1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Resizable, Scalable, Concurrent Hash Table
5 * Copyright (c) 2015-2016 Herbert Xu <herbert@gondor.apana.org.au>
6 * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
7 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
9 * Code partially derived from nft_hash
10 * Rewritten with rehash code from br_multicast plus single list
11 * pointer as suggested by Josh Triplett
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License version 2 as
15 * published by the Free Software Foundation.
18 #ifndef _LINUX_RHASHTABLE_H
19 #define _LINUX_RHASHTABLE_H
21 #include <linux/err.h>
22 #include <linux/errno.h>
23 #include <linux/jhash.h>
24 #include <linux/list_nulls.h>
25 #include <linux/rcupdate.h>
26 #include <linux/workqueue.h>
27 #include <linux/rculist.h>
28 #include <linux/bit_spinlock.h>
30 #include <linux/rhashtable-types.h>
32 * Objects in an rhashtable have an embedded struct rhash_head
33 * which is linked into as hash chain from the hash table - or one
34 * of two or more hash tables when the rhashtable is being resized.
35 * The end of the chain is marked with a special nulls marks which has
36 * the least significant bit set but otherwise stores the address of
37 * the hash bucket. This allows us to be sure we've found the end
39 * The value stored in the hash bucket has BIT(0) used as a lock bit.
40 * This bit must be atomically set before any changes are made to
41 * the chain. To avoid dereferencing this pointer without clearing
42 * the bit first, we use an opaque 'struct rhash_lock_head *' for the
43 * pointer stored in the bucket. This struct needs to be defined so
44 * that rcu_dereference() works on it, but it has no content so a
45 * cast is needed for it to be useful. This ensures it isn't
46 * used by mistake with clearing the lock bit first.
48 struct rhash_lock_head {};
50 /* Maximum chain length before rehash
52 * The maximum (not average) chain length grows with the size of the hash
53 * table, at a rate of (log N)/(log log N).
55 * The value of 16 is selected so that even if the hash table grew to
56 * 2^32 you would not expect the maximum chain length to exceed it
57 * unless we are under attack (or extremely unlucky).
59 * As this limit is only to detect attacks, we don't need to set it to a
60 * lower value as you'd need the chain length to vastly exceed 16 to have
61 * any real effect on the system.
63 #define RHT_ELASTICITY 16u
66 * struct bucket_table - Table of hash buckets
67 * @size: Number of hash buckets
68 * @nest: Number of bits of first-level nested table.
69 * @rehash: Current bucket being rehashed
70 * @hash_rnd: Random seed to fold into hash
71 * @walkers: List of active walkers
72 * @rcu: RCU structure for freeing the table
73 * @future_tbl: Table under construction during rehashing
74 * @ntbl: Nested table used when out of memory.
75 * @buckets: size * hash buckets
81 struct list_head walkers;
84 struct bucket_table __rcu *future_tbl;
86 struct rhash_lock_head __rcu *buckets[] ____cacheline_aligned_in_smp;
90 * NULLS_MARKER() expects a hash value with the low
91 * bits mostly likely to be significant, and it discards
93 * We give it an address, in which the bottom bit is
94 * always 0, and the msb might be significant.
95 * So we shift the address down one bit to align with
96 * expectations and avoid losing a significant bit.
98 * We never store the NULLS_MARKER in the hash table
99 * itself as we need the lsb for locking.
100 * Instead we store a NULL
102 #define RHT_NULLS_MARKER(ptr) \
103 ((void *)NULLS_MARKER(((unsigned long) (ptr)) >> 1))
104 #define INIT_RHT_NULLS_HEAD(ptr) \
107 static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
109 return ((unsigned long) ptr & 1);
112 static inline void *rht_obj(const struct rhashtable *ht,
113 const struct rhash_head *he)
115 return (char *)he - ht->p.head_offset;
118 static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
121 return hash & (tbl->size - 1);
124 static inline unsigned int rht_key_get_hash(struct rhashtable *ht,
125 const void *key, const struct rhashtable_params params,
126 unsigned int hash_rnd)
130 /* params must be equal to ht->p if it isn't constant. */
131 if (!__builtin_constant_p(params.key_len))
132 hash = ht->p.hashfn(key, ht->key_len, hash_rnd);
133 else if (params.key_len) {
134 unsigned int key_len = params.key_len;
137 hash = params.hashfn(key, key_len, hash_rnd);
138 else if (key_len & (sizeof(u32) - 1))
139 hash = jhash(key, key_len, hash_rnd);
141 hash = jhash2(key, key_len / sizeof(u32), hash_rnd);
143 unsigned int key_len = ht->p.key_len;
146 hash = params.hashfn(key, key_len, hash_rnd);
148 hash = jhash(key, key_len, hash_rnd);
154 static inline unsigned int rht_key_hashfn(
155 struct rhashtable *ht, const struct bucket_table *tbl,
156 const void *key, const struct rhashtable_params params)
158 unsigned int hash = rht_key_get_hash(ht, key, params, tbl->hash_rnd);
160 return rht_bucket_index(tbl, hash);
163 static inline unsigned int rht_head_hashfn(
164 struct rhashtable *ht, const struct bucket_table *tbl,
165 const struct rhash_head *he, const struct rhashtable_params params)
167 const char *ptr = rht_obj(ht, he);
169 return likely(params.obj_hashfn) ?
