]>
Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
7e1e7763 TG |
2 | /* |
3 | * Resizable, Scalable, Concurrent Hash Table | |
4 | * | |
02fd97c3 | 5 | * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au> |
a5ec68e3 | 6 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> |
7e1e7763 TG |
7 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> |
8 | * | |
7e1e7763 | 9 | * Code partially derived from nft_hash |
02fd97c3 HX |
10 | * Rewritten with rehash code from br_multicast plus single list |
11 | * pointer as suggested by Josh Triplett | |
7e1e7763 TG |
12 | */ |
13 | ||
07ee0722 | 14 | #include <linux/atomic.h> |
7e1e7763 TG |
15 | #include <linux/kernel.h> |
16 | #include <linux/init.h> | |
17 | #include <linux/log2.h> | |
5beb5c90 | 18 | #include <linux/sched.h> |
b2d09103 | 19 | #include <linux/rculist.h> |
7e1e7763 TG |
20 | #include <linux/slab.h> |
21 | #include <linux/vmalloc.h> | |
22 | #include <linux/mm.h> | |
87545899 | 23 | #include <linux/jhash.h> |
7e1e7763 TG |
24 | #include <linux/random.h> |
25 | #include <linux/rhashtable.h> | |
61d7b097 | 26 | #include <linux/err.h> |
6d795413 | 27 | #include <linux/export.h> |
7e1e7763 TG |
28 | |
29 | #define HASH_DEFAULT_SIZE 64UL | |
c2e213cf | 30 | #define HASH_MIN_SIZE 4U |
97defe1e | 31 | |
da20420f HX |
32 | union nested_table { |
33 | union nested_table __rcu *table; | |
ba6306e3 | 34 | struct rhash_lock_head *bucket; |
da20420f HX |
35 | }; |
36 | ||
988dfbd7 | 37 | static u32 head_hashfn(struct rhashtable *ht, |
8d24c0b4 TG |
38 | const struct bucket_table *tbl, |
39 | const struct rhash_head *he) | |
7e1e7763 | 40 | { |
02fd97c3 | 41 | return rht_head_hashfn(ht, tbl, he, ht->p); |
7e1e7763 TG |
42 | } |
43 | ||
a03eaec0 | 44 | #ifdef CONFIG_PROVE_LOCKING |
a03eaec0 | 45 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) |
a03eaec0 TG |
46 | |
47 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
48 | { | |
49 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; | |
50 | } | |
51 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
52 | ||
53 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
54 | { | |
8f0db018 N |
55 | if (!debug_locks) |
56 | return 1; | |
57 | if (unlikely(tbl->nest)) | |
58 | return 1; | |
ca0b709d | 59 | return bit_spin_is_locked(0, (unsigned long *)&tbl->buckets[hash]); |
a03eaec0 TG |
60 | } |
61 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
62 | #else | |
63 | #define ASSERT_RHT_MUTEX(HT) | |
a03eaec0 TG |
64 | #endif |
65 | ||
4a3084aa HX |
66 | static inline union nested_table *nested_table_top( |
67 | const struct bucket_table *tbl) | |
68 | { | |
69 | /* The top-level bucket entry does not need RCU protection | |
70 | * because it's set at the same time as tbl->nest. | |
71 | */ | |
72 | return (void *)rcu_dereference_protected(tbl->buckets[0], 1); | |
73 | } | |
74 | ||
da20420f HX |
75 | static void nested_table_free(union nested_table *ntbl, unsigned int size) |
76 | { | |
77 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
78 | const unsigned int len = 1 << shift; | |
79 | unsigned int i; | |
80 | ||
4a3084aa | 81 | ntbl = rcu_dereference_protected(ntbl->table, 1); |
da20420f HX |
82 | if (!ntbl) |
83 | return; | |
84 | ||
85 | if (size > len) { | |
86 | size >>= shift; | |
87 | for (i = 0; i < len; i++) | |
88 | nested_table_free(ntbl + i, size); | |
89 | } | |
90 | ||
91 | kfree(ntbl); | |
92 | } | |
93 | ||
94 | static void nested_bucket_table_free(const struct bucket_table *tbl) | |
95 | { | |
96 | unsigned int size = tbl->size >> tbl->nest; | |
97 | unsigned int len = 1 << tbl->nest; | |
98 | union nested_table *ntbl; | |
99 | unsigned int i; | |
100 | ||
4a3084aa | 101 | ntbl = nested_table_top(tbl); |
da20420f HX |
102 | |
103 | for (i = 0; i < len; i++) | |
104 | nested_table_free(ntbl + i, size); | |
105 | ||
106 | kfree(ntbl); | |
107 | } | |
108 | ||
97defe1e TG |
109 | static void bucket_table_free(const struct bucket_table *tbl) |
110 | { | |
da20420f HX |
111 | if (tbl->nest) |
112 | nested_bucket_table_free(tbl); | |
113 | ||
97defe1e TG |
114 | kvfree(tbl); |
115 | } | |
116 | ||
9d901bc0 HX |
117 | static void bucket_table_free_rcu(struct rcu_head *head) |
118 | { | |
119 | bucket_table_free(container_of(head, struct bucket_table, rcu)); | |
120 | } | |
121 | ||
da20420f HX |
122 | static union nested_table *nested_table_alloc(struct rhashtable *ht, |
123 | union nested_table __rcu **prev, | |
5af68ef7 | 124 | bool leaf) |
da20420f HX |
125 | { |
126 | union nested_table *ntbl; | |
127 | int i; | |
128 | ||
129 | ntbl = rcu_dereference(*prev); | |
130 | if (ntbl) | |
131 | return ntbl; | |
132 | ||
133 | ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC); | |
134 | ||
5af68ef7 N |
135 | if (ntbl && leaf) { |
