Commit | Line | Data |
---|---|---|
7e1e7763 TG |
1 | /* |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
02fd97c3 | 4 | * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au> |
a5ec68e3 | 5 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> |
7e1e7763 TG |
6 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> |
7 | * | |
7e1e7763 | 8 | * Code partially derived from nft_hash |
02fd97c3 HX |
9 | * Rewritten with rehash code from br_multicast plus single list |
10 | * pointer as suggested by Josh Triplett | |
7e1e7763 TG |
11 | * |
12 | * This program is free software; you can redistribute it and/or modify | |
13 | * it under the terms of the GNU General Public License version 2 as | |
14 | * published by the Free Software Foundation. | |
15 | */ | |
16 | ||
07ee0722 | 17 | #include <linux/atomic.h> |
7e1e7763 TG |
18 | #include <linux/kernel.h> |
19 | #include <linux/init.h> | |
20 | #include <linux/log2.h> | |
5beb5c90 | 21 | #include <linux/sched.h> |
b2d09103 | 22 | #include <linux/rculist.h> |
7e1e7763 TG |
23 | #include <linux/slab.h> |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/mm.h> | |
87545899 | 26 | #include <linux/jhash.h> |
7e1e7763 TG |
27 | #include <linux/random.h> |
28 | #include <linux/rhashtable.h> | |
61d7b097 | 29 | #include <linux/err.h> |
6d795413 | 30 | #include <linux/export.h> |
7e1e7763 TG |
31 | |
32 | #define HASH_DEFAULT_SIZE 64UL | |
c2e213cf | 33 | #define HASH_MIN_SIZE 4U |
4cf0b354 | 34 | #define BUCKET_LOCKS_PER_CPU 32UL |
97defe1e | 35 | |
da20420f HX |
36 | union nested_table { |
37 | union nested_table __rcu *table; | |
38 | struct rhash_head __rcu *bucket; | |
39 | }; | |
40 | ||
988dfbd7 | 41 | static u32 head_hashfn(struct rhashtable *ht, |
8d24c0b4 TG |
42 | const struct bucket_table *tbl, |
43 | const struct rhash_head *he) | |
7e1e7763 | 44 | { |
02fd97c3 | 45 | return rht_head_hashfn(ht, tbl, he, ht->p); |
7e1e7763 TG |
46 | } |
47 | ||
a03eaec0 | 48 | #ifdef CONFIG_PROVE_LOCKING |
a03eaec0 | 49 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) |
a03eaec0 TG |
50 | |
51 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
52 | { | |
53 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; | |
54 | } | |
55 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
56 | ||
57 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
58 | { | |
02fd97c3 | 59 | spinlock_t *lock = rht_bucket_lock(tbl, hash); |
a03eaec0 TG |
60 | |
61 | return (debug_locks) ? lockdep_is_held(lock) : 1; | |
62 | } | |
63 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
64 | #else | |
65 | #define ASSERT_RHT_MUTEX(HT) | |
a03eaec0 TG |
66 | #endif |
67 | ||
da20420f HX |
68 | static void nested_table_free(union nested_table *ntbl, unsigned int size) |
69 | { | |
70 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
71 | const unsigned int len = 1 << shift; | |
72 | unsigned int i; | |
73 | ||
74 | ntbl = rcu_dereference_raw(ntbl->table); | |
75 | if (!ntbl) | |
76 | return; | |
77 | ||
78 | if (size > len) { | |
79 | size >>= shift; | |
80 | for (i = 0; i < len; i++) | |
81 | nested_table_free(ntbl + i, size); | |
82 | } | |
83 | ||
84 | kfree(ntbl); | |
85 | } | |
86 | ||
87 | static void nested_bucket_table_free(const struct bucket_table *tbl) | |
88 | { | |
89 | unsigned int size = tbl->size >> tbl->nest; | |
90 | unsigned int len = 1 << tbl->nest; | |
91 | union nested_table *ntbl; | |
92 | unsigned int i; | |
93 | ||
94 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
95 | ||
96 | for (i = 0; i < len; i++) | |
97 | nested_table_free(ntbl + i, size); | |
98 | ||
99 | kfree(ntbl); | |
100 | } | |
101 | ||
97defe1e TG |
102 | static void bucket_table_free(const struct bucket_table *tbl) |
103 | { | |
da20420f HX |
104 | if (tbl->nest) |
105 | nested_bucket_table_free(tbl); | |
106 | ||
64e0cd0d | 107 | free_bucket_spinlocks(tbl->locks); |
97defe1e TG |
108 | kvfree(tbl); |
109 | } | |
110 | ||
9d901bc0 HX |
111 | static void bucket_table_free_rcu(struct rcu_head *head) |
112 | { | |
113 | bucket_table_free(container_of(head, struct bucket_table, rcu)); | |
114 | } | |
115 | ||
da20420f HX |
116 | static union nested_table *nested_table_alloc(struct rhashtable *ht, |
117 | union nested_table __rcu **prev, | |
5af68ef7 | 118 | bool leaf) |
da20420f HX |
119 | { |
120 | union nested_table *ntbl; | |
121 | int i; | |
122 | ||
123 | ntbl = rcu_dereference(*prev); | |
124 | if (ntbl) | |
125 | return ntbl; | |
126 | ||
127 | ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC); | |
128 | ||
5af68ef7 N |
129 | if (ntbl && leaf) { |
130 | for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0]); i++) | |
9b4f64a2 | 131 | INIT_RHT_NULLS_HEAD(ntbl[i].bucket); |
da20420f HX |
132 | } |
133 | ||
134 | rcu_assign_pointer(*prev, ntbl); | |
135 | ||
136 | return ntbl; | |
137 | } | |
138 | ||
139 | static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht, | |
140 | size_t nbuckets, | |
141 | gfp_t gfp) | |
142 | { | |
143 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
144 | struct bucket_table *tbl; | |
145 | size_t size; | |
146 | ||
147 | if (nbuckets < (1 << (shift + 1))) | |
148 | return NULL; | |
149 | ||
150 | size = sizeof(*tbl) + sizeof(tbl->buckets[0]); | |
151 | ||
152 | tbl = kzalloc(size, gfp); | |
153 | if (!tbl) | |
154 | return NULL; | |
155 | ||
156 | if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets, | |
5af68ef7 | 157 | false)) { |
da20420f HX |
158 | kfree(tbl); |
159 | return NULL; | |
160 | } | |
161 | ||