170 rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
173 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
177 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
179 * @tbl: current table
181 static inline bool rht_grow_above_75(const struct rhashtable *ht,
182 const struct bucket_table *tbl)
184 /* Expand table when exceeding 75% load */
185 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
186 (!ht->p.max_size || tbl->size < ht->p.max_size);
190 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
192 * @tbl: current table
194 static inline bool rht_shrink_below_30(const struct rhashtable *ht,
195 const struct bucket_table *tbl)
197 /* Shrink table beneath 30% load */
198 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
199 tbl->size > ht->p.min_size;
203 * rht_grow_above_100 - returns true if nelems > table-size
205 * @tbl: current table
207 static inline bool rht_grow_above_100(const struct rhashtable *ht,
208 const struct bucket_table *tbl)
210 return atomic_read(&ht->nelems) > tbl->size &&
211 (!ht->p.max_size || tbl->size < ht->p.max_size);
215 * rht_grow_above_max - returns true if table is above maximum
217 * @tbl: current table
219 static inline bool rht_grow_above_max(const struct rhashtable *ht,
220 const struct bucket_table *tbl)
222 return atomic_read(&ht->nelems) >= ht->max_elems;
225 #ifdef CONFIG_PROVE_LOCKING
226 int lockdep_rht_mutex_is_held(struct rhashtable *ht);
227 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
229 static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
234 static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
239 #endif /* CONFIG_PROVE_LOCKING */
241 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
242 struct rhash_head *obj);
244 void rhashtable_walk_enter(struct rhashtable *ht,
245 struct rhashtable_iter *iter);
246 void rhashtable_walk_exit(struct rhashtable_iter *iter);
247 int rhashtable_walk_start_check(struct rhashtable_iter *iter) __acquires(RCU);
249 static inline void rhashtable_walk_start(struct rhashtable_iter *iter)
251 (void)rhashtable_walk_start_check(iter);
254 void *rhashtable_walk_next(struct rhashtable_iter *iter);
255 void *rhashtable_walk_peek(struct rhashtable_iter *iter);
256 void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
258 void rhashtable_free_and_destroy(struct rhashtable *ht,
259 void (*free_fn)(void *ptr, void *arg),
261 void rhashtable_destroy(struct rhashtable *ht);
263 struct rhash_lock_head __rcu **rht_bucket_nested(
264 const struct bucket_table *tbl, unsigned int hash);
265 struct rhash_lock_head __rcu **__rht_bucket_nested(
266 const struct bucket_table *tbl, unsigned int hash);
267 struct rhash_lock_head __rcu **rht_bucket_nested_insert(
268 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash);
270 #define rht_dereference(p, ht) \
273 #define rht_dereference_rcu(p, ht) \
276 #define rht_dereference_bucket(p, tbl, hash) \
279 #define rht_dereference_bucket_rcu(p, tbl, hash) \
282 #define rht_entry(tpos, pos, member) \
283 ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
285 static inline struct rhash_lock_head __rcu *const *rht_bucket(
286 const struct bucket_table *tbl, unsigned int hash)
288 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
292 static inline struct rhash_lock_head __rcu **rht_bucket_var(
293 struct bucket_table *tbl, unsigned int hash)
295 return unlikely(tbl->nest) ? __rht_bucket_nested(tbl, hash) :
299 static inline struct rhash_lock_head __rcu **rht_bucket_insert(
300 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
302 return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
307 * We lock a bucket by setting BIT(0) in the pointer - this is always
308 * zero in real pointers. The NULLS mark is never stored in the bucket,
309 * rather we store NULL if the bucket is empty.
310 * bit_spin_locks do not handle contention well, but the whole point
311 * of the hashtable design is to achieve minimum per-bucket contention.
