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/slab.h>
21 #include <linux/vmalloc.h>
23 #include <linux/jhash.h>
24 #include <linux/random.h>
25 #include <linux/rhashtable.h>
26 #include <linux/err.h>
28 #define HASH_DEFAULT_SIZE 64UL
29 #define HASH_MIN_SIZE 4UL
30 #define BUCKET_LOCKS_PER_CPU 128UL
32 /* Base bits plus 1 bit for nulls marker */
33 #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 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
44 * a single lock always covers both buckets which may both contains
45 * entries which link to the same bucket of the old table during resizing.
46 * This allows to simplify the locking as locking the bucket in both
47 * tables during resize always guarantee protection.
49 * IMPORTANT: When holding the bucket lock of both the old and new table
50 * during expansions and shrinking, the old bucket lock must always be
53 static spinlock_t *bucket_lock(const struct bucket_table *tbl, u32 hash)
55 return &tbl->locks[hash & tbl->locks_mask];
58 static void *rht_obj(const struct rhashtable *ht, const struct rhash_head *he)
60 return (void *) he - ht->p.head_offset;
63 static u32 rht_bucket_index(const struct bucket_table *tbl, u32 hash)
65 return hash & (tbl->size - 1);
68 static u32 obj_raw_hashfn(const struct rhashtable *ht, const void *ptr)
72 if (unlikely(!ht->p.key_len))
73 hash = ht->p.obj_hashfn(ptr, ht->p.hash_rnd);
75 hash = ht->p.hashfn(ptr + ht->p.key_offset, ht->p.key_len,
78 return hash >> HASH_RESERVED_SPACE;
81 static u32 key_hashfn(struct rhashtable *ht, const void *key, u32 len)
83 return ht->p.hashfn(key, len, ht->p.hash_rnd) >> HASH_RESERVED_SPACE;
86 static u32 head_hashfn(const struct rhashtable *ht,
87 const struct bucket_table *tbl,
88 const struct rhash_head *he)
90 return rht_bucket_index(tbl, obj_raw_hashfn(ht, rht_obj(ht, he)));
93 #ifdef CONFIG_PROVE_LOCKING
94 static void debug_dump_buckets(const struct rhashtable *ht,
95 const struct bucket_table *tbl)
97 struct rhash_head *he;
100 for (i = 0; i < tbl->size; i++) {
101 pr_warn(" [Bucket %d] ", i);
102 rht_for_each_rcu(he, tbl, i) {
103 hash = head_hashfn(ht, tbl, he);
104 pr_cont("[hash = %#x, lock = %p] ",
105 hash, bucket_lock(tbl, hash));
112 static void debug_dump_table(struct rhashtable *ht,
113 const struct bucket_table *tbl,
116 struct bucket_table *old_tbl, *future_tbl;
118 pr_emerg("BUG: lock for hash %#x in table %p not held\n",
122 future_tbl = rht_dereference_rcu(ht->future_tbl, ht);
123 old_tbl = rht_dereference_rcu(ht->tbl, ht);
124 if (future_tbl != old_tbl) {
125 pr_warn("Future table %p (size: %zd)\n",
126 future_tbl, future_tbl->size);
127 debug_dump_buckets(ht, future_tbl);
130 pr_warn("Table %p (size: %zd)\n", old_tbl, old_tbl->size);
131 debug_dump_buckets(ht, old_tbl);
136 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
137 #define ASSERT_BUCKET_LOCK(HT, TBL, HASH) \
139 if (unlikely(!lockdep_rht_bucket_is_held(TBL, HASH))) { \
140 debug_dump_table(HT, TBL, HASH); \
145 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
147 return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
149 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
151 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
153 spinlock_t *lock = bucket_lock(tbl, hash);
155 return (debug_locks) ? lockdep_is_held(lock) : 1;
157 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
159 #define ASSERT_RHT_MUTEX(HT)
160 #define ASSERT_BUCKET_LOCK(HT, TBL, HASH)
164 static struct rhash_head __rcu **bucket_tail(struct bucket_table *tbl, u32 n)
166 struct rhash_head __rcu **pprev;
168 for (pprev = &tbl->buckets[n];
169 !rht_is_a_nulls(rht_dereference_bucket(*pprev, tbl, n));
