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 | ||
07ee0722 | 17 | #include <linux/atomic.h> |
7e1e7763 TG |
18 | #include <linux/kernel.h> |
19 | #include <linux/init.h> | |
20 | #include <linux/log2.h> | |
5beb5c90 | 21 | #include <linux/sched.h> |
7e1e7763 TG |
22 | #include <linux/slab.h> |
23 | #include <linux/vmalloc.h> | |
24 | #include <linux/mm.h> | |
87545899 | 25 | #include <linux/jhash.h> |
7e1e7763 TG |
26 | #include <linux/random.h> |
27 | #include <linux/rhashtable.h> | |
61d7b097 | 28 | #include <linux/err.h> |
7e1e7763 TG |
29 | |
30 | #define HASH_DEFAULT_SIZE 64UL | |
c2e213cf | 31 | #define HASH_MIN_SIZE 4U |
97defe1e TG |
32 | #define BUCKET_LOCKS_PER_CPU 128UL |
33 | ||
988dfbd7 | 34 | static u32 head_hashfn(struct rhashtable *ht, |
8d24c0b4 TG |
35 | const struct bucket_table *tbl, |
36 | const struct rhash_head *he) | |
7e1e7763 | 37 | { |
02fd97c3 | 38 | return rht_head_hashfn(ht, tbl, he, ht->p); |
7e1e7763 TG |
39 | } |
40 | ||
a03eaec0 | 41 | #ifdef CONFIG_PROVE_LOCKING |
a03eaec0 | 42 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) |
a03eaec0 TG |
43 | |
44 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
45 | { | |
46 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; | |
47 | } | |
48 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
49 | ||
50 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
51 | { | |
02fd97c3 | 52 | spinlock_t *lock = rht_bucket_lock(tbl, hash); |
a03eaec0 TG |
53 | |
54 | return (debug_locks) ? lockdep_is_held(lock) : 1; | |
55 | } | |
56 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
57 | #else | |
58 | #define ASSERT_RHT_MUTEX(HT) | |
a03eaec0 TG |
59 | #endif |
60 | ||
61 | ||
b9ecfdaa HX |
62 | static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl, |
63 | gfp_t gfp) | |
97defe1e TG |
64 | { |
65 | unsigned int i, size; | |
66 | #if defined(CONFIG_PROVE_LOCKING) | |
67 | unsigned int nr_pcpus = 2; | |
68 | #else | |
69 | unsigned int nr_pcpus = num_possible_cpus(); | |
70 | #endif | |
71 | ||
72 | nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL); | |
73 | size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul); | |
74 | ||
a5ec68e3 TG |
75 | /* Never allocate more than 0.5 locks per bucket */ |
76 | size = min_t(unsigned int, size, tbl->size >> 1); | |
97defe1e TG |
77 | |
78 | if (sizeof(spinlock_t) != 0) { | |
79 | #ifdef CONFIG_NUMA | |
b9ecfdaa HX |
80 | if (size * sizeof(spinlock_t) > PAGE_SIZE && |
81 | gfp == GFP_KERNEL) | |
97defe1e TG |
82 | tbl->locks = vmalloc(size * sizeof(spinlock_t)); |
83 | else | |
84 | #endif | |
85 | tbl->locks = kmalloc_array(size, sizeof(spinlock_t), | |
b9ecfdaa | 86 | gfp); |
97defe1e TG |
87 | if (!tbl->locks) |
88 | return -ENOMEM; | |
89 | for (i = 0; i < size; i++) | |
90 | spin_lock_init(&tbl->locks[i]); | |
91 | } | |
92 | tbl->locks_mask = size - 1; | |
93 | ||
94 | return 0; | |
95 | } | |
96 | ||
97 | static void bucket_table_free(const struct bucket_table *tbl) | |
98 | { | |
99 | if (tbl) | |
100 | kvfree(tbl->locks); | |
101 | ||
102 | kvfree(tbl); | |
103 | } | |
104 | ||
9d901bc0 HX |
105 | static void bucket_table_free_rcu(struct rcu_head *head) |
106 | { | |
107 | bucket_table_free(container_of(head, struct bucket_table, rcu)); | |
108 | } | |
109 | ||
97defe1e | 110 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
b9ecfdaa HX |
111 | size_t nbuckets, |
112 | gfp_t gfp) | |
7e1e7763 | 113 | { |
eb6d1abf | 114 | struct bucket_table *tbl = NULL; |
7e1e7763 | 115 | size_t size; |
f89bd6f8 | 116 | int i; |
7e1e7763 TG |
117 | |
118 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
b9ecfdaa HX |
119 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) || |