312 * A nested hash table might not have a bucket pointer. In that case
313 * we cannot get a lock. For remove and replace the bucket cannot be
314 * interesting and doesn't need locking.
315 * For insert we allocate the bucket if this is the last bucket_table,
316 * and then take the lock.
317 * Sometimes we unlock a bucket by writing a new pointer there. In that
318 * case we don't need to unlock, but we do need to reset state such as
319 * local_bh. For that we have rht_assign_unlock(). As rcu_assign_pointer()
320 * provides the same release semantics that bit_spin_unlock() provides,
322 * When we write to a bucket without unlocking, we use rht_assign_locked().
325 static inline void rht_lock(struct bucket_table *tbl,
326 struct rhash_lock_head __rcu **bkt)
328 bit_spin_lock(0, (unsigned long *)bkt);
331 static inline void rht_lock_nested(struct bucket_table *tbl,
332 struct rhash_lock_head __rcu **bucket,
333 unsigned int subclass)
335 bit_spin_lock(0, (unsigned long *)bucket);
338 static inline void rht_unlock(struct bucket_table *tbl,
339 struct rhash_lock_head __rcu **bkt)
341 bit_spin_unlock(0, (unsigned long *)bkt);
344 static inline struct rhash_head *__rht_ptr(
345 struct rhash_lock_head *p, struct rhash_lock_head __rcu *const *bkt)
347 return (struct rhash_head *)
348 ((unsigned long)p & ~BIT(0) ?:
349 (unsigned long)RHT_NULLS_MARKER(bkt));
353 * Where 'bkt' is a bucket and might be locked:
354 * rht_ptr_rcu() dereferences that pointer and clears the lock bit.
355 * rht_ptr() dereferences in a context where the bucket is locked.
356 * rht_ptr_exclusive() dereferences in a context where exclusive
357 * access is guaranteed, such as when destroying the table.
359 static inline struct rhash_head *rht_ptr_rcu(
360 struct rhash_lock_head __rcu *const *bkt)
362 return __rht_ptr(rcu_dereference(*bkt), bkt);
365 static inline struct rhash_head *rht_ptr(
366 struct rhash_lock_head __rcu *const *bkt,
367 struct bucket_table *tbl,
370 return __rht_ptr(rht_dereference_bucket(*bkt, tbl, hash), bkt);
373 static inline struct rhash_head *rht_ptr_exclusive(
374 struct rhash_lock_head __rcu *const *bkt)
376 return __rht_ptr(rcu_dereference(*bkt), bkt);
379 static inline void rht_assign_locked(struct rhash_lock_head __rcu **bkt,
380 struct rhash_head *obj)
382 if (rht_is_a_nulls(obj))
384 rcu_assign_pointer(*bkt, (void *)((unsigned long)obj | BIT(0)));
387 static inline void rht_assign_unlock(struct bucket_table *tbl,
388 struct rhash_lock_head __rcu **bkt,
389 struct rhash_head *obj)
391 if (rht_is_a_nulls(obj))
393 rcu_assign_pointer(*bkt, (void *)obj);
396 bit_spin_wake(0, (unsigned long *) bkt);
400 * rht_for_each_from - iterate over hash chain from given head
401 * @pos: the &struct rhash_head to use as a loop cursor.
402 * @head: the &struct rhash_head to start from
403 * @tbl: the &struct bucket_table
404 * @hash: the hash value / bucket index
406 #define rht_for_each_from(pos, head, tbl, hash) \
408 !rht_is_a_nulls(pos); \
409 pos = rht_dereference_bucket((pos)->next, tbl, hash))
412 * rht_for_each - iterate over hash chain
413 * @pos: the &struct rhash_head to use as a loop cursor.
414 * @tbl: the &struct bucket_table
415 * @hash: the hash value / bucket index
417 #define rht_for_each(pos, tbl, hash) \
418 rht_for_each_from(pos, rht_ptr(rht_bucket(tbl, hash), tbl, hash), \
422 * rht_for_each_entry_from - iterate over hash chain from given head
423 * @tpos: the type * to use as a loop cursor.
424 * @pos: the &struct rhash_head to use as a loop cursor.
425 * @head: the &struct rhash_head to start from
426 * @tbl: the &struct bucket_table
427 * @hash: the hash value / bucket index
428 * @member: name of the &struct rhash_head within the hashable struct.
430 #define rht_for_each_entry_from(tpos, pos, head, tbl, hash, member) \
432 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
433 pos = rht_dereference_bucket((pos)->next, tbl, hash))
436 * rht_for_each_entry - iterate over hash chain of given type
437 * @tpos: the type * to use as a loop cursor.
