]>
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 | ||
07ee0722 | 20 | #include <linux/atomic.h> |
f2dba9c6 | 21 | #include <linux/compiler.h> |
3cf92222 | 22 | #include <linux/err.h> |
6626af69 | 23 | #include <linux/errno.h> |
31ccde2d | 24 | #include <linux/jhash.h> |
f89bd6f8 | 25 | #include <linux/list_nulls.h> |
97defe1e | 26 | #include <linux/workqueue.h> |
86b35b64 | 27 | #include <linux/mutex.h> |
02fd97c3 | 28 | #include <linux/rcupdate.h> |
7e1e7763 | 29 | |
f89bd6f8 TG |
30 | /* |
31 | * The end of the chain is marked with a special nulls marks which has | |
32 | * the following format: | |
33 | * | |
34 | * +-------+-----------------------------------------------------+-+ | |
35 | * | Base | Hash |1| | |
36 | * +-------+-----------------------------------------------------+-+ | |
37 | * | |
38 | * Base (4 bits) : Reserved to distinguish between multiple tables. | |
39 | * Specified via &struct rhashtable_params.nulls_base. | |
40 | * Hash (27 bits): Full hash (unmasked) of first element added to bucket | |
41 | * 1 (1 bit) : Nulls marker (always set) | |
42 | * | |
43 | * The remaining bits of the next pointer remain unused for now. | |
44 | */ | |
45 | #define RHT_BASE_BITS 4 | |
46 | #define RHT_HASH_BITS 27 | |
47 | #define RHT_BASE_SHIFT RHT_HASH_BITS | |
48 | ||
02fd97c3 HX |
49 | /* Base bits plus 1 bit for nulls marker */ |
50 | #define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1) | |
51 | ||
7e1e7763 | 52 | struct rhash_head { |
5300fdcb | 53 | struct rhash_head __rcu *next; |
7e1e7763 TG |
54 | }; |
55 | ||
97defe1e TG |
56 | /** |
57 | * struct bucket_table - Table of hash buckets | |
58 | * @size: Number of hash buckets | |
63d512d0 | 59 | * @rehash: Current bucket being rehashed |
988dfbd7 | 60 | * @hash_rnd: Random seed to fold into hash |
97defe1e TG |
61 | * @locks_mask: Mask to apply before accessing locks[] |
62 | * @locks: Array of spinlocks protecting individual buckets | |
eddee5ba | 63 | * @walkers: List of active walkers |
9d901bc0 | 64 | * @rcu: RCU structure for freeing the table |
c4db8848 | 65 | * @future_tbl: Table under construction during rehashing |
97defe1e TG |
66 | * @buckets: size * hash buckets |
67 | */ | |
7e1e7763 | 68 | struct bucket_table { |
63d512d0 HX |
69 | unsigned int size; |
70 | unsigned int rehash; | |
988dfbd7 | 71 | u32 hash_rnd; |
b9ebafbe ED |
72 | unsigned int locks_mask; |
73 | spinlock_t *locks; | |
eddee5ba | 74 | struct list_head walkers; |
9d901bc0 | 75 | struct rcu_head rcu; |
b9ebafbe | 76 | |
c4db8848 HX |
77 | struct bucket_table __rcu *future_tbl; |
78 | ||
b9ebafbe | 79 | struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp; |
7e1e7763 TG |
80 | }; |
81 | ||
02fd97c3 HX |
82 | /** |
83 | * struct rhashtable_compare_arg - Key for the function rhashtable_compare | |
84 | * @ht: Hash table | |
85 | * @key: Key to compare against | |
86 | */ | |
87 | struct rhashtable_compare_arg { | |
88 | struct rhashtable *ht; | |
89 | const void *key; | |
90 | }; | |
91 | ||
7e1e7763 | 92 | typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); |
49f7b33e | 93 | typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed); |
02fd97c3 HX |
94 | typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg, |
95 | const void *obj); | |
7e1e7763 TG |
96 | |
97 | struct rhashtable; | |
98 | ||
99 | /** | |
100 | * struct rhashtable_params - Hash table construction parameters | |
101 | * @nelem_hint: Hint on number of elements, should be 75% of desired size | |
102 | * @key_len: Length of key | |
103 | * @key_offset: Offset of key in struct to be hashed | |
104 | * @head_offset: Offset of rhash_head in struct to be hashed | |
07ee0722 | 105 | * @insecure_max_entries: Maximum number of entries (may be exceeded) |
c2e213cf HX |
106 | * @max_size: Maximum size while expanding |
107 | * @min_size: Minimum size while shrinking | |
f89bd6f8 | 108 | * @nulls_base: Base value to generate nulls marker |
ccd57b1b | 109 | * @insecure_elasticity: Set to true to disable chain length checks |
b5e2c150 | 110 | * @automatic_shrinking: Enable automatic shrinking of tables |
97defe1e | 111 | * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) |
31ccde2d | 112 | * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash) |
7e1e7763 | 113 | * @obj_hashfn: Function to hash object |
02fd97c3 | 114 | * @obj_cmpfn: Function to compare key with object |
7e1e7763 TG |
115 | */ |
116 | struct rhashtable_params { | |
117 | size_t nelem_hint; | |
118 | size_t key_len; | |
119 | size_t key_offset; | |
120 | size_t head_offset; | |
07ee0722 | 121 | unsigned int insecure_max_entries; |
c2e213cf HX |
122 | unsigned int max_size; |
123 | unsigned int min_size; | |
f89bd6f8 | 124 | u32 nulls_base; |
