2 * Resizable, Scalable, Concurrent Hash Table
4 * Copyright (c) 2014 Thomas Graf <tgraf@suug.ch>
5 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
7 * Based on the following paper:
8 * https://www.usenix.org/legacy/event/atc11/tech/final_files/Triplett.pdf
10 * Code partially derived from nft_hash
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/kernel.h>
18 #include <linux/init.h>
19 #include <linux/log2.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
23 #include <linux/jhash.h>
24 #include <linux/random.h>
25 #include <linux/rhashtable.h>
27 #define HASH_DEFAULT_SIZE 64UL
28 #define HASH_MIN_SIZE 4UL
29 #define BUCKET_LOCKS_PER_CPU 128UL
31 /* Base bits plus 1 bit for nulls marker */
32 #define HASH_RESERVED_SPACE (RHT_BASE_BITS + 1)
40 /* The bucket lock is selected based on the hash and protects mutations
41 * on a group of hash buckets.
43 * IMPORTANT: When holding the bucket lock of both the old and new table
44 * during expansions and shrinking, the old bucket lock must always be
47 static spinlock_t *bucket_lock(const struct bucket_table *tbl, u32 hash)
49 return &tbl->locks[hash & tbl->locks_mask];
52 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
53 #define ASSERT_BUCKET_LOCK(TBL, HASH) \
54 BUG_ON(!lockdep_rht_bucket_is_held(TBL, HASH))
56 #ifdef CONFIG_PROVE_LOCKING
57 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
59 return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
61 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
63 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
65 spinlock_t *lock = bucket_lock(tbl, hash);
67 return (debug_locks) ? lockdep_is_held(lock) : 1;
69 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
72 static void *rht_obj(const struct rhashtable *ht, const struct rhash_head *he)
74 return (void *) he - ht->p.head_offset;
77 static u32 rht_bucket_index(const struct bucket_table *tbl, u32 hash)
79 return hash & (tbl->size - 1);
82 static u32 obj_raw_hashfn(const struct rhashtable *ht, const void *ptr)
86 if (unlikely(!ht->p.key_len))
87 hash = ht->p.obj_hashfn(ptr, ht->p.hash_rnd);
89 hash = ht->p.hashfn(ptr + ht->p.key_offset, ht->p.key_len,
92 return hash >> HASH_RESERVED_SPACE;
95 static u32 key_hashfn(struct rhashtable *ht, const void *key, u32 len)
97 struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
100 hash = ht->p.hashfn(key, len, ht->p.hash_rnd);
101 hash >>= HASH_RESERVED_SPACE;
103 return rht_bucket_index(tbl, hash);
106 static u32 head_hashfn(const struct rhashtable *ht,
107 const struct bucket_table *tbl,
108 const struct rhash_head *he)
110 return rht_bucket_index(tbl, obj_raw_hashfn(ht, rht_obj(ht, he)));
113 static struct rhash_head __rcu **bucket_tail(struct bucket_table *tbl, u32 n)
115 struct rhash_head __rcu **pprev;
117 for (pprev = &tbl->buckets[n];
118 !rht_is_a_nulls(rht_dereference_bucket(*pprev, tbl, n));
119 pprev = &rht_dereference_bucket(*pprev, tbl, n)->next)
125 static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl)
127 unsigned int i, size;
128 #if defined(CONFIG_PROVE_LOCKING)
129 unsigned int nr_pcpus = 2;
131 unsigned int nr_pcpus = num_possible_cpus();
134 nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL);
135 size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
137 /* Never allocate more than one lock per bucket */
138 size = min_t(unsigned int, size, tbl->size);
140 if (sizeof(spinlock_t) != 0) {
142 if (size * sizeof(spinlock_t) > PAGE_SIZE)
143 tbl->locks = vmalloc(size * sizeof(spinlock_t));
146 tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
150 for (i = 0; i < size; i++)
151 spin_lock_init(&tbl->locks[i]);
153 tbl->locks_mask = size - 1;
158 static void bucket_table_free(const struct bucket_table *tbl)
166 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
169 struct bucket_table *tbl;
173 size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
174 tbl = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
181 tbl->size = nbuckets;
183 if (alloc_bucket_locks(ht, tbl) < 0) {
184 bucket_table_free(tbl);
188 for (i = 0; i < nbuckets; i++)