120 | gfp != GFP_KERNEL) | |
121 | tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY); | |
122 | if (tbl == NULL && gfp == GFP_KERNEL) | |
7e1e7763 | 123 | tbl = vzalloc(size); |
7e1e7763 TG |
124 | if (tbl == NULL) |
125 | return NULL; | |
126 | ||
127 | tbl->size = nbuckets; | |
128 | ||
b9ecfdaa | 129 | if (alloc_bucket_locks(ht, tbl, gfp) < 0) { |
97defe1e TG |
130 | bucket_table_free(tbl); |
131 | return NULL; | |
132 | } | |
7e1e7763 | 133 | |
eddee5ba HX |
134 | INIT_LIST_HEAD(&tbl->walkers); |
135 | ||
5269b53d HX |
136 | get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd)); |
137 | ||
f89bd6f8 TG |
138 | for (i = 0; i < nbuckets; i++) |
139 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); | |
140 | ||
97defe1e | 141 | return tbl; |
7e1e7763 TG |
142 | } |
143 | ||
b824478b HX |
144 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
145 | struct bucket_table *tbl) | |
146 | { | |
147 | struct bucket_table *new_tbl; | |
148 | ||
149 | do { | |
150 | new_tbl = tbl; | |
151 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
152 | } while (tbl); | |
153 | ||
154 | return new_tbl; | |
155 | } | |
156 | ||
299e5c32 | 157 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash) |
a5ec68e3 | 158 | { |
aa34a6cb | 159 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
b824478b HX |
160 | struct bucket_table *new_tbl = rhashtable_last_table(ht, |
161 | rht_dereference_rcu(old_tbl->future_tbl, ht)); | |
aa34a6cb HX |
162 | struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash]; |
163 | int err = -ENOENT; | |
164 | struct rhash_head *head, *next, *entry; | |
165 | spinlock_t *new_bucket_lock; | |
299e5c32 | 166 | unsigned int new_hash; |
aa34a6cb HX |
167 | |
168 | rht_for_each(entry, old_tbl, old_hash) { | |
169 | err = 0; | |
170 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
171 | ||
172 | if (rht_is_a_nulls(next)) | |
173 | break; | |
a5ec68e3 | 174 | |
aa34a6cb HX |
175 | pprev = &entry->next; |
176 | } | |
a5ec68e3 | 177 | |
aa34a6cb HX |
178 | if (err) |
179 | goto out; | |
97defe1e | 180 | |
aa34a6cb | 181 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 182 | |
02fd97c3 | 183 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 184 | |
8f2484bd | 185 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
186 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
187 | new_tbl, new_hash); | |
97defe1e | 188 | |
aa34a6cb HX |
189 | if (rht_is_a_nulls(head)) |
190 | INIT_RHT_NULLS_HEAD(entry->next, ht, new_hash); | |
191 | else | |
192 | RCU_INIT_POINTER(entry->next, head); | |
a5ec68e3 | 193 | |
aa34a6cb HX |
194 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
195 | spin_unlock(new_bucket_lock); | |
97defe1e | 196 | |
aa34a6cb | 197 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 198 | |
aa34a6cb HX |
199 | out: |
200 | return err; | |
201 | } | |
97defe1e | 202 | |
299e5c32 TG |
203 | static void rhashtable_rehash_chain(struct rhashtable *ht, |
204 | unsigned int old_hash) | |
aa34a6cb HX |
205 | { |
206 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
207 | spinlock_t *old_bucket_lock; | |
208 | ||
02fd97c3 | 209 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 210 | |
aa34a6cb HX |
211 | spin_lock_bh(old_bucket_lock); |
212 | while (!rhashtable_rehash_one(ht, old_hash)) | |
213 | ; | |
63d512d0 | 214 | old_tbl->rehash++; |
aa34a6cb | 215 | spin_unlock_bh(old_bucket_lock); |
97defe1e TG |
216 | } |
217 | ||
b824478b HX |
218 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
219 | struct bucket_table *old_tbl, | |
220 | struct bucket_table *new_tbl) | |
97defe1e | 221 | { |
b824478b HX |
222 | /* Protect future_tbl using the first bucket lock. */ |
223 | spin_lock_bh(old_tbl->locks); | |
