usb: Don't die twice if PCI xhci host is not responding in resume
[linux/fpc-iii.git] / lib / rhashtable.c
blob101dac085c62e532d843b74a5856ba35b4b336bb
1 /*
2 * Resizable, Scalable, Concurrent Hash Table
4 * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au>
5 * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
6 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
8 * Code partially derived from nft_hash
9 * Rewritten with rehash code from br_multicast plus single list
10 * pointer as suggested by Josh Triplett
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.
17 #include <linux/atomic.h>
18 #include <linux/kernel.h>
19 #include <linux/init.h>
20 #include <linux/log2.h>
21 #include <linux/sched.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/mm.h>
25 #include <linux/jhash.h>
26 #include <linux/random.h>
27 #include <linux/rhashtable.h>
28 #include <linux/err.h>
29 #include <linux/export.h>
31 #define HASH_DEFAULT_SIZE 64UL
32 #define HASH_MIN_SIZE 4U
33 #define BUCKET_LOCKS_PER_CPU 32UL
35 static u32 head_hashfn(struct rhashtable *ht,
36 const struct bucket_table *tbl,
37 const struct rhash_head *he)
39 return rht_head_hashfn(ht, tbl, he, ht->p);
42 #ifdef CONFIG_PROVE_LOCKING
43 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
45 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
47 return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
49 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
51 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
53 spinlock_t *lock = rht_bucket_lock(tbl, hash);
55 return (debug_locks) ? lockdep_is_held(lock) : 1;
57 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
58 #else
59 #define ASSERT_RHT_MUTEX(HT)
60 #endif
63 static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl,
64 gfp_t gfp)
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
73 nr_pcpus = min_t(unsigned int, nr_pcpus, 64UL);
74 size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
76 /* Never allocate more than 0.5 locks per bucket */
77 size = min_t(unsigned int, size, tbl->size >> 1);
79 if (sizeof(spinlock_t) != 0) {
80 tbl->locks = NULL;
81 #ifdef CONFIG_NUMA
82 if (size * sizeof(spinlock_t) > PAGE_SIZE &&
83 gfp == GFP_KERNEL)
84 tbl->locks = vmalloc(size * sizeof(spinlock_t));
85 #endif
86 if (gfp != GFP_KERNEL)
87 gfp |= __GFP_NOWARN | __GFP_NORETRY;
89 if (!tbl->locks)
90 tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
91 gfp);
92 if (!tbl->locks)
93 return -ENOMEM;
94 for (i = 0; i < size; i++)
95 spin_lock_init(&tbl->locks[i]);
97 tbl->locks_mask = size - 1;
99 return 0;
102 static void bucket_table_free(const struct bucket_table *tbl)
104 if (tbl)
105 kvfree(tbl->locks);
107 kvfree(tbl);
110 static void bucket_table_free_rcu(struct rcu_head *head)
112 bucket_table_free(container_of(head, struct bucket_table, rcu));
115 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
116 size_t nbuckets,
117 gfp_t gfp)
119 struct bucket_table *tbl = NULL;
120 size_t size;
121 int i;
123 size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
124 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) ||
125 gfp != GFP_KERNEL)
126 tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
127 if (tbl == NULL && gfp == GFP_KERNEL)
128 tbl = vzalloc(size);
129 if (tbl == NULL)
130 return NULL;
132 tbl->size = nbuckets;
134 if (alloc_bucket_locks(ht, tbl, gfp) < 0) {
135 bucket_table_free(tbl);
136 return NULL;
139 INIT_LIST_HEAD(&tbl->walkers);
141 get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
143 for (i = 0; i < nbuckets; i++)
144 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
146 return tbl;
149 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
150 struct bucket_table *tbl)
152 struct bucket_table *new_tbl;
154 do {
155 new_tbl = tbl;
156 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
157 } while (tbl);
159 return new_tbl;
162 static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
164 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
165 struct bucket_table *new_tbl = rhashtable_last_table(ht,
166 rht_dereference_rcu(old_tbl->future_tbl, ht));
167 struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash];
168 int err = -ENOENT;
169 struct rhash_head *head, *next, *entry;
170 spinlock_t *new_bucket_lock;
171 unsigned int new_hash;
173 rht_for_each(entry, old_tbl, old_hash) {
174 err = 0;
175 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
177 if (rht_is_a_nulls(next))
178 break;
180 pprev = &entry->next;
183 if (err)
184 goto out;
186 new_hash = head_hashfn(ht, new_tbl, entry);
188 new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
190 spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
191 head = rht_dereference_bucket(new_tbl->buckets[new_hash],
192 new_tbl, new_hash);
194 RCU_INIT_POINTER(entry->next, head);
196 rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
197 spin_unlock(new_bucket_lock);
199 rcu_assign_pointer(*pprev, next);
201 out:
202 return err;
205 static void rhashtable_rehash_chain(struct rhashtable *ht,
206 unsigned int old_hash)
208 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
209 spinlock_t *old_bucket_lock;
211 old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
213 spin_lock_bh(old_bucket_lock);
214 while (!rhashtable_rehash_one(ht, old_hash))
216 old_tbl->rehash++;
217 spin_unlock_bh(old_bucket_lock);
220 static int rhashtable_rehash_attach(struct rhashtable *ht,
221 struct bucket_table *old_tbl,
222 struct bucket_table *new_tbl)
224 /* Protect future_tbl using the first bucket lock. */
225 spin_lock_bh(old_tbl->locks);
227 /* Did somebody beat us to it? */
228 if (rcu_access_pointer(old_tbl->future_tbl)) {
229 spin_unlock_bh(old_tbl->locks);
230 return -EEXIST;
233 /* Make insertions go into the new, empty table right away. Deletions
234 * and lookups will be attempted in both tables until we synchronize.
