]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - lib/rhashtable.c
mm/slub: calculate start order with reserved in consideration
[mirror_ubuntu-zesty-kernel.git] / lib / rhashtable.c
CommitLineData
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>
7e1e7763
TG
22#include <linux/slab.h>
23#include <linux/vmalloc.h>
24#include <linux/mm.h>
87545899 25#include <linux/jhash.h>
7e1e7763
TG
26#include <linux/random.h>
27#include <linux/rhashtable.h>
61d7b097 28#include <linux/err.h>
6d795413 29#include <linux/export.h>
7e1e7763
TG
30
31#define HASH_DEFAULT_SIZE 64UL
c2e213cf 32#define HASH_MIN_SIZE 4U
97defe1e
TG
33#define BUCKET_LOCKS_PER_CPU 128UL
34
988dfbd7 35static u32 head_hashfn(struct rhashtable *ht,
8d24c0b4
TG
36 const struct bucket_table *tbl,
37 const struct rhash_head *he)
7e1e7763 38{
02fd97c3 39 return rht_head_hashfn(ht, tbl, he, ht->p);
7e1e7763
TG
40}
41
a03eaec0 42#ifdef CONFIG_PROVE_LOCKING
a03eaec0 43#define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
a03eaec0
TG
44
45int lockdep_rht_mutex_is_held(struct rhashtable *ht)
46{
47 return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
48}
49EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
50
51int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
52{
02fd97c3 53 spinlock_t *lock = rht_bucket_lock(tbl, hash);
a03eaec0
TG
54
55 return (debug_locks) ? lockdep_is_held(lock) : 1;
56}
57EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
58#else
59#define ASSERT_RHT_MUTEX(HT)
a03eaec0
TG
60#endif
61
62
b9ecfdaa
HX
63static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl,
64 gfp_t gfp)
97defe1e
TG
65{
66 unsigned int i, size;
67#if defined(CONFIG_PROVE_LOCKING)
68 unsigned int nr_pcpus = 2;
69#else
70 unsigned int nr_pcpus = num_possible_cpus();
71#endif
72
73 nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL);
74 size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
75
a5ec68e3
TG
76 /* Never allocate more than 0.5 locks per bucket */
77 size = min_t(unsigned int, size, tbl->size >> 1);
97defe1e
TG
78
79 if (sizeof(spinlock_t) != 0) {
80#ifdef CONFIG_NUMA
b9ecfdaa
HX
81 if (size * sizeof(spinlock_t) > PAGE_SIZE &&
82 gfp == GFP_KERNEL)
97defe1e
TG
83 tbl->locks = vmalloc(size * sizeof(spinlock_t));
84 else
85#endif
86 tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
b9ecfdaa 87 gfp);
97defe1e
TG
88 if (!tbl->locks)
89 return -ENOMEM;
90 for (i = 0; i < size; i++)
91 spin_lock_init(&tbl->locks[i]);
92 }
93 tbl->locks_mask = size - 1;
94
95 return 0;
96}
97
98static void bucket_table_free(const struct bucket_table *tbl)
99{
100 if (tbl)
101 kvfree(tbl->locks);
102
103 kvfree(tbl);
104}
105
9d901bc0
HX
106static void bucket_table_free_rcu(struct rcu_head *head)
107{
108 bucket_table_free(container_of(head, struct bucket_table, rcu));
109}
110
97defe1e 111static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
b9ecfdaa
HX
112 size_t nbuckets,
113 gfp_t gfp)
7e1e7763 114{
eb6d1abf 115 struct bucket_table *tbl = NULL;
7e1e7763 116 size_t size;
f89bd6f8 117 int i;
7e1e7763
TG
118
119 size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
b9ecfdaa
HX
120 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) ||
121 gfp != GFP_KERNEL)
122 tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
123 if (tbl == NULL && gfp == GFP_KERNEL)
7e1e7763 124 tbl = vzalloc(size);
7e1e7763
TG
125 if (tbl == NULL)
126 return NULL;
127
128 tbl->size = nbuckets;
129
b9ecfdaa 130 if (alloc_bucket_locks(ht, tbl, gfp) < 0) {
97defe1e
TG
131 bucket_table_free(tbl);
132 return NULL;
133 }
7e1e7763 134
eddee5ba
HX
135 INIT_LIST_HEAD(&tbl->walkers);
136
5269b53d
HX
137 get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
138
f89bd6f8
TG
139 for (i = 0; i < nbuckets; i++)
140 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
141
97defe1e 142 return tbl;
7e1e7763
TG
143}
144
b824478b
HX
145static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
146 struct bucket_table *tbl)
147{
148 struct bucket_table *new_tbl;
149
150 do {
151 new_tbl = tbl;
152 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
153 } while (tbl);
154
155 return new_tbl;
156}
157
299e5c32 158static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
a5ec68e3 159{
aa34a6cb 160 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
b824478b
HX
161 struct bucket_table *new_tbl = rhashtable_last_table(ht,
162 rht_dereference_rcu(old_tbl->future_tbl, ht));
aa34a6cb
HX
163 struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash];
164 int err = -ENOENT;
165 struct rhash_head *head, *next, *entry;
166 spinlock_t *new_bucket_lock;
299e5c32 167 unsigned int new_hash;
aa34a6cb
HX
168
169 rht_for_each(entry, old_tbl, old_hash) {
170 err = 0;
171 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
172
173 if (rht_is_a_nulls(next))
174 break;
a5ec68e3 175
aa34a6cb
HX
176 pprev = &entry->next;
177 }
a5ec68e3 178
aa34a6cb
HX
179 if (err)
180 goto out;
97defe1e 181
aa34a6cb 182 new_hash = head_hashfn(ht, new_tbl, entry);
7e1e7763 183
02fd97c3 184 new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
7e1e7763 185
8f2484bd 186 spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
aa34a6cb
HX
187 head = rht_dereference_bucket(new_tbl->buckets[new_hash],
188 new_tbl, new_hash);
97defe1e 189
7def0f95 190 RCU_INIT_POINTER(entry->next, head);
a5ec68e3 191
aa34a6cb
HX
192 rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
193 spin_unlock(new_bucket_lock);
97defe1e 194
aa34a6cb 195 rcu_assign_pointer(*pprev, next);
7e1e7763 196
aa34a6cb
HX
197out:
198 return err;
199}
97defe1e 200
299e5c32
TG
201static void rhashtable_rehash_chain(struct rhashtable *ht,
202 unsigned int old_hash)
aa34a6cb
HX
203{
204 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
205 spinlock_t *old_bucket_lock;
206
02fd97c3 207 old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
a5ec68e3 208
aa34a6cb
HX
209 spin_lock_bh(old_bucket_lock);
210 while (!rhashtable_rehash_one(ht, old_hash))
211 ;
63d512d0 212 old_tbl->rehash++;
aa34a6cb 213 spin_unlock_bh(old_bucket_lock);
97defe1e
TG
214}
215
b824478b
HX
216static int rhashtable_rehash_attach(struct rhashtable *ht,
217 struct bucket_table *old_tbl,
218 struct bucket_table *new_tbl)
97defe1e 219{
b824478b
HX
220 /* Protect future_tbl using the first bucket lock. */
221 spin_lock_bh(old_tbl->locks);
222
223 /* Did somebody beat us to it? */
224 if (rcu_access_pointer(old_tbl->future_tbl)) {
225 spin_unlock_bh(old_tbl->locks);
226 return -EEXIST;
227 }
7cd10db8 228
aa34a6cb
HX
229 /* Make insertions go into the new, empty table right away. Deletions
230 * and lookups will be attempted in both tables until we synchronize.
aa34a6cb 231 */
c4db8848 232 rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
aa34a6cb 233
9497df88
HX
234 /* Ensure the new table is visible to readers. */
235 smp_wmb();
236
b824478b
HX
237 spin_unlock_bh(old_tbl->locks);
238
239 return 0;
240}
241
242static int rhashtable_rehash_table(struct rhashtable *ht)
243{
244 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
245 struct bucket_table *new_tbl;
246 struct rhashtable_walker *walker;
299e5c32 247 unsigned int old_hash;
b824478b
HX
248
249 new_tbl = rht_dereference(old_tbl->future_tbl, ht);
250 if (!new_tbl)
251 return 0;
252
aa34a6cb
HX
253 for (old_hash = 0; old_hash < old_tbl->size; old_hash++)
254 rhashtable_rehash_chain(ht, old_hash);
255
256 /* Publish the new table pointer. */
257 rcu_assign_pointer(ht->tbl, new_tbl);
258
ba7c95ea 259 spin_lock(&ht->lock);
eddee5ba
HX
260 list_for_each_entry(walker, &old_tbl->walkers, list)
261 walker->tbl = NULL;
ba7c95ea 262 spin_unlock(&ht->lock);
eddee5ba 263
aa34a6cb
HX
264 /* Wait for readers. All new readers will see the new
265 * table, and thus no references to the old table will
266 * remain.
267 */
9d901bc0 268 call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
b824478b
HX
269
270 return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
7e1e7763
TG
271}
272
273/**
274 * rhashtable_expand - Expand hash table while allowing concurrent lookups
275 * @ht: the hash table to expand
7e1e7763 276 *
aa34a6cb 277 * A secondary bucket array is allocated and the hash entries are migrated.
7e1e7763
TG
278 *
279 * This function may only be called in a context where it is safe to call
280 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
281 *
97defe1e
TG
282 * The caller must ensure that no concurrent resizing occurs by holding
283 * ht->mutex.
284 *
285 * It is valid to have concurrent insertions and deletions protected by per
286 * bucket locks or concurrent RCU protected lookups and traversals.
7e1e7763 287 */
b824478b 288static int rhashtable_expand(struct rhashtable *ht)
7e1e7763
TG
289{
290 struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
b824478b 291 int err;
7e1e7763
TG
292
293 ASSERT_RHT_MUTEX(ht);
294
b824478b
HX
295 old_tbl = rhashtable_last_table(ht, old_tbl);
296
b9ecfdaa 297 new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL);
7e1e7763
TG
298 if (new_tbl == NULL)
299 return -ENOMEM;
300
b824478b
HX
301 err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
302 if (err)
303 bucket_table_free(new_tbl);
304
305 return err;
7e1e7763 306}
7e1e7763
TG
307
308/**
309 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
310 * @ht: the hash table to shrink
7e1e7763 311 *
18093d1c
HX
312 * This function shrinks the hash table to fit, i.e., the smallest
313 * size would not cause it to expand right away automatically.
7e1e7763 314 *
97defe1e
TG
315 * The caller must ensure that no concurrent resizing occurs by holding
316 * ht->mutex.
317 *
7e1e7763
TG
318 * The caller must ensure that no concurrent table mutations take place.
319 * It is however valid to have concurrent lookups if they are RCU protected.
97defe1e
TG
320 *
321 * It is valid to have concurrent insertions and deletions protected by per
322 * bucket locks or concurrent RCU protected lookups and traversals.
7e1e7763 323 */
b824478b 324static int rhashtable_shrink(struct rhashtable *ht)
7e1e7763 325{
a5b6846f 326 struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
299e5c32 327 unsigned int size;
b824478b 328 int err;
7e1e7763
TG
329
330 ASSERT_RHT_MUTEX(ht);
331
299e5c32 332 size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2);
18093d1c
HX
333 if (size < ht->p.min_size)
334 size = ht->p.min_size;
335
336 if (old_tbl->size <= size)
337 return 0;
338
b824478b
HX
339 if (rht_dereference(old_tbl->future_tbl, ht))
340 return -EEXIST;
341
b9ecfdaa 342 new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
97defe1e 343 if (new_tbl == NULL)
7e1e7763
TG
344 return -ENOMEM;
345
b824478b
HX
346 err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
347 if (err)
348 bucket_table_free(new_tbl);
349
350 return err;
7e1e7763 351}
7e1e7763 352
97defe1e
TG
353static void rht_deferred_worker(struct work_struct *work)
354{
355 struct rhashtable *ht;
356 struct bucket_table *tbl;
b824478b 357 int err = 0;
97defe1e 358
57699a40 359 ht = container_of(work, struct rhashtable, run_work);
97defe1e 360 mutex_lock(&ht->mutex);
28134a53 361
97defe1e 362 tbl = rht_dereference(ht->tbl, ht);
b824478b 363 tbl = rhashtable_last_table(ht, tbl);
97defe1e 364
a5b6846f 365 if (rht_grow_above_75(ht, tbl))
97defe1e 366 rhashtable_expand(ht);
b5e2c150 367 else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
97defe1e 368 rhashtable_shrink(ht);
b824478b
HX
369
370 err = rhashtable_rehash_table(ht);
371
97defe1e 372 mutex_unlock(&ht->mutex);
b824478b
HX
373
374 if (err)
375 schedule_work(&ht->run_work);
97defe1e
TG
376}
377
ccd57b1b
HX
378static bool rhashtable_check_elasticity(struct rhashtable *ht,
379 struct bucket_table *tbl,
299e5c32 380 unsigned int hash)
ccd57b1b 381{
299e5c32 382 unsigned int elasticity = ht->elasticity;
ccd57b1b
HX
383 struct rhash_head *head;
384
385 rht_for_each(head, tbl, hash)
386 if (!--elasticity)
387 return true;
388
389 return false;
390}
391
392int rhashtable_insert_rehash(struct rhashtable *ht)
393{
394 struct bucket_table *old_tbl;
395 struct bucket_table *new_tbl;
396 struct bucket_table *tbl;
397 unsigned int size;
398 int err;
399
400 old_tbl = rht_dereference_rcu(ht->tbl, ht);
401 tbl = rhashtable_last_table(ht, old_tbl);
402
403 size = tbl->size;
404
405 if (rht_grow_above_75(ht, tbl))
406 size *= 2;
a87b9ebf
TG
407 /* Do not schedule more than one rehash */
408 else if (old_tbl != tbl)
ccd57b1b
HX
409 return -EBUSY;
410
411 new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
e2307ed6
TG
412 if (new_tbl == NULL) {
413 /* Schedule async resize/rehash to try allocation
414 * non-atomic context.
415 */
416 schedule_work(&ht->run_work);
ccd57b1b 417 return -ENOMEM;
e2307ed6 418 }
ccd57b1b
HX
419
420 err = rhashtable_rehash_attach(ht, tbl, new_tbl);
421 if (err) {
422 bucket_table_free(new_tbl);
423 if (err == -EEXIST)
424 err = 0;
425 } else
426 schedule_work(&ht->run_work);
427
428 return err;
429}
430EXPORT_SYMBOL_GPL(rhashtable_insert_rehash);
431
02fd97c3
HX
432int rhashtable_insert_slow(struct rhashtable *ht, const void *key,
433 struct rhash_head *obj,
434 struct bucket_table *tbl)
435{
436 struct rhash_head *head;
299e5c32 437 unsigned int hash;
ccd57b1b 438 int err;
02fd97c3 439
b824478b 440 tbl = rhashtable_last_table(ht, tbl);
02fd97c3
HX
441 hash = head_hashfn(ht, tbl, obj);
442 spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING);
443
ccd57b1b 444 err = -EEXIST;
02fd97c3
HX
445 if (key && rhashtable_lookup_fast(ht, key, ht->p))
446 goto exit;
447
07ee0722
HX
448 err = -E2BIG;
449 if (unlikely(rht_grow_above_max(ht, tbl)))
450 goto exit;
451
ccd57b1b
HX
452 err = -EAGAIN;
453 if (rhashtable_check_elasticity(ht, tbl, hash) ||
454 rht_grow_above_100(ht, tbl))
455 goto exit;
456
02fd97c3
HX
457 err = 0;
458
459 head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
460
461 RCU_INIT_POINTER(obj->next, head);
462
463 rcu_assign_pointer(tbl->buckets[hash], obj);
464
465 atomic_inc(&ht->nelems);
466
467exit:
468 spin_unlock(rht_bucket_lock(tbl, hash));
469
470 return err;
471}
472EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
473
f2dba9c6
HX
474/**
475 * rhashtable_walk_init - Initialise an iterator
476 * @ht: Table to walk over
477 * @iter: Hash table Iterator
478 *
479 * This function prepares a hash table walk.
480 *
481 * Note that if you restart a walk after rhashtable_walk_stop you
482 * may see the same object twice. Also, you may miss objects if
483 * there are removals in between rhashtable_walk_stop and the next
484 * call to rhashtable_walk_start.
485 *
486 * For a completely stable walk you should construct your own data
487 * structure outside the hash table.
488 *
489 * This function may sleep so you must not call it from interrupt
490 * context or with spin locks held.
491 *
492 * You must call rhashtable_walk_exit if this function returns
493 * successfully.
494 */
495int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter)
496{
497 iter->ht = ht;
498 iter->p = NULL;
499 iter->slot = 0;
500 iter->skip = 0;
501
502 iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL);
503 if (!iter->walker)
504 return -ENOMEM;
505
506 mutex_lock(&ht->mutex);
eddee5ba
HX
507 iter->walker->tbl = rht_dereference(ht->tbl, ht);
508 list_add(&iter->walker->list, &iter->walker->tbl->walkers);
f2dba9c6
HX
509 mutex_unlock(&ht->mutex);
510
511 return 0;
512}
513EXPORT_SYMBOL_GPL(rhashtable_walk_init);
514
515/**
516 * rhashtable_walk_exit - Free an iterator
517 * @iter: Hash table Iterator
518 *
519 * This function frees resources allocated by rhashtable_walk_init.
520 */
521void rhashtable_walk_exit(struct rhashtable_iter *iter)
522{
523 mutex_lock(&iter->ht->mutex);
eddee5ba
HX
524 if (iter->walker->tbl)
525 list_del(&iter->walker->list);
f2dba9c6
HX
526 mutex_unlock(&iter->ht->mutex);
527 kfree(iter->walker);
528}
529EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
530
531/**
532 * rhashtable_walk_start - Start a hash table walk
533 * @iter: Hash table iterator
534 *
535 * Start a hash table walk. Note that we take the RCU lock in all
536 * cases including when we return an error. So you must always call
537 * rhashtable_walk_stop to clean up.
538 *
539 * Returns zero if successful.
540 *
541 * Returns -EAGAIN if resize event occured. Note that the iterator
542 * will rewind back to the beginning and you may use it immediately
543 * by calling rhashtable_walk_next.
544 */
545int rhashtable_walk_start(struct rhashtable_iter *iter)
db4374f4 546 __acquires(RCU)
f2dba9c6 547{
eddee5ba
HX
548 struct rhashtable *ht = iter->ht;
549
550 mutex_lock(&ht->mutex);
551
552 if (iter->walker->tbl)
553 list_del(&iter->walker->list);
554
f2dba9c6
HX
555 rcu_read_lock();
556
eddee5ba
HX
557 mutex_unlock(&ht->mutex);
558
559 if (!iter->walker->tbl) {
560 iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht);
f2dba9c6
HX
561 return -EAGAIN;
562 }
563
564 return 0;
565}
566EXPORT_SYMBOL_GPL(rhashtable_walk_start);
567
568/**
569 * rhashtable_walk_next - Return the next object and advance the iterator
570 * @iter: Hash table iterator
571 *
572 * Note that you must call rhashtable_walk_stop when you are finished
573 * with the walk.
574 *
575 * Returns the next object or NULL when the end of the table is reached.
576 *
577 * Returns -EAGAIN if resize event occured. Note that the iterator
578 * will rewind back to the beginning and you may continue to use it.
579 */
580void *rhashtable_walk_next(struct rhashtable_iter *iter)
581{
eddee5ba 582 struct bucket_table *tbl = iter->walker->tbl;
f2dba9c6
HX
583 struct rhashtable *ht = iter->ht;
584 struct rhash_head *p = iter->p;
f2dba9c6 585
f2dba9c6
HX
586 if (p) {
587 p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot);
588 goto next;
589 }
590
591 for (; iter->slot < tbl->size; iter->slot++) {
592 int skip = iter->skip;
593
594 rht_for_each_rcu(p, tbl, iter->slot) {
595 if (!skip)
596 break;
597 skip--;
598 }
599
600next:
601 if (!rht_is_a_nulls(p)) {
602 iter->skip++;
603 iter->p = p;
c936a79f 604 return rht_obj(ht, p);
f2dba9c6
HX
605 }
606
607 iter->skip = 0;
608 }
609
142b942a
PS
610 iter->p = NULL;
611
d88252f9
HX
612 /* Ensure we see any new tables. */
613 smp_rmb();
614
c4db8848
HX
615 iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht);
616 if (iter->walker->tbl) {
f2dba9c6
HX
617 iter->slot = 0;
618 iter->skip = 0;
f2dba9c6
HX
619 return ERR_PTR(-EAGAIN);
620 }
621
c936a79f 622 return NULL;
f2dba9c6
HX
623}
624EXPORT_SYMBOL_GPL(rhashtable_walk_next);
625
626/**
627 * rhashtable_walk_stop - Finish a hash table walk
628 * @iter: Hash table iterator
629 *
630 * Finish a hash table walk.
631 */
632void rhashtable_walk_stop(struct rhashtable_iter *iter)
db4374f4 633 __releases(RCU)
f2dba9c6 634{
eddee5ba
HX
635 struct rhashtable *ht;
636 struct bucket_table *tbl = iter->walker->tbl;
637
eddee5ba 638 if (!tbl)
963ecbd4 639 goto out;
eddee5ba
HX
640
641 ht = iter->ht;
642
ba7c95ea 643 spin_lock(&ht->lock);
c4db8848 644 if (tbl->rehash < tbl->size)
eddee5ba
HX
645 list_add(&iter->walker->list, &tbl->walkers);
646 else
647 iter->walker->tbl = NULL;
ba7c95ea 648 spin_unlock(&ht->lock);
eddee5ba 649
f2dba9c6 650 iter->p = NULL;
963ecbd4
HX
651
652out:
653 rcu_read_unlock();
f2dba9c6
HX
654}
655EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
656
488fb86e 657static size_t rounded_hashtable_size(const struct rhashtable_params *params)
7e1e7763 658{
94000176 659 return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
e2e21c1c 660 (unsigned long)params->min_size);
7e1e7763
TG
661}
662
31ccde2d
HX
663static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
664{
665 return jhash2(key, length, seed);
666}
667
7e1e7763
TG
668/**
669 * rhashtable_init - initialize a new hash table
670 * @ht: hash table to be initialized
671 * @params: configuration parameters
672 *
673 * Initializes a new hash table based on the provided configuration
674 * parameters. A table can be configured either with a variable or
675 * fixed length key:
676 *
677 * Configuration Example 1: Fixed length keys
678 * struct test_obj {
679 * int key;
680 * void * my_member;
681 * struct rhash_head node;
682 * };
683 *
684 * struct rhashtable_params params = {
685 * .head_offset = offsetof(struct test_obj, node),
686 * .key_offset = offsetof(struct test_obj, key),
687 * .key_len = sizeof(int),
87545899 688 * .hashfn = jhash,
f89bd6f8 689 * .nulls_base = (1U << RHT_BASE_SHIFT),
7e1e7763
TG
690 * };
691 *
692 * Configuration Example 2: Variable length keys
693 * struct test_obj {
694 * [...]
695 * struct rhash_head node;
696 * };
697 *
49f7b33e 698 * u32 my_hash_fn(const void *data, u32 len, u32 seed)
7e1e7763
TG
699 * {
700 * struct test_obj *obj = data;
701 *
702 * return [... hash ...];
703 * }
704 *
705 * struct rhashtable_params params = {
706 * .head_offset = offsetof(struct test_obj, node),
87545899 707 * .hashfn = jhash,
7e1e7763 708 * .obj_hashfn = my_hash_fn,
7e1e7763
TG
709 * };
710 */
488fb86e
HX
711int rhashtable_init(struct rhashtable *ht,
712 const struct rhashtable_params *params)
7e1e7763
TG
713{
714 struct bucket_table *tbl;
715 size_t size;
716
717 size = HASH_DEFAULT_SIZE;
718
31ccde2d 719 if ((!params->key_len && !params->obj_hashfn) ||
02fd97c3 720 (params->obj_hashfn && !params->obj_cmpfn))
7e1e7763
TG
721 return -EINVAL;
722
f89bd6f8
TG
723 if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
724 return -EINVAL;
725
7e1e7763 726 if (params->nelem_hint)
94000176 727 size = rounded_hashtable_size(params);
7e1e7763 728
97defe1e
TG
729 memset(ht, 0, sizeof(*ht));
730 mutex_init(&ht->mutex);
ba7c95ea 731 spin_lock_init(&ht->lock);
97defe1e
TG
732 memcpy(&ht->p, params, sizeof(*params));
733
a998f712
TG
734 if (params->min_size)
735 ht->p.min_size = roundup_pow_of_two(params->min_size);
736
737 if (params->max_size)
738 ht->p.max_size = rounddown_pow_of_two(params->max_size);
739
07ee0722
HX
740 if (params->insecure_max_entries)
741 ht->p.insecure_max_entries =
742 rounddown_pow_of_two(params->insecure_max_entries);
743 else
744 ht->p.insecure_max_entries = ht->p.max_size * 2;
745
488fb86e 746 ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE);
a998f712 747
27ed44a5
HX
748 /* The maximum (not average) chain length grows with the
749 * size of the hash table, at a rate of (log N)/(log log N).
750 * The value of 16 is selected so that even if the hash
751 * table grew to 2^32 you would not expect the maximum
752 * chain length to exceed it unless we are under attack
753 * (or extremely unlucky).
754 *
755 * As this limit is only to detect attacks, we don't need
756 * to set it to a lower value as you'd need the chain
757 * length to vastly exceed 16 to have any real effect
758 * on the system.
759 */
ccd57b1b
HX
760 if (!params->insecure_elasticity)
761 ht->elasticity = 16;
762
97defe1e
TG
763 if (params->locks_mul)
764 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
765 else
766 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
767
31ccde2d
HX
768 ht->key_len = ht->p.key_len;
769 if (!params->hashfn) {
770 ht->p.hashfn = jhash;
771
772 if (!(ht->key_len & (sizeof(u32) - 1))) {
773 ht->key_len /= sizeof(u32);
774 ht->p.hashfn = rhashtable_jhash2;
775 }
776 }
777
b9ecfdaa 778 tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
7e1e7763
TG
779 if (tbl == NULL)
780 return -ENOMEM;
781
545a148e 782 atomic_set(&ht->nelems, 0);
a5b6846f 783
7e1e7763
TG
784 RCU_INIT_POINTER(ht->tbl, tbl);
785
4c4b52d9 786 INIT_WORK(&ht->run_work, rht_deferred_worker);
97defe1e 787
7e1e7763
TG
788 return 0;
789}
790EXPORT_SYMBOL_GPL(rhashtable_init);
791
792/**
6b6f302c 793 * rhashtable_free_and_destroy - free elements and destroy hash table
7e1e7763 794 * @ht: the hash table to destroy
6b6f302c
TG
795 * @free_fn: callback to release resources of element
796 * @arg: pointer passed to free_fn
7e1e7763 797 *
6b6f302c
TG
798 * Stops an eventual async resize. If defined, invokes free_fn for each
799 * element to releasal resources. Please note that RCU protected
800 * readers may still be accessing the elements. Releasing of resources
801 * must occur in a compatible manner. Then frees the bucket array.
802 *
803 * This function will eventually sleep to wait for an async resize
804 * to complete. The caller is responsible that no further write operations
805 * occurs in parallel.
7e1e7763 806 */
6b6f302c
TG
807void rhashtable_free_and_destroy(struct rhashtable *ht,
808 void (*free_fn)(void *ptr, void *arg),
809 void *arg)
7e1e7763 810{
6b6f302c
TG
811 const struct bucket_table *tbl;
812 unsigned int i;
97defe1e 813
4c4b52d9 814 cancel_work_sync(&ht->run_work);
97defe1e 815
57699a40 816 mutex_lock(&ht->mutex);
6b6f302c
TG
817 tbl = rht_dereference(ht->tbl, ht);
818 if (free_fn) {
819 for (i = 0; i < tbl->size; i++) {
820 struct rhash_head *pos, *next;
821
822 for (pos = rht_dereference(tbl->buckets[i], ht),
823 next = !rht_is_a_nulls(pos) ?
824 rht_dereference(pos->next, ht) : NULL;
825 !rht_is_a_nulls(pos);
826 pos = next,
827 next = !rht_is_a_nulls(pos) ?
828 rht_dereference(pos->next, ht) : NULL)
829 free_fn(rht_obj(ht, pos), arg);
830 }
831 }
832
833 bucket_table_free(tbl);
97defe1e 834 mutex_unlock(&ht->mutex);
7e1e7763 835}
6b6f302c
TG
836EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
837
838void rhashtable_destroy(struct rhashtable *ht)
839{
840 return rhashtable_free_and_destroy(ht, NULL, NULL);
841}
7e1e7763 842EXPORT_SYMBOL_GPL(rhashtable_destroy);