170 pprev = &rht_dereference_bucket(*pprev, tbl, n)->next)
176 static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl)
178 unsigned int i, size;
179 #if defined(CONFIG_PROVE_LOCKING)
180 unsigned int nr_pcpus = 2;
182 unsigned int nr_pcpus = num_possible_cpus();
185 nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL);
186 size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
188 /* Never allocate more than 0.5 locks per bucket */
189 size = min_t(unsigned int, size, tbl->size >> 1);
191 if (sizeof(spinlock_t) != 0) {
193 if (size * sizeof(spinlock_t) > PAGE_SIZE)
194 tbl->locks = vmalloc(size * sizeof(spinlock_t));
197 tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
201 for (i = 0; i < size; i++)
202 spin_lock_init(&tbl->locks[i]);
204 tbl->locks_mask = size - 1;
209 static void bucket_table_free(const struct bucket_table *tbl)
217 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
220 struct bucket_table *tbl;
224 size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
225 tbl = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
232 tbl->size = nbuckets;
234 if (alloc_bucket_locks(ht, tbl) < 0) {
235 bucket_table_free(tbl);
239 for (i = 0; i < nbuckets; i++)
240 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
246 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
248 * @new_size: new table size
250 bool rht_grow_above_75(const struct rhashtable *ht, size_t new_size)
252 /* Expand table when exceeding 75% load */
253 return atomic_read(&ht->nelems) > (new_size / 4 * 3) &&
254 (ht->p.max_shift && atomic_read(&ht->shift) < ht->p.max_shift);
256 EXPORT_SYMBOL_GPL(rht_grow_above_75);
259 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
261 * @new_size: new table size
263 bool rht_shrink_below_30(const struct rhashtable *ht, size_t new_size)
265 /* Shrink table beneath 30% load */
266 return atomic_read(&ht->nelems) < (new_size * 3 / 10) &&
267 (atomic_read(&ht->shift) > ht->p.min_shift);
269 EXPORT_SYMBOL_GPL(rht_shrink_below_30);
271 static void lock_buckets(struct bucket_table *new_tbl,
272 struct bucket_table *old_tbl, unsigned int hash)
273 __acquires(old_bucket_lock)
275 spin_lock_bh(bucket_lock(old_tbl, hash));
276 if (new_tbl != old_tbl)
277 spin_lock_bh_nested(bucket_lock(new_tbl, hash),
281 static void unlock_buckets(struct bucket_table *new_tbl,
282 struct bucket_table *old_tbl, unsigned int hash)
283 __releases(old_bucket_lock)
285 if (new_tbl != old_tbl)
286 spin_unlock_bh(bucket_lock(new_tbl, hash));
287 spin_unlock_bh(bucket_lock(old_tbl, hash));
291 * Unlink entries on bucket which hash to different bucket.
293 * Returns true if no more work needs to be performed on the bucket.
295 static bool hashtable_chain_unzip(struct rhashtable *ht,
296 const struct bucket_table *new_tbl,
297 struct bucket_table *old_tbl,
300 struct rhash_head *he, *p, *next;
301 unsigned int new_hash, new_hash2;
303 ASSERT_BUCKET_LOCK(ht, old_tbl, old_hash);
305 /* Old bucket empty, no work needed. */
306 p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl,
308 if (rht_is_a_nulls(p))
311 new_hash = head_hashfn(ht, new_tbl, p);
312 ASSERT_BUCKET_LOCK(ht, new_tbl, new_hash);
314 /* Advance the old bucket pointer one or more times until it
315 * reaches a node that doesn't hash to the same bucket as the
316 * previous node p. Call the previous node p;
318 rht_for_each_continue(he, p->next, old_tbl, old_hash) {
319 new_hash2 = head_hashfn(ht, new_tbl, he);
320 ASSERT_BUCKET_LOCK(ht, new_tbl, new_hash2);
322 if (new_hash != new_hash2)
326 rcu_assign_pointer(old_tbl->buckets[old_hash], p->next);
328 /* Find the subsequent node which does hash to the same
329 * bucket as node P, or NULL if no such node exists.
331 INIT_RHT_NULLS_HEAD(next, ht, old_hash);
332 if (!rht_is_a_nulls(he)) {
333 rht_for_each_continue(he, he->next, old_tbl, old_hash) {
334 if (head_hashfn(ht, new_tbl, he) == new_hash) {
341 /* Set p's next pointer to that subsequent node pointer,
342 * bypassing the nodes which do not hash to p's bucket
344 rcu_assign_pointer(p->next, next);
346 p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl,
349 return !rht_is_a_nulls(p);
352 static void link_old_to_new(struct rhashtable *ht, struct bucket_table *new_tbl,
353 unsigned int new_hash, struct rhash_head *entry)
355 ASSERT_BUCKET_LOCK(ht, new_tbl, new_hash);
357 rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), entry);
361 * rhashtable_expand - Expand hash table while allowing concurrent lookups
362 * @ht: the hash table to expand
364 * A secondary bucket array is allocated and the hash entries are migrated
365 * while keeping them on both lists until the end of the RCU grace period.
367 * This function may only be called in a context where it is safe to call
368 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
370 * The caller must ensure that no concurrent resizing occurs by holding
373 * It is valid to have concurrent insertions and deletions protected by per
374 * bucket locks or concurrent RCU protected lookups and traversals.
376 int rhashtable_expand(struct rhashtable *ht)
378 struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
379 struct rhash_head *he;
380 unsigned int new_hash, old_hash;
381 bool complete = false;
383 ASSERT_RHT_MUTEX(ht);
385 new_tbl = bucket_table_alloc(ht, old_tbl->size * 2);
389 atomic_inc(&ht->shift);
391 /* Make insertions go into the new, empty table right away. Deletions
392 * and lookups will be attempted in both tables until we synchronize.
393 * The synchronize_rcu() guarantees for the new table to be picked up
394 * so no new additions go into the old table while we relink.
396 rcu_assign_pointer(ht->future_tbl, new_tbl);
399 /* For each new bucket, search the corresponding old bucket for the
400 * first entry that hashes to the new bucket, and link the end of
401 * newly formed bucket chain (containing entries added to future
402 * table) to that entry. Since all the entries which will end up in
403 * the new bucket appear in the same old bucket, this constructs an
404 * entirely valid new hash table, but with multiple buckets
405 * "zipped" together into a single imprecise chain.
407 for (new_hash = 0; new_hash < new_tbl->size; new_hash++) {
408 old_hash = rht_bucket_index(old_tbl, new_hash);
409 lock_buckets(new_tbl, old_tbl, new_hash);
410 rht_for_each(he, old_tbl, old_hash) {
411 if (head_hashfn(ht, new_tbl, he) == new_hash) {
412 link_old_to_new(ht, new_tbl, new_hash, he);
416 unlock_buckets(new_tbl, old_tbl, new_hash);
419 /* Unzip interleaved hash chains */
420 while (!complete && !ht->being_destroyed) {
421 /* Wait for readers. All new readers will see the new
422 * table, and thus no references to the old table will
427 /* For each bucket in the old table (each of which
428 * contains items from multiple buckets of the new
432 for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
433 lock_buckets(new_tbl, old_tbl, old_hash);
435 if (hashtable_chain_unzip(ht, new_tbl, old_tbl,
439 unlock_buckets(new_tbl, old_tbl, old_hash);
443 rcu_assign_pointer(ht->tbl, new_tbl);
446 bucket_table_free(old_tbl);
449 EXPORT_SYMBOL_GPL(rhashtable_expand);
452 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
453 * @ht: the hash table to shrink
455 * This function may only be called in a context where it is safe to call
456 * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
458 * The caller must ensure that no concurrent resizing occurs by holding
461 * The caller must ensure that no concurrent table mutations take place.
