]>
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; | |
240 | ||
aa34a6cb HX |
241 | rht_for_each(entry, old_tbl, old_hash) { |
242 | err = 0; | |
243 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
244 | ||
245 | if (rht_is_a_nulls(next)) | |
246 | break; | |
a5ec68e3 | 247 | |
aa34a6cb HX |
248 | pprev = &entry->next; |
249 | } | |
a5ec68e3 | 250 | |
aa34a6cb HX |
251 | if (err) |
252 | goto out; | |
97defe1e | 253 | |
aa34a6cb | 254 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 255 | |
02fd97c3 | 256 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 257 | |
8f2484bd | 258 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
259 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
260 | new_tbl, new_hash); | |
97defe1e | 261 | |
7def0f95 | 262 | RCU_INIT_POINTER(entry->next, head); |
a5ec68e3 | 263 | |
aa34a6cb HX |
264 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
265 | spin_unlock(new_bucket_lock); | |
97defe1e | 266 | |
aa34a6cb | 267 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 268 | |
aa34a6cb HX |
269 | out: |
270 | return err; | |
271 | } | |
97defe1e | 272 | |
da20420f | 273 | static int rhashtable_rehash_chain(struct rhashtable *ht, |
299e5c32 | 274 | unsigned int old_hash) |
aa34a6cb HX |
275 | { |
276 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
277 | spinlock_t *old_bucket_lock; | |
da20420f | 278 | int err; |
aa34a6cb | 279 | |
02fd97c3 | 280 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 281 | |
aa34a6cb | 282 | spin_lock_bh(old_bucket_lock); |
da20420f | 283 | while (!(err = rhashtable_rehash_one(ht, old_hash))) |
aa34a6cb | 284 | ; |
da20420f | 285 | |
4feb7c7a | 286 | if (err == -ENOENT) |
da20420f | 287 | err = 0; |
4feb7c7a | 288 | |
aa34a6cb | 289 | spin_unlock_bh(old_bucket_lock); |
da20420f HX |
290 | |
291 | return err; | |
97defe1e TG |
292 | } |
293 | ||
b824478b HX |
294 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
295 | struct bucket_table *old_tbl, | |
296 | struct bucket_table *new_tbl) | |
97defe1e | 297 | { |
aa34a6cb HX |
298 | /* Make insertions go into the new, empty table right away. Deletions |
299 | * and lookups will be attempted in both tables until we synchronize. | |
0ad66449 N |
300 | * As cmpxchg() provides strong barriers, we do not need |
301 | * rcu_assign_pointer(). | |
aa34a6cb | 302 | */ |
aa34a6cb | 303 | |
0ad66449 N |
304 | if (cmpxchg(&old_tbl->future_tbl, NULL, new_tbl) != NULL) |
305 | return -EEXIST; | |
b824478b HX |
306 | |
307 | return 0; | |
308 | } | |
309 | ||
310 | static int rhashtable_rehash_table(struct rhashtable *ht) | |
311 | { | |
312 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
313 | struct bucket_table *new_tbl; | |
314 | struct rhashtable_walker *walker; | |
299e5c32 | 315 | unsigned int old_hash; |
da20420f | 316 | int err; |
b824478b HX |
317 | |
318 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); | |
319 | if (!new_tbl) | |
320 | return 0; | |
321 | ||
da20420f HX |
322 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
323 | err = rhashtable_rehash_chain(ht, old_hash); | |
324 | if (err) | |
325 | return err; | |
ae6da1f5 | 326 | cond_resched(); |
da20420f | 327 | } |
aa34a6cb HX |
328 | |
329 | /* Publish the new table pointer. */ | |
330 | rcu_assign_pointer(ht->tbl, new_tbl); | |
331 | ||
ba7c95ea | 332 | spin_lock(&ht->lock); |
eddee5ba HX |
333 | list_for_each_entry(walker, &old_tbl->walkers, list) |
334 | walker->tbl = NULL; | |
335 | ||
aa34a6cb HX |
336 | /* Wait for readers. All new readers will see the new |
337 | * table, and thus no references to the old table will | |
338 | * remain. | |
4feb7c7a N |
339 | * We do this inside the locked region so that |
340 | * rhashtable_walk_stop() can use rcu_head_after_call_rcu() | |
341 | * to check if it should not re-link the table. | |
aa34a6cb | 342 | */ |
9d901bc0 | 343 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
4feb7c7a | 344 | spin_unlock(&ht->lock); |
b824478b HX |
345 | |
346 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; | |
7e1e7763 TG |
347 | } |
348 | ||
da20420f HX |
349 | static int rhashtable_rehash_alloc(struct rhashtable *ht, |
350 | struct bucket_table *old_tbl, | |
351 | unsigned int size) | |
7e1e7763 | 352 | { |
da20420f | 353 | struct bucket_table *new_tbl; |
b824478b | 354 | int err; |
7e1e7763 TG |
355 | |
356 | ASSERT_RHT_MUTEX(ht); | |
357 | ||
da20420f | 358 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
359 | if (new_tbl == NULL) |
360 | return -ENOMEM; | |
361 | ||
b824478b HX |
362 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
363 | if (err) | |
364 | bucket_table_free(new_tbl); | |
365 | ||
366 | return err; | |
7e1e7763 | 367 | } |
7e1e7763 TG |
368 | |
369 | /** | |
370 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
371 | * @ht: the hash table to shrink | |
7e1e7763 | 372 | * |
18093d1c HX |
373 | * This function shrinks the hash table to fit, i.e., the smallest |
374 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 375 | * |
97defe1e TG |
376 | * The caller must ensure that no concurrent resizing occurs by holding |
377 | * ht->mutex. | |
378 | * | |
7e1e7763 TG |
379 | * The caller must ensure that no concurrent table mutations take place. |
380 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
381 | * |
382 | * It is valid to have concurrent insertions and deletions protected by per | |
383 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 384 | */ |
b824478b | 385 | static int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 386 | { |
da20420f | 387 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
12311959 VN |
388 | unsigned int nelems = atomic_read(&ht->nelems); |
389 | unsigned int size = 0; | |
7e1e7763 | 390 | |
12311959 VN |
391 | if (nelems) |
392 | size = roundup_pow_of_two(nelems * 3 / 2); | |
18093d1c HX |
393 | if (size < ht->p.min_size) |
394 | size = ht->p.min_size; | |
395 | ||
396 | if (old_tbl->size <= size) | |
397 | return 0; | |
398 | ||
b824478b HX |
399 | if (rht_dereference(old_tbl->future_tbl, ht)) |
400 | return -EEXIST; | |
401 | ||
da20420f | 402 | return rhashtable_rehash_alloc(ht, old_tbl, size); |
7e1e7763 | 403 | } |
7e1e7763 | 404 | |
97defe1e TG |
405 | static void rht_deferred_worker(struct work_struct *work) |
406 | { | |
407 | struct rhashtable *ht; | |
408 | struct bucket_table *tbl; | |
b824478b | 409 | int err = 0; |
97defe1e | 410 | |
57699a40 | 411 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 412 | mutex_lock(&ht->mutex); |
28134a53 | 413 | |
97defe1e | 414 | tbl = rht_dereference(ht->tbl, ht); |
b824478b | 415 | tbl = rhashtable_last_table(ht, tbl); |
97defe1e | 416 | |
a5b6846f | 417 | if (rht_grow_above_75(ht, tbl)) |
da20420f | 418 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2); |
b5e2c150 | 419 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
da20420f HX |
420 | err = rhashtable_shrink(ht); |
421 | else if (tbl->nest) | |
422 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size); | |
b824478b | 423 | |
408f13ef HX |
424 | if (!err || err == -EEXIST) { |
425 | int nerr; | |
426 | ||
427 | nerr = rhashtable_rehash_table(ht); | |
428 | err = err ?: nerr; | |
429 | } | |
b824478b | 430 | |
97defe1e | 431 | mutex_unlock(&ht->mutex); |
b824478b HX |
432 | |
433 | if (err) | |
434 | schedule_work(&ht->run_work); | |
97defe1e TG |
435 | } |
436 | ||
ca26893f HX |
437 | static int rhashtable_insert_rehash(struct rhashtable *ht, |
438 | struct bucket_table *tbl) | |
ccd57b1b HX |
439 | { |
440 | struct bucket_table *old_tbl; | |
441 | struct bucket_table *new_tbl; | |
ccd57b1b HX |
442 | unsigned int size; |
443 | int err; | |
444 | ||
445 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
ccd57b1b HX |
446 | |
447 | size = tbl->size; | |
448 | ||
3cf92222 HX |
449 | err = -EBUSY; |
450 | ||
ccd57b1b HX |
451 | if (rht_grow_above_75(ht, tbl)) |
452 | size *= 2; | |
a87b9ebf TG |
453 | /* Do not schedule more than one rehash */ |
454 | else if (old_tbl != tbl) | |
3cf92222 HX |
455 | goto fail; |
456 | ||
457 | err = -ENOMEM; | |
ccd57b1b | 458 | |
93f976b5 | 459 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC | __GFP_NOWARN); |
3cf92222 HX |
460 | if (new_tbl == NULL) |
461 | goto fail; | |
ccd57b1b HX |
462 | |
463 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); | |
464 | if (err) { | |
465 | bucket_table_free(new_tbl); | |
466 | if (err == -EEXIST) | |
467 | err = 0; | |
468 | } else | |
469 | schedule_work(&ht->run_work); | |
470 | ||
471 | return err; | |
3cf92222 HX |
472 | |
473 | fail: | |
474 | /* Do not fail the insert if someone else did a rehash. */ | |
c0690016 | 475 | if (likely(rcu_access_pointer(tbl->future_tbl))) |
3cf92222 HX |
476 | return 0; |
477 | ||
478 | /* Schedule async rehash to retry allocation in process context. */ | |
479 | if (err == -ENOMEM) | |
480 | schedule_work(&ht->run_work); | |
481 | ||
482 | return err; | |
ccd57b1b | 483 | } |
ccd57b1b | 484 | |
ca26893f HX |
485 | static void *rhashtable_lookup_one(struct rhashtable *ht, |
486 | struct bucket_table *tbl, unsigned int hash, | |
487 | const void *key, struct rhash_head *obj) | |
02fd97c3 | 488 | { |
ca26893f HX |
489 | struct rhashtable_compare_arg arg = { |
490 | .ht = ht, | |
491 | .key = key, | |
492 | }; | |
493 | struct rhash_head __rcu **pprev; | |
02fd97c3 | 494 | struct rhash_head *head; |
ca26893f | 495 | int elasticity; |
02fd97c3 | 496 | |
5f8ddeab | 497 | elasticity = RHT_ELASTICITY; |
da20420f | 498 | pprev = rht_bucket_var(tbl, hash); |
f7ad68bf | 499 | rht_for_each_from(head, *pprev, tbl, hash) { |
ca26893f HX |
500 | struct rhlist_head *list; |
501 | struct rhlist_head *plist; | |
502 | ||
503 | elasticity--; | |
504 | if (!key || | |
505 | (ht->p.obj_cmpfn ? | |
506 | ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
d3dcf8eb PB |
507 | rhashtable_compare(&arg, rht_obj(ht, head)))) { |
508 | pprev = &head->next; | |
ca26893f | 509 | continue; |
d3dcf8eb | 510 | } |
ca26893f HX |
511 | |
512 | if (!ht->rhlist) | |
513 | return rht_obj(ht, head); | |
514 | ||
515 | list = container_of(obj, struct rhlist_head, rhead); | |
516 | plist = container_of(head, struct rhlist_head, rhead); | |
517 | ||
518 | RCU_INIT_POINTER(list->next, plist); | |
519 | head = rht_dereference_bucket(head->next, tbl, hash); | |
520 | RCU_INIT_POINTER(list->rhead.next, head); | |
521 | rcu_assign_pointer(*pprev, obj); | |
522 | ||
523 | return NULL; | |
5ca8cc5b | 524 | } |
02fd97c3 | 525 | |
ca26893f HX |
526 | if (elasticity <= 0) |
527 | return ERR_PTR(-EAGAIN); | |
528 | ||
529 | return ERR_PTR(-ENOENT); | |
530 | } | |
531 | ||
532 | static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht, | |
533 | struct bucket_table *tbl, | |
534 | unsigned int hash, | |
535 | struct rhash_head *obj, | |
536 | void *data) | |
537 | { | |
da20420f | 538 | struct rhash_head __rcu **pprev; |
ca26893f HX |
539 | struct bucket_table *new_tbl; |
540 | struct rhash_head *head; | |
541 | ||
542 | if (!IS_ERR_OR_NULL(data)) | |
543 | return ERR_PTR(-EEXIST); | |
07ee0722 | 544 | |
ca26893f HX |
545 | if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT) |
546 | return ERR_CAST(data); | |
ccd57b1b | 547 | |
c0690016 | 548 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
ca26893f HX |
549 | if (new_tbl) |
550 | return new_tbl; | |
551 | ||
552 | if (PTR_ERR(data) != -ENOENT) | |
553 | return ERR_CAST(data); | |
554 | ||
555 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
556 | return ERR_PTR(-E2BIG); | |
557 | ||
558 | if (unlikely(rht_grow_above_100(ht, tbl))) | |
559 | return ERR_PTR(-EAGAIN); | |
02fd97c3 | 560 | |
da20420f HX |
561 | pprev = rht_bucket_insert(ht, tbl, hash); |
562 | if (!pprev) | |
563 | return ERR_PTR(-ENOMEM); | |
564 | ||
565 | head = rht_dereference_bucket(*pprev, tbl, hash); | |
02fd97c3 HX |
566 | |
567 | RCU_INIT_POINTER(obj->next, head); | |
ca26893f HX |
568 | if (ht->rhlist) { |
569 | struct rhlist_head *list; | |
570 | ||
571 | list = container_of(obj, struct rhlist_head, rhead); | |
572 | RCU_INIT_POINTER(list->next, NULL); | |
573 | } | |
02fd97c3 | 574 | |
da20420f | 575 | rcu_assign_pointer(*pprev, obj); |
02fd97c3 HX |
576 | |
577 | atomic_inc(&ht->nelems); | |
ca26893f HX |
578 | if (rht_grow_above_75(ht, tbl)) |
579 | schedule_work(&ht->run_work); | |
02fd97c3 | 580 | |
ca26893f HX |
581 | return NULL; |
582 | } | |
02fd97c3 | 583 | |
ca26893f HX |
584 | static void *rhashtable_try_insert(struct rhashtable *ht, const void *key, |
585 | struct rhash_head *obj) | |
586 | { | |
587 | struct bucket_table *new_tbl; | |
588 | struct bucket_table *tbl; | |
589 | unsigned int hash; | |
ca26893f HX |
590 | void *data; |
591 | ||
4feb7c7a | 592 | new_tbl = rcu_dereference(ht->tbl); |
ca26893f | 593 | |
4feb7c7a | 594 | do { |
ca26893f HX |
595 | tbl = new_tbl; |
596 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
4feb7c7a | 597 | spin_lock_bh(rht_bucket_lock(tbl, hash)); |
ca26893f HX |
598 | |
599 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
600 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
601 | if (PTR_ERR(new_tbl) != -EEXIST) | |
602 | data = ERR_CAST(new_tbl); | |
603 | ||
4feb7c7a N |
604 | spin_unlock_bh(rht_bucket_lock(tbl, hash)); |
605 | } while (!IS_ERR_OR_NULL(new_tbl)); | |
ca26893f HX |
606 | |
607 | if (PTR_ERR(data) == -EAGAIN) | |
608 | data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?: | |
609 | -EAGAIN); | |
610 | ||
611 | return data; | |
612 | } | |
613 | ||
614 | void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
615 | struct rhash_head *obj) | |
616 | { | |
617 | void *data; | |
618 | ||
619 | do { | |
620 | rcu_read_lock(); | |
621 | data = rhashtable_try_insert(ht, key, obj); | |
622 | rcu_read_unlock(); | |
623 | } while (PTR_ERR(data) == -EAGAIN); | |
624 | ||
625 | return data; | |
02fd97c3 HX |
626 | } |
627 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
628 | ||
f2dba9c6 | 629 | /** |
246779dd | 630 | * rhashtable_walk_enter - Initialise an iterator |
f2dba9c6 HX |
631 | * @ht: Table to walk over |
632 | * @iter: Hash table Iterator | |
633 | * | |
634 | * This function prepares a hash table walk. | |
635 | * | |
636 | * Note that if you restart a walk after rhashtable_walk_stop you | |
637 | * may see the same object twice. Also, you may miss objects if | |
638 | * there are removals in between rhashtable_walk_stop and the next | |
639 | * call to rhashtable_walk_start. | |
640 | * | |
641 | * For a completely stable walk you should construct your own data | |
642 | * structure outside the hash table. | |
643 | * | |
82266e98 N |
644 | * This function may be called from any process context, including |
645 | * non-preemptable context, but cannot be called from softirq or | |
646 | * hardirq context. | |
f2dba9c6 | 647 | * |
246779dd | 648 | * You must call rhashtable_walk_exit after this function returns. |
f2dba9c6 | 649 | */ |
246779dd | 650 | void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter) |
f2dba9c6 HX |
651 | { |
652 | iter->ht = ht; | |
653 | iter->p = NULL; | |
654 | iter->slot = 0; | |
655 | iter->skip = 0; | |
2db54b47 | 656 | iter->end_of_table = 0; |
f2dba9c6 | 657 | |
c6ff5268 | 658 | spin_lock(&ht->lock); |
246779dd | 659 | iter->walker.tbl = |
179ccc0a | 660 | rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock)); |
246779dd | 661 | list_add(&iter->walker.list, &iter->walker.tbl->walkers); |
c6ff5268 | 662 | spin_unlock(&ht->lock); |
f2dba9c6 | 663 | } |
246779dd | 664 | EXPORT_SYMBOL_GPL(rhashtable_walk_enter); |
f2dba9c6 HX |
665 | |
666 | /** | |
667 | * rhashtable_walk_exit - Free an iterator | |
668 | * @iter: Hash table Iterator | |
669 | * | |
6c4128f6 | 670 | * This function frees resources allocated by rhashtable_walk_enter. |
f2dba9c6 HX |
671 | */ |
672 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
673 | { | |
c6ff5268 | 674 | spin_lock(&iter->ht->lock); |
246779dd HX |
675 | if (iter->walker.tbl) |
676 | list_del(&iter->walker.list); | |
c6ff5268 | 677 | spin_unlock(&iter->ht->lock); |
f2dba9c6 HX |
678 | } |
679 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
680 | ||
681 | /** | |
97a6ec4a | 682 | * rhashtable_walk_start_check - Start a hash table walk |
f2dba9c6 HX |
683 | * @iter: Hash table iterator |
684 | * | |
0647169c AG |
685 | * Start a hash table walk at the current iterator position. Note that we take |
686 | * the RCU lock in all cases including when we return an error. So you must | |
687 | * always call rhashtable_walk_stop to clean up. | |
f2dba9c6 HX |
688 | * |
689 | * Returns zero if successful. | |
690 | * | |
691 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
692 | * will rewind back to the beginning and you may use it immediately | |
693 | * by calling rhashtable_walk_next. | |
97a6ec4a TH |
694 | * |
695 | * rhashtable_walk_start is defined as an inline variant that returns | |
696 | * void. This is preferred in cases where the caller would ignore | |
697 | * resize events and always continue. | |
f2dba9c6 | 698 | */ |
97a6ec4a | 699 | int rhashtable_walk_start_check(struct rhashtable_iter *iter) |
db4374f4 | 700 | __acquires(RCU) |
f2dba9c6 | 701 | { |
eddee5ba | 702 | struct rhashtable *ht = iter->ht; |
5d240a89 | 703 | bool rhlist = ht->rhlist; |
eddee5ba | 704 | |
c6ff5268 | 705 | rcu_read_lock(); |
eddee5ba | 706 | |
c6ff5268 | 707 | spin_lock(&ht->lock); |
246779dd HX |
708 | if (iter->walker.tbl) |
709 | list_del(&iter->walker.list); | |
c6ff5268 | 710 | spin_unlock(&ht->lock); |
eddee5ba | 711 | |
5d240a89 N |
712 | if (iter->end_of_table) |
713 | return 0; | |
714 | if (!iter->walker.tbl) { | |
246779dd | 715 | iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht); |
b41cc04b N |
716 | iter->slot = 0; |
717 | iter->skip = 0; | |
f2dba9c6 HX |
718 | return -EAGAIN; |
719 | } | |
720 | ||
5d240a89 N |
721 | if (iter->p && !rhlist) { |
722 | /* | |
723 | * We need to validate that 'p' is still in the table, and | |
724 | * if so, update 'skip' | |
725 | */ | |
726 | struct rhash_head *p; | |
727 | int skip = 0; | |
728 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
729 | skip++; | |
730 | if (p == iter->p) { | |
731 | iter->skip = skip; | |
732 | goto found; | |
733 | } | |
734 | } | |
735 | iter->p = NULL; | |
736 | } else if (iter->p && rhlist) { | |
737 | /* Need to validate that 'list' is still in the table, and | |
738 | * if so, update 'skip' and 'p'. | |
739 | */ | |
740 | struct rhash_head *p; | |
741 | struct rhlist_head *list; | |
742 | int skip = 0; | |
743 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
744 | for (list = container_of(p, struct rhlist_head, rhead); | |
745 | list; | |
746 | list = rcu_dereference(list->next)) { | |
747 | skip++; | |
748 | if (list == iter->list) { | |
749 | iter->p = p; | |
c643ecf3 | 750 | iter->skip = skip; |
5d240a89 N |
751 | goto found; |
752 | } | |
753 | } | |
754 | } | |
755 | iter->p = NULL; | |
756 | } | |
757 | found: | |
f2dba9c6 HX |
758 | return 0; |
759 | } | |
97a6ec4a | 760 | EXPORT_SYMBOL_GPL(rhashtable_walk_start_check); |
f2dba9c6 HX |
761 | |
762 | /** | |
2db54b47 TH |
763 | * __rhashtable_walk_find_next - Find the next element in a table (or the first |
764 | * one in case of a new walk). | |
765 | * | |
f2dba9c6 HX |
766 | * @iter: Hash table iterator |
767 | * | |
2db54b47 | 768 | * Returns the found object or NULL when the end of the table is reached. |
f2dba9c6 | 769 | * |
2db54b47 | 770 | * Returns -EAGAIN if resize event occurred. |
f2dba9c6 | 771 | */ |
2db54b47 | 772 | static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter) |
f2dba9c6 | 773 | { |
246779dd | 774 | struct bucket_table *tbl = iter->walker.tbl; |
ca26893f | 775 | struct rhlist_head *list = iter->list; |
f2dba9c6 HX |
776 | struct rhashtable *ht = iter->ht; |
777 | struct rhash_head *p = iter->p; | |
ca26893f | 778 | bool rhlist = ht->rhlist; |
f2dba9c6 | 779 | |
2db54b47 TH |
780 | if (!tbl) |
781 | return NULL; | |
f2dba9c6 HX |
782 | |
783 | for (; iter->slot < tbl->size; iter->slot++) { | |
784 | int skip = iter->skip; | |
785 | ||
786 | rht_for_each_rcu(p, tbl, iter->slot) { | |
ca26893f HX |
787 | if (rhlist) { |
788 | list = container_of(p, struct rhlist_head, | |
789 | rhead); | |
790 | do { | |
791 | if (!skip) | |
792 | goto next; | |
793 | skip--; | |
794 | list = rcu_dereference(list->next); | |
795 | } while (list); | |
796 | ||
797 | continue; | |
798 | } | |
f2dba9c6 HX |
799 | if (!skip) |
800 | break; | |
801 | skip--; | |
802 | } | |
803 | ||
804 | next: | |
805 | if (!rht_is_a_nulls(p)) { | |
806 | iter->skip++; | |
807 | iter->p = p; | |
ca26893f HX |
808 | iter->list = list; |
809 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
f2dba9c6 HX |
810 | } |
811 | ||
812 | iter->skip = 0; | |
813 | } | |
814 | ||
142b942a PS |
815 | iter->p = NULL; |
816 | ||
d88252f9 HX |
817 | /* Ensure we see any new tables. */ |
818 | smp_rmb(); | |
819 | ||
246779dd HX |
820 | iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
821 | if (iter->walker.tbl) { | |
f2dba9c6 HX |
822 | iter->slot = 0; |
823 | iter->skip = 0; | |
f2dba9c6 | 824 | return ERR_PTR(-EAGAIN); |
2db54b47 TH |
825 | } else { |
826 | iter->end_of_table = true; | |
f2dba9c6 HX |
827 | } |
828 | ||
c936a79f | 829 | return NULL; |
f2dba9c6 | 830 | } |
2db54b47 TH |
831 | |
832 | /** | |
833 | * rhashtable_walk_next - Return the next object and advance the iterator | |
834 | * @iter: Hash table iterator | |
835 | * | |
836 | * Note that you must call rhashtable_walk_stop when you are finished | |
837 | * with the walk. | |
838 | * | |
839 | * Returns the next object or NULL when the end of the table is reached. | |
840 | * | |
841 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
842 | * will rewind back to the beginning and you may continue to use it. | |
843 | */ | |
844 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
845 | { | |
846 | struct rhlist_head *list = iter->list; | |
847 | struct rhashtable *ht = iter->ht; | |
848 | struct rhash_head *p = iter->p; | |
849 | bool rhlist = ht->rhlist; | |
850 | ||
851 | if (p) { | |
852 | if (!rhlist || !(list = rcu_dereference(list->next))) { | |
853 | p = rcu_dereference(p->next); | |
854 | list = container_of(p, struct rhlist_head, rhead); | |
855 | } | |
856 | if (!rht_is_a_nulls(p)) { | |
857 | iter->skip++; | |
858 | iter->p = p; | |
859 | iter->list = list; | |
860 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
861 | } | |
862 | ||
863 | /* At the end of this slot, switch to next one and then find | |
864 | * next entry from that point. | |
865 | */ | |
866 | iter->skip = 0; | |
867 | iter->slot++; | |
868 | } | |
869 | ||
870 | return __rhashtable_walk_find_next(iter); | |
871 | } | |
f2dba9c6 HX |
872 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); |
873 | ||
2db54b47 TH |
874 | /** |
875 | * rhashtable_walk_peek - Return the next object but don't advance the iterator | |
876 | * @iter: Hash table iterator | |
877 | * | |
878 | * Returns the next object or NULL when the end of the table is reached. | |
879 | * | |
880 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
881 | * will rewind back to the beginning and you may continue to use it. | |
882 | */ | |
883 | void *rhashtable_walk_peek(struct rhashtable_iter *iter) | |
884 | { | |
885 | struct rhlist_head *list = iter->list; | |
886 | struct rhashtable *ht = iter->ht; | |
887 | struct rhash_head *p = iter->p; | |
888 | ||
889 | if (p) | |
890 | return rht_obj(ht, ht->rhlist ? &list->rhead : p); | |
891 | ||
892 | /* No object found in current iter, find next one in the table. */ | |
893 | ||
894 | if (iter->skip) { | |
895 | /* A nonzero skip value points to the next entry in the table | |
896 | * beyond that last one that was found. Decrement skip so | |
897 | * we find the current value. __rhashtable_walk_find_next | |
898 | * will restore the original value of skip assuming that | |
899 | * the table hasn't changed. | |
900 | */ | |
901 | iter->skip--; | |
902 | } | |
903 | ||
904 | return __rhashtable_walk_find_next(iter); | |
905 | } | |
906 | EXPORT_SYMBOL_GPL(rhashtable_walk_peek); | |
907 | ||
f2dba9c6 HX |
908 | /** |
909 | * rhashtable_walk_stop - Finish a hash table walk | |
910 | * @iter: Hash table iterator | |
911 | * | |
0647169c AG |
912 | * Finish a hash table walk. Does not reset the iterator to the start of the |
913 | * hash table. | |
f2dba9c6 HX |
914 | */ |
915 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 916 | __releases(RCU) |
f2dba9c6 | 917 | { |
eddee5ba | 918 | struct rhashtable *ht; |
246779dd | 919 | struct bucket_table *tbl = iter->walker.tbl; |
eddee5ba | 920 | |
eddee5ba | 921 | if (!tbl) |
963ecbd4 | 922 | goto out; |
eddee5ba HX |
923 | |
924 | ht = iter->ht; | |
925 | ||
ba7c95ea | 926 | spin_lock(&ht->lock); |
4feb7c7a N |
927 | if (rcu_head_after_call_rcu(&tbl->rcu, bucket_table_free_rcu)) |
928 | /* This bucket table is being freed, don't re-link it. */ | |
246779dd | 929 | iter->walker.tbl = NULL; |
4feb7c7a N |
930 | else |
931 | list_add(&iter->walker.list, &tbl->walkers); | |
ba7c95ea | 932 | spin_unlock(&ht->lock); |
eddee5ba | 933 | |
963ecbd4 HX |
934 | out: |
935 | rcu_read_unlock(); | |
f2dba9c6 HX |
936 | } |
937 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
938 | ||
488fb86e | 939 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 940 | { |
107d01f5 DB |
941 | size_t retsize; |
942 | ||
943 | if (params->nelem_hint) | |
944 | retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3), | |
945 | (unsigned long)params->min_size); | |
946 | else | |
947 | retsize = max(HASH_DEFAULT_SIZE, | |
948 | (unsigned long)params->min_size); | |
949 | ||
950 | return retsize; | |
7e1e7763 TG |
951 | } |
952 | ||
31ccde2d HX |
953 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
954 | { | |
955 | return jhash2(key, length, seed); | |
956 | } | |
957 | ||
7e1e7763 TG |
958 | /** |
959 | * rhashtable_init - initialize a new hash table | |
960 | * @ht: hash table to be initialized | |
961 | * @params: configuration parameters | |
962 | * | |
963 | * Initializes a new hash table based on the provided configuration | |
964 | * parameters. A table can be configured either with a variable or | |
965 | * fixed length key: | |
966 | * | |
967 | * Configuration Example 1: Fixed length keys | |
968 | * struct test_obj { | |
969 | * int key; | |
970 | * void * my_member; | |
971 | * struct rhash_head node; | |
972 | * }; | |
973 | * | |
974 | * struct rhashtable_params params = { | |
975 | * .head_offset = offsetof(struct test_obj, node), | |
976 | * .key_offset = offsetof(struct test_obj, key), | |
977 | * .key_len = sizeof(int), | |
87545899 | 978 | * .hashfn = jhash, |
7e1e7763 TG |
979 | * }; |
980 | * | |
981 | * Configuration Example 2: Variable length keys | |
982 | * struct test_obj { | |
983 | * [...] | |
984 | * struct rhash_head node; | |
985 | * }; | |
986 | * | |
49f7b33e | 987 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
988 | * { |
989 | * struct test_obj *obj = data; | |
990 | * | |
991 | * return [... hash ...]; | |
992 | * } | |
993 | * | |
994 | * struct rhashtable_params params = { | |
995 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 996 | * .hashfn = jhash, |
7e1e7763 | 997 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
998 | * }; |
999 | */ | |
488fb86e HX |
1000 | int rhashtable_init(struct rhashtable *ht, |
1001 | const struct rhashtable_params *params) | |
7e1e7763 TG |
1002 | { |
1003 | struct bucket_table *tbl; | |
1004 | size_t size; | |
1005 | ||
31ccde2d | 1006 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 1007 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
1008 | return -EINVAL; |
1009 | ||
97defe1e TG |
1010 | memset(ht, 0, sizeof(*ht)); |
1011 | mutex_init(&ht->mutex); | |
ba7c95ea | 1012 | spin_lock_init(&ht->lock); |
97defe1e TG |
1013 | memcpy(&ht->p, params, sizeof(*params)); |
1014 | ||
a998f712 TG |
1015 | if (params->min_size) |
1016 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
1017 | ||
6d684e54 HX |
1018 | /* Cap total entries at 2^31 to avoid nelems overflow. */ |
1019 | ht->max_elems = 1u << 31; | |
2d2ab658 HX |
1020 | |
1021 | if (params->max_size) { | |
1022 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
1023 | if (ht->p.max_size < ht->max_elems / 2) | |
1024 | ht->max_elems = ht->p.max_size * 2; | |
1025 | } | |
6d684e54 | 1026 | |
48e75b43 | 1027 | ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 1028 | |
107d01f5 | 1029 | size = rounded_hashtable_size(&ht->p); |
3a324606 | 1030 | |
97defe1e TG |
1031 | if (params->locks_mul) |
1032 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
1033 | else | |
1034 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
1035 | ||
31ccde2d HX |
1036 | ht->key_len = ht->p.key_len; |
1037 | if (!params->hashfn) { | |
1038 | ht->p.hashfn = jhash; | |
1039 | ||
1040 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
1041 | ht->key_len /= sizeof(u32); | |
1042 | ht->p.hashfn = rhashtable_jhash2; | |
1043 | } | |
1044 | } | |
1045 | ||
2d22ecf6 DB |
1046 | /* |
1047 | * This is api initialization and thus we need to guarantee the | |
1048 | * initial rhashtable allocation. Upon failure, retry with the | |
1049 | * smallest possible size with __GFP_NOFAIL semantics. | |
1050 | */ | |
b9ecfdaa | 1051 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
2d22ecf6 DB |
1052 | if (unlikely(tbl == NULL)) { |
1053 | size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); | |
1054 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL); | |
1055 | } | |
7e1e7763 | 1056 | |
545a148e | 1057 | atomic_set(&ht->nelems, 0); |
a5b6846f | 1058 | |
7e1e7763 TG |
1059 | RCU_INIT_POINTER(ht->tbl, tbl); |
1060 | ||
4c4b52d9 | 1061 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1062 | |
7e1e7763 TG |
1063 | return 0; |
1064 | } | |
1065 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1066 | ||
ca26893f HX |
1067 | /** |
1068 | * rhltable_init - initialize a new hash list table | |
1069 | * @hlt: hash list table to be initialized | |
1070 | * @params: configuration parameters | |
1071 | * | |
1072 | * Initializes a new hash list table. | |
1073 | * | |
1074 | * See documentation for rhashtable_init. | |
1075 | */ | |
1076 | int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params) | |
1077 | { | |
1078 | int err; | |
1079 | ||
ca26893f HX |
1080 | err = rhashtable_init(&hlt->ht, params); |
1081 | hlt->ht.rhlist = true; | |
1082 | return err; | |
1083 | } | |
1084 | EXPORT_SYMBOL_GPL(rhltable_init); | |
1085 | ||
1086 | static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj, | |
1087 | void (*free_fn)(void *ptr, void *arg), | |
1088 | void *arg) | |
1089 | { | |
1090 | struct rhlist_head *list; | |
1091 | ||
1092 | if (!ht->rhlist) { | |
1093 | free_fn(rht_obj(ht, obj), arg); | |
1094 | return; | |
1095 | } | |
1096 | ||
1097 | list = container_of(obj, struct rhlist_head, rhead); | |
1098 | do { | |
1099 | obj = &list->rhead; | |
1100 | list = rht_dereference(list->next, ht); | |
1101 | free_fn(rht_obj(ht, obj), arg); | |
1102 | } while (list); | |
1103 | } | |
1104 | ||
7e1e7763 | 1105 | /** |
6b6f302c | 1106 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 1107 | * @ht: the hash table to destroy |
6b6f302c TG |
1108 | * @free_fn: callback to release resources of element |
1109 | * @arg: pointer passed to free_fn | |
7e1e7763 | 1110 | * |
6b6f302c TG |
1111 | * Stops an eventual async resize. If defined, invokes free_fn for each |
1112 | * element to releasal resources. Please note that RCU protected | |
1113 | * readers may still be accessing the elements. Releasing of resources | |
1114 | * must occur in a compatible manner. Then frees the bucket array. | |
1115 | * | |
1116 | * This function will eventually sleep to wait for an async resize | |
1117 | * to complete. The caller is responsible that no further write operations | |
1118 | * occurs in parallel. | |
7e1e7763 | 1119 | */ |
6b6f302c TG |
1120 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
1121 | void (*free_fn)(void *ptr, void *arg), | |
1122 | void *arg) | |
7e1e7763 | 1123 | { |
0026129c | 1124 | struct bucket_table *tbl, *next_tbl; |
6b6f302c | 1125 | unsigned int i; |
97defe1e | 1126 | |
4c4b52d9 | 1127 | cancel_work_sync(&ht->run_work); |
97defe1e | 1128 | |
57699a40 | 1129 | mutex_lock(&ht->mutex); |
6b6f302c | 1130 | tbl = rht_dereference(ht->tbl, ht); |
0026129c | 1131 | restart: |
6b6f302c TG |
1132 | if (free_fn) { |
1133 | for (i = 0; i < tbl->size; i++) { | |
1134 | struct rhash_head *pos, *next; | |
1135 | ||
ae6da1f5 | 1136 | cond_resched(); |
da20420f | 1137 | for (pos = rht_dereference(*rht_bucket(tbl, i), ht), |
6b6f302c TG |
1138 | next = !rht_is_a_nulls(pos) ? |
1139 | rht_dereference(pos->next, ht) : NULL; | |
1140 | !rht_is_a_nulls(pos); | |
1141 | pos = next, | |
1142 | next = !rht_is_a_nulls(pos) ? | |
1143 | rht_dereference(pos->next, ht) : NULL) | |
ca26893f | 1144 | rhashtable_free_one(ht, pos, free_fn, arg); |
6b6f302c TG |
1145 | } |
1146 | } | |
1147 | ||
0026129c | 1148 | next_tbl = rht_dereference(tbl->future_tbl, ht); |
6b6f302c | 1149 | bucket_table_free(tbl); |
0026129c TY |
1150 | if (next_tbl) { |
1151 | tbl = next_tbl; | |
1152 | goto restart; | |
1153 | } | |
97defe1e | 1154 | mutex_unlock(&ht->mutex); |
7e1e7763 | 1155 | } |
6b6f302c TG |
1156 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
1157 | ||
1158 | void rhashtable_destroy(struct rhashtable *ht) | |
1159 | { | |
1160 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
1161 | } | |
7e1e7763 | 1162 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |
da20420f HX |
1163 | |
1164 | struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl, | |
1165 | unsigned int hash) | |
1166 | { | |
1167 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
82208d0d | 1168 | static struct rhash_head __rcu *rhnull; |
da20420f HX |
1169 | unsigned int index = hash & ((1 << tbl->nest) - 1); |
1170 | unsigned int size = tbl->size >> tbl->nest; | |
1171 | unsigned int subhash = hash; | |
1172 | union nested_table *ntbl; | |
1173 | ||
1174 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
c4d2603d | 1175 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash); |
da20420f HX |
1176 | subhash >>= tbl->nest; |
1177 | ||
1178 | while (ntbl && size > (1 << shift)) { | |
1179 | index = subhash & ((1 << shift) - 1); | |
c4d2603d HX |
1180 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, |
1181 | tbl, hash); | |
da20420f HX |
1182 | size >>= shift; |
1183 | subhash >>= shift; | |
1184 | } | |
1185 | ||
82208d0d N |
1186 | if (!ntbl) { |
1187 | if (!rhnull) | |
1188 | INIT_RHT_NULLS_HEAD(rhnull); | |
da20420f | 1189 | return &rhnull; |
82208d0d | 1190 | } |
da20420f HX |
1191 | |
1192 | return &ntbl[subhash].bucket; | |
1193 | ||
1194 | } | |
1195 | EXPORT_SYMBOL_GPL(rht_bucket_nested); | |
1196 | ||
1197 | struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht, | |
1198 | struct bucket_table *tbl, | |
1199 | unsigned int hash) | |
1200 | { | |
1201 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1202 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1203 | unsigned int size = tbl->size >> tbl->nest; | |
1204 | union nested_table *ntbl; | |
da20420f HX |
1205 | |
1206 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
1207 | hash >>= tbl->nest; | |
da20420f | 1208 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1209 | size <= (1 << shift)); |
da20420f HX |
1210 | |
1211 | while (ntbl && size > (1 << shift)) { | |
1212 | index = hash & ((1 << shift) - 1); | |
1213 | size >>= shift; | |
1214 | hash >>= shift; | |
da20420f | 1215 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1216 | size <= (1 << shift)); |
da20420f HX |
1217 | } |
1218 | ||
1219 | if (!ntbl) | |
1220 | return NULL; | |
1221 | ||
1222 | return &ntbl[hash].bucket; | |
1223 | ||
1224 | } | |
1225 | EXPORT_SYMBOL_GPL(rht_bucket_nested_insert); |