]>
Commit | Line | Data |
---|---|---|
7e1e7763 TG |
1 | /* |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
4 | * Copyright (c) 2014 Thomas Graf <tgraf@suug.ch> | |
5 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> | |
6 | * | |
7 | * Based on the following paper by Josh Triplett, Paul E. McKenney | |
8 | * and Jonathan Walpole: | |
9 | * https://www.usenix.org/legacy/event/atc11/tech/final_files/Triplett.pdf | |
10 | * | |
11 | * Code partially derived from nft_hash | |
12 | * | |
13 | * This program is free software; you can redistribute it and/or modify | |
14 | * it under the terms of the GNU General Public License version 2 as | |
15 | * published by the Free Software Foundation. | |
16 | */ | |
17 | ||
18 | #ifndef _LINUX_RHASHTABLE_H | |
19 | #define _LINUX_RHASHTABLE_H | |
20 | ||
f2dba9c6 | 21 | #include <linux/compiler.h> |
f89bd6f8 | 22 | #include <linux/list_nulls.h> |
97defe1e | 23 | #include <linux/workqueue.h> |
86b35b64 | 24 | #include <linux/mutex.h> |
7e1e7763 | 25 | |
f89bd6f8 TG |
26 | /* |
27 | * The end of the chain is marked with a special nulls marks which has | |
28 | * the following format: | |
29 | * | |
30 | * +-------+-----------------------------------------------------+-+ | |
31 | * | Base | Hash |1| | |
32 | * +-------+-----------------------------------------------------+-+ | |
33 | * | |
34 | * Base (4 bits) : Reserved to distinguish between multiple tables. | |
35 | * Specified via &struct rhashtable_params.nulls_base. | |
36 | * Hash (27 bits): Full hash (unmasked) of first element added to bucket | |
37 | * 1 (1 bit) : Nulls marker (always set) | |
38 | * | |
39 | * The remaining bits of the next pointer remain unused for now. | |
40 | */ | |
41 | #define RHT_BASE_BITS 4 | |
42 | #define RHT_HASH_BITS 27 | |
43 | #define RHT_BASE_SHIFT RHT_HASH_BITS | |
44 | ||
7e1e7763 | 45 | struct rhash_head { |
5300fdcb | 46 | struct rhash_head __rcu *next; |
7e1e7763 TG |
47 | }; |
48 | ||
97defe1e TG |
49 | /** |
50 | * struct bucket_table - Table of hash buckets | |
51 | * @size: Number of hash buckets | |
63d512d0 | 52 | * @rehash: Current bucket being rehashed |
988dfbd7 | 53 | * @hash_rnd: Random seed to fold into hash |
97defe1e TG |
54 | * @locks_mask: Mask to apply before accessing locks[] |
55 | * @locks: Array of spinlocks protecting individual buckets | |
eddee5ba | 56 | * @walkers: List of active walkers |
9d901bc0 | 57 | * @rcu: RCU structure for freeing the table |
c4db8848 | 58 | * @future_tbl: Table under construction during rehashing |
97defe1e TG |
59 | * @buckets: size * hash buckets |
60 | */ | |
7e1e7763 | 61 | struct bucket_table { |
63d512d0 HX |
62 | unsigned int size; |
63 | unsigned int rehash; | |
988dfbd7 | 64 | u32 hash_rnd; |
b9ebafbe ED |
65 | unsigned int locks_mask; |
66 | spinlock_t *locks; | |
eddee5ba | 67 | struct list_head walkers; |
9d901bc0 | 68 | struct rcu_head rcu; |
b9ebafbe | 69 | |
c4db8848 HX |
70 | struct bucket_table __rcu *future_tbl; |
71 | ||
b9ebafbe | 72 | struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp; |
7e1e7763 TG |
73 | }; |
74 | ||
75 | typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); | |
76 | typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 seed); | |
77 | ||
78 | struct rhashtable; | |
79 | ||
80 | /** | |
81 | * struct rhashtable_params - Hash table construction parameters | |
82 | * @nelem_hint: Hint on number of elements, should be 75% of desired size | |
83 | * @key_len: Length of key | |
84 | * @key_offset: Offset of key in struct to be hashed | |
85 | * @head_offset: Offset of rhash_head in struct to be hashed | |
7e1e7763 | 86 | * @max_shift: Maximum number of shifts while expanding |
94000176 | 87 | * @min_shift: Minimum number of shifts while shrinking |
f89bd6f8 | 88 | * @nulls_base: Base value to generate nulls marker |
97defe1e | 89 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) |
7e1e7763 TG |
90 | * @hashfn: Function to hash key |
91 | * @obj_hashfn: Function to hash object | |
7e1e7763 TG |
92 | */ |
93 | struct rhashtable_params { | |
94 | size_t nelem_hint; | |
95 | size_t key_len; | |
96 | size_t key_offset; | |
97 | size_t head_offset; | |
7e1e7763 | 98 | size_t max_shift; |
94000176 | 99 | size_t min_shift; |
f89bd6f8 | 100 | u32 nulls_base; |
97defe1e | 101 | size_t locks_mul; |
7e1e7763 TG |
102 | rht_hashfn_t hashfn; |
103 | rht_obj_hashfn_t obj_hashfn; | |
7e1e7763 TG |
104 | }; |
105 | ||
106 | /** | |
107 | * struct rhashtable - Hash table handle | |
108 | * @tbl: Bucket table | |
109 | * @nelems: Number of elements in table | |
7e1e7763 | 110 | * @p: Configuration parameters |
97defe1e TG |
111 | * @run_work: Deferred worker to expand/shrink asynchronously |
112 | * @mutex: Mutex to protect current/future table swapping | |
113 | * @being_destroyed: True if table is set up for destruction | |
7e1e7763 TG |
114 | */ |
115 | struct rhashtable { | |
116 | struct bucket_table __rcu *tbl; | |
97defe1e | 117 | atomic_t nelems; |
a5b6846f | 118 | bool being_destroyed; |
7e1e7763 | 119 | struct rhashtable_params p; |
57699a40 | 120 | struct work_struct run_work; |
97defe1e | 121 | struct mutex mutex; |