ccd57b1b | 125 | bool insecure_elasticity; |
b5e2c150 | 126 | bool automatic_shrinking; |
97defe1e | 127 | size_t locks_mul; |
7e1e7763 TG |
128 | rht_hashfn_t hashfn; |
129 | rht_obj_hashfn_t obj_hashfn; | |
02fd97c3 | 130 | rht_obj_cmpfn_t obj_cmpfn; |
7e1e7763 TG |
131 | }; |
132 | ||
133 | /** | |
134 | * struct rhashtable - Hash table handle | |
135 | * @tbl: Bucket table | |
136 | * @nelems: Number of elements in table | |
31ccde2d | 137 | * @key_len: Key length for hashfn |
ccd57b1b | 138 | * @elasticity: Maximum chain length before rehash |
7e1e7763 | 139 | * @p: Configuration parameters |
97defe1e TG |
140 | * @run_work: Deferred worker to expand/shrink asynchronously |
141 | * @mutex: Mutex to protect current/future table swapping | |
ba7c95ea | 142 | * @lock: Spin lock to protect walker list |
7e1e7763 TG |
143 | */ |
144 | struct rhashtable { | |
145 | struct bucket_table __rcu *tbl; | |
97defe1e | 146 | atomic_t nelems; |
31ccde2d | 147 | unsigned int key_len; |
ccd57b1b | 148 | unsigned int elasticity; |
7e1e7763 | 149 | struct rhashtable_params p; |
57699a40 | 150 | struct work_struct run_work; |
97defe1e | 151 | struct mutex mutex; |
ba7c95ea | 152 | spinlock_t lock; |
7e1e7763 TG |
153 | }; |
154 | ||
f2dba9c6 HX |
155 | /** |
156 | * struct rhashtable_walker - Hash table walker | |
157 | * @list: List entry on list of walkers | |
eddee5ba | 158 | * @tbl: The table that we were walking over |
f2dba9c6 HX |
159 | */ |
160 | struct rhashtable_walker { | |
161 | struct list_head list; | |
eddee5ba | 162 | struct bucket_table *tbl; |
f2dba9c6 HX |
163 | }; |
164 | ||
165 | /** | |
166 | * struct rhashtable_iter - Hash table iterator, fits into netlink cb | |
167 | * @ht: Table to iterate through | |
168 | * @p: Current pointer | |
169 | * @walker: Associated rhashtable walker | |
170 | * @slot: Current slot | |
171 | * @skip: Number of entries to skip in slot | |
172 | */ | |
173 | struct rhashtable_iter { | |
174 | struct rhashtable *ht; | |
175 | struct rhash_head *p; | |
246779dd | 176 | struct rhashtable_walker walker; |
f2dba9c6 HX |
177 | unsigned int slot; |
178 | unsigned int skip; | |
179 | }; | |
180 | ||
f89bd6f8 TG |
181 | static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) |
182 | { | |
183 | return NULLS_MARKER(ht->p.nulls_base + hash); | |
184 | } | |
185 | ||
186 | #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ | |
187 | ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) | |
188 | ||
189 | static inline bool rht_is_a_nulls(const struct rhash_head *ptr) | |
190 | { | |
191 | return ((unsigned long) ptr & 1); | |
192 | } | |
193 | ||
194 | static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) | |
195 | { | |
196 | return ((unsigned long) ptr) >> 1; | |
197 | } | |
198 | ||
02fd97c3 HX |
199 | static inline void *rht_obj(const struct rhashtable *ht, |
200 | const struct rhash_head *he) | |
201 | { | |
202 | return (char *)he - ht->p.head_offset; | |
203 | } | |
204 | ||
205 | static inline unsigned int rht_bucket_index(const struct bucket_table *tbl, | |
206 | unsigned int hash) | |
207 | { | |
208 | return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1); | |
209 | } | |
210 | ||
211 | static inline unsigned int rht_key_hashfn( | |
212 | struct rhashtable *ht, const struct bucket_table *tbl, | |
213 | const void *key, const struct rhashtable_params params) | |
214 | { | |
299e5c32 | 215 | unsigned int hash; |
de91b25c | 216 | |
31ccde2d HX |
217 | /* params must be equal to ht->p if it isn't constant. */ |
218 | if (!__builtin_constant_p(params.key_len)) | |
219 | hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd); | |
220 | else if (params.key_len) { | |
299e5c32 | 221 | unsigned int key_len = params.key_len; |
31ccde2d HX |
222 | |
223 | if (params.hashfn) | |
224 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
225 | else if (key_len & (sizeof(u32) - 1)) | |
226 | hash = jhash(key, key_len, tbl->hash_rnd); | |
227 | else | |
228 | hash = jhash2(key, key_len / sizeof(u32), | |
229 | tbl->hash_rnd); | |
230 | } else { | |
299e5c32 | 231 | unsigned int key_len = ht->p.key_len; |
31ccde2d HX |
232 | |
233 | if (params.hashfn) | |
234 | hash = params.hashfn(key, key_len, tbl->hash_rnd); | |
235 | else | |
236 | hash = jhash(key, key_len, tbl->hash_rnd); | |
237 | } | |
238 | ||
239 | return rht_bucket_index(tbl, hash); | |
02fd97c3 HX |
240 | } |
241 | ||
242 | static inline unsigned int rht_head_hashfn( | |
243 | struct rhashtable *ht, const struct bucket_table *tbl, | |
244 | const struct rhash_head *he, const struct rhashtable_params params) | |
245 | { | |
246 | const char *ptr = rht_obj(ht, he); | |
247 | ||
248 | return likely(params.obj_hashfn) ? | |
49f7b33e PM |