189 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
195 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
197 * @new_size: new table size
199 bool rht_grow_above_75(const struct rhashtable *ht, size_t new_size)
201 /* Expand table when exceeding 75% load */
202 return atomic_read(&ht->nelems) > (new_size / 4 * 3);
204 EXPORT_SYMBOL_GPL(rht_grow_above_75);
207 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
209 * @new_size: new table size
211 bool rht_shrink_below_30(const struct rhashtable *ht, size_t new_size)
213 /* Shrink table beneath 30% load */
214 return atomic_read(&ht->nelems) < (new_size * 3 / 10);
216 EXPORT_SYMBOL_GPL(rht_shrink_below_30);
218 static void hashtable_chain_unzip(const struct rhashtable *ht,
219 const struct bucket_table *new_tbl,
220 struct bucket_table *old_tbl,
223 struct rhash_head *he, *p, *next;
224 spinlock_t *new_bucket_lock, *new_bucket_lock2 = NULL;
225 unsigned int new_hash, new_hash2;
227 ASSERT_BUCKET_LOCK(old_tbl, old_hash);
229 /* Old bucket empty, no work needed. */
230 p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl,
232 if (rht_is_a_nulls(p))
235 new_hash = new_hash2 = head_hashfn(ht, new_tbl, p);
236 new_bucket_lock = bucket_lock(new_tbl, new_hash);
238 /* Advance the old bucket pointer one or more times until it
239 * reaches a node that doesn't hash to the same bucket as the
240 * previous node p. Call the previous node p;
242 rht_for_each_continue(he, p->next, old_tbl, old_hash) {
243 new_hash2 = head_hashfn(ht, new_tbl, he);
244 if (new_hash != new_hash2)
248 rcu_assign_pointer(old_tbl->buckets[old_hash], p->next);
250 spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED);
252 /* If we have encountered an entry that maps to a different bucket in
253 * the new table, lock down that bucket as well as we might cut off
254 * the end of the chain.
256 new_bucket_lock2 = bucket_lock(new_tbl, new_hash);
257 if (new_bucket_lock != new_bucket_lock2)
258 spin_lock_bh_nested(new_bucket_lock2, RHT_LOCK_NESTED2);
260 /* Find the subsequent node which does hash to the same
261 * bucket as node P, or NULL if no such node exists.
263 INIT_RHT_NULLS_HEAD(next, ht, old_hash);
264 if (!rht_is_a_nulls(he)) {
265 rht_for_each_continue(he, he->next, old_tbl, old_hash) {
266 if (head_hashfn(ht, new_tbl, he) == new_hash) {
273 /* Set p's next pointer to that subsequent node pointer,
274 * bypassing the nodes which do not hash to p's bucket
276 rcu_assign_pointer(p->next, next);
278 if (new_bucket_lock != new_bucket_lock2)
279 spin_unlock_bh(new_bucket_lock2);
280 spin_unlock_bh(new_bucket_lock);
283 static void link_old_to_new(struct bucket_table *new_tbl,
284 unsigned int new_hash, struct rhash_head *entry)
286 spinlock_t *new_bucket_lock;
288 new_bucket_lock = bucket_lock(new_tbl, new_hash);
290 spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED);
291 rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), entry);
292 spin_unlock_bh(new_bucket_lock);
296 * rhashtable_expand - Expand hash table while allowing concurrent lookups
297 * @ht: the hash table to expand
299 * A secondary bucket array is allocated and the hash entries are migrated
300 * while keeping them on both lists until the end of the RCU grace period.
302 * This function may only be called in a context where it is safe to call
303 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
305 * The caller must ensure that no concurrent resizing occurs by holding
308 * It is valid to have concurrent insertions and deletions protected by per
309 * bucket locks or concurrent RCU protected lookups and traversals.
311 int rhashtable_expand(struct rhashtable *ht)
313 struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
314 struct rhash_head *he;
315 spinlock_t *old_bucket_lock;
316 unsigned int new_hash, old_hash;
317 bool complete = false;
319 ASSERT_RHT_MUTEX(ht);
321 if (ht->p.max_shift && ht->shift >= ht->p.max_shift)
324 new_tbl = bucket_table_alloc(ht, old_tbl->size * 2);
330 /* Make insertions go into the new, empty table right away. Deletions
331 * and lookups will be attempted in both tables until we synchronize.