224 | ||
225 | /* Did somebody beat us to it? */ | |
226 | if (rcu_access_pointer(old_tbl->future_tbl)) { | |
227 | spin_unlock_bh(old_tbl->locks); | |
228 | return -EEXIST; | |
229 | } | |
7cd10db8 | 230 | |
aa34a6cb HX |
231 | /* Make insertions go into the new, empty table right away. Deletions |
232 | * and lookups will be attempted in both tables until we synchronize. | |
aa34a6cb | 233 | */ |
c4db8848 | 234 | rcu_assign_pointer(old_tbl->future_tbl, new_tbl); |
aa34a6cb | 235 | |
9497df88 HX |
236 | /* Ensure the new table is visible to readers. */ |
237 | smp_wmb(); | |
238 | ||
b824478b HX |
239 | spin_unlock_bh(old_tbl->locks); |
240 | ||
241 | return 0; | |
242 | } | |
243 | ||
244 | static int rhashtable_rehash_table(struct rhashtable *ht) | |
245 | { | |
246 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
247 | struct bucket_table *new_tbl; | |
248 | struct rhashtable_walker *walker; | |
299e5c32 | 249 | unsigned int old_hash; |
b824478b HX |
250 | |
251 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); | |
252 | if (!new_tbl) | |
253 | return 0; | |
254 | ||
aa34a6cb HX |
255 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) |
256 | rhashtable_rehash_chain(ht, old_hash); | |
257 | ||
258 | /* Publish the new table pointer. */ | |
259 | rcu_assign_pointer(ht->tbl, new_tbl); | |
260 | ||
ba7c95ea | 261 | spin_lock(&ht->lock); |
eddee5ba HX |
262 | list_for_each_entry(walker, &old_tbl->walkers, list) |
263 | walker->tbl = NULL; | |
ba7c95ea | 264 | spin_unlock(&ht->lock); |
eddee5ba | 265 | |
aa34a6cb HX |
266 | /* Wait for readers. All new readers will see the new |
267 | * table, and thus no references to the old table will | |
268 | * remain. | |
269 | */ | |
9d901bc0 | 270 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
b824478b HX |
271 | |
272 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; | |
7e1e7763 TG |
273 | } |
274 | ||
275 | /** | |
276 | * rhashtable_expand - Expand hash table while allowing concurrent lookups | |
277 | * @ht: the hash table to expand | |
7e1e7763 | 278 | * |
aa34a6cb | 279 | * A secondary bucket array is allocated and the hash entries are migrated. |
7e1e7763 TG |
280 | * |
281 | * This function may only be called in a context where it is safe to call | |
282 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. | |
283 | * | |
97defe1e TG |
284 | * The caller must ensure that no concurrent resizing occurs by holding |
285 | * ht->mutex. | |
286 | * | |
287 | * It is valid to have concurrent insertions and deletions protected by per | |
288 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 289 | */ |
b824478b | 290 | static int rhashtable_expand(struct rhashtable *ht) |
7e1e7763 TG |
291 | { |
292 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); | |
b824478b | 293 | int err; |
7e1e7763 TG |
294 | |
295 | ASSERT_RHT_MUTEX(ht); | |
296 | ||
b824478b HX |
297 | old_tbl = rhashtable_last_table(ht, old_tbl); |
298 | ||
b9ecfdaa | 299 | new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL); |
7e1e7763 TG |
300 | if (new_tbl == NULL) |
301 | return -ENOMEM; | |
302 | ||
b824478b HX |
303 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
304 | if (err) | |
305 | bucket_table_free(new_tbl); | |
306 | ||
307 | return err; | |
7e1e7763 | 308 | } |
7e1e7763 TG |
309 | |
310 | /** | |
311 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
312 | * @ht: the hash table to shrink | |
7e1e7763 | 313 | * |
18093d1c HX |
314 | * This function shrinks the hash table to fit, i.e., the smallest |
315 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 316 | * |
97defe1e TG |
317 | * The caller must ensure that no concurrent resizing occurs by holding |