249 | rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?: |
250 | ht->p.key_len, | |
251 | tbl->hash_rnd)) : | |
02fd97c3 HX |
252 | rht_key_hashfn(ht, tbl, ptr + params.key_offset, params); |
253 | } | |
254 | ||
255 | /** | |
256 | * rht_grow_above_75 - returns true if nelems > 0.75 * table-size | |
257 | * @ht: hash table | |
258 | * @tbl: current table | |
259 | */ | |
260 | static inline bool rht_grow_above_75(const struct rhashtable *ht, | |
261 | const struct bucket_table *tbl) | |
262 | { | |
263 | /* Expand table when exceeding 75% load */ | |
264 | return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) && | |
265 | (!ht->p.max_size || tbl->size < ht->p.max_size); | |
266 | } | |
267 | ||
268 | /** | |
269 | * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size | |
270 | * @ht: hash table | |
271 | * @tbl: current table | |
272 | */ | |
273 | static inline bool rht_shrink_below_30(const struct rhashtable *ht, | |
274 | const struct bucket_table *tbl) | |
275 | { | |
276 | /* Shrink table beneath 30% load */ | |
277 | return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) && | |
278 | tbl->size > ht->p.min_size; | |
279 | } | |
280 | ||
ccd57b1b HX |
281 | /** |
282 | * rht_grow_above_100 - returns true if nelems > table-size | |
283 | * @ht: hash table | |
284 | * @tbl: current table | |
285 | */ | |
286 | static inline bool rht_grow_above_100(const struct rhashtable *ht, | |
287 | const struct bucket_table *tbl) | |
288 | { | |
1d8dc3d3 JB |
289 | return atomic_read(&ht->nelems) > tbl->size && |
290 | (!ht->p.max_size || tbl->size < ht->p.max_size); | |
ccd57b1b HX |
291 | } |
292 | ||
07ee0722 HX |
293 | /** |
294 | * rht_grow_above_max - returns true if table is above maximum | |
295 | * @ht: hash table | |
296 | * @tbl: current table | |
297 | */ | |
298 | static inline bool rht_grow_above_max(const struct rhashtable *ht, | |
299 | const struct bucket_table *tbl) | |
300 | { | |
301 | return ht->p.insecure_max_entries && | |
302 | atomic_read(&ht->nelems) >= ht->p.insecure_max_entries; | |
303 | } | |
304 | ||
02fd97c3 HX |
305 | /* The bucket lock is selected based on the hash and protects mutations |
306 | * on a group of hash buckets. | |
307 | * | |
308 | * A maximum of tbl->size/2 bucket locks is allocated. This ensures that | |
309 | * a single lock always covers both buckets which may both contains | |
310 | * entries which link to the same bucket of the old table during resizing. | |
311 | * This allows to simplify the locking as locking the bucket in both | |
312 | * tables during resize always guarantee protection. | |
313 | * | |
314 | * IMPORTANT: When holding the bucket lock of both the old and new table | |
315 | * during expansions and shrinking, the old bucket lock must always be | |
316 | * acquired first. | |
317 | */ | |
318 | static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl, | |
319 | unsigned int hash) | |
320 | { | |
321 | return &tbl->locks[hash & tbl->locks_mask]; | |
322 | } | |
323 | ||
7e1e7763 | 324 | #ifdef CONFIG_PROVE_LOCKING |
97defe1e | 325 | int lockdep_rht_mutex_is_held(struct rhashtable *ht); |
88d6ed15 | 326 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); |
7e1e7763 | 327 | #else |
97defe1e | 328 | static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 TG |
329 | { |
330 | return 1; | |
331 | } | |
88d6ed15 TG |
332 | |
333 | static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, | |
334 | u32 hash) | |
335 | { | |
336 | return 1; | |
337 | } | |
7e1e7763 TG |
338 | #endif /* CONFIG_PROVE_LOCKING */ |
339 | ||
488fb86e HX |
340 | int rhashtable_init(struct rhashtable *ht, |
341 | const struct rhashtable_params *params); | |
7e1e7763 | 342 | |
3cf92222 HX |
343 | struct bucket_table *rhashtable_insert_slow(struct rhashtable *ht, |
344 | const void *key, | |
345 | struct rhash_head *obj, | |
5ca8cc5b PNA |
346 | struct bucket_table *old_tbl, |
347 | void **data); | |
3cf92222 | 348 | int rhashtable_insert_rehash(struct rhashtable *ht, struct bucket_table *tbl); |
7e1e7763 | 349 | |
246779dd HX |
350 | void rhashtable_walk_enter(struct rhashtable *ht, |
351 | struct rhashtable_iter *iter); | |
f2dba9c6 HX |
352 | void rhashtable_walk_exit(struct rhashtable_iter *iter); |
353 | int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); | |
354 | void *rhashtable_walk_next(struct rhashtable_iter *iter); | |
355 | void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); | |
356 | ||
6b6f302c TG |
357 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
358 | void (*free_fn)(void *ptr, void *arg), | |
359 | void *arg); | |
97defe1e | 360 | void rhashtable_destroy(struct rhashtable *ht); |
7e1e7763 TG |
361 | |
362 | #define rht_dereference(p, ht) \ | |