462 * It is however valid to have concurrent lookups if they are RCU protected.
464 * It is valid to have concurrent insertions and deletions protected by per
465 * bucket locks or concurrent RCU protected lookups and traversals.
467 int rhashtable_shrink(struct rhashtable *ht)
469 struct bucket_table *new_tbl, *tbl = rht_dereference(ht->tbl, ht);
470 unsigned int new_hash;
472 ASSERT_RHT_MUTEX(ht);
474 new_tbl = bucket_table_alloc(ht, tbl->size / 2);
478 rcu_assign_pointer(ht->future_tbl, new_tbl);
481 /* Link the first entry in the old bucket to the end of the
482 * bucket in the new table. As entries are concurrently being
483 * added to the new table, lock down the new bucket. As we
484 * always divide the size in half when shrinking, each bucket
485 * in the new table maps to exactly two buckets in the old
488 for (new_hash = 0; new_hash < new_tbl->size; new_hash++) {
489 lock_buckets(new_tbl, tbl, new_hash);
491 rcu_assign_pointer(*bucket_tail(new_tbl, new_hash),
492 tbl->buckets[new_hash]);
493 ASSERT_BUCKET_LOCK(ht, tbl, new_hash + new_tbl->size);
494 rcu_assign_pointer(*bucket_tail(new_tbl, new_hash),
495 tbl->buckets[new_hash + new_tbl->size]);
497 unlock_buckets(new_tbl, tbl, new_hash);
500 /* Publish the new, valid hash table */
501 rcu_assign_pointer(ht->tbl, new_tbl);
502 atomic_dec(&ht->shift);
504 /* Wait for readers. No new readers will have references to the
509 bucket_table_free(tbl);
513 EXPORT_SYMBOL_GPL(rhashtable_shrink);
515 static void rht_deferred_worker(struct work_struct *work)
517 struct rhashtable *ht;
518 struct bucket_table *tbl;
519 struct rhashtable_walker *walker;
521 ht = container_of(work, struct rhashtable, run_work);
522 mutex_lock(&ht->mutex);
523 if (ht->being_destroyed)
526 tbl = rht_dereference(ht->tbl, ht);
528 list_for_each_entry(walker, &ht->walkers, list)
529 walker->resize = true;
531 if (ht->p.grow_decision && ht->p.grow_decision(ht, tbl->size))
532 rhashtable_expand(ht);
533 else if (ht->p.shrink_decision && ht->p.shrink_decision(ht, tbl->size))
534 rhashtable_shrink(ht);
537 mutex_unlock(&ht->mutex);
540 static void rhashtable_probe_expand(struct rhashtable *ht)
542 const struct bucket_table *new_tbl = rht_dereference_rcu(ht->future_tbl, ht);
543 const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
545 /* Only adjust the table if no resizing is currently in progress. */
546 if (tbl == new_tbl && ht->p.grow_decision &&
547 ht->p.grow_decision(ht, tbl->size))
548 schedule_work(&ht->run_work);
551 static void rhashtable_probe_shrink(struct rhashtable *ht)
553 const struct bucket_table *new_tbl = rht_dereference_rcu(ht->future_tbl, ht);
554 const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
556 /* Only adjust the table if no resizing is currently in progress. */
557 if (tbl == new_tbl && ht->p.shrink_decision &&
558 ht->p.shrink_decision(ht, tbl->size))
559 schedule_work(&ht->run_work);
562 static void __rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj,
563 struct bucket_table *tbl, u32 hash)
565 struct rhash_head *head;
567 hash = rht_bucket_index(tbl, hash);
568 head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
570 ASSERT_BUCKET_LOCK(ht, tbl, hash);
572 if (rht_is_a_nulls(head))
573 INIT_RHT_NULLS_HEAD(obj->next, ht, hash);
575 RCU_INIT_POINTER(obj->next, head);
577 rcu_assign_pointer(tbl->buckets[hash], obj);
579 atomic_inc(&ht->nelems);
581 rhashtable_probe_expand(ht);
585 * rhashtable_insert - insert object into hash table
587 * @obj: pointer to hash head inside object
589 * Will take a per bucket spinlock to protect against mutual mutations
590 * on the same bucket. Multiple insertions may occur in parallel unless
591 * they map to the same bucket lock.