318 | * ht->mutex. | |
319 | * | |
7e1e7763 TG |
320 | * The caller must ensure that no concurrent table mutations take place. |
321 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
322 | * |
323 | * It is valid to have concurrent insertions and deletions protected by per | |
324 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 325 | */ |
b824478b | 326 | static int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 327 | { |
a5b6846f | 328 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); |
299e5c32 | 329 | unsigned int size; |
b824478b | 330 | int err; |
7e1e7763 TG |
331 | |
332 | ASSERT_RHT_MUTEX(ht); | |
333 | ||
299e5c32 | 334 | size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2); |
18093d1c HX |
335 | if (size < ht->p.min_size) |
336 | size = ht->p.min_size; | |
337 | ||
338 | if (old_tbl->size <= size) | |
339 | return 0; | |
340 | ||
b824478b HX |
341 | if (rht_dereference(old_tbl->future_tbl, ht)) |
342 | return -EEXIST; | |
343 | ||
b9ecfdaa | 344 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
97defe1e | 345 | if (new_tbl == NULL) |
7e1e7763 TG |
346 | return -ENOMEM; |
347 | ||
b824478b HX |
348 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
349 | if (err) | |
350 | bucket_table_free(new_tbl); | |
351 | ||
352 | return err; | |
7e1e7763 | 353 | } |
7e1e7763 | 354 | |
97defe1e TG |
355 | static void rht_deferred_worker(struct work_struct *work) |
356 | { | |
357 | struct rhashtable *ht; | |
358 | struct bucket_table *tbl; | |
b824478b | 359 | int err = 0; |
97defe1e | 360 | |
57699a40 | 361 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 362 | mutex_lock(&ht->mutex); |
28134a53 | 363 | |
97defe1e | 364 | tbl = rht_dereference(ht->tbl, ht); |
b824478b | 365 | tbl = rhashtable_last_table(ht, tbl); |
97defe1e | 366 | |
a5b6846f | 367 | if (rht_grow_above_75(ht, tbl)) |
97defe1e | 368 | rhashtable_expand(ht); |
b5e2c150 | 369 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
97defe1e | 370 | rhashtable_shrink(ht); |
b824478b HX |
371 | |
372 | err = rhashtable_rehash_table(ht); | |
373 | ||
97defe1e | 374 | mutex_unlock(&ht->mutex); |
b824478b HX |
375 | |
376 | if (err) | |
377 | schedule_work(&ht->run_work); | |
97defe1e TG |
378 | } |
379 | ||
ccd57b1b HX |
380 | static bool rhashtable_check_elasticity(struct rhashtable *ht, |
381 | struct bucket_table *tbl, | |
299e5c32 | 382 | unsigned int hash) |
ccd57b1b | 383 | { |
299e5c32 | 384 | unsigned int elasticity = ht->elasticity; |
ccd57b1b HX |
385 | struct rhash_head *head; |
386 | ||
387 | rht_for_each(head, tbl, hash) | |
388 | if (!--elasticity) | |
389 | return true; | |
390 | ||
391 | return false; | |
392 | } | |
393 | ||
394 | int rhashtable_insert_rehash(struct rhashtable *ht) | |
395 | { | |
396 | struct bucket_table *old_tbl; | |
397 | struct bucket_table *new_tbl; | |
398 | struct bucket_table *tbl; | |
399 | unsigned int size; | |
400 | int err; | |
401 | ||
402 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
403 | tbl = rhashtable_last_table(ht, old_tbl); | |
404 | ||
405 | size = tbl->size; | |
406 | ||
407 | if (rht_grow_above_75(ht, tbl)) | |
408 | size *= 2; | |
a87b9ebf TG |
409 | /* Do not schedule more than one rehash */ |
410 | else if (old_tbl != tbl) | |
ccd57b1b HX |
411 | return -EBUSY; |
412 | ||
413 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC); | |
e2307ed6 TG |
414 | if (new_tbl == NULL) { |
415 | /* Schedule async resize/rehash to try allocation | |
416 | * non-atomic context. | |
417 | */ | |
418 | schedule_work(&ht->run_work); | |
ccd57b1b | 419 | return -ENOMEM; |
e2307ed6 | 420 | } |
ccd57b1b HX |
421 | |
422 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); | |
423 | if (err) { | |