438 * @pos: the &struct rhash_head to use as a loop cursor.
439 * @tbl: the &struct bucket_table
440 * @hash: the hash value / bucket index
441 * @member: name of the &struct rhash_head within the hashable struct.
443 #define rht_for_each_entry(tpos, pos, tbl, hash, member) \
444 rht_for_each_entry_from(tpos, pos, \
445 rht_ptr(rht_bucket(tbl, hash), tbl, hash), \
449 * rht_for_each_entry_safe - safely iterate over hash chain of given type
450 * @tpos: the type * to use as a loop cursor.
451 * @pos: the &struct rhash_head to use as a loop cursor.
452 * @next: the &struct rhash_head to use as next in loop cursor.
453 * @tbl: the &struct bucket_table
454 * @hash: the hash value / bucket index
455 * @member: name of the &struct rhash_head within the hashable struct.
457 * This hash chain list-traversal primitive allows for the looped code to
458 * remove the loop cursor from the list.
460 #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \
461 for (pos = rht_ptr(rht_bucket(tbl, hash), tbl, hash), \
462 next = !rht_is_a_nulls(pos) ? \
463 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
464 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
466 next = !rht_is_a_nulls(pos) ? \
467 rht_dereference_bucket(pos->next, tbl, hash) : NULL)
470 * rht_for_each_rcu_from - iterate over rcu hash chain from given head
471 * @pos: the &struct rhash_head to use as a loop cursor.
472 * @head: the &struct rhash_head to start from
473 * @tbl: the &struct bucket_table
474 * @hash: the hash value / bucket index
476 * This hash chain list-traversal primitive may safely run concurrently with
477 * the _rcu mutation primitives such as rhashtable_insert() as long as the
478 * traversal is guarded by rcu_read_lock().
480 #define rht_for_each_rcu_from(pos, head, tbl, hash) \
481 for (({barrier(); }), \
483 !rht_is_a_nulls(pos); \
484 pos = rcu_dereference_raw(pos->next))
487 * rht_for_each_rcu - iterate over rcu hash chain
488 * @pos: the &struct rhash_head to use as a loop cursor.
489 * @tbl: the &struct bucket_table
490 * @hash: the hash value / bucket index
492 * This hash chain list-traversal primitive may safely run concurrently with
493 * the _rcu mutation primitives such as rhashtable_insert() as long as the
494 * traversal is guarded by rcu_read_lock().
496 #define rht_for_each_rcu(pos, tbl, hash) \
497 for (({barrier(); }), \
498 pos = rht_ptr_rcu(rht_bucket(tbl, hash)); \
499 !rht_is_a_nulls(pos); \
500 pos = rcu_dereference_raw(pos->next))
503 * rht_for_each_entry_rcu_from - iterated over rcu hash chain from given head
504 * @tpos: the type * to use as a loop cursor.
505 * @pos: the &struct rhash_head to use as a loop cursor.
506 * @head: the &struct rhash_head to start from
507 * @tbl: the &struct bucket_table
508 * @hash: the hash value / bucket index
509 * @member: name of the &struct rhash_head within the hashable struct.
511 * This hash chain list-traversal primitive may safely run concurrently with
512 * the _rcu mutation primitives such as rhashtable_insert() as long as the
513 * traversal is guarded by rcu_read_lock().
515 #define rht_for_each_entry_rcu_from(tpos, pos, head, tbl, hash, member) \
516 for (({barrier(); }), \
518 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
519 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
522 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
523 * @tpos: the type * to use as a loop cursor.
524 * @pos: the &struct rhash_head to use as a loop cursor.
525 * @tbl: the &struct bucket_table
526 * @hash: the hash value / bucket index
527 * @member: name of the &struct rhash_head within the hashable struct.
529 * This hash chain list-traversal primitive may safely run concurrently with
530 * the _rcu mutation primitives such as rhashtable_insert() as long as the
531 * traversal is guarded by rcu_read_lock().
533 #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \
534 rht_for_each_entry_rcu_from(tpos, pos, \
535 rht_ptr_rcu(rht_bucket(tbl, hash)), \
539 * rhl_for_each_rcu - iterate over rcu hash table list
540 * @pos: the &struct rlist_head to use as a loop cursor.
541 * @list: the head of the list
543 * This hash chain list-traversal primitive should be used on the
544 * list returned by rhltable_lookup.
546 #define rhl_for_each_rcu(pos, list) \
547 for (pos = list; pos; pos = rcu_dereference_raw(pos->next))
550 * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
551 * @tpos: the type * to use as a loop cursor.
