | |
667 | * | |
668 | * This lookup function may only be used for fixed key hash table (key_len | |
669 | * parameter set). It will BUG() if used inappropriately. | |
670 | * | |
671 | * It is safe to call this function from atomic context. | |
672 | * | |
673 | * Will trigger an automatic deferred table resizing if the size grows | |
674 | * beyond the watermark indicated by grow_decision() which can be passed | |
675 | * to rhashtable_init(). | |
676 | */ | |
677 | static inline int rhashtable_lookup_insert_fast( | |
678 | struct rhashtable *ht, struct rhash_head *obj, | |
679 | const struct rhashtable_params params) | |
680 | { | |
681 | const char *key = rht_obj(ht, obj); | |
682 | ||
683 | BUG_ON(ht->p.obj_hashfn); | |
684 | ||
685 | return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, | |
686 | params); | |
687 | } | |
688 | ||
689 | /** | |
690 | * rhashtable_lookup_insert_key - search and insert object to hash table | |
691 | * with explicit key | |
692 | * @ht: hash table | |
693 | * @key: key | |
694 | * @obj: pointer to hash head inside object | |
695 | * @params: hash table parameters | |
696 | * | |
697 | * Locks down the bucket chain in both the old and new table if a resize | |
698 | * is in progress to ensure that writers can't remove from the old table | |
699 | * and can't insert to the new table during the atomic operation of search | |
700 | * and insertion. Searches for duplicates in both the old and new table if | |
701 | * a resize is in progress. | |
702 | * | |
703 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
704 | * | |
705 | * Will trigger an automatic deferred table resizing if the size grows | |
706 | * beyond the watermark indicated by grow_decision() which can be passed | |
707 | * to rhashtable_init(). | |
708 | * | |
709 | * Returns zero on success. | |
710 | */ | |
711 | static inline int rhashtable_lookup_insert_key( | |
712 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
713 | const struct rhashtable_params params) | |
714 | { | |
715 | BUG_ON(!ht->p.obj_hashfn || !key); | |
716 | ||
717 | return __rhashtable_insert_fast(ht, key, obj, params); | |
718 | } | |
719 | ||
ac833bdd | 720 | /* Internal function, please use rhashtable_remove_fast() instead */ |
02fd97c3 HX |
721 | static inline int __rhashtable_remove_fast( |
722 | struct rhashtable *ht, struct bucket_table *tbl, | |
723 | struct rhash_head *obj, const struct rhashtable_params params) | |
724 | { | |
725 | struct rhash_head __rcu **pprev; | |
726 | struct rhash_head *he; | |
727 | spinlock_t * lock; | |
299e5c32 | 728 | unsigned int hash; |
02fd97c3 HX |
729 | int err = -ENOENT; |
730 | ||
731 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
732 | lock = rht_bucket_lock(tbl, hash); | |
733 | ||
734 | spin_lock_bh(lock); | |
735 | ||
736 | pprev = &tbl->buckets[hash]; | |
737 | rht_for_each(he, tbl, hash) { | |
738 | if (he != obj) { | |
739 | pprev = &he->next; | |
740 | continue; | |
741 | } | |
742 | ||
743 | rcu_assign_pointer(*pprev, obj->next); | |
744 | err = 0; | |
745 | break; | |
746 | } | |
747 | ||
748 | spin_unlock_bh(lock); | |
749 | ||
750 | return err; | |
751 | } | |
752 | ||
753 | /** | |
754 | * rhashtable_remove_fast - remove object from hash table | |
755 | * @ht: hash table | |
756 | * @obj: pointer to hash head inside object | |
757 | * @params: hash table parameters | |
758 | * | |
759 | * Since the hash chain is single linked, the removal operation needs to | |
760 | * walk the bucket chain upon removal. The removal operation is thus | |
761 | * considerable slow if the hash table is not correctly sized. | |
762 | * | |
763 | * Will automatically shrink the table via rhashtable_expand() if the | |
764 | * shrink_decision function specified at rhashtable_init() returns true. | |
765 | * | |
766 | * Returns zero on success, -ENOENT if the entry could not be found. | |
767 | */ | |
768 | static inline int rhashtable_remove_fast( | |
769 | struct rhashtable *ht, struct rhash_head *obj, | |
770 | const struct rhashtable_params params) | |
771 | { | |
772 | struct bucket_table *tbl; | |
773 | int err; | |
774 | ||
775 | rcu_read_lock(); | |
776 | ||
777 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
778 | ||
779 | /* Because we have already taken (and released) the bucket | |
780 | * lock in old_tbl, if we find that future_tbl is not yet | |
781 | * visible then that guarantees the entry to still be in | |
782 | * the old tbl if it exists. | |
783 | */ | |
784 | while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) && | |
785 | (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) | |
786 | ; | |
787 | ||
788 | if (err) | |
789 | goto out; | |
790 | ||
791 | atomic_dec(&ht->nelems); | |
b5e2c150 TG |
792 | if (unlikely(ht->p.automatic_shrinking && |
793 | rht_shrink_below_30(ht, tbl))) | |
02fd97c3 HX |
794 | schedule_work(&ht->run_work); |
795 | ||
796 | out: | |
797 | rcu_read_unlock(); | |
798 | ||
799 | return err; | |
800 | } | |
801 | ||
7e1e7763 | 802 | #endif /* _LINUX_RHASHTABLE_H */ |