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 | |
52 | * @locks_mask: Mask to apply before accessing locks[] | |
53 | * @locks: Array of spinlocks protecting individual buckets | |
54 | * @buckets: size * hash buckets | |
55 | */ | |
7e1e7763 TG |
56 | struct bucket_table { |
57 | size_t size; | |
97defe1e TG |
58 | unsigned int locks_mask; |
59 | spinlock_t *locks; | |
7e1e7763 TG |
60 | struct rhash_head __rcu *buckets[]; |
61 | }; | |
62 | ||
63 | typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); | |
64 | typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 seed); | |
65 | ||
66 | struct rhashtable; | |
67 | ||
68 | /** | |
69 | * struct rhashtable_params - Hash table construction parameters | |
70 | * @nelem_hint: Hint on number of elements, should be 75% of desired size | |
71 | * @key_len: Length of key | |
72 | * @key_offset: Offset of key in struct to be hashed | |
73 | * @head_offset: Offset of rhash_head in struct to be hashed | |
74 | * @hash_rnd: Seed to use while hashing | |
75 | * @max_shift: Maximum number of shifts while expanding | |
94000176 | 76 | * @min_shift: Minimum number of shifts while shrinking |
f89bd6f8 | 77 | * @nulls_base: Base value to generate nulls marker |
97defe1e | 78 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) |
7e1e7763 TG |
79 | * @hashfn: Function to hash key |
80 | * @obj_hashfn: Function to hash object | |
81 | * @grow_decision: If defined, may return true if table should expand | |
82 | * @shrink_decision: If defined, may return true if table should shrink | |
6f73d3b1 YX |
83 | * |
84 | * Note: when implementing the grow and shrink decision function, min/max | |
85 | * shift must be enforced, otherwise, resizing watermarks they set may be | |
86 | * useless. | |
7e1e7763 TG |
87 | */ |
88 | struct rhashtable_params { | |
89 | size_t nelem_hint; | |
90 | size_t key_len; | |
91 | size_t key_offset; | |
92 | size_t head_offset; | |
93 | u32 hash_rnd; | |
94 | size_t max_shift; | |
94000176 | 95 | size_t min_shift; |
f89bd6f8 | 96 | u32 nulls_base; |
97defe1e | 97 | size_t locks_mul; |
7e1e7763 TG |
98 | rht_hashfn_t hashfn; |
99 | rht_obj_hashfn_t obj_hashfn; | |
100 | bool (*grow_decision)(const struct rhashtable *ht, | |
101 | size_t new_size); | |
102 | bool (*shrink_decision)(const struct rhashtable *ht, | |
103 | size_t new_size); | |
7e1e7763 TG |
104 | }; |
105 | ||
106 | /** | |
107 | * struct rhashtable - Hash table handle | |
108 | * @tbl: Bucket table | |
97defe1e | 109 | * @future_tbl: Table under construction during expansion/shrinking |
7e1e7763 TG |
110 | * @nelems: Number of elements in table |
111 | * @shift: Current size (1 << shift) | |
112 | * @p: Configuration parameters | |
97defe1e TG |
113 | * @run_work: Deferred worker to expand/shrink asynchronously |
114 | * @mutex: Mutex to protect current/future table swapping | |
f2dba9c6 | 115 | * @walkers: List of active walkers |
97defe1e | 116 | * @being_destroyed: True if table is set up for destruction |
7e1e7763 TG |
117 | */ |
118 | struct rhashtable { | |
119 | struct bucket_table __rcu *tbl; | |
97defe1e TG |
120 | struct bucket_table __rcu *future_tbl; |
121 | atomic_t nelems; | |
c0c09bfd | 122 | atomic_t shift; |
7e1e7763 | 123 | struct rhashtable_params p; |
57699a40 | 124 | struct work_struct run_work; |
97defe1e | 125 | struct mutex mutex; |
f2dba9c6 | 126 | struct list_head walkers; |
97defe1e | 127 | bool being_destroyed; |
7e1e7763 TG |
128 | }; |
129 | ||
f2dba9c6 HX |
130 | /** |
131 | * struct rhashtable_walker - Hash table walker | |
132 | * @list: List entry on list of walkers | |
133 | * @resize: Resize event occured | |
134 | */ | |
135 | struct rhashtable_walker { | |
136 | struct list_head list; | |
137 | bool resize; | |
138 | }; | |
139 | ||
140 | /** | |
141 | * struct rhashtable_iter - Hash table iterator, fits into netlink cb | |
142 | * @ht: Table to iterate through | |
143 | * @p: Current pointer | |
144 | * @walker: Associated rhashtable walker | |
145 | * @slot: Current slot | |
146 | * @skip: Number of entries to skip in slot | |
147 | */ | |
148 | struct rhashtable_iter { | |
149 | struct rhashtable *ht; | |
150 | struct rhash_head *p; | |
151 | struct rhashtable_walker *walker; | |
152 | unsigned int slot; | |
153 | unsigned int skip; | |
154 | }; | |
155 | ||
f89bd6f8 TG |
156 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) |
157 | { | |
158 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
159 | } | |
160 | ||
161 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
162 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
163 | ||