552 * @pos: the &struct rlist_head to use as a loop cursor.
553 * @list: the head of the list
554 * @member: name of the &struct rlist_head within the hashable struct.
556 * This hash chain list-traversal primitive should be used on the
557 * list returned by rhltable_lookup.
559 #define rhl_for_each_entry_rcu(tpos, pos, list, member) \
560 for (pos = list; pos && rht_entry(tpos, pos, member); \
561 pos = rcu_dereference_raw(pos->next))
563 static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
566 struct rhashtable *ht = arg->ht;
567 const char *ptr = obj;
569 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
572 /* Internal function, do not use. */
573 static inline struct rhash_head *__rhashtable_lookup(
574 struct rhashtable *ht, const void *key,
575 const struct rhashtable_params params)
577 struct rhashtable_compare_arg arg = {
581 struct rhash_lock_head __rcu *const *bkt;
582 struct bucket_table *tbl;
583 struct rhash_head *he;
586 tbl = rht_dereference_rcu(ht->tbl, ht);
588 hash = rht_key_hashfn(ht, tbl, key, params);
589 bkt = rht_bucket(tbl, hash);
591 rht_for_each_rcu_from(he, rht_ptr_rcu(bkt), tbl, hash) {
592 if (params.obj_cmpfn ?
593 params.obj_cmpfn(&arg, rht_obj(ht, he)) :
594 rhashtable_compare(&arg, rht_obj(ht, he)))
598 /* An object might have been moved to a different hash chain,
599 * while we walk along it - better check and retry.
601 } while (he != RHT_NULLS_MARKER(bkt));
603 /* Ensure we see any new tables. */
606 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
614 * rhashtable_lookup - search hash table
616 * @key: the pointer to the key
617 * @params: hash table parameters
619 * Computes the hash value for the key and traverses the bucket chain looking
620 * for a entry with an identical key. The first matching entry is returned.
622 * This must only be called under the RCU read lock.
624 * Returns the first entry on which the compare function returned true.
626 static inline void *rhashtable_lookup(
627 struct rhashtable *ht, const void *key,
628 const struct rhashtable_params params)
630 struct rhash_head *he = __rhashtable_lookup(ht, key, params);
632 return he ? rht_obj(ht, he) : NULL;
636 * rhashtable_lookup_fast - search hash table, without RCU read lock
638 * @key: the pointer to the key
639 * @params: hash table parameters
641 * Computes the hash value for the key and traverses the bucket chain looking
642 * for a entry with an identical key. The first matching entry is returned.
644 * Only use this function when you have other mechanisms guaranteeing
645 * that the object won't go away after the RCU read lock is released.
647 * Returns the first entry on which the compare function returned true.
649 static inline void *rhashtable_lookup_fast(
650 struct rhashtable *ht, const void *key,
651 const struct rhashtable_params params)
656 obj = rhashtable_lookup(ht, key, params);
663 * rhltable_lookup - search hash list table
665 * @key: the pointer to the key
666 * @params: hash table parameters
668 * Computes the hash value for the key and traverses the bucket chain looking
669 * for a entry with an identical key. All matching entries are returned
672 * This must only be called under the RCU read lock.
674 * Returns the list of entries that match the given key.
676 static inline struct rhlist_head *rhltable_lookup(
677 struct rhltable *hlt, const void *key,
678 const struct rhashtable_params params)
680 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params);
682 return he ? container_of(he, struct rhlist_head, rhead) : NULL;
685 /* Internal function, please use rhashtable_insert_fast() instead. This
686 * function returns the existing element already in hashes in there is a clash,
687 * otherwise it returns an error via ERR_PTR().