363 | rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) | |
364 | ||
365 | #define rht_dereference_rcu(p, ht) \ | |
366 | rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) | |
367 | ||
88d6ed15 TG |
368 | #define rht_dereference_bucket(p, tbl, hash) \ |
369 | rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
7e1e7763 | 370 | |
88d6ed15 TG |
371 | #define rht_dereference_bucket_rcu(p, tbl, hash) \ |
372 | rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) | |
373 | ||
374 | #define rht_entry(tpos, pos, member) \ | |
375 | ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) | |
7e1e7763 TG |
376 | |
377 | /** | |
88d6ed15 TG |
378 | * rht_for_each_continue - continue iterating over hash chain |
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 | |
7e1e7763 | 383 | */ |
88d6ed15 TG |
384 | #define rht_for_each_continue(pos, head, tbl, hash) \ |
385 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 386 | !rht_is_a_nulls(pos); \ |
88d6ed15 TG |
387 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
388 | ||
389 | /** | |
390 | * rht_for_each - iterate over hash chain | |
391 | * @pos: the &struct rhash_head to use as a loop cursor. | |
392 | * @tbl: the &struct bucket_table | |
393 | * @hash: the hash value / bucket index | |
394 | */ | |
395 | #define rht_for_each(pos, tbl, hash) \ | |
396 | rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
397 | ||
398 | /** | |
399 | * rht_for_each_entry_continue - continue iterating over hash chain | |
400 | * @tpos: the type * to use as a loop cursor. | |
401 | * @pos: the &struct rhash_head to use as a loop cursor. | |
402 | * @head: the previous &struct rhash_head to continue from | |
403 | * @tbl: the &struct bucket_table | |
404 | * @hash: the hash value / bucket index | |
405 | * @member: name of the &struct rhash_head within the hashable struct. | |
406 | */ | |
407 | #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ | |
408 | for (pos = rht_dereference_bucket(head, tbl, hash); \ | |
f89bd6f8 | 409 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 410 | pos = rht_dereference_bucket((pos)->next, tbl, hash)) |
7e1e7763 TG |
411 | |
412 | /** | |
413 | * rht_for_each_entry - iterate over hash chain of given type | |
88d6ed15 TG |
414 | * @tpos: the type * to use as a loop cursor. |
415 | * @pos: the &struct rhash_head to use as a loop cursor. | |
416 | * @tbl: the &struct bucket_table | |
417 | * @hash: the hash value / bucket index | |
418 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 | 419 | */ |
88d6ed15 TG |
420 | #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ |
421 | rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \ | |
422 | tbl, hash, member) | |
7e1e7763 TG |
423 | |
424 | /** | |
425 | * rht_for_each_entry_safe - safely iterate over hash chain of given type | |
88d6ed15 TG |
426 | * @tpos: the type * to use as a loop cursor. |
427 | * @pos: the &struct rhash_head to use as a loop cursor. | |
428 | * @next: the &struct rhash_head to use as next in loop cursor. | |
429 | * @tbl: the &struct bucket_table | |
430 | * @hash: the hash value / bucket index | |
431 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
432 | * |
433 | * This hash chain list-traversal primitive allows for the looped code to | |
434 | * remove the loop cursor from the list. | |
435 | */ | |
88d6ed15 TG |
436 | #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ |
437 | for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \ | |
f89bd6f8 TG |
438 | next = !rht_is_a_nulls(pos) ? \ |
439 | rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ | |
440 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ | |
607954b0 PM |
441 | pos = next, \ |
442 | next = !rht_is_a_nulls(pos) ? \ | |
443 | rht_dereference_bucket(pos->next, tbl, hash) : NULL) | |
88d6ed15 TG |
444 | |
445 | /** | |
446 | * rht_for_each_rcu_continue - continue iterating over rcu hash chain | |
447 | * @pos: the &struct rhash_head to use as a loop cursor. | |
448 | * @head: the previous &struct rhash_head to continue from | |
449 | * @tbl: the &struct bucket_table | |
450 | * @hash: the hash value / bucket index | |
451 | * | |
452 | * This hash chain list-traversal primitive may safely run concurrently with | |
453 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
454 | * traversal is guarded by rcu_read_lock(). | |
455 | */ | |
456 | #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ | |
457 | for (({barrier(); }), \ | |
458 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 459 | !rht_is_a_nulls(pos); \ |
88d6ed15 | 460 | pos = rcu_dereference_raw(pos->next)) |
7e1e7763 TG |
461 | |
462 | /** | |
463 | * rht_for_each_rcu - iterate over rcu hash chain | |
88d6ed15 TG |