164 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
165 | { | |
166 | return ((unsigned long) ptr & 1); | |
167 | } | |
168 | ||
169 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
170 | { | |
171 | return ((unsigned long) ptr) >> 1; | |
172 | } | |
173 | ||
7e1e7763 | 174 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 175 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 176 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 177 | #else |
97defe1e | 178 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
179 | { |
180 | return 1; | |
181 | } | |
88d6ed15 TG |
182 | |
183 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
184 | u32 hash) | |
185 | { | |
186 | return 1; | |
187 | } | |
7e1e7763 TG |
188 | #endif /* CONFIG_PROVE_LOCKING */ |
189 | ||
190 | int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params); | |
191 | ||
6eba8224 TG |
192 | void rhashtable_insert(struct rhashtable *ht, struct rhash_head *node); |
193 | bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *node); | |
7e1e7763 TG |
194 | |
195 | bool rht_grow_above_75(const struct rhashtable *ht, size_t new_size); | |
196 | bool rht_shrink_below_30(const struct rhashtable *ht, size_t new_size); | |
197 | ||
6eba8224 TG |
198 | int rhashtable_expand(struct rhashtable *ht); |
199 | int rhashtable_shrink(struct rhashtable *ht); | |
7e1e7763 | 200 | |
97defe1e TG |
201 | void *rhashtable_lookup(struct rhashtable *ht, const void *key); |
202 | void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key, | |
7e1e7763 | 203 | bool (*compare)(void *, void *), void *arg); |
7a868d1e | 204 | |
db304854 | 205 | bool rhashtable_lookup_insert(struct rhashtable *ht, struct rhash_head *obj); |
7a868d1e YX |
206 | bool rhashtable_lookup_compare_insert(struct rhashtable *ht, |
207 | struct rhash_head *obj, | |
208 | bool (*compare)(void *, void *), | |
209 | void *arg); | |
7e1e7763 | 210 | |
f2dba9c6 HX |
211 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter); |
212 | void rhashtable_walk_exit(struct rhashtable_iter *iter); | |
213 | int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); | |
214 | void *rhashtable_walk_next(struct rhashtable_iter *iter); | |
215 | void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); | |
216 | ||
97defe1e | 217 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
218 | |
219 | #define rht_dereference(p, ht) \ | |
220 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
221 | ||
222 | #define rht_dereference_rcu(p, ht) \ | |
223 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
224 | ||
88d6ed15 TG |
225 | #define rht_dereference_bucket(p, tbl, hash) \ |
226 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 227 | |
88d6ed15 TG |
228 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
229 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
230 | ||
231 | #define rht_entry(tpos, pos, member) \ | |
232 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
233 | |
234 | /** | |
88d6ed15 TG |
235 | * rht_for_each_continue - continue iterating over hash chain |
236 | * @pos: the &struct rhash_head to use as a loop cursor. | |
237 | * @head: the previous &struct rhash_head to continue from | |
238 | * @tbl: the &struct bucket_table | |
239 | * @hash: the hash value / bucket index | |
7e1e7763 | 240 | */ |
88d6ed15 TG |
241 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
242 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 243 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
244 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
245 | ||
246 | /** | |
247 | * rht_for_each - iterate over hash chain | |
248 | * @pos: the &struct rhash_head to use as a loop cursor. | |
249 | * @tbl: the &struct bucket_table | |
250 | * @hash: the hash value / bucket index | |
251 | */ | |
252 | #define rht_for_each(pos, tbl, hash) \ | |
253 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
254 | ||
255 | /** | |
256 | * rht_for_each_entry_continue - continue iterating over hash chain | |
257 | * @tpos: the type * to use as a loop cursor. | |
258 | * @pos: the &struct rhash_head to use as a loop cursor. | |
259 | * @head: the previous &struct rhash_head to continue from | |
260 | * @tbl: the &struct bucket_table | |
261 | * @hash: the hash value / bucket index | |
262 | * @member: name of the &struct rhash_head within the hashable struct. | |
263 | */ | |
264 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
265 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 266 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 267 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
