]>
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> |
b2d09103 | 22 | #include <linux/rculist.h> |
7e1e7763 TG |
23 | #include <linux/slab.h> |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/mm.h> | |
87545899 | 26 | #include <linux/jhash.h> |
7e1e7763 TG |
27 | #include <linux/random.h> |
28 | #include <linux/rhashtable.h> | |
61d7b097 | 29 | #include <linux/err.h> |
6d795413 | 30 | #include <linux/export.h> |
7e1e7763 TG |
31 | |
32 | #define HASH_DEFAULT_SIZE 64UL | |
c2e213cf | 33 | #define HASH_MIN_SIZE 4U |
4cf0b354 | 34 | #define BUCKET_LOCKS_PER_CPU 32UL |
97defe1e | 35 | |
da20420f HX |
36 | union nested_table { |
37 | union nested_table __rcu *table; | |
38 | struct rhash_head __rcu *bucket; | |
39 | }; | |
40 | ||
988dfbd7 | 41 | static u32 head_hashfn(struct rhashtable *ht, |
8d24c0b4 TG |
42 | const struct bucket_table *tbl, |
43 | const struct rhash_head *he) | |
7e1e7763 | 44 | { |
02fd97c3 | 45 | return rht_head_hashfn(ht, tbl, he, ht->p); |
7e1e7763 TG |
46 | } |
47 | ||
a03eaec0 | 48 | #ifdef CONFIG_PROVE_LOCKING |
a03eaec0 | 49 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) |
a03eaec0 TG |
50 | |
51 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
52 | { | |
53 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; | |
54 | } | |
55 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
56 | ||
57 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
58 | { | |
02fd97c3 | 59 | spinlock_t *lock = rht_bucket_lock(tbl, hash); |
a03eaec0 TG |
60 | |
61 | return (debug_locks) ? lockdep_is_held(lock) : 1; | |
62 | } | |
63 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
64 | #else | |
65 | #define ASSERT_RHT_MUTEX(HT) | |
a03eaec0 TG |
66 | #endif |
67 | ||
da20420f HX |
68 | static void nested_table_free(union nested_table *ntbl, unsigned int size) |
69 | { | |
70 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
71 | const unsigned int len = 1 << shift; | |
72 | unsigned int i; | |
73 | ||
74 | ntbl = rcu_dereference_raw(ntbl->table); | |
75 | if (!ntbl) | |
76 | return; | |
77 | ||
78 | if (size > len) { | |
79 | size >>= shift; | |
80 | for (i = 0; i < len; i++) | |
81 | nested_table_free(ntbl + i, size); | |
82 | } | |
83 | ||
84 | kfree(ntbl); | |
85 | } | |
86 | ||
87 | static void nested_bucket_table_free(const struct bucket_table *tbl) | |
88 | { | |
89 | unsigned int size = tbl->size >> tbl->nest; | |
90 | unsigned int len = 1 << tbl->nest; | |
91 | union nested_table *ntbl; | |
92 | unsigned int i; | |
93 | ||
94 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
95 | ||
96 | for (i = 0; i < len; i++) | |
97 | nested_table_free(ntbl + i, size); | |
98 | ||
99 | kfree(ntbl); | |
100 | } | |
101 | ||
97defe1e TG |
102 | static void bucket_table_free(const struct bucket_table *tbl) |
103 | { | |
da20420f HX |
104 | if (tbl->nest) |
105 | nested_bucket_table_free(tbl); | |
106 | ||
64e0cd0d | 107 | free_bucket_spinlocks(tbl->locks); |
97defe1e TG |
108 | kvfree(tbl); |
109 | } | |
110 | ||
9d901bc0 HX |
111 | static void bucket_table_free_rcu(struct rcu_head *head) |
112 | { | |
113 | bucket_table_free(container_of(head, struct bucket_table, rcu)); | |
114 | } | |
115 | ||
da20420f HX |
116 | static union nested_table *nested_table_alloc(struct rhashtable *ht, |
117 | union nested_table __rcu **prev, | |
5af68ef7 | 118 | bool leaf) |
da20420f HX |
119 | { |
120 | union nested_table *ntbl; | |
121 | int i; | |
122 | ||
123 | ntbl = rcu_dereference(*prev); | |
124 | if (ntbl) | |
125 | return ntbl; | |
126 | ||
127 | ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC); | |
128 | ||
5af68ef7 N |
129 | if (ntbl && leaf) { |
130 | for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0]); i++) | |
9b4f64a2 | 131 | INIT_RHT_NULLS_HEAD(ntbl[i].bucket); |
da20420f HX |
132 | } |
133 | ||
7a41c294 N |
134 | if (cmpxchg(prev, NULL, ntbl) == NULL) |
135 | return ntbl; | |
136 | /* Raced with another thread. */ | |
137 | kfree(ntbl); | |
138 | return rcu_dereference(*prev); | |
da20420f HX |
139 | } |
140 | ||
141 | static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht, | |
142 | size_t nbuckets, | |
143 | gfp_t gfp) | |
144 | { | |
145 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
146 | struct bucket_table *tbl; | |
147 | size_t size; | |
148 | ||
149 | if (nbuckets < (1 << (shift + 1))) | |
150 | return NULL; | |
151 | ||
152 | size = sizeof(*tbl) + sizeof(tbl->buckets[0]); | |
153 | ||
154 | tbl = kzalloc(size, gfp); | |
155 | if (!tbl) | |
156 | return NULL; | |
157 | ||
158 | if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets, | |
5af68ef7 | 159 | false)) { |
da20420f HX |
160 | kfree(tbl); |
161 | return NULL; | |
162 | } | |
163 | ||
164 | tbl->nest = (ilog2(nbuckets) - 1) % shift + 1; | |
165 | ||
166 | return tbl; | |
167 | } | |
168 | ||
97defe1e | 169 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
b9ecfdaa HX |
170 | size_t nbuckets, |
171 | gfp_t gfp) | |
7e1e7763 | 172 | { |
eb6d1abf | 173 | struct bucket_table *tbl = NULL; |
64e0cd0d | 174 | size_t size, max_locks; |
f89bd6f8 | 175 | int i; |
7e1e7763 TG |
176 | |
177 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
93f976b5 | 178 | tbl = kvzalloc(size, gfp); |
da20420f HX |
179 | |
180 | size = nbuckets; | |
181 | ||
2d22ecf6 | 182 | if (tbl == NULL && (gfp & ~__GFP_NOFAIL) != GFP_KERNEL) { |
da20420f HX |
183 | tbl = nested_bucket_table_alloc(ht, nbuckets, gfp); |
184 | nbuckets = 0; | |
185 | } | |