136 | for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0]); i++) | |
9b4f64a2 | 137 | INIT_RHT_NULLS_HEAD(ntbl[i].bucket); |
da20420f HX |
138 | } |
139 | ||
e9458a4e | 140 | if (cmpxchg((union nested_table **)prev, NULL, ntbl) == NULL) |
7a41c294 N |
141 | return ntbl; |
142 | /* Raced with another thread. */ | |
143 | kfree(ntbl); | |
144 | return rcu_dereference(*prev); | |
da20420f HX |
145 | } |
146 | ||
147 | static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht, | |
148 | size_t nbuckets, | |
149 | gfp_t gfp) | |
150 | { | |
151 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
152 | struct bucket_table *tbl; | |
153 | size_t size; | |
154 | ||
155 | if (nbuckets < (1 << (shift + 1))) | |
156 | return NULL; | |
157 | ||
158 | size = sizeof(*tbl) + sizeof(tbl->buckets[0]); | |
159 | ||
160 | tbl = kzalloc(size, gfp); | |
161 | if (!tbl) | |
162 | return NULL; | |
163 | ||
164 | if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets, | |
5af68ef7 | 165 | false)) { |
da20420f HX |
166 | kfree(tbl); |
167 | return NULL; | |
168 | } | |
169 | ||
170 | tbl->nest = (ilog2(nbuckets) - 1) % shift + 1; | |
171 | ||
172 | return tbl; | |
173 | } | |
174 | ||
97defe1e | 175 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
b9ecfdaa HX |
176 | size_t nbuckets, |
177 | gfp_t gfp) | |
7e1e7763 | 178 | { |
eb6d1abf | 179 | struct bucket_table *tbl = NULL; |
8f0db018 | 180 | size_t size; |
f89bd6f8 | 181 | int i; |
149212f0 | 182 | static struct lock_class_key __key; |
7e1e7763 | 183 | |
c252aa3e | 184 | tbl = kvzalloc(struct_size(tbl, buckets, nbuckets), gfp); |
da20420f HX |
185 | |
186 | size = nbuckets; | |
187 | ||
2d22ecf6 | 188 | if (tbl == NULL && (gfp & ~__GFP_NOFAIL) != GFP_KERNEL) { |
da20420f HX |
189 | tbl = nested_bucket_table_alloc(ht, nbuckets, gfp); |
190 | nbuckets = 0; | |
191 | } | |
2d22ecf6 | 192 | |
7e1e7763 TG |
193 | if (tbl == NULL) |
194 | return NULL; | |
195 | ||
149212f0 N |
196 | lockdep_init_map(&tbl->dep_map, "rhashtable_bucket", &__key, 0); |
197 | ||
da20420f | 198 | tbl->size = size; |
7e1e7763 | 199 | |
4feb7c7a | 200 | rcu_head_init(&tbl->rcu); |
eddee5ba HX |
201 | INIT_LIST_HEAD(&tbl->walkers); |
202 | ||
d48ad080 | 203 | tbl->hash_rnd = get_random_u32(); |
5269b53d | 204 | |
f89bd6f8 | 205 | for (i = 0; i < nbuckets; i++) |
9b4f64a2 | 206 | INIT_RHT_NULLS_HEAD(tbl->buckets[i]); |
f89bd6f8 | 207 | |
97defe1e | 208 | return tbl; |
7e1e7763 TG |
209 | } |
210 | ||
b824478b HX |
211 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
212 | struct bucket_table *tbl) | |
213 | { | |
214 | struct bucket_table *new_tbl; | |
215 | ||
216 | do { | |
217 | new_tbl = tbl; | |
218 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
219 | } while (tbl); | |
220 | ||
221 | return new_tbl; | |
222 | } | |
223 | ||
8f0db018 | 224 | static int rhashtable_rehash_one(struct rhashtable *ht, |
ba6306e3 | 225 | struct rhash_lock_head **bkt, |
8f0db018 | 226 | 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 | 230 | int err = -EAGAIN; |
aa34a6cb | 231 | struct rhash_head *head, *next, *entry; |
e4edbe3c | 232 | struct rhash_head __rcu **pprev = NULL; |
299e5c32 | 233 | unsigned int new_hash; |
aa34a6cb | 234 | |
da20420f HX |
235 | if (new_tbl->nest) |
236 | goto out; | |
237 | ||
238 | err = -ENOENT; | |
239 | ||
adc6a3ab N |
240 | rht_for_each_from(entry, rht_ptr(bkt, old_tbl, old_hash), |
241 | old_tbl, old_hash) { | |
aa34a6cb HX |
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 | |
149212f0 | 256 | rht_lock_nested(new_tbl, &new_tbl->buckets[new_hash], SINGLE_DEPTH_NESTING); |
7e1e7763 | 257 | |
adc6a3ab | 258 | head = rht_ptr(new_tbl->buckets + new_hash, new_tbl, new_hash); |
97defe1e | 259 | |
7def0f95 | 260 | RCU_INIT_POINTER(entry->next, head); |
a5ec68e3 | 261 | |
149212f0 | 262 | rht_assign_unlock(new_tbl, &new_tbl->buckets[new_hash], entry); |
97defe1e | 263 | |
8f0db018 N |
264 | if (pprev) |
265 | rcu_assign_pointer(*pprev, next); | |
266 | else | |
267 | /* Need to preserved the bit lock. */ | |
f4712b46 | 268 | rht_assign_locked(bkt, next); |
7e1e7763 | 269 | |
aa34a6cb HX |
270 | out: |
271 | return err; | |
272 | } | |
97defe1e | 273 | |
da20420f | 274 | static int rhashtable_rehash_chain(struct rhashtable *ht, |
299e5c32 | 275 | unsigned int old_hash) |
aa34a6cb HX |
276 | { |
277 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
ba6306e3 | 278 | struct rhash_lock_head **bkt = rht_bucket_var(old_tbl, old_hash); |
da20420f | 279 | int err; |
aa34a6cb | 280 | |
8f0db018 N |
281 | if (!bkt) |
282 | return 0; | |
149212f0 | 283 | rht_lock(old_tbl, bkt); |
a5ec68e3 | 284 | |
8f0db018 | 285 | while (!(err = rhashtable_rehash_one(ht, bkt, old_hash))) |
aa34a6cb | 286 | ; |
da20420f | 287 | |
4feb7c7a | 288 | if (err == -ENOENT) |