268 | |
269 | /** | |
270 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
271 | * @tpos: the type * to use as a loop cursor. |
272 | * @pos: the &struct rhash_head to use as a loop cursor. | |
273 | * @tbl: the &struct bucket_table | |
274 | * @hash: the hash value / bucket index | |
275 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 276 | */ |
88d6ed15 TG |
277 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
278 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
279 | tbl, hash, member) | |
7e1e7763 TG |
280 | |
281 | /** | |
282 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
283 | * @tpos: the type * to use as a loop cursor. |
284 | * @pos: the &struct rhash_head to use as a loop cursor. | |
285 | * @next: the &struct rhash_head to use as next in loop cursor. | |
286 | * @tbl: the &struct bucket_table | |
287 | * @hash: the hash value / bucket index | |
288 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
289 | * |
290 | * This hash chain list-traversal primitive allows for the looped code to | |
291 | * remove the loop cursor from the list. | |
292 | */ | |
88d6ed15 TG |
293 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
294 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
295 | next = !rht_is_a_nulls(pos) ? \ |
296 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
297 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
607954b0 PM |
298 | pos = next, \ |
299 | next = !rht_is_a_nulls(pos) ? \ | |
300 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
88d6ed15 TG |
301 | |
302 | /** | |
303 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
304 | * @pos: the &struct rhash_head to use as a loop cursor. | |
305 | * @head: the previous &struct rhash_head to continue from | |
306 | * @tbl: the &struct bucket_table | |
307 | * @hash: the hash value / bucket index | |
308 | * | |
309 | * This hash chain list-traversal primitive may safely run concurrently with | |
310 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
311 | * traversal is guarded by rcu_read_lock(). | |
312 | */ | |
313 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
314 | for (({barrier(); }), \ | |
315 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 316 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 317 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
318 | |
319 | /** | |
320 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
321 | * @pos: the &struct rhash_head to use as a loop cursor. |
322 | * @tbl: the &struct bucket_table | |
323 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
324 | * |
325 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 326 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
327 | * traversal is guarded by rcu_read_lock(). |
328 | */ | |
88d6ed15 TG |
329 | #define rht_for_each_rcu(pos, tbl, hash) \ |
330 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
331 | ||
332 | /** | |
333 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
334 | * @tpos: the type * to use as a loop cursor. | |
335 | * @pos: the &struct rhash_head to use as a loop cursor. | |
336 | * @head: the previous &struct rhash_head to continue from | |
337 | * @tbl: the &struct bucket_table | |
338 | * @hash: the hash value / bucket index | |
339 | * @member: name of the &struct rhash_head within the hashable struct. | |
340 | * | |
341 | * This hash chain list-traversal primitive may safely run concurrently with | |
342 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
343 | * traversal is guarded by rcu_read_lock(). | |
344 | */ | |
345 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
346 | for (({barrier(); }), \ | |
347 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 348 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 349 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
350 | |
351 | /** | |
352 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
353 | * @tpos: the type * to use as a loop cursor. |
354 | * @pos: the &struct rhash_head to use as a loop cursor. | |
355 | * @tbl: the &struct bucket_table | |
356 | * @hash: the hash value / bucket index | |
357 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
358 | * |
359 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 360 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
361 | * traversal is guarded by rcu_read_lock(). |
362 | */ | |
88d6ed15 TG |
363 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
364 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
365 | tbl, hash, member) | |
7e1e7763 TG |
366 | |
367 | #endif /* _LINUX_RHASHTABLE_H */ |