464 | * @pos: the &struct rhash_head to use as a loop cursor. |
465 | * @tbl: the &struct bucket_table | |
466 | * @hash: the hash value / bucket index | |
7e1e7763 TG |
467 | * |
468 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 469 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
470 | * traversal is guarded by rcu_read_lock(). |
471 | */ | |
88d6ed15 TG |
472 | #define rht_for_each_rcu(pos, tbl, hash) \ |
473 | rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash) | |
474 | ||
475 | /** | |
476 | * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain | |
477 | * @tpos: the type * to use as a loop cursor. | |
478 | * @pos: the &struct rhash_head to use as a loop cursor. | |
479 | * @head: the previous &struct rhash_head to continue from | |
480 | * @tbl: the &struct bucket_table | |
481 | * @hash: the hash value / bucket index | |
482 | * @member: name of the &struct rhash_head within the hashable struct. | |
483 | * | |
484 | * This hash chain list-traversal primitive may safely run concurrently with | |
485 | * the _rcu mutation primitives such as rhashtable_insert() as long as the | |
486 | * traversal is guarded by rcu_read_lock(). | |
487 | */ | |
488 | #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ | |
489 | for (({barrier(); }), \ | |
490 | pos = rht_dereference_bucket_rcu(head, tbl, hash); \ | |
f89bd6f8 | 491 | (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ |
88d6ed15 | 492 | pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) |
7e1e7763 TG |
493 | |
494 | /** | |
495 | * rht_for_each_entry_rcu - iterate over rcu hash chain of given type | |
88d6ed15 TG |
496 | * @tpos: the type * to use as a loop cursor. |
497 | * @pos: the &struct rhash_head to use as a loop cursor. | |
498 | * @tbl: the &struct bucket_table | |
499 | * @hash: the hash value / bucket index | |
500 | * @member: name of the &struct rhash_head within the hashable struct. | |
7e1e7763 TG |
501 | * |
502 | * This hash chain list-traversal primitive may safely run concurrently with | |
88d6ed15 | 503 | * the _rcu mutation primitives such as rhashtable_insert() as long as the |
7e1e7763 TG |
504 | * traversal is guarded by rcu_read_lock(). |
505 | */ | |
88d6ed15 TG |
506 | #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ |
507 | rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\ | |
508 | tbl, hash, member) | |
7e1e7763 | 509 | |
02fd97c3 HX |
510 | static inline int rhashtable_compare(struct rhashtable_compare_arg *arg, |
511 | const void *obj) | |
512 | { | |
513 | struct rhashtable *ht = arg->ht; | |
514 | const char *ptr = obj; | |
515 | ||
516 | return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len); | |
517 | } | |
518 | ||
519 | /** | |
520 | * rhashtable_lookup_fast - search hash table, inlined version | |
521 | * @ht: hash table | |
522 | * @key: the pointer to the key | |
523 | * @params: hash table parameters | |
524 | * | |
525 | * Computes the hash value for the key and traverses the bucket chain looking | |
526 | * for a entry with an identical key. The first matching entry is returned. | |
527 | * | |
528 | * Returns the first entry on which the compare function returned true. | |
529 | */ | |
530 | static inline void *rhashtable_lookup_fast( | |
531 | struct rhashtable *ht, const void *key, | |
532 | const struct rhashtable_params params) | |
533 | { | |
534 | struct rhashtable_compare_arg arg = { | |
535 | .ht = ht, | |
536 | .key = key, | |
537 | }; | |
538 | const struct bucket_table *tbl; | |
539 | struct rhash_head *he; | |
299e5c32 | 540 | unsigned int hash; |
02fd97c3 HX |
541 | |
542 | rcu_read_lock(); | |
543 | ||
544 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
545 | restart: | |
546 | hash = rht_key_hashfn(ht, tbl, key, params); | |
547 | rht_for_each_rcu(he, tbl, hash) { | |
548 | if (params.obj_cmpfn ? | |
549 | params.obj_cmpfn(&arg, rht_obj(ht, he)) : | |
550 | rhashtable_compare(&arg, rht_obj(ht, he))) | |
551 | continue; | |
552 | rcu_read_unlock(); | |
553 | return rht_obj(ht, he); | |
554 | } | |
555 | ||
556 | /* Ensure we see any new tables. */ | |
557 | smp_rmb(); | |
558 | ||
559 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
560 | if (unlikely(tbl)) | |
561 | goto restart; | |
562 | rcu_read_unlock(); | |
563 | ||
564 | return NULL; | |
565 | } | |
566 | ||
5ca8cc5b PNA |
567 | /* Internal function, please use rhashtable_insert_fast() instead. This |
568 | * function returns the existing element already in hashes in there is a clash, | |
569 | * otherwise it returns an error via ERR_PTR(). | |
570 | */ | |
571 | static inline void *__rhashtable_insert_fast( | |
02fd97c3 HX |
572 | struct rhashtable *ht, const void *key, struct rhash_head *obj, |
