]>
Commit | Line | Data |
---|---|---|
7e1e7763 TG |
1 | /* |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
02fd97c3 | 4 | * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au> |
a5ec68e3 | 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 |
02fd97c3 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 | #include <linux/kernel.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/log2.h> | |
5beb5c90 | 20 | #include <linux/sched.h> |
7e1e7763 TG |
21 | #include <linux/slab.h> |
22 | #include <linux/vmalloc.h> | |
23 | #include <linux/mm.h> | |
87545899 | 24 | #include <linux/jhash.h> |
7e1e7763 TG |
25 | #include <linux/random.h> |
26 | #include <linux/rhashtable.h> | |
61d7b097 | 27 | #include <linux/err.h> |
7e1e7763 TG |
28 | |
29 | #define HASH_DEFAULT_SIZE 64UL | |
c2e213cf | 30 | #define HASH_MIN_SIZE 4U |
97defe1e TG |
31 | #define BUCKET_LOCKS_PER_CPU 128UL |
32 | ||
988dfbd7 | 33 | static u32 head_hashfn(struct rhashtable *ht, |
8d24c0b4 TG |
34 | const struct bucket_table *tbl, |
35 | const struct rhash_head *he) | |
7e1e7763 | 36 | { |
02fd97c3 | 37 | return rht_head_hashfn(ht, tbl, he, ht->p); |
7e1e7763 TG |
38 | } |
39 | ||
a03eaec0 | 40 | #ifdef CONFIG_PROVE_LOCKING |
a03eaec0 | 41 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) |
a03eaec0 TG |
42 | |
43 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
44 | { | |
45 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; | |
46 | } | |
47 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
48 | ||
49 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
50 | { | |
02fd97c3 | 51 | spinlock_t *lock = rht_bucket_lock(tbl, hash); |
a03eaec0 TG |
52 | |
53 | return (debug_locks) ? lockdep_is_held(lock) : 1; | |
54 | } | |
55 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
56 | #else | |
57 | #define ASSERT_RHT_MUTEX(HT) | |
a03eaec0 TG |
58 | #endif |
59 | ||
60 | ||
97defe1e TG |
61 | static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl) |
62 | { | |
63 | unsigned int i, size; | |
64 | #if defined(CONFIG_PROVE_LOCKING) | |
65 | unsigned int nr_pcpus = 2; | |
66 | #else | |
67 | unsigned int nr_pcpus = num_possible_cpus(); | |
68 | #endif | |
69 | ||
70 | nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL); | |
71 | size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul); | |
72 | ||
a5ec68e3 TG |
73 | /* Never allocate more than 0.5 locks per bucket */ |
74 | size = min_t(unsigned int, size, tbl->size >> 1); | |
97defe1e TG |
75 | |
76 | if (sizeof(spinlock_t) != 0) { | |
77 | #ifdef CONFIG_NUMA | |
78 | if (size * sizeof(spinlock_t) > PAGE_SIZE) | |
79 | tbl->locks = vmalloc(size * sizeof(spinlock_t)); | |
80 | else | |
81 | #endif | |
82 | tbl->locks = kmalloc_array(size, sizeof(spinlock_t), | |
83 | GFP_KERNEL); | |
84 | if (!tbl->locks) | |
85 | return -ENOMEM; | |
86 | for (i = 0; i < size; i++) | |
87 | spin_lock_init(&tbl->locks[i]); | |
88 | } | |
89 | tbl->locks_mask = size - 1; | |
90 | ||
91 | return 0; | |
92 | } | |
93 | ||
94 | static void bucket_table_free(const struct bucket_table *tbl) | |
95 | { | |
96 | if (tbl) | |
97 | kvfree(tbl->locks); | |
98 | ||
99 | kvfree(tbl); | |
100 | } | |
101 | ||
9d901bc0 HX |
102 | static void bucket_table_free_rcu(struct rcu_head *head) |
103 | { | |
104 | bucket_table_free(container_of(head, struct bucket_table, rcu)); | |
105 | } | |
106 | ||
97defe1e | 107 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
5269b53d | 108 | size_t nbuckets) |
7e1e7763 | 109 | { |
eb6d1abf | 110 | struct bucket_table *tbl = NULL; |
7e1e7763 | 111 | size_t size; |
f89bd6f8 | 112 | int i; |
7e1e7763 TG |
113 | |
114 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
eb6d1abf DB |
115 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) |
116 | tbl = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); | |
7e1e7763 TG |
117 | if (tbl == NULL) |
118 | tbl = vzalloc(size); | |