da20420f | 289 | err = 0; |
149212f0 | 290 | rht_unlock(old_tbl, bkt); |
da20420f HX |
291 | |
292 | return err; | |
97defe1e TG |
293 | } |
294 | ||
b824478b HX |
295 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
296 | struct bucket_table *old_tbl, | |
297 | struct bucket_table *new_tbl) | |
97defe1e | 298 | { |
aa34a6cb HX |
299 | /* Make insertions go into the new, empty table right away. Deletions |
300 | * and lookups will be attempted in both tables until we synchronize. | |
0ad66449 N |
301 | * As cmpxchg() provides strong barriers, we do not need |
302 | * rcu_assign_pointer(). | |
aa34a6cb | 303 | */ |
aa34a6cb | 304 | |
e9458a4e HX |
305 | if (cmpxchg((struct bucket_table **)&old_tbl->future_tbl, NULL, |
306 | new_tbl) != NULL) | |
0ad66449 | 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 | 487 | static void *rhashtable_lookup_one(struct rhashtable *ht, |
ba6306e3 | 488 | struct rhash_lock_head **bkt, |
ca26893f HX |
489 | struct bucket_table *tbl, unsigned int hash, |
490 | const void *key, struct rhash_head *obj) | |
02fd97c3 | 491 | { |
ca26893f HX |
492 | struct rhashtable_compare_arg arg = { |
493 | .ht = ht, | |
494 | .key = key, | |
495 | }; | |
e4edbe3c | 496 | struct rhash_head __rcu **pprev = NULL; |
02fd97c3 | 497 | struct rhash_head *head; |
ca26893f | 498 | int elasticity; |
02fd97c3 | 499 | |
5f8ddeab | 500 | elasticity = RHT_ELASTICITY; |
adc6a3ab | 501 | rht_for_each_from(head, rht_ptr(bkt, tbl, hash), tbl, hash) { |
ca26893f HX |
502 | struct rhlist_head *list; |
503 | struct rhlist_head *plist; | |
504 | ||
505 | elasticity--; | |
506 | if (!key || | |
507 | (ht->p.obj_cmpfn ? | |
508 | ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
d3dcf8eb PB |
509 | rhashtable_compare(&arg, rht_obj(ht, head)))) { |
510 | pprev = &head->next; | |
ca26893f | 511 | continue; |
d3dcf8eb | 512 | } |
ca26893f HX |
513 | |
514 | if (!ht->rhlist) | |
515 | return rht_obj(ht, head); | |
516 | ||
517 | list = container_of(obj, struct rhlist_head, rhead); | |
518 | plist = container_of(head, struct rhlist_head, rhead); | |
519 | ||
520 | RCU_INIT_POINTER(list->next, plist); | |
521 | head = rht_dereference_bucket(head->next, tbl, hash); | |
522 | RCU_INIT_POINTER(list->rhead.next, head); | |
8f0db018 N |
523 | if (pprev) |
524 | rcu_assign_pointer(*pprev, obj); | |
525 | else | |
526 | /* Need to preserve the bit lock */ | |
f4712b46 | 527 | rht_assign_locked(bkt, obj); |
ca26893f HX |
528 | |
529 | return NULL; | |
5ca8cc5b | 530 | } |
02fd97c3 | 531 | |
ca26893f HX |
532 | if (elasticity <= 0) |
533 | return ERR_PTR(-EAGAIN); | |
534 | ||
535 | return ERR_PTR(-ENOENT); | |
536 | } | |
537 | ||
538 | static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht, | |
ba6306e3 | 539 | struct rhash_lock_head **bkt, |
ca26893f HX |
540 | struct bucket_table *tbl, |
541 | unsigned int hash, | |
542 | struct rhash_head *obj, | |
543 | void *data) | |
544 | { | |
545 | struct bucket_table *new_tbl; | |
546 | struct rhash_head *head; | |
547 | ||
548 | if (!IS_ERR_OR_NULL(data)) | |
549 | return ERR_PTR(-EEXIST); | |
07ee0722 | 550 | |
ca26893f HX |
551 | if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT) |
552 | return ERR_CAST(data); | |
ccd57b1b | 553 | |
c0690016 | 554 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
ca26893f HX |
555 | if (new_tbl) |
556 | return new_tbl; | |
557 | ||
558 | if (PTR_ERR(data) != -ENOENT) | |
559 | return ERR_CAST(data); | |
560 | ||
561 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
562 | return ERR_PTR(-E2BIG); | |
563 | ||
564 | if (unlikely(rht_grow_above_100(ht, tbl))) | |
565 | return ERR_PTR(-EAGAIN); | |
02fd97c3 | 566 | |
adc6a3ab | 567 | head = rht_ptr(bkt, tbl, hash); |
02fd97c3 HX |
568 | |
569 | RCU_INIT_POINTER(obj->next, head); | |
ca26893f HX |
570 | if (ht->rhlist) { |
571 | struct rhlist_head *list; | |
572 | ||
573 | list = container_of(obj, struct rhlist_head, rhead); | |
574 | RCU_INIT_POINTER(list->next, NULL); | |
575 | } | |
02fd97c3 | 576 | |
8f0db018 N |
577 | /* bkt is always the head of the list, so it holds |
578 | * the lock, which we need to preserve | |
579 | */ | |
f4712b46 | 580 | rht_assign_locked(bkt, obj); |
02fd97c3 HX |
581 | |
582 | atomic_inc(&ht->nelems); | |
ca26893f HX |
583 | if (rht_grow_above_75(ht, tbl)) |
584 | schedule_work(&ht->run_work); | |
02fd97c3 | 585 | |
ca26893f HX |
586 | return NULL; |
587 | } | |
02fd97c3 | 588 | |
ca26893f HX |
589 | static void *rhashtable_try_insert(struct rhashtable *ht, const void *key, |
590 | struct rhash_head *obj) | |
591 | { | |
592 | struct bucket_table *new_tbl; | |
593 | struct bucket_table *tbl; | |
ba6306e3 | 594 | struct rhash_lock_head **bkt; |
ca26893f | 595 | unsigned int hash; |
ca26893f HX |