424 | bucket_table_free(new_tbl); | |
425 | if (err == -EEXIST) | |
426 | err = 0; | |
427 | } else | |
428 | schedule_work(&ht->run_work); | |
429 | ||
430 | return err; | |
431 | } | |
432 | EXPORT_SYMBOL_GPL(rhashtable_insert_rehash); | |
433 | ||
02fd97c3 HX |
434 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, |
435 | struct rhash_head *obj, | |
436 | struct bucket_table *tbl) | |
437 | { | |
438 | struct rhash_head *head; | |
299e5c32 | 439 | unsigned int hash; |
ccd57b1b | 440 | int err; |
02fd97c3 | 441 | |
b824478b | 442 | tbl = rhashtable_last_table(ht, tbl); |
02fd97c3 HX |
443 | hash = head_hashfn(ht, tbl, obj); |
444 | spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING); | |
445 | ||
ccd57b1b | 446 | err = -EEXIST; |
02fd97c3 HX |
447 | if (key && rhashtable_lookup_fast(ht, key, ht->p)) |
448 | goto exit; | |
449 | ||
07ee0722 HX |
450 | err = -E2BIG; |
451 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
452 | goto exit; | |
453 | ||
ccd57b1b HX |
454 | err = -EAGAIN; |
455 | if (rhashtable_check_elasticity(ht, tbl, hash) || | |
456 | rht_grow_above_100(ht, tbl)) | |
457 | goto exit; | |
458 | ||
02fd97c3 HX |
459 | err = 0; |
460 | ||
461 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
462 | ||
463 | RCU_INIT_POINTER(obj->next, head); | |
464 | ||
465 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
466 | ||
467 | atomic_inc(&ht->nelems); | |
468 | ||
469 | exit: | |
470 | spin_unlock(rht_bucket_lock(tbl, hash)); | |
471 | ||
472 | return err; | |
473 | } | |
474 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
475 | ||
f2dba9c6 HX |
476 | /** |
477 | * rhashtable_walk_init - Initialise an iterator | |
478 | * @ht: Table to walk over | |
479 | * @iter: Hash table Iterator | |
480 | * | |
481 | * This function prepares a hash table walk. | |
482 | * | |
483 | * Note that if you restart a walk after rhashtable_walk_stop you | |
484 | * may see the same object twice. Also, you may miss objects if | |
485 | * there are removals in between rhashtable_walk_stop and the next | |
486 | * call to rhashtable_walk_start. | |
487 | * | |
488 | * For a completely stable walk you should construct your own data | |
489 | * structure outside the hash table. | |
490 | * | |
491 | * This function may sleep so you must not call it from interrupt | |
492 | * context or with spin locks held. | |
493 | * | |
494 | * You must call rhashtable_walk_exit if this function returns | |
495 | * successfully. | |
496 | */ | |
497 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter) | |
498 | { | |
499 | iter->ht = ht; | |
500 | iter->p = NULL; | |
501 | iter->slot = 0; | |
502 | iter->skip = 0; | |
503 | ||
504 | iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL); | |
505 | if (!iter->walker) | |
506 | return -ENOMEM; | |
507 | ||
508 | mutex_lock(&ht->mutex); | |
eddee5ba HX |
509 | iter->walker->tbl = rht_dereference(ht->tbl, ht); |
510 | list_add(&iter->walker->list, &iter->walker->tbl->walkers); | |
f2dba9c6 HX |
511 | mutex_unlock(&ht->mutex); |
512 | ||
513 | return 0; | |
514 | } | |
515 | EXPORT_SYMBOL_GPL(rhashtable_walk_init); | |
516 | ||
517 | /** | |
518 | * rhashtable_walk_exit - Free an iterator | |
519 | * @iter: Hash table Iterator | |
520 | * | |
521 | * This function frees resources allocated by rhashtable_walk_init. | |
522 | */ | |
523 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
524 | { | |
525 | mutex_lock(&iter->ht->mutex); | |
eddee5ba HX |
526 | if (iter->walker->tbl) |
527 | list_del(&iter->walker->list); | |
f2dba9c6 HX |
528 | mutex_unlock(&iter->ht->mutex); |
529 | kfree(iter->walker); | |
530 | } | |
531 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
532 | ||