7e1e7763 TG |
119 | if (tbl == NULL) |
120 | return NULL; | |
121 | ||
122 | tbl->size = nbuckets; | |
123 | ||
97defe1e TG |
124 | if (alloc_bucket_locks(ht, tbl) < 0) { |
125 | bucket_table_free(tbl); | |
126 | return NULL; | |
127 | } | |
7e1e7763 | 128 | |
eddee5ba HX |
129 | INIT_LIST_HEAD(&tbl->walkers); |
130 | ||
5269b53d HX |
131 | get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd)); |
132 | ||
f89bd6f8 TG |
133 | for (i = 0; i < nbuckets; i++) |
134 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); | |
135 | ||
97defe1e | 136 | return tbl; |
7e1e7763 TG |
137 | } |
138 | ||
aa34a6cb | 139 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned old_hash) |
a5ec68e3 | 140 | { |
aa34a6cb | 141 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
c4db8848 HX |
142 | struct bucket_table *new_tbl = |
143 | rht_dereference(old_tbl->future_tbl, ht) ?: old_tbl; | |
aa34a6cb HX |
144 | struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash]; |
145 | int err = -ENOENT; | |
146 | struct rhash_head *head, *next, *entry; | |
147 | spinlock_t *new_bucket_lock; | |
148 | unsigned new_hash; | |
149 | ||
150 | rht_for_each(entry, old_tbl, old_hash) { | |
151 | err = 0; | |
152 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
153 | ||
154 | if (rht_is_a_nulls(next)) | |
155 | break; | |
a5ec68e3 | 156 | |
aa34a6cb HX |
157 | pprev = &entry->next; |
158 | } | |
a5ec68e3 | 159 | |
aa34a6cb HX |
160 | if (err) |
161 | goto out; | |
97defe1e | 162 | |
aa34a6cb | 163 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 164 | |
02fd97c3 | 165 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 166 | |
8f2484bd | 167 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
168 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
169 | new_tbl, new_hash); | |
97defe1e | 170 | |
aa34a6cb HX |
171 | if (rht_is_a_nulls(head)) |
172 | INIT_RHT_NULLS_HEAD(entry->next, ht, new_hash); | |
173 | else | |
174 | RCU_INIT_POINTER(entry->next, head); | |
a5ec68e3 | 175 | |
aa34a6cb HX |
176 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
177 | spin_unlock(new_bucket_lock); | |
97defe1e | 178 | |
aa34a6cb | 179 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 180 | |
aa34a6cb HX |
181 | out: |
182 | return err; | |
183 | } | |
97defe1e | 184 | |
aa34a6cb HX |
185 | static void rhashtable_rehash_chain(struct rhashtable *ht, unsigned old_hash) |
186 | { | |
187 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
188 | spinlock_t *old_bucket_lock; | |
189 | ||
02fd97c3 | 190 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 191 | |
aa34a6cb HX |
192 | spin_lock_bh(old_bucket_lock); |
193 | while (!rhashtable_rehash_one(ht, old_hash)) | |
194 | ; | |
63d512d0 | 195 | old_tbl->rehash++; |
aa34a6cb | 196 | spin_unlock_bh(old_bucket_lock); |
97defe1e TG |
197 | } |
198 | ||
aa34a6cb HX |
199 | static void rhashtable_rehash(struct rhashtable *ht, |
200 | struct bucket_table *new_tbl) | |
97defe1e | 201 | { |
aa34a6cb | 202 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
eddee5ba | 203 | struct rhashtable_walker *walker; |
aa34a6cb | 204 | unsigned old_hash; |
7cd10db8 | 205 | |
aa34a6cb HX |
206 | /* Make insertions go into the new, empty table right away. Deletions |
207 | * and lookups will be attempted in both tables until we synchronize. | |
aa34a6cb | 208 | */ |
c4db8848 | 209 | rcu_assign_pointer(old_tbl->future_tbl, new_tbl); |
aa34a6cb | 210 | |
9497df88 HX |
211 | /* Ensure the new table is visible to readers. */ |
212 | smp_wmb(); | |
213 | ||
aa34a6cb HX |
214 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) |
215 | rhashtable_rehash_chain(ht, old_hash); | |
216 | ||
217 | /* Publish the new table pointer. */ | |
218 | rcu_assign_pointer(ht->tbl, new_tbl); | |
219 | ||
eddee5ba HX |
220 | list_for_each_entry(walker, &old_tbl->walkers, list) |
221 | walker->tbl = NULL; | |
222 | ||
aa34a6cb HX |
223 | /* Wait for readers. All new readers will see the new |
224 | * table, and thus no references to the old table will | |
225 | * remain. | |
226 | */ | |
9d901bc0 | 227 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
7e1e7763 TG |
228 | } |
229 | ||
230 | /** | |
231 | * rhashtable_expand - Expand hash table while allowing concurrent lookups | |
232 | * @ht: the hash table to expand | |
7e1e7763 | 233 | * |
aa34a6cb | 234 | * A secondary bucket array is allocated and the hash entries are migrated. |
7e1e7763 TG |
235 | * |
236 | * This function may only be called in a context where it is safe to call | |
237 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. | |
238 | * | |
97defe1e TG |
239 | * The caller must ensure that no concurrent resizing occurs by holding |
240 | * ht->mutex. | |
241 | * | |
242 | * It is valid to have concurrent insertions and deletions protected by per | |
243 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 244 | */ |
6eba8224 | 245 | int rhashtable_expand(struct rhashtable *ht) |
7e1e7763 TG |
246 | { |
247 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); | |
7e1e7763 TG |
248 | |
249 | ASSERT_RHT_MUTEX(ht); | |
250 | ||
5269b53d | 251 | new_tbl = bucket_table_alloc(ht, old_tbl->size * 2); |
7e1e7763 TG |
252 | if (new_tbl == NULL) |
253 | return -ENOMEM; | |
254 | ||
aa34a6cb | 255 | rhashtable_rehash(ht, new_tbl); |
7e1e7763 TG |
256 | return 0; |
257 | } | |
258 | EXPORT_SYMBOL_GPL(rhashtable_expand); | |
259 | ||
260 | /** | |
261 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
262 | * @ht: the hash table to shrink | |
7e1e7763 | 263 | * |
18093d1c HX |
264 | * This function shrinks the hash table to fit, i.e., the smallest |
265 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 266 | * |
97defe1e TG |
267 | * The caller must ensure that no concurrent resizing occurs by holding |
268 | * ht->mutex. | |
269 | * | |
7e1e7763 TG |
270 | * The caller must ensure that no concurrent table mutations take place. |
271 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
272 | * |
273 | * It is valid to have concurrent insertions and deletions protected by per | |
274 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 275 | */ |
6eba8224 | 276 | int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 277 | { |
a5b6846f | 278 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); |
18093d1c | 279 | unsigned size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2); |
7e1e7763 TG |
280 | |
281 | ASSERT_RHT_MUTEX(ht); | |
282 | ||
18093d1c HX |
283 | if (size < ht->p.min_size) |
284 | size = ht->p.min_size; | |
285 | ||
286 | if (old_tbl->size <= size) | |
287 | return 0; | |
288 | ||
289 | new_tbl = bucket_table_alloc(ht, size); | |
97defe1e | 290 | if (new_tbl == NULL) |
7e1e7763 TG |
291 | return -ENOMEM; |
292 | ||
aa34a6cb | 293 | rhashtable_rehash(ht, new_tbl); |
7e1e7763 TG |
294 | return 0; |
295 | } | |
296 | EXPORT_SYMBOL_GPL(rhashtable_shrink); | |
297 | ||
97defe1e TG |
298 | static void rht_deferred_worker(struct work_struct *work) |
299 | { | |
300 | struct rhashtable *ht; | |
301 | struct bucket_table *tbl; | |
302 | ||
57699a40 | 303 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 304 | mutex_lock(&ht->mutex); |
28134a53 HX |
305 | if (ht->being_destroyed) |
306 | goto unlock; | |
307 | ||
97defe1e TG |
308 | tbl = rht_dereference(ht->tbl, ht); |
309 | ||
a5b6846f | 310 | if (rht_grow_above_75(ht, tbl)) |
97defe1e | 311 | rhashtable_expand(ht); |
a5b6846f | 312 | else if (rht_shrink_below_30(ht, tbl)) |
97defe1e | 313 | rhashtable_shrink(ht); |
28134a53 | 314 | unlock: |
97defe1e TG |
315 | mutex_unlock(&ht->mutex); |
316 | } | |
317 | ||
02fd97c3 HX |
318 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, |
319 | struct rhash_head *obj, | |
320 | struct bucket_table *tbl) | |
321 | { | |
322 | struct rhash_head *head; | |
323 | unsigned hash; | |
324 | int err = -EEXIST; | |
325 | ||
326 | hash = head_hashfn(ht, tbl, obj); | |
327 | spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING); | |
328 | ||
329 | if (key && rhashtable_lookup_fast(ht, key, ht->p)) | |
330 | goto exit; | |
331 | ||
332 | err = 0; | |
333 | ||
334 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
335 | ||
336 | RCU_INIT_POINTER(obj->next, head); | |
337 | ||
338 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
339 | ||
340 | atomic_inc(&ht->nelems); | |
341 | ||
342 | exit: | |
343 | spin_unlock(rht_bucket_lock(tbl, hash)); | |
344 | ||
345 | return err; | |
346 | } | |
347 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
348 | ||
f2dba9c6 HX |
349 | /** |
350 | * rhashtable_walk_init - Initialise an iterator | |
351 | * @ht: Table to walk over | |
352 | * @iter: Hash table Iterator | |
353 | * | |
354 | * This function prepares a hash table walk. | |
355 | * | |
356 | * Note that if you restart a walk after rhashtable_walk_stop you | |
357 | * may see the same object twice. Also, you may miss objects if | |
358 | * there are removals in between rhashtable_walk_stop and the next | |
359 | * call to rhashtable_walk_start. | |
360 | * | |
361 | * For a completely stable walk you should construct your own data | |
362 | * structure outside the hash table. | |
363 | * | |
364 | * This function may sleep so you must not call it from interrupt | |
365 | * context or with spin locks held. | |
366 | * | |
367 | * You must call rhashtable_walk_exit if this function returns | |
368 | * successfully. | |
369 | */ | |
370 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter) | |
371 | { | |
372 | iter->ht = ht; | |
373 | iter->p = NULL; | |
374 | iter->slot = 0; | |
375 | iter->skip = 0; | |
376 | ||
377 | iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL); | |
378 | if (!iter->walker) | |
379 | return -ENOMEM; | |
380 | ||
381 | mutex_lock(&ht->mutex); | |
eddee5ba HX |
382 | iter->walker->tbl = rht_dereference(ht->tbl, ht); |
383 | list_add(&iter->walker->list, &iter->walker->tbl->walkers); | |
f2dba9c6 HX |
384 | mutex_unlock(&ht->mutex); |
385 | ||
386 | return 0; | |
387 | } | |
388 | EXPORT_SYMBOL_GPL(rhashtable_walk_init); | |
389 | ||
390 | /** | |
391 | * rhashtable_walk_exit - Free an iterator | |
392 | * @iter: Hash table Iterator | |
393 | * | |
394 | * This function frees resources allocated by rhashtable_walk_init. | |
395 | */ | |
396 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
397 | { | |
398 | mutex_lock(&iter->ht->mutex); | |
eddee5ba HX |
399 | if (iter->walker->tbl) |
400 | list_del(&iter->walker->list); | |
f2dba9c6 HX |
401 | mutex_unlock(&iter->ht->mutex); |
402 | kfree(iter->walker); | |
403 | } | |
404 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
405 | ||
406 | /** | |
407 | * rhashtable_walk_start - Start a hash table walk | |
408 | * @iter: Hash table iterator | |
409 | * | |
410 | * Start a hash table walk. Note that we take the RCU lock in all | |
411 | * cases including when we return an error. So you must always call | |
412 | * rhashtable_walk_stop to clean up. | |
413 | * | |
414 | * Returns zero if successful. | |
415 | * | |
416 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
417 | * will rewind back to the beginning and you may use it immediately | |
418 | * by calling rhashtable_walk_next. | |
419 | */ | |
420 | int rhashtable_walk_start(struct rhashtable_iter *iter) | |
db4374f4 | 421 | __acquires(RCU) |
f2dba9c6 | 422 | { |
eddee5ba HX |
423 | struct rhashtable *ht = iter->ht; |
424 | ||
425 | mutex_lock(&ht->mutex); | |
426 | ||
427 | if (iter->walker->tbl) | |
428 | list_del(&iter->walker->list); | |
429 | ||
f2dba9c6 HX |
430 | rcu_read_lock(); |
431 | ||
eddee5ba HX |
432 | mutex_unlock(&ht->mutex); |
433 | ||
434 | if (!iter->walker->tbl) { | |
435 | iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht); | |
f2dba9c6 HX |
436 | return -EAGAIN; |
437 | } | |
438 | ||
439 | return 0; | |
440 | } | |
441 | EXPORT_SYMBOL_GPL(rhashtable_walk_start); | |
442 | ||
443 | /** | |
444 | * rhashtable_walk_next - Return the next object and advance the iterator | |
445 | * @iter: Hash table iterator | |
446 | * | |
447 | * Note that you must call rhashtable_walk_stop when you are finished | |
448 | * with the walk. | |
449 | * | |
450 | * Returns the next object or NULL when the end of the table is reached. | |
451 | * | |
452 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
453 | * will rewind back to the beginning and you may continue to use it. | |
454 | */ | |
455 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
456 | { | |
eddee5ba | 457 | struct bucket_table *tbl = iter->walker->tbl; |
f2dba9c6 HX |
458 | struct rhashtable *ht = iter->ht; |
459 | struct rhash_head *p = iter->p; | |
460 | void *obj = NULL; | |
461 | ||
f2dba9c6 HX |
462 | if (p) { |
463 | p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot); | |
464 | goto next; | |
465 | } | |
466 | ||
467 | for (; iter->slot < tbl->size; iter->slot++) { | |
468 | int skip = iter->skip; | |
469 | ||
470 | rht_for_each_rcu(p, tbl, iter->slot) { | |
471 | if (!skip) | |
472 | break; | |
473 | skip--; | |
474 | } | |
475 | ||
476 | next: | |
477 | if (!rht_is_a_nulls(p)) { | |
478 | iter->skip++; | |
479 | iter->p = p; | |
480 | obj = rht_obj(ht, p); | |
481 | goto out; | |
482 | } | |
483 | ||
484 | iter->skip = 0; | |
485 | } | |
486 | ||
d88252f9 HX |
487 | /* Ensure we see any new tables. */ |
488 | smp_rmb(); | |
489 | ||
c4db8848 HX |
490 | iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
491 | if (iter->walker->tbl) { | |
f2dba9c6 HX |
492 | iter->slot = 0; |
493 | iter->skip = 0; | |
f2dba9c6 HX |
494 | return ERR_PTR(-EAGAIN); |
495 | } | |
496 | ||
eddee5ba HX |
497 | iter->p = NULL; |
498 | ||
499 | out: | |
500 | ||
f2dba9c6 HX |
501 | return obj; |
502 | } | |
503 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); | |
504 | ||
505 | /** | |
506 | * rhashtable_walk_stop - Finish a hash table walk | |
507 | * @iter: Hash table iterator | |
508 | * | |
509 | * Finish a hash table walk. | |
510 | */ | |
511 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 512 | __releases(RCU) |
f2dba9c6 | 513 | { |
eddee5ba HX |
514 | struct rhashtable *ht; |
515 | struct bucket_table *tbl = iter->walker->tbl; | |
516 | ||
eddee5ba | 517 | if (!tbl) |
963ecbd4 | 518 | goto out; |
eddee5ba HX |
519 | |
520 | ht = iter->ht; | |
521 | ||
522 | mutex_lock(&ht->mutex); | |
c4db8848 | 523 | if (tbl->rehash < tbl->size) |
eddee5ba HX |
524 | list_add(&iter->walker->list, &tbl->walkers); |
525 | else | |
526 | iter->walker->tbl = NULL; | |
527 | mutex_unlock(&ht->mutex); | |
528 | ||
f2dba9c6 | 529 | iter->p = NULL; |
963ecbd4 HX |
530 | |
531 | out: | |
532 | rcu_read_unlock(); | |
f2dba9c6 HX |
533 | } |
534 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
535 | ||
488fb86e | 536 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 537 | { |
94000176 | 538 | return max(roundup_pow_of_two(params->nelem_hint * 4 / 3), |
e2e21c1c | 539 | (unsigned long)params->min_size); |
7e1e7763 TG |
540 | } |
541 | ||
31ccde2d HX |
542 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
543 | { | |
544 | return jhash2(key, length, seed); | |
545 | } | |
546 | ||
7e1e7763 TG |
547 | /** |
548 | * rhashtable_init - initialize a new hash table | |
549 | * @ht: hash table to be initialized | |
550 | * @params: configuration parameters | |
551 | * | |
552 | * Initializes a new hash table based on the provided configuration | |
553 | * parameters. A table can be configured either with a variable or | |
554 | * fixed length key: | |
555 | * | |
556 | * Configuration Example 1: Fixed length keys | |
557 | * struct test_obj { | |
558 | * int key; | |
559 | * void * my_member; | |
560 | * struct rhash_head node; | |
561 | * }; | |
562 | * | |
563 | * struct rhashtable_params params = { | |
564 | * .head_offset = offsetof(struct test_obj, node), | |
565 | * .key_offset = offsetof(struct test_obj, key), | |
566 | * .key_len = sizeof(int), | |
87545899 | 567 | * .hashfn = jhash, |
f89bd6f8 | 568 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
7e1e7763 TG |
569 | * }; |
570 | * | |
571 | * Configuration Example 2: Variable length keys | |
572 | * struct test_obj { | |
573 | * [...] | |
574 | * struct rhash_head node; | |
575 | * }; | |
576 | * | |
577 | * u32 my_hash_fn(const void *data, u32 seed) | |
578 | * { | |
579 | * struct test_obj *obj = data; | |
580 | * | |
581 | * return [... hash ...]; | |
582 | * } | |
583 | * | |
584 | * struct rhashtable_params params = { | |
585 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 586 | * .hashfn = jhash, |
7e1e7763 | 587 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
588 | * }; |
589 | */ | |
488fb86e HX |
590 | int rhashtable_init(struct rhashtable *ht, |
591 | const struct rhashtable_params *params) | |
7e1e7763 TG |
592 | { |
593 | struct bucket_table *tbl; | |
594 | size_t size; | |
595 | ||
596 | size = HASH_DEFAULT_SIZE; | |
597 | ||
31ccde2d | 598 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 599 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
600 | return -EINVAL; |
601 | ||
f89bd6f8 TG |
602 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
603 | return -EINVAL; | |
604 | ||
7e1e7763 | 605 | if (params->nelem_hint) |
94000176 | 606 | size = rounded_hashtable_size(params); |
7e1e7763 | 607 | |
97defe1e TG |
608 | memset(ht, 0, sizeof(*ht)); |
609 | mutex_init(&ht->mutex); | |
610 | memcpy(&ht->p, params, sizeof(*params)); | |
611 | ||
a998f712 TG |
612 | if (params->min_size) |
613 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
614 | ||
615 | if (params->max_size) | |
616 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
617 | ||
488fb86e | 618 | ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 619 | |
97defe1e TG |
620 | if (params->locks_mul) |
621 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
622 | else | |
623 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
624 | ||
31ccde2d HX |
625 | ht->key_len = ht->p.key_len; |
626 | if (!params->hashfn) { | |
627 | ht->p.hashfn = jhash; | |
628 | ||
629 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
630 | ht->key_len /= sizeof(u32); | |
631 | ht->p.hashfn = rhashtable_jhash2; | |
632 | } | |
633 | } | |
634 | ||
5269b53d | 635 | tbl = bucket_table_alloc(ht, size); |
7e1e7763 TG |
636 | if (tbl == NULL) |
637 | return -ENOMEM; | |
638 | ||
545a148e | 639 | atomic_set(&ht->nelems, 0); |
a5b6846f | 640 | |
7e1e7763 TG |
641 | RCU_INIT_POINTER(ht->tbl, tbl); |
642 | ||
4c4b52d9 | 643 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 644 | |
7e1e7763 TG |
645 | return 0; |
646 | } | |
647 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
648 | ||
649 | /** | |
650 | * rhashtable_destroy - destroy hash table | |
651 | * @ht: the hash table to destroy | |
652 | * | |
ae82ddcf PNA |
653 | * Frees the bucket array. This function is not rcu safe, therefore the caller |
654 | * has to make sure that no resizing may happen by unpublishing the hashtable | |
655 | * and waiting for the quiescent cycle before releasing the bucket array. | |
7e1e7763 | 656 | */ |
97defe1e | 657 | void rhashtable_destroy(struct rhashtable *ht) |
7e1e7763 | 658 | { |
97defe1e TG |
659 | ht->being_destroyed = true; |
660 | ||
4c4b52d9 | 661 | cancel_work_sync(&ht->run_work); |
97defe1e | 662 | |
57699a40 | 663 | mutex_lock(&ht->mutex); |
97defe1e | 664 | bucket_table_free(rht_dereference(ht->tbl, ht)); |
97defe1e | 665 | mutex_unlock(&ht->mutex); |
7e1e7763 TG |
666 | } |
667 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |