]>
Commit | Line | Data |
---|---|---|
7e1e7763 TG |
1 | /* |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
a5ec68e3 | 4 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> |
7e1e7763 TG |
5 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> |
6 | * | |
7 | * Based on the following paper: | |
8 | * https://www.usenix.org/legacy/event/atc11/tech/final_files/Triplett.pdf | |
9 | * | |
10 | * Code partially derived from nft_hash | |
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 | #include <linux/kernel.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/log2.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/vmalloc.h> | |
22 | #include <linux/mm.h> | |
87545899 | 23 | #include <linux/jhash.h> |
7e1e7763 TG |
24 | #include <linux/random.h> |
25 | #include <linux/rhashtable.h> | |
7e1e7763 TG |
26 | |
27 | #define HASH_DEFAULT_SIZE 64UL | |
28 | #define HASH_MIN_SIZE 4UL | |
97defe1e TG |
29 | #define BUCKET_LOCKS_PER_CPU 128UL |
30 | ||
f89bd6f8 TG |
31 | /* Base bits plus 1 bit for nulls marker */ |
32 | #define HASH_RESERVED_SPACE (RHT_BASE_BITS + 1) | |
33 | ||
97defe1e TG |
34 | enum { |
35 | RHT_LOCK_NORMAL, | |
36 | RHT_LOCK_NESTED, | |
97defe1e TG |
37 | }; |
38 | ||
39 | /* The bucket lock is selected based on the hash and protects mutations | |
40 | * on a group of hash buckets. | |
41 | * | |
a5ec68e3 TG |
42 | * A maximum of tbl->size/2 bucket locks is allocated. This ensures that |
43 | * a single lock always covers both buckets which may both contains | |
44 | * entries which link to the same bucket of the old table during resizing. | |
45 | * This allows to simplify the locking as locking the bucket in both | |
46 | * tables during resize always guarantee protection. | |
47 | * | |
97defe1e TG |
48 | * IMPORTANT: When holding the bucket lock of both the old and new table |
49 | * during expansions and shrinking, the old bucket lock must always be | |
50 | * acquired first. | |
51 | */ | |
52 | static spinlock_t *bucket_lock(const struct bucket_table *tbl, u32 hash) | |
53 | { | |
54 | return &tbl->locks[hash & tbl->locks_mask]; | |
55 | } | |
7e1e7763 TG |
56 | |
57 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) | |
97defe1e TG |
58 | #define ASSERT_BUCKET_LOCK(TBL, HASH) \ |
59 | BUG_ON(!lockdep_rht_bucket_is_held(TBL, HASH)) | |
7e1e7763 TG |
60 | |
61 | #ifdef CONFIG_PROVE_LOCKING | |
97defe1e | 62 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
7e1e7763 | 63 | { |
97defe1e | 64 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; |
7e1e7763 TG |
65 | } |
66 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
88d6ed15 TG |
67 | |
68 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
69 | { | |
97defe1e TG |
70 | spinlock_t *lock = bucket_lock(tbl, hash); |
71 | ||
72 | return (debug_locks) ? lockdep_is_held(lock) : 1; | |
88d6ed15 TG |
73 | } |
74 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
7e1e7763 TG |
75 | #endif |
76 | ||
c91eee56 | 77 | static void *rht_obj(const struct rhashtable *ht, const struct rhash_head *he) |
7e1e7763 TG |
78 | { |
79 | return (void *) he - ht->p.head_offset; | |
80 | } | |
7e1e7763 | 81 | |
8d24c0b4 | 82 | static u32 rht_bucket_index(const struct bucket_table *tbl, u32 hash) |
7e1e7763 | 83 | { |
8d24c0b4 | 84 | return hash & (tbl->size - 1); |
7e1e7763 | 85 | } |
7e1e7763 | 86 | |
8d24c0b4 | 87 | static u32 obj_raw_hashfn(const struct rhashtable *ht, const void *ptr) |
7e1e7763 | 88 | { |
8d24c0b4 | 89 | u32 hash; |
7e1e7763 | 90 | |
8d24c0b4 TG |
91 | if (unlikely(!ht->p.key_len)) |
92 | hash = ht->p.obj_hashfn(ptr, ht->p.hash_rnd); | |
93 | else | |
94 | hash = ht->p.hashfn(ptr + ht->p.key_offset, ht->p.key_len, | |
95 | ht->p.hash_rnd); | |
7e1e7763 | 96 | |
f89bd6f8 | 97 | return hash >> HASH_RESERVED_SPACE; |
7e1e7763 TG |
98 | } |
99 | ||
97defe1e | 100 | static u32 key_hashfn(struct rhashtable *ht, const void *key, u32 len) |
7e1e7763 | 101 | { |
c88455ce | 102 | return ht->p.hashfn(key, len, ht->p.hash_rnd) >> HASH_RESERVED_SPACE; |
7e1e7763 | 103 | } |
7e1e7763 TG |
104 | |
105 | static u32 head_hashfn(const struct rhashtable *ht, | |
8d24c0b4 TG |
106 | const struct bucket_table *tbl, |
107 | const struct rhash_head *he) | |
7e1e7763 | 108 | { |
8d24c0b4 | 109 | return rht_bucket_index(tbl, obj_raw_hashfn(ht, rht_obj(ht, he))); |
7e1e7763 TG |
110 | } |
111 | ||
b8e1943e TG |
112 | static struct rhash_head __rcu **bucket_tail(struct bucket_table *tbl, u32 n) |
113 | { | |
114 | struct rhash_head __rcu **pprev; | |
115 | ||
116 | for (pprev = &tbl->buckets[n]; | |
f89bd6f8 | 117 | !rht_is_a_nulls(rht_dereference_bucket(*pprev, tbl, n)); |
b8e1943e TG |
118 | pprev = &rht_dereference_bucket(*pprev, tbl, n)->next) |
119 | ; | |
120 | ||
121 | return pprev; | |
122 | } | |
123 | ||
97defe1e TG |
124 | static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl) |
125 | { | |
126 | unsigned int i, size; | |
127 | #if defined(CONFIG_PROVE_LOCKING) | |
128 | unsigned int nr_pcpus = 2; | |
129 | #else | |
130 | unsigned int nr_pcpus = num_possible_cpus(); | |
131 | #endif | |
132 | ||
133 | nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL); | |
134 | size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul); | |
135 | ||
a5ec68e3 TG |
136 | /* Never allocate more than 0.5 locks per bucket */ |
137 | size = min_t(unsigned int, size, tbl->size >> 1); | |
97defe1e TG |
138 | |
139 | if (sizeof(spinlock_t) != 0) { | |
140 | #ifdef CONFIG_NUMA | |
141 | if (size * sizeof(spinlock_t) > PAGE_SIZE) | |
142 | tbl->locks = vmalloc(size * sizeof(spinlock_t)); | |
143 | else | |
144 | #endif | |
145 | tbl->locks = kmalloc_array(size, sizeof(spinlock_t), | |
146 | GFP_KERNEL); | |
147 | if (!tbl->locks) | |
148 | return -ENOMEM; | |
149 | for (i = 0; i < size; i++) | |
150 | spin_lock_init(&tbl->locks[i]); | |
151 | } | |
152 | tbl->locks_mask = size - 1; | |
153 | ||
154 | return 0; | |
155 | } | |
156 | ||
157 | static void bucket_table_free(const struct bucket_table *tbl) | |
158 | { | |
159 | if (tbl) | |
160 | kvfree(tbl->locks); | |
161 | ||
162 | kvfree(tbl); | |
163 | } | |
164 | ||
165 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, | |
166 | size_t nbuckets) | |
7e1e7763 TG |
167 | { |
168 | struct bucket_table *tbl; | |
169 | size_t size; | |
f89bd6f8 | 170 | int i; |
7e1e7763 TG |
171 | |
172 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
6eba8224 | 173 | tbl = kzalloc(size, GFP_KERNEL | __GFP_NOWARN); |
7e1e7763 TG |
174 | if (tbl == NULL) |
175 | tbl = vzalloc(size); | |
176 | ||
177 | if (tbl == NULL) | |
178 | return NULL; | |
179 | ||
180 | tbl->size = nbuckets; | |
181 | ||
97defe1e TG |
182 | if (alloc_bucket_locks(ht, tbl) < 0) { |
183 | bucket_table_free(tbl); | |
184 | return NULL; | |
185 | } | |
7e1e7763 | 186 | |
f89bd6f8 TG |
187 | for (i = 0; i < nbuckets; i++) |
188 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); | |
189 | ||
97defe1e | 190 | return tbl; |
7e1e7763 TG |
191 | } |
192 | ||
193 | /** | |
194 | * rht_grow_above_75 - returns true if nelems > 0.75 * table-size | |
195 | * @ht: hash table | |
196 | * @new_size: new table size | |
197 | */ | |
198 | bool rht_grow_above_75(const struct rhashtable *ht, size_t new_size) | |
199 | { | |
200 | /* Expand table when exceeding 75% load */ | |
c0c09bfd YX |
201 | return atomic_read(&ht->nelems) > (new_size / 4 * 3) && |
202 | (ht->p.max_shift && atomic_read(&ht->shift) < ht->p.max_shift); | |
7e1e7763 TG |
203 | } |
204 | EXPORT_SYMBOL_GPL(rht_grow_above_75); | |
205 | ||
206 | /** | |
207 | * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size | |
208 | * @ht: hash table | |
209 | * @new_size: new table size | |
210 | */ | |
211 | bool rht_shrink_below_30(const struct rhashtable *ht, size_t new_size) | |
212 | { | |
213 | /* Shrink table beneath 30% load */ | |
c0c09bfd YX |
214 | return atomic_read(&ht->nelems) < (new_size * 3 / 10) && |
215 | (atomic_read(&ht->shift) > ht->p.min_shift); | |
7e1e7763 TG |
216 | } |
217 | EXPORT_SYMBOL_GPL(rht_shrink_below_30); | |
218 | ||
a5ec68e3 TG |
219 | static void lock_buckets(struct bucket_table *new_tbl, |
220 | struct bucket_table *old_tbl, unsigned int hash) | |
221 | __acquires(old_bucket_lock) | |
222 | { | |
223 | spin_lock_bh(bucket_lock(old_tbl, hash)); | |
224 | if (new_tbl != old_tbl) | |
225 | spin_lock_bh_nested(bucket_lock(new_tbl, hash), | |
226 | RHT_LOCK_NESTED); | |
227 | } | |
228 | ||
229 | static void unlock_buckets(struct bucket_table *new_tbl, | |
230 | struct bucket_table *old_tbl, unsigned int hash) | |
231 | __releases(old_bucket_lock) | |
232 | { | |
233 | if (new_tbl != old_tbl) | |
234 | spin_unlock_bh(bucket_lock(new_tbl, hash)); | |
235 | spin_unlock_bh(bucket_lock(old_tbl, hash)); | |
236 | } | |
237 | ||
238 | /** | |
239 | * Unlink entries on bucket which hash to different bucket. | |
240 | * | |
241 | * Returns true if no more work needs to be performed on the bucket. | |
242 | */ | |
243 | static bool hashtable_chain_unzip(const struct rhashtable *ht, | |
7e1e7763 | 244 | const struct bucket_table *new_tbl, |
97defe1e TG |
245 | struct bucket_table *old_tbl, |
246 | size_t old_hash) | |
7e1e7763 TG |
247 | { |
248 | struct rhash_head *he, *p, *next; | |
97defe1e TG |
249 | unsigned int new_hash, new_hash2; |
250 | ||
251 | ASSERT_BUCKET_LOCK(old_tbl, old_hash); | |
7e1e7763 TG |
252 | |
253 | /* Old bucket empty, no work needed. */ | |
97defe1e TG |
254 | p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl, |
255 | old_hash); | |
f89bd6f8 | 256 | if (rht_is_a_nulls(p)) |
a5ec68e3 | 257 | return false; |
7e1e7763 | 258 | |
a5ec68e3 TG |
259 | new_hash = head_hashfn(ht, new_tbl, p); |
260 | ASSERT_BUCKET_LOCK(new_tbl, new_hash); | |
97defe1e | 261 | |
7e1e7763 TG |
262 | /* Advance the old bucket pointer one or more times until it |
263 | * reaches a node that doesn't hash to the same bucket as the | |
264 | * previous node p. Call the previous node p; | |
265 | */ | |
97defe1e TG |
266 | rht_for_each_continue(he, p->next, old_tbl, old_hash) { |
267 | new_hash2 = head_hashfn(ht, new_tbl, he); | |
a5ec68e3 TG |
268 | ASSERT_BUCKET_LOCK(new_tbl, new_hash2); |
269 | ||
97defe1e | 270 | if (new_hash != new_hash2) |
7e1e7763 TG |
271 | break; |
272 | p = he; | |
273 | } | |
97defe1e TG |
274 | rcu_assign_pointer(old_tbl->buckets[old_hash], p->next); |
275 | ||
7e1e7763 TG |
276 | /* Find the subsequent node which does hash to the same |
277 | * bucket as node P, or NULL if no such node exists. | |
278 | */ | |
f89bd6f8 TG |
279 | INIT_RHT_NULLS_HEAD(next, ht, old_hash); |
280 | if (!rht_is_a_nulls(he)) { | |
97defe1e TG |
281 | rht_for_each_continue(he, he->next, old_tbl, old_hash) { |
282 | if (head_hashfn(ht, new_tbl, he) == new_hash) { | |
7e1e7763 TG |
283 | next = he; |
284 | break; | |
285 | } | |
286 | } | |
287 | } | |
288 | ||
289 | /* Set p's next pointer to that subsequent node pointer, | |
290 | * bypassing the nodes which do not hash to p's bucket | |
291 | */ | |
97defe1e TG |
292 | rcu_assign_pointer(p->next, next); |
293 | ||
a5ec68e3 TG |
294 | p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl, |
295 | old_hash); | |
296 | ||
297 | return !rht_is_a_nulls(p); | |
97defe1e TG |
298 | } |
299 | ||
300 | static void link_old_to_new(struct bucket_table *new_tbl, | |
301 | unsigned int new_hash, struct rhash_head *entry) | |
302 | { | |
97defe1e | 303 | rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), entry); |
7e1e7763 TG |
304 | } |
305 | ||
306 | /** | |
307 | * rhashtable_expand - Expand hash table while allowing concurrent lookups | |
308 | * @ht: the hash table to expand | |
7e1e7763 TG |
309 | * |
310 | * A secondary bucket array is allocated and the hash entries are migrated | |
311 | * while keeping them on both lists until the end of the RCU grace period. | |
312 | * | |
313 | * This function may only be called in a context where it is safe to call | |
314 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. | |
315 | * | |
97defe1e TG |
316 | * The caller must ensure that no concurrent resizing occurs by holding |
317 | * ht->mutex. | |
318 | * | |
319 | * It is valid to have concurrent insertions and deletions protected by per | |
320 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 321 | */ |
6eba8224 | 322 | int rhashtable_expand(struct rhashtable *ht) |
7e1e7763 TG |
323 | { |
324 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); | |
325 | struct rhash_head *he; | |
97defe1e TG |
326 | unsigned int new_hash, old_hash; |
327 | bool complete = false; | |
7e1e7763 TG |
328 | |
329 | ASSERT_RHT_MUTEX(ht); | |
330 | ||
97defe1e | 331 | new_tbl = bucket_table_alloc(ht, old_tbl->size * 2); |
7e1e7763 TG |
332 | if (new_tbl == NULL) |
333 | return -ENOMEM; | |
334 | ||
c0c09bfd | 335 | atomic_inc(&ht->shift); |
7e1e7763 | 336 | |
97defe1e TG |
337 | /* Make insertions go into the new, empty table right away. Deletions |
338 | * and lookups will be attempted in both tables until we synchronize. | |
339 | * The synchronize_rcu() guarantees for the new table to be picked up | |
340 | * so no new additions go into the old table while we relink. | |
341 | */ | |
342 | rcu_assign_pointer(ht->future_tbl, new_tbl); | |
343 | synchronize_rcu(); | |
344 | ||
345 | /* For each new bucket, search the corresponding old bucket for the | |
346 | * first entry that hashes to the new bucket, and link the end of | |
347 | * newly formed bucket chain (containing entries added to future | |
348 | * table) to that entry. Since all the entries which will end up in | |
349 | * the new bucket appear in the same old bucket, this constructs an | |
350 | * entirely valid new hash table, but with multiple buckets | |
351 | * "zipped" together into a single imprecise chain. | |
7e1e7763 | 352 | */ |
97defe1e TG |
353 | for (new_hash = 0; new_hash < new_tbl->size; new_hash++) { |
354 | old_hash = rht_bucket_index(old_tbl, new_hash); | |
a5ec68e3 | 355 | lock_buckets(new_tbl, old_tbl, new_hash); |
97defe1e TG |
356 | rht_for_each(he, old_tbl, old_hash) { |
357 | if (head_hashfn(ht, new_tbl, he) == new_hash) { | |
358 | link_old_to_new(new_tbl, new_hash, he); | |
7e1e7763 TG |
359 | break; |
360 | } | |
361 | } | |
a5ec68e3 | 362 | unlock_buckets(new_tbl, old_tbl, new_hash); |
7e1e7763 TG |
363 | } |
364 | ||
365 | /* Publish the new table pointer. Lookups may now traverse | |
0c828f2f HX |
366 | * the new table, but they will not benefit from any |
367 | * additional efficiency until later steps unzip the buckets. | |
7e1e7763 TG |
368 | */ |
369 | rcu_assign_pointer(ht->tbl, new_tbl); | |
370 | ||
371 | /* Unzip interleaved hash chains */ | |
97defe1e | 372 | while (!complete && !ht->being_destroyed) { |
7e1e7763 TG |
373 | /* Wait for readers. All new readers will see the new |
374 | * table, and thus no references to the old table will | |
375 | * remain. | |
376 | */ | |
377 | synchronize_rcu(); | |
378 | ||
379 | /* For each bucket in the old table (each of which | |
380 | * contains items from multiple buckets of the new | |
381 | * table): ... | |
382 | */ | |
383 | complete = true; | |
97defe1e | 384 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
a5ec68e3 | 385 | lock_buckets(new_tbl, old_tbl, old_hash); |
97defe1e | 386 | |
a5ec68e3 TG |
387 | if (hashtable_chain_unzip(ht, new_tbl, old_tbl, |
388 | old_hash)) | |
7e1e7763 | 389 | complete = false; |
97defe1e | 390 | |
a5ec68e3 | 391 | unlock_buckets(new_tbl, old_tbl, old_hash); |
7e1e7763 | 392 | } |
97defe1e | 393 | } |
7e1e7763 | 394 | |
2af4b529 TG |
395 | synchronize_rcu(); |
396 | ||
7e1e7763 TG |
397 | bucket_table_free(old_tbl); |
398 | return 0; | |
399 | } | |
400 | EXPORT_SYMBOL_GPL(rhashtable_expand); | |
401 | ||
402 | /** | |
403 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
404 | * @ht: the hash table to shrink | |
7e1e7763 TG |
405 | * |
406 | * This function may only be called in a context where it is safe to call | |
407 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. | |
408 | * | |
97defe1e TG |
409 | * The caller must ensure that no concurrent resizing occurs by holding |
410 | * ht->mutex. | |
411 | * | |
7e1e7763 TG |
412 | * The caller must ensure that no concurrent table mutations take place. |
413 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
414 | * |
415 | * It is valid to have concurrent insertions and deletions protected by per | |
416 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 417 | */ |
6eba8224 | 418 | int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 419 | { |
97defe1e | 420 | struct bucket_table *new_tbl, *tbl = rht_dereference(ht->tbl, ht); |
97defe1e | 421 | unsigned int new_hash; |
7e1e7763 TG |
422 | |
423 | ASSERT_RHT_MUTEX(ht); | |
424 | ||
97defe1e TG |
425 | new_tbl = bucket_table_alloc(ht, tbl->size / 2); |
426 | if (new_tbl == NULL) | |
7e1e7763 TG |
427 | return -ENOMEM; |
428 | ||
97defe1e TG |
429 | rcu_assign_pointer(ht->future_tbl, new_tbl); |
430 | synchronize_rcu(); | |
7e1e7763 | 431 | |
97defe1e TG |
432 | /* Link the first entry in the old bucket to the end of the |
433 | * bucket in the new table. As entries are concurrently being | |
434 | * added to the new table, lock down the new bucket. As we | |
435 | * always divide the size in half when shrinking, each bucket | |
436 | * in the new table maps to exactly two buckets in the old | |
437 | * table. | |
7e1e7763 | 438 | */ |
97defe1e | 439 | for (new_hash = 0; new_hash < new_tbl->size; new_hash++) { |
a5ec68e3 | 440 | lock_buckets(new_tbl, tbl, new_hash); |
97defe1e TG |
441 | |
442 | rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), | |
443 | tbl->buckets[new_hash]); | |
444 | rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), | |
445 | tbl->buckets[new_hash + new_tbl->size]); | |
446 | ||
a5ec68e3 | 447 | unlock_buckets(new_tbl, tbl, new_hash); |
7e1e7763 TG |
448 | } |
449 | ||
450 | /* Publish the new, valid hash table */ | |
97defe1e | 451 | rcu_assign_pointer(ht->tbl, new_tbl); |
c0c09bfd | 452 | atomic_dec(&ht->shift); |
7e1e7763 TG |
453 | |
454 | /* Wait for readers. No new readers will have references to the | |
455 | * old hash table. | |
456 | */ | |
457 | synchronize_rcu(); | |
458 | ||
459 | bucket_table_free(tbl); | |
460 | ||
461 | return 0; | |
462 | } | |
463 | EXPORT_SYMBOL_GPL(rhashtable_shrink); | |
464 | ||
97defe1e TG |
465 | static void rht_deferred_worker(struct work_struct *work) |
466 | { | |
467 | struct rhashtable *ht; | |
468 | struct bucket_table *tbl; | |
f2dba9c6 | 469 | struct rhashtable_walker *walker; |
97defe1e | 470 | |
57699a40 | 471 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 472 | mutex_lock(&ht->mutex); |
28134a53 HX |
473 | if (ht->being_destroyed) |
474 | goto unlock; | |
475 | ||
97defe1e TG |
476 | tbl = rht_dereference(ht->tbl, ht); |
477 | ||
f2dba9c6 HX |
478 | list_for_each_entry(walker, &ht->walkers, list) |
479 | walker->resize = true; | |
480 | ||
97defe1e TG |
481 | if (ht->p.grow_decision && ht->p.grow_decision(ht, tbl->size)) |
482 | rhashtable_expand(ht); | |
483 | else if (ht->p.shrink_decision && ht->p.shrink_decision(ht, tbl->size)) | |
484 | rhashtable_shrink(ht); | |
485 | ||
28134a53 | 486 | unlock: |
97defe1e TG |
487 | mutex_unlock(&ht->mutex); |
488 | } | |
489 | ||
54c5b7d3 YX |
490 | static void rhashtable_wakeup_worker(struct rhashtable *ht) |
491 | { | |
492 | struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht); | |
493 | struct bucket_table *new_tbl = rht_dereference_rcu(ht->future_tbl, ht); | |
494 | size_t size = tbl->size; | |
495 | ||
496 | /* Only adjust the table if no resizing is currently in progress. */ | |
497 | if (tbl == new_tbl && | |
498 | ((ht->p.grow_decision && ht->p.grow_decision(ht, size)) || | |
499 | (ht->p.shrink_decision && ht->p.shrink_decision(ht, size)))) | |
57699a40 | 500 | schedule_work(&ht->run_work); |
54c5b7d3 YX |
501 | } |
502 | ||
db304854 YX |
503 | static void __rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj, |
504 | struct bucket_table *tbl, u32 hash) | |
505 | { | |
506 | struct rhash_head *head = rht_dereference_bucket(tbl->buckets[hash], | |
507 | tbl, hash); | |
508 | ||
509 | if (rht_is_a_nulls(head)) | |
510 | INIT_RHT_NULLS_HEAD(obj->next, ht, hash); | |
511 | else | |
512 | RCU_INIT_POINTER(obj->next, head); | |
513 | ||
514 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
515 | ||
516 | atomic_inc(&ht->nelems); | |
517 | ||
518 | rhashtable_wakeup_worker(ht); | |
519 | } | |
520 | ||
7e1e7763 | 521 | /** |
db304854 | 522 | * rhashtable_insert - insert object into hash table |
7e1e7763 TG |
523 | * @ht: hash table |
524 | * @obj: pointer to hash head inside object | |
7e1e7763 | 525 | * |
97defe1e TG |
526 | * Will take a per bucket spinlock to protect against mutual mutations |
527 | * on the same bucket. Multiple insertions may occur in parallel unless | |
528 | * they map to the same bucket lock. | |
7e1e7763 | 529 | * |
97defe1e TG |
530 | * It is safe to call this function from atomic context. |
531 | * | |
532 | * Will trigger an automatic deferred table resizing if the size grows | |
533 | * beyond the watermark indicated by grow_decision() which can be passed | |
534 | * to rhashtable_init(). | |
7e1e7763 | 535 | */ |
6eba8224 | 536 | void rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj) |
7e1e7763 | 537 | { |
a5ec68e3 | 538 | struct bucket_table *tbl, *old_tbl; |
97defe1e | 539 | unsigned hash; |
7e1e7763 | 540 | |
97defe1e | 541 | rcu_read_lock(); |
7e1e7763 | 542 | |
97defe1e | 543 | tbl = rht_dereference_rcu(ht->future_tbl, ht); |
a5ec68e3 | 544 | old_tbl = rht_dereference_rcu(ht->tbl, ht); |
8d24c0b4 | 545 | hash = head_hashfn(ht, tbl, obj); |
97defe1e | 546 | |
a5ec68e3 | 547 | lock_buckets(tbl, old_tbl, hash); |
db304854 | 548 | __rhashtable_insert(ht, obj, tbl, hash); |
a5ec68e3 | 549 | unlock_buckets(tbl, old_tbl, hash); |
7e1e7763 | 550 | |
97defe1e | 551 | rcu_read_unlock(); |
7e1e7763 TG |
552 | } |
553 | EXPORT_SYMBOL_GPL(rhashtable_insert); | |
554 | ||
7e1e7763 TG |
555 | /** |
556 | * rhashtable_remove - remove object from hash table | |
557 | * @ht: hash table | |
558 | * @obj: pointer to hash head inside object | |
7e1e7763 TG |
559 | * |
560 | * Since the hash chain is single linked, the removal operation needs to | |
561 | * walk the bucket chain upon removal. The removal operation is thus | |
562 | * considerable slow if the hash table is not correctly sized. | |
563 | * | |
db304854 | 564 | * Will automatically shrink the table via rhashtable_expand() if the |
7e1e7763 TG |
565 | * shrink_decision function specified at rhashtable_init() returns true. |
566 | * | |
567 | * The caller must ensure that no concurrent table mutations occur. It is | |
568 | * however valid to have concurrent lookups if they are RCU protected. | |
569 | */ | |
6eba8224 | 570 | bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *obj) |
7e1e7763 | 571 | { |
a5ec68e3 | 572 | struct bucket_table *tbl, *new_tbl, *old_tbl; |
7e1e7763 TG |
573 | struct rhash_head __rcu **pprev; |
574 | struct rhash_head *he; | |
a5ec68e3 | 575 | unsigned int hash, new_hash; |
fe6a043c | 576 | bool ret = false; |
7e1e7763 | 577 | |
97defe1e | 578 | rcu_read_lock(); |
a5ec68e3 TG |
579 | tbl = old_tbl = rht_dereference_rcu(ht->tbl, ht); |
580 | new_tbl = rht_dereference_rcu(ht->future_tbl, ht); | |
581 | new_hash = head_hashfn(ht, new_tbl, obj); | |
7e1e7763 | 582 | |
a5ec68e3 | 583 | lock_buckets(new_tbl, old_tbl, new_hash); |
97defe1e | 584 | restart: |
a5ec68e3 | 585 | hash = rht_bucket_index(tbl, new_hash); |
97defe1e TG |
586 | pprev = &tbl->buckets[hash]; |
587 | rht_for_each(he, tbl, hash) { | |
7e1e7763 TG |
588 | if (he != obj) { |
589 | pprev = &he->next; | |
590 | continue; | |
591 | } | |
592 | ||
97defe1e | 593 | rcu_assign_pointer(*pprev, obj->next); |
897362e4 | 594 | |
fe6a043c TG |
595 | ret = true; |
596 | break; | |
7e1e7763 TG |
597 | } |
598 | ||
fe6a043c TG |
599 | /* The entry may be linked in either 'tbl', 'future_tbl', or both. |
600 | * 'future_tbl' only exists for a short period of time during | |
601 | * resizing. Thus traversing both is fine and the added cost is | |
602 | * very rare. | |
603 | */ | |
a5ec68e3 TG |
604 | if (tbl != new_tbl) { |
605 | tbl = new_tbl; | |
97defe1e TG |
606 | goto restart; |
607 | } | |
608 | ||
a5ec68e3 | 609 | unlock_buckets(new_tbl, old_tbl, new_hash); |
fe6a043c TG |
610 | |
611 | if (ret) { | |
612 | atomic_dec(&ht->nelems); | |
613 | rhashtable_wakeup_worker(ht); | |
614 | } | |
615 | ||
97defe1e TG |
616 | rcu_read_unlock(); |
617 | ||
fe6a043c | 618 | return ret; |
7e1e7763 TG |
619 | } |
620 | EXPORT_SYMBOL_GPL(rhashtable_remove); | |
621 | ||
efb975a6 YX |
622 | struct rhashtable_compare_arg { |
623 | struct rhashtable *ht; | |
624 | const void *key; | |
625 | }; | |
626 | ||
627 | static bool rhashtable_compare(void *ptr, void *arg) | |
628 | { | |
629 | struct rhashtable_compare_arg *x = arg; | |
630 | struct rhashtable *ht = x->ht; | |
631 | ||
632 | return !memcmp(ptr + ht->p.key_offset, x->key, ht->p.key_len); | |
633 | } | |
634 | ||
7e1e7763 TG |
635 | /** |
636 | * rhashtable_lookup - lookup key in hash table | |
637 | * @ht: hash table | |
638 | * @key: pointer to key | |
639 | * | |
640 | * Computes the hash value for the key and traverses the bucket chain looking | |
641 | * for a entry with an identical key. The first matching entry is returned. | |
642 | * | |
643 | * This lookup function may only be used for fixed key hash table (key_len | |
db304854 | 644 | * parameter set). It will BUG() if used inappropriately. |
7e1e7763 | 645 | * |
97defe1e | 646 | * Lookups may occur in parallel with hashtable mutations and resizing. |
7e1e7763 | 647 | */ |
97defe1e | 648 | void *rhashtable_lookup(struct rhashtable *ht, const void *key) |
7e1e7763 | 649 | { |
efb975a6 YX |
650 | struct rhashtable_compare_arg arg = { |
651 | .ht = ht, | |
652 | .key = key, | |
653 | }; | |
7e1e7763 TG |
654 | |
655 | BUG_ON(!ht->p.key_len); | |
656 | ||
efb975a6 | 657 | return rhashtable_lookup_compare(ht, key, &rhashtable_compare, &arg); |
7e1e7763 TG |
658 | } |
659 | EXPORT_SYMBOL_GPL(rhashtable_lookup); | |
660 | ||
661 | /** | |
662 | * rhashtable_lookup_compare - search hash table with compare function | |
663 | * @ht: hash table | |
8d24c0b4 | 664 | * @key: the pointer to the key |
7e1e7763 TG |
665 | * @compare: compare function, must return true on match |
666 | * @arg: argument passed on to compare function | |
667 | * | |
668 | * Traverses the bucket chain behind the provided hash value and calls the | |
669 | * specified compare function for each entry. | |
670 | * | |
97defe1e | 671 | * Lookups may occur in parallel with hashtable mutations and resizing. |
7e1e7763 TG |
672 | * |
673 | * Returns the first entry on which the compare function returned true. | |
674 | */ | |
97defe1e | 675 | void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key, |
7e1e7763 TG |
676 | bool (*compare)(void *, void *), void *arg) |
677 | { | |
97defe1e | 678 | const struct bucket_table *tbl, *old_tbl; |
7e1e7763 | 679 | struct rhash_head *he; |
8d24c0b4 | 680 | u32 hash; |
7e1e7763 | 681 | |
97defe1e TG |
682 | rcu_read_lock(); |
683 | ||
684 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
685 | tbl = rht_dereference_rcu(ht->future_tbl, ht); | |
8d24c0b4 | 686 | hash = key_hashfn(ht, key, ht->p.key_len); |
97defe1e TG |
687 | restart: |
688 | rht_for_each_rcu(he, tbl, rht_bucket_index(tbl, hash)) { | |
7e1e7763 TG |
689 | if (!compare(rht_obj(ht, he), arg)) |
690 | continue; | |
97defe1e | 691 | rcu_read_unlock(); |
a4b18cda | 692 | return rht_obj(ht, he); |
7e1e7763 TG |
693 | } |
694 | ||
97defe1e TG |
695 | if (unlikely(tbl != old_tbl)) { |
696 | tbl = old_tbl; | |
697 | goto restart; | |
698 | } | |
699 | rcu_read_unlock(); | |
700 | ||
7e1e7763 TG |
701 | return NULL; |
702 | } | |
703 | EXPORT_SYMBOL_GPL(rhashtable_lookup_compare); | |
704 | ||
db304854 YX |
705 | /** |
706 | * rhashtable_lookup_insert - lookup and insert object into hash table | |
707 | * @ht: hash table | |
708 | * @obj: pointer to hash head inside object | |
709 | * | |
710 | * Locks down the bucket chain in both the old and new table if a resize | |
711 | * is in progress to ensure that writers can't remove from the old table | |
712 | * and can't insert to the new table during the atomic operation of search | |
713 | * and insertion. Searches for duplicates in both the old and new table if | |
714 | * a resize is in progress. | |
715 | * | |
716 | * This lookup function may only be used for fixed key hash table (key_len | |
717 | * parameter set). It will BUG() if used inappropriately. | |
718 | * | |
719 | * It is safe to call this function from atomic context. | |
720 | * | |
721 | * Will trigger an automatic deferred table resizing if the size grows | |
722 | * beyond the watermark indicated by grow_decision() which can be passed | |
723 | * to rhashtable_init(). | |
724 | */ | |
725 | bool rhashtable_lookup_insert(struct rhashtable *ht, struct rhash_head *obj) | |
7a868d1e YX |
726 | { |
727 | struct rhashtable_compare_arg arg = { | |
728 | .ht = ht, | |
729 | .key = rht_obj(ht, obj) + ht->p.key_offset, | |
730 | }; | |
731 | ||
732 | BUG_ON(!ht->p.key_len); | |
733 | ||
734 | return rhashtable_lookup_compare_insert(ht, obj, &rhashtable_compare, | |
735 | &arg); | |
736 | } | |
737 | EXPORT_SYMBOL_GPL(rhashtable_lookup_insert); | |
738 | ||
739 | /** | |
740 | * rhashtable_lookup_compare_insert - search and insert object to hash table | |
741 | * with compare function | |
742 | * @ht: hash table | |
743 | * @obj: pointer to hash head inside object | |
744 | * @compare: compare function, must return true on match | |
745 | * @arg: argument passed on to compare function | |
746 | * | |
747 | * Locks down the bucket chain in both the old and new table if a resize | |
748 | * is in progress to ensure that writers can't remove from the old table | |
749 | * and can't insert to the new table during the atomic operation of search | |
750 | * and insertion. Searches for duplicates in both the old and new table if | |
751 | * a resize is in progress. | |
752 | * | |
753 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
754 | * | |
755 | * Will trigger an automatic deferred table resizing if the size grows | |
756 | * beyond the watermark indicated by grow_decision() which can be passed | |
757 | * to rhashtable_init(). | |
758 | */ | |
759 | bool rhashtable_lookup_compare_insert(struct rhashtable *ht, | |
760 | struct rhash_head *obj, | |
761 | bool (*compare)(void *, void *), | |
762 | void *arg) | |
db304854 YX |
763 | { |
764 | struct bucket_table *new_tbl, *old_tbl; | |
a5ec68e3 | 765 | u32 new_hash; |
db304854 YX |
766 | bool success = true; |
767 | ||
768 | BUG_ON(!ht->p.key_len); | |
769 | ||
770 | rcu_read_lock(); | |
db304854 | 771 | old_tbl = rht_dereference_rcu(ht->tbl, ht); |
db304854 YX |
772 | new_tbl = rht_dereference_rcu(ht->future_tbl, ht); |
773 | new_hash = head_hashfn(ht, new_tbl, obj); | |
a5ec68e3 TG |
774 | |
775 | lock_buckets(new_tbl, old_tbl, new_hash); | |
db304854 | 776 | |
7a868d1e YX |
777 | if (rhashtable_lookup_compare(ht, rht_obj(ht, obj) + ht->p.key_offset, |
778 | compare, arg)) { | |
db304854 YX |
779 | success = false; |
780 | goto exit; | |
781 | } | |
782 | ||
783 | __rhashtable_insert(ht, obj, new_tbl, new_hash); | |
784 | ||
785 | exit: | |
a5ec68e3 | 786 | unlock_buckets(new_tbl, old_tbl, new_hash); |
db304854 YX |
787 | rcu_read_unlock(); |
788 | ||
789 | return success; | |
790 | } | |
7a868d1e | 791 | EXPORT_SYMBOL_GPL(rhashtable_lookup_compare_insert); |
db304854 | 792 | |
f2dba9c6 HX |
793 | /** |
794 | * rhashtable_walk_init - Initialise an iterator | |
795 | * @ht: Table to walk over | |
796 | * @iter: Hash table Iterator | |
797 | * | |
798 | * This function prepares a hash table walk. | |
799 | * | |
800 | * Note that if you restart a walk after rhashtable_walk_stop you | |
801 | * may see the same object twice. Also, you may miss objects if | |
802 | * there are removals in between rhashtable_walk_stop and the next | |
803 | * call to rhashtable_walk_start. | |
804 | * | |
805 | * For a completely stable walk you should construct your own data | |
806 | * structure outside the hash table. | |
807 | * | |
808 | * This function may sleep so you must not call it from interrupt | |
809 | * context or with spin locks held. | |
810 | * | |
811 | * You must call rhashtable_walk_exit if this function returns | |
812 | * successfully. | |
813 | */ | |
814 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter) | |
815 | { | |
816 | iter->ht = ht; | |
817 | iter->p = NULL; | |
818 | iter->slot = 0; | |
819 | iter->skip = 0; | |
820 | ||
821 | iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL); | |
822 | if (!iter->walker) | |
823 | return -ENOMEM; | |
824 | ||
825 | mutex_lock(&ht->mutex); | |
826 | list_add(&iter->walker->list, &ht->walkers); | |
827 | mutex_unlock(&ht->mutex); | |
828 | ||
829 | return 0; | |
830 | } | |
831 | EXPORT_SYMBOL_GPL(rhashtable_walk_init); | |
832 | ||
833 | /** | |
834 | * rhashtable_walk_exit - Free an iterator | |
835 | * @iter: Hash table Iterator | |
836 | * | |
837 | * This function frees resources allocated by rhashtable_walk_init. | |
838 | */ | |
839 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
840 | { | |
841 | mutex_lock(&iter->ht->mutex); | |
842 | list_del(&iter->walker->list); | |
843 | mutex_unlock(&iter->ht->mutex); | |
844 | kfree(iter->walker); | |
845 | } | |
846 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
847 | ||
848 | /** | |
849 | * rhashtable_walk_start - Start a hash table walk | |
850 | * @iter: Hash table iterator | |
851 | * | |
852 | * Start a hash table walk. Note that we take the RCU lock in all | |
853 | * cases including when we return an error. So you must always call | |
854 | * rhashtable_walk_stop to clean up. | |
855 | * | |
856 | * Returns zero if successful. | |
857 | * | |
858 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
859 | * will rewind back to the beginning and you may use it immediately | |
860 | * by calling rhashtable_walk_next. | |
861 | */ | |
862 | int rhashtable_walk_start(struct rhashtable_iter *iter) | |
863 | { | |
864 | rcu_read_lock(); | |
865 | ||
866 | if (iter->walker->resize) { | |
867 | iter->slot = 0; | |
868 | iter->skip = 0; | |
869 | iter->walker->resize = false; | |
870 | return -EAGAIN; | |
871 | } | |
872 | ||
873 | return 0; | |
874 | } | |
875 | EXPORT_SYMBOL_GPL(rhashtable_walk_start); | |
876 | ||
877 | /** | |
878 | * rhashtable_walk_next - Return the next object and advance the iterator | |
879 | * @iter: Hash table iterator | |
880 | * | |
881 | * Note that you must call rhashtable_walk_stop when you are finished | |
882 | * with the walk. | |
883 | * | |
884 | * Returns the next object or NULL when the end of the table is reached. | |
885 | * | |
886 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
887 | * will rewind back to the beginning and you may continue to use it. | |
888 | */ | |
889 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
890 | { | |
891 | const struct bucket_table *tbl; | |
892 | struct rhashtable *ht = iter->ht; | |
893 | struct rhash_head *p = iter->p; | |
894 | void *obj = NULL; | |
895 | ||
896 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
897 | ||
898 | if (p) { | |
899 | p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot); | |
900 | goto next; | |
901 | } | |
902 | ||
903 | for (; iter->slot < tbl->size; iter->slot++) { | |
904 | int skip = iter->skip; | |
905 | ||
906 | rht_for_each_rcu(p, tbl, iter->slot) { | |
907 | if (!skip) | |
908 | break; | |
909 | skip--; | |
910 | } | |
911 | ||
912 | next: | |
913 | if (!rht_is_a_nulls(p)) { | |
914 | iter->skip++; | |
915 | iter->p = p; | |
916 | obj = rht_obj(ht, p); | |
917 | goto out; | |
918 | } | |
919 | ||
920 | iter->skip = 0; | |
921 | } | |
922 | ||
923 | iter->p = NULL; | |
924 | ||
925 | out: | |
926 | if (iter->walker->resize) { | |
927 | iter->p = NULL; | |
928 | iter->slot = 0; | |
929 | iter->skip = 0; | |
930 | iter->walker->resize = false; | |
931 | return ERR_PTR(-EAGAIN); | |
932 | } | |
933 | ||
934 | return obj; | |
935 | } | |
936 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); | |
937 | ||
938 | /** | |
939 | * rhashtable_walk_stop - Finish a hash table walk | |
940 | * @iter: Hash table iterator | |
941 | * | |
942 | * Finish a hash table walk. | |
943 | */ | |
944 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
945 | { | |
946 | rcu_read_unlock(); | |
947 | iter->p = NULL; | |
948 | } | |
949 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
950 | ||
94000176 | 951 | static size_t rounded_hashtable_size(struct rhashtable_params *params) |
7e1e7763 | 952 | { |
94000176 YX |
953 | return max(roundup_pow_of_two(params->nelem_hint * 4 / 3), |
954 | 1UL << params->min_shift); | |
7e1e7763 TG |
955 | } |
956 | ||
957 | /** | |
958 | * rhashtable_init - initialize a new hash table | |
959 | * @ht: hash table to be initialized | |
960 | * @params: configuration parameters | |
961 | * | |
962 | * Initializes a new hash table based on the provided configuration | |
963 | * parameters. A table can be configured either with a variable or | |
964 | * fixed length key: | |
965 | * | |
966 | * Configuration Example 1: Fixed length keys | |
967 | * struct test_obj { | |
968 | * int key; | |
969 | * void * my_member; | |
970 | * struct rhash_head node; | |
971 | * }; | |
972 | * | |
973 | * struct rhashtable_params params = { | |
974 | * .head_offset = offsetof(struct test_obj, node), | |
975 | * .key_offset = offsetof(struct test_obj, key), | |
976 | * .key_len = sizeof(int), | |
87545899 | 977 | * .hashfn = jhash, |
f89bd6f8 | 978 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
7e1e7763 TG |
979 | * }; |
980 | * | |
981 | * Configuration Example 2: Variable length keys | |
982 | * struct test_obj { | |
983 | * [...] | |
984 | * struct rhash_head node; | |
985 | * }; | |
986 | * | |
987 | * u32 my_hash_fn(const void *data, u32 seed) | |
988 | * { | |
989 | * struct test_obj *obj = data; | |
990 | * | |
991 | * return [... hash ...]; | |
992 | * } | |
993 | * | |
994 | * struct rhashtable_params params = { | |
995 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 996 | * .hashfn = jhash, |
7e1e7763 | 997 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
998 | * }; |
999 | */ | |
1000 | int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params) | |
1001 | { | |
1002 | struct bucket_table *tbl; | |
1003 | size_t size; | |
1004 | ||
1005 | size = HASH_DEFAULT_SIZE; | |
1006 | ||
1007 | if ((params->key_len && !params->hashfn) || | |
1008 | (!params->key_len && !params->obj_hashfn)) | |
1009 | return -EINVAL; | |
1010 | ||
f89bd6f8 TG |
1011 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
1012 | return -EINVAL; | |
1013 | ||
94000176 YX |
1014 | params->min_shift = max_t(size_t, params->min_shift, |
1015 | ilog2(HASH_MIN_SIZE)); | |
1016 | ||
7e1e7763 | 1017 | if (params->nelem_hint) |
94000176 | 1018 | size = rounded_hashtable_size(params); |
7e1e7763 | 1019 | |
97defe1e TG |
1020 | memset(ht, 0, sizeof(*ht)); |
1021 | mutex_init(&ht->mutex); | |
1022 | memcpy(&ht->p, params, sizeof(*params)); | |
f2dba9c6 | 1023 | INIT_LIST_HEAD(&ht->walkers); |
97defe1e TG |
1024 | |
1025 | if (params->locks_mul) | |
1026 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
1027 | else | |
1028 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
1029 | ||
1030 | tbl = bucket_table_alloc(ht, size); | |
7e1e7763 TG |
1031 | if (tbl == NULL) |
1032 | return -ENOMEM; | |
1033 | ||
545a148e | 1034 | atomic_set(&ht->nelems, 0); |
c0c09bfd | 1035 | atomic_set(&ht->shift, ilog2(tbl->size)); |
7e1e7763 | 1036 | RCU_INIT_POINTER(ht->tbl, tbl); |
97defe1e | 1037 | RCU_INIT_POINTER(ht->future_tbl, tbl); |
7e1e7763 TG |
1038 | |
1039 | if (!ht->p.hash_rnd) | |
1040 | get_random_bytes(&ht->p.hash_rnd, sizeof(ht->p.hash_rnd)); | |
1041 | ||
97defe1e | 1042 | if (ht->p.grow_decision || ht->p.shrink_decision) |
57699a40 | 1043 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1044 | |
7e1e7763 TG |
1045 | return 0; |
1046 | } | |
1047 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1048 | ||
1049 | /** | |
1050 | * rhashtable_destroy - destroy hash table | |
1051 | * @ht: the hash table to destroy | |
1052 | * | |
ae82ddcf PNA |
1053 | * Frees the bucket array. This function is not rcu safe, therefore the caller |
1054 | * has to make sure that no resizing may happen by unpublishing the hashtable | |
1055 | * and waiting for the quiescent cycle before releasing the bucket array. | |
7e1e7763 | 1056 | */ |
97defe1e | 1057 | void rhashtable_destroy(struct rhashtable *ht) |
7e1e7763 | 1058 | { |
97defe1e TG |
1059 | ht->being_destroyed = true; |
1060 | ||
57699a40 YX |
1061 | if (ht->p.grow_decision || ht->p.shrink_decision) |
1062 | cancel_work_sync(&ht->run_work); | |
97defe1e | 1063 | |
57699a40 | 1064 | mutex_lock(&ht->mutex); |
97defe1e | 1065 | bucket_table_free(rht_dereference(ht->tbl, ht)); |
97defe1e | 1066 | mutex_unlock(&ht->mutex); |
7e1e7763 TG |
1067 | } |
1068 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |