]>
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 | ||
68 | ||
b9ecfdaa HX |
69 | static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl, |
70 | gfp_t gfp) | |
97defe1e TG |
71 | { |
72 | unsigned int i, size; | |
73 | #if defined(CONFIG_PROVE_LOCKING) | |
74 | unsigned int nr_pcpus = 2; | |
75 | #else | |
76 | unsigned int nr_pcpus = num_possible_cpus(); | |
77 | #endif | |
78 | ||
4cf0b354 | 79 | nr_pcpus = min_t(unsigned int, nr_pcpus, 64UL); |
97defe1e TG |
80 | size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul); |
81 | ||
a5ec68e3 TG |
82 | /* Never allocate more than 0.5 locks per bucket */ |
83 | size = min_t(unsigned int, size, tbl->size >> 1); | |
97defe1e | 84 | |
da20420f HX |
85 | if (tbl->nest) |
86 | size = min(size, 1U << tbl->nest); | |
87 | ||
97defe1e | 88 | if (sizeof(spinlock_t) != 0) { |
43ca5bc4 MH |
89 | if (gfpflags_allow_blocking(gfp)) |
90 | tbl->locks = kvmalloc(size * sizeof(spinlock_t), gfp); | |
91 | else | |
9dbeea7f ED |
92 | tbl->locks = kmalloc_array(size, sizeof(spinlock_t), |
93 | gfp); | |
97defe1e TG |
94 | if (!tbl->locks) |
95 | return -ENOMEM; | |
96 | for (i = 0; i < size; i++) | |
97 | spin_lock_init(&tbl->locks[i]); | |
98 | } | |
99 | tbl->locks_mask = size - 1; | |
100 | ||
101 | return 0; | |
102 | } | |
103 | ||
da20420f HX |
104 | static void nested_table_free(union nested_table *ntbl, unsigned int size) |
105 | { | |
106 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
107 | const unsigned int len = 1 << shift; | |
108 | unsigned int i; | |
109 | ||
110 | ntbl = rcu_dereference_raw(ntbl->table); | |
111 | if (!ntbl) | |
112 | return; | |
113 | ||
114 | if (size > len) { | |
115 | size >>= shift; | |
116 | for (i = 0; i < len; i++) | |
117 | nested_table_free(ntbl + i, size); | |
118 | } | |
119 | ||
120 | kfree(ntbl); | |
121 | } | |
122 | ||
123 | static void nested_bucket_table_free(const struct bucket_table *tbl) | |
124 | { | |
125 | unsigned int size = tbl->size >> tbl->nest; | |
126 | unsigned int len = 1 << tbl->nest; | |
127 | union nested_table *ntbl; | |
128 | unsigned int i; | |
129 | ||
130 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
131 | ||
132 | for (i = 0; i < len; i++) | |
133 | nested_table_free(ntbl + i, size); | |
134 | ||
135 | kfree(ntbl); | |
136 | } | |
137 | ||
97defe1e TG |
138 | static void bucket_table_free(const struct bucket_table *tbl) |
139 | { | |
da20420f HX |
140 | if (tbl->nest) |
141 | nested_bucket_table_free(tbl); | |
142 | ||
ca435407 | 143 | kvfree(tbl->locks); |
97defe1e TG |
144 | kvfree(tbl); |
145 | } | |
146 | ||
9d901bc0 HX |
147 | static void bucket_table_free_rcu(struct rcu_head *head) |
148 | { | |
149 | bucket_table_free(container_of(head, struct bucket_table, rcu)); | |
150 | } | |
151 | ||
da20420f HX |
152 | static union nested_table *nested_table_alloc(struct rhashtable *ht, |
153 | union nested_table __rcu **prev, | |
154 | unsigned int shifted, | |
155 | unsigned int nhash) | |
156 | { | |
157 | union nested_table *ntbl; | |
158 | int i; | |
159 | ||
160 | ntbl = rcu_dereference(*prev); | |
161 | if (ntbl) | |
162 | return ntbl; | |
163 | ||
164 | ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC); | |
165 | ||
166 | if (ntbl && shifted) { | |
167 | for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0].bucket); i++) | |
168 | INIT_RHT_NULLS_HEAD(ntbl[i].bucket, ht, | |
169 | (i << shifted) | nhash); | |
170 | } | |
171 | ||
172 | rcu_assign_pointer(*prev, ntbl); | |
173 | ||
174 | return ntbl; | |
175 | } | |
176 | ||
177 | static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht, | |
178 | size_t nbuckets, | |
179 | gfp_t gfp) | |
180 | { | |
181 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
182 | struct bucket_table *tbl; | |
183 | size_t size; | |
184 | ||
185 | if (nbuckets < (1 << (shift + 1))) | |
186 | return NULL; | |
187 | ||
188 | size = sizeof(*tbl) + sizeof(tbl->buckets[0]); | |
189 | ||
190 | tbl = kzalloc(size, gfp); | |
191 | if (!tbl) | |
192 | return NULL; | |
193 | ||
194 | if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets, | |
195 | 0, 0)) { | |
196 | kfree(tbl); | |
197 | return NULL; | |
198 | } | |
199 | ||
200 | tbl->nest = (ilog2(nbuckets) - 1) % shift + 1; | |
201 | ||
202 | return tbl; | |
203 | } | |
204 | ||
97defe1e | 205 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
b9ecfdaa HX |
206 | size_t nbuckets, |
207 | gfp_t gfp) | |
7e1e7763 | 208 | { |
eb6d1abf | 209 | struct bucket_table *tbl = NULL; |
7e1e7763 | 210 | size_t size; |
f89bd6f8 | 211 | int i; |
7e1e7763 TG |
212 | |
213 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
12e8fd6f | 214 | if (gfp != GFP_KERNEL) |
b9ecfdaa | 215 | tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY); |
12e8fd6f MH |
216 | else |
217 | tbl = kvzalloc(size, gfp); | |
da20420f HX |
218 | |
219 | size = nbuckets; | |
220 | ||
221 | if (tbl == NULL && gfp != GFP_KERNEL) { | |
222 | tbl = nested_bucket_table_alloc(ht, nbuckets, gfp); | |
223 | nbuckets = 0; | |
224 | } | |
7e1e7763 TG |
225 | if (tbl == NULL) |
226 | return NULL; | |
227 | ||
da20420f | 228 | tbl->size = size; |
7e1e7763 | 229 | |
b9ecfdaa | 230 | if (alloc_bucket_locks(ht, tbl, gfp) < 0) { |
97defe1e TG |
231 | bucket_table_free(tbl); |
232 | return NULL; | |
233 | } | |
7e1e7763 | 234 | |
eddee5ba HX |
235 | INIT_LIST_HEAD(&tbl->walkers); |
236 | ||
d48ad080 | 237 | tbl->hash_rnd = get_random_u32(); |
5269b53d | 238 | |
f89bd6f8 TG |
239 | for (i = 0; i < nbuckets; i++) |
240 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); | |
241 | ||
97defe1e | 242 | return tbl; |
7e1e7763 TG |
243 | } |
244 | ||
b824478b HX |
245 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
246 | struct bucket_table *tbl) | |
247 | { | |
248 | struct bucket_table *new_tbl; | |
249 | ||
250 | do { | |
251 | new_tbl = tbl; | |
252 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
253 | } while (tbl); | |
254 | ||
255 | return new_tbl; | |
256 | } | |
257 | ||
299e5c32 | 258 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash) |
a5ec68e3 | 259 | { |
aa34a6cb | 260 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
b824478b HX |
261 | struct bucket_table *new_tbl = rhashtable_last_table(ht, |
262 | rht_dereference_rcu(old_tbl->future_tbl, ht)); | |
da20420f HX |
263 | struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash); |
264 | int err = -EAGAIN; | |
aa34a6cb HX |
265 | struct rhash_head *head, *next, *entry; |
266 | spinlock_t *new_bucket_lock; | |
299e5c32 | 267 | unsigned int new_hash; |
aa34a6cb | 268 | |
da20420f HX |
269 | if (new_tbl->nest) |
270 | goto out; | |
271 | ||
272 | err = -ENOENT; | |
273 | ||
aa34a6cb HX |
274 | rht_for_each(entry, old_tbl, old_hash) { |
275 | err = 0; | |
276 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
277 | ||
278 | if (rht_is_a_nulls(next)) | |
279 | break; | |
a5ec68e3 | 280 | |
aa34a6cb HX |
281 | pprev = &entry->next; |
282 | } | |
a5ec68e3 | 283 | |
aa34a6cb HX |
284 | if (err) |
285 | goto out; | |
97defe1e | 286 | |
aa34a6cb | 287 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 288 | |
02fd97c3 | 289 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 290 | |
8f2484bd | 291 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
292 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
293 | new_tbl, new_hash); | |
97defe1e | 294 | |
7def0f95 | 295 | RCU_INIT_POINTER(entry->next, head); |
a5ec68e3 | 296 | |
aa34a6cb HX |
297 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
298 | spin_unlock(new_bucket_lock); | |
97defe1e | 299 | |
aa34a6cb | 300 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 301 | |
aa34a6cb HX |
302 | out: |
303 | return err; | |
304 | } | |
97defe1e | 305 | |
da20420f | 306 | static int rhashtable_rehash_chain(struct rhashtable *ht, |
299e5c32 | 307 | unsigned int old_hash) |
aa34a6cb HX |
308 | { |
309 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
310 | spinlock_t *old_bucket_lock; | |
da20420f | 311 | int err; |
aa34a6cb | 312 | |
02fd97c3 | 313 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 314 | |
aa34a6cb | 315 | spin_lock_bh(old_bucket_lock); |
da20420f | 316 | while (!(err = rhashtable_rehash_one(ht, old_hash))) |
aa34a6cb | 317 | ; |
da20420f HX |
318 | |
319 | if (err == -ENOENT) { | |
320 | old_tbl->rehash++; | |
321 | err = 0; | |
322 | } | |
aa34a6cb | 323 | spin_unlock_bh(old_bucket_lock); |
da20420f HX |
324 | |
325 | return err; | |
97defe1e TG |
326 | } |
327 | ||
b824478b HX |
328 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
329 | struct bucket_table *old_tbl, | |
330 | struct bucket_table *new_tbl) | |
97defe1e | 331 | { |
b824478b HX |
332 | /* Protect future_tbl using the first bucket lock. */ |
333 | spin_lock_bh(old_tbl->locks); | |
334 | ||
335 | /* Did somebody beat us to it? */ | |
336 | if (rcu_access_pointer(old_tbl->future_tbl)) { | |
337 | spin_unlock_bh(old_tbl->locks); | |
338 | return -EEXIST; | |
339 | } | |
7cd10db8 | 340 | |
aa34a6cb HX |
341 | /* Make insertions go into the new, empty table right away. Deletions |
342 | * and lookups will be attempted in both tables until we synchronize. | |
aa34a6cb | 343 | */ |
c4db8848 | 344 | rcu_assign_pointer(old_tbl->future_tbl, new_tbl); |
aa34a6cb | 345 | |
b824478b HX |
346 | spin_unlock_bh(old_tbl->locks); |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
351 | static int rhashtable_rehash_table(struct rhashtable *ht) | |
352 | { | |
353 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
354 | struct bucket_table *new_tbl; | |
355 | struct rhashtable_walker *walker; | |
299e5c32 | 356 | unsigned int old_hash; |
da20420f | 357 | int err; |
b824478b HX |
358 | |
359 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); | |
360 | if (!new_tbl) | |
361 | return 0; | |
362 | ||
da20420f HX |
363 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
364 | err = rhashtable_rehash_chain(ht, old_hash); | |
365 | if (err) | |
366 | return err; | |
2ccf21bc | 367 | cond_resched(); |
da20420f | 368 | } |
aa34a6cb HX |
369 | |
370 | /* Publish the new table pointer. */ | |
371 | rcu_assign_pointer(ht->tbl, new_tbl); | |
372 | ||
ba7c95ea | 373 | spin_lock(&ht->lock); |
eddee5ba HX |
374 | list_for_each_entry(walker, &old_tbl->walkers, list) |
375 | walker->tbl = NULL; | |
ba7c95ea | 376 | spin_unlock(&ht->lock); |
eddee5ba | 377 | |
aa34a6cb HX |
378 | /* Wait for readers. All new readers will see the new |
379 | * table, and thus no references to the old table will | |
380 | * remain. | |
381 | */ | |
9d901bc0 | 382 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
b824478b HX |
383 | |
384 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; | |
7e1e7763 TG |
385 | } |
386 | ||
da20420f HX |
387 | static int rhashtable_rehash_alloc(struct rhashtable *ht, |
388 | struct bucket_table *old_tbl, | |
389 | unsigned int size) | |
7e1e7763 | 390 | { |
da20420f | 391 | struct bucket_table *new_tbl; |
b824478b | 392 | int err; |
7e1e7763 TG |
393 | |
394 | ASSERT_RHT_MUTEX(ht); | |
395 | ||
da20420f | 396 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
397 | if (new_tbl == NULL) |
398 | return -ENOMEM; | |
399 | ||
b824478b HX |
400 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
401 | if (err) | |
402 | bucket_table_free(new_tbl); | |
403 | ||
404 | return err; | |
7e1e7763 | 405 | } |
7e1e7763 TG |
406 | |
407 | /** | |
408 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
409 | * @ht: the hash table to shrink | |
7e1e7763 | 410 | * |
18093d1c HX |
411 | * This function shrinks the hash table to fit, i.e., the smallest |
412 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 413 | * |
97defe1e TG |
414 | * The caller must ensure that no concurrent resizing occurs by holding |
415 | * ht->mutex. | |
416 | * | |
7e1e7763 TG |
417 | * The caller must ensure that no concurrent table mutations take place. |
418 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
419 | * |
420 | * It is valid to have concurrent insertions and deletions protected by per | |
421 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 422 | */ |
b824478b | 423 | static int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 424 | { |
da20420f | 425 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
12311959 VN |
426 | unsigned int nelems = atomic_read(&ht->nelems); |
427 | unsigned int size = 0; | |
7e1e7763 | 428 | |
12311959 VN |
429 | if (nelems) |
430 | size = roundup_pow_of_two(nelems * 3 / 2); | |
18093d1c HX |
431 | if (size < ht->p.min_size) |
432 | size = ht->p.min_size; | |
433 | ||
434 | if (old_tbl->size <= size) | |
435 | return 0; | |
436 | ||
b824478b HX |
437 | if (rht_dereference(old_tbl->future_tbl, ht)) |
438 | return -EEXIST; | |
439 | ||
da20420f | 440 | return rhashtable_rehash_alloc(ht, old_tbl, size); |
7e1e7763 | 441 | } |
7e1e7763 | 442 | |
97defe1e TG |
443 | static void rht_deferred_worker(struct work_struct *work) |
444 | { | |
445 | struct rhashtable *ht; | |
446 | struct bucket_table *tbl; | |
b824478b | 447 | int err = 0; |
97defe1e | 448 | |
57699a40 | 449 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 450 | mutex_lock(&ht->mutex); |
28134a53 | 451 | |
97defe1e | 452 | tbl = rht_dereference(ht->tbl, ht); |
b824478b | 453 | tbl = rhashtable_last_table(ht, tbl); |
97defe1e | 454 | |
a5b6846f | 455 | if (rht_grow_above_75(ht, tbl)) |
da20420f | 456 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2); |
b5e2c150 | 457 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
da20420f HX |
458 | err = rhashtable_shrink(ht); |
459 | else if (tbl->nest) | |
460 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size); | |
b824478b | 461 | |
f322610c HX |
462 | if (!err || err == -EEXIST) { |
463 | int nerr; | |
464 | ||
465 | nerr = rhashtable_rehash_table(ht); | |
466 | err = err ?: nerr; | |
467 | } | |
b824478b | 468 | |
97defe1e | 469 | mutex_unlock(&ht->mutex); |
b824478b HX |
470 | |
471 | if (err) | |
472 | schedule_work(&ht->run_work); | |
97defe1e TG |
473 | } |
474 | ||
ca26893f HX |
475 | static int rhashtable_insert_rehash(struct rhashtable *ht, |
476 | struct bucket_table *tbl) | |
ccd57b1b HX |
477 | { |
478 | struct bucket_table *old_tbl; | |
479 | struct bucket_table *new_tbl; | |
ccd57b1b HX |
480 | unsigned int size; |
481 | int err; | |
482 | ||
483 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
ccd57b1b HX |
484 | |
485 | size = tbl->size; | |
486 | ||
3cf92222 HX |
487 | err = -EBUSY; |
488 | ||
ccd57b1b HX |
489 | if (rht_grow_above_75(ht, tbl)) |
490 | size *= 2; | |
a87b9ebf TG |
491 | /* Do not schedule more than one rehash */ |
492 | else if (old_tbl != tbl) | |
3cf92222 HX |
493 | goto fail; |
494 | ||
495 | err = -ENOMEM; | |
ccd57b1b HX |
496 | |
497 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC); | |
3cf92222 HX |
498 | if (new_tbl == NULL) |
499 | goto fail; | |
ccd57b1b HX |
500 | |
501 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); | |
502 | if (err) { | |
503 | bucket_table_free(new_tbl); | |
504 | if (err == -EEXIST) | |
505 | err = 0; | |
506 | } else | |
507 | schedule_work(&ht->run_work); | |
508 | ||
509 | return err; | |
3cf92222 HX |
510 | |
511 | fail: | |
512 | /* Do not fail the insert if someone else did a rehash. */ | |
513 | if (likely(rcu_dereference_raw(tbl->future_tbl))) | |
514 | return 0; | |
515 | ||
516 | /* Schedule async rehash to retry allocation in process context. */ | |
517 | if (err == -ENOMEM) | |
518 | schedule_work(&ht->run_work); | |
519 | ||
520 | return err; | |
ccd57b1b | 521 | } |
ccd57b1b | 522 | |
ca26893f HX |
523 | static void *rhashtable_lookup_one(struct rhashtable *ht, |
524 | struct bucket_table *tbl, unsigned int hash, | |
525 | const void *key, struct rhash_head *obj) | |
02fd97c3 | 526 | { |
ca26893f HX |
527 | struct rhashtable_compare_arg arg = { |
528 | .ht = ht, | |
529 | .key = key, | |
530 | }; | |
531 | struct rhash_head __rcu **pprev; | |
02fd97c3 | 532 | struct rhash_head *head; |
ca26893f | 533 | int elasticity; |
02fd97c3 | 534 | |
5f8ddeab | 535 | elasticity = RHT_ELASTICITY; |
da20420f HX |
536 | pprev = rht_bucket_var(tbl, hash); |
537 | rht_for_each_continue(head, *pprev, tbl, hash) { | |
ca26893f HX |
538 | struct rhlist_head *list; |
539 | struct rhlist_head *plist; | |
540 | ||
541 | elasticity--; | |
542 | if (!key || | |
543 | (ht->p.obj_cmpfn ? | |
544 | ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
7ca08768 PB |
545 | rhashtable_compare(&arg, rht_obj(ht, head)))) { |
546 | pprev = &head->next; | |
ca26893f | 547 | continue; |
7ca08768 | 548 | } |
ca26893f HX |
549 | |
550 | if (!ht->rhlist) | |
551 | return rht_obj(ht, head); | |
552 | ||
553 | list = container_of(obj, struct rhlist_head, rhead); | |
554 | plist = container_of(head, struct rhlist_head, rhead); | |
555 | ||
556 | RCU_INIT_POINTER(list->next, plist); | |
557 | head = rht_dereference_bucket(head->next, tbl, hash); | |
558 | RCU_INIT_POINTER(list->rhead.next, head); | |
559 | rcu_assign_pointer(*pprev, obj); | |
560 | ||
561 | return NULL; | |
5ca8cc5b | 562 | } |
02fd97c3 | 563 | |
ca26893f HX |
564 | if (elasticity <= 0) |
565 | return ERR_PTR(-EAGAIN); | |
566 | ||
567 | return ERR_PTR(-ENOENT); | |
568 | } | |
569 | ||
570 | static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht, | |
571 | struct bucket_table *tbl, | |
572 | unsigned int hash, | |
573 | struct rhash_head *obj, | |
574 | void *data) | |
575 | { | |
da20420f | 576 | struct rhash_head __rcu **pprev; |
ca26893f HX |
577 | struct bucket_table *new_tbl; |
578 | struct rhash_head *head; | |
579 | ||
580 | if (!IS_ERR_OR_NULL(data)) | |
581 | return ERR_PTR(-EEXIST); | |
07ee0722 | 582 | |
ca26893f HX |
583 | if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT) |
584 | return ERR_CAST(data); | |
ccd57b1b | 585 | |
ca26893f HX |
586 | new_tbl = rcu_dereference(tbl->future_tbl); |
587 | if (new_tbl) | |
588 | return new_tbl; | |
589 | ||
590 | if (PTR_ERR(data) != -ENOENT) | |
591 | return ERR_CAST(data); | |
592 | ||
593 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
594 | return ERR_PTR(-E2BIG); | |
595 | ||
596 | if (unlikely(rht_grow_above_100(ht, tbl))) | |
597 | return ERR_PTR(-EAGAIN); | |
02fd97c3 | 598 | |
da20420f HX |
599 | pprev = rht_bucket_insert(ht, tbl, hash); |
600 | if (!pprev) | |
601 | return ERR_PTR(-ENOMEM); | |
602 | ||
603 | head = rht_dereference_bucket(*pprev, tbl, hash); | |
02fd97c3 HX |
604 | |
605 | RCU_INIT_POINTER(obj->next, head); | |
ca26893f HX |
606 | if (ht->rhlist) { |
607 | struct rhlist_head *list; | |
608 | ||
609 | list = container_of(obj, struct rhlist_head, rhead); | |
610 | RCU_INIT_POINTER(list->next, NULL); | |
611 | } | |
02fd97c3 | 612 | |
da20420f | 613 | rcu_assign_pointer(*pprev, obj); |
02fd97c3 HX |
614 | |
615 | atomic_inc(&ht->nelems); | |
ca26893f HX |
616 | if (rht_grow_above_75(ht, tbl)) |
617 | schedule_work(&ht->run_work); | |
02fd97c3 | 618 | |
ca26893f HX |
619 | return NULL; |
620 | } | |
02fd97c3 | 621 | |
ca26893f HX |
622 | static void *rhashtable_try_insert(struct rhashtable *ht, const void *key, |
623 | struct rhash_head *obj) | |
624 | { | |
625 | struct bucket_table *new_tbl; | |
626 | struct bucket_table *tbl; | |
627 | unsigned int hash; | |
628 | spinlock_t *lock; | |
629 | void *data; | |
630 | ||
631 | tbl = rcu_dereference(ht->tbl); | |
632 | ||
633 | /* All insertions must grab the oldest table containing | |
634 | * the hashed bucket that is yet to be rehashed. | |
635 | */ | |
636 | for (;;) { | |
637 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
638 | lock = rht_bucket_lock(tbl, hash); | |
639 | spin_lock_bh(lock); | |
640 | ||
641 | if (tbl->rehash <= hash) | |
642 | break; | |
643 | ||
644 | spin_unlock_bh(lock); | |
645 | tbl = rcu_dereference(tbl->future_tbl); | |
646 | } | |
647 | ||
648 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
649 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
650 | if (PTR_ERR(new_tbl) != -EEXIST) | |
651 | data = ERR_CAST(new_tbl); | |
652 | ||
653 | while (!IS_ERR_OR_NULL(new_tbl)) { | |
654 | tbl = new_tbl; | |
655 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
656 | spin_lock_nested(rht_bucket_lock(tbl, hash), | |
657 | SINGLE_DEPTH_NESTING); | |
658 | ||
659 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
660 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
661 | if (PTR_ERR(new_tbl) != -EEXIST) | |
662 | data = ERR_CAST(new_tbl); | |
663 | ||
664 | spin_unlock(rht_bucket_lock(tbl, hash)); | |
665 | } | |
666 | ||
667 | spin_unlock_bh(lock); | |
668 | ||
669 | if (PTR_ERR(data) == -EAGAIN) | |
670 | data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?: | |
671 | -EAGAIN); | |
672 | ||
673 | return data; | |
674 | } | |
675 | ||
676 | void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
677 | struct rhash_head *obj) | |
678 | { | |
679 | void *data; | |
680 | ||
681 | do { | |
682 | rcu_read_lock(); | |
683 | data = rhashtable_try_insert(ht, key, obj); | |
684 | rcu_read_unlock(); | |
685 | } while (PTR_ERR(data) == -EAGAIN); | |
686 | ||
687 | return data; | |
02fd97c3 HX |
688 | } |
689 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
690 | ||
f2dba9c6 | 691 | /** |
246779dd | 692 | * rhashtable_walk_enter - Initialise an iterator |
f2dba9c6 HX |
693 | * @ht: Table to walk over |
694 | * @iter: Hash table Iterator | |
695 | * | |
696 | * This function prepares a hash table walk. | |
697 | * | |
698 | * Note that if you restart a walk after rhashtable_walk_stop you | |
699 | * may see the same object twice. Also, you may miss objects if | |
700 | * there are removals in between rhashtable_walk_stop and the next | |
701 | * call to rhashtable_walk_start. | |
702 | * | |
703 | * For a completely stable walk you should construct your own data | |
704 | * structure outside the hash table. | |
705 | * | |
706 | * This function may sleep so you must not call it from interrupt | |
707 | * context or with spin locks held. | |
708 | * | |
246779dd | 709 | * You must call rhashtable_walk_exit after this function returns. |
f2dba9c6 | 710 | */ |
246779dd | 711 | void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter) |
f2dba9c6 HX |
712 | { |
713 | iter->ht = ht; | |
714 | iter->p = NULL; | |
715 | iter->slot = 0; | |
716 | iter->skip = 0; | |
717 | ||
c6ff5268 | 718 | spin_lock(&ht->lock); |
246779dd | 719 | iter->walker.tbl = |
179ccc0a | 720 | rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock)); |
246779dd | 721 | list_add(&iter->walker.list, &iter->walker.tbl->walkers); |
c6ff5268 | 722 | spin_unlock(&ht->lock); |
f2dba9c6 | 723 | } |
246779dd | 724 | EXPORT_SYMBOL_GPL(rhashtable_walk_enter); |
f2dba9c6 HX |
725 | |
726 | /** | |
727 | * rhashtable_walk_exit - Free an iterator | |
728 | * @iter: Hash table Iterator | |
729 | * | |
730 | * This function frees resources allocated by rhashtable_walk_init. | |
731 | */ | |
732 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
733 | { | |
c6ff5268 | 734 | spin_lock(&iter->ht->lock); |
246779dd HX |
735 | if (iter->walker.tbl) |
736 | list_del(&iter->walker.list); | |
c6ff5268 | 737 | spin_unlock(&iter->ht->lock); |
f2dba9c6 HX |
738 | } |
739 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
740 | ||
741 | /** | |
742 | * rhashtable_walk_start - Start a hash table walk | |
743 | * @iter: Hash table iterator | |
744 | * | |
0647169c AG |
745 | * Start a hash table walk at the current iterator position. Note that we take |
746 | * the RCU lock in all cases including when we return an error. So you must | |
747 | * always call rhashtable_walk_stop to clean up. | |
f2dba9c6 HX |
748 | * |
749 | * Returns zero if successful. | |
750 | * | |
751 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
752 | * will rewind back to the beginning and you may use it immediately | |
753 | * by calling rhashtable_walk_next. | |
754 | */ | |
755 | int rhashtable_walk_start(struct rhashtable_iter *iter) | |
db4374f4 | 756 | __acquires(RCU) |
f2dba9c6 | 757 | { |
eddee5ba HX |
758 | struct rhashtable *ht = iter->ht; |
759 | ||
c6ff5268 | 760 | rcu_read_lock(); |
eddee5ba | 761 | |
c6ff5268 | 762 | spin_lock(&ht->lock); |
246779dd HX |
763 | if (iter->walker.tbl) |
764 | list_del(&iter->walker.list); | |
c6ff5268 | 765 | spin_unlock(&ht->lock); |
eddee5ba | 766 | |
246779dd HX |
767 | if (!iter->walker.tbl) { |
768 | iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht); | |
f2dba9c6 HX |
769 | return -EAGAIN; |
770 | } | |
771 | ||
772 | return 0; | |
773 | } | |
774 | EXPORT_SYMBOL_GPL(rhashtable_walk_start); | |
775 | ||
776 | /** | |
777 | * rhashtable_walk_next - Return the next object and advance the iterator | |
778 | * @iter: Hash table iterator | |
779 | * | |
780 | * Note that you must call rhashtable_walk_stop when you are finished | |
781 | * with the walk. | |
782 | * | |
783 | * Returns the next object or NULL when the end of the table is reached. | |
784 | * | |
785 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
786 | * will rewind back to the beginning and you may continue to use it. | |
787 | */ | |
788 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
789 | { | |
246779dd | 790 | struct bucket_table *tbl = iter->walker.tbl; |
ca26893f | 791 | struct rhlist_head *list = iter->list; |
f2dba9c6 HX |
792 | struct rhashtable *ht = iter->ht; |
793 | struct rhash_head *p = iter->p; | |
ca26893f | 794 | bool rhlist = ht->rhlist; |
f2dba9c6 | 795 | |
f2dba9c6 | 796 | if (p) { |
ca26893f HX |
797 | if (!rhlist || !(list = rcu_dereference(list->next))) { |
798 | p = rcu_dereference(p->next); | |
799 | list = container_of(p, struct rhlist_head, rhead); | |
800 | } | |
f2dba9c6 HX |
801 | goto next; |
802 | } | |
803 | ||
804 | for (; iter->slot < tbl->size; iter->slot++) { | |
805 | int skip = iter->skip; | |
806 | ||
807 | rht_for_each_rcu(p, tbl, iter->slot) { | |
ca26893f HX |
808 | if (rhlist) { |
809 | list = container_of(p, struct rhlist_head, | |
810 | rhead); | |
811 | do { | |
812 | if (!skip) | |
813 | goto next; | |
814 | skip--; | |
815 | list = rcu_dereference(list->next); | |
816 | } while (list); | |
817 | ||
818 | continue; | |
819 | } | |
f2dba9c6 HX |
820 | if (!skip) |
821 | break; | |
822 | skip--; | |
823 | } | |
824 | ||
825 | next: | |
826 | if (!rht_is_a_nulls(p)) { | |
827 | iter->skip++; | |
828 | iter->p = p; | |
ca26893f HX |
829 | iter->list = list; |
830 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
f2dba9c6 HX |
831 | } |
832 | ||
833 | iter->skip = 0; | |
834 | } | |
835 | ||
142b942a PS |
836 | iter->p = NULL; |
837 | ||
d88252f9 HX |
838 | /* Ensure we see any new tables. */ |
839 | smp_rmb(); | |
840 | ||
246779dd HX |
841 | iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
842 | if (iter->walker.tbl) { | |
f2dba9c6 HX |
843 | iter->slot = 0; |
844 | iter->skip = 0; | |
f2dba9c6 HX |
845 | return ERR_PTR(-EAGAIN); |
846 | } | |
847 | ||
c936a79f | 848 | return NULL; |
f2dba9c6 HX |
849 | } |
850 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); | |
851 | ||
852 | /** | |
853 | * rhashtable_walk_stop - Finish a hash table walk | |
854 | * @iter: Hash table iterator | |
855 | * | |
0647169c AG |
856 | * Finish a hash table walk. Does not reset the iterator to the start of the |
857 | * hash table. | |
f2dba9c6 HX |
858 | */ |
859 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 860 | __releases(RCU) |
f2dba9c6 | 861 | { |
eddee5ba | 862 | struct rhashtable *ht; |
246779dd | 863 | struct bucket_table *tbl = iter->walker.tbl; |
eddee5ba | 864 | |
eddee5ba | 865 | if (!tbl) |
963ecbd4 | 866 | goto out; |
eddee5ba HX |
867 | |
868 | ht = iter->ht; | |
869 | ||
ba7c95ea | 870 | spin_lock(&ht->lock); |
c4db8848 | 871 | if (tbl->rehash < tbl->size) |
246779dd | 872 | list_add(&iter->walker.list, &tbl->walkers); |
eddee5ba | 873 | else |
246779dd | 874 | iter->walker.tbl = NULL; |
ba7c95ea | 875 | spin_unlock(&ht->lock); |
eddee5ba | 876 | |
f2dba9c6 | 877 | iter->p = NULL; |
963ecbd4 HX |
878 | |
879 | out: | |
880 | rcu_read_unlock(); | |
f2dba9c6 HX |
881 | } |
882 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
883 | ||
488fb86e | 884 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 885 | { |
cc6fce1f DB |
886 | size_t retsize; |
887 | ||
888 | if (params->nelem_hint) | |
889 | retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3), | |
890 | (unsigned long)params->min_size); | |
891 | else | |
892 | retsize = max(HASH_DEFAULT_SIZE, | |
893 | (unsigned long)params->min_size); | |
894 | ||
895 | return retsize; | |
7e1e7763 TG |
896 | } |
897 | ||
31ccde2d HX |
898 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
899 | { | |
900 | return jhash2(key, length, seed); | |
901 | } | |
902 | ||
7e1e7763 TG |
903 | /** |
904 | * rhashtable_init - initialize a new hash table | |
905 | * @ht: hash table to be initialized | |
906 | * @params: configuration parameters | |
907 | * | |
908 | * Initializes a new hash table based on the provided configuration | |
909 | * parameters. A table can be configured either with a variable or | |
910 | * fixed length key: | |
911 | * | |
912 | * Configuration Example 1: Fixed length keys | |
913 | * struct test_obj { | |
914 | * int key; | |
915 | * void * my_member; | |
916 | * struct rhash_head node; | |
917 | * }; | |
918 | * | |
919 | * struct rhashtable_params params = { | |
920 | * .head_offset = offsetof(struct test_obj, node), | |
921 | * .key_offset = offsetof(struct test_obj, key), | |
922 | * .key_len = sizeof(int), | |
87545899 | 923 | * .hashfn = jhash, |
f89bd6f8 | 924 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
7e1e7763 TG |
925 | * }; |
926 | * | |
927 | * Configuration Example 2: Variable length keys | |
928 | * struct test_obj { | |
929 | * [...] | |
930 | * struct rhash_head node; | |
931 | * }; | |
932 | * | |
49f7b33e | 933 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
934 | * { |
935 | * struct test_obj *obj = data; | |
936 | * | |
937 | * return [... hash ...]; | |
938 | * } | |
939 | * | |
940 | * struct rhashtable_params params = { | |
941 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 942 | * .hashfn = jhash, |
7e1e7763 | 943 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
944 | * }; |
945 | */ | |
488fb86e HX |
946 | int rhashtable_init(struct rhashtable *ht, |
947 | const struct rhashtable_params *params) | |
7e1e7763 TG |
948 | { |
949 | struct bucket_table *tbl; | |
950 | size_t size; | |
951 | ||
31ccde2d | 952 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 953 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
954 | return -EINVAL; |
955 | ||
f89bd6f8 TG |
956 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
957 | return -EINVAL; | |
958 | ||
97defe1e TG |
959 | memset(ht, 0, sizeof(*ht)); |
960 | mutex_init(&ht->mutex); | |
ba7c95ea | 961 | spin_lock_init(&ht->lock); |
97defe1e TG |
962 | memcpy(&ht->p, params, sizeof(*params)); |
963 | ||
a998f712 TG |
964 | if (params->min_size) |
965 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
966 | ||
6d684e54 HX |
967 | /* Cap total entries at 2^31 to avoid nelems overflow. */ |
968 | ht->max_elems = 1u << 31; | |
2d2ab658 HX |
969 | |
970 | if (params->max_size) { | |
971 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
972 | if (ht->p.max_size < ht->max_elems / 2) | |
973 | ht->max_elems = ht->p.max_size * 2; | |
974 | } | |
6d684e54 | 975 | |
48e75b43 | 976 | ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 977 | |
cc6fce1f | 978 | size = rounded_hashtable_size(&ht->p); |
3a324606 | 979 | |
97defe1e TG |
980 | if (params->locks_mul) |
981 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
982 | else | |
983 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
984 | ||
31ccde2d HX |
985 | ht->key_len = ht->p.key_len; |
986 | if (!params->hashfn) { | |
987 | ht->p.hashfn = jhash; | |
988 | ||
989 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
990 | ht->key_len /= sizeof(u32); | |
991 | ht->p.hashfn = rhashtable_jhash2; | |
992 | } | |
993 | } | |
994 | ||
b9ecfdaa | 995 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
996 | if (tbl == NULL) |
997 | return -ENOMEM; | |
998 | ||
545a148e | 999 | atomic_set(&ht->nelems, 0); |
a5b6846f | 1000 | |
7e1e7763 TG |
1001 | RCU_INIT_POINTER(ht->tbl, tbl); |
1002 | ||
4c4b52d9 | 1003 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1004 | |
7e1e7763 TG |
1005 | return 0; |
1006 | } | |
1007 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1008 | ||
ca26893f HX |
1009 | /** |
1010 | * rhltable_init - initialize a new hash list table | |
1011 | * @hlt: hash list table to be initialized | |
1012 | * @params: configuration parameters | |
1013 | * | |
1014 | * Initializes a new hash list table. | |
1015 | * | |
1016 | * See documentation for rhashtable_init. | |
1017 | */ | |
1018 | int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params) | |
1019 | { | |
1020 | int err; | |
1021 | ||
1022 | /* No rhlist NULLs marking for now. */ | |
1023 | if (params->nulls_base) | |
1024 | return -EINVAL; | |
1025 | ||
1026 | err = rhashtable_init(&hlt->ht, params); | |
1027 | hlt->ht.rhlist = true; | |
1028 | return err; | |
1029 | } | |
1030 | EXPORT_SYMBOL_GPL(rhltable_init); | |
1031 | ||
1032 | static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj, | |
1033 | void (*free_fn)(void *ptr, void *arg), | |
1034 | void *arg) | |
1035 | { | |
1036 | struct rhlist_head *list; | |
1037 | ||
1038 | if (!ht->rhlist) { | |
1039 | free_fn(rht_obj(ht, obj), arg); | |
1040 | return; | |
1041 | } | |
1042 | ||
1043 | list = container_of(obj, struct rhlist_head, rhead); | |
1044 | do { | |
1045 | obj = &list->rhead; | |
1046 | list = rht_dereference(list->next, ht); | |
1047 | free_fn(rht_obj(ht, obj), arg); | |
1048 | } while (list); | |
1049 | } | |
1050 | ||
7e1e7763 | 1051 | /** |
6b6f302c | 1052 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 1053 | * @ht: the hash table to destroy |
6b6f302c TG |
1054 | * @free_fn: callback to release resources of element |
1055 | * @arg: pointer passed to free_fn | |
7e1e7763 | 1056 | * |
6b6f302c TG |
1057 | * Stops an eventual async resize. If defined, invokes free_fn for each |
1058 | * element to releasal resources. Please note that RCU protected | |
1059 | * readers may still be accessing the elements. Releasing of resources | |
1060 | * must occur in a compatible manner. Then frees the bucket array. | |
1061 | * | |
1062 | * This function will eventually sleep to wait for an async resize | |
1063 | * to complete. The caller is responsible that no further write operations | |
1064 | * occurs in parallel. | |
7e1e7763 | 1065 | */ |
6b6f302c TG |
1066 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
1067 | void (*free_fn)(void *ptr, void *arg), | |
1068 | void *arg) | |
7e1e7763 | 1069 | { |
7a295b55 | 1070 | struct bucket_table *tbl, *next_tbl; |
6b6f302c | 1071 | unsigned int i; |
97defe1e | 1072 | |
4c4b52d9 | 1073 | cancel_work_sync(&ht->run_work); |
97defe1e | 1074 | |
57699a40 | 1075 | mutex_lock(&ht->mutex); |
6b6f302c | 1076 | tbl = rht_dereference(ht->tbl, ht); |
7a295b55 | 1077 | restart: |
6b6f302c TG |
1078 | if (free_fn) { |
1079 | for (i = 0; i < tbl->size; i++) { | |
1080 | struct rhash_head *pos, *next; | |
1081 | ||
2ccf21bc | 1082 | cond_resched(); |
da20420f | 1083 | for (pos = rht_dereference(*rht_bucket(tbl, i), ht), |
6b6f302c TG |
1084 | next = !rht_is_a_nulls(pos) ? |
1085 | rht_dereference(pos->next, ht) : NULL; | |
1086 | !rht_is_a_nulls(pos); | |
1087 | pos = next, | |
1088 | next = !rht_is_a_nulls(pos) ? | |
1089 | rht_dereference(pos->next, ht) : NULL) | |
ca26893f | 1090 | rhashtable_free_one(ht, pos, free_fn, arg); |
6b6f302c TG |
1091 | } |
1092 | } | |
1093 | ||
7a295b55 | 1094 | next_tbl = rht_dereference(tbl->future_tbl, ht); |
6b6f302c | 1095 | bucket_table_free(tbl); |
7a295b55 TY |
1096 | if (next_tbl) { |
1097 | tbl = next_tbl; | |
1098 | goto restart; | |
1099 | } | |
97defe1e | 1100 | mutex_unlock(&ht->mutex); |
7e1e7763 | 1101 | } |
6b6f302c TG |
1102 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
1103 | ||
1104 | void rhashtable_destroy(struct rhashtable *ht) | |
1105 | { | |
1106 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
1107 | } | |
7e1e7763 | 1108 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |
da20420f HX |
1109 | |
1110 | struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl, | |
1111 | unsigned int hash) | |
1112 | { | |
1113 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1114 | static struct rhash_head __rcu *rhnull = | |
1115 | (struct rhash_head __rcu *)NULLS_MARKER(0); | |
1116 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1117 | unsigned int size = tbl->size >> tbl->nest; | |
1118 | unsigned int subhash = hash; | |
1119 | union nested_table *ntbl; | |
1120 | ||
1121 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
c4d2603d | 1122 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash); |
da20420f HX |
1123 | subhash >>= tbl->nest; |
1124 | ||
1125 | while (ntbl && size > (1 << shift)) { | |
1126 | index = subhash & ((1 << shift) - 1); | |
c4d2603d HX |
1127 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, |
1128 | tbl, hash); | |
da20420f HX |
1129 | size >>= shift; |
1130 | subhash >>= shift; | |
1131 | } | |
1132 | ||
1133 | if (!ntbl) | |
1134 | return &rhnull; | |
1135 | ||
1136 | return &ntbl[subhash].bucket; | |
1137 | ||
1138 | } | |
1139 | EXPORT_SYMBOL_GPL(rht_bucket_nested); | |
1140 | ||
1141 | struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht, | |
1142 | struct bucket_table *tbl, | |
1143 | unsigned int hash) | |
1144 | { | |
1145 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1146 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1147 | unsigned int size = tbl->size >> tbl->nest; | |
1148 | union nested_table *ntbl; | |
1149 | unsigned int shifted; | |
1150 | unsigned int nhash; | |
1151 | ||
1152 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
1153 | hash >>= tbl->nest; | |
1154 | nhash = index; | |
1155 | shifted = tbl->nest; | |
1156 | ntbl = nested_table_alloc(ht, &ntbl[index].table, | |
1157 | size <= (1 << shift) ? shifted : 0, nhash); | |
1158 | ||
1159 | while (ntbl && size > (1 << shift)) { | |
1160 | index = hash & ((1 << shift) - 1); | |
1161 | size >>= shift; | |
1162 | hash >>= shift; | |
1163 | nhash |= index << shifted; | |
1164 | shifted += shift; | |
1165 | ntbl = nested_table_alloc(ht, &ntbl[index].table, | |
1166 | size <= (1 << shift) ? shifted : 0, | |
1167 | nhash); | |
1168 | } | |
1169 | ||
1170 | if (!ntbl) | |
1171 | return NULL; | |
1172 | ||
1173 | return &ntbl[hash].bucket; | |
1174 | ||
1175 | } | |
1176 | EXPORT_SYMBOL_GPL(rht_bucket_nested_insert); |