596 | void *data; |
597 | ||
4feb7c7a | 598 | new_tbl = rcu_dereference(ht->tbl); |
ca26893f | 599 | |
4feb7c7a | 600 | do { |
ca26893f HX |
601 | tbl = new_tbl; |
602 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
8f0db018 N |
603 | if (rcu_access_pointer(tbl->future_tbl)) |
604 | /* Failure is OK */ | |
605 | bkt = rht_bucket_var(tbl, hash); | |
606 | else | |
607 | bkt = rht_bucket_insert(ht, tbl, hash); | |
608 | if (bkt == NULL) { | |
609 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
610 | data = ERR_PTR(-EAGAIN); | |
611 | } else { | |
149212f0 | 612 | rht_lock(tbl, bkt); |
8f0db018 N |
613 | data = rhashtable_lookup_one(ht, bkt, tbl, |
614 | hash, key, obj); | |
615 | new_tbl = rhashtable_insert_one(ht, bkt, tbl, | |
616 | hash, obj, data); | |
617 | if (PTR_ERR(new_tbl) != -EEXIST) | |
618 | data = ERR_CAST(new_tbl); | |
619 | ||
149212f0 | 620 | rht_unlock(tbl, bkt); |
8f0db018 | 621 | } |
4feb7c7a | 622 | } while (!IS_ERR_OR_NULL(new_tbl)); |
ca26893f HX |
623 | |
624 | if (PTR_ERR(data) == -EAGAIN) | |
625 | data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?: | |
626 | -EAGAIN); | |
627 | ||
628 | return data; | |
629 | } | |
630 | ||
631 | void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
632 | struct rhash_head *obj) | |
633 | { | |
634 | void *data; | |
635 | ||
636 | do { | |
637 | rcu_read_lock(); | |
638 | data = rhashtable_try_insert(ht, key, obj); | |
639 | rcu_read_unlock(); | |
640 | } while (PTR_ERR(data) == -EAGAIN); | |
641 | ||
642 | return data; | |
02fd97c3 HX |
643 | } |
644 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
645 | ||
f2dba9c6 | 646 | /** |
246779dd | 647 | * rhashtable_walk_enter - Initialise an iterator |
f2dba9c6 HX |
648 | * @ht: Table to walk over |
649 | * @iter: Hash table Iterator | |
650 | * | |
651 | * This function prepares a hash table walk. | |
652 | * | |
653 | * Note that if you restart a walk after rhashtable_walk_stop you | |
654 | * may see the same object twice. Also, you may miss objects if | |
655 | * there are removals in between rhashtable_walk_stop and the next | |
656 | * call to rhashtable_walk_start. | |
657 | * | |
658 | * For a completely stable walk you should construct your own data | |
659 | * structure outside the hash table. | |
660 | * | |
82266e98 N |
661 | * This function may be called from any process context, including |
662 | * non-preemptable context, but cannot be called from softirq or | |
663 | * hardirq context. | |
f2dba9c6 | 664 | * |
246779dd | 665 | * You must call rhashtable_walk_exit after this function returns. |
f2dba9c6 | 666 | */ |
246779dd | 667 | void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter) |
f2dba9c6 HX |
668 | { |
669 | iter->ht = ht; | |
670 | iter->p = NULL; | |
671 | iter->slot = 0; | |
672 | iter->skip = 0; | |
2db54b47 | 673 | iter->end_of_table = 0; |
f2dba9c6 | 674 | |
c6ff5268 | 675 | spin_lock(&ht->lock); |
246779dd | 676 | iter->walker.tbl = |
179ccc0a | 677 | rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock)); |
246779dd | 678 | list_add(&iter->walker.list, &iter->walker.tbl->walkers); |
c6ff5268 | 679 | spin_unlock(&ht->lock); |
f2dba9c6 | 680 | } |
246779dd | 681 | EXPORT_SYMBOL_GPL(rhashtable_walk_enter); |
f2dba9c6 HX |
682 | |
683 | /** | |
684 | * rhashtable_walk_exit - Free an iterator | |
685 | * @iter: Hash table Iterator | |
686 | * | |
6c4128f6 | 687 | * This function frees resources allocated by rhashtable_walk_enter. |
f2dba9c6 HX |
688 | */ |
689 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
690 | { | |
c6ff5268 | 691 | spin_lock(&iter->ht->lock); |
246779dd HX |
692 | if (iter->walker.tbl) |
693 | list_del(&iter->walker.list); | |
c6ff5268 | 694 | spin_unlock(&iter->ht->lock); |
f2dba9c6 HX |
695 | } |
696 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
697 | ||
698 | /** | |
97a6ec4a | 699 | * rhashtable_walk_start_check - Start a hash table walk |
f2dba9c6 HX |
700 | * @iter: Hash table iterator |
701 | * | |
0647169c AG |
702 | * Start a hash table walk at the current iterator position. Note that we take |
703 | * the RCU lock in all cases including when we return an error. So you must | |
704 | * always call rhashtable_walk_stop to clean up. | |
f2dba9c6 HX |
705 | * |
706 | * Returns zero if successful. | |
707 | * | |
708 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
709 | * will rewind back to the beginning and you may use it immediately | |
710 | * by calling rhashtable_walk_next. | |
97a6ec4a TH |
711 | * |
712 | * rhashtable_walk_start is defined as an inline variant that returns | |
713 | * void. This is preferred in cases where the caller would ignore | |
714 | * resize events and always continue. | |
f2dba9c6 | 715 | */ |
97a6ec4a | 716 | int rhashtable_walk_start_check(struct rhashtable_iter *iter) |
db4374f4 | 717 | __acquires(RCU) |
f2dba9c6 | 718 | { |
eddee5ba | 719 | struct rhashtable *ht = iter->ht; |
5d240a89 | 720 | bool rhlist = ht->rhlist; |
eddee5ba | 721 | |
c6ff5268 | 722 | rcu_read_lock(); |
eddee5ba | 723 | |
c6ff5268 | 724 | spin_lock(&ht->lock); |
246779dd HX |
725 | if (iter->walker.tbl) |
726 | list_del(&iter->walker.list); | |
c6ff5268 | 727 | spin_unlock(&ht->lock); |
eddee5ba | 728 | |
5d240a89 N |
729 | if (iter->end_of_table) |
730 | return 0; | |
731 | if (!iter->walker.tbl) { | |
246779dd | 732 | iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht); |
b41cc04b N |
733 | iter->slot = 0; |
734 | iter->skip = 0; | |
f2dba9c6 HX |
735 | return -EAGAIN; |
736 | } | |
737 | ||
5d240a89 N |
738 | if (iter->p && !rhlist) { |
739 | /* | |
740 | * We need to validate that 'p' is still in the table, and | |
741 | * if so, update 'skip' | |
742 | */ | |
743 | struct rhash_head *p; | |
744 | int skip = 0; | |
745 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
746 | skip++; | |
747 | if (p == iter->p) { | |
748 | iter->skip = skip; | |
749 | goto found; | |
750 | } | |
751 | } | |
752 | iter->p = NULL; | |
753 | } else if (iter->p && rhlist) { | |
754 | /* Need to validate that 'list' is still in the table, and | |
755 | * if so, update 'skip' and 'p'. | |
756 | */ | |
757 | struct rhash_head *p; | |
758 | struct rhlist_head *list; | |
759 | int skip = 0; | |
760 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
761 | for (list = container_of(p, struct rhlist_head, rhead); | |
762 | list; | |
763 | list = rcu_dereference(list->next)) { | |
764 | skip++; | |
765 | if (list == iter->list) { | |
766 | iter->p = p; | |
c643ecf3 | 767 | iter->skip = skip; |
5d240a89 N |
768 | goto found; |
769 | } | |
770 | } | |
771 | } | |
772 | iter->p = NULL; | |
773 | } | |
774 | found: | |
f2dba9c6 HX |
775 | return 0; |
776 | } | |
97a6ec4a | 777 | EXPORT_SYMBOL_GPL(rhashtable_walk_start_check); |
f2dba9c6 HX |
778 | |
779 | /** | |
2db54b47 TH |
780 | * __rhashtable_walk_find_next - Find the next element in a table (or the first |
781 | * one in case of a new walk). | |
782 | * | |
f2dba9c6 HX |
783 | * @iter: Hash table iterator |
784 | * | |
2db54b47 | 785 | * Returns the found object or NULL when the end of the table is reached. |
f2dba9c6 | 786 | * |
2db54b47 | 787 | * Returns -EAGAIN if resize event occurred. |
f2dba9c6 | 788 | */ |
2db54b47 | 789 | static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter) |
f2dba9c6 | 790 | { |
246779dd | 791 | struct bucket_table *tbl = iter->walker.tbl; |
ca26893f | 792 | struct rhlist_head *list = iter->list; |
f2dba9c6 HX |
793 | struct rhashtable *ht = iter->ht; |
794 | struct rhash_head *p = iter->p; | |
ca26893f | 795 | bool rhlist = ht->rhlist; |
f2dba9c6 | 796 | |
2db54b47 TH |
797 | if (!tbl) |
798 | return NULL; | |
f2dba9c6 HX |
799 | |
800 | for (; iter->slot < tbl->size; iter->slot++) { | |
801 | int skip = iter->skip; | |
802 | ||
803 | rht_for_each_rcu(p, tbl, iter->slot) { | |
ca26893f HX |
804 | if (rhlist) { |
805 | list = container_of(p, struct rhlist_head, | |
806 | rhead); | |
807 | do { | |
808 | if (!skip) | |
809 | goto next; | |
810 | skip--; | |
811 | list = rcu_dereference(list->next); | |
812 | } while (list); | |
813 | ||
814 | continue; | |
815 | } | |
f2dba9c6 HX |
816 | if (!skip) |
817 | break; | |
818 | skip--; | |
819 | } | |
820 | ||
821 | next: | |
822 | if (!rht_is_a_nulls(p)) { | |
823 | iter->skip++; | |
824 | iter->p = p; | |
ca26893f HX |
825 | iter->list = list; |
826 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
f2dba9c6 HX |
827 | } |
828 | ||
829 | iter->skip = 0; | |
830 | } | |
831 | ||
142b942a PS |
832 | iter->p = NULL; |
833 | ||
d88252f9 HX |
834 | /* Ensure we see any new tables. */ |
835 | smp_rmb(); | |
836 | ||
246779dd HX |
837 | iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
838 | if (iter->walker.tbl) { | |
f2dba9c6 HX |
839 | iter->slot = 0; |
840 | iter->skip = 0; | |
f2dba9c6 | 841 | return ERR_PTR(-EAGAIN); |
2db54b47 TH |
842 | } else { |
843 | iter->end_of_table = true; | |
f2dba9c6 HX |
844 | } |
845 | ||
c936a79f | 846 | return NULL; |
f2dba9c6 | 847 | } |
2db54b47 TH |
848 | |
849 | /** | |
850 | * rhashtable_walk_next - Return the next object and advance the iterator | |
851 | * @iter: Hash table iterator | |
852 | * | |
853 | * Note that you must call rhashtable_walk_stop when you are finished | |
854 | * with the walk. | |
855 | * | |
856 | * Returns the next object or NULL when the end of the table is reached. | |
857 | * | |
858 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
859 | * will rewind back to the beginning and you may continue to use it. | |
860 | */ | |
861 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
862 | { | |
863 | struct rhlist_head *list = iter->list; | |
864 | struct rhashtable *ht = iter->ht; | |
865 | struct rhash_head *p = iter->p; | |
866 | bool rhlist = ht->rhlist; | |
867 | ||
868 | if (p) { | |
869 | if (!rhlist || !(list = rcu_dereference(list->next))) { | |
870 | p = rcu_dereference(p->next); | |
871 | list = container_of(p, struct rhlist_head, rhead); | |
872 | } | |
873 | if (!rht_is_a_nulls(p)) { | |
874 | iter->skip++; | |
875 | iter->p = p; | |
876 | iter->list = list; | |
877 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
878 | } | |
879 | ||
880 | /* At the end of this slot, switch to next one and then find | |
881 | * next entry from that point. | |
882 | */ | |
883 | iter->skip = 0; | |
884 | iter->slot++; | |
885 | } | |
886 | ||
887 | return __rhashtable_walk_find_next(iter); | |
888 | } | |
f2dba9c6 HX |
889 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); |
890 | ||
2db54b47 TH |
891 | /** |
892 | * rhashtable_walk_peek - Return the next object but don't advance the iterator | |
893 | * @iter: Hash table iterator | |
894 | * | |
895 | * Returns the next object or NULL when the end of the table is reached. | |
896 | * | |
897 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
898 | * will rewind back to the beginning and you may continue to use it. | |
899 | */ | |
900 | void *rhashtable_walk_peek(struct rhashtable_iter *iter) | |
901 | { | |
902 | struct rhlist_head *list = iter->list; | |
903 | struct rhashtable *ht = iter->ht; | |
904 | struct rhash_head *p = iter->p; | |
905 | ||
906 | if (p) | |
907 | return rht_obj(ht, ht->rhlist ? &list->rhead : p); | |
908 | ||
909 | /* No object found in current iter, find next one in the table. */ | |
910 | ||
911 | if (iter->skip) { | |
912 | /* A nonzero skip value points to the next entry in the table | |
913 | * beyond that last one that was found. Decrement skip so | |
914 | * we find the current value. __rhashtable_walk_find_next | |
915 | * will restore the original value of skip assuming that | |
916 | * the table hasn't changed. | |
917 | */ | |
918 | iter->skip--; | |
919 | } | |
920 | ||
921 | return __rhashtable_walk_find_next(iter); | |
922 | } | |
923 | EXPORT_SYMBOL_GPL(rhashtable_walk_peek); | |
924 | ||
f2dba9c6 HX |
925 | /** |
926 | * rhashtable_walk_stop - Finish a hash table walk | |
927 | * @iter: Hash table iterator | |
928 | * | |
0647169c AG |
929 | * Finish a hash table walk. Does not reset the iterator to the start of the |
930 | * hash table. | |
f2dba9c6 HX |
931 | */ |
932 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 933 | __releases(RCU) |
f2dba9c6 | 934 | { |
eddee5ba | 935 | struct rhashtable *ht; |
246779dd | 936 | struct bucket_table *tbl = iter->walker.tbl; |
eddee5ba | 937 | |
eddee5ba | 938 | if (!tbl) |
963ecbd4 | 939 | goto out; |
eddee5ba HX |
940 | |
941 | ht = iter->ht; | |
942 | ||
ba7c95ea | 943 | spin_lock(&ht->lock); |
4feb7c7a N |
944 | if (rcu_head_after_call_rcu(&tbl->rcu, bucket_table_free_rcu)) |
945 | /* This bucket table is being freed, don't re-link it. */ | |
246779dd | 946 | iter->walker.tbl = NULL; |
4feb7c7a N |
947 | else |
948 | list_add(&iter->walker.list, &tbl->walkers); | |
ba7c95ea | 949 | spin_unlock(&ht->lock); |
eddee5ba | 950 | |
963ecbd4 HX |
951 | out: |
952 | rcu_read_unlock(); | |
f2dba9c6 HX |
953 | } |
954 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
955 | ||
488fb86e | 956 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 957 | { |
107d01f5 DB |
958 | size_t retsize; |
959 | ||
960 | if (params->nelem_hint) | |
961 | retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3), | |
962 | (unsigned long)params->min_size); | |
963 | else | |
964 | retsize = max(HASH_DEFAULT_SIZE, | |
965 | (unsigned long)params->min_size); | |
966 | ||
967 | return retsize; | |
7e1e7763 TG |
968 | } |
969 | ||
31ccde2d HX |
970 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
971 | { | |
972 | return jhash2(key, length, seed); | |
973 | } | |
974 | ||
7e1e7763 TG |
975 | /** |
976 | * rhashtable_init - initialize a new hash table | |
977 | * @ht: hash table to be initialized | |
978 | * @params: configuration parameters | |
979 | * | |
980 | * Initializes a new hash table based on the provided configuration | |
981 | * parameters. A table can be configured either with a variable or | |
982 | * fixed length key: | |
983 | * | |
984 | * Configuration Example 1: Fixed length keys | |
985 | * struct test_obj { | |
986 | * int key; | |
987 | * void * my_member; | |
988 | * struct rhash_head node; | |
989 | * }; | |
990 | * | |
991 | * struct rhashtable_params params = { | |
992 | * .head_offset = offsetof(struct test_obj, node), | |
993 | * .key_offset = offsetof(struct test_obj, key), | |
994 | * .key_len = sizeof(int), | |
87545899 | 995 | * .hashfn = jhash, |
7e1e7763 TG |
996 | * }; |
997 | * | |
998 | * Configuration Example 2: Variable length keys | |
999 | * struct test_obj { | |
1000 | * [...] | |
1001 | * struct rhash_head node; | |
1002 | * }; | |
1003 | * | |
49f7b33e | 1004 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
1005 | * { |
1006 | * struct test_obj *obj = data; | |
1007 | * | |
1008 | * return [... hash ...]; | |
1009 | * } | |
1010 | * | |
1011 | * struct rhashtable_params params = { | |
1012 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 1013 | * .hashfn = jhash, |
7e1e7763 | 1014 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
1015 | * }; |
1016 | */ | |
488fb86e HX |
1017 | int rhashtable_init(struct rhashtable *ht, |
1018 | const struct rhashtable_params *params) | |
7e1e7763 TG |
1019 | { |
1020 | struct bucket_table *tbl; | |
1021 | size_t size; | |
1022 | ||
31ccde2d | 1023 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 1024 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
1025 | return -EINVAL; |
1026 | ||
97defe1e TG |
1027 | memset(ht, 0, sizeof(*ht)); |
1028 | mutex_init(&ht->mutex); | |
ba7c95ea | 1029 | spin_lock_init(&ht->lock); |
97defe1e TG |
1030 | memcpy(&ht->p, params, sizeof(*params)); |
1031 | ||
a998f712 TG |
1032 | if (params->min_size) |
1033 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
1034 | ||
6d684e54 HX |
1035 | /* Cap total entries at 2^31 to avoid nelems overflow. */ |
1036 | ht->max_elems = 1u << 31; | |
2d2ab658 HX |
1037 | |
1038 | if (params->max_size) { | |
1039 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
1040 | if (ht->p.max_size < ht->max_elems / 2) | |
1041 | ht->max_elems = ht->p.max_size * 2; | |
1042 | } | |
6d684e54 | 1043 | |
48e75b43 | 1044 | ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 1045 | |
107d01f5 | 1046 | size = rounded_hashtable_size(&ht->p); |
3a324606 | 1047 | |
31ccde2d HX |
1048 | ht->key_len = ht->p.key_len; |
1049 | if (!params->hashfn) { | |
1050 | ht->p.hashfn = jhash; | |
1051 | ||
1052 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
1053 | ht->key_len /= sizeof(u32); | |
1054 | ht->p.hashfn = rhashtable_jhash2; | |
1055 | } | |
1056 | } | |
1057 | ||
2d22ecf6 DB |
1058 | /* |
1059 | * This is api initialization and thus we need to guarantee the | |
1060 | * initial rhashtable allocation. Upon failure, retry with the | |
1061 | * smallest possible size with __GFP_NOFAIL semantics. | |
1062 | */ | |
b9ecfdaa | 1063 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
2d22ecf6 DB |
1064 | if (unlikely(tbl == NULL)) { |
1065 | size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); | |
1066 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL); | |
1067 | } | |
7e1e7763 | 1068 | |
545a148e | 1069 | atomic_set(&ht->nelems, 0); |
a5b6846f | 1070 | |
7e1e7763 TG |
1071 | RCU_INIT_POINTER(ht->tbl, tbl); |
1072 | ||
4c4b52d9 | 1073 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1074 | |
7e1e7763 TG |
1075 | return 0; |
1076 | } | |
1077 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1078 | ||
ca26893f HX |
1079 | /** |
1080 | * rhltable_init - initialize a new hash list table | |
1081 | * @hlt: hash list table to be initialized | |
1082 | * @params: configuration parameters | |
1083 | * | |
1084 | * Initializes a new hash list table. | |
1085 | * | |
1086 | * See documentation for rhashtable_init. | |
1087 | */ | |
1088 | int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params) | |
1089 | { | |
1090 | int err; | |
1091 | ||
ca26893f HX |
1092 | err = rhashtable_init(&hlt->ht, params); |
1093 | hlt->ht.rhlist = true; | |
1094 | return err; | |
1095 | } | |
1096 | EXPORT_SYMBOL_GPL(rhltable_init); | |
1097 | ||
1098 | static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj, | |
1099 | void (*free_fn)(void *ptr, void *arg), | |
1100 | void *arg) | |
1101 | { | |
1102 | struct rhlist_head *list; | |
1103 | ||
1104 | if (!ht->rhlist) { | |
1105 | free_fn(rht_obj(ht, obj), arg); | |
1106 | return; | |
1107 | } | |
1108 | ||
1109 | list = container_of(obj, struct rhlist_head, rhead); | |
1110 | do { | |
1111 | obj = &list->rhead; | |
1112 | list = rht_dereference(list->next, ht); | |
1113 | free_fn(rht_obj(ht, obj), arg); | |
1114 | } while (list); | |
1115 | } | |
1116 | ||
7e1e7763 | 1117 | /** |
6b6f302c | 1118 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 1119 | * @ht: the hash table to destroy |
6b6f302c TG |
1120 | * @free_fn: callback to release resources of element |
1121 | * @arg: pointer passed to free_fn | |
7e1e7763 | 1122 | * |
6b6f302c TG |
1123 | * Stops an eventual async resize. If defined, invokes free_fn for each |
1124 | * element to releasal resources. Please note that RCU protected | |
1125 | * readers may still be accessing the elements. Releasing of resources | |
1126 | * must occur in a compatible manner. Then frees the bucket array. | |
1127 | * | |
1128 | * This function will eventually sleep to wait for an async resize | |
1129 | * to complete. The caller is responsible that no further write operations | |
1130 | * occurs in parallel. | |
7e1e7763 | 1131 | */ |
6b6f302c TG |
1132 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
1133 | void (*free_fn)(void *ptr, void *arg), | |
1134 | void *arg) | |
7e1e7763 | 1135 | { |
0026129c | 1136 | struct bucket_table *tbl, *next_tbl; |
6b6f302c | 1137 | unsigned int i; |
97defe1e | 1138 | |
4c4b52d9 | 1139 | cancel_work_sync(&ht->run_work); |
97defe1e | 1140 | |
57699a40 | 1141 | mutex_lock(&ht->mutex); |
6b6f302c | 1142 | tbl = rht_dereference(ht->tbl, ht); |
0026129c | 1143 | restart: |
6b6f302c TG |
1144 | if (free_fn) { |
1145 | for (i = 0; i < tbl->size; i++) { | |
1146 | struct rhash_head *pos, *next; | |
1147 | ||
ae6da1f5 | 1148 | cond_resched(); |
adc6a3ab | 1149 | for (pos = rht_ptr_exclusive(rht_bucket(tbl, i)), |
6b6f302c TG |
1150 | next = !rht_is_a_nulls(pos) ? |
1151 | rht_dereference(pos->next, ht) : NULL; | |
1152 | !rht_is_a_nulls(pos); | |
1153 | pos = next, | |
1154 | next = !rht_is_a_nulls(pos) ? | |
1155 | rht_dereference(pos->next, ht) : NULL) | |
ca26893f | 1156 | rhashtable_free_one(ht, pos, free_fn, arg); |
6b6f302c TG |
1157 | } |
1158 | } | |
1159 | ||
0026129c | 1160 | next_tbl = rht_dereference(tbl->future_tbl, ht); |
6b6f302c | 1161 | bucket_table_free(tbl); |
0026129c TY |
1162 | if (next_tbl) { |
1163 | tbl = next_tbl; | |
1164 | goto restart; | |
1165 | } | |
97defe1e | 1166 | mutex_unlock(&ht->mutex); |
7e1e7763 | 1167 | } |
6b6f302c TG |
1168 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
1169 | ||
1170 | void rhashtable_destroy(struct rhashtable *ht) | |
1171 | { | |
1172 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
1173 | } | |
7e1e7763 | 1174 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |
da20420f | 1175 | |
ba6306e3 HX |
1176 | struct rhash_lock_head **__rht_bucket_nested(const struct bucket_table *tbl, |
1177 | unsigned int hash) | |
da20420f HX |
1178 | { |
1179 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
da20420f HX |
1180 | unsigned int index = hash & ((1 << tbl->nest) - 1); |
1181 | unsigned int size = tbl->size >> tbl->nest; | |
1182 | unsigned int subhash = hash; | |
1183 | union nested_table *ntbl; | |
1184 | ||
4a3084aa | 1185 | ntbl = nested_table_top(tbl); |
c4d2603d | 1186 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash); |
da20420f HX |
1187 | subhash >>= tbl->nest; |
1188 | ||
1189 | while (ntbl && size > (1 << shift)) { | |
1190 | index = subhash & ((1 << shift) - 1); | |
c4d2603d HX |
1191 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, |
1192 | tbl, hash); | |
da20420f HX |
1193 | size >>= shift; |
1194 | subhash >>= shift; | |
1195 | } | |
1196 | ||
ff302db9 N |
1197 | if (!ntbl) |
1198 | return NULL; | |
da20420f HX |
1199 | |
1200 | return &ntbl[subhash].bucket; | |
1201 | ||
1202 | } | |
ff302db9 N |
1203 | EXPORT_SYMBOL_GPL(__rht_bucket_nested); |
1204 | ||
ba6306e3 HX |
1205 | struct rhash_lock_head **rht_bucket_nested(const struct bucket_table *tbl, |
1206 | unsigned int hash) | |
ff302db9 | 1207 | { |
ba6306e3 | 1208 | static struct rhash_lock_head *rhnull; |
ff302db9 N |
1209 | |
1210 | if (!rhnull) | |
1211 | INIT_RHT_NULLS_HEAD(rhnull); | |
1212 | return __rht_bucket_nested(tbl, hash) ?: &rhnull; | |
1213 | } | |
da20420f HX |
1214 | EXPORT_SYMBOL_GPL(rht_bucket_nested); |
1215 | ||
ba6306e3 HX |
1216 | struct rhash_lock_head **rht_bucket_nested_insert(struct rhashtable *ht, |
1217 | struct bucket_table *tbl, | |
1218 | unsigned int hash) | |
da20420f HX |
1219 | { |
1220 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1221 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1222 | unsigned int size = tbl->size >> tbl->nest; | |
1223 | union nested_table *ntbl; | |
da20420f | 1224 | |
4a3084aa | 1225 | ntbl = nested_table_top(tbl); |
da20420f | 1226 | hash >>= tbl->nest; |
da20420f | 1227 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1228 | size <= (1 << shift)); |
da20420f HX |
1229 | |
1230 | while (ntbl && size > (1 << shift)) { | |
1231 | index = hash & ((1 << shift) - 1); | |
1232 | size >>= shift; | |
1233 | hash >>= shift; | |
da20420f | 1234 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1235 | size <= (1 << shift)); |
da20420f HX |
1236 | } |
1237 | ||
1238 | if (!ntbl) | |
1239 | return NULL; | |
1240 | ||
1241 | return &ntbl[hash].bucket; | |
1242 | ||
1243 | } | |
1244 | EXPORT_SYMBOL_GPL(rht_bucket_nested_insert); |