332 * The synchronize_rcu() guarantees for the new table to be picked up
333 * so no new additions go into the old table while we relink.
335 rcu_assign_pointer(ht->future_tbl, new_tbl);
338 /* For each new bucket, search the corresponding old bucket for the
339 * first entry that hashes to the new bucket, and link the end of
340 * newly formed bucket chain (containing entries added to future
341 * table) to that entry. Since all the entries which will end up in
342 * the new bucket appear in the same old bucket, this constructs an
343 * entirely valid new hash table, but with multiple buckets
344 * "zipped" together into a single imprecise chain.
346 for (new_hash = 0; new_hash < new_tbl->size; new_hash++) {
347 old_hash = rht_bucket_index(old_tbl, new_hash);
348 old_bucket_lock = bucket_lock(old_tbl, old_hash);
350 spin_lock_bh(old_bucket_lock);
351 rht_for_each(he, old_tbl, old_hash) {
352 if (head_hashfn(ht, new_tbl, he) == new_hash) {
353 link_old_to_new(new_tbl, new_hash, he);
357 spin_unlock_bh(old_bucket_lock);
360 /* Publish the new table pointer. Lookups may now traverse
361 * the new table, but they will not benefit from any
362 * additional efficiency until later steps unzip the buckets.
364 rcu_assign_pointer(ht->tbl, new_tbl);
366 /* Unzip interleaved hash chains */
367 while (!complete && !ht->being_destroyed) {
368 /* Wait for readers. All new readers will see the new
369 * table, and thus no references to the old table will
374 /* For each bucket in the old table (each of which
375 * contains items from multiple buckets of the new
379 for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
380 struct rhash_head *head;
382 old_bucket_lock = bucket_lock(old_tbl, old_hash);
383 spin_lock_bh(old_bucket_lock);
385 hashtable_chain_unzip(ht, new_tbl, old_tbl, old_hash);
386 head = rht_dereference_bucket(old_tbl->buckets[old_hash],
388 if (!rht_is_a_nulls(head))
391 spin_unlock_bh(old_bucket_lock);
395 bucket_table_free(old_tbl);
398 EXPORT_SYMBOL_GPL(rhashtable_expand);
401 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
402 * @ht: the hash table to shrink
404 * This function may only be called in a context where it is safe to call
405 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
407 * The caller must ensure that no concurrent resizing occurs by holding
410 * The caller must ensure that no concurrent table mutations take place.
411 * It is however valid to have concurrent lookups if they are RCU protected.
413 * It is valid to have concurrent insertions and deletions protected by per
414 * bucket locks or concurrent RCU protected lookups and traversals.
416 int rhashtable_shrink(struct rhashtable *ht)
418 struct bucket_table *new_tbl, *tbl = rht_dereference(ht->tbl, ht);
419 spinlock_t *new_bucket_lock, *old_bucket_lock1, *old_bucket_lock2;
420 unsigned int new_hash;
422 ASSERT_RHT_MUTEX(ht);
424 if (ht->shift <= ht->p.min_shift)
427 new_tbl = bucket_table_alloc(ht, tbl->size / 2);
431 rcu_assign_pointer(ht->future_tbl, new_tbl);
434 /* Link the first entry in the old bucket to the end of the
435 * bucket in the new table. As entries are concurrently being
436 * added to the new table, lock down the new bucket. As we
437 * always divide the size in half when shrinking, each bucket
438 * in the new table maps to exactly two buckets in the old
441 * As removals can occur concurrently on the old table, we need
442 * to lock down both matching buckets in the old table.
444 for (new_hash = 0; new_hash < new_tbl->size; new_hash++) {
445 old_bucket_lock1 = bucket_lock(tbl, new_hash);
446 old_bucket_lock2 = bucket_lock(tbl, new_hash + new_tbl->size);
447 new_bucket_lock = bucket_lock(new_tbl, new_hash);
449 spin_lock_bh(old_bucket_lock1);
450 spin_lock_bh_nested(old_bucket_lock2, RHT_LOCK_NESTED);
451 spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED2);
453 rcu_assign_pointer(*bucket_tail(new_tbl, new_hash),
454 tbl->buckets[new_hash]);
455 rcu_assign_pointer(*bucket_tail(new_tbl, new_hash),
456 tbl->buckets[new_hash + new_tbl->size]);
458 spin_unlock_bh(new_bucket_lock);
459 spin_unlock_bh(old_bucket_lock2);
460 spin_unlock_bh(old_bucket_lock1);
463 /* Publish the new, valid hash table */
464 rcu_assign_pointer(ht->tbl, new_tbl);
467 /* Wait for readers. No new readers will have references to the
472 bucket_table_free(tbl);
476 EXPORT_SYMBOL_GPL(rhashtable_shrink);
478 static void rht_deferred_worker(struct work_struct *work)
480 struct rhashtable *ht;
481 struct bucket_table *tbl;
483 ht = container_of(work, struct rhashtable, run_work.work);
484 mutex_lock(&ht->mutex);
485 tbl = rht_dereference(ht->tbl, ht);
487 if (ht->p.grow_decision && ht->p.grow_decision(ht, tbl->size))
488 rhashtable_expand(ht);
489 else if (ht->p.shrink_decision && ht->p.shrink_decision(ht, tbl->size))
490 rhashtable_shrink(ht);
492 mutex_unlock(&ht->mutex);
495 static void rhashtable_wakeup_worker(struct rhashtable *ht)
497 struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
498 struct bucket_table *new_tbl = rht_dereference_rcu(ht->future_tbl, ht);
499 size_t size = tbl->size;
501 /* Only adjust the table if no resizing is currently in progress. */
502 if (tbl == new_tbl &&
503 ((ht->p.grow_decision && ht->p.grow_decision(ht, size)) ||
504 (ht->p.shrink_decision && ht->p.shrink_decision(ht, size))))
505 schedule_delayed_work(&ht->run_work, 0);
508 static void __rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj,
509 struct bucket_table *tbl, u32 hash)
511 struct rhash_head *head = rht_dereference_bucket(tbl->buckets[hash],
514 if (rht_is_a_nulls(head))
515 INIT_RHT_NULLS_HEAD(obj->next, ht, hash);
517 RCU_INIT_POINTER(obj->next, head);
519 rcu_assign_pointer(tbl->buckets[hash], obj);
521 atomic_inc(&ht->nelems);
523 rhashtable_wakeup_worker(ht);
527 * rhashtable_insert - insert object into hash table
529 * @obj: pointer to hash head inside object
531 * Will take a per bucket spinlock to protect against mutual mutations
532 * on the same bucket. Multiple insertions may occur in parallel unless
533 * they map to the same bucket lock.
535 * It is safe to call this function from atomic context.
537 * Will trigger an automatic deferred table resizing if the size grows
538 * beyond the watermark indicated by grow_decision() which can be passed
539 * to rhashtable_init().
541 void rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj)
543 struct bucket_table *tbl;
549 tbl = rht_dereference_rcu(ht->future_tbl, ht);
550 hash = head_hashfn(ht, tbl, obj);
551 lock = bucket_lock(tbl, hash);
554 __rhashtable_insert(ht, obj, tbl, hash);
555 spin_unlock_bh(lock);
559 EXPORT_SYMBOL_GPL(rhashtable_insert);
562 * rhashtable_remove - remove object from hash table
564 * @obj: pointer to hash head inside object
566 * Since the hash chain is single linked, the removal operation needs to
567 * walk the bucket chain upon removal. The removal operation is thus
568 * considerable slow if the hash table is not correctly sized.
570 * Will automatically shrink the table via rhashtable_expand() if the
571 * shrink_decision function specified at rhashtable_init() returns true.
573 * The caller must ensure that no concurrent table mutations occur. It is
574 * however valid to have concurrent lookups if they are RCU protected.
