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, | |
118 | unsigned int shifted, | |
119 | unsigned int nhash) | |
120 | { | |
121 | union nested_table *ntbl; | |
122 | int i; | |
123 | ||
124 | ntbl = rcu_dereference(*prev); | |
125 | if (ntbl) | |
126 | return ntbl; | |
127 | ||
128 | ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC); | |
129 | ||
130 | if (ntbl && shifted) { | |
131 | for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0].bucket); i++) | |
132 | INIT_RHT_NULLS_HEAD(ntbl[i].bucket, ht, | |
133 | (i << shifted) | nhash); | |
134 | } | |
135 | ||
136 | rcu_assign_pointer(*prev, ntbl); | |
137 | ||
138 | return ntbl; | |
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, | |
159 | 0, 0)) { | |
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]); | |
12e8fd6f | 178 | if (gfp != GFP_KERNEL) |
b9ecfdaa | 179 | tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY); |
12e8fd6f MH |
180 | else |
181 | tbl = kvzalloc(size, gfp); | |
da20420f HX |
182 | |
183 | size = nbuckets; | |
184 | ||
185 | if (tbl == NULL && gfp != GFP_KERNEL) { | |
186 | tbl = nested_bucket_table_alloc(ht, nbuckets, gfp); | |
187 | nbuckets = 0; | |
188 | } | |
7e1e7763 TG |
189 | if (tbl == NULL) |
190 | return NULL; | |
191 | ||
da20420f | 192 | tbl->size = size; |
7e1e7763 | 193 | |
64e0cd0d TH |
194 | max_locks = size >> 1; |
195 | if (tbl->nest) | |
196 | max_locks = min_t(size_t, max_locks, 1U << tbl->nest); | |
197 | ||
198 | if (alloc_bucket_spinlocks(&tbl->locks, &tbl->locks_mask, max_locks, | |
199 | ht->p.locks_mul, gfp) < 0) { | |
97defe1e TG |
200 | bucket_table_free(tbl); |
201 | return NULL; | |
202 | } | |
7e1e7763 | 203 | |
eddee5ba HX |
204 | INIT_LIST_HEAD(&tbl->walkers); |
205 | ||
d48ad080 | 206 | tbl->hash_rnd = get_random_u32(); |
5269b53d | 207 | |
f89bd6f8 TG |
208 | for (i = 0; i < nbuckets; i++) |
209 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); | |
210 | ||
97defe1e | 211 | return tbl; |
7e1e7763 TG |
212 | } |
213 | ||
b824478b HX |
214 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
215 | struct bucket_table *tbl) | |
216 | { | |
217 | struct bucket_table *new_tbl; | |
218 | ||
219 | do { | |
220 | new_tbl = tbl; | |
221 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
222 | } while (tbl); | |
223 | ||
224 | return new_tbl; | |
225 | } | |
226 | ||
299e5c32 | 227 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash) |
a5ec68e3 | 228 | { |
aa34a6cb | 229 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
b824478b HX |
230 | struct bucket_table *new_tbl = rhashtable_last_table(ht, |
231 | rht_dereference_rcu(old_tbl->future_tbl, ht)); | |
da20420f HX |
232 | struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash); |
233 | int err = -EAGAIN; | |
aa34a6cb HX |
234 | struct rhash_head *head, *next, *entry; |
235 | spinlock_t *new_bucket_lock; | |
299e5c32 | 236 | unsigned int new_hash; |
aa34a6cb | 237 | |
da20420f HX |
238 | if (new_tbl->nest) |
239 | goto out; | |
240 | ||
241 | err = -ENOENT; | |
242 | ||
aa34a6cb HX |
243 | rht_for_each(entry, old_tbl, old_hash) { |
244 | err = 0; | |
245 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
246 | ||
247 | if (rht_is_a_nulls(next)) | |
248 | break; | |
a5ec68e3 | 249 | |
aa34a6cb HX |
250 | pprev = &entry->next; |
251 | } | |
a5ec68e3 | 252 | |
aa34a6cb HX |
253 | if (err) |
254 | goto out; | |
97defe1e | 255 | |
aa34a6cb | 256 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 257 | |
02fd97c3 | 258 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 259 | |
8f2484bd | 260 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
261 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
262 | new_tbl, new_hash); | |
97defe1e | 263 | |
7def0f95 | 264 | RCU_INIT_POINTER(entry->next, head); |
a5ec68e3 | 265 | |
aa34a6cb HX |
266 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
267 | spin_unlock(new_bucket_lock); | |
97defe1e | 268 | |
aa34a6cb | 269 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 270 | |
aa34a6cb HX |
271 | out: |
272 | return err; | |
273 | } | |
97defe1e | 274 | |
da20420f | 275 | static int rhashtable_rehash_chain(struct rhashtable *ht, |
299e5c32 | 276 | unsigned int old_hash) |
aa34a6cb HX |
277 | { |
278 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
279 | spinlock_t *old_bucket_lock; | |
da20420f | 280 | int err; |
aa34a6cb | 281 | |
02fd97c3 | 282 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 283 | |
aa34a6cb | 284 | spin_lock_bh(old_bucket_lock); |
da20420f | 285 | while (!(err = rhashtable_rehash_one(ht, old_hash))) |
aa34a6cb | 286 | ; |
da20420f HX |
287 | |
288 | if (err == -ENOENT) { | |
289 | old_tbl->rehash++; | |
290 | err = 0; | |
291 | } | |
aa34a6cb | 292 | spin_unlock_bh(old_bucket_lock); |
da20420f HX |
293 | |
294 | return err; | |
97defe1e TG |
295 | } |
296 | ||