689 static inline void *__rhashtable_insert_fast(
690 struct rhashtable *ht, const void *key, struct rhash_head *obj,
691 const struct rhashtable_params params, bool rhlist)
693 struct rhashtable_compare_arg arg = {
697 struct rhash_lock_head __rcu **bkt;
698 struct rhash_head __rcu **pprev;
699 struct bucket_table *tbl;
700 struct rhash_head *head;
707 tbl = rht_dereference_rcu(ht->tbl, ht);
708 hash = rht_head_hashfn(ht, tbl, obj, params);
709 elasticity = RHT_ELASTICITY;
710 bkt = rht_bucket_insert(ht, tbl, hash);
711 data = ERR_PTR(-ENOMEM);
717 if (unlikely(rcu_access_pointer(tbl->future_tbl))) {
719 rht_unlock(tbl, bkt);
721 return rhashtable_insert_slow(ht, key, obj);
724 rht_for_each_from(head, rht_ptr(bkt, tbl, hash), tbl, hash) {
725 struct rhlist_head *plist;
726 struct rhlist_head *list;
731 params.obj_cmpfn(&arg, rht_obj(ht, head)) :
732 rhashtable_compare(&arg, rht_obj(ht, head)))) {
737 data = rht_obj(ht, head);
743 list = container_of(obj, struct rhlist_head, rhead);
744 plist = container_of(head, struct rhlist_head, rhead);
746 RCU_INIT_POINTER(list->next, plist);
747 head = rht_dereference_bucket(head->next, tbl, hash);
748 RCU_INIT_POINTER(list->rhead.next, head);
750 rcu_assign_pointer(*pprev, obj);
751 rht_unlock(tbl, bkt);
753 rht_assign_unlock(tbl, bkt, obj);
761 data = ERR_PTR(-E2BIG);
762 if (unlikely(rht_grow_above_max(ht, tbl)))
765 if (unlikely(rht_grow_above_100(ht, tbl)))
768 /* Inserting at head of list makes unlocking free. */
769 head = rht_ptr(bkt, tbl, hash);
771 RCU_INIT_POINTER(obj->next, head);
773 struct rhlist_head *list;
775 list = container_of(obj, struct rhlist_head, rhead);
776 RCU_INIT_POINTER(list->next, NULL);
779 atomic_inc(&ht->nelems);
780 rht_assign_unlock(tbl, bkt, obj);
782 if (rht_grow_above_75(ht, tbl))
783 schedule_work(&ht->run_work);
792 rht_unlock(tbl, bkt);
797 * rhashtable_insert_fast - insert object into hash table
799 * @obj: pointer to hash head inside object
800 * @params: hash table parameters
802 * Will take the per bucket bitlock to protect against mutual mutations
803 * on the same bucket. Multiple insertions may occur in parallel unless
804 * they map to the same bucket.
806 * It is safe to call this function from atomic context.
808 * Will trigger an automatic deferred table resizing if residency in the
809 * table grows beyond 70%.
811 static inline int rhashtable_insert_fast(
812 struct rhashtable *ht, struct rhash_head *obj,
813 const struct rhashtable_params params)
817 ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
821 return ret == NULL ? 0 : -EEXIST;
825 * rhltable_insert_key - insert object into hash list table
826 * @hlt: hash list table
827 * @key: the pointer to the key
828 * @list: pointer to hash list head inside object
829 * @params: hash table parameters
831 * Will take the per bucket bitlock to protect against mutual mutations
832 * on the same bucket. Multiple insertions may occur in parallel unless
833 * they map to the same bucket.
835 * It is safe to call this function from atomic context.
837 * Will trigger an automatic deferred table resizing if residency in the
838 * table grows beyond 70%.
840 static inline int rhltable_insert_key(
841 struct rhltable *hlt, const void *key, struct rhlist_head *list,
842 const struct rhashtable_params params)
844 return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
849 * rhltable_insert - insert object into hash list table
850 * @hlt: hash list table
851 * @list: pointer to hash list head inside object
852 * @params: hash table parameters
854 * Will take the per bucket bitlock to protect against mutual mutations
855 * on the same bucket. Multiple insertions may occur in parallel unless
856 * they map to the same bucket.
858 * It is safe to call this function from atomic context.
860 * Will trigger an automatic deferred table resizing if residency in the
861 * table grows beyond 70%.
863 static inline int rhltable_insert(
864 struct rhltable *hlt, struct rhlist_head *list,
865 const struct rhashtable_params params)
867 const char *key = rht_obj(&hlt->ht, &list->rhead);
869 key += params.key_offset;
871 return rhltable_insert_key(hlt, key, list, params);
875 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
877 * @obj: pointer to hash head inside object
878 * @params: hash table parameters
880 * This lookup function may only be used for fixed key hash table (key_len
881 * parameter set). It will BUG() if used inappropriately.
883 * It is safe to call this function from atomic context.
885 * Will trigger an automatic deferred table resizing if residency in the
886 * table grows beyond 70%.
888 static inline int rhashtable_lookup_insert_fast(
889 struct rhashtable *ht, struct rhash_head *obj,
890 const struct rhashtable_params params)
892 const char *key = rht_obj(ht, obj);
895 BUG_ON(ht->p.obj_hashfn);
897 ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
902 return ret == NULL ? 0 : -EEXIST;
906 * rhashtable_lookup_get_insert_fast - lookup and insert object into hash table
908 * @obj: pointer to hash head inside object
909 * @params: hash table parameters
911 * Just like rhashtable_lookup_insert_fast(), but this function returns the
912 * object if it exists, NULL if it did not and the insertion was successful,
913 * and an ERR_PTR otherwise.
915 static inline void *rhashtable_lookup_get_insert_fast(
916 struct rhashtable *ht, struct rhash_head *obj,
917 const struct rhashtable_params params)
919 const char *key = rht_obj(ht, obj);
921 BUG_ON(ht->p.obj_hashfn);
923 return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
928 * rhashtable_lookup_insert_key - search and insert object to hash table
932 * @obj: pointer to hash head inside object
933 * @params: hash table parameters
935 * Lookups may occur in parallel with hashtable mutations and resizing.
937 * Will trigger an automatic deferred table resizing if residency in the
938 * table grows beyond 70%.
940 * Returns zero on success.
942 static inline int rhashtable_lookup_insert_key(
943 struct rhashtable *ht, const void *key, struct rhash_head *obj,
944 const struct rhashtable_params params)
948 BUG_ON(!ht->p.obj_hashfn || !key);
950 ret = __rhashtable_insert_fast(ht, key, obj, params, false);
954 return ret == NULL ? 0 : -EEXIST;
958 * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
961 * @obj: pointer to hash head inside object
962 * @params: hash table parameters
964 * Just like rhashtable_lookup_insert_key(), but this function returns the
965 * object if it exists, NULL if it does not and the insertion was successful,
966 * and an ERR_PTR otherwise.
968 static inline void *rhashtable_lookup_get_insert_key(
969 struct rhashtable *ht, const void *key, struct rhash_head *obj,
970 const struct rhashtable_params params)
972 BUG_ON(!ht->p.obj_hashfn || !key);
974 return __rhashtable_insert_fast(ht, key, obj, params, false);
977 /* Internal function, please use rhashtable_remove_fast() instead */
978 static inline int __rhashtable_remove_fast_one(
979 struct rhashtable *ht, struct bucket_table *tbl,
980 struct rhash_head *obj, const struct rhashtable_params params,
983 struct rhash_lock_head __rcu **bkt;
984 struct rhash_head __rcu **pprev;
985 struct rhash_head *he;
989 hash = rht_head_hashfn(ht, tbl, obj, params);
990 bkt = rht_bucket_var(tbl, hash);
996 rht_for_each_from(he, rht_ptr(bkt, tbl, hash), tbl, hash) {
997 struct rhlist_head *list;
999 list = container_of(he, struct rhlist_head, rhead);
1002 struct rhlist_head __rcu **lpprev;
1010 lpprev = &list->next;
1011 list = rht_dereference_bucket(list->next,
1013 } while (list && obj != &list->rhead);
1018 list = rht_dereference_bucket(list->next, tbl, hash);
1019 RCU_INIT_POINTER(*lpprev, list);
1024 obj = rht_dereference_bucket(obj->next, tbl, hash);
1028 list = rht_dereference_bucket(list->next, tbl, hash);
1030 RCU_INIT_POINTER(list->rhead.next, obj);
1037 rcu_assign_pointer(*pprev, obj);
1038 rht_unlock(tbl, bkt);
1040 rht_assign_unlock(tbl, bkt, obj);
1045 rht_unlock(tbl, bkt);
1048 atomic_dec(&ht->nelems);
1049 if (unlikely(ht->p.automatic_shrinking &&
1050 rht_shrink_below_30(ht, tbl)))
1051 schedule_work(&ht->run_work);
1058 /* Internal function, please use rhashtable_remove_fast() instead */
1059 static inline int __rhashtable_remove_fast(
1060 struct rhashtable *ht, struct rhash_head *obj,
1061 const struct rhashtable_params params, bool rhlist)
1063 struct bucket_table *tbl;
1068 tbl = rht_dereference_rcu(ht->tbl, ht);
1070 /* Because we have already taken (and released) the bucket
1071 * lock in old_tbl, if we find that future_tbl is not yet
1072 * visible then that guarantees the entry to still be in
1073 * the old tbl if it exists.
1075 while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
1077 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1086 * rhashtable_remove_fast - remove object from hash table
1088 * @obj: pointer to hash head inside object
1089 * @params: hash table parameters
1091 * Since the hash chain is single linked, the removal operation needs to
1092 * walk the bucket chain upon removal. The removal operation is thus
1093 * considerable slow if the hash table is not correctly sized.
1095 * Will automatically shrink the table if permitted when residency drops
1098 * Returns zero on success, -ENOENT if the entry could not be found.
1100 static inline int rhashtable_remove_fast(
1101 struct rhashtable *ht, struct rhash_head *obj,
1102 const struct rhashtable_params params)
1104 return __rhashtable_remove_fast(ht, obj, params, false);
1108 * rhltable_remove - remove object from hash list table
1109 * @hlt: hash list table
1110 * @list: pointer to hash list head inside object
1111 * @params: hash table parameters
1113 * Since the hash chain is single linked, the removal operation needs to
1114 * walk the bucket chain upon removal. The removal operation is thus
1115 * considerable slow if the hash table is not correctly sized.
1117 * Will automatically shrink the table if permitted when residency drops
1120 * Returns zero on success, -ENOENT if the entry could not be found.
1122 static inline int rhltable_remove(
1123 struct rhltable *hlt, struct rhlist_head *list,
1124 const struct rhashtable_params params)
1126 return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
1129 /* Internal function, please use rhashtable_replace_fast() instead */
1130 static inline int __rhashtable_replace_fast(
1131 struct rhashtable *ht, struct bucket_table *tbl,
1132 struct rhash_head *obj_old, struct rhash_head *obj_new,
1133 const struct rhashtable_params params)
1135 struct rhash_lock_head __rcu **bkt;
1136 struct rhash_head __rcu **pprev;
1137 struct rhash_head *he;
1141 /* Minimally, the old and new objects must have same hash
1142 * (which should mean identifiers are the same).
1144 hash = rht_head_hashfn(ht, tbl, obj_old, params);
1145 if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1148 bkt = rht_bucket_var(tbl, hash);
1155 rht_for_each_from(he, rht_ptr(bkt, tbl, hash), tbl, hash) {
1156 if (he != obj_old) {
1161 rcu_assign_pointer(obj_new->next, obj_old->next);
1163 rcu_assign_pointer(*pprev, obj_new);
1164 rht_unlock(tbl, bkt);
1166 rht_assign_unlock(tbl, bkt, obj_new);
1172 rht_unlock(tbl, bkt);
1179 * rhashtable_replace_fast - replace an object in hash table
1181 * @obj_old: pointer to hash head inside object being replaced
1182 * @obj_new: pointer to hash head inside object which is new
1183 * @params: hash table parameters
1185 * Replacing an object doesn't affect the number of elements in the hash table
1186 * or bucket, so we don't need to worry about shrinking or expanding the
1189 * Returns zero on success, -ENOENT if the entry could not be found,
1190 * -EINVAL if hash is not the same for the old and new objects.
1192 static inline int rhashtable_replace_fast(
1193 struct rhashtable *ht, struct rhash_head *obj_old,
1194 struct rhash_head *obj_new,
1195 const struct rhashtable_params params)
1197 struct bucket_table *tbl;
1202 tbl = rht_dereference_rcu(ht->tbl, ht);
1204 /* Because we have already taken (and released) the bucket
1205 * lock in old_tbl, if we find that future_tbl is not yet
1206 * visible then that guarantees the entry to still be in
1207 * the old tbl if it exists.
1209 while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1210 obj_new, params)) &&
1211 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1220 * rhltable_walk_enter - Initialise an iterator
1221 * @hlt: Table to walk over
1222 * @iter: Hash table Iterator
1224 * This function prepares a hash table walk.
1226 * Note that if you restart a walk after rhashtable_walk_stop you
1227 * may see the same object twice. Also, you may miss objects if
1228 * there are removals in between rhashtable_walk_stop and the next
1229 * call to rhashtable_walk_start.
1231 * For a completely stable walk you should construct your own data
1232 * structure outside the hash table.
1234 * This function may be called from any process context, including
1235 * non-preemptable context, but cannot be called from softirq or
1238 * You must call rhashtable_walk_exit after this function returns.
1240 static inline void rhltable_walk_enter(struct rhltable *hlt,
1241 struct rhashtable_iter *iter)
1243 return rhashtable_walk_enter(&hlt->ht, iter);
1247 * rhltable_free_and_destroy - free elements and destroy hash list table
1248 * @hlt: the hash list table to destroy
1249 * @free_fn: callback to release resources of element
1250 * @arg: pointer passed to free_fn
1252 * See documentation for rhashtable_free_and_destroy.
1254 static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1255 void (*free_fn)(void *ptr,
1259 return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1262 static inline void rhltable_destroy(struct rhltable *hlt)
1264 return rhltable_free_and_destroy(hlt, NULL, NULL);
1267 #endif /* _LINUX_RHASHTABLE_H */