533 | /** | |
534 | * rhashtable_walk_start - Start a hash table walk | |
535 | * @iter: Hash table iterator | |
536 | * | |
537 | * Start a hash table walk. Note that we take the RCU lock in all | |
538 | * cases including when we return an error. So you must always call | |
539 | * rhashtable_walk_stop to clean up. | |
540 | * | |
541 | * Returns zero if successful. | |
542 | * | |
543 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
544 | * will rewind back to the beginning and you may use it immediately | |
545 | * by calling rhashtable_walk_next. | |
546 | */ | |
547 | int rhashtable_walk_start(struct rhashtable_iter *iter) | |
db4374f4 | 548 | __acquires(RCU) |
f2dba9c6 | 549 | { |
eddee5ba HX |
550 | struct rhashtable *ht = iter->ht; |
551 | ||
552 | mutex_lock(&ht->mutex); | |
553 | ||
554 | if (iter->walker->tbl) | |
555 | list_del(&iter->walker->list); | |
556 | ||
f2dba9c6 HX |
557 | rcu_read_lock(); |
558 | ||
eddee5ba HX |
559 | mutex_unlock(&ht->mutex); |
560 | ||
561 | if (!iter->walker->tbl) { | |
562 | iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht); | |
f2dba9c6 HX |
563 | return -EAGAIN; |
564 | } | |
565 | ||
566 | return 0; | |
567 | } | |
568 | EXPORT_SYMBOL_GPL(rhashtable_walk_start); | |
569 | ||
570 | /** | |
571 | * rhashtable_walk_next - Return the next object and advance the iterator | |
572 | * @iter: Hash table iterator | |
573 | * | |
574 | * Note that you must call rhashtable_walk_stop when you are finished | |
575 | * with the walk. | |
576 | * | |
577 | * Returns the next object or NULL when the end of the table is reached. | |
578 | * | |
579 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
580 | * will rewind back to the beginning and you may continue to use it. | |
581 | */ | |
582 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
583 | { | |
eddee5ba | 584 | struct bucket_table *tbl = iter->walker->tbl; |
f2dba9c6 HX |
585 | struct rhashtable *ht = iter->ht; |
586 | struct rhash_head *p = iter->p; | |
f2dba9c6 | 587 | |
f2dba9c6 HX |
588 | if (p) { |
589 | p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot); | |
590 | goto next; | |
591 | } | |
592 | ||
593 | for (; iter->slot < tbl->size; iter->slot++) { | |
594 | int skip = iter->skip; | |
595 | ||
596 | rht_for_each_rcu(p, tbl, iter->slot) { | |
597 | if (!skip) | |
598 | break; | |
599 | skip--; | |
600 | } | |
601 | ||
602 | next: | |
603 | if (!rht_is_a_nulls(p)) { | |
604 | iter->skip++; | |
605 | iter->p = p; | |
c936a79f | 606 | return rht_obj(ht, p); |
f2dba9c6 HX |
607 | } |
608 | ||
609 | iter->skip = 0; | |
610 | } | |
611 | ||
d88252f9 HX |
612 | /* Ensure we see any new tables. */ |
613 | smp_rmb(); | |
614 | ||
c4db8848 HX |
615 | iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
616 | if (iter->walker->tbl) { | |
f2dba9c6 HX |
617 | iter->slot = 0; |
618 | iter->skip = 0; | |
f2dba9c6 HX |
619 | return ERR_PTR(-EAGAIN); |
620 | } | |
621 | ||
eddee5ba HX |
622 | iter->p = NULL; |
623 | ||
c936a79f | 624 | return NULL; |
f2dba9c6 HX |
625 | } |
626 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); | |
627 | ||
628 | /** | |
629 | * rhashtable_walk_stop - Finish a hash table walk | |
630 | * @iter: Hash table iterator | |
631 | * | |
632 | * Finish a hash table walk. | |
633 | */ | |
634 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 635 | __releases(RCU) |
f2dba9c6 | 636 | { |
eddee5ba HX |
637 | struct rhashtable *ht; |
638 | struct bucket_table *tbl = iter->walker->tbl; | |
639 | ||
eddee5ba | 640 | if (!tbl) |
963ecbd4 | 641 | goto out; |
eddee5ba HX |
642 | |
643 | ht = iter->ht; | |
644 | ||
ba7c95ea | 645 | spin_lock(&ht->lock); |
c4db8848 | 646 | if (tbl->rehash < tbl->size) |
eddee5ba HX |