573 | const struct rhashtable_params params) | |
574 | { | |
575 | struct rhashtable_compare_arg arg = { | |
576 | .ht = ht, | |
577 | .key = key, | |
578 | }; | |
02fd97c3 HX |
579 | struct bucket_table *tbl, *new_tbl; |
580 | struct rhash_head *head; | |
581 | spinlock_t *lock; | |
299e5c32 TG |
582 | unsigned int elasticity; |
583 | unsigned int hash; | |
5ca8cc5b | 584 | void *data = NULL; |
ccd57b1b | 585 | int err; |
02fd97c3 | 586 | |
ccd57b1b | 587 | restart: |
02fd97c3 HX |
588 | rcu_read_lock(); |
589 | ||
590 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
02fd97c3 | 591 | |
b824478b HX |
592 | /* All insertions must grab the oldest table containing |
593 | * the hashed bucket that is yet to be rehashed. | |
02fd97c3 | 594 | */ |
b824478b HX |
595 | for (;;) { |
596 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
597 | lock = rht_bucket_lock(tbl, hash); | |
598 | spin_lock_bh(lock); | |
599 | ||
600 | if (tbl->rehash <= hash) | |
601 | break; | |
602 | ||
603 | spin_unlock_bh(lock); | |
604 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
605 | } | |
606 | ||
02fd97c3 HX |
607 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
608 | if (unlikely(new_tbl)) { | |
5ca8cc5b | 609 | tbl = rhashtable_insert_slow(ht, key, obj, new_tbl, &data); |
3cf92222 | 610 | if (!IS_ERR_OR_NULL(tbl)) |
ccd57b1b | 611 | goto slow_path; |
3cf92222 HX |
612 | |
613 | err = PTR_ERR(tbl); | |
5ca8cc5b PNA |
614 | if (err == -EEXIST) |
615 | err = 0; | |
616 | ||
02fd97c3 HX |
617 | goto out; |
618 | } | |
619 | ||
07ee0722 HX |
620 | err = -E2BIG; |
621 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
622 | goto out; | |
623 | ||
ccd57b1b HX |
624 | if (unlikely(rht_grow_above_100(ht, tbl))) { |
625 | slow_path: | |
626 | spin_unlock_bh(lock); | |
3cf92222 | 627 | err = rhashtable_insert_rehash(ht, tbl); |
58be8a58 | 628 | rcu_read_unlock(); |
ccd57b1b | 629 | if (err) |
5ca8cc5b | 630 | return ERR_PTR(err); |
ccd57b1b HX |
631 | |
632 | goto restart; | |
633 | } | |
02fd97c3 | 634 | |
5ca8cc5b | 635 | err = 0; |
ccd57b1b | 636 | elasticity = ht->elasticity; |
02fd97c3 | 637 | rht_for_each(head, tbl, hash) { |
ccd57b1b HX |
638 | if (key && |
639 | unlikely(!(params.obj_cmpfn ? | |
02fd97c3 | 640 | params.obj_cmpfn(&arg, rht_obj(ht, head)) : |
5ca8cc5b PNA |
641 | rhashtable_compare(&arg, rht_obj(ht, head))))) { |
642 | data = rht_obj(ht, head); | |
02fd97c3 | 643 | goto out; |
5ca8cc5b | 644 | } |
ccd57b1b HX |
645 | if (!--elasticity) |
646 | goto slow_path; | |
02fd97c3 HX |
647 | } |
648 | ||
02fd97c3 HX |
649 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); |
650 | ||
651 | RCU_INIT_POINTER(obj->next, head); | |
652 | ||
653 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
654 | ||
655 | atomic_inc(&ht->nelems); | |
656 | if (rht_grow_above_75(ht, tbl)) | |
657 | schedule_work(&ht->run_work); | |
658 | ||
659 | out: | |
660 | spin_unlock_bh(lock); | |
661 | rcu_read_unlock(); | |
662 | ||
5ca8cc5b | 663 | return err ? ERR_PTR(err) : data; |
02fd97c3 HX |
664 | } |
665 | ||
666 | /** | |
667 | * rhashtable_insert_fast - insert object into hash table | |
668 | * @ht: hash table | |
669 | * @obj: pointer to hash head inside object | |
670 | * @params: hash table parameters | |
671 | * | |
672 | * Will take a per bucket spinlock to protect against mutual mutations | |
673 | * on the same bucket. Multiple insertions may occur in parallel unless | |
674 | * they map to the same bucket lock. | |
675 | * | |
676 | * It is safe to call this function from atomic context. | |
677 | * | |
678 | * Will trigger an automatic deferred table resizing if the size grows | |
679 | * beyond the watermark indicated by grow_decision() which can be passed | |
680 | * to rhashtable_init(). | |
681 | */ | |
682 | static inline int rhashtable_insert_fast( | |
683 | struct rhashtable *ht, struct rhash_head *obj, | |
684 | const struct rhashtable_params params) | |
685 | { | |
5ca8cc5b PNA |
686 | void *ret; |
687 | ||
688 | ret = __rhashtable_insert_fast(ht, NULL, obj, params); | |
689 | if (IS_ERR(ret)) | |
690 | return PTR_ERR(ret); | |
691 | ||
692 | return ret == NULL ? 0 : -EEXIST; | |
02fd97c3 HX |
693 | } |
694 | ||
695 | /** | |
696 | * rhashtable_lookup_insert_fast - lookup and insert object into hash table | |
697 | * @ht: hash table | |
698 | * @obj: pointer to hash head inside object | |
699 | * @params: hash table parameters | |
700 | * | |
701 | * Locks down the bucket chain in both the old and new table if a resize | |
702 | * is in progress to ensure that writers can't remove from the old table | |
703 | * and can't insert to the new table during the atomic operation of search | |
704 | * and insertion. Searches for duplicates in both the old and new table if | |
