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/rculist.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
26 #include <linux/jhash.h>
27 #include <linux/random.h>
28 #include <linux/rhashtable.h>
29 #include <linux/err.h>
30 #include <linux/export.h>
32 #define HASH_DEFAULT_SIZE 64UL
33 #define HASH_MIN_SIZE 4U
34 #define BUCKET_LOCKS_PER_CPU 32UL
37 union nested_table __rcu
*table
;
38 struct rhash_head __rcu
*bucket
;
41 static u32
head_hashfn(struct rhashtable
*ht
,
42 const struct bucket_table
*tbl
,
43 const struct rhash_head
*he
)
45 return rht_head_hashfn(ht
, tbl
, he
, ht
->p
);
48 #ifdef CONFIG_PROVE_LOCKING
49 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
51 int lockdep_rht_mutex_is_held(struct rhashtable
*ht
)
53 return (debug_locks
) ? lockdep_is_held(&ht
->mutex
) : 1;
55 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held
);
57 int lockdep_rht_bucket_is_held(const struct bucket_table
*tbl
, u32 hash
)
59 spinlock_t
*lock
= rht_bucket_lock(tbl
, hash
);
61 return (debug_locks
) ? lockdep_is_held(lock
) : 1;
63 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held
);
65 #define ASSERT_RHT_MUTEX(HT)
69 static int alloc_bucket_locks(struct rhashtable
*ht
, struct bucket_table
*tbl
,
73 #if defined(CONFIG_PROVE_LOCKING)
74 unsigned int nr_pcpus
= 2;
76 unsigned int nr_pcpus
= num_possible_cpus();
79 nr_pcpus
= min_t(unsigned int, nr_pcpus
, 64UL);
80 size
= roundup_pow_of_two(nr_pcpus
* ht
->p
.locks_mul
);
82 /* Never allocate more than 0.5 locks per bucket */
83 size
= min_t(unsigned int, size
, tbl
->size
>> 1);
86 size
= min(size
, 1U << tbl
->nest
);
88 if (sizeof(spinlock_t
) != 0) {
89 if (gfpflags_allow_blocking(gfp
))
90 tbl
->locks
= kvmalloc(size
* sizeof(spinlock_t
), gfp
);
92 tbl
->locks
= kmalloc_array(size
, sizeof(spinlock_t
),
96 for (i
= 0; i
< size
; i
++)
97 spin_lock_init(&tbl
->locks
[i
]);
99 tbl
->locks_mask
= size
- 1;
104 static void nested_table_free(union nested_table
*ntbl
, unsigned int size
)
106 const unsigned int shift
= PAGE_SHIFT
- ilog2(sizeof(void *));
107 const unsigned int len
= 1 << shift
;
110 ntbl
= rcu_dereference_raw(ntbl
->table
);
116 for (i
= 0; i
< len
; i
++)
117 nested_table_free(ntbl
+ i
, size
);
123 static void nested_bucket_table_free(const struct bucket_table
*tbl
)
125 unsigned int size
= tbl
->size
>> tbl
->nest
;
126 unsigned int len
= 1 << tbl
->nest
;
127 union nested_table
*ntbl
;
130 ntbl
= (union nested_table
*)rcu_dereference_raw(tbl
->buckets
[0]);
132 for (i
= 0; i
< len
; i
++)
133 nested_table_free(ntbl
+ i
, size
);
138 static void bucket_table_free(const struct bucket_table
*tbl
)
141 nested_bucket_table_free(tbl
);
147 static void bucket_table_free_rcu(struct rcu_head
*head
)
149 bucket_table_free(container_of(head
, struct bucket_table
, rcu
));
152 static union nested_table
*nested_table_alloc(struct rhashtable
*ht
,
153 union nested_table __rcu
**prev
,
154 unsigned int shifted
,
157 union nested_table
*ntbl
;
160 ntbl
= rcu_dereference(*prev
);
164 ntbl
= kzalloc(PAGE_SIZE
, GFP_ATOMIC
);
166 if (ntbl
&& shifted
) {
167 for (i
= 0; i
< PAGE_SIZE
/ sizeof(ntbl
[0].bucket
); i
++)
168 INIT_RHT_NULLS_HEAD(ntbl
[i
].bucket
, ht
,
169 (i
<< shifted
) | nhash
);
172 rcu_assign_pointer(*prev