647 | list_add(&iter->walker->list, &tbl->walkers); |
648 | else | |
649 | iter->walker->tbl = NULL; | |
ba7c95ea | 650 | spin_unlock(&ht->lock); |
eddee5ba | 651 | |
f2dba9c6 | 652 | iter->p = NULL; |
963ecbd4 HX |
653 | |
654 | out: | |
655 | rcu_read_unlock(); | |
f2dba9c6 HX |
656 | } |
657 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
658 | ||
488fb86e | 659 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 660 | { |
94000176 | 661 | return max(roundup_pow_of_two(params->nelem_hint * 4 / 3), |
e2e21c1c | 662 | (unsigned long)params->min_size); |
7e1e7763 TG |
663 | } |
664 | ||
31ccde2d HX |
665 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
666 | { | |
667 | return jhash2(key, length, seed); | |
668 | } | |
669 | ||
7e1e7763 TG |
670 | /** |
671 | * rhashtable_init - initialize a new hash table | |
672 | * @ht: hash table to be initialized | |
673 | * @params: configuration parameters | |
674 | * | |
675 | * Initializes a new hash table based on the provided configuration | |
676 | * parameters. A table can be configured either with a variable or | |
677 | * fixed length key: | |
678 | * | |
679 | * Configuration Example 1: Fixed length keys | |
680 | * struct test_obj { | |
681 | * int key; | |
682 | * void * my_member; | |
683 | * struct rhash_head node; | |
684 | * }; | |
685 | * | |
686 | * struct rhashtable_params params = { | |
687 | * .head_offset = offsetof(struct test_obj, node), | |
688 | * .key_offset = offsetof(struct test_obj, key), | |
689 | * .key_len = sizeof(int), | |
87545899 | 690 | * .hashfn = jhash, |
f89bd6f8 | 691 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
7e1e7763 TG |
692 | * }; |
693 | * | |
694 | * Configuration Example 2: Variable length keys | |
695 | * struct test_obj { | |
696 | * [...] | |
697 | * struct rhash_head node; | |
698 | * }; | |
699 | * | |
49f7b33e | 700 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
701 | * { |
702 | * struct test_obj *obj = data; | |
703 | * | |
704 | * return [... hash ...]; | |
705 | * } | |
706 | * | |
707 | * struct rhashtable_params params = { | |
708 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 709 | * .hashfn = jhash, |
7e1e7763 | 710 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
711 | * }; |
712 | */ | |
488fb86e HX |
713 | int rhashtable_init(struct rhashtable *ht, |
714 | const struct rhashtable_params *params) | |
7e1e7763 TG |
715 | { |
716 | struct bucket_table *tbl; | |
717 | size_t size; | |
718 | ||
719 | size = HASH_DEFAULT_SIZE; | |
720 | ||
31ccde2d | 721 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 722 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
723 | return -EINVAL; |
724 | ||
f89bd6f8 TG |
725 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
726 | return -EINVAL; | |
727 | ||
7e1e7763 | 728 | if (params->nelem_hint) |
94000176 | 729 | size = rounded_hashtable_size(params); |
7e1e7763 | 730 | |
97defe1e TG |
731 | memset(ht, 0, sizeof(*ht)); |
732 | mutex_init(&ht->mutex); | |
ba7c95ea | 733 | spin_lock_init(&ht->lock); |
97defe1e TG |
734 | memcpy(&ht->p, params, sizeof(*params)); |
735 | ||
a998f712 TG |
736 | if (params->min_size) |
737 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
738 | ||
739 | if (params->max_size) | |
740 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
741 | ||
07ee0722 HX |
742 | if (params->insecure_max_entries) |
743 | ht->p.insecure_max_entries = | |
744 | rounddown_pow_of_two(params->insecure_max_entries); | |
745 | else | |
746 | ht->p.insecure_max_entries = ht->p.max_size * 2; | |
747 | ||
488fb86e | 748 | ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 749 | |
27ed44a5 HX |
750 | /* The maximum (not average) chain length grows with the |
