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 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 |
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) ? | |
245 | rht_bucket_index(tbl, params.obj_hashfn(ptr, tbl->hash_rnd)) : | |
246 | rht_key_hashfn(ht, tbl, ptr + params.key_offset, params); | |
247 | } | |
248 | ||
249 | /** | |
250 | * rht_grow_above_75 - returns true if nelems > 0.75 * table-size | |
251 | * @ht: hash table | |
252 | * @tbl: current table | |
253 | */ | |
254 | static inline bool rht_grow_above_75(const struct rhashtable *ht, | |
255 | const struct bucket_table *tbl) | |
256 | { | |
257 | /* Expand table when exceeding 75% load */ | |
258 | return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) && | |
259 | (!ht->p.max_size || tbl->size < ht->p.max_size); | |
260 | } | |
261 | ||
262 | /** | |
263 | * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size | |
264 | * @ht: hash table | |
265 | * @tbl: current table | |
266 | */ | |
267 | static inline bool rht_shrink_below_30(const struct rhashtable *ht, | |
268 | const struct bucket_table *tbl) | |
269 | { | |
270 | /* Shrink table beneath 30% load */ | |
271 | return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) && | |
272 | tbl->size > ht->p.min_size; | |
273 | } | |
274 | ||
ccd57b1b HX |
275 | /** |
276 | * rht_grow_above_100 - returns true if nelems > table-size | |
277 | * @ht: hash table | |
278 | * @tbl: current table | |
279 | */ | |
280 | static inline bool rht_grow_above_100(const struct rhashtable *ht, | |
281 | const struct bucket_table *tbl) | |
282 | { | |
283 | return atomic_read(&ht->nelems) > tbl->size; | |
284 | } | |
285 | ||
02fd97c3 HX |
286 | /* The bucket lock is selected based on the hash and protects mutations |
287 | * on a group of hash buckets. | |
288 | * | |
289 | * A maximum of tbl->size/2 bucket locks is allocated. This ensures that | |
290 | * a single lock always covers both buckets which may both contains | |
291 | * entries which link to the same bucket of the old table during resizing. | |
292 | * This allows to simplify the locking as locking the bucket in both | |
293 | * tables during resize always guarantee protection. | |
294 | * | |
295 | * IMPORTANT: When holding the bucket lock of both the old and new table | |
296 | * during expansions and shrinking, the old bucket lock must always be | |
297 | * acquired first. | |
298 | */ | |
299 | static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl, | |
300 | unsigned int hash) | |
301 | { | |
302 | return &tbl->locks[hash & tbl->locks_mask]; | |
303 | } | |
304 | ||
7e1e7763 | 305 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 306 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 307 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 308 | #else |
97defe1e | 309 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
310 | { |
311 | return 1; | |
312 | } | |
88d6ed15 TG |
313 | |
314 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
315 | u32 hash) | |
316 | { | |
317 | return 1; | |
318 | } | |
7e1e7763 TG |
319 | #endif /* CONFIG_PROVE_LOCKING */ |
320 | ||
488fb86e HX |
321 | int rhashtable_init(struct rhashtable *ht, |
322 | const struct rhashtable_params *params); | |
7e1e7763 | 323 | |
02fd97c3 HX |
324 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, |
325 | struct rhash_head *obj, | |
326 | struct bucket_table *old_tbl); | |
ccd57b1b | 327 | int rhashtable_insert_rehash(struct rhashtable *ht); |
7e1e7763 | 328 | |
f2dba9c6 HX |
329 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter); |
330 | void rhashtable_walk_exit(struct rhashtable_iter *iter); | |
331 | int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); | |
332 | void *rhashtable_walk_next(struct rhashtable_iter *iter); | |
333 | void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); | |
334 | ||
6b6f302c TG |
335 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
336 | void (*free_fn)(void *ptr, void *arg), | |
337 | void *arg); | |
97defe1e | 338 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
339 | |
340 | #define rht_dereference(p, ht) \ | |
341 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
342 | ||
343 | #define rht_dereference_rcu(p, ht) \ | |
344 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
345 | ||
88d6ed15 TG |
346 | #define rht_dereference_bucket(p, tbl, hash) \ |
347 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 348 | |
88d6ed15 TG |