2d22ecf6 | 186 | |
7e1e7763 TG |
187 | if (tbl == NULL) |
188 | return NULL; | |
189 | ||
da20420f | 190 | tbl->size = size; |
7e1e7763 | 191 | |
64e0cd0d TH |
192 | max_locks = size >> 1; |
193 | if (tbl->nest) | |
194 | max_locks = min_t(size_t, max_locks, 1U << tbl->nest); | |
195 | ||
196 | if (alloc_bucket_spinlocks(&tbl->locks, &tbl->locks_mask, max_locks, | |
197 | ht->p.locks_mul, gfp) < 0) { | |
97defe1e TG |
198 | bucket_table_free(tbl); |
199 | return NULL; | |
200 | } | |
7e1e7763 | 201 | |
4feb7c7a | 202 | rcu_head_init(&tbl->rcu); |
eddee5ba HX |
203 | INIT_LIST_HEAD(&tbl->walkers); |
204 | ||
d48ad080 | 205 | tbl->hash_rnd = get_random_u32(); |
5269b53d | 206 | |
f89bd6f8 | 207 | for (i = 0; i < nbuckets; i++) |
9b4f64a2 | 208 | INIT_RHT_NULLS_HEAD(tbl->buckets[i]); |
f89bd6f8 | 209 | |
97defe1e | 210 | return tbl; |
7e1e7763 TG |
211 | } |
212 | ||
b824478b HX |
213 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
214 | struct bucket_table *tbl) | |
215 | { | |
216 | struct bucket_table *new_tbl; | |
217 | ||
218 | do { | |
219 | new_tbl = tbl; | |
220 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
221 | } while (tbl); | |
222 | ||
223 | return new_tbl; | |
224 | } | |
225 | ||
299e5c32 | 226 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash) |
a5ec68e3 | 227 | { |
aa34a6cb | 228 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
c0690016 | 229 | struct bucket_table *new_tbl = rhashtable_last_table(ht, old_tbl); |
da20420f HX |
230 | struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash); |
231 | int err = -EAGAIN; | |
aa34a6cb HX |
232 | struct rhash_head *head, *next, *entry; |
233 | spinlock_t *new_bucket_lock; | |
299e5c32 | 234 | unsigned int new_hash; |
aa34a6cb | 235 | |
da20420f HX |
236 | if (new_tbl->nest) |
237 | goto out; | |
238 | ||
239 | err = -ENOENT; | |
ff302db9 N |
240 | if (!pprev) |
241 | goto out; | |
da20420f | 242 | |
ff302db9 | 243 | rht_for_each_from(entry, *pprev, old_tbl, old_hash) { |
aa34a6cb HX |
244 | err = 0; |
245 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
246 | ||
247 | if (rht_is_a_nulls(next)) | |
248 | break; | |
a5ec68e3 | 249 | |
aa34a6cb HX |
250 | pprev = &entry->next; |
251 | } | |
a5ec68e3 | 252 | |
aa34a6cb HX |
253 | if (err) |
254 | goto out; | |
97defe1e | 255 | |
aa34a6cb | 256 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 257 | |
02fd97c3 | 258 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 259 | |
8f2484bd | 260 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
261 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
262 | new_tbl, new_hash); | |
97defe1e | 263 | |
7def0f95 | 264 | RCU_INIT_POINTER(entry->next, head); |
a5ec68e3 | 265 | |
aa34a6cb HX |
266 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
267 | spin_unlock(new_bucket_lock); | |
97defe1e | 268 | |
aa34a6cb | 269 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 270 | |
aa34a6cb HX |
271 | out: |
272 | return err; | |
273 | } | |
97defe1e | 274 | |
da20420f | 275 | static int rhashtable_rehash_chain(struct rhashtable *ht, |
299e5c32 | 276 | unsigned int old_hash) |
aa34a6cb HX |
277 | { |
278 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
279 | spinlock_t *old_bucket_lock; | |
da20420f | 280 | int err; |
aa34a6cb | 281 | |
02fd97c3 | 282 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 283 | |
aa34a6cb | 284 | spin_lock_bh(old_bucket_lock); |
da20420f | 285 | while (!(err = rhashtable_rehash_one(ht, old_hash))) |
aa34a6cb | 286 | ; |
da20420f | 287 | |
4feb7c7a | 288 | if (err == -ENOENT) |
da20420f | 289 | err = 0; |
4feb7c7a | 290 | |
aa34a6cb | 291 | spin_unlock_bh(old_bucket_lock); |
da20420f HX |
292 | |
293 | return err; | |
97defe1e TG |
294 | } |
295 | ||
b824478b HX |
296 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
297 | struct bucket_table *old_tbl, | |
298 | struct bucket_table *new_tbl) | |
97defe1e | 299 | { |
aa34a6cb HX |
300 | /* Make insertions go into the new, empty table right away. Deletions |
301 | * and lookups will be attempted in both tables until we synchronize. | |
0ad66449 N |
302 | * As cmpxchg() provides strong barriers, we do not need |
303 | * rcu_assign_pointer(). | |
aa34a6cb | 304 | */ |
aa34a6cb | 305 | |
0ad66449 N |
306 | if (cmpxchg(&old_tbl->future_tbl, NULL, new_tbl) != NULL) |
307 | return -EEXIST; | |
b824478b HX |
308 | |
309 | return 0; | |
310 | } | |
311 | ||
312 | static int rhashtable_rehash_table(struct rhashtable *ht) | |
313 | { | |
314 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
315 | struct bucket_table *new_tbl; | |
316 | struct rhashtable_walker *walker; | |
299e5c32 | 317 | unsigned int old_hash; |
da20420f | 318 | int err; |
b824478b HX |
319 | |
320 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); | |
321 | if (!new_tbl) | |
322 | return 0; | |
323 | ||
da20420f HX |
324 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
325 | err = rhashtable_rehash_chain(ht, old_hash); | |
326 | if (err) | |
327 | return err; | |
ae6da1f5 | 328 | cond_resched(); |
da20420f | 329 | } |
aa34a6cb HX |
330 | |
331 | /* Publish the new table pointer. */ | |
332 | rcu_assign_pointer(ht->tbl, new_tbl); | |
333 | ||
ba7c95ea | 334 | spin_lock(&ht->lock); |
eddee5ba HX |
335 | list_for_each_entry(walker, &old_tbl->walkers, list) |
336 | walker->tbl = NULL; | |
337 | ||
aa34a6cb HX |
338 | /* Wait for readers. All new readers will see the new |