705 | * a resize is in progress. | |
706 | * | |
707 | * This lookup function may only be used for fixed key hash table (key_len | |
708 | * parameter set). It will BUG() if used inappropriately. | |
709 | * | |
710 | * It is safe to call this function from atomic context. | |
711 | * | |
712 | * Will trigger an automatic deferred table resizing if the size grows | |
713 | * beyond the watermark indicated by grow_decision() which can be passed | |
714 | * to rhashtable_init(). | |
715 | */ | |
716 | static inline int rhashtable_lookup_insert_fast( | |
717 | struct rhashtable *ht, struct rhash_head *obj, | |
718 | const struct rhashtable_params params) | |
719 | { | |
720 | const char *key = rht_obj(ht, obj); | |
5ca8cc5b | 721 | void *ret; |
02fd97c3 HX |
722 | |
723 | BUG_ON(ht->p.obj_hashfn); | |
724 | ||
5ca8cc5b PNA |
725 | ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params); |
726 | if (IS_ERR(ret)) | |
727 | return PTR_ERR(ret); | |
728 | ||
729 | return ret == NULL ? 0 : -EEXIST; | |
02fd97c3 HX |
730 | } |
731 | ||
732 | /** | |
733 | * rhashtable_lookup_insert_key - search and insert object to hash table | |
734 | * with explicit key | |
735 | * @ht: hash table | |
736 | * @key: key | |
737 | * @obj: pointer to hash head inside object | |
738 | * @params: hash table parameters | |
739 | * | |
740 | * Locks down the bucket chain in both the old and new table if a resize | |
741 | * is in progress to ensure that writers can't remove from the old table | |
742 | * and can't insert to the new table during the atomic operation of search | |
743 | * and insertion. Searches for duplicates in both the old and new table if | |
744 | * a resize is in progress. | |
745 | * | |
746 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
747 | * | |
748 | * Will trigger an automatic deferred table resizing if the size grows | |
749 | * beyond the watermark indicated by grow_decision() which can be passed | |
750 | * to rhashtable_init(). | |
751 | * | |
752 | * Returns zero on success. | |
753 | */ | |
754 | static inline int rhashtable_lookup_insert_key( | |
755 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
756 | const struct rhashtable_params params) | |
5ca8cc5b PNA |
757 | { |
758 | void *ret; | |
759 | ||
760 | BUG_ON(!ht->p.obj_hashfn || !key); | |
761 | ||
762 | ret = __rhashtable_insert_fast(ht, key, obj, params); | |
763 | if (IS_ERR(ret)) | |
764 | return PTR_ERR(ret); | |
765 | ||
766 | return ret == NULL ? 0 : -EEXIST; | |
767 | } | |
768 | ||
769 | /** | |
770 | * rhashtable_lookup_get_insert_key - lookup and insert object into hash table | |
771 | * @ht: hash table | |
772 | * @obj: pointer to hash head inside object | |
773 | * @params: hash table parameters | |
774 | * @data: pointer to element data already in hashes | |
775 | * | |
776 | * Just like rhashtable_lookup_insert_key(), but this function returns the | |
777 | * object if it exists, NULL if it does not and the insertion was successful, | |
778 | * and an ERR_PTR otherwise. | |
779 | */ | |
780 | static inline void *rhashtable_lookup_get_insert_key( | |
781 | struct rhashtable *ht, const void *key, struct rhash_head *obj, | |
782 | const struct rhashtable_params params) | |
02fd97c3 HX |
783 | { |
784 | BUG_ON(!ht->p.obj_hashfn || !key); | |
785 | ||
786 | return __rhashtable_insert_fast(ht, key, obj, params); | |
787 | } | |
788 | ||
ac833bdd | 789 | /* Internal function, please use rhashtable_remove_fast() instead */ |
02fd97c3 HX |
790 | static inline int __rhashtable_remove_fast( |
791 | struct rhashtable *ht, struct bucket_table *tbl, | |
792 | struct rhash_head *obj, const struct rhashtable_params params) | |
793 | { | |
794 | struct rhash_head __rcu **pprev; | |
795 | struct rhash_head *he; | |
796 | spinlock_t * lock; | |
299e5c32 | 797 | unsigned int hash; |
02fd97c3 HX |
798 | int err = -ENOENT; |
799 | ||
800 | hash = rht_head_hashfn(ht, tbl, obj, params); | |
801 | lock = rht_bucket_lock(tbl, hash); | |
802 | ||
803 | spin_lock_bh(lock); | |
804 | ||
805 | pprev = &tbl->buckets[hash]; | |
806 | rht_for_each(he, tbl, hash) { | |
807 | if (he != obj) { | |
808 | pprev = &he->next; | |
809 | continue; | |
810 | } | |
811 | ||
812 | rcu_assign_pointer(*pprev, obj->next); | |
813 | err = 0; | |
814 | break; | |
815 | } | |
816 | ||
817 | spin_unlock_bh(lock); | |
818 | ||
819 | return err; | |
820 | } | |
821 | ||
822 | /** | |
823 | * rhashtable_remove_fast - remove object from hash table | |
824 | * @ht: hash table | |
825 | * @obj: pointer to hash head inside object | |
826 | * @params: hash table parameters | |
827 | * | |
828 | * Since the hash chain is single linked, the removal operation needs to | |