b824478b HX |
297 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
298 | struct bucket_table *old_tbl, | |
299 | struct bucket_table *new_tbl) | |
97defe1e | 300 | { |
b824478b HX |
301 | /* Protect future_tbl using the first bucket lock. */ |
302 | spin_lock_bh(old_tbl->locks); | |
303 | ||
304 | /* Did somebody beat us to it? */ | |
305 | if (rcu_access_pointer(old_tbl->future_tbl)) { | |
306 | spin_unlock_bh(old_tbl->locks); | |
307 | return -EEXIST; | |
308 | } | |
7cd10db8 | 309 | |
aa34a6cb HX |
310 | /* Make insertions go into the new, empty table right away. Deletions |
311 | * and lookups will be attempted in both tables until we synchronize. | |
aa34a6cb | 312 | */ |
c4db8848 | 313 | rcu_assign_pointer(old_tbl->future_tbl, new_tbl); |
aa34a6cb | 314 | |
b824478b HX |
315 | spin_unlock_bh(old_tbl->locks); |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
320 | static int rhashtable_rehash_table(struct rhashtable *ht) | |
321 | { | |
322 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
323 | struct bucket_table *new_tbl; | |
324 | struct rhashtable_walker *walker; | |
299e5c32 | 325 | unsigned int old_hash; |
da20420f | 326 | int err; |
b824478b HX |
327 | |
328 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); | |
329 | if (!new_tbl) | |
330 | return 0; | |
331 | ||
da20420f HX |
332 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
333 | err = rhashtable_rehash_chain(ht, old_hash); | |
334 | if (err) | |
335 | return err; | |
ae6da1f5 | 336 | cond_resched(); |
da20420f | 337 | } |
aa34a6cb HX |
338 | |
339 | /* Publish the new table pointer. */ | |
340 | rcu_assign_pointer(ht->tbl, new_tbl); | |
341 | ||
ba7c95ea | 342 | spin_lock(&ht->lock); |
eddee5ba HX |
343 | list_for_each_entry(walker, &old_tbl->walkers, list) |
344 | walker->tbl = NULL; | |
ba7c95ea | 345 | spin_unlock(&ht->lock); |
eddee5ba | 346 | |
aa34a6cb HX |
347 | /* Wait for readers. All new readers will see the new |
348 | * table, and thus no references to the old table will | |
349 | * remain. | |
350 | */ | |
9d901bc0 | 351 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
b824478b HX |
352 | |
353 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; | |
7e1e7763 TG |
354 | } |
355 | ||
da20420f HX |
356 | static int rhashtable_rehash_alloc(struct rhashtable *ht, |
357 | struct bucket_table *old_tbl, | |
358 | unsigned int size) | |
7e1e7763 | 359 | { |
da20420f | 360 | struct bucket_table *new_tbl; |
b824478b | 361 | int err; |
7e1e7763 TG |
362 | |
363 | ASSERT_RHT_MUTEX(ht); | |
364 | ||
da20420f | 365 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
366 | if (new_tbl == NULL) |
367 | return -ENOMEM; | |
368 | ||
b824478b HX |
369 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
370 | if (err) | |
371 | bucket_table_free(new_tbl); | |
372 | ||
373 | return err; | |
7e1e7763 | 374 | } |
7e1e7763 TG |
375 | |
376 | /** | |
377 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
378 | * @ht: the hash table to shrink | |
7e1e7763 | 379 | * |
18093d1c HX |
380 | * This function shrinks the hash table to fit, i.e., the smallest |
381 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 382 | * |
97defe1e TG |
383 | * The caller must ensure that no concurrent resizing occurs by holding |
384 | * ht->mutex. | |
385 | * | |
7e1e7763 TG |
386 | * The caller must ensure that no concurrent table mutations take place. |
387 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
388 | * |
389 | * It is valid to have concurrent insertions and deletions protected by per | |
390 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 391 | */ |
b824478b | 392 | static int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 393 | { |
da20420f | 394 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
12311959 VN |
395 | unsigned int nelems = atomic_read(&ht->nelems); |
396 | unsigned int size = 0; | |
7e1e7763 | 397 | |
12311959 VN |
398 | if (nelems) |
399 | size = roundup_pow_of_two(nelems * 3 / 2); | |
18093d1c HX |
400 | if (size < ht->p.min_size) |
401 | size = ht->p.min_size; | |
402 | ||
403 | if (old_tbl->size <= size) | |
404 | return 0; | |
405 | ||
b824478b HX |
406 | if (rht_dereference(old_tbl->future_tbl, ht)) |
407 | return -EEXIST; | |
408 | ||
da20420f | 409 | return rhashtable_rehash_alloc(ht, old_tbl, size); |
7e1e7763 | 410 | } |
7e1e7763 | 411 | |
97defe1e TG |
412 | static void rht_deferred_worker(struct work_struct *work) |
413 | { | |
414 | struct rhashtable *ht; | |
415 | struct bucket_table *tbl; | |
b824478b | 416 | int err = 0; |
97defe1e | 417 | |
57699a40 | 418 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 419 | mutex_lock(&ht->mutex); |
28134a53 | 420 | |
97defe1e | 421 | tbl = rht_dereference(ht->tbl, ht); |
b824478b | 422 | tbl = rhashtable_last_table(ht, tbl); |
97defe1e | 423 | |
a5b6846f | 424 | if (rht_grow_above_75(ht, tbl)) |