339 | * table, and thus no references to the old table will | |
340 | * remain. | |
4feb7c7a N |
341 | * We do this inside the locked region so that |
342 | * rhashtable_walk_stop() can use rcu_head_after_call_rcu() | |
343 | * to check if it should not re-link the table. | |
aa34a6cb | 344 | */ |
9d901bc0 | 345 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
4feb7c7a | 346 | spin_unlock(&ht->lock); |
b824478b HX |
347 | |
348 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; | |
7e1e7763 TG |
349 | } |
350 | ||
da20420f HX |
351 | static int rhashtable_rehash_alloc(struct rhashtable *ht, |
352 | struct bucket_table *old_tbl, | |
353 | unsigned int size) | |
7e1e7763 | 354 | { |
da20420f | 355 | struct bucket_table *new_tbl; |
b824478b | 356 | int err; |
7e1e7763 TG |
357 | |
358 | ASSERT_RHT_MUTEX(ht); | |
359 | ||
da20420f | 360 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
361 | if (new_tbl == NULL) |
362 | return -ENOMEM; | |
363 | ||
b824478b HX |
364 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
365 | if (err) | |
366 | bucket_table_free(new_tbl); | |
367 | ||
368 | return err; | |
7e1e7763 | 369 | } |
7e1e7763 TG |
370 | |
371 | /** | |
372 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
373 | * @ht: the hash table to shrink | |
7e1e7763 | 374 | * |
18093d1c HX |
375 | * This function shrinks the hash table to fit, i.e., the smallest |
376 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 377 | * |
97defe1e TG |
378 | * The caller must ensure that no concurrent resizing occurs by holding |
379 | * ht->mutex. | |
380 | * | |
7e1e7763 TG |
381 | * The caller must ensure that no concurrent table mutations take place. |
382 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
383 | * |
384 | * It is valid to have concurrent insertions and deletions protected by per | |
385 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 386 | */ |
b824478b | 387 | static int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 388 | { |
da20420f | 389 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
12311959 VN |
390 | unsigned int nelems = atomic_read(&ht->nelems); |
391 | unsigned int size = 0; | |
7e1e7763 | 392 | |
12311959 VN |
393 | if (nelems) |
394 | size = roundup_pow_of_two(nelems * 3 / 2); | |
18093d1c HX |
395 | if (size < ht->p.min_size) |
396 | size = ht->p.min_size; | |
397 | ||
398 | if (old_tbl->size <= size) | |
399 | return 0; | |
400 | ||
b824478b HX |
401 | if (rht_dereference(old_tbl->future_tbl, ht)) |
402 | return -EEXIST; | |
403 | ||
da20420f | 404 | return rhashtable_rehash_alloc(ht, old_tbl, size); |
7e1e7763 | 405 | } |
7e1e7763 | 406 | |
97defe1e TG |
407 | static void rht_deferred_worker(struct work_struct *work) |
408 | { | |
409 | struct rhashtable *ht; | |
410 | struct bucket_table *tbl; | |
b824478b | 411 | int err = 0; |
97defe1e | 412 | |
57699a40 | 413 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 414 | mutex_lock(&ht->mutex); |
28134a53 | 415 | |
97defe1e | 416 | tbl = rht_dereference(ht->tbl, ht); |
b824478b | 417 | tbl = rhashtable_last_table(ht, tbl); |
97defe1e | 418 | |
a5b6846f | 419 | if (rht_grow_above_75(ht, tbl)) |
da20420f | 420 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2); |
b5e2c150 | 421 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
da20420f HX |
422 | err = rhashtable_shrink(ht); |
423 | else if (tbl->nest) | |
424 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size); | |
b824478b | 425 | |
408f13ef HX |
426 | if (!err || err == -EEXIST) { |
427 | int nerr; | |
428 | ||
429 | nerr = rhashtable_rehash_table(ht); | |
430 | err = err ?: nerr; | |
431 | } | |
b824478b | 432 | |
97defe1e | 433 | mutex_unlock(&ht->mutex); |
b824478b HX |
434 | |
435 | if (err) | |
436 | schedule_work(&ht->run_work); | |
97defe1e TG |
437 | } |
438 | ||
ca26893f HX |
439 | static int rhashtable_insert_rehash(struct rhashtable *ht, |
440 | struct bucket_table *tbl) | |
ccd57b1b HX |
441 | { |
442 | struct bucket_table *old_tbl; | |
443 | struct bucket_table *new_tbl; | |
ccd57b1b HX |
444 | unsigned int size; |
445 | int err; | |
446 | ||
447 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
ccd57b1b HX |
448 | |
449 | size = tbl->size; | |
450 | ||
3cf92222 HX |
451 | err = -EBUSY; |
452 | ||
ccd57b1b HX |
453 | if (rht_grow_above_75(ht, tbl)) |
454 | size *= 2; | |
a87b9ebf TG |
455 | /* Do not schedule more than one rehash */ |
456 | else if (old_tbl != tbl) | |
3cf92222 HX |
457 | goto fail; |
458 | ||
459 | err = -ENOMEM; | |
ccd57b1b | 460 | |
93f976b5 | 461 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC | __GFP_NOWARN); |
3cf92222 HX |
462 | if (new_tbl == NULL) |
463 | goto fail; | |
ccd57b1b HX |
464 | |
465 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); | |
466 | if (err) { | |
467 | bucket_table_free(new_tbl); | |
468 | if (err == -EEXIST) | |
469 | err = 0; | |
470 | } else | |
471 | schedule_work(&ht->run_work); | |
472 | ||
473 | return err; | |
3cf92222 HX |
474 | |
475 | fail: | |
476 | /* Do not fail the insert if someone else did a rehash. */ | |
c0690016 | 477 | if (likely(rcu_access_pointer(tbl->future_tbl))) |
3cf92222 HX |
478 | return 0; |
479 | ||
480 | /* Schedule async rehash to retry allocation in process context. */ | |
481 | if (err == -ENOMEM) | |
482 | schedule_work(&ht->run_work); | |
483 | ||
484 | return err; | |
ccd57b1b | 485 | } |
ccd57b1b | 486 | |
ca26893f HX |
487 | static void *rhashtable_lookup_one(struct rhashtable *ht, |
488 | struct bucket_table *tbl, unsigned int hash, | |
489 | const void *key, struct rhash_head *obj) | |
02fd97c3 | 490 | { |
ca26893f HX |
491 | struct rhashtable_compare_arg arg = { |
492 | .ht = ht, | |
493 | .key = key, | |
494 | }; | |
495 | struct rhash_head __rcu **pprev; | |
02fd97c3 | 496 | struct rhash_head *head; |
ca26893f | 497 | int elasticity; |
02fd97c3 | 498 | |
5f8ddeab | 499 | elasticity = RHT_ELASTICITY; |
da20420f | 500 | pprev = rht_bucket_var(tbl, hash); |
ff302db9 N |
501 | if (!pprev) |
502 | return ERR_PTR(-ENOENT); | |
f7ad68bf | 503 | rht_for_each_from(head, *pprev, tbl, hash) { |
ca26893f HX |
504 | struct rhlist_head *list; |
505 | struct rhlist_head *plist; | |
506 | ||
507 | elasticity--; | |
508 | if (!key || | |
509 | (ht->p.obj_cmpfn ? | |
510 | ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
d3dcf8eb PB |
511 | rhashtable_compare(&arg, rht_obj(ht, head)))) { |
512 | pprev = &head->next; | |
ca26893f | 513 | continue; |
d3dcf8eb | 514 | } |
ca26893f HX |
515 | |
516 | if (!ht->rhlist) | |
517 | return rht_obj(ht, head); | |
518 | ||
519 | list = container_of(obj, struct rhlist_head, rhead); | |
520 | plist = container_of(head, struct rhlist_head, rhead); | |
521 | ||
522 | RCU_INIT_POINTER(list->next, plist); | |
523 | head = rht_dereference_bucket(head->next, tbl, hash); | |
524 | RCU_INIT_POINTER(list->rhead.next, head); | |
525 | rcu_assign_pointer(*pprev, obj); | |
526 | ||
527 | return NULL; | |
5ca8cc5b | 528 | } |
02fd97c3 | 529 | |
ca26893f HX |
530 | if (elasticity <= 0) |
531 | return ERR_PTR(-EAGAIN); | |
532 | ||
533 | return ERR_PTR(-ENOENT); | |
534 | } | |
535 | ||
536 | static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht, | |
537 | struct bucket_table *tbl, | |
538 | unsigned int hash, | |
539 | struct rhash_head *obj, | |
540 | void *data) | |
541 | { | |
da20420f | 542 | struct rhash_head __rcu **pprev; |
ca26893f HX |
543 | struct bucket_table *new_tbl; |
544 | struct rhash_head *head; | |
545 | ||
546 | if (!IS_ERR_OR_NULL(data)) | |
547 | return ERR_PTR(-EEXIST); | |
07ee0722 | 548 | |
ca26893f HX |
549 | if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT) |
550 | return ERR_CAST(data); | |
ccd57b1b | 551 | |
c0690016 | 552 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
ca26893f HX |
553 | if (new_tbl) |
554 | return new_tbl; | |
555 | ||
556 | if (PTR_ERR(data) != -ENOENT) | |
557 | return ERR_CAST(data); | |
558 | ||
559 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
560 | return ERR_PTR(-E2BIG); | |
561 | ||
562 | if (unlikely(rht_grow_above_100(ht, tbl))) | |
563 | return ERR_PTR(-EAGAIN); | |
02fd97c3 | 564 | |
da20420f HX |
565 | pprev = rht_bucket_insert(ht, tbl, hash); |
566 | if (!pprev) | |
567 | return ERR_PTR(-ENOMEM); | |
568 | ||
569 | head = rht_dereference_bucket(*pprev, tbl, hash); | |
02fd97c3 HX |
570 | |
571 | RCU_INIT_POINTER(obj->next, head); | |
ca26893f HX |
572 | if (ht->rhlist) { |
573 | struct rhlist_head *list; | |
574 | ||
575 | list = container_of(obj, struct rhlist_head, rhead); | |
576 | RCU_INIT_POINTER(list->next, NULL); | |
577 | } | |
02fd97c3 | 578 | |
da20420f | 579 | rcu_assign_pointer(*pprev, obj); |
02fd97c3 HX |
580 | |
581 | atomic_inc(&ht->nelems); | |
ca26893f HX |
582 | if (rht_grow_above_75(ht, tbl)) |
583 | schedule_work(&ht->run_work); | |
02fd97c3 | 584 | |
ca26893f HX |
585 | return NULL; |
586 | } | |
02fd97c3 | 587 | |
ca26893f HX |
588 | static void *rhashtable_try_insert(struct rhashtable *ht, const void *key, |
589 | struct rhash_head *obj) | |
590 | { | |
591 | struct bucket_table *new_tbl; | |
592 | struct bucket_table *tbl; | |
593 | unsigned int hash; | |
ca26893f HX |
594 | void *data; |
595 | ||
4feb7c7a | 596 | new_tbl = rcu_dereference(ht->tbl); |
ca26893f | 597 | |
4feb7c7a | 598 | do { |
ca26893f HX |
599 | tbl = new_tbl; |
600 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
4feb7c7a | 601 | spin_lock_bh(rht_bucket_lock(tbl, hash)); |
ca26893f HX |
602 | |
603 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
604 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
605 | if (PTR_ERR(new_tbl) != -EEXIST) | |
606 | data = ERR_CAST(new_tbl); | |
607 | ||
4feb7c7a N |
608 | spin_unlock_bh(rht_bucket_lock(tbl, hash)); |
609 | } while (!IS_ERR_OR_NULL(new_tbl)); | |
ca26893f HX |
610 | |
611 | if (PTR_ERR(data) == -EAGAIN) | |
612 | data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?: | |
613 | -EAGAIN); | |
614 | ||
615 | return data; | |
616 | } | |
617 | ||
618 | void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
619 | struct rhash_head *obj) | |
620 | { | |
621 | void *data; | |
622 | ||
623 | do { | |
624 | rcu_read_lock(); | |
625 | data = rhashtable_try_insert(ht, key, obj); | |
626 | rcu_read_unlock(); | |
627 | } while (PTR_ERR(data) == -EAGAIN); | |
628 | ||
629 | return data; | |
02fd97c3 HX |
630 | } |
631 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
632 | ||
f2dba9c6 | 633 | /** |
246779dd | 634 | * rhashtable_walk_enter - Initialise an iterator |
f2dba9c6 HX |
635 | * @ht: Table to walk over |
636 | * @iter: Hash table Iterator | |
637 | * | |
638 | * This function prepares a hash table walk. | |
639 | * | |
640 | * Note that if you restart a walk after rhashtable_walk_stop you | |
641 | * may see the same object twice. Also, you may miss objects if | |
642 | * there are removals in between rhashtable_walk_stop and the next | |
643 | * call to rhashtable_walk_start. | |
644 | * | |
645 | * For a completely stable walk you should construct your own data | |
646 | * structure outside the hash table. | |
647 | * | |
82266e98 N |
648 | * This function may be called from any process context, including |
649 | * non-preemptable context, but cannot be called from softirq or | |
650 | * hardirq context. | |
f2dba9c6 | 651 | * |
246779dd | 652 | * You must call rhashtable_walk_exit after this function returns. |
f2dba9c6 | 653 | */ |
246779dd | 654 | void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter) |
f2dba9c6 HX |
655 | { |
656 | iter->ht = ht; | |
657 | iter->p = NULL; | |
658 | iter->slot = 0; | |
659 | iter->skip = 0; | |
2db54b47 | 660 | iter->end_of_table = 0; |
f2dba9c6 | 661 | |
c6ff5268 | 662 | spin_lock(&ht->lock); |
246779dd | 663 | iter->walker.tbl = |
179ccc0a | 664 | rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock)); |
246779dd | 665 | list_add(&iter->walker.list, &iter->walker.tbl->walkers); |
c6ff5268 | 666 | spin_unlock(&ht->lock); |
f2dba9c6 | 667 | } |
246779dd | 668 | EXPORT_SYMBOL_GPL(rhashtable_walk_enter); |
f2dba9c6 HX |
669 | |
670 | /** | |
671 | * rhashtable_walk_exit - Free an iterator | |
672 | * @iter: Hash table Iterator | |
673 | * | |
6c4128f6 | 674 | * This function frees resources allocated by rhashtable_walk_enter. |
f2dba9c6 HX |
675 | */ |
676 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
677 | { | |
c6ff5268 | 678 | spin_lock(&iter->ht->lock); |
246779dd HX |
679 | if (iter->walker.tbl) |
680 | list_del(&iter->walker.list); | |
c6ff5268 | 681 | spin_unlock(&iter->ht->lock); |
f2dba9c6 HX |
682 | } |
683 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
684 | ||
685 | /** | |
97a6ec4a | 686 | * rhashtable_walk_start_check - Start a hash table walk |
f2dba9c6 HX |
687 | * @iter: Hash table iterator |
688 | * | |
0647169c AG |
689 | * Start a hash table walk at the current iterator position. Note that we take |
690 | * the RCU lock in all cases including when we return an error. So you must | |
691 | * always call rhashtable_walk_stop to clean up. | |
f2dba9c6 HX |
692 | * |
693 | * Returns zero if successful. | |
694 | * | |
695 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
696 | * will rewind back to the beginning and you may use it immediately | |
697 | * by calling rhashtable_walk_next. | |
97a6ec4a TH |
698 | * |
699 | * rhashtable_walk_start is defined as an inline variant that returns | |
700 | * void. This is preferred in cases where the caller would ignore | |
701 | * resize events and always continue. | |
f2dba9c6 | 702 | */ |
97a6ec4a | 703 | int rhashtable_walk_start_check(struct rhashtable_iter *iter) |
db4374f4 | 704 | __acquires(RCU) |
f2dba9c6 | 705 | { |
eddee5ba | 706 | struct rhashtable *ht = iter->ht; |
5d240a89 | 707 | bool rhlist = ht->rhlist; |
eddee5ba | 708 | |
c6ff5268 | 709 | rcu_read_lock(); |
eddee5ba | 710 | |
c6ff5268 | 711 | spin_lock(&ht->lock); |
246779dd HX |
712 | if (iter->walker.tbl) |
713 | list_del(&iter->walker.list); | |
c6ff5268 | 714 | spin_unlock(&ht->lock); |
eddee5ba | 715 | |
5d240a89 N |
716 | if (iter->end_of_table) |
717 | return 0; | |
718 | if (!iter->walker.tbl) { | |
246779dd | 719 | iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht); |
b41cc04b N |
720 | iter->slot = 0; |
721 | iter->skip = 0; | |
f2dba9c6 HX |
722 | return -EAGAIN; |
723 | } | |
724 | ||
5d240a89 N |
725 | if (iter->p && !rhlist) { |
726 | /* | |
727 | * We need to validate that 'p' is still in the table, and | |
728 | * if so, update 'skip' | |
729 | */ | |
730 | struct rhash_head *p; | |
731 | int skip = 0; | |
732 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
733 | skip++; | |
734 | if (p == iter->p) { | |
735 | iter->skip = skip; | |
736 | goto found; | |
737 | } | |
738 | } | |
739 | iter->p = NULL; | |
740 | } else if (iter->p && rhlist) { | |
741 | /* Need to validate that 'list' is still in the table, and | |
742 | * if so, update 'skip' and 'p'. | |
743 | */ | |
744 | struct rhash_head *p; | |
745 | struct rhlist_head *list; | |
746 | int skip = 0; | |
747 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
748 | for (list = container_of(p, struct rhlist_head, rhead); | |
749 | list; | |
750 | list = rcu_dereference(list->next)) { | |
751 | skip++; | |
752 | if (list == iter->list) { | |
753 | iter->p = p; | |
c643ecf3 | 754 | iter->skip = skip; |
5d240a89 N |
755 | goto found; |
756 | } | |
757 | } | |
758 | } | |
759 | iter->p = NULL; | |
760 | } | |
761 | found: | |
f2dba9c6 HX |
762 | return 0; |
763 | } | |
97a6ec4a | 764 | EXPORT_SYMBOL_GPL(rhashtable_walk_start_check); |
f2dba9c6 HX |
765 | |
766 | /** | |
2db54b47 TH |
767 | * __rhashtable_walk_find_next - Find the next element in a table (or the first |
768 | * one in case of a new walk). | |
769 | * | |
f2dba9c6 HX |
770 | * @iter: Hash table iterator |
771 | * | |
2db54b47 | 772 | * Returns the found object or NULL when the end of the table is reached. |
f2dba9c6 | 773 | * |
2db54b47 | 774 | * Returns -EAGAIN if resize event occurred. |
f2dba9c6 | 775 | */ |
2db54b47 | 776 | static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter) |
f2dba9c6 | 777 | { |
246779dd | 778 | struct bucket_table *tbl = iter->walker.tbl; |
ca26893f | 779 | struct rhlist_head *list = iter->list; |
f2dba9c6 HX |
780 | struct rhashtable *ht = iter->ht; |
781 | struct rhash_head *p = iter->p; | |
ca26893f | 782 | bool rhlist = ht->rhlist; |
f2dba9c6 | 783 | |
2db54b47 TH |
784 | if (!tbl) |
785 | return NULL; | |
f2dba9c6 HX |
786 | |
787 | for (; iter->slot < tbl->size; iter->slot++) { | |
788 | int skip = iter->skip; | |
789 | ||
790 | rht_for_each_rcu(p, tbl, iter->slot) { | |
ca26893f HX |
791 | if (rhlist) { |
792 | list = container_of(p, struct rhlist_head, | |
793 | rhead); | |
794 | do { | |
795 | if (!skip) | |
796 | goto next; | |
797 | skip--; | |
798 | list = rcu_dereference(list->next); | |
799 | } while (list); | |
800 | ||
801 | continue; | |
802 | } | |
f2dba9c6 HX |
803 | if (!skip) |
804 | break; | |
805 | skip--; | |
806 | } | |
807 | ||
808 | next: | |
809 | if (!rht_is_a_nulls(p)) { | |
810 | iter->skip++; | |
811 | iter->p = p; | |
ca26893f HX |
812 | iter->list = list; |
813 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
f2dba9c6 HX |
814 | } |
815 | ||
816 | iter->skip = 0; | |
817 | } | |
818 | ||
142b942a PS |
819 | iter->p = NULL; |
820 | ||
d88252f9 HX |
821 | /* Ensure we see any new tables. */ |
822 | smp_rmb(); | |
823 | ||
246779dd HX |
824 | iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
825 | if (iter->walker.tbl) { | |
f2dba9c6 HX |
826 | iter->slot = 0; |
827 | iter->skip = 0; | |
f2dba9c6 | 828 | return ERR_PTR(-EAGAIN); |
2db54b47 TH |
829 | } else { |
830 | iter->end_of_table = true; | |
f2dba9c6 HX |
831 | } |
832 | ||
c936a79f | 833 | return NULL; |
f2dba9c6 | 834 | } |
2db54b47 TH |
835 | |
836 | /** | |
837 | * rhashtable_walk_next - Return the next object and advance the iterator | |
838 | * @iter: Hash table iterator | |
839 | * | |
840 | * Note that you must call rhashtable_walk_stop when you are finished | |
841 | * with the walk. | |
842 | * | |
843 | * Returns the next object or NULL when the end of the table is reached. | |
844 | * | |
845 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
846 | * will rewind back to the beginning and you may continue to use it. | |
847 | */ | |
848 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
849 | { | |
850 | struct rhlist_head *list = iter->list; | |
851 | struct rhashtable *ht = iter->ht; | |
852 | struct rhash_head *p = iter->p; | |
853 | bool rhlist = ht->rhlist; | |
854 | ||
855 | if (p) { | |
856 | if (!rhlist || !(list = rcu_dereference(list->next))) { | |
857 | p = rcu_dereference(p->next); | |
858 | list = container_of(p, struct rhlist_head, rhead); | |
859 | } | |
860 | if (!rht_is_a_nulls(p)) { | |
861 | iter->skip++; | |
862 | iter->p = p; | |
863 | iter->list = list; | |
864 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
865 | } | |
866 | ||
867 | /* At the end of this slot, switch to next one and then find | |
868 | * next entry from that point. | |
869 | */ | |
870 | iter->skip = 0; | |
871 | iter->slot++; | |
872 | } | |
873 | ||
874 | return __rhashtable_walk_find_next(iter); | |
875 | } | |
f2dba9c6 HX |
876 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); |
877 | ||
2db54b47 TH |
878 | /** |
879 | * rhashtable_walk_peek - Return the next object but don't advance the iterator | |
880 | * @iter: Hash table iterator | |
881 | * | |
882 | * Returns the next object or NULL when the end of the table is reached. | |
883 | * | |
884 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
885 | * will rewind back to the beginning and you may continue to use it. | |
886 | */ | |
887 | void *rhashtable_walk_peek(struct rhashtable_iter *iter) | |
888 | { | |
889 | struct rhlist_head *list = iter->list; | |
890 | struct rhashtable *ht = iter->ht; | |
891 | struct rhash_head *p = iter->p; | |
892 | ||
893 | if (p) | |
894 | return rht_obj(ht, ht->rhlist ? &list->rhead : p); | |
895 | ||
896 | /* No object found in current iter, find next one in the table. */ | |
897 | ||
898 | if (iter->skip) { | |
899 | /* A nonzero skip value points to the next entry in the table | |
900 | * beyond that last one that was found. Decrement skip so | |
901 | * we find the current value. __rhashtable_walk_find_next | |
902 | * will restore the original value of skip assuming that | |
903 | * the table hasn't changed. | |
904 | */ | |
905 | iter->skip--; | |
906 | } | |
907 | ||
908 | return __rhashtable_walk_find_next(iter); | |
909 | } | |
910 | EXPORT_SYMBOL_GPL(rhashtable_walk_peek); | |
911 | ||
f2dba9c6 HX |
912 | /** |
913 | * rhashtable_walk_stop - Finish a hash table walk | |
914 | * @iter: Hash table iterator | |
915 | * | |
0647169c AG |
916 | * Finish a hash table walk. Does not reset the iterator to the start of the |
917 | * hash table. | |
f2dba9c6 HX |
918 | */ |
919 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 920 | __releases(RCU) |
f2dba9c6 | 921 | { |
eddee5ba | 922 | struct rhashtable *ht; |
246779dd | 923 | struct bucket_table *tbl = iter->walker.tbl; |
eddee5ba | 924 | |
eddee5ba | 925 | if (!tbl) |
963ecbd4 | 926 | goto out; |
eddee5ba HX |
927 | |
928 | ht = iter->ht; | |
929 | ||
ba7c95ea | 930 | spin_lock(&ht->lock); |
4feb7c7a N |
931 | if (rcu_head_after_call_rcu(&tbl->rcu, bucket_table_free_rcu)) |
932 | /* This bucket table is being freed, don't re-link it. */ | |
246779dd | 933 | iter->walker.tbl = NULL; |
4feb7c7a N |
934 | else |
935 | list_add(&iter->walker.list, &tbl->walkers); | |
ba7c95ea | 936 | spin_unlock(&ht->lock); |
eddee5ba | 937 | |
963ecbd4 HX |
938 | out: |
939 | rcu_read_unlock(); | |
f2dba9c6 HX |
940 | } |
941 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
942 | ||
488fb86e | 943 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 944 | { |
107d01f5 DB |
945 | size_t retsize; |
946 | ||
947 | if (params->nelem_hint) | |
948 | retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3), | |
949 | (unsigned long)params->min_size); | |
950 | else | |
951 | retsize = max(HASH_DEFAULT_SIZE, | |
952 | (unsigned long)params->min_size); | |
953 | ||
954 | return retsize; | |
7e1e7763 TG |
955 | } |
956 | ||
31ccde2d HX |
957 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
958 | { | |
959 | return jhash2(key, length, seed); | |
960 | } | |
961 | ||
7e1e7763 TG |
962 | /** |
963 | * rhashtable_init - initialize a new hash table | |
964 | * @ht: hash table to be initialized | |
965 | * @params: configuration parameters | |
966 | * | |
967 | * Initializes a new hash table based on the provided configuration | |
968 | * parameters. A table can be configured either with a variable or | |
969 | * fixed length key: | |
970 | * | |
971 | * Configuration Example 1: Fixed length keys | |
972 | * struct test_obj { | |
973 | * int key; | |
974 | * void * my_member; | |
975 | * struct rhash_head node; | |
976 | * }; | |
977 | * | |
978 | * struct rhashtable_params params = { | |
979 | * .head_offset = offsetof(struct test_obj, node), | |
980 | * .key_offset = offsetof(struct test_obj, key), | |
981 | * .key_len = sizeof(int), | |
87545899 | 982 | * .hashfn = jhash, |
7e1e7763 TG |
983 | * }; |
984 | * | |
985 | * Configuration Example 2: Variable length keys | |
986 | * struct test_obj { | |
987 | * [...] | |
988 | * struct rhash_head node; | |
989 | * }; | |
990 | * | |
49f7b33e | 991 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
992 | * { |
993 | * struct test_obj *obj = data; | |
994 | * | |
995 | * return [... hash ...]; | |
996 | * } | |
997 | * | |
998 | * struct rhashtable_params params = { | |
999 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 1000 | * .hashfn = jhash, |
7e1e7763 | 1001 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
1002 | * }; |
1003 | */ | |
488fb86e HX |
1004 | int rhashtable_init(struct rhashtable *ht, |
1005 | const struct rhashtable_params *params) | |
7e1e7763 TG |
1006 | { |
1007 | struct bucket_table *tbl; | |
1008 | size_t size; | |
1009 | ||
31ccde2d | 1010 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 1011 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
1012 | return -EINVAL; |
1013 | ||
97defe1e TG |
1014 | memset(ht, 0, sizeof(*ht)); |
1015 | mutex_init(&ht->mutex); | |
ba7c95ea | 1016 | spin_lock_init(&ht->lock); |
97defe1e TG |
1017 | memcpy(&ht->p, params, sizeof(*params)); |
1018 | ||
a998f712 TG |
1019 | if (params->min_size) |
1020 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
1021 | ||
6d684e54 HX |
1022 | /* Cap total entries at 2^31 to avoid nelems overflow. */ |
1023 | ht->max_elems = 1u << 31; | |
2d2ab658 HX |
1024 | |
1025 | if (params->max_size) { | |
1026 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
1027 | if (ht->p.max_size < ht->max_elems / 2) | |
1028 | ht->max_elems = ht->p.max_size * 2; | |
1029 | } | |
6d684e54 | 1030 | |
48e75b43 | 1031 | ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 1032 | |
107d01f5 | 1033 | size = rounded_hashtable_size(&ht->p); |
3a324606 | 1034 | |
97defe1e TG |
1035 | if (params->locks_mul) |
1036 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
1037 | else | |
1038 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
1039 | ||
31ccde2d HX |
1040 | ht->key_len = ht->p.key_len; |
1041 | if (!params->hashfn) { | |
1042 | ht->p.hashfn = jhash; | |
1043 | ||
1044 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
1045 | ht->key_len /= sizeof(u32); | |
1046 | ht->p.hashfn = rhashtable_jhash2; | |
1047 | } | |
1048 | } | |
1049 | ||
2d22ecf6 DB |
1050 | /* |
1051 | * This is api initialization and thus we need to guarantee the | |
1052 | * initial rhashtable allocation. Upon failure, retry with the | |
1053 | * smallest possible size with __GFP_NOFAIL semantics. | |
1054 | */ | |
b9ecfdaa | 1055 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
2d22ecf6 DB |
1056 | if (unlikely(tbl == NULL)) { |
1057 | size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); | |
1058 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL); | |
1059 | } | |
7e1e7763 | 1060 | |
545a148e | 1061 | atomic_set(&ht->nelems, 0); |
a5b6846f | 1062 | |
7e1e7763 TG |
1063 | RCU_INIT_POINTER(ht->tbl, tbl); |
1064 | ||
4c4b52d9 | 1065 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1066 | |
7e1e7763 TG |
1067 | return 0; |
1068 | } | |
1069 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1070 | ||
ca26893f HX |
1071 | /** |
1072 | * rhltable_init - initialize a new hash list table | |
1073 | * @hlt: hash list table to be initialized | |
1074 | * @params: configuration parameters | |
1075 | * | |
1076 | * Initializes a new hash list table. | |
1077 | * | |
1078 | * See documentation for rhashtable_init. | |
1079 | */ | |
1080 | int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params) | |
1081 | { | |
1082 | int err; | |
1083 | ||
ca26893f HX |
1084 | err = rhashtable_init(&hlt->ht, params); |
1085 | hlt->ht.rhlist = true; | |
1086 | return err; | |
1087 | } | |
1088 | EXPORT_SYMBOL_GPL(rhltable_init); | |
1089 | ||
1090 | static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj, | |
1091 | void (*free_fn)(void *ptr, void *arg), | |
1092 | void *arg) | |
1093 | { | |
1094 | struct rhlist_head *list; | |
1095 | ||
1096 | if (!ht->rhlist) { | |
1097 | free_fn(rht_obj(ht, obj), arg); | |
1098 | return; | |
1099 | } | |
1100 | ||
1101 | list = container_of(obj, struct rhlist_head, rhead); | |
1102 | do { | |
1103 | obj = &list->rhead; | |
1104 | list = rht_dereference(list->next, ht); | |
1105 | free_fn(rht_obj(ht, obj), arg); | |
1106 | } while (list); | |
1107 | } | |
1108 | ||
7e1e7763 | 1109 | /** |
6b6f302c | 1110 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 1111 | * @ht: the hash table to destroy |
6b6f302c TG |
1112 | * @free_fn: callback to release resources of element |
1113 | * @arg: pointer passed to free_fn | |
7e1e7763 | 1114 | * |
6b6f302c TG |
1115 | * Stops an eventual async resize. If defined, invokes free_fn for each |
1116 | * element to releasal resources. Please note that RCU protected | |
1117 | * readers may still be accessing the elements. Releasing of resources | |
1118 | * must occur in a compatible manner. Then frees the bucket array. | |
1119 | * | |
1120 | * This function will eventually sleep to wait for an async resize | |
1121 | * to complete. The caller is responsible that no further write operations | |
1122 | * occurs in parallel. | |
7e1e7763 | 1123 | */ |
6b6f302c TG |
1124 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
1125 | void (*free_fn)(void *ptr, void *arg), | |
1126 | void *arg) | |
7e1e7763 | 1127 | { |
0026129c | 1128 | struct bucket_table *tbl, *next_tbl; |
6b6f302c | 1129 | unsigned int i; |
97defe1e | 1130 | |
4c4b52d9 | 1131 | cancel_work_sync(&ht->run_work); |
97defe1e | 1132 | |
57699a40 | 1133 | mutex_lock(&ht->mutex); |
6b6f302c | 1134 | tbl = rht_dereference(ht->tbl, ht); |
0026129c | 1135 | restart: |
6b6f302c TG |
1136 | if (free_fn) { |
1137 | for (i = 0; i < tbl->size; i++) { | |
1138 | struct rhash_head *pos, *next; | |
1139 | ||
ae6da1f5 | 1140 | cond_resched(); |
da20420f | 1141 | for (pos = rht_dereference(*rht_bucket(tbl, i), ht), |
6b6f302c TG |
1142 | next = !rht_is_a_nulls(pos) ? |
1143 | rht_dereference(pos->next, ht) : NULL; | |
1144 | !rht_is_a_nulls(pos); | |
1145 | pos = next, | |
1146 | next = !rht_is_a_nulls(pos) ? | |
1147 | rht_dereference(pos->next, ht) : NULL) | |
ca26893f | 1148 | rhashtable_free_one(ht, pos, free_fn, arg); |
6b6f302c TG |
1149 | } |
1150 | } | |
1151 | ||
0026129c | 1152 | next_tbl = rht_dereference(tbl->future_tbl, ht); |
6b6f302c | 1153 | bucket_table_free(tbl); |
0026129c TY |
1154 | if (next_tbl) { |
1155 | tbl = next_tbl; | |
1156 | goto restart; | |
1157 | } | |
97defe1e | 1158 | mutex_unlock(&ht->mutex); |
7e1e7763 | 1159 | } |
6b6f302c TG |
1160 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
1161 | ||
1162 | void rhashtable_destroy(struct rhashtable *ht) | |
1163 | { | |
1164 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
1165 | } | |
7e1e7763 | 1166 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |
da20420f | 1167 | |
ff302db9 N |
1168 | struct rhash_head __rcu **__rht_bucket_nested(const struct bucket_table *tbl, |
1169 | unsigned int hash) | |
da20420f HX |
1170 | { |
1171 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
da20420f HX |
1172 | unsigned int index = hash & ((1 << tbl->nest) - 1); |
1173 | unsigned int size = tbl->size >> tbl->nest; | |
1174 | unsigned int subhash = hash; | |
1175 | union nested_table *ntbl; | |
1176 | ||
1177 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
c4d2603d | 1178 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash); |
da20420f HX |
1179 | subhash >>= tbl->nest; |
1180 | ||
1181 | while (ntbl && size > (1 << shift)) { | |
1182 | index = subhash & ((1 << shift) - 1); | |
c4d2603d HX |
1183 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, |
1184 | tbl, hash); | |
da20420f HX |
1185 | size >>= shift; |
1186 | subhash >>= shift; | |
1187 | } | |
1188 | ||
ff302db9 N |
1189 | if (!ntbl) |
1190 | return NULL; | |
da20420f HX |
1191 | |
1192 | return &ntbl[subhash].bucket; | |
1193 | ||
1194 | } | |
ff302db9 N |
1195 | EXPORT_SYMBOL_GPL(__rht_bucket_nested); |
1196 | ||
1197 | struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl, | |
1198 | unsigned int hash) | |
1199 | { | |
1200 | static struct rhash_head __rcu *rhnull; | |
1201 | ||
1202 | if (!rhnull) | |
1203 | INIT_RHT_NULLS_HEAD(rhnull); | |
1204 | return __rht_bucket_nested(tbl, hash) ?: &rhnull; | |
1205 | } | |
da20420f HX |
1206 | EXPORT_SYMBOL_GPL(rht_bucket_nested); |
1207 | ||
1208 | struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht, | |
1209 | struct bucket_table *tbl, | |
1210 | unsigned int hash) | |
1211 | { | |
1212 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1213 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1214 | unsigned int size = tbl->size >> tbl->nest; | |
1215 | union nested_table *ntbl; | |
da20420f HX |
1216 | |
1217 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
1218 | hash >>= tbl->nest; | |
da20420f | 1219 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1220 | size <= (1 << shift)); |
da20420f HX |
1221 | |
1222 | while (ntbl && size > (1 << shift)) { | |
1223 | index = hash & ((1 << shift) - 1); | |
1224 | size >>= shift; | |
1225 | hash >>= shift; | |
da20420f | 1226 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1227 | size <= (1 << shift)); |
da20420f HX |
1228 | } |
1229 | ||
1230 | if (!ntbl) | |
1231 | return NULL; | |
1232 | ||
1233 | return &ntbl[hash].bucket; | |
1234 | ||
1235 | } | |
1236 | EXPORT_SYMBOL_GPL(rht_bucket_nested_insert); |