593 * It is safe to call this function from atomic context.
595 * Will trigger an automatic deferred table resizing if the size grows
596 * beyond the watermark indicated by grow_decision() which can be passed
597 * to rhashtable_init().
599 void rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj)
601 struct bucket_table *tbl, *old_tbl;
606 tbl = rht_dereference_rcu(ht->future_tbl, ht);
607 old_tbl = rht_dereference_rcu(ht->tbl, ht);
608 hash = obj_raw_hashfn(ht, rht_obj(ht, obj));
610 lock_buckets(tbl, old_tbl, hash);
611 __rhashtable_insert(ht, obj, tbl, hash);
612 unlock_buckets(tbl, old_tbl, hash);
616 EXPORT_SYMBOL_GPL(rhashtable_insert);
619 * rhashtable_remove - remove object from hash table
621 * @obj: pointer to hash head inside object
623 * Since the hash chain is single linked, the removal operation needs to
624 * walk the bucket chain upon removal. The removal operation is thus
625 * considerable slow if the hash table is not correctly sized.
627 * Will automatically shrink the table via rhashtable_expand() if the
628 * shrink_decision function specified at rhashtable_init() returns true.
630 * The caller must ensure that no concurrent table mutations occur. It is
631 * however valid to have concurrent lookups if they are RCU protected.
633 bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *obj)
635 struct bucket_table *tbl, *new_tbl, *old_tbl;
636 struct rhash_head __rcu **pprev;
637 struct rhash_head *he, *he2;
638 unsigned int hash, new_hash;
642 old_tbl = rht_dereference_rcu(ht->tbl, ht);
643 tbl = new_tbl = rht_dereference_rcu(ht->future_tbl, ht);
644 new_hash = obj_raw_hashfn(ht, rht_obj(ht, obj));
646 lock_buckets(new_tbl, old_tbl, new_hash);
648 hash = rht_bucket_index(tbl, new_hash);
649 pprev = &tbl->buckets[hash];
650 rht_for_each(he, tbl, hash) {
656 ASSERT_BUCKET_LOCK(ht, tbl, hash);
658 if (old_tbl->size > new_tbl->size && tbl == old_tbl &&
659 !rht_is_a_nulls(obj->next) &&
660 head_hashfn(ht, tbl, obj->next) != hash) {
661 rcu_assign_pointer(*pprev, (struct rhash_head *) rht_marker(ht, hash));
662 } else if (unlikely(old_tbl->size < new_tbl->size && tbl == new_tbl)) {
663 rht_for_each_continue(he2, obj->next, tbl, hash) {
664 if (head_hashfn(ht, tbl, he2) == hash) {
665 rcu_assign_pointer(*pprev, he2);
670 rcu_assign_pointer(*pprev, (struct rhash_head *) rht_marker(ht, hash));
672 rcu_assign_pointer(*pprev, obj->next);
680 /* The entry may be linked in either 'tbl', 'future_tbl', or both.
681 * 'future_tbl' only exists for a short period of time during
682 * resizing. Thus traversing both is fine and the added cost is
685 if (tbl != old_tbl) {
690 unlock_buckets(new_tbl, old_tbl, new_hash);
693 atomic_dec(&ht->nelems);
694 rhashtable_probe_shrink(ht);
701 EXPORT_SYMBOL_GPL(rhashtable_remove);
703 struct rhashtable_compare_arg {
704 struct rhashtable *ht;
708 static bool rhashtable_compare(void *ptr, void *arg)
710 struct rhashtable_compare_arg *x = arg;
711 struct rhashtable *ht = x->ht;
713 return !memcmp(ptr + ht->p.key_offset, x->key, ht->p.key_len);
717 * rhashtable_lookup - lookup key in hash table
719 * @key: pointer to key
721 * Computes the hash value for the key and traverses the bucket chain looking
722 * for a entry with an identical key. The first matching entry is returned.
724 * This lookup function may only be used for fixed key hash table (key_len
725 * parameter set). It will BUG() if used inappropriately.
727 * Lookups may occur in parallel with hashtable mutations and resizing.
729 void *rhashtable_lookup(struct rhashtable *ht, const void *key)
731 struct rhashtable_compare_arg arg = {
736 BUG_ON(!ht->p.key_len);
738 return rhashtable_lookup_compare(ht, key, &rhashtable_compare, &arg);
740 EXPORT_SYMBOL_GPL(rhashtable_lookup);
743 * rhashtable_lookup_compare - search hash table with compare function
745 * @key: the pointer to the key
746 * @compare: compare function, must return true on match
747 * @arg: argument passed on to compare function
749 * Traverses the bucket chain behind the provided hash value and calls the
750 * specified compare function for each entry.
752 * Lookups may occur in parallel with hashtable mutations and resizing.
754 * Returns the first entry on which the compare function returned true.
756 void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key,
757 bool (*compare)(void *, void *), void *arg)
759 const struct bucket_table *tbl, *old_tbl;
760 struct rhash_head *he;
765 old_tbl = rht_dereference_rcu(ht->tbl, ht);
766 tbl = rht_dereference_rcu(ht->future_tbl, ht);
767 hash = key_hashfn(ht, key, ht->p.key_len);
769 rht_for_each_rcu(he, tbl, rht_bucket_index(tbl, hash)) {
770 if (!compare(rht_obj(ht, he), arg))
773 return rht_obj(ht, he);
776 if (unlikely(tbl != old_tbl)) {
784 EXPORT_SYMBOL_GPL(rhashtable_lookup_compare);
787 * rhashtable_lookup_insert - lookup and insert object into hash table
789 * @obj: pointer to hash head inside object
791 * Locks down the bucket chain in both the old and new table if a resize
792 * is in progress to ensure that writers can't remove from the old table
793 * and can't insert to the new table during the atomic operation of search
794 * and insertion. Searches for duplicates in both the old and new table if
795 * a resize is in progress.
797 * This lookup function may only be used for fixed key hash table (key_len
798 * parameter set). It will BUG() if used inappropriately.
800 * It is safe to call this function from atomic context.
802 * Will trigger an automatic deferred table resizing if the size grows
803 * beyond the watermark indicated by grow_decision() which can be passed
804 * to rhashtable_init().
806 bool rhashtable_lookup_insert(struct rhashtable *ht, struct rhash_head *obj)
808 struct rhashtable_compare_arg arg = {
810 .key = rht_obj(ht, obj) + ht->p.key_offset,
813 BUG_ON(!ht->p.key_len);
815 return rhashtable_lookup_compare_insert(ht, obj, &rhashtable_compare,
818 EXPORT_SYMBOL_GPL(rhashtable_lookup_insert);
821 * rhashtable_lookup_compare_insert - search and insert object to hash table
822 * with compare function
824 * @obj: pointer to hash head inside object
825 * @compare: compare function, must return true on match
826 * @arg: argument passed on to compare function
828 * Locks down the bucket chain in both the old and new table if a resize
829 * is in progress to ensure that writers can't remove from the old table
830 * and can't insert to the new table during the atomic operation of search
831 * and insertion. Searches for duplicates in both the old and new table if
832 * a resize is in progress.
834 * Lookups may occur in parallel with hashtable mutations and resizing.
836 * Will trigger an automatic deferred table resizing if the size grows
837 * beyond the watermark indicated by grow_decision() which can be passed
838 * to rhashtable_init().
840 bool rhashtable_lookup_compare_insert(struct rhashtable *ht,
841 struct rhash_head *obj,
842 bool (*compare)(void *, void *),
845 struct bucket_table *new_tbl, *old_tbl;
849 BUG_ON(!ht->p.key_len);
852 old_tbl = rht_dereference_rcu(ht->tbl, ht);
853 new_tbl = rht_dereference_rcu(ht->future_tbl, ht);
854 new_hash = obj_raw_hashfn(ht, rht_obj(ht, obj));
856 lock_buckets(new_tbl, old_tbl, new_hash);
858 if (rhashtable_lookup_compare(ht, rht_obj(ht, obj) + ht->p.key_offset,
864 __rhashtable_insert(ht, obj, new_tbl, new_hash);
867 unlock_buckets(new_tbl, old_tbl, new_hash);
872 EXPORT_SYMBOL_GPL(rhashtable_lookup_compare_insert);
875 * rhashtable_walk_init - Initialise an iterator
876 * @ht: Table to walk over
877 * @iter: Hash table Iterator
879 * This function prepares a hash table walk.
881 * Note that if you restart a walk after rhashtable_walk_stop you
882 * may see the same object twice. Also, you may miss objects if
883 * there are removals in between rhashtable_walk_stop and the next
884 * call to rhashtable_walk_start.
886 * For a completely stable walk you should construct your own data
887 * structure outside the hash table.
889 * This function may sleep so you must not call it from interrupt
890 * context or with spin locks held.
892 * You must call rhashtable_walk_exit if this function returns
895 int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter)
902 iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL);
906 mutex_lock(&ht->mutex);
907 list_add(&iter->walker->list, &ht->walkers);
908 mutex_unlock(&ht->mutex);
912 EXPORT_SYMBOL_GPL(rhashtable_walk_init);
915 * rhashtable_walk_exit - Free an iterator
916 * @iter: Hash table Iterator
918 * This function frees resources allocated by rhashtable_walk_init.
920 void rhashtable_walk_exit(struct rhashtable_iter *iter)
922 mutex_lock(&iter->ht->mutex);
923 list_del(&iter->walker->list);
924 mutex_unlock(&iter->ht->mutex);
927 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
930 * rhashtable_walk_start - Start a hash table walk
931 * @iter: Hash table iterator
933 * Start a hash table walk. Note that we take the RCU lock in all
934 * cases including when we return an error. So you must always call
935 * rhashtable_walk_stop to clean up.
937 * Returns zero if successful.
939 * Returns -EAGAIN if resize event occured. Note that the iterator
940 * will rewind back to the beginning and you may use it immediately
941 * by calling rhashtable_walk_next.
943 int rhashtable_walk_start(struct rhashtable_iter *iter)
947 if (iter->walker->resize) {
950 iter->walker->resize = false;
956 EXPORT_SYMBOL_GPL(rhashtable_walk_start);
959 * rhashtable_walk_next - Return the next object and advance the iterator
960 * @iter: Hash table iterator
962 * Note that you must call rhashtable_walk_stop when you are finished
965 * Returns the next object or NULL when the end of the table is reached.
967 * Returns -EAGAIN if resize event occured. Note that the iterator
968 * will rewind back to the beginning and you may continue to use it.
970 void *rhashtable_walk_next(struct rhashtable_iter *iter)
972 const struct bucket_table *tbl;
973 struct rhashtable *ht = iter->ht;
974 struct rhash_head *p = iter->p;
977 tbl = rht_dereference_rcu(ht->tbl, ht);
980 p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot);
984 for (; iter->slot < tbl->size; iter->slot++) {
985 int skip = iter->skip;
987 rht_for_each_rcu(p, tbl, iter->slot) {
994 if (!rht_is_a_nulls(p)) {
997 obj = rht_obj(ht, p);
1007 if (iter->walker->resize) {
1011 iter->walker->resize = false;
1012 return ERR_PTR(-EAGAIN);
1017 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
1020 * rhashtable_walk_stop - Finish a hash table walk
1021 * @iter: Hash table iterator
1023 * Finish a hash table walk.
1025 void rhashtable_walk_stop(struct rhashtable_iter *iter)
1030 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
1032 static size_t rounded_hashtable_size(struct rhashtable_params *params)
1034 return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
1035 1UL << params->min_shift);
1039 * rhashtable_init - initialize a new hash table
1040 * @ht: hash table to be initialized
1041 * @params: configuration parameters
1043 * Initializes a new hash table based on the provided configuration
1044 * parameters. A table can be configured either with a variable or
1047 * Configuration Example 1: Fixed length keys
1051 * struct rhash_head node;
1054 * struct rhashtable_params params = {
1055 * .head_offset = offsetof(struct test_obj, node),
1056 * .key_offset = offsetof(struct test_obj, key),
1057 * .key_len = sizeof(int),
1059 * .nulls_base = (1U << RHT_BASE_SHIFT),
1062 * Configuration Example 2: Variable length keys
1065 * struct rhash_head node;
1068 * u32 my_hash_fn(const void *data, u32 seed)
1070 * struct test_obj *obj = data;
1072 * return [... hash ...];
1075 * struct rhashtable_params params = {
1076 * .head_offset = offsetof(struct test_obj, node),
1078 * .obj_hashfn = my_hash_fn,
1081 int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params)
1083 struct bucket_table *tbl;
1086 size = HASH_DEFAULT_SIZE;
1088 if ((params->key_len && !params->hashfn) ||
1089 (!params->key_len && !params->obj_hashfn))
1092 if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
1095 params->min_shift = max_t(size_t, params->min_shift,
1096 ilog2(HASH_MIN_SIZE));
1098 if (params->nelem_hint)
1099 size = rounded_hashtable_size(params);
1101 memset(ht, 0, sizeof(*ht));
1102 mutex_init(&ht->mutex);
1103 memcpy(&ht->p, params, sizeof(*params));
1104 INIT_LIST_HEAD(&ht->walkers);
1106 if (params->locks_mul)
1107 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
1109 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
1111 tbl = bucket_table_alloc(ht, size);
1115 atomic_set(&ht->nelems, 0);
1116 atomic_set(&ht->shift, ilog2(tbl->size));
1117 RCU_INIT_POINTER(ht->tbl, tbl);
1118 RCU_INIT_POINTER(ht->future_tbl, tbl);
1120 if (!ht->p.hash_rnd)
1121 get_random_bytes(&ht->p.hash_rnd, sizeof(ht->p.hash_rnd));
1123 if (ht->p.grow_decision || ht->p.shrink_decision)
1124 INIT_WORK(&ht->run_work, rht_deferred_worker);
1128 EXPORT_SYMBOL_GPL(rhashtable_init);
1131 * rhashtable_destroy - destroy hash table
1132 * @ht: the hash table to destroy
1134 * Frees the bucket array. This function is not rcu safe, therefore the caller
1135 * has to make sure that no resizing may happen by unpublishing the hashtable
1136 * and waiting for the quiescent cycle before releasing the bucket array.
1138 void rhashtable_destroy(struct rhashtable *ht)
1140 ht->being_destroyed = true;
1142 if (ht->p.grow_decision || ht->p.shrink_decision)
1143 cancel_work_sync(&ht->run_work);
1145 mutex_lock(&ht->mutex);
1146 bucket_table_free(rht_dereference(ht->tbl, ht));
1147 mutex_unlock(&ht->mutex);
1149 EXPORT_SYMBOL_GPL(rhashtable_destroy);