7e1e7763 TG |
122 | }; |
123 | ||
f2dba9c6 HX |
124 | /** |
125 | * struct rhashtable_walker - Hash table walker | |
126 | * @list: List entry on list of walkers | |
eddee5ba | 127 | * @tbl: The table that we were walking over |
f2dba9c6 HX |
128 | */ |
129 | struct rhashtable_walker { | |
130 | struct list_head list; | |
eddee5ba | 131 | struct bucket_table *tbl; |
f2dba9c6 HX |
132 | }; |
133 | ||
134 | /** | |
135 | * struct rhashtable_iter - Hash table iterator, fits into netlink cb | |
136 | * @ht: Table to iterate through | |
137 | * @p: Current pointer | |
138 | * @walker: Associated rhashtable walker | |
139 | * @slot: Current slot | |
140 | * @skip: Number of entries to skip in slot | |
141 | */ | |
142 | struct rhashtable_iter { | |
143 | struct rhashtable *ht; | |
144 | struct rhash_head *p; | |
145 | struct rhashtable_walker *walker; | |
146 | unsigned int slot; | |
147 | unsigned int skip; | |
148 | }; | |
149 | ||
f89bd6f8 TG |
150 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) |
151 | { | |
152 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
153 | } | |
154 | ||
155 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
156 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
157 | ||
158 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
159 | { | |
160 | return ((unsigned long) ptr & 1); | |
161 | } | |
162 | ||
163 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
164 | { | |
165 | return ((unsigned long) ptr) >> 1; | |
166 | } | |
167 | ||
7e1e7763 | 168 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 169 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 170 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 171 | #else |
97defe1e | 172 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
173 | { |
174 | return 1; | |
175 | } | |
88d6ed15 TG |
176 | |
177 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
178 | u32 hash) | |
179 | { | |
180 | return 1; | |
181 | } | |
7e1e7763 TG |
182 | #endif /* CONFIG_PROVE_LOCKING */ |
183 | ||
184 | int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params); | |
185 | ||
6eba8224 TG |
186 | void rhashtable_insert(struct rhashtable *ht, struct rhash_head *node); |
187 | bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *node); | |
7e1e7763 | 188 | |
6eba8224 TG |
189 | int rhashtable_expand(struct rhashtable *ht); |
190 | int rhashtable_shrink(struct rhashtable *ht); | |
7e1e7763 | 191 | |
97defe1e TG |
192 | void *rhashtable_lookup(struct rhashtable *ht, const void *key); |
193 | void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key, | |
7e1e7763 | 194 | bool (*compare)(void *, void *), void *arg); |
7a868d1e | 195 | |
db304854 | 196 | bool rhashtable_lookup_insert(struct rhashtable *ht, struct rhash_head *obj); |
7a868d1e YX |
197 | bool rhashtable_lookup_compare_insert(struct rhashtable *ht, |
198 | struct rhash_head *obj, | |
199 | bool (*compare)(void *, void *), | |
200 | void *arg); | |
7e1e7763 | 201 | |
f2dba9c6 HX |
202 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter); |
203 | void rhashtable_walk_exit(struct rhashtable_iter *iter); | |
204 | int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); | |
205 | void *rhashtable_walk_next(struct rhashtable_iter *iter); | |
206 | void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); | |
207 | ||
97defe1e | 208 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
209 | |
210 | #define rht_dereference(p, ht) \ | |
211 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
212 | ||
213 | #define rht_dereference_rcu(p, ht) \ | |
214 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
215 | ||
88d6ed15 TG |
216 | #define rht_dereference_bucket(p, tbl, hash) \ |
217 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 218 | |
88d6ed15 TG |
219 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
220 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
221 | ||
222 | #define rht_entry(tpos, pos, member) \ | |
223 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
224 | |
225 | /** | |
88d6ed15 TG |
226 | * rht_for_each_continue - continue iterating over hash chain |
227 | * @pos: the &struct rhash_head to use as a loop cursor. | |
228 | * @head: the previous &struct rhash_head to continue from | |
229 | * @tbl: the &struct bucket_table | |
230 | * @hash: the hash value / bucket index | |
7e1e7763 | 231 | */ |
88d6ed15 TG |
232 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
233 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 234 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
235 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
236 | ||
237 | /** | |
238 | * rht_for_each - iterate over hash chain | |
239 | * @pos: the &struct rhash_head to use as a loop cursor. | |
240 | * @tbl: the &struct bucket_table | |
241 | * @hash: the hash value / bucket index | |
242 | */ | |
243 | #define rht_for_each(pos, tbl, hash) \ | |
244 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
245 | ||
246 | /** | |
247 | * rht_for_each_entry_continue - continue iterating over hash chain | |
248 | * @tpos: the type * to use as a loop cursor. | |