162 | tbl->nest = (ilog2(nbuckets) - 1) % shift + 1; | |
163 | ||
164 | return tbl; | |
165 | } | |
166 | ||
97defe1e | 167 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
b9ecfdaa HX |
168 | size_t nbuckets, |
169 | gfp_t gfp) | |
7e1e7763 | 170 | { |
eb6d1abf | 171 | struct bucket_table *tbl = NULL; |
64e0cd0d | 172 | size_t size, max_locks; |
f89bd6f8 | 173 | int i; |
7e1e7763 TG |
174 | |
175 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
93f976b5 | 176 | tbl = kvzalloc(size, gfp); |
da20420f HX |
177 | |
178 | size = nbuckets; | |
179 | ||
2d22ecf6 | 180 | if (tbl == NULL && (gfp & ~__GFP_NOFAIL) != GFP_KERNEL) { |
da20420f HX |
181 | tbl = nested_bucket_table_alloc(ht, nbuckets, gfp); |
182 | nbuckets = 0; | |
183 | } | |
2d22ecf6 | 184 | |
7e1e7763 TG |
185 | if (tbl == NULL) |
186 | return NULL; | |
187 | ||
da20420f | 188 | tbl->size = size; |
7e1e7763 | 189 | |
64e0cd0d TH |
190 | max_locks = size >> 1; |
191 | if (tbl->nest) | |
192 | max_locks = min_t(size_t, max_locks, 1U << tbl->nest); | |
193 | ||
194 | if (alloc_bucket_spinlocks(&tbl->locks, &tbl->locks_mask, max_locks, | |
195 | ht->p.locks_mul, gfp) < 0) { | |
97defe1e TG |
196 | bucket_table_free(tbl); |
197 | return NULL; | |
198 | } | |
7e1e7763 | 199 | |
eddee5ba HX |
200 | INIT_LIST_HEAD(&tbl->walkers); |
201 | ||
d48ad080 | 202 | tbl->hash_rnd = get_random_u32(); |
5269b53d | 203 | |
f89bd6f8 | 204 | for (i = 0; i < nbuckets; i++) |
9b4f64a2 | 205 | INIT_RHT_NULLS_HEAD(tbl->buckets[i]); |
f89bd6f8 | 206 | |
97defe1e | 207 | return tbl; |
7e1e7763 TG |
208 | } |
209 | ||
b824478b HX |
210 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
211 | struct bucket_table *tbl) | |
212 | { | |
213 | struct bucket_table *new_tbl; | |
214 | ||
215 | do { | |
216 | new_tbl = tbl; | |
217 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); | |
218 | } while (tbl); | |
219 | ||
220 | return new_tbl; | |
221 | } | |
222 | ||
299e5c32 | 223 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash) |
a5ec68e3 | 224 | { |
aa34a6cb | 225 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
c0690016 | 226 | struct bucket_table *new_tbl = rhashtable_last_table(ht, old_tbl); |
da20420f HX |
227 | struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash); |
228 | int err = -EAGAIN; | |
aa34a6cb HX |
229 | struct rhash_head *head, *next, *entry; |
230 | spinlock_t *new_bucket_lock; | |
299e5c32 | 231 | unsigned int new_hash; |
aa34a6cb | 232 | |
da20420f HX |
233 | if (new_tbl->nest) |
234 | goto out; | |
235 | ||
236 | err = -ENOENT; | |
237 | ||
aa34a6cb HX |
238 | rht_for_each(entry, old_tbl, old_hash) { |
239 | err = 0; | |
240 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); | |
241 | ||
242 | if (rht_is_a_nulls(next)) | |
243 | break; | |
a5ec68e3 | 244 | |
aa34a6cb HX |
245 | pprev = &entry->next; |
246 | } | |
a5ec68e3 | 247 | |
aa34a6cb HX |
248 | if (err) |
249 | goto out; | |
97defe1e | 250 | |
aa34a6cb | 251 | new_hash = head_hashfn(ht, new_tbl, entry); |
7e1e7763 | 252 | |
02fd97c3 | 253 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
7e1e7763 | 254 | |
8f2484bd | 255 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
aa34a6cb HX |
256 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
257 | new_tbl, new_hash); | |
97defe1e | 258 | |
7def0f95 | 259 | RCU_INIT_POINTER(entry->next, head); |
a5ec68e3 | 260 | |
aa34a6cb HX |
261 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
262 | spin_unlock(new_bucket_lock); | |
97defe1e | 263 | |
aa34a6cb | 264 | rcu_assign_pointer(*pprev, next); |
7e1e7763 | 265 | |
aa34a6cb HX |
266 | out: |
267 | return err; | |
268 | } | |
97defe1e | 269 | |
da20420f | 270 | static int rhashtable_rehash_chain(struct rhashtable *ht, |
299e5c32 | 271 | unsigned int old_hash) |
aa34a6cb HX |
272 | { |
273 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
274 | spinlock_t *old_bucket_lock; | |
da20420f | 275 | int err; |
aa34a6cb | 276 | |
02fd97c3 | 277 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
a5ec68e3 | 278 | |
aa34a6cb | 279 | spin_lock_bh(old_bucket_lock); |
da20420f | 280 | while (!(err = rhashtable_rehash_one(ht, old_hash))) |
aa34a6cb | 281 | ; |
da20420f HX |
282 | |
283 | if (err == -ENOENT) { | |
284 | old_tbl->rehash++; | |
285 | err = 0; | |
286 | } | |
aa34a6cb | 287 | spin_unlock_bh(old_bucket_lock); |
da20420f HX |
288 | |
289 | return err; | |
97defe1e TG |
290 | } |
291 | ||
b824478b HX |
292 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
293 | struct bucket_table *old_tbl, | |
294 | struct bucket_table *new_tbl) | |
97defe1e | 295 | { |
aa34a6cb HX |
296 | /* Make insertions go into the new, empty table right away. Deletions |
297 | * and lookups will be attempted in both tables until we synchronize. | |
0ad66449 N |
298 | * As cmpxchg() provides strong barriers, we do not need |
299 | * rcu_assign_pointer(). | |
aa34a6cb | 300 | */ |
aa34a6cb | 301 | |
0ad66449 N |
302 | if (cmpxchg(&old_tbl->future_tbl, NULL, new_tbl) != NULL) |
303 | return -EEXIST; | |
b824478b HX |
304 | |
305 | return 0; | |
306 | } | |
307 | ||
308 | static int rhashtable_rehash_table(struct rhashtable *ht) | |
309 | { | |
310 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); | |
311 | struct bucket_table *new_tbl; | |
312 | struct rhashtable_walker *walker; | |
299e5c32 | 313 | unsigned int old_hash; |
da20420f | 314 | int err; |
b824478b HX |
315 | |
316 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); | |
317 | if (!new_tbl) | |
318 | return 0; | |
319 | ||
da20420f HX |
320 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
321 | err = rhashtable_rehash_chain(ht, old_hash); | |
322 | if (err) | |
323 | return err; | |
ae6da1f5 | 324 | cond_resched(); |
da20420f | 325 | } |
aa34a6cb HX |
326 | |
327 | /* Publish the new table pointer. */ | |
328 | rcu_assign_pointer(ht->tbl, new_tbl); | |
329 | ||
ba7c95ea | 330 | spin_lock(&ht->lock); |
eddee5ba HX |
331 | list_for_each_entry(walker, &old_tbl->walkers, list) |
332 | walker->tbl = NULL; | |
ba7c95ea | 333 | spin_unlock(&ht->lock); |
eddee5ba | 334 | |
aa34a6cb HX |
335 | /* Wait for readers. All new readers will see the new |
336 | * table, and thus no references to the old table will | |
337 | * remain. | |
338 | */ | |
9d901bc0 | 339 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
b824478b HX |
340 | |
341 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; | |
7e1e7763 TG |
342 | } |
343 | ||
da20420f HX |
344 | static int rhashtable_rehash_alloc(struct rhashtable *ht, |
345 | struct bucket_table *old_tbl, | |
346 | unsigned int size) | |
7e1e7763 | 347 | { |
da20420f | 348 | struct bucket_table *new_tbl; |
b824478b | 349 | int err; |
7e1e7763 TG |
350 | |
351 | ASSERT_RHT_MUTEX(ht); | |
352 | ||
da20420f | 353 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
7e1e7763 TG |
354 | if (new_tbl == NULL) |
355 | return -ENOMEM; | |
356 | ||
b824478b HX |
357 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
358 | if (err) | |
359 | bucket_table_free(new_tbl); | |
360 | ||
361 | return err; | |
7e1e7763 | 362 | } |
7e1e7763 TG |
363 | |
364 | /** | |
365 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
366 | * @ht: the hash table to shrink | |
7e1e7763 | 367 | * |
18093d1c HX |
368 | * This function shrinks the hash table to fit, i.e., the smallest |
369 | * size would not cause it to expand right away automatically. | |
7e1e7763 | 370 | * |
97defe1e TG |
371 | * The caller must ensure that no concurrent resizing occurs by holding |
372 | * ht->mutex. | |
373 | * | |
7e1e7763 TG |
374 | * The caller must ensure that no concurrent table mutations take place. |
375 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
376 | * |
377 | * It is valid to have concurrent insertions and deletions protected by per | |
378 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 379 | */ |
b824478b | 380 | static int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 381 | { |
da20420f | 382 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
12311959 VN |
383 | unsigned int nelems = atomic_read(&ht->nelems); |
384 | unsigned int size = 0; | |
7e1e7763 | 385 | |
12311959 VN |
386 | if (nelems) |
387 | size = roundup_pow_of_two(nelems * 3 / 2); | |
18093d1c HX |
388 | if (size < ht->p.min_size) |
389 | size = ht->p.min_size; | |
390 | ||
391 | if (old_tbl->size <= size) | |
392 | return 0; | |
393 | ||
b824478b HX |
394 | if (rht_dereference(old_tbl->future_tbl, ht)) |
395 | return -EEXIST; | |
396 | ||
da20420f | 397 | return rhashtable_rehash_alloc(ht, old_tbl, size); |
7e1e7763 | 398 | } |
7e1e7763 | 399 | |
97defe1e TG |
400 | static void rht_deferred_worker(struct work_struct *work) |
401 | { | |
402 | struct rhashtable *ht; | |
403 | struct bucket_table *tbl; | |
b824478b | 404 | int err = 0; |
97defe1e | 405 | |
57699a40 | 406 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 407 | mutex_lock(&ht->mutex); |
28134a53 | 408 | |
97defe1e | 409 | tbl = rht_dereference(ht->tbl, ht); |
b824478b | 410 | tbl = rhashtable_last_table(ht, tbl); |
97defe1e | 411 | |
a5b6846f | 412 | if (rht_grow_above_75(ht, tbl)) |
da20420f | 413 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2); |
b5e2c150 | 414 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
da20420f HX |
415 | err = rhashtable_shrink(ht); |
416 | else if (tbl->nest) | |
417 | err = rhashtable_rehash_alloc(ht, tbl, tbl->size); | |
b824478b | 418 | |
da20420f HX |
419 | if (!err) |
420 | err = rhashtable_rehash_table(ht); | |
b824478b | 421 | |
97defe1e | 422 | mutex_unlock(&ht->mutex); |
b824478b HX |
423 | |
424 | if (err) | |
425 | schedule_work(&ht->run_work); | |
97defe1e TG |
426 | } |
427 | ||
ca26893f HX |
428 | static int rhashtable_insert_rehash(struct rhashtable *ht, |
429 | struct bucket_table *tbl) | |
ccd57b1b HX |
430 | { |
431 | struct bucket_table *old_tbl; | |
432 | struct bucket_table *new_tbl; | |
ccd57b1b HX |
433 | unsigned int size; |
434 | int err; | |
435 | ||
436 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
ccd57b1b HX |
437 | |
438 | size = tbl->size; | |
439 | ||
3cf92222 HX |
440 | err = -EBUSY; |
441 | ||
ccd57b1b HX |
442 | if (rht_grow_above_75(ht, tbl)) |
443 | size *= 2; | |
a87b9ebf TG |
444 | /* Do not schedule more than one rehash */ |
445 | else if (old_tbl != tbl) | |
3cf92222 HX |
446 | goto fail; |
447 | ||
448 | err = -ENOMEM; | |
ccd57b1b | 449 | |
93f976b5 | 450 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC | __GFP_NOWARN); |
3cf92222 HX |
451 | if (new_tbl == NULL) |
452 | goto fail; | |
ccd57b1b HX |
453 | |
454 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); | |
455 | if (err) { | |
456 | bucket_table_free(new_tbl); | |
457 | if (err == -EEXIST) | |
458 | err = 0; | |
459 | } else | |
460 | schedule_work(&ht->run_work); | |
461 | ||
462 | return err; | |
3cf92222 HX |
463 | |
464 | fail: | |
465 | /* Do not fail the insert if someone else did a rehash. */ | |
c0690016 | 466 | if (likely(rcu_access_pointer(tbl->future_tbl))) |
3cf92222 HX |
467 | return 0; |
468 | ||
469 | /* Schedule async rehash to retry allocation in process context. */ | |
470 | if (err == -ENOMEM) | |
471 | schedule_work(&ht->run_work); | |
472 | ||
473 | return err; | |
ccd57b1b | 474 | } |
ccd57b1b | 475 | |
ca26893f HX |
476 | static void *rhashtable_lookup_one(struct rhashtable *ht, |
477 | struct bucket_table *tbl, unsigned int hash, | |
478 | const void *key, struct rhash_head *obj) | |
02fd97c3 | 479 | { |
ca26893f HX |
480 | struct rhashtable_compare_arg arg = { |
481 | .ht = ht, | |
482 | .key = key, | |
483 | }; | |
484 | struct rhash_head __rcu **pprev; | |
02fd97c3 | 485 | struct rhash_head *head; |
ca26893f | 486 | int elasticity; |
02fd97c3 | 487 | |
5f8ddeab | 488 | elasticity = RHT_ELASTICITY; |
da20420f HX |
489 | pprev = rht_bucket_var(tbl, hash); |
490 | rht_for_each_continue(head, *pprev, tbl, hash) { | |
ca26893f HX |
491 | struct rhlist_head *list; |
492 | struct rhlist_head *plist; | |
493 | ||
494 | elasticity--; | |
495 | if (!key || | |
496 | (ht->p.obj_cmpfn ? | |
497 | ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) : | |
d3dcf8eb PB |
498 | rhashtable_compare(&arg, rht_obj(ht, head)))) { |
499 | pprev = &head->next; | |
ca26893f | 500 | continue; |
d3dcf8eb | 501 | } |
ca26893f HX |
502 | |
503 | if (!ht->rhlist) | |
504 | return rht_obj(ht, head); | |
505 | ||
506 | list = container_of(obj, struct rhlist_head, rhead); | |
507 | plist = container_of(head, struct rhlist_head, rhead); | |
508 | ||
509 | RCU_INIT_POINTER(list->next, plist); | |
510 | head = rht_dereference_bucket(head->next, tbl, hash); | |
511 | RCU_INIT_POINTER(list->rhead.next, head); | |
512 | rcu_assign_pointer(*pprev, obj); | |
513 | ||
514 | return NULL; | |
5ca8cc5b | 515 | } |
02fd97c3 | 516 | |
ca26893f HX |
517 | if (elasticity <= 0) |
518 | return ERR_PTR(-EAGAIN); | |
519 | ||
520 | return ERR_PTR(-ENOENT); | |
521 | } | |
522 | ||
523 | static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht, | |
524 | struct bucket_table *tbl, | |
525 | unsigned int hash, | |
526 | struct rhash_head *obj, | |
527 | void *data) | |
528 | { | |
da20420f | 529 | struct rhash_head __rcu **pprev; |
ca26893f HX |
530 | struct bucket_table *new_tbl; |
531 | struct rhash_head *head; | |
532 | ||
533 | if (!IS_ERR_OR_NULL(data)) | |
534 | return ERR_PTR(-EEXIST); | |
07ee0722 | 535 | |
ca26893f HX |
536 | if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT) |
537 | return ERR_CAST(data); | |
ccd57b1b | 538 | |
c0690016 | 539 | new_tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
ca26893f HX |
540 | if (new_tbl) |
541 | return new_tbl; | |
542 | ||
543 | if (PTR_ERR(data) != -ENOENT) | |
544 | return ERR_CAST(data); | |
545 | ||
546 | if (unlikely(rht_grow_above_max(ht, tbl))) | |
547 | return ERR_PTR(-E2BIG); | |
548 | ||
549 | if (unlikely(rht_grow_above_100(ht, tbl))) | |
550 | return ERR_PTR(-EAGAIN); | |
02fd97c3 | 551 | |
da20420f HX |
552 | pprev = rht_bucket_insert(ht, tbl, hash); |
553 | if (!pprev) | |
554 | return ERR_PTR(-ENOMEM); | |
555 | ||
556 | head = rht_dereference_bucket(*pprev, tbl, hash); | |
02fd97c3 HX |
557 | |
558 | RCU_INIT_POINTER(obj->next, head); | |
ca26893f HX |
559 | if (ht->rhlist) { |
560 | struct rhlist_head *list; | |
561 | ||
562 | list = container_of(obj, struct rhlist_head, rhead); | |
563 | RCU_INIT_POINTER(list->next, NULL); | |
564 | } | |
02fd97c3 | 565 | |
da20420f | 566 | rcu_assign_pointer(*pprev, obj); |
02fd97c3 HX |
567 | |
568 | atomic_inc(&ht->nelems); | |
ca26893f HX |
569 | if (rht_grow_above_75(ht, tbl)) |
570 | schedule_work(&ht->run_work); | |
02fd97c3 | 571 | |
ca26893f HX |
572 | return NULL; |
573 | } | |
02fd97c3 | 574 | |
ca26893f HX |
575 | static void *rhashtable_try_insert(struct rhashtable *ht, const void *key, |
576 | struct rhash_head *obj) | |
577 | { | |
578 | struct bucket_table *new_tbl; | |
579 | struct bucket_table *tbl; | |
580 | unsigned int hash; | |
581 | spinlock_t *lock; | |
582 | void *data; | |
583 | ||
584 | tbl = rcu_dereference(ht->tbl); | |
585 | ||
586 | /* All insertions must grab the oldest table containing | |
587 | * the hashed bucket that is yet to be rehashed. | |
588 | */ | |
589 | for (;;) { | |
590 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
591 | lock = rht_bucket_lock(tbl, hash); | |
592 | spin_lock_bh(lock); | |
593 | ||
594 | if (tbl->rehash <= hash) | |
595 | break; | |
596 | ||
597 | spin_unlock_bh(lock); | |
c0690016 | 598 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
ca26893f HX |
599 | } |
600 | ||
601 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
602 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
603 | if (PTR_ERR(new_tbl) != -EEXIST) | |
604 | data = ERR_CAST(new_tbl); | |
605 | ||
606 | while (!IS_ERR_OR_NULL(new_tbl)) { | |
607 | tbl = new_tbl; | |
608 | hash = rht_head_hashfn(ht, tbl, obj, ht->p); | |
609 | spin_lock_nested(rht_bucket_lock(tbl, hash), | |
610 | SINGLE_DEPTH_NESTING); | |
611 | ||
612 | data = rhashtable_lookup_one(ht, tbl, hash, key, obj); | |
613 | new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data); | |
614 | if (PTR_ERR(new_tbl) != -EEXIST) | |
615 | data = ERR_CAST(new_tbl); | |
616 | ||
617 | spin_unlock(rht_bucket_lock(tbl, hash)); | |
618 | } | |
619 | ||
620 | spin_unlock_bh(lock); | |
621 | ||
622 | if (PTR_ERR(data) == -EAGAIN) | |
623 | data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?: | |
624 | -EAGAIN); | |
625 | ||
626 | return data; | |
627 | } | |
628 | ||
629 | void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, | |
630 | struct rhash_head *obj) | |
631 | { | |
632 | void *data; | |
633 | ||
634 | do { | |
635 | rcu_read_lock(); | |
636 | data = rhashtable_try_insert(ht, key, obj); | |
637 | rcu_read_unlock(); | |
638 | } while (PTR_ERR(data) == -EAGAIN); | |
639 | ||
640 | return data; | |
02fd97c3 HX |
641 | } |
642 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); | |
643 | ||
f2dba9c6 | 644 | /** |
246779dd | 645 | * rhashtable_walk_enter - Initialise an iterator |
f2dba9c6 HX |
646 | * @ht: Table to walk over |
647 | * @iter: Hash table Iterator | |
648 | * | |
649 | * This function prepares a hash table walk. | |
650 | * | |
651 | * Note that if you restart a walk after rhashtable_walk_stop you | |
652 | * may see the same object twice. Also, you may miss objects if | |
653 | * there are removals in between rhashtable_walk_stop and the next | |
654 | * call to rhashtable_walk_start. | |
655 | * | |
656 | * For a completely stable walk you should construct your own data | |
657 | * structure outside the hash table. | |
658 | * | |
82266e98 N |
659 | * This function may be called from any process context, including |
660 | * non-preemptable context, but cannot be called from softirq or | |
661 | * hardirq context. | |
f2dba9c6 | 662 | * |
246779dd | 663 | * You must call rhashtable_walk_exit after this function returns. |
f2dba9c6 | 664 | */ |
246779dd | 665 | void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter) |
f2dba9c6 HX |
666 | { |
667 | iter->ht = ht; | |
668 | iter->p = NULL; | |
669 | iter->slot = 0; | |
670 | iter->skip = 0; | |
2db54b47 | 671 | iter->end_of_table = 0; |
f2dba9c6 | 672 | |
c6ff5268 | 673 | spin_lock(&ht->lock); |
246779dd | 674 | iter->walker.tbl = |
179ccc0a | 675 | rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock)); |
246779dd | 676 | list_add(&iter->walker.list, &iter->walker.tbl->walkers); |
c6ff5268 | 677 | spin_unlock(&ht->lock); |
f2dba9c6 | 678 | } |
246779dd | 679 | EXPORT_SYMBOL_GPL(rhashtable_walk_enter); |
f2dba9c6 HX |
680 | |
681 | /** | |
682 | * rhashtable_walk_exit - Free an iterator | |
683 | * @iter: Hash table Iterator | |
684 | * | |
685 | * This function frees resources allocated by rhashtable_walk_init. | |
686 | */ | |
687 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
688 | { | |
c6ff5268 | 689 | spin_lock(&iter->ht->lock); |
246779dd HX |
690 | if (iter->walker.tbl) |
691 | list_del(&iter->walker.list); | |
c6ff5268 | 692 | spin_unlock(&iter->ht->lock); |
f2dba9c6 HX |
693 | } |
694 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
695 | ||
696 | /** | |
97a6ec4a | 697 | * rhashtable_walk_start_check - Start a hash table walk |
f2dba9c6 HX |
698 | * @iter: Hash table iterator |
699 | * | |
0647169c AG |
700 | * Start a hash table walk at the current iterator position. Note that we take |
701 | * the RCU lock in all cases including when we return an error. So you must | |
702 | * always call rhashtable_walk_stop to clean up. | |
f2dba9c6 HX |
703 | * |
704 | * Returns zero if successful. | |
705 | * | |
706 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
707 | * will rewind back to the beginning and you may use it immediately | |
708 | * by calling rhashtable_walk_next. | |
97a6ec4a TH |
709 | * |
710 | * rhashtable_walk_start is defined as an inline variant that returns | |
711 | * void. This is preferred in cases where the caller would ignore | |
712 | * resize events and always continue. | |
f2dba9c6 | 713 | */ |
97a6ec4a | 714 | int rhashtable_walk_start_check(struct rhashtable_iter *iter) |
db4374f4 | 715 | __acquires(RCU) |
f2dba9c6 | 716 | { |
eddee5ba | 717 | struct rhashtable *ht = iter->ht; |
5d240a89 | 718 | bool rhlist = ht->rhlist; |
eddee5ba | 719 | |
c6ff5268 | 720 | rcu_read_lock(); |
eddee5ba | 721 | |
c6ff5268 | 722 | spin_lock(&ht->lock); |
246779dd HX |
723 | if (iter->walker.tbl) |
724 | list_del(&iter->walker.list); | |
c6ff5268 | 725 | spin_unlock(&ht->lock); |
eddee5ba | 726 | |
5d240a89 N |
727 | if (iter->end_of_table) |
728 | return 0; | |
729 | if (!iter->walker.tbl) { | |
246779dd | 730 | iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht); |
b41cc04b N |
731 | iter->slot = 0; |
732 | iter->skip = 0; | |
f2dba9c6 HX |
733 | return -EAGAIN; |
734 | } | |
735 | ||
5d240a89 N |
736 | if (iter->p && !rhlist) { |
737 | /* | |
738 | * We need to validate that 'p' is still in the table, and | |
739 | * if so, update 'skip' | |
740 | */ | |
741 | struct rhash_head *p; | |
742 | int skip = 0; | |
743 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
744 | skip++; | |
745 | if (p == iter->p) { | |
746 | iter->skip = skip; | |
747 | goto found; | |
748 | } | |
749 | } | |
750 | iter->p = NULL; | |
751 | } else if (iter->p && rhlist) { | |
752 | /* Need to validate that 'list' is still in the table, and | |
753 | * if so, update 'skip' and 'p'. | |
754 | */ | |
755 | struct rhash_head *p; | |
756 | struct rhlist_head *list; | |
757 | int skip = 0; | |
758 | rht_for_each_rcu(p, iter->walker.tbl, iter->slot) { | |
759 | for (list = container_of(p, struct rhlist_head, rhead); | |
760 | list; | |
761 | list = rcu_dereference(list->next)) { | |
762 | skip++; | |
763 | if (list == iter->list) { | |
764 | iter->p = p; | |
c643ecf3 | 765 | iter->skip = skip; |
5d240a89 N |
766 | goto found; |
767 | } | |
768 | } | |
769 | } | |
770 | iter->p = NULL; | |
771 | } | |
772 | found: | |
f2dba9c6 HX |
773 | return 0; |
774 | } | |
97a6ec4a | 775 | EXPORT_SYMBOL_GPL(rhashtable_walk_start_check); |
f2dba9c6 HX |
776 | |
777 | /** | |
2db54b47 TH |
778 | * __rhashtable_walk_find_next - Find the next element in a table (or the first |
779 | * one in case of a new walk). | |
780 | * | |
f2dba9c6 HX |
781 | * @iter: Hash table iterator |
782 | * | |
2db54b47 | 783 | * Returns the found object or NULL when the end of the table is reached. |
f2dba9c6 | 784 | * |
2db54b47 | 785 | * Returns -EAGAIN if resize event occurred. |
f2dba9c6 | 786 | */ |
2db54b47 | 787 | static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter) |
f2dba9c6 | 788 | { |
246779dd | 789 | struct bucket_table *tbl = iter->walker.tbl; |
ca26893f | 790 | struct rhlist_head *list = iter->list; |
f2dba9c6 HX |
791 | struct rhashtable *ht = iter->ht; |
792 | struct rhash_head *p = iter->p; | |
ca26893f | 793 | bool rhlist = ht->rhlist; |
f2dba9c6 | 794 | |
2db54b47 TH |
795 | if (!tbl) |
796 | return NULL; | |
f2dba9c6 HX |
797 | |
798 | for (; iter->slot < tbl->size; iter->slot++) { | |
799 | int skip = iter->skip; | |
800 | ||
801 | rht_for_each_rcu(p, tbl, iter->slot) { | |
ca26893f HX |
802 | if (rhlist) { |
803 | list = container_of(p, struct rhlist_head, | |
804 | rhead); | |
805 | do { | |
806 | if (!skip) | |
807 | goto next; | |
808 | skip--; | |
809 | list = rcu_dereference(list->next); | |
810 | } while (list); | |
811 | ||
812 | continue; | |
813 | } | |
f2dba9c6 HX |
814 | if (!skip) |
815 | break; | |
816 | skip--; | |
817 | } | |
818 | ||
819 | next: | |
820 | if (!rht_is_a_nulls(p)) { | |
821 | iter->skip++; | |
822 | iter->p = p; | |
ca26893f HX |
823 | iter->list = list; |
824 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
f2dba9c6 HX |
825 | } |
826 | ||
827 | iter->skip = 0; | |
828 | } | |
829 | ||
142b942a PS |
830 | iter->p = NULL; |
831 | ||
d88252f9 HX |
832 | /* Ensure we see any new tables. */ |
833 | smp_rmb(); | |
834 | ||
246779dd HX |
835 | iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
836 | if (iter->walker.tbl) { | |
f2dba9c6 HX |
837 | iter->slot = 0; |
838 | iter->skip = 0; | |
f2dba9c6 | 839 | return ERR_PTR(-EAGAIN); |
2db54b47 TH |
840 | } else { |
841 | iter->end_of_table = true; | |
f2dba9c6 HX |
842 | } |
843 | ||
c936a79f | 844 | return NULL; |
f2dba9c6 | 845 | } |
2db54b47 TH |
846 | |
847 | /** | |
848 | * rhashtable_walk_next - Return the next object and advance the iterator | |
849 | * @iter: Hash table iterator | |
850 | * | |
851 | * Note that you must call rhashtable_walk_stop when you are finished | |
852 | * with the walk. | |
853 | * | |
854 | * Returns the next object or NULL when the end of the table is reached. | |
855 | * | |
856 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
857 | * will rewind back to the beginning and you may continue to use it. | |
858 | */ | |
859 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
860 | { | |
861 | struct rhlist_head *list = iter->list; | |
862 | struct rhashtable *ht = iter->ht; | |
863 | struct rhash_head *p = iter->p; | |
864 | bool rhlist = ht->rhlist; | |
865 | ||
866 | if (p) { | |
867 | if (!rhlist || !(list = rcu_dereference(list->next))) { | |
868 | p = rcu_dereference(p->next); | |
869 | list = container_of(p, struct rhlist_head, rhead); | |
870 | } | |
871 | if (!rht_is_a_nulls(p)) { | |
872 | iter->skip++; | |
873 | iter->p = p; | |
874 | iter->list = list; | |
875 | return rht_obj(ht, rhlist ? &list->rhead : p); | |
876 | } | |
877 | ||
878 | /* At the end of this slot, switch to next one and then find | |
879 | * next entry from that point. | |
880 | */ | |
881 | iter->skip = 0; | |
882 | iter->slot++; | |
883 | } | |
884 | ||
885 | return __rhashtable_walk_find_next(iter); | |
886 | } | |
f2dba9c6 HX |
887 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); |
888 | ||
2db54b47 TH |
889 | /** |
890 | * rhashtable_walk_peek - Return the next object but don't advance the iterator | |
891 | * @iter: Hash table iterator | |
892 | * | |
893 | * Returns the next object or NULL when the end of the table is reached. | |
894 | * | |
895 | * Returns -EAGAIN if resize event occurred. Note that the iterator | |
896 | * will rewind back to the beginning and you may continue to use it. | |
897 | */ | |
898 | void *rhashtable_walk_peek(struct rhashtable_iter *iter) | |
899 | { | |
900 | struct rhlist_head *list = iter->list; | |
901 | struct rhashtable *ht = iter->ht; | |
902 | struct rhash_head *p = iter->p; | |
903 | ||
904 | if (p) | |
905 | return rht_obj(ht, ht->rhlist ? &list->rhead : p); | |
906 | ||
907 | /* No object found in current iter, find next one in the table. */ | |
908 | ||
909 | if (iter->skip) { | |
910 | /* A nonzero skip value points to the next entry in the table | |
911 | * beyond that last one that was found. Decrement skip so | |
912 | * we find the current value. __rhashtable_walk_find_next | |
913 | * will restore the original value of skip assuming that | |
914 | * the table hasn't changed. | |
915 | */ | |
916 | iter->skip--; | |
917 | } | |
918 | ||
919 | return __rhashtable_walk_find_next(iter); | |
920 | } | |
921 | EXPORT_SYMBOL_GPL(rhashtable_walk_peek); | |
922 | ||
f2dba9c6 HX |
923 | /** |
924 | * rhashtable_walk_stop - Finish a hash table walk | |
925 | * @iter: Hash table iterator | |
926 | * | |
0647169c AG |
927 | * Finish a hash table walk. Does not reset the iterator to the start of the |
928 | * hash table. | |
f2dba9c6 HX |
929 | */ |
930 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
db4374f4 | 931 | __releases(RCU) |
f2dba9c6 | 932 | { |
eddee5ba | 933 | struct rhashtable *ht; |
246779dd | 934 | struct bucket_table *tbl = iter->walker.tbl; |
eddee5ba | 935 | |
eddee5ba | 936 | if (!tbl) |
963ecbd4 | 937 | goto out; |
eddee5ba HX |
938 | |
939 | ht = iter->ht; | |
940 | ||
ba7c95ea | 941 | spin_lock(&ht->lock); |
c4db8848 | 942 | if (tbl->rehash < tbl->size) |
246779dd | 943 | list_add(&iter->walker.list, &tbl->walkers); |
eddee5ba | 944 | else |
246779dd | 945 | iter->walker.tbl = NULL; |
ba7c95ea | 946 | spin_unlock(&ht->lock); |
eddee5ba | 947 | |
963ecbd4 HX |
948 | out: |
949 | rcu_read_unlock(); | |
f2dba9c6 HX |
950 | } |
951 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
952 | ||
488fb86e | 953 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
7e1e7763 | 954 | { |
107d01f5 DB |
955 | size_t retsize; |
956 | ||
957 | if (params->nelem_hint) | |
958 | retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3), | |
959 | (unsigned long)params->min_size); | |
960 | else | |
961 | retsize = max(HASH_DEFAULT_SIZE, | |
962 | (unsigned long)params->min_size); | |
963 | ||
964 | return retsize; | |
7e1e7763 TG |
965 | } |
966 | ||
31ccde2d HX |
967 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
968 | { | |
969 | return jhash2(key, length, seed); | |
970 | } | |
971 | ||
7e1e7763 TG |
972 | /** |
973 | * rhashtable_init - initialize a new hash table | |
974 | * @ht: hash table to be initialized | |
975 | * @params: configuration parameters | |
976 | * | |
977 | * Initializes a new hash table based on the provided configuration | |
978 | * parameters. A table can be configured either with a variable or | |
979 | * fixed length key: | |
980 | * | |
981 | * Configuration Example 1: Fixed length keys | |
982 | * struct test_obj { | |
983 | * int key; | |
984 | * void * my_member; | |
985 | * struct rhash_head node; | |
986 | * }; | |
987 | * | |
988 | * struct rhashtable_params params = { | |
989 | * .head_offset = offsetof(struct test_obj, node), | |
990 | * .key_offset = offsetof(struct test_obj, key), | |
991 | * .key_len = sizeof(int), | |
87545899 | 992 | * .hashfn = jhash, |
7e1e7763 TG |
993 | * }; |
994 | * | |
995 | * Configuration Example 2: Variable length keys | |
996 | * struct test_obj { | |
997 | * [...] | |
998 | * struct rhash_head node; | |
999 | * }; | |
1000 | * | |
49f7b33e | 1001 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
7e1e7763 TG |
1002 | * { |
1003 | * struct test_obj *obj = data; | |
1004 | * | |
1005 | * return [... hash ...]; | |
1006 | * } | |
1007 | * | |
1008 | * struct rhashtable_params params = { | |
1009 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 1010 | * .hashfn = jhash, |
7e1e7763 | 1011 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
1012 | * }; |
1013 | */ | |
488fb86e HX |
1014 | int rhashtable_init(struct rhashtable *ht, |
1015 | const struct rhashtable_params *params) | |
7e1e7763 TG |
1016 | { |
1017 | struct bucket_table *tbl; | |
1018 | size_t size; | |
1019 | ||
31ccde2d | 1020 | if ((!params->key_len && !params->obj_hashfn) || |
02fd97c3 | 1021 | (params->obj_hashfn && !params->obj_cmpfn)) |
7e1e7763 TG |
1022 | return -EINVAL; |
1023 | ||
97defe1e TG |
1024 | memset(ht, 0, sizeof(*ht)); |
1025 | mutex_init(&ht->mutex); | |
ba7c95ea | 1026 | spin_lock_init(&ht->lock); |
97defe1e TG |
1027 | memcpy(&ht->p, params, sizeof(*params)); |
1028 | ||
a998f712 TG |
1029 | if (params->min_size) |
1030 | ht->p.min_size = roundup_pow_of_two(params->min_size); | |
1031 | ||
6d684e54 HX |
1032 | /* Cap total entries at 2^31 to avoid nelems overflow. */ |
1033 | ht->max_elems = 1u << 31; | |
2d2ab658 HX |
1034 | |
1035 | if (params->max_size) { | |
1036 | ht->p.max_size = rounddown_pow_of_two(params->max_size); | |
1037 | if (ht->p.max_size < ht->max_elems / 2) | |
1038 | ht->max_elems = ht->p.max_size * 2; | |
1039 | } | |
6d684e54 | 1040 | |
48e75b43 | 1041 | ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); |
a998f712 | 1042 | |
107d01f5 | 1043 | size = rounded_hashtable_size(&ht->p); |
3a324606 | 1044 | |
97defe1e TG |
1045 | if (params->locks_mul) |
1046 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
1047 | else | |
1048 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
1049 | ||
31ccde2d HX |
1050 | ht->key_len = ht->p.key_len; |
1051 | if (!params->hashfn) { | |
1052 | ht->p.hashfn = jhash; | |
1053 | ||
1054 | if (!(ht->key_len & (sizeof(u32) - 1))) { | |
1055 | ht->key_len /= sizeof(u32); | |
1056 | ht->p.hashfn = rhashtable_jhash2; | |
1057 | } | |
1058 | } | |
1059 | ||
2d22ecf6 DB |
1060 | /* |
1061 | * This is api initialization and thus we need to guarantee the | |
1062 | * initial rhashtable allocation. Upon failure, retry with the | |
1063 | * smallest possible size with __GFP_NOFAIL semantics. | |
1064 | */ | |
b9ecfdaa | 1065 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
2d22ecf6 DB |
1066 | if (unlikely(tbl == NULL)) { |
1067 | size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE); | |
1068 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL); | |
1069 | } | |
7e1e7763 | 1070 | |
545a148e | 1071 | atomic_set(&ht->nelems, 0); |
a5b6846f | 1072 | |
7e1e7763 TG |
1073 | RCU_INIT_POINTER(ht->tbl, tbl); |
1074 | ||
4c4b52d9 | 1075 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1076 | |
7e1e7763 TG |
1077 | return 0; |
1078 | } | |
1079 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1080 | ||
ca26893f HX |
1081 | /** |
1082 | * rhltable_init - initialize a new hash list table | |
1083 | * @hlt: hash list table to be initialized | |
1084 | * @params: configuration parameters | |
1085 | * | |
1086 | * Initializes a new hash list table. | |
1087 | * | |
1088 | * See documentation for rhashtable_init. | |
1089 | */ | |
1090 | int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params) | |
1091 | { | |
1092 | int err; | |
1093 | ||
ca26893f HX |
1094 | err = rhashtable_init(&hlt->ht, params); |
1095 | hlt->ht.rhlist = true; | |
1096 | return err; | |
1097 | } | |
1098 | EXPORT_SYMBOL_GPL(rhltable_init); | |
1099 | ||
1100 | static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj, | |
1101 | void (*free_fn)(void *ptr, void *arg), | |
1102 | void *arg) | |
1103 | { | |
1104 | struct rhlist_head *list; | |
1105 | ||
1106 | if (!ht->rhlist) { | |
1107 | free_fn(rht_obj(ht, obj), arg); | |
1108 | return; | |
1109 | } | |
1110 | ||
1111 | list = container_of(obj, struct rhlist_head, rhead); | |
1112 | do { | |
1113 | obj = &list->rhead; | |
1114 | list = rht_dereference(list->next, ht); | |
1115 | free_fn(rht_obj(ht, obj), arg); | |
1116 | } while (list); | |
1117 | } | |
1118 | ||
7e1e7763 | 1119 | /** |
6b6f302c | 1120 | * rhashtable_free_and_destroy - free elements and destroy hash table |
7e1e7763 | 1121 | * @ht: the hash table to destroy |
6b6f302c TG |
1122 | * @free_fn: callback to release resources of element |
1123 | * @arg: pointer passed to free_fn | |
7e1e7763 | 1124 | * |
6b6f302c TG |
1125 | * Stops an eventual async resize. If defined, invokes free_fn for each |
1126 | * element to releasal resources. Please note that RCU protected | |
1127 | * readers may still be accessing the elements. Releasing of resources | |
1128 | * must occur in a compatible manner. Then frees the bucket array. | |
1129 | * | |
1130 | * This function will eventually sleep to wait for an async resize | |
1131 | * to complete. The caller is responsible that no further write operations | |
1132 | * occurs in parallel. | |
7e1e7763 | 1133 | */ |
6b6f302c TG |
1134 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
1135 | void (*free_fn)(void *ptr, void *arg), | |
1136 | void *arg) | |
7e1e7763 | 1137 | { |
0026129c | 1138 | struct bucket_table *tbl, *next_tbl; |
6b6f302c | 1139 | unsigned int i; |
97defe1e | 1140 | |
4c4b52d9 | 1141 | cancel_work_sync(&ht->run_work); |
97defe1e | 1142 | |
57699a40 | 1143 | mutex_lock(&ht->mutex); |
6b6f302c | 1144 | tbl = rht_dereference(ht->tbl, ht); |
0026129c | 1145 | restart: |
6b6f302c TG |
1146 | if (free_fn) { |
1147 | for (i = 0; i < tbl->size; i++) { | |
1148 | struct rhash_head *pos, *next; | |
1149 | ||
ae6da1f5 | 1150 | cond_resched(); |
da20420f | 1151 | for (pos = rht_dereference(*rht_bucket(tbl, i), ht), |
6b6f302c TG |
1152 | next = !rht_is_a_nulls(pos) ? |
1153 | rht_dereference(pos->next, ht) : NULL; | |
1154 | !rht_is_a_nulls(pos); | |
1155 | pos = next, | |
1156 | next = !rht_is_a_nulls(pos) ? | |
1157 | rht_dereference(pos->next, ht) : NULL) | |
ca26893f | 1158 | rhashtable_free_one(ht, pos, free_fn, arg); |
6b6f302c TG |
1159 | } |
1160 | } | |
1161 | ||
0026129c | 1162 | next_tbl = rht_dereference(tbl->future_tbl, ht); |
6b6f302c | 1163 | bucket_table_free(tbl); |
0026129c TY |
1164 | if (next_tbl) { |
1165 | tbl = next_tbl; | |
1166 | goto restart; | |
1167 | } | |
97defe1e | 1168 | mutex_unlock(&ht->mutex); |
7e1e7763 | 1169 | } |
6b6f302c TG |
1170 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
1171 | ||
1172 | void rhashtable_destroy(struct rhashtable *ht) | |
1173 | { | |
1174 | return rhashtable_free_and_destroy(ht, NULL, NULL); | |
1175 | } | |
7e1e7763 | 1176 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |
da20420f HX |
1177 | |
1178 | struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl, | |
1179 | unsigned int hash) | |
1180 | { | |
1181 | const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *)); | |
82208d0d | 1182 | static struct rhash_head __rcu *rhnull; |
da20420f HX |
1183 | unsigned int index = hash & ((1 << tbl->nest) - 1); |
1184 | unsigned int size = tbl->size >> tbl->nest; | |
1185 | unsigned int subhash = hash; | |
1186 | union nested_table *ntbl; | |
1187 | ||
1188 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
c4d2603d | 1189 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash); |
da20420f HX |
1190 | subhash >>= tbl->nest; |
1191 | ||
1192 | while (ntbl && size > (1 << shift)) { | |
1193 | index = subhash & ((1 << shift) - 1); | |
c4d2603d HX |
1194 | ntbl = rht_dereference_bucket_rcu(ntbl[index].table, |
1195 | tbl, hash); | |
da20420f HX |
1196 | size >>= shift; |
1197 | subhash >>= shift; | |
1198 | } | |
1199 | ||
82208d0d N |
1200 | if (!ntbl) { |
1201 | if (!rhnull) | |
1202 | INIT_RHT_NULLS_HEAD(rhnull); | |
da20420f | 1203 | return &rhnull; |
82208d0d | 1204 | } |
da20420f HX |
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; | |
da20420f HX |
1219 | |
1220 | ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]); | |
1221 | hash >>= tbl->nest; | |
da20420f | 1222 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1223 | size <= (1 << shift)); |
da20420f HX |
1224 | |
1225 | while (ntbl && size > (1 << shift)) { | |
1226 | index = hash & ((1 << shift) - 1); | |
1227 | size >>= shift; | |
1228 | hash >>= shift; | |
da20420f | 1229 | ntbl = nested_table_alloc(ht, &ntbl[index].table, |
5af68ef7 | 1230 | size <= (1 << shift)); |
da20420f HX |
1231 | } |
1232 | ||
1233 | if (!ntbl) | |
1234 | return NULL; | |
1235 | ||
1236 | return &ntbl[hash].bucket; | |
1237 | ||
1238 | } | |
1239 | EXPORT_SYMBOL_GPL(rht_bucket_nested_insert); |