751 | * size of the hash table, at a rate of (log N)/(log log N). | |
752 | * The value of 16 is selected so that even if the hash | |
753 | * table grew to 2^32 you would not expect the maximum | |
754 | * chain length to exceed it unless we are under attack | |
755 | * (or extremely unlucky). | |
756 | * | |
757 | * As this limit is only to detect attacks, we don't need | |
758 | * to set it to a lower value as you'd need the chain | |
759 | * length to vastly exceed 16 to have any real effect | |
760 | * on the system. | |
761 | */ | |
ccd57b1b HX |
762 | if (!params->insecure_elasticity) |
763 | ht->elasticity = 16; | |
764 | ||
97defe1e TG |
765 | if (params->locks_mul) |
766 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
767 | else | |
768 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
769 | ||
31ccde2d HX |
770 | ht->key_len = ht->p.key_len; |
771 | if (!params->hashfn) { | |
772 | ht->p.hashfn = jhash; | |
773 | ||
774 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
775 | ht->key_len /= sizeof(u32); | |
776 | ht->p.hashfn = rhashtable_jhash2; | |
777 | } | |
778 | } | |
779 | ||
b9ecfdaa | 780 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
781 | if (tbl == NULL) |
782 | return -ENOMEM; | |
783 | ||
545a148e | 784 | atomic_set(&ht->nelems, 0); |
a5b6846f | 785 | |
7e1e7763 TG |
786 | RCU_INIT_POINTER(ht->tbl, tbl); |
787 | ||
4c4b52d9 | 788 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 789 | |
7e1e7763 TG |
790 | return 0; |
791 | } | |
792 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
793 | ||
794 | /** | |
6b6f302c | 795 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 796 | * @ht: the hash table to destroy |
6b6f302c TG |
797 | * @free_fn: callback to release resources of element |
798 | * @arg: pointer passed to free_fn | |
7e1e7763 | 799 | * |
6b6f302c TG |
800 | * Stops an eventual async resize. If defined, invokes free_fn for each |
801 | * element to releasal resources. Please note that RCU protected | |
802 | * readers may still be accessing the elements. Releasing of resources | |
803 | * must occur in a compatible manner. Then frees the bucket array. | |
804 | * | |
805 | * This function will eventually sleep to wait for an async resize | |
806 | * to complete. The caller is responsible that no further write operations | |
807 | * occurs in parallel. | |
7e1e7763 | 808 | */ |
6b6f302c TG |
809 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
810 | void (*free_fn)(void *ptr, void *arg), | |
811 | void *arg) | |
7e1e7763 | 812 | { |
6b6f302c TG |
813 | const struct bucket_table *tbl; |
814 | unsigned int i; | |
97defe1e | 815 | |
4c4b52d9 | 816 | cancel_work_sync(&ht->run_work); |
97defe1e | 817 | |
57699a40 | 818 | mutex_lock(&ht->mutex); |
6b6f302c TG |
819 | tbl = rht_dereference(ht->tbl, ht); |
820 | if (free_fn) { | |
821 | for (i = 0; i < tbl->size; i++) { | |
822 | struct rhash_head *pos, *next; | |
823 | ||
824 | for (pos = rht_dereference(tbl->buckets[i], ht), | |
825 | next = !rht_is_a_nulls(pos) ? | |
826 | rht_dereference(pos->next, ht) : NULL; | |
827 | !rht_is_a_nulls(pos); | |
828 | pos = next, | |
829 | next = !rht_is_a_nulls(pos) ? | |
830 | rht_dereference(pos->next, ht) : NULL) | |
831 | free_fn(rht_obj(ht, pos), arg); | |
832 | } | |
833 | } | |
834 | ||
835 | bucket_table_free(tbl); | |
97defe1e | 836 | mutex_unlock(&ht->mutex); |
7e1e7763 | 837 | } |
6b6f302c TG |
838 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
839 | ||
840 | void rhashtable_destroy(struct rhashtable *ht) | |
841 | { | |
842 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
843 | } | |
7e1e7763 | 844 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |