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 | ||
f89bd6f8 | 21 | #include <linux/list_nulls.h> |
97defe1e | 22 | #include <linux/workqueue.h> |
7e1e7763 | 23 | |
f89bd6f8 TG |
24 | /* |
25 | * The end of the chain is marked with a special nulls marks which has | |
26 | * the following format: | |
27 | * | |
28 | * +-------+-----------------------------------------------------+-+ | |
29 | * | Base | Hash |1| | |
30 | * +-------+-----------------------------------------------------+-+ | |
31 | * | |
32 | * Base (4 bits) : Reserved to distinguish between multiple tables. | |
33 | * Specified via &struct rhashtable_params.nulls_base. | |
34 | * Hash (27 bits): Full hash (unmasked) of first element added to bucket | |
35 | * 1 (1 bit) : Nulls marker (always set) | |
36 | * | |
37 | * The remaining bits of the next pointer remain unused for now. | |
38 | */ | |
39 | #define RHT_BASE_BITS 4 | |
40 | #define RHT_HASH_BITS 27 | |
41 | #define RHT_BASE_SHIFT RHT_HASH_BITS | |
42 | ||
7e1e7763 | 43 | struct rhash_head { |
5300fdcb | 44 | struct rhash_head __rcu *next; |
7e1e7763 TG |
45 | }; |
46 | ||
97defe1e TG |
47 | /** |
48 | * struct bucket_table - Table of hash buckets | |
49 | * @size: Number of hash buckets | |
50 | * @locks_mask: Mask to apply before accessing locks[] | |
51 | * @locks: Array of spinlocks protecting individual buckets | |
52 | * @buckets: size * hash buckets | |
53 | */ | |
7e1e7763 TG |
54 | struct bucket_table { |
55 | size_t size; | |
97defe1e TG |
56 | unsigned int locks_mask; |
57 | spinlock_t *locks; | |
7e1e7763 TG |
58 | struct rhash_head __rcu *buckets[]; |
59 | }; | |
60 | ||
61 | typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); | |
62 | typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 seed); | |
63 | ||
64 | struct rhashtable; | |
65 | ||
66 | /** | |
67 | * struct rhashtable_params - Hash table construction parameters | |
68 | * @nelem_hint: Hint on number of elements, should be 75% of desired size | |
69 | * @key_len: Length of key | |
70 | * @key_offset: Offset of key in struct to be hashed | |
71 | * @head_offset: Offset of rhash_head in struct to be hashed | |
72 | * @hash_rnd: Seed to use while hashing | |
73 | * @max_shift: Maximum number of shifts while expanding | |
94000176 | 74 | * @min_shift: Minimum number of shifts while shrinking |
f89bd6f8 | 75 | * @nulls_base: Base value to generate nulls marker |
97defe1e | 76 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) |
7e1e7763 TG |
77 | * @hashfn: Function to hash key |
78 | * @obj_hashfn: Function to hash object | |
79 | * @grow_decision: If defined, may return true if table should expand | |
80 | * @shrink_decision: If defined, may return true if table should shrink | |
7e1e7763 TG |
81 | */ |
82 | struct rhashtable_params { | |
83 | size_t nelem_hint; | |
84 | size_t key_len; | |
85 | size_t key_offset; | |
86 | size_t head_offset; | |
87 | u32 hash_rnd; | |
88 | size_t max_shift; | |
94000176 | 89 | size_t min_shift; |
f89bd6f8 | 90 | u32 nulls_base; |
97defe1e | 91 | size_t locks_mul; |
7e1e7763 TG |
92 | rht_hashfn_t hashfn; |
93 | rht_obj_hashfn_t obj_hashfn; | |
94 | bool (*grow_decision)(const struct rhashtable *ht, | |
95 | size_t new_size); | |
96 | bool (*shrink_decision)(const struct rhashtable *ht, | |
97 | size_t new_size); | |
7e1e7763 TG |
98 | }; |
99 | ||
100 | /** | |
101 | * struct rhashtable - Hash table handle | |
102 | * @tbl: Bucket table | |
97defe1e | 103 | * @future_tbl: Table under construction during expansion/shrinking |
7e1e7763 TG |
104 | * @nelems: Number of elements in table |
105 | * @shift: Current size (1 << shift) | |
106 | * @p: Configuration parameters | |
97defe1e TG |
107 | * @run_work: Deferred worker to expand/shrink asynchronously |
108 | * @mutex: Mutex to protect current/future table swapping | |
109 | * @being_destroyed: True if table is set up for destruction | |
7e1e7763 TG |
110 | */ |
111 | struct rhashtable { | |
112 | struct bucket_table __rcu *tbl; | |
97defe1e TG |
113 | struct bucket_table __rcu *future_tbl; |
114 | atomic_t nelems; | |
7e1e7763 TG |
115 | size_t shift; |
116 | struct rhashtable_params p; | |
97defe1e TG |
117 | struct delayed_work run_work; |
118 | struct mutex mutex; | |
119 | bool being_destroyed; | |
7e1e7763 TG |
120 | }; |
121 | ||
f89bd6f8 TG |
122 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) |
123 | { | |
124 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
125 | } | |
126 | ||
127 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
128 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
129 | ||
130 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
131 | { | |
132 | return ((unsigned long) ptr & 1); | |
133 | } | |
134 | ||
135 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
136 | { | |
137 | return ((unsigned long) ptr) >> 1; | |
138 | } | |
139 | ||
7e1e7763 | 140 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 141 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 142 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 143 | #else |
97defe1e | 144 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
145 | { |
146 | return 1; | |
147 | } | |
88d6ed15 TG |
148 | |
149 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
150 | u32 hash) | |
151 | { | |
152 | return 1; | |
153 | } | |
7e1e7763 TG |
154 | #endif /* CONFIG_PROVE_LOCKING */ |
155 | ||
156 | int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params); | |
157 | ||
6eba8224 TG |
158 | void rhashtable_insert(struct rhashtable *ht, struct rhash_head *node); |
159 | bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *node); | |
7e1e7763 TG |
160 | |
161 | bool rht_grow_above_75(const struct rhashtable *ht, size_t new_size); | |
162 | bool rht_shrink_below_30(const struct rhashtable *ht, size_t new_size); | |
163 | ||
6eba8224 TG |
164 | int rhashtable_expand(struct rhashtable *ht); |
165 | int rhashtable_shrink(struct rhashtable *ht); | |
7e1e7763 | 166 | |
97defe1e TG |
167 | void *rhashtable_lookup(struct rhashtable *ht, const void *key); |
168 | void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key, | |
7e1e7763 TG |
169 | bool (*compare)(void *, void *), void *arg); |
170 | ||
97defe1e | 171 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
172 | |
173 | #define rht_dereference(p, ht) \ | |
174 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
175 | ||
176 | #define rht_dereference_rcu(p, ht) \ | |
177 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
178 | ||
88d6ed15 TG |
179 | #define rht_dereference_bucket(p, tbl, hash) \ |
180 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 181 | |
88d6ed15 TG |
182 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
183 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
184 | ||
185 | #define rht_entry(tpos, pos, member) \ | |
186 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
187 | |
188 | /** | |
88d6ed15 TG |
189 | * rht_for_each_continue - continue iterating over hash chain |
190 | * @pos: the &struct rhash_head to use as a loop cursor. | |
191 | * @head: the previous &struct rhash_head to continue from | |
192 | * @tbl: the &struct bucket_table | |
193 | * @hash: the hash value / bucket index | |
7e1e7763 | 194 | */ |
88d6ed15 TG |
195 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
196 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 197 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
198 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
199 | ||
200 | /** | |
201 | * rht_for_each - iterate over hash chain | |
202 | * @pos: the &struct rhash_head to use as a loop cursor. | |
203 | * @tbl: the &struct bucket_table | |
204 | * @hash: the hash value / bucket index | |
205 | */ | |
206 | #define rht_for_each(pos, tbl, hash) \ | |
207 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
208 | ||
209 | /** | |
210 | * rht_for_each_entry_continue - continue iterating over hash chain | |
211 | * @tpos: the type * to use as a loop cursor. | |
212 | * @pos: the &struct rhash_head to use as a loop cursor. | |
213 | * @head: the previous &struct rhash_head to continue from | |
214 | * @tbl: the &struct bucket_table | |
215 | * @hash: the hash value / bucket index | |
216 | * @member: name of the &struct rhash_head within the hashable struct. | |
217 | */ | |
218 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
219 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 220 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 221 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
222 | |
223 | /** | |
224 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
225 | * @tpos: the type * to use as a loop cursor. |