, ntbl
);
177 static struct bucket_table
*nested_bucket_table_alloc(struct rhashtable
*ht
,
181 const unsigned int shift
= PAGE_SHIFT
- ilog2(sizeof(void *));
182 struct bucket_table
*tbl
;
185 if (nbuckets
< (1 << (shift
+ 1)))
188 size
= sizeof(*tbl
) + sizeof(tbl
->buckets
[0]);
190 tbl
= kzalloc(size
, gfp
);
194 if (!nested_table_alloc(ht
, (union nested_table __rcu
**)tbl
->buckets
,
200 tbl
->nest
= (ilog2(nbuckets
) - 1) % shift
+ 1;
205 static struct bucket_table
*bucket_table_alloc(struct rhashtable
*ht
,
209 struct bucket_table
*tbl
= NULL
;
213 size
= sizeof(*tbl
) + nbuckets
* sizeof(tbl
->buckets
[0]);
214 if (gfp
!= GFP_KERNEL
)
215 tbl
= kzalloc(size
, gfp
| __GFP_NOWARN
| __GFP_NORETRY
);
217 tbl
= kvzalloc(size
, gfp
);
221 if (tbl
== NULL
&& gfp
!= GFP_KERNEL
) {
222 tbl
= nested_bucket_table_alloc(ht
, nbuckets
, gfp
);
230 if (alloc_bucket_locks(ht
, tbl
, gfp
) < 0) {
231 bucket_table_free(tbl
);
235 INIT_LIST_HEAD(&tbl
->walkers
);
237 tbl
->hash_rnd
= get_random_u32();
239 for (i
= 0; i
< nbuckets
; i
++)
240 INIT_RHT_NULLS_HEAD(tbl
->buckets
[i
], ht
, i
);
245 static struct bucket_table
*rhashtable_last_table(struct rhashtable
*ht
,
246 struct bucket_table
*tbl
)
248 struct bucket_table
*new_tbl
;
252 tbl
= rht_dereference_rcu(tbl
->future_tbl
, ht
);
258 static int rhashtable_rehash_one(struct rhashtable
*ht
, unsigned int old_hash
)
260 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
261 struct bucket_table
*new_tbl
= rhashtable_last_table(ht
,
262 rht_dereference_rcu(old_tbl
->future_tbl
, ht
));
263 struct rhash_head __rcu
**pprev
= rht_bucket_var(old_tbl
, old_hash
);
265 struct rhash_head
*head
, *next
, *entry
;
266 spinlock_t
*new_bucket_lock
;
267 unsigned int new_hash
;
274 rht_for_each(entry
, old_tbl
, old_hash
) {
276 next
= rht_dereference_bucket(entry
->next
, old_tbl
, old_hash
);
278 if (rht_is_a_nulls(next
))
281 pprev
= &entry
->next
;
287 new_hash
= head_hashfn(ht
, new_tbl
, entry
);
289 new_bucket_lock
= rht_bucket_lock(new_tbl
, new_hash
);
291 spin_lock_nested(new_bucket_lock
, SINGLE_DEPTH_NESTING
);
292 head
= rht_dereference_bucket(new_tbl
->buckets
[new_hash
],
295 RCU_INIT_POINTER(entry
->next
, head
);
297 rcu_assign_pointer(new_tbl
->buckets
[new_hash
], entry
);
298 spin_unlock(new_bucket_lock
);
300 rcu_assign_pointer(*pprev
, next
);
306 static int rhashtable_rehash_chain(struct rhashtable
*ht
,
307 unsigned int old_hash
)
309 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
310 spinlock_t
*old_bucket_lock
;
313 old_bucket_lock
= rht_bucket_lock(old_tbl
, old_hash
);
315 spin_lock_bh(old_bucket_lock
);
316 while (!(err
= rhashtable_rehash_one(ht
, old_hash
)))
319 if (err
== -ENOENT
) {
323 spin_unlock_bh(old_bucket_lock
);
328 static int rhashtable_rehash_attach(struct rhashtable
*ht
,
329 struct bucket_table
*old_tbl
,
330 struct bucket_table
*new_tbl
)
332 /* Protect future_tbl using the first bucket lock. */
333 spin_lock_bh(old_tbl
->locks
);
335 /* Did somebody beat us to it? */
336 if (rcu_access_pointer(old_tbl
->future_tbl
)) {
337 spin_unlock_bh(old_tbl
->locks
);
341 /* Make insertions go into the new, empty table right away. Deletions
342 * and lookups will be attempted in both tables until we synchronize.
344 rcu_assign_pointer(old_tbl
->future_tbl
, new_tbl
);
346 spin_unlock_bh(old_tbl
->locks
);
351 static int rhashtable_rehash_table(struct rhashtable
*ht
)
353 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
354 struct bucket_table
*new_tbl
;
355 struct rhashtable_walker
*walker
;
356 unsigned int old_hash
;
359 new_tbl
= rht_dereference(old_tbl
->future_tbl
, ht
);
363 for (old_hash
= 0; old_hash
< old_tbl
->size
; old_hash
++) {
364 err
= rhashtable_rehash_chain(ht
, old_hash
);
369 /* Publish the new table pointer. */
370 rcu_assign_pointer(ht
->tbl
, new_tbl
);
372 spin_lock(&ht
->lock
);
373 list_for_each_entry(walker
, &old_tbl
->walkers
, list
)
375 spin_unlock(&ht
->lock
);
377 /* Wait for readers. All new readers will see the new
378 * table, and thus no references to the old table will
381 call_rcu(&old_tbl
->rcu
, bucket_table_free_rcu
);
383 return rht_dereference(new_tbl
->future_tbl
, ht
) ? -EAGAIN
: 0;
386 static int rhashtable_rehash_alloc(struct rhashtable
*ht
,
387 struct bucket_table
*old_tbl
,
390 struct bucket_table
*new_tbl
;
393 ASSERT_RHT_MUTEX(ht
);
395 new_tbl
= bucket_table_alloc(ht
, size
, GFP_KERNEL
);
399 err
= rhashtable_rehash_attach(ht
, old_tbl
, new_tbl
);
401 bucket_table_free(new_tbl
);
407 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
408 * @ht: the hash table to shrink
410 * This function shrinks the hash table to fit, i.e., the smallest
411 * size would not cause it to expand right away automatically.
413 * The caller must ensure that no concurrent resizing occurs by holding
416 * The caller must ensure that no concurrent table mutations take place.
417 * It is however valid to have concurrent lookups if they are RCU protected.
419 * It is valid to have concurrent insertions and deletions protected by per
420 * bucket locks or concurrent RCU protected lookups and traversals.
422 static int rhashtable_shrink(struct rhashtable
*ht
)
424 struct bucket_table
*old_tbl
= rht_dereference(ht
->tbl
, ht
);
425 unsigned int nelems
= atomic_read(&ht
->nelems
);
426 unsigned int size
= 0;
429 size
= roundup_pow_of_two(nelems
* 3 / 2);
430 if (size
< ht
->p
.min_size
)
431 size
= ht
->p
.min_size
;
433 if (old_tbl
->size
<= size
)
436 if (rht_dereference(old_tbl
->future_tbl
, ht
))
439 return rhashtable_rehash_alloc(ht
, old_tbl
, size
);
442 static void rht_deferred_worker(struct work_struct
*work
)
444 struct rhashtable
*ht
;
445 struct bucket_table
*tbl
;
448 ht
= container_of(work
, struct rhashtable
, run_work
);
449 mutex_lock(&ht
->mutex
);
451 tbl
= rht_dereference(ht
->tbl
, ht
);
452 tbl
= rhashtable_last_table(ht
, tbl
);
454 if (rht_grow_above_75(ht
, tbl
))
455 err
= rhashtable_rehash_alloc(ht
, tbl
, tbl
->size
* 2);
456 else if (ht
->p
.automatic_shrinking
&& rht_shrink_below_30(ht
, tbl
))
457 err
= rhashtable_shrink(ht
);
459 err
= rhashtable_rehash_alloc(ht
, tbl
, tbl
->size
);
462 err
= rhashtable_rehash_table(ht
);
464 mutex_unlock(&ht
->mutex
);
467 schedule_work(&ht
->run_work
);
470 static int rhashtable_insert_rehash(struct rhashtable
*ht
,
471 struct bucket_table
*tbl
)
473 struct bucket_table
*old_tbl
;
474 struct bucket_table
*new_tbl
;
478 old_tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
484 if (rht_grow_above_75(ht
, tbl
))
486 /* Do not schedule more than one rehash */
487 else if (old_tbl
!= tbl
)
492 new_tbl
= bucket_table_alloc(ht
, size
, GFP_ATOMIC
);
496 err
= rhashtable_rehash_attach(ht
, tbl
, new_tbl
);
498 bucket_table_free(new_tbl
);
502 schedule_work(&ht
->run_work
);
507 /* Do not fail the insert if someone else did a rehash. */
508 if (likely(rcu_dereference_raw(tbl
->future_tbl
)))
511 /* Schedule async rehash to retry allocation in process context. */
513 schedule_work(&ht
->run_work
);
518 static void *rhashtable_lookup_one(struct rhashtable
*ht
,
519 struct bucket_table
*tbl
, unsigned int hash
,
520 const void *key
, struct rhash_head
*obj
)
522 struct rhashtable_compare_arg arg
= {
526 struct rhash_head __rcu
**pprev
;
527 struct rhash_head
*head
;
530 elasticity
= RHT_ELASTICITY
;
531 pprev
= rht_bucket_var(tbl
, hash
);
532 rht_for_each_continue(head
, *pprev
, tbl
, hash
) {
533 struct rhlist_head
*list
;
534 struct rhlist_head
*plist
;
539 ht
->p
.obj_cmpfn(&arg
, rht_obj(ht
, head
)) :
540 rhashtable_compare(&arg
, rht_obj(ht
, head
)))) {
546 return rht_obj(ht
, head
);
548 list
= container_of(obj
, struct rhlist_head
, rhead
);
549 plist
= container_of(head
, struct rhlist_head
, rhead
);
551 RCU_INIT_POINTER(list
->next
, plist
);
552 head
= rht_dereference_bucket(head
->next
, tbl
, hash
);
553 RCU_INIT_POINTER(list
->rhead
.next
, head
);
554 rcu_assign_pointer(*pprev
, obj
);
560 return ERR_PTR(-EAGAIN
);
562 return ERR_PTR(-ENOENT
);
565 static struct bucket_table
*rhashtable_insert_one(struct rhashtable
*ht
,
566 struct bucket_table
*tbl
,
568 struct rhash_head
*obj
,
571 struct rhash_head __rcu
**pprev
;
572 struct bucket_table
*new_tbl
;
573 struct rhash_head
*head
;
575 if (!IS_ERR_OR_NULL(data
))
576 return ERR_PTR(-EEXIST
);
578 if (PTR_ERR(data
) != -EAGAIN
&& PTR_ERR(data
) != -ENOENT
)
579 return ERR_CAST(data
);
581 new_tbl
= rcu_dereference(tbl
->future_tbl
);
585 if (PTR_ERR(data
) != -ENOENT
)
586 return ERR_CAST(data
);
588 if (unlikely(rht_grow_above_max(ht
, tbl
)))
589 return ERR_PTR(-E2BIG
);
591 if (unlikely(rht_grow_above_100(ht
, tbl
)))
592 return ERR_PTR(-EAGAIN
);
594 pprev
= rht_bucket_insert(ht
, tbl
, hash
);
596 return ERR_PTR(-ENOMEM
);
598 head
= rht_dereference_bucket(*pprev
, tbl
, hash
);
600 RCU_INIT_POINTER(obj
->next
, head
);
602 struct rhlist_head
*list
;
604 list
= container_of(obj
, struct rhlist_head
, rhead
);
605 RCU_INIT_POINTER(list
->next
, NULL
);
608 rcu_assign_pointer(*pprev
, obj
);
610 atomic_inc(&ht
->nelems
);
611 if (rht_grow_above_75(ht
, tbl
))
612 schedule_work(&ht
->run_work
);
617 static void *rhashtable_try_insert(struct rhashtable
*ht
, const void *key
,
618 struct rhash_head
*obj
)
620 struct bucket_table
*new_tbl
;
621 struct bucket_table
*tbl
;
626 tbl
= rcu_dereference(ht
->tbl
);
628 /* All insertions must grab the oldest table containing
629 * the hashed bucket that is yet to be rehashed.
632 hash
= rht_head_hashfn(ht
, tbl
, obj
, ht
->p
);
633 lock
= rht_bucket_lock(tbl
, hash
);
636 if (tbl
->rehash
<= hash
)
639 spin_unlock_bh(lock
);
640 tbl
= rcu_dereference(tbl
->future_tbl
);
643 data
= rhashtable_lookup_one(ht
, tbl
, hash
, key
, obj
);
644 new_tbl
= rhashtable_insert_one(ht
, tbl
, hash
, obj
, data
);
645 if (PTR_ERR(new_tbl
) != -EEXIST
)
646 data
= ERR_CAST(new_tbl
);
648 while (!IS_ERR_OR_NULL(new_tbl
)) {
650 hash
= rht_head_hashfn(ht
, tbl
, obj
, ht
->p
);
651 spin_lock_nested(rht_bucket_lock(tbl
, hash
),
652 SINGLE_DEPTH_NESTING
);
654 data
= rhashtable_lookup_one(ht
, tbl
, hash
, key
, obj
);
655 new_tbl
= rhashtable_insert_one(ht
, tbl
, hash
, obj
, data
);
656 if (PTR_ERR(new_tbl
) != -EEXIST
)
657 data
= ERR_CAST(new_tbl
);
659 spin_unlock(rht_bucket_lock(tbl
, hash
));
662 spin_unlock_bh(lock
);
664 if (PTR_ERR(data
) == -EAGAIN
)
665 data
= ERR_PTR(rhashtable_insert_rehash(ht
, tbl
) ?:
671 void *rhashtable_insert_slow(struct rhashtable
*ht
, const void *key
,
672 struct rhash_head
*obj
)
678 data
= rhashtable_try_insert(ht
, key
, obj
);
680 } while (PTR_ERR(data
) == -EAGAIN
);
684 EXPORT_SYMBOL_GPL(rhashtable_insert_slow
);
687 * rhashtable_walk_enter - Initialise an iterator
688 * @ht: Table to walk over
689 * @iter: Hash table Iterator
691 * This function prepares a hash table walk.
693 * Note that if you restart a walk after rhashtable_walk_stop you
694 * may see the same object twice. Also, you may miss objects if
695 * there are removals in between rhashtable_walk_stop and the next
696 * call to rhashtable_walk_start.
698 * For a completely stable walk you should construct your own data
699 * structure outside the hash table.
701 * This function may sleep so you must not call it from interrupt
702 * context or with spin locks held.
704 * You must call rhashtable_walk_exit after this function returns.
706 void rhashtable_walk_enter(struct rhashtable
*ht
, struct rhashtable_iter
*iter
)
713 spin_lock(&ht
->lock
);
715 rcu_dereference_protected(ht
->tbl
, lockdep_is_held(&ht
->lock
));
716 list_add(&iter
->walker
.list
, &iter
->walker
.tbl
->walkers
);
717 spin_unlock(&ht
->lock
);
719 EXPORT_SYMBOL_GPL(rhashtable_walk_enter
);
722 * rhashtable_walk_exit - Free an iterator
723 * @iter: Hash table Iterator
725 * This function frees resources allocated by rhashtable_walk_init.
727 void rhashtable_walk_exit(struct rhashtable_iter
*iter
)
729 spin_lock(&iter
->ht
->lock
);
730 if (iter
->walker
.tbl
)
731 list_del(&iter
->walker
.list
);
732 spin_unlock(&iter
->ht
->lock
);
734 EXPORT_SYMBOL_GPL(rhashtable_walk_exit
);
737 * rhashtable_walk_start - Start a hash table walk
738 * @iter: Hash table iterator
740 * Start a hash table walk at the current iterator position. Note that we take
741 * the RCU lock in all cases including when we return an error. So you must
742 * always call rhashtable_walk_stop to clean up.
744 * Returns zero if successful.
746 * Returns -EAGAIN if resize event occured. Note that the iterator
747 * will rewind back to the beginning and you may use it immediately
748 * by calling rhashtable_walk_next.
750 int rhashtable_walk_start(struct rhashtable_iter
*iter
)
753 struct rhashtable
*ht
= iter
->ht
;
757 spin_lock(&ht
->lock
);
758 if (iter
->walker
.tbl
)
759 list_del(&iter
->walker
.list
);
760 spin_unlock(&ht
->lock
);
762 if (!iter
->walker
.tbl
) {
763 iter
->walker
.tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
769 EXPORT_SYMBOL_GPL(rhashtable_walk_start
);
772 * rhashtable_walk_next - Return the next object and advance the iterator
773 * @iter: Hash table iterator
775 * Note that you must call rhashtable_walk_stop when you are finished
778 * Returns the next object or NULL when the end of the table is reached.
780 * Returns -EAGAIN if resize event occured. Note that the iterator
781 * will rewind back to the beginning and you may continue to use it.
783 void *rhashtable_walk_next(struct rhashtable_iter
*iter
)
785 struct bucket_table
*tbl
= iter
->walker
.tbl
;
786 struct rhlist_head
*list
= iter
->list
;
787 struct rhashtable
*ht
= iter
->ht
;
788 struct rhash_head
*p
= iter
->p
;
789 bool rhlist
= ht
->rhlist
;
792 if (!rhlist
|| !(list
= rcu_dereference(list
->next
))) {
793 p
= rcu_dereference(p
->next
);
794 list
= container_of(p
, struct rhlist_head
, rhead
);
799 for (; iter
->slot
< tbl
->size
; iter
->slot
++) {
800 int skip
= iter
->skip
;
802 rht_for_each_rcu(p
, tbl
, iter
->slot
) {
804 list
= container_of(p
, struct rhlist_head
,
810 list
= rcu_dereference(list
->next
);
821 if (!rht_is_a_nulls(p
)) {
825 return rht_obj(ht
, rhlist
? &list
->rhead
: p
);
833 /* Ensure we see any new tables. */
836 iter
->walker
.tbl
= rht_dereference_rcu(tbl
->future_tbl
, ht
);
837 if (iter
->walker
.tbl
) {
840 return ERR_PTR(-EAGAIN
);
845 EXPORT_SYMBOL_GPL(rhashtable_walk_next
);
848 * rhashtable_walk_stop - Finish a hash table walk
849 * @iter: Hash table iterator
851 * Finish a hash table walk. Does not reset the iterator to the start of the
854 void rhashtable_walk_stop(struct rhashtable_iter
*iter
)
857 struct rhashtable
*ht
;
858 struct bucket_table
*tbl
= iter
->walker
.tbl
;
865 spin_lock(&ht
->lock
);
866 if (tbl
->rehash
< tbl
->size
)
867 list_add(&iter
->walker
.list
, &tbl
->walkers
);
869 iter
->walker
.tbl
= NULL
;
870 spin_unlock(&ht
->lock
);
877 EXPORT_SYMBOL_GPL(rhashtable_walk_stop
);
879 static size_t rounded_hashtable_size(const struct rhashtable_params
*params
)
883 if (params
->nelem_hint
)
884 retsize
= max(roundup_pow_of_two(params
->nelem_hint
* 4 / 3),
885 (unsigned long)params
->min_size
);
887 retsize
= max(HASH_DEFAULT_SIZE
,
888 (unsigned long)params
->min_size
);
893 static u32
rhashtable_jhash2(const void *key
, u32 length
, u32 seed
)
895 return jhash2(key
, length
, seed
);
899 * rhashtable_init - initialize a new hash table
900 * @ht: hash table to be initialized
901 * @params: configuration parameters
903 * Initializes a new hash table based on the provided configuration
904 * parameters. A table can be configured either with a variable or
907 * Configuration Example 1: Fixed length keys
911 * struct rhash_head node;
914 * struct rhashtable_params params = {
915 * .head_offset = offsetof(struct test_obj, node),
916 * .key_offset = offsetof(struct test_obj, key),
917 * .key_len = sizeof(int),
919 * .nulls_base = (1U << RHT_BASE_SHIFT),
922 * Configuration Example 2: Variable length keys
925 * struct rhash_head node;
928 * u32 my_hash_fn(const void *data, u32 len, u32 seed)
930 * struct test_obj *obj = data;
932 * return [... hash ...];
935 * struct rhashtable_params params = {
936 * .head_offset = offsetof(struct test_obj, node),
938 * .obj_hashfn = my_hash_fn,
941 int rhashtable_init(struct rhashtable
*ht
,
942 const struct rhashtable_params
*params
)
944 struct bucket_table
*tbl
;
947 if ((!params
->key_len
&& !params
->obj_hashfn
) ||
948 (params
->obj_hashfn
&& !params
->obj_cmpfn
))
951 if (params
->nulls_base
&& params
->nulls_base
< (1U << RHT_BASE_SHIFT
))
954 memset(ht
, 0, sizeof(*ht
));
955 mutex_init(&ht
->mutex
);
956 spin_lock_init(&ht
->lock
);
957 memcpy(&ht
->p
, params
, sizeof(*params
));
959 if (params
->min_size
)
960 ht
->p
.min_size
= roundup_pow_of_two(params
->min_size
);
962 /* Cap total entries at 2^31 to avoid nelems overflow. */
963 ht
->max_elems
= 1u << 31;
965 if (params
->max_size
) {
966 ht
->p
.max_size
= rounddown_pow_of_two(params
->max_size
);
967 if (ht
->p
.max_size
< ht
->max_elems
/ 2)
968 ht
->max_elems
= ht
->p
.max_size
* 2;
971 ht
->p
.min_size
= max_t(u16
, ht
->p
.min_size
, HASH_MIN_SIZE
);
973 size
= rounded_hashtable_size(&ht
->p
);
975 if (params
->locks_mul
)
976 ht
->p
.locks_mul
= roundup_pow_of_two(params
->locks_mul
);
978 ht
->p
.locks_mul
= BUCKET_LOCKS_PER_CPU
;
980 ht
->key_len
= ht
->p
.key_len
;
981 if (!params
->hashfn
) {
982 ht
->p
.hashfn
= jhash
;
984 if (!(ht
->key_len
& (sizeof(u32
) - 1))) {
985 ht
->key_len
/= sizeof(u32
);
986 ht
->p
.hashfn
= rhashtable_jhash2
;
990 tbl
= bucket_table_alloc(ht
, size
, GFP_KERNEL
);
994 atomic_set(&ht
->nelems
, 0);
996 RCU_INIT_POINTER(ht
->tbl
, tbl
);
998 INIT_WORK(&ht
->run_work
, rht_deferred_worker
);
1002 EXPORT_SYMBOL_GPL(rhashtable_init
);
1005 * rhltable_init - initialize a new hash list table
1006 * @hlt: hash list table to be initialized
1007 * @params: configuration parameters
1009 * Initializes a new hash list table.
1011 * See documentation for rhashtable_init.
1013 int rhltable_init(struct rhltable
*hlt
, const struct rhashtable_params
*params
)
1017 /* No rhlist NULLs marking for now. */
1018 if (params
->nulls_base
)
1021 err
= rhashtable_init(&hlt
->ht
, params
);
1022 hlt
->ht
.rhlist
= true;
1025 EXPORT_SYMBOL_GPL(rhltable_init
);
1027 static void rhashtable_free_one(struct rhashtable
*ht
, struct rhash_head
*obj
,
1028 void (*free_fn
)(void *ptr
, void *arg
),
1031 struct rhlist_head
*list
;
1034 free_fn(rht_obj(ht
, obj
), arg
);
1038 list
= container_of(obj
, struct rhlist_head
, rhead
);
1041 list
= rht_dereference(list
->next
, ht
);
1042 free_fn(rht_obj(ht
, obj
), arg
);
1047 * rhashtable_free_and_destroy - free elements and destroy hash table
1048 * @ht: the hash table to destroy
1049 * @free_fn: callback to release resources of element
1050 * @arg: pointer passed to free_fn
1052 * Stops an eventual async resize. If defined, invokes free_fn for each
1053 * element to releasal resources. Please note that RCU protected
1054 * readers may still be accessing the elements. Releasing of resources
1055 * must occur in a compatible manner. Then frees the bucket array.
1057 * This function will eventually sleep to wait for an async resize
1058 * to complete. The caller is responsible that no further write operations
1059 * occurs in parallel.
1061 void rhashtable_free_and_destroy(struct rhashtable
*ht
,
1062 void (*free_fn
)(void *ptr
, void *arg
),
1065 struct bucket_table
*tbl
, *next_tbl
;
1068 cancel_work_sync(&ht
->run_work
);
1070 mutex_lock(&ht
->mutex
);
1071 tbl
= rht_dereference(ht
->tbl
, ht
);
1074 for (i
= 0; i
< tbl
->size
; i
++) {
1075 struct rhash_head
*pos
, *next
;
1077 for (pos
= rht_dereference(*rht_bucket(tbl
, i
), ht
),
1078 next
= !rht_is_a_nulls(pos
) ?
1079 rht_dereference(pos
->next
, ht
) : NULL
;
1080 !rht_is_a_nulls(pos
);
1082 next
= !rht_is_a_nulls(pos
) ?
1083 rht_dereference(pos
->next
, ht
) : NULL
)
1084 rhashtable_free_one(ht
, pos
, free_fn
, arg
);
1088 next_tbl
= rht_dereference(tbl
->future_tbl
, ht
);
1089 bucket_table_free(tbl
);
1094 mutex_unlock(&ht
->mutex
);
1096 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy
);
1098 void rhashtable_destroy(struct rhashtable
*ht
)
1100 return rhashtable_free_and_destroy(ht
, NULL
, NULL
);
1102 EXPORT_SYMBOL_GPL(rhashtable_destroy
);
1104 struct rhash_head __rcu
**rht_bucket_nested(const struct bucket_table
*tbl
,
1107 const unsigned int shift
= PAGE_SHIFT
- ilog2(sizeof(void *));
1108 static struct rhash_head __rcu
*rhnull
=
1109 (struct rhash_head __rcu
*)NULLS_MARKER(0);
1110 unsigned int index
= hash
& ((1 << tbl
->nest
) - 1);
1111 unsigned int size
= tbl
->size
>> tbl
->nest
;
1112 unsigned int subhash
= hash
;
1113 union nested_table
*ntbl
;
1115 ntbl
= (union nested_table
*)rcu_dereference_raw(tbl
->buckets
[0]);
1116 ntbl
= rht_dereference_bucket_rcu(ntbl
[index
].table
, tbl
, hash
);
1117 subhash
>>= tbl
->nest
;
1119 while (ntbl
&& size
> (1 << shift
)) {
1120 index
= subhash
& ((1 << shift
) - 1);
1121 ntbl
= rht_dereference_bucket_rcu(ntbl
[index
].table
,
1130 return &ntbl
[subhash
].bucket
;
1133 EXPORT_SYMBOL_GPL(rht_bucket_nested
);
1135 struct rhash_head __rcu
**rht_bucket_nested_insert(struct rhashtable
*ht
,
1136 struct bucket_table
*tbl
,
1139 const unsigned int shift
= PAGE_SHIFT
- ilog2(sizeof(void *));
1140 unsigned int index
= hash
& ((1 << tbl
->nest
) - 1);
1141 unsigned int size
= tbl
->size
>> tbl
->nest
;
1142 union nested_table
*ntbl
;
1143 unsigned int shifted
;
1146 ntbl
= (union nested_table
*)rcu_dereference_raw(tbl
->buckets
[0]);
1149 shifted
= tbl
->nest
;
1150 ntbl
= nested_table_alloc(ht
, &ntbl
[index
].table
,
1151 size
<= (1 << shift
) ? shifted
: 0, nhash
);
1153 while (ntbl
&& size
> (1 << shift
)) {
1154 index
= hash
& ((1 << shift
) - 1);
1157 nhash
|= index
<< shifted
;
1159 ntbl
= nested_table_alloc(ht
, &ntbl
[index
].table
,
1160 size
<= (1 << shift
) ? shifted
: 0,
1167 return &ntbl
[hash
].bucket
;
1170 EXPORT_SYMBOL_GPL(rht_bucket_nested_insert
);