249 | * @pos: the &struct rhash_head to use as a loop cursor. | |
250 | * @head: the previous &struct rhash_head to continue from | |
251 | * @tbl: the &struct bucket_table | |
252 | * @hash: the hash value / bucket index | |
253 | * @member: name of the &struct rhash_head within the hashable struct. | |
254 | */ | |
255 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
256 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 257 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 258 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
259 | |
260 | /** | |
261 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
262 | * @tpos: the type * to use as a loop cursor. |
263 | * @pos: the &struct rhash_head to use as a loop cursor. | |
264 | * @tbl: the &struct bucket_table | |
265 | * @hash: the hash value / bucket index | |
266 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 267 | */ |
88d6ed15 TG |
268 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
269 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
270 | tbl, hash, member) | |
7e1e7763 TG |
271 | |
272 | /** | |
273 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
274 | * @tpos: the type * to use as a loop cursor. |
275 | * @pos: the &struct rhash_head to use as a loop cursor. | |
276 | * @next: the &struct rhash_head to use as next in loop cursor. | |
277 | * @tbl: the &struct bucket_table | |
278 | * @hash: the hash value / bucket index | |
279 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
280 | * |
281 | * This hash chain list-traversal primitive allows for the looped code to | |
282 | * remove the loop cursor from the list. | |
283 | */ | |
88d6ed15 TG |
284 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
285 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
286 | next = !rht_is_a_nulls(pos) ? \ |
287 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
288 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
607954b0 PM |
289 | pos = next, \ |
290 | next = !rht_is_a_nulls(pos) ? \ | |
291 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
88d6ed15 TG |
292 | |
293 | /** | |
294 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
295 | * @pos: the &struct rhash_head to use as a loop cursor. | |
296 | * @head: the previous &struct rhash_head to continue from | |
297 | * @tbl: the &struct bucket_table | |
298 | * @hash: the hash value / bucket index | |
299 | * | |
300 | * This hash chain list-traversal primitive may safely run concurrently with | |
301 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
302 | * traversal is guarded by rcu_read_lock(). | |
303 | */ | |
304 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
305 | for (({barrier(); }), \ | |
306 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 307 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 308 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
309 | |
310 | /** | |
311 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
312 | * @pos: the &struct rhash_head to use as a loop cursor. |
313 | * @tbl: the &struct bucket_table | |
314 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
315 | * |
316 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 317 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
318 | * traversal is guarded by rcu_read_lock(). |
319 | */ | |
88d6ed15 TG |
320 | #define rht_for_each_rcu(pos, tbl, hash) \ |
321 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
322 | ||
323 | /** | |
324 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
325 | * @tpos: the type * to use as a loop cursor. | |
326 | * @pos: the &struct rhash_head to use as a loop cursor. | |
327 | * @head: the previous &struct rhash_head to continue from | |
328 | * @tbl: the &struct bucket_table | |
329 | * @hash: the hash value / bucket index | |
330 | * @member: name of the &struct rhash_head within the hashable struct. | |
331 | * | |
332 | * This hash chain list-traversal primitive may safely run concurrently with | |
333 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
334 | * traversal is guarded by rcu_read_lock(). | |
335 | */ | |
336 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
337 | for (({barrier(); }), \ | |
338 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 339 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 340 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
341 | |
342 | /** | |
343 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
344 | * @tpos: the type * to use as a loop cursor. |
345 | * @pos: the &struct rhash_head to use as a loop cursor. | |
346 | * @tbl: the &struct bucket_table | |
347 | * @hash: the hash value / bucket index | |
348 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
349 | * |
350 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 351 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
352 | * traversal is guarded by rcu_read_lock(). |
353 | */ | |
88d6ed15 TG |
354 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
355 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
356 | tbl, hash, member) | |
7e1e7763 TG |
357 | |
358 | #endif /* _LINUX_RHASHTABLE_H */ |