da20420f | 425 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2); |
b5e2c150 | 426 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
da20420f HX |
427 | err = rhashtable_shrink(ht); |
428 | else if (tbl->nest) | |
429 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size); | |
b824478b | 430 | |
da20420f HX |
431 | if (!err) |
432 | err = rhashtable_rehash_table(ht); | |
b824478b | 433 | |
97defe1e | 434 | mutex_unlock(&ht->mutex); |
b824478b HX |
435 | |
436 | if (err) | |
437 | schedule_work(&ht->run_work); | |
97defe1e TG |
438 | } |
439 | ||
ca26893f HX |
440 | static int rhashtable_insert_rehash(struct rhashtable *ht, |
441 | struct bucket_table *tbl) | |
ccd57b1b HX |
442 | { |
443 | struct bucket_table *old_tbl; | |
444 | struct bucket_table *new_tbl; | |
ccd57b1b HX |
445 | unsigned int size; |
446 | int err; | |
447 | ||
448 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
ccd57b1b HX |
449 | |
450 | size = tbl->size; | |
451 | ||
3cf92222 HX |
452 | err = -EBUSY; |
453 | ||
ccd57b1b HX |
454 | if (rht_grow_above_75(ht, tbl)) |
455 | size *= 2; | |
a87b9ebf TG |
456 | /* Do not schedule more than one rehash */ |
457 | else if (old_tbl != tbl) | |
3cf92222 HX |
458 | goto fail; |
459 | ||
460 | err = -ENOMEM; | |
ccd57b1b HX |
461 | |
462 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC); | |
3cf92222 HX |
463 | if (new_tbl == NULL) |
464 | goto fail; | |
ccd57b1b HX |
465 | |
466 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); | |
467 | if (err) { | |
468 | bucket_table_free(new_tbl); | |
469 | if (err == -EEXIST) | |
470 | err = 0; | |
471 | } else | |
472 | schedule_work(&ht->run_work); | |
473 | ||
474 | return err; | |
3cf92222 HX |
475 | |
476 | fail: | |
477 | /* Do not fail the insert if someone else did a rehash. */ | |
478 | if (likely(rcu_dereference_raw(tbl->future_tbl))) | |
479 | return 0; | |
480 | ||
481 | /* Schedule async rehash to retry allocation in process context. */ | |
482 | if (err == -ENOMEM) | |
483 | schedule_work(&ht->run_work); | |
484 | ||
485 | return err; | |
ccd57b1b | 486 | } |
ccd57b1b | 487 | |
ca26893f HX |
488 | static void *rhashtable_lookup_one(struct rhashtable *ht, |
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 | }; | |
496 | struct rhash_head __rcu **pprev; | |
02fd97c3 | 497 | struct rhash_head *head; |
ca26893f | 498 | int elasticity; |
02fd97c3 | 499 | |
5f8ddeab | 500 | elasticity = RHT_ELASTICITY; |
da20420f HX |
501 | pprev = rht_bucket_var(tbl, hash); |
502 | rht_for_each_continue(head, *pprev, tbl, hash) { | |
ca26893f HX |
503 | struct rhlist_head *list; |
504 | struct rhlist_head *plist; | |
505 | ||
506 | elasticity--; | |
507 | if (!key || | |
508 | (ht->p.obj_cmpfn ? | |
509 | ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
d3dcf8eb PB |
510 | rhashtable_compare(&arg, rht_obj(ht, head)))) { |
511 | pprev = &head->next; | |
ca26893f | 512 | continue; |
d3dcf8eb | 513 | } |
ca26893f HX |
514 | |
515 | if (!ht->rhlist) | |
516 | return rht_obj(ht, head); | |
517 | ||
518 | list = container_of(obj, struct rhlist_head, rhead); | |
519 | plist = container_of(head, struct rhlist_head, rhead); | |
520 | ||
521 | RCU_INIT_POINTER(list->next, plist); | |
522 | head = rht_dereference_bucket(head->next, tbl, hash); | |
523 | RCU_INIT_POINTER(list->rhead.next, head); | |
524 | rcu_assign_pointer(*pprev, obj); | |
525 | ||
526 | return NULL; | |
5ca8cc5b | 527 | } |
02fd97c3 | 528 | |
ca26893f HX |
529 | if (elasticity <= 0) |
530 | return ERR_PTR(-EAGAIN); | |
531 | ||
532 | return ERR_PTR(-ENOENT); | |
533 | } | |
534 | ||
535 | static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht, | |
536 | struct bucket_table *tbl, | |
537 | unsigned int hash, | |
538 | struct rhash_head *obj, | |
539 | void *data) | |
540 | { | |
da20420f | 541 | struct rhash_head __rcu **pprev; |
ca26893f HX |
542 | struct bucket_table *new_tbl; |
543 | struct rhash_head *head; | |
544 | ||
545 | if (!IS_ERR_OR_NULL(data)) | |
546 | return ERR_PTR(-EEXIST); | |
07ee0722 | 547 | |
ca26893f HX |
548 | if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT) |
549 | return ERR_CAST(data); | |
ccd57b1b | 550 | |
ca26893f HX |
551 | new_tbl = rcu_dereference(tbl->future_tbl); |
552 | if (new_tbl) | |
553 | return new_tbl; | |
554 | ||
555 | if (PTR_ERR(data) != -ENOENT) | |
556 | return ERR_CAST(data); | |
557 | ||
558 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
559 | return ERR_PTR(-E2BIG); | |
560 | ||
561 | if (unlikely(rht_grow_above_100(ht, tbl))) | |
562 | return ERR_PTR(-EAGAIN); | |
02fd97c3 | 563 | |
da20420f HX |
564 | pprev = rht_bucket_insert(ht, tbl, hash); |
565 | if (!pprev) | |
566 | return ERR_PTR(-ENOMEM); | |
567 | ||
568 | head = rht_dereference_bucket(*pprev, tbl, hash); | |
02fd97c3 HX |
569 | |
570 | RCU_INIT_POINTER(obj->next, head); | |
ca26893f HX |
571 | if (ht->rhlist) { |
572 | struct rhlist_head *list; | |
573 | ||
574 | list = container_of(obj, struct rhlist_head, rhead); | |
575 | RCU_INIT_POINTER(list->next, NULL); | |
576 | } | |
02fd97c3 | 577 | |
da20420f | 578 | rcu_assign_pointer(*pprev, obj); |
02fd97c3 HX |
579 | |
580 | atomic_inc(&ht->nelems); | |
ca26893f HX |
581 | if (rht_grow_above_75(ht, tbl)) |
582 | schedule_work(&ht->run_work); | |
02fd97c3 | 583 | |
ca26893f HX |
584 | return NULL; |
585 | } | |
02fd97c3 | 586 | |
ca26893f HX |
587 | static void *rhashtable_try_insert(struct rhashtable *ht, const void *key, |
588 | struct rhash_head *obj) | |
589 | { | |
590 | struct bucket_table *new_tbl; | |
591 | struct bucket_table *tbl; | |
592 | unsigned int hash; | |
593 | spinlock_t *lock; | |
594 | void *data; | |
595 | ||
596 | tbl = rcu_dereference(ht->tbl); | |
597 | ||
598 | /* All insertions must grab the oldest table containing | |
599 | * the hashed bucket that is yet to be rehashed. | |
600 | */ | |
601 | for (;;) { | |
602 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
603 | lock = rht_bucket_lock(tbl, hash); | |
604 | spin_lock_bh(lock); | |
605 | ||
606 | if (tbl->rehash <= hash) | |
607 | break; | |
608 | ||
609 | spin_unlock_bh(lock); | |
610 | tbl = rcu_dereference(tbl->future_tbl); | |
611 | } | |
612 | ||
613 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
614 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
615 | if (PTR_ERR(new_tbl) != -EEXIST) | |
616 | data = ERR_CAST(new_tbl); | |
617 | ||
618 | while (!IS_ERR_OR_NULL(new_tbl)) { | |
619 | tbl = new_tbl; | |
620 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
621 | spin_lock_nested(rht_bucket_lock(tbl, hash), | |
622 | SINGLE_DEPTH_NESTING); | |
623 | ||
624 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
625 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
626 | if (PTR_ERR(new_tbl) != -EEXIST) | |
627 | data = ERR_CAST(new_tbl); | |
628 | ||
629 | spin_unlock(rht_bucket_lock(tbl, hash)); | |
630 | } | |
631 | ||
632 | spin_unlock_bh(lock); | |
633 | ||
634 | if (PTR_ERR(data) == -EAGAIN) | |
635 | data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?: | |
636 | -EAGAIN); | |
637 | ||
638 | return data; | |
639 | } | |
640 | ||
641 | void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
642 | struct rhash_head *obj) | |
643 | { | |
644 | void *data; | |
645 | ||
646 | do { | |
647 | rcu_read_lock(); | |
648 | data = rhashtable_try_insert(ht, key, obj); | |
649 | rcu_read_unlock(); | |
650 | } while (PTR_ERR(data) == -EAGAIN); | |
651 | ||
652 | return data; | |
02fd97c3 HX |
653 | } |
654 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
655 | ||
f2dba9c6 | 656 | /** |
246779dd | 657 | * rhashtable_walk_enter - Initialise an iterator |
f2dba9c6 HX |
658 | * @ht: Table to walk over |
659 | * @iter: Hash table Iterator | |
660 | * | |
661 | * This function prepares a hash table walk. | |
662 | * | |
663 | * Note that if you restart a walk after rhashtable_walk_stop you | |
664 | * may see the same object twice. Also, you may miss objects if | |
665 | * there are removals in between rhashtable_walk_stop and the next | |
666 | * call to rhashtable_walk_start. | |
667 | * | |
668 | * For a completely stable walk you should construct your own data | |
669 | * structure outside the hash table. | |
670 | * | |
82266e98 N |
671 | * This function may be called from any process context, including |
672 | * non-preemptable context, but cannot be called from softirq or | |
673 | * hardirq context. | |
f2dba9c6 | 674 | * |
246779dd | 675 | * You must call rhashtable_walk_exit after this function returns. |
f2dba9c6 | 676 | */ |
246779dd | 677 | void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter) |
f2dba9c6 HX |
678 | { |
679 | iter->ht = ht; | |
680 | iter->p = NULL; | |
681 | iter->slot = 0; | |
682 | iter->skip = 0; | |
2db54b47 | 683 | iter->end_of_table = 0; |
f2dba9c6 | 684 | |
c6ff5268 | 685 | spin_lock(&ht->lock); |
246779dd | 686 | iter->walker.tbl = |
179ccc0a | 687 | rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock)); |
246779dd | 688 | list_add(&iter->walker.list, &iter->walker.tbl->walkers); |
c6ff5268 | 689 | spin_unlock(&ht->lock); |
f2dba9c6 | 690 | } |
246779dd | 691 | EXPORT_SYMBOL_GPL(rhashtable_walk_enter); |
f2dba9c6 HX |
692 | |
693 | /** | |
694 | * rhashtable_walk_exit - Free an iterator | |
695 | * @iter: Hash table Iterator | |
696 | * | |
697 | * This function frees resources allocated by rhashtable_walk_init. | |
698 | */ | |
699 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
700 | { | |
c6ff5268 | 701 | spin_lock(&iter->ht->lock); |
246779dd HX |
702 | if (iter->walker.tbl) |
703 | list_del(&iter->walker.list); | |
c6ff5268 | 704 | spin_unlock(&iter->ht->lock); |
f2dba9c6 HX |
705 | } |
706 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
707 | ||
708 | /** | |
97a6ec4a | 709 | * rhashtable_walk_start_check - Start a hash table walk |
f2dba9c6 HX |
710 | * @iter: Hash table iterator |
711 | * | |
0647169c AG |
712 | * Start a hash table walk at the current iterator position. Note that we take |
713 | * the RCU lock in all cases including when we return an error. So you must | |
714 | * always call rhashtable_walk_stop to clean up. | |
f2dba9c6 HX |
715 | * |
716 | * Returns zero if successful. | |
717 | * | |
718 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
719 | * will rewind back to the beginning and you may use it immediately | |
720 | * by calling rhashtable_walk_next. | |
97a6ec4a TH |
721 | * |
722 | * rhashtable_walk_start is defined as an inline variant that returns | |
723 | * void. This is preferred in cases where the caller would ignore | |
724 | * resize events and always continue. | |
f2dba9c6 | 725 | */ |
97a6ec4a | 726 | int rhashtable_walk_start_check(struct rhashtable_iter *iter) |
db4374f4 | 727 | __acquires(RCU) |
f2dba9c6 | 728 | { |
eddee5ba | 729 | struct rhashtable *ht = iter->ht; |
5d240a89 | 730 | bool rhlist = ht->rhlist; |
eddee5ba | 731 | |
c6ff5268 | 732 | rcu_read_lock(); |
eddee5ba | 733 | |
c6ff5268 | 734 | spin_lock(&ht->lock); |
246779dd HX |
735 | if (iter->walker.tbl) |
736 | list_del(&iter->walker.list); | |
c6ff5268 | 737 | spin_unlock(&ht->lock); |
eddee5ba | 738 | |
5d240a89 N |
739 | if (iter->end_of_table) |
740 | return 0; | |
741 | if (!iter->walker.tbl) { | |
246779dd | 742 | iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht); |
b41cc04b N |
743 | iter->slot = 0; |
744 | iter->skip = 0; | |
f2dba9c6 HX |
745 | return -EAGAIN; |
746 | } | |
747 | ||
5d240a89 N |
748 | if (iter->p && !rhlist) { |
749 | /* | |
750 | * We need to validate that 'p' is still in the table, and | |
751 | * if so, update 'skip' | |
752 | */ | |
753 | struct rhash_head *p; | |
754 | int skip = 0; | |
755 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
756 | skip++; | |
757 | if (p == iter->p) { | |
758 | iter->skip = skip; | |
759 | goto found; | |
760 | } | |
761 | } | |
762 | iter->p = NULL; | |
763 | } else if (iter->p && rhlist) { | |
764 | /* Need to validate that 'list' is still in the table, and | |
765 | * if so, update 'skip' and 'p'. | |
766 | */ | |
767 | struct rhash_head *p; | |
768 | struct rhlist_head *list; | |
769 | int skip = 0; | |
770 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
771 | for (list = container_of(p, struct rhlist_head, rhead); | |
772 | list; | |
773 | list = rcu_dereference(list->next)) { | |
774 | skip++; | |
775 | if (list == iter->list) { | |
776 | iter->p = p; | |
777 | skip = skip; | |
778 | goto found; | |
779 | } | |
780 | } | |
781 | } | |
782 | iter->p = NULL; | |
783 | } | |
784 | found: | |
f2dba9c6 HX |
785 | return 0; |
786 | } | |
97a6ec4a | 787 | EXPORT_SYMBOL_GPL(rhashtable_walk_start_check); |
f2dba9c6 HX |
788 | |
789 | /** | |
2db54b47 TH |
790 | * __rhashtable_walk_find_next - Find the next element in a table (or the first |
791 | * one in case of a new walk). | |
792 | * | |
f2dba9c6 HX |
793 | * @iter: Hash table iterator |
794 | * | |
2db54b47 | 795 | * Returns the found object or NULL when the end of the table is reached. |
f2dba9c6 | 796 | * |
2db54b47 | 797 | * Returns -EAGAIN if resize event occurred. |
f2dba9c6 | 798 | */ |
2db54b47 | 799 | static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter) |
f2dba9c6 | 800 | { |
246779dd | 801 | struct bucket_table *tbl = iter->walker.tbl; |
ca26893f | 802 | struct rhlist_head *list = iter->list; |
f2dba9c6 HX |
803 | struct rhashtable *ht = iter->ht; |
804 | struct rhash_head *p = iter->p; | |
ca26893f | 805 | bool rhlist = ht->rhlist; |
f2dba9c6 | 806 | |
2db54b47 TH |
807 | if (!tbl) |
808 | return NULL; | |
f2dba9c6 HX |
809 | |
810 | for (; iter->slot < tbl->size; iter->slot++) { | |
811 | int skip = iter->skip; | |
812 | ||
813 | rht_for_each_rcu(p, tbl, iter->slot) { | |
ca26893f HX |
814 | if (rhlist) { |
815 | list = container_of(p, struct rhlist_head, | |
816 | rhead); | |
817 | do { | |
818 | if (!skip) | |
819 | goto next; | |
820 | skip--; | |
821 | list = rcu_dereference(list->next); | |
822 | } while (list); | |
823 | ||
824 | continue; | |
825 | } | |
f2dba9c6 HX |
826 | if (!skip) |
827 | break; | |
828 | skip--; | |
829 | } | |
830 | ||
831 | next: | |
832 | if (!rht_is_a_nulls(p)) { | |
833 | iter->skip++; | |
834 | iter->p = p; | |
ca26893f HX |
835 | iter->list = list; |
836 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
f2dba9c6 HX |
837 | } |
838 | ||
839 | iter->skip = 0; | |
840 | } | |
841 | ||
142b942a PS |
842 | iter->p = NULL; |
843 | ||
d88252f9 HX |
844 | /* Ensure we see any new tables. */ |
845 | smp_rmb(); | |
846 | ||
246779dd HX |
847 | iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
848 | if (iter->walker.tbl) { | |
f2dba9c6 HX |
849 | iter->slot = 0; |
850 | iter->skip = 0; | |
f2dba9c6 | 851 | return ERR_PTR(-EAGAIN); |
2db54b47 TH |
852 | } else { |
853 | iter->end_of_table = true; | |
f2dba9c6 HX |
854 | } |
855 | ||
c936a79f | 856 | return NULL; |
f2dba9c6 | 857 | } |
2db54b47 TH |
858 | |
859 | /** | |
860 | * rhashtable_walk_next - Return the next object and advance the iterator | |
861 | * @iter: Hash table iterator | |
862 | * | |
863 | * Note that you must call rhashtable_walk_stop when you are finished | |
864 | * with the walk. | |
865 | * | |
866 | * Returns the next object or NULL when the end of the table is reached. | |
867 | * | |
868 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
869 | * will rewind back to the beginning and you may continue to use it. | |
870 | */ | |
871 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
872 | { | |
873 | struct rhlist_head *list = iter->list; | |
874 | struct rhashtable *ht = iter->ht; | |
875 | struct rhash_head *p = iter->p; | |
876 | bool rhlist = ht->rhlist; | |
877 | ||
878 | if (p) { | |
879 | if (!rhlist || !(list = rcu_dereference(list->next))) { | |
880 | p = rcu_dereference(p->next); | |
881 | list = container_of(p, struct rhlist_head, rhead); | |
882 | } | |
883 | if (!rht_is_a_nulls(p)) { | |
884 | iter->skip++; | |
885 | iter->p = p; | |
886 | iter->list = list; | |
887 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
888 | } | |
889 | ||
890 | /* At the end of this slot, switch to next one and then find | |
891 | * next entry from that point. | |
892 | */ | |
893 | iter->skip = 0; | |
894 | iter->slot++; | |
895 | } | |
896 | ||
897 | return __rhashtable_walk_find_next(iter); | |
898 | } | |
f2dba9c6 HX |
899 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); |
900 | ||
2db54b47 TH |
901 | /** |
902 | * rhashtable_walk_peek - Return the next object but don't advance the iterator | |
903 | * @iter: Hash table iterator | |
904 | * | |
905 | * Returns the next object or NULL when the end of the table is reached. | |
906 | * | |
907 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
908 | * will rewind back to the beginning and you may continue to use it. | |
909 | */ | |
910 | void *rhashtable_walk_peek(struct rhashtable_iter *iter) | |
911 | { | |
912 | struct rhlist_head *list = iter->list; | |
913 | struct rhashtable *ht = iter->ht; | |
914 | struct rhash_head *p = iter->p; | |
915 | ||
916 | if (p) | |
917 | return rht_obj(ht, ht->rhlist ? &list->rhead : p); | |
918 | ||
919 | /* No object found in current iter, find next one in the table. */ | |
920 | ||
921 | if (iter->skip) { | |
922 | /* A nonzero skip value points to the next entry in the table | |
923 | * beyond that last one that was found. Decrement skip so | |
924 | * we find the current value. __rhashtable_walk_find_next | |
925 | * will restore the original value of skip assuming that | |
926 | * the table hasn't changed. | |
927 | */ | |
928 | iter->skip--; | |
929 | } | |
930 | ||
931 | return __rhashtable_walk_find_next(iter); | |
932 | } | |
933 | EXPORT_SYMBOL_GPL(rhashtable_walk_peek); | |
934 | ||
f2dba9c6 HX |
935 | /** |
936 | * rhashtable_walk_stop - Finish a hash table walk | |
937 | * @iter: Hash table iterator | |
938 | * | |
0647169c AG |
939 | * Finish a hash table walk. Does not reset the iterator to the start of the |
940 | * hash table. | |
f2dba9c6 HX |
941 | */ |
942 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 943 | __releases(RCU) |
f2dba9c6 | 944 | { |
eddee5ba | 945 | struct rhashtable *ht; |
246779dd | 946 | struct bucket_table *tbl = iter->walker.tbl; |
eddee5ba | 947 | |
eddee5ba | 948 | if (!tbl) |
963ecbd4 | 949 | goto out; |
eddee5ba HX |
950 | |
951 | ht = iter->ht; | |
952 | ||
ba7c95ea | 953 | spin_lock(&ht->lock); |
c4db8848 | 954 | if (tbl->rehash < tbl->size) |
246779dd | 955 | list_add(&iter->walker.list, &tbl->walkers); |
eddee5ba | 956 | else |
246779dd | 957 | iter->walker.tbl = NULL; |
ba7c95ea | 958 | spin_unlock(&ht->lock); |
eddee5ba | 959 | |
963ecbd4 HX |
960 | out: |
961 | rcu_read_unlock(); | |
f2dba9c6 HX |
962 | } |
963 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
964 | ||
488fb86e | 965 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 966 | { |
94000176 | 967 | return max(roundup_pow_of_two(params->nelem_hint * 4 / 3), |
e2e21c1c | 968 | (unsigned long)params->min_size); |
7e1e7763 TG |
969 | } |
970 | ||
31ccde2d HX |
971 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
972 | { | |
973 | return jhash2(key, length, seed); | |
974 | } | |
975 | ||
7e1e7763 TG |
976 | /** |
977 | * rhashtable_init - initialize a new hash table | |
978 | * @ht: hash table to be initialized | |
979 | * @params: configuration parameters | |
980 | * | |
981 | * Initializes a new hash table based on the provided configuration | |
982 | * parameters. A table can be configured either with a variable or | |
983 | * fixed length key: | |
984 | * | |
985 | * Configuration Example 1: Fixed length keys | |
986 | * struct test_obj { | |
987 | * int key; | |
988 | * void * my_member; | |
989 | * struct rhash_head node; | |
990 | * }; | |
991 | * | |
992 | * struct rhashtable_params params = { | |
993 | * .head_offset = offsetof(struct test_obj, node), | |
994 | * .key_offset = offsetof(struct test_obj, key), | |
995 | * .key_len = sizeof(int), | |
87545899 | 996 | * .hashfn = jhash, |
f89bd6f8 | 997 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
7e1e7763 TG |
998 | * }; |
999 | * | |
1000 | * Configuration Example 2: Variable length keys | |
1001 | * struct test_obj { | |
1002 | * [...] | |
1003 | * struct rhash_head node; | |
1004 | * }; | |
1005 | * | |
49f7b33e | 1006 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
1007 | * { |
1008 | * struct test_obj *obj = data; | |
1009 | * | |
1010 | * return [... hash ...]; | |
1011 | * } | |
1012 | * | |
1013 | * struct rhashtable_params params = { | |
1014 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 1015 | * .hashfn = jhash, |
7e1e7763 | 1016 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
1017 | * }; |
1018 | */ | |
488fb86e HX |
1019 | int rhashtable_init(struct rhashtable *ht, |
1020 | const struct rhashtable_params *params) | |
7e1e7763 TG |
1021 | { |
1022 | struct bucket_table *tbl; | |
1023 | size_t size; | |
1024 | ||
1025 | size = HASH_DEFAULT_SIZE; | |
1026 | ||
31ccde2d | 1027 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 1028 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
1029 | return -EINVAL; |
1030 | ||
f89bd6f8 TG |
1031 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
1032 | return -EINVAL; | |
1033 | ||
97defe1e TG |
1034 | memset(ht, 0, sizeof(*ht)); |
1035 | mutex_init(&ht->mutex); | |
ba7c95ea | 1036 | spin_lock_init(&ht->lock); |
97defe1e TG |
1037 | memcpy(&ht->p, params, sizeof(*params)); |
1038 | ||
a998f712 TG |
1039 | if (params->min_size) |
1040 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
1041 | ||
6d684e54 HX |
1042 | /* Cap total entries at 2^31 to avoid nelems overflow. */ |
1043 | ht->max_elems = 1u << 31; | |
2d2ab658 HX |
1044 | |
1045 | if (params->max_size) { | |
1046 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
1047 | if (ht->p.max_size < ht->max_elems / 2) | |
1048 | ht->max_elems = ht->p.max_size * 2; | |
1049 | } | |
6d684e54 | 1050 | |
48e75b43 | 1051 | ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 1052 | |
3a324606 HX |
1053 | if (params->nelem_hint) |
1054 | size = rounded_hashtable_size(&ht->p); | |
1055 | ||
97defe1e TG |
1056 | if (params->locks_mul) |
1057 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
1058 | else | |
1059 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
1060 | ||
31ccde2d HX |
1061 | ht->key_len = ht->p.key_len; |
1062 | if (!params->hashfn) { | |
1063 | ht->p.hashfn = jhash; | |
1064 | ||
1065 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
1066 | ht->key_len /= sizeof(u32); | |
1067 | ht->p.hashfn = rhashtable_jhash2; | |
1068 | } | |
1069 | } | |
1070 | ||
b9ecfdaa | 1071 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
1072 | if (tbl == NULL) |
1073 | return -ENOMEM; | |
1074 | ||
545a148e | 1075 | atomic_set(&ht->nelems, 0); |
a5b6846f | 1076 | |
7e1e7763 TG |
1077 | RCU_INIT_POINTER(ht->tbl, tbl); |
1078 | ||
4c4b52d9 | 1079 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1080 | |
7e1e7763 TG |
1081 | return 0; |
1082 | } | |
1083 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1084 | ||
ca26893f HX |
1085 | /** |
1086 | * rhltable_init - initialize a new hash list table | |
1087 | * @hlt: hash list table to be initialized | |
1088 | * @params: configuration parameters | |
1089 | * | |
1090 | * Initializes a new hash list table. | |
1091 | * | |
1092 | * See documentation for rhashtable_init. | |
1093 | */ | |
1094 | int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params) | |
1095 | { | |
1096 | int err; | |
1097 | ||
1098 | /* No rhlist NULLs marking for now. */ | |
1099 | if (params->nulls_base) | |
1100 | return -EINVAL; | |
1101 | ||
1102 | err = rhashtable_init(&hlt->ht, params); | |
1103 | hlt->ht.rhlist = true; | |
1104 | return err; | |
1105 | } | |
1106 | EXPORT_SYMBOL_GPL(rhltable_init); | |
1107 | ||
1108 | static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj, | |
1109 | void (*free_fn)(void *ptr, void *arg), | |
1110 | void *arg) | |
1111 | { | |
1112 | struct rhlist_head *list; | |
1113 | ||
1114 | if (!ht->rhlist) { | |
1115 | free_fn(rht_obj(ht, obj), arg); | |
1116 | return; | |
1117 | } | |
1118 | ||
1119 | list = container_of(obj, struct rhlist_head, rhead); | |
1120 | do { | |
1121 | obj = &list->rhead; | |
1122 | list = rht_dereference(list->next, ht); | |
1123 | free_fn(rht_obj(ht, obj), arg); | |
1124 | } while (list); | |
1125 | } | |
1126 | ||
7e1e7763 | 1127 | /** |
6b6f302c | 1128 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 1129 | * @ht: the hash table to destroy |
6b6f302c TG |
1130 | * @free_fn: callback to release resources of element |
1131 | * @arg: pointer passed to free_fn | |
7e1e7763 | 1132 | * |
6b6f302c TG |
1133 | * Stops an eventual async resize. If defined, invokes free_fn for each |
1134 | * element to releasal resources. Please note that RCU protected | |
1135 | * readers may still be accessing the elements. Releasing of resources | |
1136 | * must occur in a compatible manner. Then frees the bucket array. | |
1137 | * | |
1138 | * This function will eventually sleep to wait for an async resize | |
1139 | * to complete. The caller is responsible that no further write operations | |
1140 | * occurs in parallel. | |
7e1e7763 | 1141 | */ |
6b6f302c TG |
1142 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
1143 | void (*free_fn)(void *ptr, void *arg), | |
1144 | void *arg) | |
7e1e7763 | 1145 | { |
da20420f | 1146 | struct bucket_table *tbl; |
6b6f302c | 1147 | unsigned int i; |
97defe1e | 1148 | |
4c4b52d9 | 1149 | cancel_work_sync(&ht->run_work); |
97defe1e | 1150 | |
57699a40 | 1151 | mutex_lock(&ht->mutex); |
6b6f302c TG |
1152 | tbl = rht_dereference(ht->tbl, ht); |
1153 | if (free_fn) { | |
1154 | for (i = 0; i < tbl->size; i++) { | |
1155 | struct rhash_head *pos, *next; | |
1156 | ||
ae6da1f5 | 1157 | cond_resched(); |
da20420f | 1158 | for (pos = rht_dereference(*rht_bucket(tbl, i), ht), |
6b6f302c TG |
1159 | next = !rht_is_a_nulls(pos) ? |
1160 | rht_dereference(pos->next, ht) : NULL; | |
1161 | !rht_is_a_nulls(pos); | |
1162 | pos = next, | |
1163 | next = !rht_is_a_nulls(pos) ? | |
1164 | rht_dereference(pos->next, ht) : NULL) | |
ca26893f | 1165 | rhashtable_free_one(ht, pos, free_fn, arg); |
6b6f302c TG |
1166 | } |
1167 | } | |
1168 | ||
1169 | bucket_table_free(tbl); | |
97defe1e | 1170 | mutex_unlock(&ht->mutex); |
7e1e7763 | 1171 | } |
6b6f302c TG |
1172 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
1173 | ||
1174 | void rhashtable_destroy(struct rhashtable *ht) | |
1175 | { | |
1176 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
1177 | } | |
7e1e7763 | 1178 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |
da20420f HX |
1179 | |
1180 | struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl, | |
1181 | unsigned int hash) | |
1182 | { | |
1183 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1184 | static struct rhash_head __rcu *rhnull = | |
1185 | (struct rhash_head __rcu *)NULLS_MARKER(0); | |
1186 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1187 | unsigned int size = tbl->size >> tbl->nest; | |
1188 | unsigned int subhash = hash; | |
1189 | union nested_table *ntbl; | |
1190 | ||
1191 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
c4d2603d | 1192 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash); |
da20420f HX |
1193 | subhash >>= tbl->nest; |
1194 | ||
1195 | while (ntbl && size > (1 << shift)) { | |
1196 | index = subhash & ((1 << shift) - 1); | |
c4d2603d HX |
1197 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, |
1198 | tbl, hash); | |
da20420f HX |
1199 | size >>= shift; |
1200 | subhash >>= shift; | |
1201 | } | |
1202 | ||
1203 | if (!ntbl) | |
1204 | return &rhnull; | |
1205 | ||
1206 | return &ntbl[subhash].bucket; | |
1207 | ||
1208 | } | |
1209 | EXPORT_SYMBOL_GPL(rht_bucket_nested); | |
1210 | ||
1211 | struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht, | |
1212 | struct bucket_table *tbl, | |
1213 | unsigned int hash) | |
1214 | { | |
1215 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
1216 | unsigned int index = hash & ((1 << tbl->nest) - 1); | |
1217 | unsigned int size = tbl->size >> tbl->nest; | |
1218 | union nested_table *ntbl; | |
1219 | unsigned int shifted; | |
1220 | unsigned int nhash; | |
1221 | ||
1222 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
1223 | hash >>= tbl->nest; | |
1224 | nhash = index; | |
1225 | shifted = tbl->nest; | |
1226 | ntbl = nested_table_alloc(ht, &ntbl[index].table, | |
1227 | size <= (1 << shift) ? shifted : 0, nhash); | |
1228 | ||
1229 | while (ntbl && size > (1 << shift)) { | |
1230 | index = hash & ((1 << shift) - 1); | |
1231 | size >>= shift; | |
1232 | hash >>= shift; | |
1233 | nhash |= index << shifted; | |
1234 | shifted += shift; | |
1235 | ntbl = nested_table_alloc(ht, &ntbl[index].table, | |
1236 | size <= (1 << shift) ? shifted : 0, | |
1237 | nhash); | |
1238 | } | |
1239 | ||
1240 | if (!ntbl) | |
1241 | return NULL; | |
1242 | ||
1243 | return &ntbl[hash].bucket; | |
1244 | ||
1245 | } | |
1246 | EXPORT_SYMBOL_GPL(rht_bucket_nested_insert); |