2 * Resizable, Scalable, Concurrent Hash Table
4 * Copyright (c) 2014-2015 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/sched.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
24 #include <linux/jhash.h>
25 #include <linux/random.h>
26 #include <linux/rhashtable.h>
27 #include <linux/err.h>
29 #define HASH_DEFAULT_SIZE 64UL
30 #define HASH_MIN_SIZE 4UL
31 #define BUCKET_LOCKS_PER_CPU 128UL
33 /* Base bits plus 1 bit for nulls marker */
34 #define HASH_RESERVED_SPACE (RHT_BASE_BITS + 1)
41 /* The bucket lock is selected based on the hash and protects mutations
42 * on a group of hash buckets.
44 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
45 * a single lock always covers both buckets which may both contains
46 * entries which link to the same bucket of the old table during resizing.
47 * This allows to simplify the locking as locking the bucket in both
48 * tables during resize always guarantee protection.
50 * IMPORTANT: When holding the bucket lock of both the old and new table
51 * during expansions and shrinking, the old bucket lock must always be
54 static spinlock_t
*bucket_lock(const struct bucket_table
*tbl
, u32 hash
)
56 return &tbl
->locks
[hash
& tbl
->locks_mask
];
59 static void *rht_obj(const struct rhashtable
*ht
, const struct rhash_head
*he
)
61 return (void *) he
- ht
->p
.head_offset
;
64 static u32
rht_bucket_index(const struct bucket_table
*tbl
, u32 hash
)
66 return (hash
>> HASH_RESERVED_SPACE
) & (tbl
->size
- 1);
69 static u32
key_hashfn(struct rhashtable
*ht
, const struct bucket_table
*tbl
,
72 return rht_bucket_index(tbl
, ht
->p
.hashfn(key
, ht
->p
.key_len
,
76 static u32
head_hashfn(struct rhashtable
*ht
,
77 const struct bucket_table
*tbl
,
78 const struct rhash_head
*he
)
80 const char *ptr
= rht_obj(ht
, he
);
82 return likely(ht
->p
.key_len
) ?
83 key_hashfn(ht
, tbl
, ptr
+ ht
->p
.key_offset
) :
84 rht_bucket_index(tbl
, ht
->p
.obj_hashfn(ptr
, tbl
->hash_rnd
));
87 #ifdef CONFIG_PROVE_LOCKING
88 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
90 int lockdep_rht_mutex_is_held(struct rhashtable
*ht
)
92 return (debug_locks
) ? lockdep_is_held(&ht
->mutex
) : 1;
94 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held
);
96 int lockdep_rht_bucket_is_held(const struct bucket_table
*tbl
, u32 hash
)
98 spinlock_t
*lock
= bucket_lock(tbl
, hash
);
100 return (debug_locks
) ? lockdep_is_held(lock
) : 1;
102 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held
);
104 #define ASSERT_RHT_MUTEX(HT)
108 static int alloc_bucket_locks(struct rhashtable
*ht
, struct bucket_table
*tbl
)
110 unsigned int i
, size
;
111 #if defined(CONFIG_PROVE_LOCKING)
112 unsigned int nr_pcpus
= 2;
114 unsigned int nr_pcpus
= num_possible_cpus();
117 nr_pcpus
= min_t(unsigned int, nr_pcpus
, 32UL);
118 size
= roundup_pow_of_two(nr_pcpus
* ht
->p
.locks_mul
);
120 /* Never allocate more than 0.5 locks per bucket */
121 size
= min_t(unsigned int, size
, tbl
->size
>> 1);
123 if (sizeof(spinlock_t
) != 0) {
125 if (size
* sizeof(spinlock_t
) > PAGE_SIZE
)
126 tbl
->locks
= vmalloc(size
* sizeof(spinlock_t
));
129 tbl
->locks
= kmalloc_array(size
, sizeof(spinlock_t
),
133 for (i
= 0; i
< size
; i
++)
134 spin_lock_init(&tbl
->locks
[i
]);
136 tbl
->locks_mask
= size
- 1;
141 static void bucket_table_free(const struct bucket_table
*tbl
)
149 static struct bucket_table
*bucket_table_alloc(struct rhashtable
*ht
,
150 size_t nbuckets
, u32 hash_rnd
)
152 struct bucket_table
*tbl
= NULL
;
156 size
= sizeof(*tbl
) + nbuckets
* sizeof(tbl
->buckets
[0]);
157 if (size
<= (PAGE_SIZE
<< PAGE_ALLOC_COSTLY_ORDER
))
158 tbl
= kzalloc(size
, GFP_KERNEL
| __GFP_NOWARN
| __GFP_NORETRY
);
164 tbl
->size
= nbuckets
;
165 tbl
->shift
= ilog2(nbuckets
);
166 tbl
->hash_rnd
= hash_rnd
;
168 if (alloc_bucket_locks(ht
, tbl
) < 0) {
169 bucket_table_free(tbl
);
173 for (i
= 0; i
< nbuckets
; i
++)
174 INIT_RHT_NULLS_HEAD(tbl
->buckets
[i
], ht
, i
);
180 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
182 * @tbl: current table
184 static bool rht_grow_above_75(const struct rhashtable
*ht
,
185 const struct bucket_table
*tbl
)
187 /* Expand table when exceeding 75% load */
188 return atomic_read(&ht
->nelems
) > (tbl
->size
/ 4 * 3) &&
189 (!ht
->p
.max_shift
|| tbl
->shift
< ht
->p
.max_shift
);
193 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
195 * @tbl: current table
197 static bool rht_shrink_below_30(const struct rhashtable
*ht
,
198 const struct bucket_table
*tbl
)
200 /* Shrink table beneath 30% load */
201 return atomic_read(&ht
->nelems
) < (tbl
->size
* 3 / 10) &&
202 tbl
->shift
> ht
->p
.min_shift
;
205 static int rhashtable_rehash_one(struct rhashtable
*ht
, unsigned old_hash
)
207 struct bucket_table
*new_tbl
= rht_dereference(ht
->future_tbl
, ht
);
208 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
209 struct rhash_head __rcu
**pprev
= &old_tbl
->buckets
[old_hash
];
211 struct rhash_head
*head
, *next
, *entry
;
212 spinlock_t
*new_bucket_lock
;
215 rht_for_each(entry
, old_tbl
, old_hash
) {
217 next
= rht_dereference_bucket(entry
->next
, old_tbl
, old_hash
);
219 if (rht_is_a_nulls(next
))
222 pprev
= &entry
->next
;
228 new_hash
= head_hashfn(ht
, new_tbl
, entry
);
230 new_bucket_lock
= bucket_lock(new_tbl
, new_hash
);
232 spin_lock_nested(new_bucket_lock
, RHT_LOCK_NESTED
);
233 head
= rht_dereference_bucket(new_tbl
->buckets
[new_hash
],
236 if (rht_is_a_nulls(head
))
237 INIT_RHT_NULLS_HEAD(entry
->next
, ht
, new_hash
);
239 RCU_INIT_POINTER(entry
->next
, head
);
241 rcu_assign_pointer(new_tbl
->buckets
[new_hash
], entry
);
242 spin_unlock(new_bucket_lock
);
244 rcu_assign_pointer(*pprev
, next
);
250 static void rhashtable_rehash_chain(struct rhashtable
*ht
, unsigned old_hash
)
252 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
253 spinlock_t
*old_bucket_lock
;
255 old_bucket_lock
= bucket_lock(old_tbl
, old_hash
);
257 spin_lock_bh(old_bucket_lock
);
258 while (!rhashtable_rehash_one(ht
, old_hash
))
260 spin_unlock_bh(old_bucket_lock
);
263 static void rhashtable_rehash(struct rhashtable
*ht
,
264 struct bucket_table
*new_tbl
)
266 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
269 get_random_bytes(&new_tbl
->hash_rnd
, sizeof(new_tbl
->hash_rnd
));
271 /* Make insertions go into the new, empty table right away. Deletions
272 * and lookups will be attempted in both tables until we synchronize.
273 * The synchronize_rcu() guarantees for the new table to be picked up
274 * so no new additions go into the old table while we relink.
276 rcu_assign_pointer(ht
->future_tbl
, new_tbl
);
278 /* Ensure the new table is visible to readers. */
281 for (old_hash
= 0; old_hash
< old_tbl
->size
; old_hash
++)
282 rhashtable_rehash_chain(ht
, old_hash
);
284 /* Publish the new table pointer. */
285 rcu_assign_pointer(ht
->tbl
, new_tbl
);
287 /* Wait for readers. All new readers will see the new
288 * table, and thus no references to the old table will
293 bucket_table_free(old_tbl
);
297 * rhashtable_expand - Expand hash table while allowing concurrent lookups
298 * @ht: the hash table to expand
300 * A secondary bucket array is allocated and the hash entries are migrated.
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
);
315 ASSERT_RHT_MUTEX(ht
);
317 new_tbl
= bucket_table_alloc(ht
, old_tbl
->size
* 2, old_tbl
->hash_rnd
);
321 rhashtable_rehash(ht
, new_tbl
);
324 EXPORT_SYMBOL_GPL(rhashtable_expand
);
327 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
328 * @ht: the hash table to shrink
330 * This function may only be called in a context where it is safe to call
331 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
333 * The caller must ensure that no concurrent resizing occurs by holding
336 * The caller must ensure that no concurrent table mutations take place.
337 * It is however valid to have concurrent lookups if they are RCU protected.
339 * It is valid to have concurrent insertions and deletions protected by per
340 * bucket locks or concurrent RCU protected lookups and traversals.
342 int rhashtable_shrink(struct rhashtable
*ht
)
344 struct bucket_table
*new_tbl
, *old_tbl
= rht_dereference(ht
->tbl
, ht
);
346 ASSERT_RHT_MUTEX(ht
);
348 new_tbl
= bucket_table_alloc(ht
, old_tbl
->size
/ 2, old_tbl
->hash_rnd
);
352 rhashtable_rehash(ht
, new_tbl
);
355 EXPORT_SYMBOL_GPL(rhashtable_shrink
);
357 static void rht_deferred_worker(struct work_struct
*work
)
359 struct rhashtable
*ht
;
360 struct bucket_table
*tbl
;
361 struct rhashtable_walker
*walker
;
363 ht
= container_of(work
, struct rhashtable
, run_work
);
364 mutex_lock(&ht
->mutex
);
365 if (ht
->being_destroyed
)
368 tbl
= rht_dereference(ht
->tbl
, ht
);
370 list_for_each_entry(walker
, &ht
->walkers
, list
)
371 walker
->resize
= true;
373 if (rht_grow_above_75(ht
, tbl
))
374 rhashtable_expand(ht
);
375 else if (rht_shrink_below_30(ht
, tbl
))
376 rhashtable_shrink(ht
);
378 mutex_unlock(&ht
->mutex
);
381 static bool __rhashtable_insert(struct rhashtable
*ht
, struct rhash_head
*obj
,
382 bool (*compare
)(void *, void *), void *arg
)
384 struct bucket_table
*tbl
, *old_tbl
;
385 struct rhash_head
*head
;
386 bool no_resize_running
;
392 old_tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
393 hash
= head_hashfn(ht
, old_tbl
, obj
);
395 spin_lock_bh(bucket_lock(old_tbl
, hash
));
397 /* Because we have already taken the bucket lock in old_tbl,
398 * if we find that future_tbl is not yet visible then that
399 * guarantees all other insertions of the same entry will
400 * also grab the bucket lock in old_tbl because until the
401 * rehash completes ht->tbl won't be changed.
403 tbl
= rht_dereference_rcu(ht
->future_tbl
, ht
);
404 if (tbl
!= old_tbl
) {
405 hash
= head_hashfn(ht
, tbl
, obj
);
406 spin_lock_nested(bucket_lock(tbl
, hash
), RHT_LOCK_NESTED
);
410 rhashtable_lookup_compare(ht
, rht_obj(ht
, obj
) + ht
->p
.key_offset
,
416 no_resize_running
= tbl
== old_tbl
;
418 head
= rht_dereference_bucket(tbl
->buckets
[hash
], tbl
, hash
);
420 if (rht_is_a_nulls(head
))
421 INIT_RHT_NULLS_HEAD(obj
->next
, ht
, hash
);
423 RCU_INIT_POINTER(obj
->next
, head
);
425 rcu_assign_pointer(tbl
->buckets
[hash
], obj
);
427 atomic_inc(&ht
->nelems
);
428 if (no_resize_running
&& rht_grow_above_75(ht
, tbl
))
429 schedule_work(&ht
->run_work
);
432 if (tbl
!= old_tbl
) {
433 hash
= head_hashfn(ht
, tbl
, obj
);
434 spin_unlock(bucket_lock(tbl
, hash
));
437 hash
= head_hashfn(ht
, old_tbl
, obj
);
438 spin_unlock_bh(bucket_lock(old_tbl
, hash
));
446 * rhashtable_insert - insert object into hash table
448 * @obj: pointer to hash head inside object
450 * Will take a per bucket spinlock to protect against mutual mutations
451 * on the same bucket. Multiple insertions may occur in parallel unless
452 * they map to the same bucket lock.
454 * It is safe to call this function from atomic context.
456 * Will trigger an automatic deferred table resizing if the size grows
457 * beyond the watermark indicated by grow_decision() which can be passed
458 * to rhashtable_init().
460 void rhashtable_insert(struct rhashtable
*ht
, struct rhash_head
*obj
)
462 __rhashtable_insert(ht
, obj
, NULL
, NULL
);
464 EXPORT_SYMBOL_GPL(rhashtable_insert
);
466 static bool __rhashtable_remove(struct rhashtable
*ht
,
467 struct bucket_table
*tbl
,
468 struct rhash_head
*obj
)
470 struct rhash_head __rcu
**pprev
;
471 struct rhash_head
*he
;
476 hash
= head_hashfn(ht
, tbl
, obj
);
477 lock
= bucket_lock(tbl
, hash
);
481 pprev
= &tbl
->buckets
[hash
];
482 rht_for_each(he
, tbl
, hash
) {
488 rcu_assign_pointer(*pprev
, obj
->next
);
493 spin_unlock_bh(lock
);
499 * rhashtable_remove - remove object from hash table
501 * @obj: pointer to hash head inside object
503 * Since the hash chain is single linked, the removal operation needs to
504 * walk the bucket chain upon removal. The removal operation is thus
505 * considerable slow if the hash table is not correctly sized.
507 * Will automatically shrink the table via rhashtable_expand() if the
508 * shrink_decision function specified at rhashtable_init() returns true.
510 * The caller must ensure that no concurrent table mutations occur. It is
511 * however valid to have concurrent lookups if they are RCU protected.
513 bool rhashtable_remove(struct rhashtable
*ht
, struct rhash_head
*obj
)
515 struct bucket_table
*tbl
, *old_tbl
;
520 old_tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
521 ret
= __rhashtable_remove(ht
, old_tbl
, obj
);
523 /* Because we have already taken (and released) the bucket
524 * lock in old_tbl, if we find that future_tbl is not yet
525 * visible then that guarantees the entry to still be in
526 * old_tbl if it exists.
528 tbl
= rht_dereference_rcu(ht
->future_tbl
, ht
);
529 if (!ret
&& old_tbl
!= tbl
)
530 ret
= __rhashtable_remove(ht
, tbl
, obj
);
533 bool no_resize_running
= tbl
== old_tbl
;
535 atomic_dec(&ht
->nelems
);
536 if (no_resize_running
&& rht_shrink_below_30(ht
, tbl
))
537 schedule_work(&ht
->run_work
);
544 EXPORT_SYMBOL_GPL(rhashtable_remove
);
546 struct rhashtable_compare_arg
{
547 struct rhashtable
*ht
;
551 static bool rhashtable_compare(void *ptr
, void *arg
)
553 struct rhashtable_compare_arg
*x
= arg
;
554 struct rhashtable
*ht
= x
->ht
;
556 return !memcmp(ptr
+ ht
->p
.key_offset
, x
->key
, ht
->p
.key_len
);
560 * rhashtable_lookup - lookup key in hash table
562 * @key: pointer to key
564 * Computes the hash value for the key and traverses the bucket chain looking
565 * for a entry with an identical key. The first matching entry is returned.
567 * This lookup function may only be used for fixed key hash table (key_len
568 * parameter set). It will BUG() if used inappropriately.
570 * Lookups may occur in parallel with hashtable mutations and resizing.
572 void *rhashtable_lookup(struct rhashtable
*ht
, const void *key
)
574 struct rhashtable_compare_arg arg
= {
579 BUG_ON(!ht
->p
.key_len
);
581 return rhashtable_lookup_compare(ht
, key
, &rhashtable_compare
, &arg
);
583 EXPORT_SYMBOL_GPL(rhashtable_lookup
);
586 * rhashtable_lookup_compare - search hash table with compare function
588 * @key: the pointer to the key
589 * @compare: compare function, must return true on match
590 * @arg: argument passed on to compare function
592 * Traverses the bucket chain behind the provided hash value and calls the
593 * specified compare function for each entry.
595 * Lookups may occur in parallel with hashtable mutations and resizing.
597 * Returns the first entry on which the compare function returned true.
599 void *rhashtable_lookup_compare(struct rhashtable
*ht
, const void *key
,
600 bool (*compare
)(void *, void *), void *arg
)
602 const struct bucket_table
*tbl
, *old_tbl
;
603 struct rhash_head
*he
;
608 tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
609 hash
= key_hashfn(ht
, tbl
, key
);
611 rht_for_each_rcu(he
, tbl
, hash
) {
612 if (!compare(rht_obj(ht
, he
), arg
))
615 return rht_obj(ht
, he
);
618 /* Ensure we see any new tables. */
622 tbl
= rht_dereference_rcu(ht
->future_tbl
, ht
);
623 if (unlikely(tbl
!= old_tbl
))
629 EXPORT_SYMBOL_GPL(rhashtable_lookup_compare
);
632 * rhashtable_lookup_insert - lookup and insert object into hash table
634 * @obj: pointer to hash head inside object
636 * Locks down the bucket chain in both the old and new table if a resize
637 * is in progress to ensure that writers can't remove from the old table
638 * and can't insert to the new table during the atomic operation of search
639 * and insertion. Searches for duplicates in both the old and new table if
640 * a resize is in progress.
642 * This lookup function may only be used for fixed key hash table (key_len
643 * parameter set). It will BUG() if used inappropriately.
645 * It is safe to call this function from atomic context.
647 * Will trigger an automatic deferred table resizing if the size grows
648 * beyond the watermark indicated by grow_decision() which can be passed
649 * to rhashtable_init().
651 bool rhashtable_lookup_insert(struct rhashtable
*ht
, struct rhash_head
*obj
)
653 struct rhashtable_compare_arg arg
= {
655 .key
= rht_obj(ht
, obj
) + ht
->p
.key_offset
,
658 BUG_ON(!ht
->p
.key_len
);
660 return rhashtable_lookup_compare_insert(ht
, obj
, &rhashtable_compare
,
663 EXPORT_SYMBOL_GPL(rhashtable_lookup_insert
);
666 * rhashtable_lookup_compare_insert - search and insert object to hash table
667 * with compare function
669 * @obj: pointer to hash head inside object
670 * @compare: compare function, must return true on match
671 * @arg: argument passed on to compare function
673 * Locks down the bucket chain in both the old and new table if a resize
674 * is in progress to ensure that writers can't remove from the old table
675 * and can't insert to the new table during the atomic operation of search
676 * and insertion. Searches for duplicates in both the old and new table if
677 * a resize is in progress.
679 * Lookups may occur in parallel with hashtable mutations and resizing.
681 * Will trigger an automatic deferred table resizing if the size grows
682 * beyond the watermark indicated by grow_decision() which can be passed
683 * to rhashtable_init().
685 bool rhashtable_lookup_compare_insert(struct rhashtable
*ht
,
686 struct rhash_head
*obj
,
687 bool (*compare
)(void *, void *),
690 BUG_ON(!ht
->p
.key_len
);
692 return __rhashtable_insert(ht
, obj
, compare
, arg
);
694 EXPORT_SYMBOL_GPL(rhashtable_lookup_compare_insert
);
697 * rhashtable_walk_init - Initialise an iterator
698 * @ht: Table to walk over
699 * @iter: Hash table Iterator
701 * This function prepares a hash table walk.
703 * Note that if you restart a walk after rhashtable_walk_stop you
704 * may see the same object twice. Also, you may miss objects if
705 * there are removals in between rhashtable_walk_stop and the next
706 * call to rhashtable_walk_start.
708 * For a completely stable walk you should construct your own data
709 * structure outside the hash table.
711 * This function may sleep so you must not call it from interrupt
712 * context or with spin locks held.
714 * You must call rhashtable_walk_exit if this function returns
717 int rhashtable_walk_init(struct rhashtable
*ht
, struct rhashtable_iter
*iter
)
724 iter
->walker
= kmalloc(sizeof(*iter
->walker
), GFP_KERNEL
);
728 INIT_LIST_HEAD(&iter
->walker
->list
);
729 iter
->walker
->resize
= false;
731 mutex_lock(&ht
->mutex
);
732 list_add(&iter
->walker
->list
, &ht
->walkers
);
733 mutex_unlock(&ht
->mutex
);
737 EXPORT_SYMBOL_GPL(rhashtable_walk_init
);
740 * rhashtable_walk_exit - Free an iterator
741 * @iter: Hash table Iterator
743 * This function frees resources allocated by rhashtable_walk_init.
745 void rhashtable_walk_exit(struct rhashtable_iter
*iter
)
747 mutex_lock(&iter
->ht
->mutex
);
748 list_del(&iter
->walker
->list
);
749 mutex_unlock(&iter
->ht
->mutex
);
752 EXPORT_SYMBOL_GPL(rhashtable_walk_exit
);
755 * rhashtable_walk_start - Start a hash table walk
756 * @iter: Hash table iterator
758 * Start a hash table walk. Note that we take the RCU lock in all
759 * cases including when we return an error. So you must always call
760 * rhashtable_walk_stop to clean up.
762 * Returns zero if successful.
764 * Returns -EAGAIN if resize event occured. Note that the iterator
765 * will rewind back to the beginning and you may use it immediately
766 * by calling rhashtable_walk_next.
768 int rhashtable_walk_start(struct rhashtable_iter
*iter
)
772 if (iter
->walker
->resize
) {
775 iter
->walker
->resize
= false;
781 EXPORT_SYMBOL_GPL(rhashtable_walk_start
);
784 * rhashtable_walk_next - Return the next object and advance the iterator
785 * @iter: Hash table iterator
787 * Note that you must call rhashtable_walk_stop when you are finished
790 * Returns the next object or NULL when the end of the table is reached.
792 * Returns -EAGAIN if resize event occured. Note that the iterator
793 * will rewind back to the beginning and you may continue to use it.
795 void *rhashtable_walk_next(struct rhashtable_iter
*iter
)
797 const struct bucket_table
*tbl
;
798 struct rhashtable
*ht
= iter
->ht
;
799 struct rhash_head
*p
= iter
->p
;
802 tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
805 p
= rht_dereference_bucket_rcu(p
->next
, tbl
, iter
->slot
);
809 for (; iter
->slot
< tbl
->size
; iter
->slot
++) {
810 int skip
= iter
->skip
;
812 rht_for_each_rcu(p
, tbl
, iter
->slot
) {
819 if (!rht_is_a_nulls(p
)) {
822 obj
= rht_obj(ht
, p
);
832 if (iter
->walker
->resize
) {
836 iter
->walker
->resize
= false;
837 return ERR_PTR(-EAGAIN
);
842 EXPORT_SYMBOL_GPL(rhashtable_walk_next
);
845 * rhashtable_walk_stop - Finish a hash table walk
846 * @iter: Hash table iterator
848 * Finish a hash table walk.
850 void rhashtable_walk_stop(struct rhashtable_iter
*iter
)
855 EXPORT_SYMBOL_GPL(rhashtable_walk_stop
);
857 static size_t rounded_hashtable_size(struct rhashtable_params
*params
)
859 return max(roundup_pow_of_two(params
->nelem_hint
* 4 / 3),
860 1UL << params
->min_shift
);
864 * rhashtable_init - initialize a new hash table
865 * @ht: hash table to be initialized
866 * @params: configuration parameters
868 * Initializes a new hash table based on the provided configuration
869 * parameters. A table can be configured either with a variable or
872 * Configuration Example 1: Fixed length keys
876 * struct rhash_head node;
879 * struct rhashtable_params params = {
880 * .head_offset = offsetof(struct test_obj, node),
881 * .key_offset = offsetof(struct test_obj, key),
882 * .key_len = sizeof(int),
884 * .nulls_base = (1U << RHT_BASE_SHIFT),
887 * Configuration Example 2: Variable length keys
890 * struct rhash_head node;
893 * u32 my_hash_fn(const void *data, u32 seed)
895 * struct test_obj *obj = data;
897 * return [... hash ...];
900 * struct rhashtable_params params = {
901 * .head_offset = offsetof(struct test_obj, node),
903 * .obj_hashfn = my_hash_fn,
906 int rhashtable_init(struct rhashtable
*ht
, struct rhashtable_params
*params
)
908 struct bucket_table
*tbl
;
912 size
= HASH_DEFAULT_SIZE
;
914 if ((params
->key_len
&& !params
->hashfn
) ||
915 (!params
->key_len
&& !params
->obj_hashfn
))
918 if (params
->nulls_base
&& params
->nulls_base
< (1U << RHT_BASE_SHIFT
))
921 params
->min_shift
= max_t(size_t, params
->min_shift
,
922 ilog2(HASH_MIN_SIZE
));
924 if (params
->nelem_hint
)
925 size
= rounded_hashtable_size(params
);
927 memset(ht
, 0, sizeof(*ht
));
928 mutex_init(&ht
->mutex
);
929 memcpy(&ht
->p
, params
, sizeof(*params
));
930 INIT_LIST_HEAD(&ht
->walkers
);
932 if (params
->locks_mul
)
933 ht
->p
.locks_mul
= roundup_pow_of_two(params
->locks_mul
);
935 ht
->p
.locks_mul
= BUCKET_LOCKS_PER_CPU
;
937 get_random_bytes(&hash_rnd
, sizeof(hash_rnd
));
939 tbl
= bucket_table_alloc(ht
, size
, hash_rnd
);
943 atomic_set(&ht
->nelems
, 0);
945 RCU_INIT_POINTER(ht
->tbl
, tbl
);
946 RCU_INIT_POINTER(ht
->future_tbl
, tbl
);
948 INIT_WORK(&ht
->run_work
, rht_deferred_worker
);
952 EXPORT_SYMBOL_GPL(rhashtable_init
);
955 * rhashtable_destroy - destroy hash table
956 * @ht: the hash table to destroy
958 * Frees the bucket array. This function is not rcu safe, therefore the caller
959 * has to make sure that no resizing may happen by unpublishing the hashtable
960 * and waiting for the quiescent cycle before releasing the bucket array.
962 void rhashtable_destroy(struct rhashtable
*ht
)
964 ht
->being_destroyed
= true;
966 cancel_work_sync(&ht
->run_work
);
968 mutex_lock(&ht
->mutex
);
969 bucket_table_free(rht_dereference(ht
->tbl
, ht
));
970 mutex_unlock(&ht
->mutex
);
972 EXPORT_SYMBOL_GPL(rhashtable_destroy
);