226 | * @pos: the &struct rhash_head to use as a loop cursor. | |
227 | * @tbl: the &struct bucket_table | |
228 | * @hash: the hash value / bucket index | |
229 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 230 | */ |
88d6ed15 TG |
231 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
232 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
233 | tbl, hash, member) | |
7e1e7763 TG |
234 | |
235 | /** | |
236 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
237 | * @tpos: the type * to use as a loop cursor. |
238 | * @pos: the &struct rhash_head to use as a loop cursor. | |
239 | * @next: the &struct rhash_head to use as next in loop cursor. | |
240 | * @tbl: the &struct bucket_table | |
241 | * @hash: the hash value / bucket index | |
242 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
243 | * |
244 | * This hash chain list-traversal primitive allows for the looped code to | |
245 | * remove the loop cursor from the list. | |
246 | */ | |
88d6ed15 TG |
247 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
248 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
249 | next = !rht_is_a_nulls(pos) ? \ |
250 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
251 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
88d6ed15 TG |
252 | pos = next) |
253 | ||
254 | /** | |
255 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
256 | * @pos: the &struct rhash_head to use as a loop cursor. | |
257 | * @head: the previous &struct rhash_head to continue from | |
258 | * @tbl: the &struct bucket_table | |
259 | * @hash: the hash value / bucket index | |
260 | * | |
261 | * This hash chain list-traversal primitive may safely run concurrently with | |
262 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
263 | * traversal is guarded by rcu_read_lock(). | |
264 | */ | |
265 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
266 | for (({barrier(); }), \ | |
267 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 268 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 269 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
270 | |
271 | /** | |
272 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
273 | * @pos: the &struct rhash_head to use as a loop cursor. |
274 | * @tbl: the &struct bucket_table | |
275 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
276 | * |
277 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 278 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
279 | * traversal is guarded by rcu_read_lock(). |
280 | */ | |
88d6ed15 TG |
281 | #define rht_for_each_rcu(pos, tbl, hash) \ |
282 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
283 | ||
284 | /** | |
285 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
286 | * @tpos: the type * to use as a loop cursor. | |
287 | * @pos: the &struct rhash_head to use as a loop cursor. | |
288 | * @head: the previous &struct rhash_head to continue from | |
289 | * @tbl: the &struct bucket_table | |
290 | * @hash: the hash value / bucket index | |
291 | * @member: name of the &struct rhash_head within the hashable struct. | |
292 | * | |
293 | * This hash chain list-traversal primitive may safely run concurrently with | |
294 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
295 | * traversal is guarded by rcu_read_lock(). | |
296 | */ | |
297 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
298 | for (({barrier(); }), \ | |
299 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 300 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 301 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
302 | |
303 | /** | |
304 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
305 | * @tpos: the type * to use as a loop cursor. |
306 | * @pos: the &struct rhash_head to use as a loop cursor. | |
307 | * @tbl: the &struct bucket_table | |
308 | * @hash: the hash value / bucket index | |
309 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
310 | * |
311 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 312 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
313 | * traversal is guarded by rcu_read_lock(). |
314 | */ | |
88d6ed15 TG |
315 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
316 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
317 | tbl, hash, member) | |
7e1e7763 TG |
318 | |
319 | #endif /* _LINUX_RHASHTABLE_H */ |