576 bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *obj)
578 struct bucket_table *tbl;
579 struct rhash_head __rcu **pprev;
580 struct rhash_head *he;
585 tbl = rht_dereference_rcu(ht->tbl, ht);
586 hash = head_hashfn(ht, tbl, obj);
588 lock = bucket_lock(tbl, hash);
592 pprev = &tbl->buckets[hash];
593 rht_for_each(he, tbl, hash) {
599 rcu_assign_pointer(*pprev, obj->next);
600 atomic_dec(&ht->nelems);
602 spin_unlock_bh(lock);
604 rhashtable_wakeup_worker(ht);
611 if (tbl != rht_dereference_rcu(ht->future_tbl, ht)) {
612 spin_unlock_bh(lock);
614 tbl = rht_dereference_rcu(ht->future_tbl, ht);
615 hash = head_hashfn(ht, tbl, obj);
617 lock = bucket_lock(tbl, hash);
622 spin_unlock_bh(lock);
627 EXPORT_SYMBOL_GPL(rhashtable_remove);
629 struct rhashtable_compare_arg {
630 struct rhashtable *ht;
634 static bool rhashtable_compare(void *ptr, void *arg)
636 struct rhashtable_compare_arg *x = arg;
637 struct rhashtable *ht = x->ht;
639 return !memcmp(ptr + ht->p.key_offset, x->key, ht->p.key_len);
643 * rhashtable_lookup - lookup key in hash table
645 * @key: pointer to key
647 * Computes the hash value for the key and traverses the bucket chain looking
648 * for a entry with an identical key. The first matching entry is returned.
650 * This lookup function may only be used for fixed key hash table (key_len
651 * parameter set). It will BUG() if used inappropriately.
653 * Lookups may occur in parallel with hashtable mutations and resizing.
655 void *rhashtable_lookup(struct rhashtable *ht, const void *key)
657 struct rhashtable_compare_arg arg = {
662 BUG_ON(!ht->p.key_len);
664 return rhashtable_lookup_compare(ht, key, &rhashtable_compare, &arg);
666 EXPORT_SYMBOL_GPL(rhashtable_lookup);
669 * rhashtable_lookup_compare - search hash table with compare function
671 * @key: the pointer to the key
672 * @compare: compare function, must return true on match
673 * @arg: argument passed on to compare function
675 * Traverses the bucket chain behind the provided hash value and calls the
676 * specified compare function for each entry.
678 * Lookups may occur in parallel with hashtable mutations and resizing.
680 * Returns the first entry on which the compare function returned true.
682 void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key,
683 bool (*compare)(void *, void *), void *arg)
685 const struct bucket_table *tbl, *old_tbl;
686 struct rhash_head *he;
691 old_tbl = rht_dereference_rcu(ht->tbl, ht);
692 tbl = rht_dereference_rcu(ht->future_tbl, ht);
693 hash = key_hashfn(ht, key, ht->p.key_len);
695 rht_for_each_rcu(he, tbl, rht_bucket_index(tbl, hash)) {
696 if (!compare(rht_obj(ht, he), arg))
699 return rht_obj(ht, he);
702 if (unlikely(tbl != old_tbl)) {
710 EXPORT_SYMBOL_GPL(rhashtable_lookup_compare);
713 * rhashtable_lookup_insert - lookup and insert object into hash table
715 * @obj: pointer to hash head inside object
717 * Locks down the bucket chain in both the old and new table if a resize
718 * is in progress to ensure that writers can't remove from the old table
719 * and can't insert to the new table during the atomic operation of search
720 * and insertion. Searches for duplicates in both the old and new table if
721 * a resize is in progress.
723 * This lookup function may only be used for fixed key hash table (key_len
724 * parameter set). It will BUG() if used inappropriately.
726 * It is safe to call this function from atomic context.
728 * Will trigger an automatic deferred table resizing if the size grows
729 * beyond the watermark indicated by grow_decision() which can be passed
730 * to rhashtable_init().
732 bool rhashtable_lookup_insert(struct rhashtable *ht, struct rhash_head *obj)
734 struct bucket_table *new_tbl, *old_tbl;
735 spinlock_t *new_bucket_lock, *old_bucket_lock;
736 u32 new_hash, old_hash;
739 BUG_ON(!ht->p.key_len);
743 old_tbl = rht_dereference_rcu(ht->tbl, ht);
744 old_hash = head_hashfn(ht, old_tbl, obj);
745 old_bucket_lock = bucket_lock(old_tbl, old_hash);
746 spin_lock_bh(old_bucket_lock);
748 new_tbl = rht_dereference_rcu(ht->future_tbl, ht);
749 new_hash = head_hashfn(ht, new_tbl, obj);
750 new_bucket_lock = bucket_lock(new_tbl, new_hash);
751 if (unlikely(old_tbl != new_tbl))
752 spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED);
754 if (rhashtable_lookup(ht, rht_obj(ht, obj) + ht->p.key_offset)) {
759 __rhashtable_insert(ht, obj, new_tbl, new_hash);
762 if (unlikely(old_tbl != new_tbl))
763 spin_unlock_bh(new_bucket_lock);
764 spin_unlock_bh(old_bucket_lock);
770 EXPORT_SYMBOL_GPL(rhashtable_lookup_insert);
772 static size_t rounded_hashtable_size(struct rhashtable_params *params)
774 return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
775 1UL << params->min_shift);
779 * rhashtable_init - initialize a new hash table
780 * @ht: hash table to be initialized
781 * @params: configuration parameters
783 * Initializes a new hash table based on the provided configuration
784 * parameters. A table can be configured either with a variable or
787 * Configuration Example 1: Fixed length keys
791 * struct rhash_head node;
794 * struct rhashtable_params params = {
795 * .head_offset = offsetof(struct test_obj, node),
796 * .key_offset = offsetof(struct test_obj, key),
797 * .key_len = sizeof(int),
799 * .nulls_base = (1U << RHT_BASE_SHIFT),
802 * Configuration Example 2: Variable length keys
805 * struct rhash_head node;
808 * u32 my_hash_fn(const void *data, u32 seed)
810 * struct test_obj *obj = data;
812 * return [... hash ...];
815 * struct rhashtable_params params = {
816 * .head_offset = offsetof(struct test_obj, node),
818 * .obj_hashfn = my_hash_fn,
821 int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params)
823 struct bucket_table *tbl;
826 size = HASH_DEFAULT_SIZE;
828 if ((params->key_len && !params->hashfn) ||
829 (!params->key_len && !params->obj_hashfn))
832 if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
835 params->min_shift = max_t(size_t, params->min_shift,
836 ilog2(HASH_MIN_SIZE));
838 if (params->nelem_hint)
839 size = rounded_hashtable_size(params);
841 memset(ht, 0, sizeof(*ht));
842 mutex_init(&ht->mutex);
843 memcpy(&ht->p, params, sizeof(*params));
845 if (params->locks_mul)
846 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
848 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
850 tbl = bucket_table_alloc(ht, size);
854 ht->shift = ilog2(tbl->size);
855 RCU_INIT_POINTER(ht->tbl, tbl);
856 RCU_INIT_POINTER(ht->future_tbl, tbl);
859 get_random_bytes(&ht->p.hash_rnd, sizeof(ht->p.hash_rnd));
861 if (ht->p.grow_decision || ht->p.shrink_decision)
862 INIT_DEFERRABLE_WORK(&ht->run_work, rht_deferred_worker);
866 EXPORT_SYMBOL_GPL(rhashtable_init);
869 * rhashtable_destroy - destroy hash table
870 * @ht: the hash table to destroy
872 * Frees the bucket array. This function is not rcu safe, therefore the caller
873 * has to make sure that no resizing may happen by unpublishing the hashtable
874 * and waiting for the quiescent cycle before releasing the bucket array.
876 void rhashtable_destroy(struct rhashtable *ht)
878 ht->being_destroyed = true;
880 mutex_lock(&ht->mutex);
882 cancel_delayed_work(&ht->run_work);
883 bucket_table_free(rht_dereference(ht->tbl, ht));
885 mutex_unlock(&ht->mutex);
887 EXPORT_SYMBOL_GPL(rhashtable_destroy);
889 /**************************************************************************
891 **************************************************************************/
893 #ifdef CONFIG_TEST_RHASHTABLE
895 #define TEST_HT_SIZE 8
896 #define TEST_ENTRIES 2048
897 #define TEST_PTR ((void *) 0xdeadbeef)
898 #define TEST_NEXPANDS 4
903 struct rhash_head node;
906 static int __init test_rht_lookup(struct rhashtable *ht)
910 for (i = 0; i < TEST_ENTRIES * 2; i++) {
911 struct test_obj *obj;
912 bool expected = !(i % 2);
915 obj = rhashtable_lookup(ht, &key);
917 if (expected && !obj) {
918 pr_warn("Test failed: Could not find key %u\n", key);
920 } else if (!expected && obj) {
921 pr_warn("Test failed: Unexpected entry found for key %u\n",
924 } else if (expected && obj) {
925 if (obj->ptr != TEST_PTR || obj->value != i) {
926 pr_warn("Test failed: Lookup value mismatch %p!=%p, %u!=%u\n",
927 obj->ptr, TEST_PTR, obj->value, i);
936 static void test_bucket_stats(struct rhashtable *ht, bool quiet)
938 unsigned int cnt, rcu_cnt, i, total = 0;
939 struct rhash_head *pos;
940 struct test_obj *obj;
941 struct bucket_table *tbl;
943 tbl = rht_dereference_rcu(ht->tbl, ht);
944 for (i = 0; i < tbl->size; i++) {
948 pr_info(" [%#4x/%zu]", i, tbl->size);
950 rht_for_each_entry_rcu(obj, pos, tbl, i, node) {
954 pr_cont(" [%p],", obj);
957 rht_for_each_entry_rcu(obj, pos, tbl, i, node)
961 pr_warn("Test failed: Chain count mismach %d != %d",
965 pr_cont("\n [%#x] first element: %p, chain length: %u\n",
966 i, tbl->buckets[i], cnt);
969 pr_info(" Traversal complete: counted=%u, nelems=%u, entries=%d\n",
970 total, atomic_read(&ht->nelems), TEST_ENTRIES);
972 if (total != atomic_read(&ht->nelems) || total != TEST_ENTRIES)
973 pr_warn("Test failed: Total count mismatch ^^^");
976 static int __init test_rhashtable(struct rhashtable *ht)
978 struct bucket_table *tbl;
979 struct test_obj *obj;
980 struct rhash_head *pos, *next;
986 * Insert TEST_ENTRIES into table with all keys even numbers
988 pr_info(" Adding %d keys\n", TEST_ENTRIES);
989 for (i = 0; i < TEST_ENTRIES; i++) {
990 struct test_obj *obj;
992 obj = kzalloc(sizeof(*obj), GFP_KERNEL);
1001 rhashtable_insert(ht, &obj->node);
1005 test_bucket_stats(ht, true);
1006 test_rht_lookup(ht);
1009 for (i = 0; i < TEST_NEXPANDS; i++) {
1010 pr_info(" Table expansion iteration %u...\n", i);
1011 mutex_lock(&ht->mutex);
1012 rhashtable_expand(ht);
1013 mutex_unlock(&ht->mutex);
1016 pr_info(" Verifying lookups...\n");
1017 test_rht_lookup(ht);
1021 for (i = 0; i < TEST_NEXPANDS; i++) {
1022 pr_info(" Table shrinkage iteration %u...\n", i);
1023 mutex_lock(&ht->mutex);
1024 rhashtable_shrink(ht);
1025 mutex_unlock(&ht->mutex);
1028 pr_info(" Verifying lookups...\n");
1029 test_rht_lookup(ht);
1034 test_bucket_stats(ht, true);
1037 pr_info(" Deleting %d keys\n", TEST_ENTRIES);
1038 for (i = 0; i < TEST_ENTRIES; i++) {
1041 obj = rhashtable_lookup(ht, &key);
1044 rhashtable_remove(ht, &obj->node);
1051 tbl = rht_dereference_rcu(ht->tbl, ht);
1052 for (i = 0; i < tbl->size; i++)
1053 rht_for_each_entry_safe(obj, pos, next, tbl, i, node)
1059 static int __init test_rht_init(void)
1061 struct rhashtable ht;
1062 struct rhashtable_params params = {
1063 .nelem_hint = TEST_HT_SIZE,
1064 .head_offset = offsetof(struct test_obj, node),
1065 .key_offset = offsetof(struct test_obj, value),
1066 .key_len = sizeof(int),
1068 .nulls_base = (3U << RHT_BASE_SHIFT),
1069 .grow_decision = rht_grow_above_75,
1070 .shrink_decision = rht_shrink_below_30,
1074 pr_info("Running resizable hashtable tests...\n");
1076 err = rhashtable_init(&ht, ¶ms);
1078 pr_warn("Test failed: Unable to initialize hashtable: %d\n",
1083 err = test_rhashtable(&ht);
1085 rhashtable_destroy(&ht);
1090 subsys_initcall(test_rht_init);
1092 #endif /* CONFIG_TEST_RHASHTABLE */