349 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
350 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
351 | ||
352 | #define rht_entry(tpos, pos, member) \ | |
353 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
354 | |
355 | /** | |
88d6ed15 TG |
356 | * rht_for_each_continue - continue iterating over hash chain |
357 | * @pos: the &struct rhash_head to use as a loop cursor. | |
358 | * @head: the previous &struct rhash_head to continue from | |
359 | * @tbl: the &struct bucket_table | |
360 | * @hash: the hash value / bucket index | |
7e1e7763 | 361 | */ |
88d6ed15 TG |
362 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
363 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 364 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
365 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
366 | ||
367 | /** | |
368 | * rht_for_each - iterate over hash chain | |
369 | * @pos: the &struct rhash_head to use as a loop cursor. | |
370 | * @tbl: the &struct bucket_table | |
371 | * @hash: the hash value / bucket index | |
372 | */ | |
373 | #define rht_for_each(pos, tbl, hash) \ | |
374 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
375 | ||
376 | /** | |
377 | * rht_for_each_entry_continue - continue iterating over hash chain | |
378 | * @tpos: the type * to use as a loop cursor. | |
379 | * @pos: the &struct rhash_head to use as a loop cursor. | |
380 | * @head: the previous &struct rhash_head to continue from | |
381 | * @tbl: the &struct bucket_table | |
382 | * @hash: the hash value / bucket index | |
383 | * @member: name of the &struct rhash_head within the hashable struct. | |
384 | */ | |
385 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
386 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 387 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 388 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
389 | |
390 | /** | |
391 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
392 | * @tpos: the type * to use as a loop cursor. |
393 | * @pos: the &struct rhash_head to use as a loop cursor. | |
394 | * @tbl: the &struct bucket_table | |
395 | * @hash: the hash value / bucket index | |
396 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 397 | */ |
88d6ed15 TG |
398 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
399 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
400 | tbl, hash, member) | |
7e1e7763 TG |
401 | |
402 | /** | |
403 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
404 | * @tpos: the type * to use as a loop cursor. |
405 | * @pos: the &struct rhash_head to use as a loop cursor. | |
406 | * @next: the &struct rhash_head to use as next in loop cursor. | |
407 | * @tbl: the &struct bucket_table | |
408 | * @hash: the hash value / bucket index | |
409 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
410 | * |
411 | * This hash chain list-traversal primitive allows for the looped code to | |
412 | * remove the loop cursor from the list. | |
413 | */ | |
88d6ed15 TG |
414 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
415 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
416 | next = !rht_is_a_nulls(pos) ? \ |
417 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
418 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
607954b0 PM |
419 | pos = next, \ |
420 | next = !rht_is_a_nulls(pos) ? \ | |
421 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
88d6ed15 TG |
422 | |
423 | /** | |
424 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
425 | * @pos: the &struct rhash_head to use as a loop cursor. | |
426 | * @head: the previous &struct rhash_head to continue from | |
427 | * @tbl: the &struct bucket_table | |
428 | * @hash: the hash value / bucket index | |
429 | * | |
430 | * This hash chain list-traversal primitive may safely run concurrently with | |
431 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
432 | * traversal is guarded by rcu_read_lock(). | |
433 | */ | |
434 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
435 | for (({barrier(); }), \ | |
436 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 437 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 438 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
439 | |
440 | /** | |
441 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
442 | * @pos: the &struct rhash_head to use as a loop cursor. |
443 | * @tbl: the &struct bucket_table | |
444 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
445 | * |
446 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 447 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
448 | * traversal is guarded by rcu_read_lock(). |
449 | */ | |
88d6ed15 TG |
450 | #define rht_for_each_rcu(pos, tbl, hash) \ |
451 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
452 | ||
453 | /** | |
454 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
455 | * @tpos: the type * to use as a loop cursor. | |
456 | * @pos: the &struct rhash_head to use as a loop cursor. | |
457 | * @head: the previous &struct rhash_head to continue from | |
458 | * @tbl: the &struct bucket_table | |
459 | * @hash: the hash value / bucket index | |
460 | * @member: name of the &struct rhash_head within the hashable struct. | |
461 | * | |
462 | * This hash chain list-traversal primitive may safely run concurrently with | |
463 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
464 | * traversal is guarded by rcu_read_lock(). | |
465 | */ | |
466 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
467 | for (({barrier(); }), \ | |
468 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 469 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 470 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
471 | |
472 | /** | |
473 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
474 | * @tpos: the type * to use as a loop cursor. |
475 | * @pos: the &struct rhash_head to use as a loop cursor. | |
476 | * @tbl: the &struct bucket_table | |
477 | * @hash: the hash value / bucket index | |
478 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
479 | * |
480 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 481 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
482 | * traversal is guarded by rcu_read_lock(). |
483 | */ | |
88d6ed15 TG |
484 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
485 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
486 | tbl, hash, member) | |
7e1e7763 | 487 | |
02fd97c3 HX |
488 | static inline int rhashtable_compare(struct rhashtable_compare_arg *arg, |
489 | const void *obj) | |
490 | { | |
491 | struct rhashtable *ht = arg->ht; | |
492 | const char *ptr = obj; | |
493 | ||
494 | return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len); | |
495 | } | |
496 | ||
497 | /** | |
498 | * rhashtable_lookup_fast - search hash table, inlined version | |
499 | * @ht: hash table | |
500 | * @key: the pointer to the key | |
501 | * @params: hash table parameters | |
502 | * | |
503 | * Computes the hash value for the key and traverses the bucket chain looking | |
504 | * for a entry with an identical key. The first matching entry is returned. | |
505 | * | |
506 | * Returns the first entry on which the compare function returned true. | |
507 | */ | |
508 | static inline void *rhashtable_lookup_fast( | |
509 | struct rhashtable *ht, const void *key, | |
510 | const struct rhashtable_params params) | |
511 | { | |
512 | struct rhashtable_compare_arg arg = { | |
513 | .ht = ht, | |
514 | .key = key, | |
515 | }; | |
516 | const struct bucket_table *tbl; | |
517 | struct rhash_head *he; | |
299e5c32 | 518 | unsigned int hash; |
02fd97c3 HX |
519 | |
520 | rcu_read_lock(); | |
521 | ||
522 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
523 | restart: | |
524 | hash = rht_key_hashfn(ht, tbl, key, params); | |
525 | rht_for_each_rcu(he, tbl, hash) { | |
526 | if (params.obj_cmpfn ? | |
527 | params.obj_cmpfn(&arg, rht_obj(ht, he)) : | |
528 | rhashtable_compare(&arg, rht_obj(ht, he))) | |
529 | continue; | |
530 | rcu_read_unlock(); | |
531 | return rht_obj(ht, he); | |
532 | } | |
533 | ||
534 | /* Ensure we see any new tables. */ | |
535 | smp_rmb(); | |
536 | ||
537 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
538 | if (unlikely(tbl)) | |
539 | goto restart; | |
540 | rcu_read_unlock(); | |
541 | ||
542 | return NULL; | |
543 | } | |
544 | ||
ac833bdd | 545 | /* Internal function, please use rhashtable_insert_fast() instead */ |
02fd97c3 HX |
546 | static inline int __rhashtable_insert_fast( |
547 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
548 | const struct rhashtable_params params) | |
549 | { | |
550 | struct rhashtable_compare_arg arg = { | |
551 | .ht = ht, | |
552 | .key = key, | |
553 | }; | |
02fd97c3 HX |
554 | struct bucket_table *tbl, *new_tbl; |
555 | struct rhash_head *head; | |
556 | spinlock_t *lock; | |
299e5c32 TG |
557 | unsigned int elasticity; |
558 | unsigned int hash; | |
ccd57b1b | 559 | int err; |
02fd97c3 | 560 | |
ccd57b1b | 561 | restart: |
02fd97c3 HX |
562 | rcu_read_lock(); |
563 | ||
564 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
02fd97c3 | 565 | |
b824478b HX |
566 | /* All insertions must grab the oldest table containing |
567 | * the hashed bucket that is yet to be rehashed. | |
02fd97c3 | 568 | */ |
b824478b HX |
569 | for (;;) { |
570 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
571 | lock = rht_bucket_lock(tbl, hash); | |
572 | spin_lock_bh(lock); | |
573 | ||
574 | if (tbl->rehash <= hash) | |
575 | break; | |
576 | ||
577 | spin_unlock_bh(lock); | |
578 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
579 | } | |
580 | ||
02fd97c3 HX |
581 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
582 | if (unlikely(new_tbl)) { | |
583 | err = rhashtable_insert_slow(ht, key, obj, new_tbl); | |
ccd57b1b HX |
584 | if (err == -EAGAIN) |
585 | goto slow_path; | |
02fd97c3 HX |
586 | goto out; |
587 | } | |
588 | ||
ccd57b1b HX |
589 | if (unlikely(rht_grow_above_100(ht, tbl))) { |
590 | slow_path: | |
591 | spin_unlock_bh(lock); | |
ccd57b1b | 592 | err = rhashtable_insert_rehash(ht); |
58be8a58 | 593 | rcu_read_unlock(); |
ccd57b1b HX |
594 | if (err) |
595 | return err; | |
596 | ||
597 | goto restart; | |
598 | } | |
02fd97c3 | 599 | |
ccd57b1b HX |
600 | err = -EEXIST; |
601 | elasticity = ht->elasticity; | |
02fd97c3 | 602 | rht_for_each(head, tbl, hash) { |
ccd57b1b HX |
603 | if (key && |
604 | unlikely(!(params.obj_cmpfn ? | |
02fd97c3 HX |
605 | params.obj_cmpfn(&arg, rht_obj(ht, head)) : |
606 | rhashtable_compare(&arg, rht_obj(ht, head))))) | |
607 | goto out; | |
ccd57b1b HX |
608 | if (!--elasticity) |
609 | goto slow_path; | |
02fd97c3 HX |
610 | } |
611 | ||
02fd97c3 HX |
612 | err = 0; |
613 | ||
614 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
615 | ||
616 | RCU_INIT_POINTER(obj->next, head); | |
617 | ||
618 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
619 | ||
620 | atomic_inc(&ht->nelems); | |
621 | if (rht_grow_above_75(ht, tbl)) | |
622 | schedule_work(&ht->run_work); | |
623 | ||
624 | out: | |
625 | spin_unlock_bh(lock); | |
626 | rcu_read_unlock(); | |
627 | ||
628 | return err; | |
629 | } | |
630 | ||
631 | /** | |
632 | * rhashtable_insert_fast - insert object into hash table | |
633 | * @ht: hash table | |
634 | * @obj: pointer to hash head inside object | |
635 | * @params: hash table parameters | |
636 | * | |
637 | * Will take a per bucket spinlock to protect against mutual mutations | |
638 | * on the same bucket. Multiple insertions may occur in parallel unless | |
639 | * they map to the same bucket lock. | |
640 | * | |
641 | * It is safe to call this function from atomic context. | |
642 | * | |
643 | * Will trigger an automatic deferred table resizing if the size grows | |
644 | * beyond the watermark indicated by grow_decision() which can be passed | |
645 | * to rhashtable_init(). | |
646 | */ | |
647 | static inline int rhashtable_insert_fast( | |
648 | struct rhashtable *ht, struct rhash_head *obj, | |
649 | const struct rhashtable_params params) | |
650 | { | |
651 | return __rhashtable_insert_fast(ht, NULL, obj, params); | |
652 | } | |
653 | ||
654 | /** | |
655 | * rhashtable_lookup_insert_fast - lookup and insert object into hash table | |
656 | * @ht: hash table | |
657 | * @obj: pointer to hash head inside object | |
658 | * @params: hash table parameters | |
659 | * | |
660 | * Locks down the bucket chain in both the old and new table if a resize | |
661 | * is in progress to ensure that writers can't remove from the old table | |
662 | * and can't insert to the new table during the atomic operation of search | |
663 | * and insertion. Searches for duplicates in both the old and new table if | |
664 | * a resize is in progress. | |
665 | * | |
666 | * This lookup function may only be used for fixed key hash table (key_len | |
667 | * parameter set). It will BUG() if used inappropriately. | |
668 | * | |
669 | * It is safe to call this function from atomic context. | |
670 | * | |
671 | * Will trigger an automatic deferred table resizing if the size grows | |
672 | * beyond the watermark indicated by grow_decision() which can be passed | |
673 | * to rhashtable_init(). | |
674 | */ | |
675 | static inline int rhashtable_lookup_insert_fast( | |
676 | struct rhashtable *ht, struct rhash_head *obj, | |
677 | const struct rhashtable_params params) | |
678 | { | |
679 | const char *key = rht_obj(ht, obj); | |
680 | ||
681 | BUG_ON(ht->p.obj_hashfn); | |
682 | ||
683 | return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, | |
684 | params); | |
685 | } | |
686 | ||
687 | /** | |
688 | * rhashtable_lookup_insert_key - search and insert object to hash table | |
689 | * with explicit key | |
690 | * @ht: hash table | |
691 | * @key: key | |
692 | * @obj: pointer to hash head inside object | |
693 | * @params: hash table parameters | |
694 | * | |
695 | * Locks down the bucket chain in both the old and new table if a resize | |
696 | * is in progress to ensure that writers can't remove from the old table | |
697 | * and can't insert to the new table during the atomic operation of search | |
698 | * and insertion. Searches for duplicates in both the old and new table if | |
699 | * a resize is in progress. | |
700 | * | |
701 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
702 | * | |
703 | * Will trigger an automatic deferred table resizing if the size grows | |
704 | * beyond the watermark indicated by grow_decision() which can be passed | |
705 | * to rhashtable_init(). | |
706 | * | |
707 | * Returns zero on success. | |
708 | */ | |
709 | static inline int rhashtable_lookup_insert_key( | |
710 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
711 | const struct rhashtable_params params) | |
712 | { | |
713 | BUG_ON(!ht->p.obj_hashfn || !key); | |
714 | ||
715 | return __rhashtable_insert_fast(ht, key, obj, params); | |
716 | } | |
717 | ||
ac833bdd | 718 | /* Internal function, please use rhashtable_remove_fast() instead */ |
02fd97c3 HX |
719 | static inline int __rhashtable_remove_fast( |
720 | struct rhashtable *ht, struct bucket_table *tbl, | |
721 | struct rhash_head *obj, const struct rhashtable_params params) | |
722 | { | |
723 | struct rhash_head __rcu **pprev; | |
724 | struct rhash_head *he; | |
725 | spinlock_t * lock; | |
299e5c32 | 726 | unsigned int hash; |
02fd97c3 HX |
727 | int err = -ENOENT; |
728 | ||
729 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
730 | lock = rht_bucket_lock(tbl, hash); | |
731 | ||
732 | spin_lock_bh(lock); | |
733 | ||
734 | pprev = &tbl->buckets[hash]; | |
735 | rht_for_each(he, tbl, hash) { | |
736 | if (he != obj) { | |
737 | pprev = &he->next; | |
738 | continue; | |
739 | } | |
740 | ||
741 | rcu_assign_pointer(*pprev, obj->next); | |
742 | err = 0; | |
743 | break; | |
744 | } | |
745 | ||
746 | spin_unlock_bh(lock); | |
747 | ||
748 | return err; | |
749 | } | |
750 | ||
751 | /** | |
752 | * rhashtable_remove_fast - remove object from hash table | |
753 | * @ht: hash table | |
754 | * @obj: pointer to hash head inside object | |
755 | * @params: hash table parameters | |
756 | * | |
757 | * Since the hash chain is single linked, the removal operation needs to | |
758 | * walk the bucket chain upon removal. The removal operation is thus | |
759 | * considerable slow if the hash table is not correctly sized. | |
760 | * | |
761 | * Will automatically shrink the table via rhashtable_expand() if the | |
762 | * shrink_decision function specified at rhashtable_init() returns true. | |
763 | * | |
764 | * Returns zero on success, -ENOENT if the entry could not be found. | |
765 | */ | |
766 | static inline int rhashtable_remove_fast( | |
767 | struct rhashtable *ht, struct rhash_head *obj, | |
768 | const struct rhashtable_params params) | |
769 | { | |
770 | struct bucket_table *tbl; | |
771 | int err; | |
772 | ||
773 | rcu_read_lock(); | |
774 | ||
775 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
776 | ||
777 | /* Because we have already taken (and released) the bucket | |
778 | * lock in old_tbl, if we find that future_tbl is not yet | |
779 | * visible then that guarantees the entry to still be in | |
780 | * the old tbl if it exists. | |
781 | */ | |
782 | while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) && | |
783 | (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) | |
784 | ; | |
785 | ||
786 | if (err) | |
787 | goto out; | |
788 | ||
789 | atomic_dec(&ht->nelems); | |
b5e2c150 TG |
790 | if (unlikely(ht->p.automatic_shrinking && |
791 | rht_shrink_below_30(ht, tbl))) | |
02fd97c3 HX |
792 | schedule_work(&ht->run_work); |
793 | ||
794 | out: | |
795 | rcu_read_unlock(); | |
796 | ||
797 | return err; | |
798 | } | |
799 | ||
7e1e7763 | 800 | #endif /* _LINUX_RHASHTABLE_H */ |