236 rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
238 spin_unlock_bh(old_tbl->locks);
240 return 0;
243 static int rhashtable_rehash_table(struct rhashtable *ht)
245 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
246 struct bucket_table *new_tbl;
247 struct rhashtable_walker *walker;
248 unsigned int old_hash;
250 new_tbl = rht_dereference(old_tbl->future_tbl, ht);
251 if (!new_tbl)
252 return 0;
254 for (old_hash = 0; old_hash < old_tbl->size; old_hash++)
255 rhashtable_rehash_chain(ht, old_hash);
257 /* Publish the new table pointer. */
258 rcu_assign_pointer(ht->tbl, new_tbl);
260 spin_lock(&ht->lock);
261 list_for_each_entry(walker, &old_tbl->walkers, list)
262 walker->tbl = NULL;
263 spin_unlock(&ht->lock);
265 /* Wait for readers. All new readers will see the new
266 * table, and thus no references to the old table will
267 * remain.
269 call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
271 return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
275 * rhashtable_expand - Expand hash table while allowing concurrent lookups
276 * @ht: the hash table to expand
278 * A secondary bucket array is allocated and the hash entries are migrated.
280 * This function may only be called in a context where it is safe to call
281 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
283 * The caller must ensure that no concurrent resizing occurs by holding
284 * ht->mutex.
286 * It is valid to have concurrent insertions and deletions protected by per
287 * bucket locks or concurrent RCU protected lookups and traversals.
289 static int rhashtable_expand(struct rhashtable *ht)
291 struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
292 int err;
294 ASSERT_RHT_MUTEX(ht);
296 old_tbl = rhashtable_last_table(ht, old_tbl);
298 new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL);
299 if (new_tbl == NULL)
300 return -ENOMEM;
302 err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
303 if (err)
304 bucket_table_free(new_tbl);
306 return err;
310 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
311 * @ht: the hash table to shrink
313 * This function shrinks the hash table to fit, i.e., the smallest
314 * size would not cause it to expand right away automatically.
316 * The caller must ensure that no concurrent resizing occurs by holding
317 * ht->mutex.
319 * The caller must ensure that no concurrent table mutations take place.
320 * It is however valid to have concurrent lookups if they are RCU protected.
322 * It is valid to have concurrent insertions and deletions protected by per
323 * bucket locks or concurrent RCU protected lookups and traversals.
325 static int rhashtable_shrink(struct rhashtable *ht)
327 struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
328 unsigned int nelems = atomic_read(&ht->nelems);
329 unsigned int size = 0;
330 int err;
332 ASSERT_RHT_MUTEX(ht);
334 if (nelems)
335 size = roundup_pow_of_two(nelems * 3 / 2);
336 if (size < ht->p.min_size)
337 size = ht->p.min_size;
339 if (old_tbl->size <= size)
340 return 0;
342 if (rht_dereference(old_tbl->future_tbl, ht))
343 return -EEXIST;
345 new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
346 if (new_tbl == NULL)
347 return -ENOMEM;
349 err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
350 if (err)
351 bucket_table_free(new_tbl);
353 return err;
356 static void rht_deferred_worker(struct work_struct *work)
358 struct rhashtable *ht;
359 struct bucket_table *tbl;
360 int err = 0;
362 ht = container_of(work, struct rhashtable, run_work);
363 mutex_lock(&ht->mutex);
365 tbl = rht_dereference(ht->tbl, ht);
366 tbl = rhashtable_last_table(ht, tbl);
368 if (rht_grow_above_75(ht, tbl))
369 rhashtable_expand(ht);
370 else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
371 rhashtable_shrink(ht);
373 err = rhashtable_rehash_table(ht);
375 mutex_unlock(&ht->mutex);
377 if (err)
378 schedule_work(&ht->run_work);
381 static int rhashtable_insert_rehash(struct rhashtable *ht,
382 struct bucket_table *tbl)
384 struct bucket_table *old_tbl;
385 struct bucket_table *new_tbl;
386 unsigned int size;
387 int err;
389 old_tbl = rht_dereference_rcu(ht->tbl, ht);
391 size = tbl->size;
393 err = -EBUSY;
395 if (rht_grow_above_75(ht, tbl))
396 size *= 2;
397 /* Do not schedule more than one rehash */
398 else if (old_tbl != tbl)
399 goto fail;
401 err = -ENOMEM;
403 new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
404 if (new_tbl == NULL)
405 goto fail;
407 err = rhashtable_rehash_attach(ht, tbl, new_tbl);
408 if (err) {
409 bucket_table_free(new_tbl);
410 if (err == -EEXIST)
411 err = 0;
412 } else
413 schedule_work(&ht->run_work);
415 return err;
417 fail:
418 /* Do not fail the insert if someone else did a rehash. */
419 if (likely(rcu_dereference_raw(tbl->future_tbl)))
420 return 0;
422 /* Schedule async rehash to retry allocation in process context. */
423 if (err == -ENOMEM)
424 schedule_work(&ht->run_work);
426 return err;
429 static void *rhashtable_lookup_one(struct rhashtable *ht,
430 struct bucket_table *tbl, unsigned int hash,
431 const void *key, struct rhash_head *obj)
433 struct rhashtable_compare_arg arg = {
434 .ht = ht,
435 .key = key,
437 struct rhash_head __rcu **pprev;
438 struct rhash_head *head;
439 int elasticity;
441 elasticity = ht->elasticity;
442 pprev = &tbl->buckets[hash];
443 rht_for_each(head, tbl, hash) {
444 struct rhlist_head *list;
445 struct rhlist_head *plist;
447 elasticity--;
448 if (!key ||
449 (ht->p.obj_cmpfn ?
450 ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
451 rhashtable_compare(&arg, rht_obj(ht, head)))) {
452 pprev = &head->next;
453 continue;
456 if (!ht->rhlist)
457 return rht_obj(ht, head);
459 list = container_of(obj, struct rhlist_head, rhead);
460 plist = container_of(head, struct rhlist_head, rhead);
462 RCU_INIT_POINTER(list->next, plist);
463 head = rht_dereference_bucket(head->next, tbl, hash);
464 RCU_INIT_POINTER(list->rhead.next, head);
465 rcu_assign_pointer(*pprev, obj);
467 return NULL;
470 if (elasticity <= 0)
471 return ERR_PTR(-EAGAIN);
473 return ERR_PTR(-ENOENT);
476 static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
477 struct bucket_table *tbl,
478 unsigned int hash,
479 struct rhash_head *obj,
480 void *data)
482 struct bucket_table *new_tbl;
483 struct rhash_head *head;
485 if (!IS_ERR_OR_NULL(data))
486 return ERR_PTR(-EEXIST);
488 if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
489 return ERR_CAST(data);
491 new_tbl = rcu_dereference(tbl->future_tbl);
492 if (new_tbl)
493 return new_tbl;
495 if (PTR_ERR(data) != -ENOENT)
496 return ERR_CAST(data);
498 if (unlikely(rht_grow_above_max(ht, tbl)))
499 return ERR_PTR(-E2BIG);
501 if (unlikely(rht_grow_above_100(ht, tbl)))
502 return ERR_PTR(-EAGAIN);
504 head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
506 RCU_INIT_POINTER(obj->next, head);
507 if (ht->rhlist) {
508 struct rhlist_head *list;
510 list = container_of(obj, struct rhlist_head, rhead);
511 RCU_INIT_POINTER(list->next, NULL);
514 rcu_assign_pointer(tbl->buckets[hash], obj);
516 atomic_inc(&ht->nelems);
517 if (rht_grow_above_75(ht, tbl))
518 schedule_work(&ht->run_work);
520 return NULL;
523 static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
524 struct rhash_head *obj)
526 struct bucket_table *new_tbl;
527 struct bucket_table *tbl;
528 unsigned int hash;
529 spinlock_t *lock;
530 void *data;
532 tbl = rcu_dereference(ht->tbl);
534 /* All insertions must grab the oldest table containing
535 * the hashed bucket that is yet to be rehashed.
537 for (;;) {
538 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
539 lock = rht_bucket_lock(tbl, hash);
540 spin_lock_bh(lock);
542 if (tbl->rehash <= hash)
543 break;
545 spin_unlock_bh(lock);
546 tbl = rcu_dereference(tbl->future_tbl);
549 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
550 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
551 if (PTR_ERR(new_tbl) != -EEXIST)
552 data = ERR_CAST(new_tbl);
554 while (!IS_ERR_OR_NULL(new_tbl)) {
555 tbl = new_tbl;
556 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
557 spin_lock_nested(rht_bucket_lock(tbl, hash),
558 SINGLE_DEPTH_NESTING);
560 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
561 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
562 if (PTR_ERR(new_tbl) != -EEXIST)
563 data = ERR_CAST(new_tbl);
565 spin_unlock(rht_bucket_lock(tbl, hash));
568 spin_unlock_bh(lock);
570 if (PTR_ERR(data) == -EAGAIN)
571 data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
572 -EAGAIN);
574 return data;
577 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
578 struct rhash_head *obj)
580 void *data;
582 do {
583 rcu_read_lock();
584 data = rhashtable_try_insert(ht, key, obj);
585 rcu_read_unlock();
586 } while (PTR_ERR(data) == -EAGAIN);
588 return data;
590 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
593 * rhashtable_walk_enter - Initialise an iterator
594 * @ht: Table to walk over
595 * @iter: Hash table Iterator
597 * This function prepares a hash table walk.
599 * Note that if you restart a walk after rhashtable_walk_stop you
600 * may see the same object twice. Also, you may miss objects if
601 * there are removals in between rhashtable_walk_stop and the next
602 * call to rhashtable_walk_start.
604 * For a completely stable walk you should construct your own data
605 * structure outside the hash table.
607 * This function may sleep so you must not call it from interrupt
608 * context or with spin locks held.
610 * You must call rhashtable_walk_exit after this function returns.
612 void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
614 iter->ht = ht;
615 iter->p = NULL;
616 iter->slot = 0;
617 iter->skip = 0;
619 spin_lock(&ht->lock);
620 iter->walker.tbl =
621 rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
622 list_add(&iter->walker.list, &iter->walker.tbl->walkers);
623 spin_unlock(&ht->lock);
625 EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
628 * rhashtable_walk_exit - Free an iterator
629 * @iter: Hash table Iterator
631 * This function frees resources allocated by rhashtable_walk_init.
633 void rhashtable_walk_exit(struct rhashtable_iter *iter)
635 spin_lock(&iter->ht->lock);
636 if (iter->walker.tbl)
637 list_del(&iter->walker.list);
638 spin_unlock(&iter->ht->lock);
640 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
643 * rhashtable_walk_start - Start a hash table walk
644 * @iter: Hash table iterator
646 * Start a hash table walk. Note that we take the RCU lock in all
647 * cases including when we return an error. So you must always call
648 * rhashtable_walk_stop to clean up.
650 * Returns zero if successful.
652 * Returns -EAGAIN if resize event occured. Note that the iterator
653 * will rewind back to the beginning and you may use it immediately
654 * by calling rhashtable_walk_next.
656 int rhashtable_walk_start(struct rhashtable_iter *iter)
657 __acquires(RCU)
659 struct rhashtable *ht = iter->ht;
661 rcu_read_lock();
663 spin_lock(&ht->lock);
664 if (iter->walker.tbl)
665 list_del(&iter->walker.list);
666 spin_unlock(&ht->lock);
668 if (!iter->walker.tbl) {
669 iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
670 return -EAGAIN;
673 return 0;
675 EXPORT_SYMBOL_GPL(rhashtable_walk_start);
678 * rhashtable_walk_next - Return the next object and advance the iterator
679 * @iter: Hash table iterator
681 * Note that you must call rhashtable_walk_stop when you are finished
682 * with the walk.
684 * Returns the next object or NULL when the end of the table is reached.
686 * Returns -EAGAIN if resize event occured. Note that the iterator
687 * will rewind back to the beginning and you may continue to use it.
689 void *rhashtable_walk_next(struct rhashtable_iter *iter)
691 struct bucket_table *tbl = iter->walker.tbl;
692 struct rhlist_head *list = iter->list;
693 struct rhashtable *ht = iter->ht;
694 struct rhash_head *p = iter->p;
695 bool rhlist = ht->rhlist;
697 if (p) {
698 if (!rhlist || !(list = rcu_dereference(list->next))) {
699 p = rcu_dereference(p->next);
700 list = container_of(p, struct rhlist_head, rhead);
702 goto next;
705 for (; iter->slot < tbl->size; iter->slot++) {
706 int skip = iter->skip;
708 rht_for_each_rcu(p, tbl, iter->slot) {
709 if (rhlist) {
710 list = container_of(p, struct rhlist_head,
711 rhead);
712 do {
713 if (!skip)
714 goto next;
715 skip--;
716 list = rcu_dereference(list->next);
717 } while (list);
719 continue;
721 if (!skip)
722 break;
723 skip--;
726 next:
727 if (!rht_is_a_nulls(p)) {
728 iter->skip++;
729 iter->p = p;
730 iter->list = list;
731 return rht_obj(ht, rhlist ? &list->rhead : p);
734 iter->skip = 0;
737 iter->p = NULL;
739 /* Ensure we see any new tables. */
740 smp_rmb();
742 iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht);
743 if (iter->walker.tbl) {
744 iter->slot = 0;
745 iter->skip = 0;
746 return ERR_PTR(-EAGAIN);
749 return NULL;
751 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
754 * rhashtable_walk_stop - Finish a hash table walk
755 * @iter: Hash table iterator
757 * Finish a hash table walk.
759 void rhashtable_walk_stop(struct rhashtable_iter *iter)
760 __releases(RCU)
762 struct rhashtable *ht;
763 struct bucket_table *tbl = iter->walker.tbl;
765 if (!tbl)
766 goto out;
768 ht = iter->ht;
770 spin_lock(&ht->lock);
771 if (tbl->rehash < tbl->size)
772 list_add(&iter->walker.list, &tbl->walkers);
773 else
774 iter->walker.tbl = NULL;
775 spin_unlock(&ht->lock);
777 iter->p = NULL;
779 out:
780 rcu_read_unlock();
782 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
784 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
786 size_t retsize;
788 if (params->nelem_hint)
789 retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
790 (unsigned long)params->min_size);
791 else
792 retsize = max(HASH_DEFAULT_SIZE,
793 (unsigned long)params->min_size);
795 return retsize;
798 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
800 return jhash2(key, length, seed);
804 * rhashtable_init - initialize a new hash table
805 * @ht: hash table to be initialized
806 * @params: configuration parameters
808 * Initializes a new hash table based on the provided configuration
809 * parameters. A table can be configured either with a variable or
810 * fixed length key:
812 * Configuration Example 1: Fixed length keys
813 * struct test_obj {
814 * int key;
815 * void * my_member;
816 * struct rhash_head node;
817 * };
819 * struct rhashtable_params params = {
820 * .head_offset = offsetof(struct test_obj, node),
821 * .key_offset = offsetof(struct test_obj, key),
822 * .key_len = sizeof(int),
823 * .hashfn = jhash,
824 * .nulls_base = (1U << RHT_BASE_SHIFT),
825 * };
827 * Configuration Example 2: Variable length keys
828 * struct test_obj {
829 * [...]
830 * struct rhash_head node;
831 * };
833 * u32 my_hash_fn(const void *data, u32 len, u32 seed)
835 * struct test_obj *obj = data;
837 * return [... hash ...];
840 * struct rhashtable_params params = {
841 * .head_offset = offsetof(struct test_obj, node),
842 * .hashfn = jhash,
843 * .obj_hashfn = my_hash_fn,
844 * };
846 int rhashtable_init(struct rhashtable *ht,
847 const struct rhashtable_params *params)
849 struct bucket_table *tbl;
850 size_t size;
852 if ((!params->key_len && !params->obj_hashfn) ||
853 (params->obj_hashfn && !params->obj_cmpfn))
854 return -EINVAL;
856 if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
857 return -EINVAL;
859 memset(ht, 0, sizeof(*ht));
860 mutex_init(&ht->mutex);
861 spin_lock_init(&ht->lock);
862 memcpy(&ht->p, params, sizeof(*params));
864 if (params->min_size)
865 ht->p.min_size = roundup_pow_of_two(params->min_size);
867 if (params->max_size)
868 ht->p.max_size = rounddown_pow_of_two(params->max_size);
870 if (params->insecure_max_entries)
871 ht->p.insecure_max_entries =
872 rounddown_pow_of_two(params->insecure_max_entries);
873 else
874 ht->p.insecure_max_entries = ht->p.max_size * 2;
876 ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE);
878 size = rounded_hashtable_size(&ht->p);
880 /* The maximum (not average) chain length grows with the
881 * size of the hash table, at a rate of (log N)/(log log N).
882 * The value of 16 is selected so that even if the hash
883 * table grew to 2^32 you would not expect the maximum
884 * chain length to exceed it unless we are under attack
885 * (or extremely unlucky).
887 * As this limit is only to detect attacks, we don't need
888 * to set it to a lower value as you'd need the chain
889 * length to vastly exceed 16 to have any real effect
890 * on the system.
892 if (!params->insecure_elasticity)
893 ht->elasticity = 16;
895 if (params->locks_mul)
896 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
897 else
898 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
900 ht->key_len = ht->p.key_len;
901 if (!params->hashfn) {
902 ht->p.hashfn = jhash;
904 if (!(ht->key_len & (sizeof(u32) - 1))) {
905 ht->key_len /= sizeof(u32);
906 ht->p.hashfn = rhashtable_jhash2;
910 tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
911 if (tbl == NULL)
912 return -ENOMEM;
914 atomic_set(&ht->nelems, 0);
916 RCU_INIT_POINTER(ht->tbl, tbl);
918 INIT_WORK(&ht->run_work, rht_deferred_worker);
920 return 0;
922 EXPORT_SYMBOL_GPL(rhashtable_init);
925 * rhltable_init - initialize a new hash list table
926 * @hlt: hash list table to be initialized
927 * @params: configuration parameters
929 * Initializes a new hash list table.
931 * See documentation for rhashtable_init.
933 int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
935 int err;
937 /* No rhlist NULLs marking for now. */
938 if (params->nulls_base)
939 return -EINVAL;
941 err = rhashtable_init(&hlt->ht, params);
942 hlt->ht.rhlist = true;
943 return err;
945 EXPORT_SYMBOL_GPL(rhltable_init);
947 static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
948 void (*free_fn)(void *ptr, void *arg),
949 void *arg)
951 struct rhlist_head *list;
953 if (!ht->rhlist) {
954 free_fn(rht_obj(ht, obj), arg);
955 return;
958 list = container_of(obj, struct rhlist_head, rhead);
959 do {
960 obj = &list->rhead;
961 list = rht_dereference(list->next, ht);
962 free_fn(rht_obj(ht, obj), arg);
963 } while (list);
967 * rhashtable_free_and_destroy - free elements and destroy hash table
968 * @ht: the hash table to destroy
969 * @free_fn: callback to release resources of element
970 * @arg: pointer passed to free_fn
972 * Stops an eventual async resize. If defined, invokes free_fn for each
973 * element to releasal resources. Please note that RCU protected
974 * readers may still be accessing the elements. Releasing of resources
975 * must occur in a compatible manner. Then frees the bucket array.
977 * This function will eventually sleep to wait for an async resize
978 * to complete. The caller is responsible that no further write operations
979 * occurs in parallel.
981 void rhashtable_free_and_destroy(struct rhashtable *ht,
982 void (*free_fn)(void *ptr, void *arg),
983 void *arg)
985 const struct bucket_table *tbl;
986 unsigned int i;
988 cancel_work_sync(&ht->run_work);
990 mutex_lock(&ht->mutex);
991 tbl = rht_dereference(ht->tbl, ht);
992 if (free_fn) {
993 for (i = 0; i < tbl->size; i++) {
994 struct rhash_head *pos, *next;
996 for (pos = rht_dereference(tbl->buckets[i], ht),
997 next = !rht_is_a_nulls(pos) ?
998 rht_dereference(pos->next, ht) : NULL;
999 !rht_is_a_nulls(pos);
1000 pos = next,
1001 next = !rht_is_a_nulls(pos) ?
1002 rht_dereference(pos->next, ht) : NULL)
1003 rhashtable_free_one(ht, pos, free_fn, arg);
1007 bucket_table_free(tbl);
1008 mutex_unlock(&ht->mutex);
1010 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
1012 void rhashtable_destroy(struct rhashtable *ht)
1014 return rhashtable_free_and_destroy(ht, NULL, NULL);
1016 EXPORT_SYMBOL_GPL(rhashtable_destroy);