829 | * walk the bucket chain upon removal. The removal operation is thus | |
830 | * considerable slow if the hash table is not correctly sized. | |
831 | * | |
832 | * Will automatically shrink the table via rhashtable_expand() if the | |
833 | * shrink_decision function specified at rhashtable_init() returns true. | |
834 | * | |
835 | * Returns zero on success, -ENOENT if the entry could not be found. | |
836 | */ | |
837 | static inline int rhashtable_remove_fast( | |
838 | struct rhashtable *ht, struct rhash_head *obj, | |
839 | const struct rhashtable_params params) | |
840 | { | |
841 | struct bucket_table *tbl; | |
842 | int err; | |
843 | ||
844 | rcu_read_lock(); | |
845 | ||
846 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
847 | ||
848 | /* Because we have already taken (and released) the bucket | |
849 | * lock in old_tbl, if we find that future_tbl is not yet | |
850 | * visible then that guarantees the entry to still be in | |
851 | * the old tbl if it exists. | |
852 | */ | |
853 | while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) && | |
854 | (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) | |
855 | ; | |
856 | ||
857 | if (err) | |
858 | goto out; | |
859 | ||
860 | atomic_dec(&ht->nelems); | |
b5e2c150 TG |
861 | if (unlikely(ht->p.automatic_shrinking && |
862 | rht_shrink_below_30(ht, tbl))) | |
02fd97c3 HX |
863 | schedule_work(&ht->run_work); |
864 | ||
865 | out: | |
866 | rcu_read_unlock(); | |
867 | ||
868 | return err; | |
869 | } | |
870 | ||
3502cad7 TH |
871 | /* Internal function, please use rhashtable_replace_fast() instead */ |
872 | static inline int __rhashtable_replace_fast( | |
873 | struct rhashtable *ht, struct bucket_table *tbl, | |
874 | struct rhash_head *obj_old, struct rhash_head *obj_new, | |
875 | const struct rhashtable_params params) | |
876 | { | |
877 | struct rhash_head __rcu **pprev; | |
878 | struct rhash_head *he; | |
879 | spinlock_t *lock; | |
880 | unsigned int hash; | |
881 | int err = -ENOENT; | |
882 | ||
883 | /* Minimally, the old and new objects must have same hash | |
884 | * (which should mean identifiers are the same). | |
885 | */ | |
886 | hash = rht_head_hashfn(ht, tbl, obj_old, params); | |
887 | if (hash != rht_head_hashfn(ht, tbl, obj_new, params)) | |
888 | return -EINVAL; | |
889 | ||
890 | lock = rht_bucket_lock(tbl, hash); | |
891 | ||
892 | spin_lock_bh(lock); | |
893 | ||
894 | pprev = &tbl->buckets[hash]; | |
895 | rht_for_each(he, tbl, hash) { | |
896 | if (he != obj_old) { | |
897 | pprev = &he->next; | |
898 | continue; | |
899 | } | |
900 | ||
901 | rcu_assign_pointer(obj_new->next, obj_old->next); | |
902 | rcu_assign_pointer(*pprev, obj_new); | |
903 | err = 0; | |
904 | break; | |
905 | } | |
906 | ||
907 | spin_unlock_bh(lock); | |
908 | ||
909 | return err; | |
910 | } | |
911 | ||
912 | /** | |
913 | * rhashtable_replace_fast - replace an object in hash table | |
914 | * @ht: hash table | |
915 | * @obj_old: pointer to hash head inside object being replaced | |
916 | * @obj_new: pointer to hash head inside object which is new | |
917 | * @params: hash table parameters | |
918 | * | |
919 | * Replacing an object doesn't affect the number of elements in the hash table | |
920 | * or bucket, so we don't need to worry about shrinking or expanding the | |
921 | * table here. | |
922 | * | |
923 | * Returns zero on success, -ENOENT if the entry could not be found, | |
924 | * -EINVAL if hash is not the same for the old and new objects. | |
925 | */ | |
926 | static inline int rhashtable_replace_fast( | |
927 | struct rhashtable *ht, struct rhash_head *obj_old, | |
928 | struct rhash_head *obj_new, | |
929 | const struct rhashtable_params params) | |
930 | { | |
931 | struct bucket_table *tbl; | |
932 | int err; | |
933 | ||
934 | rcu_read_lock(); | |
935 | ||
936 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
937 | ||
938 | /* Because we have already taken (and released) the bucket | |
939 | * lock in old_tbl, if we find that future_tbl is not yet | |
940 | * visible then that guarantees the entry to still be in | |
941 | * the old tbl if it exists. | |
942 | */ | |
943 | while ((err = __rhashtable_replace_fast(ht, tbl, obj_old, | |
944 | obj_new, params)) && | |
945 | (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) | |
946 | ; | |
947 | ||
948 | rcu_read_unlock(); | |
949 | ||
950 | return err; | |
951 | } | |
952 | ||
246779dd HX |
953 | /* Obsolete function, do not use in new code. */ |
954 | static inline int rhashtable_walk_init(struct rhashtable *ht, | |
955 | struct rhashtable_iter *iter, gfp_t gfp) | |
956 | { | |
957 | rhashtable_walk_enter(ht, iter); | |
958 | return 0; | |
959 | } | |
960 | ||
7e1e7763 | 961 | #endif /* _LINUX_RHASHTABLE_H */ |