xref: /openbmc/linux/lib/rhashtable.c (revision cd4d09ec)
1 /*
2  * Resizable, Scalable, Concurrent Hash Table
3  *
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>
7  *
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
11  *
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.
15  */
16 
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/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/mm.h>
25 #include <linux/jhash.h>
26 #include <linux/random.h>
27 #include <linux/rhashtable.h>
28 #include <linux/err.h>
29 #include <linux/export.h>
30 
31 #define HASH_DEFAULT_SIZE	64UL
32 #define HASH_MIN_SIZE		4U
33 #define BUCKET_LOCKS_PER_CPU   128UL
34 
35 static u32 head_hashfn(struct rhashtable *ht,
36 		       const struct bucket_table *tbl,
37 		       const struct rhash_head *he)
38 {
39 	return rht_head_hashfn(ht, tbl, he, ht->p);
40 }
41 
42 #ifdef CONFIG_PROVE_LOCKING
43 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
44 
45 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
46 {
47 	return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
48 }
49 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
50 
51 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
52 {
53 	spinlock_t *lock = rht_bucket_lock(tbl, hash);
54 
55 	return (debug_locks) ? lockdep_is_held(lock) : 1;
56 }
57 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
58 #else
59 #define ASSERT_RHT_MUTEX(HT)
60 #endif
61 
62 
63 static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl,
64 			      gfp_t gfp)
65 {
66 	unsigned int i, size;
67 #if defined(CONFIG_PROVE_LOCKING)
68 	unsigned int nr_pcpus = 2;
69 #else
70 	unsigned int nr_pcpus = num_possible_cpus();
71 #endif
72 
73 	nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL);
74 	size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
75 
76 	/* Never allocate more than 0.5 locks per bucket */
77 	size = min_t(unsigned int, size, tbl->size >> 1);
78 
79 	if (sizeof(spinlock_t) != 0) {
80 #ifdef CONFIG_NUMA
81 		if (size * sizeof(spinlock_t) > PAGE_SIZE &&
82 		    gfp == GFP_KERNEL)
83 			tbl->locks = vmalloc(size * sizeof(spinlock_t));
84 		else
85 #endif
86 		tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
87 					   gfp);
88 		if (!tbl->locks)
89 			return -ENOMEM;
90 		for (i = 0; i < size; i++)
91 			spin_lock_init(&tbl->locks[i]);
92 	}
93 	tbl->locks_mask = size - 1;
94 
95 	return 0;
96 }
97 
98 static void bucket_table_free(const struct bucket_table *tbl)
99 {
100 	if (tbl)
101 		kvfree(tbl->locks);
102 
103 	kvfree(tbl);
104 }
105 
106 static void bucket_table_free_rcu(struct rcu_head *head)
107 {
108 	bucket_table_free(container_of(head, struct bucket_table, rcu));
109 }
110 
111 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
112 					       size_t nbuckets,
113 					       gfp_t gfp)
114 {
115 	struct bucket_table *tbl = NULL;
116 	size_t size;
117 	int i;
118 
119 	size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
120 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) ||
121 	    gfp != GFP_KERNEL)
122 		tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
123 	if (tbl == NULL && gfp == GFP_KERNEL)
124 		tbl = vzalloc(size);
125 	if (tbl == NULL)
126 		return NULL;
127 
128 	tbl->size = nbuckets;
129 
130 	if (alloc_bucket_locks(ht, tbl, gfp) < 0) {
131 		bucket_table_free(tbl);
132 		return NULL;
133 	}
134 
135 	INIT_LIST_HEAD(&tbl->walkers);
136 
137 	get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
138 
139 	for (i = 0; i < nbuckets; i++)
140 		INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
141 
142 	return tbl;
143 }
144 
145 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
146 						  struct bucket_table *tbl)
147 {
148 	struct bucket_table *new_tbl;
149 
150 	do {
151 		new_tbl = tbl;
152 		tbl = rht_dereference_rcu(tbl->future_tbl, ht);
153 	} while (tbl);
154 
155 	return new_tbl;
156 }
157 
158 static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
159 {
160 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
161 	struct bucket_table *new_tbl = rhashtable_last_table(ht,
162 		rht_dereference_rcu(old_tbl->future_tbl, ht));
163 	struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash];
164 	int err = -ENOENT;
165 	struct rhash_head *head, *next, *entry;
166 	spinlock_t *new_bucket_lock;
167 	unsigned int new_hash;
168 
169 	rht_for_each(entry, old_tbl, old_hash) {
170 		err = 0;
171 		next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
172 
173 		if (rht_is_a_nulls(next))
174 			break;
175 
176 		pprev = &entry->next;
177 	}
178 
179 	if (err)
180 		goto out;
181 
182 	new_hash = head_hashfn(ht, new_tbl, entry);
183 
184 	new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
185 
186 	spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
187 	head = rht_dereference_bucket(new_tbl->buckets[new_hash],
188 				      new_tbl, new_hash);
189 
190 	RCU_INIT_POINTER(entry->next, head);
191 
192 	rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
193 	spin_unlock(new_bucket_lock);
194 
195 	rcu_assign_pointer(*pprev, next);
196 
197 out:
198 	return err;
199 }
200 
201 static void rhashtable_rehash_chain(struct rhashtable *ht,
202 				    unsigned int old_hash)
203 {
204 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
205 	spinlock_t *old_bucket_lock;
206 
207 	old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
208 
209 	spin_lock_bh(old_bucket_lock);
210 	while (!rhashtable_rehash_one(ht, old_hash))
211 		;
212 	old_tbl->rehash++;
213 	spin_unlock_bh(old_bucket_lock);
214 }
215 
216 static int rhashtable_rehash_attach(struct rhashtable *ht,
217 				    struct bucket_table *old_tbl,
218 				    struct bucket_table *new_tbl)
219 {
220 	/* Protect future_tbl using the first bucket lock. */
221 	spin_lock_bh(old_tbl->locks);
222 
223 	/* Did somebody beat us to it? */
224 	if (rcu_access_pointer(old_tbl->future_tbl)) {
225 		spin_unlock_bh(old_tbl->locks);
226 		return -EEXIST;
227 	}
228 
229 	/* Make insertions go into the new, empty table right away. Deletions
230 	 * and lookups will be attempted in both tables until we synchronize.
231 	 */
232 	rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
233 
234 	spin_unlock_bh(old_tbl->locks);
235 
236 	return 0;
237 }
238 
239 static int rhashtable_rehash_table(struct rhashtable *ht)
240 {
241 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
242 	struct bucket_table *new_tbl;
243 	struct rhashtable_walker *walker;
244 	unsigned int old_hash;
245 
246 	new_tbl = rht_dereference(old_tbl->future_tbl, ht);
247 	if (!new_tbl)
248 		return 0;
249 
250 	for (old_hash = 0; old_hash < old_tbl->size; old_hash++)
251 		rhashtable_rehash_chain(ht, old_hash);
252 
253 	/* Publish the new table pointer. */
254 	rcu_assign_pointer(ht->tbl, new_tbl);
255 
256 	spin_lock(&ht->lock);
257 	list_for_each_entry(walker, &old_tbl->walkers, list)
258 		walker->tbl = NULL;
259 	spin_unlock(&ht->lock);
260 
261 	/* Wait for readers. All new readers will see the new
262 	 * table, and thus no references to the old table will
263 	 * remain.
264 	 */
265 	call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
266 
267 	return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
268 }
269 
270 /**
271  * rhashtable_expand - Expand hash table while allowing concurrent lookups
272  * @ht:		the hash table to expand
273  *
274  * A secondary bucket array is allocated and the hash entries are migrated.
275  *
276  * This function may only be called in a context where it is safe to call
277  * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
278  *
279  * The caller must ensure that no concurrent resizing occurs by holding
280  * ht->mutex.
281  *
282  * It is valid to have concurrent insertions and deletions protected by per
283  * bucket locks or concurrent RCU protected lookups and traversals.
284  */
285 static int rhashtable_expand(struct rhashtable *ht)
286 {
287 	struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
288 	int err;
289 
290 	ASSERT_RHT_MUTEX(ht);
291 
292 	old_tbl = rhashtable_last_table(ht, old_tbl);
293 
294 	new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL);
295 	if (new_tbl == NULL)
296 		return -ENOMEM;
297 
298 	err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
299 	if (err)
300 		bucket_table_free(new_tbl);
301 
302 	return err;
303 }
304 
305 /**
306  * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
307  * @ht:		the hash table to shrink
308  *
309  * This function shrinks the hash table to fit, i.e., the smallest
310  * size would not cause it to expand right away automatically.
311  *
312  * The caller must ensure that no concurrent resizing occurs by holding
313  * ht->mutex.
314  *
315  * The caller must ensure that no concurrent table mutations take place.
316  * It is however valid to have concurrent lookups if they are RCU protected.
317  *
318  * It is valid to have concurrent insertions and deletions protected by per
319  * bucket locks or concurrent RCU protected lookups and traversals.
320  */
321 static int rhashtable_shrink(struct rhashtable *ht)
322 {
323 	struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
324 	unsigned int size;
325 	int err;
326 
327 	ASSERT_RHT_MUTEX(ht);
328 
329 	size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2);
330 	if (size < ht->p.min_size)
331 		size = ht->p.min_size;
332 
333 	if (old_tbl->size <= size)
334 		return 0;
335 
336 	if (rht_dereference(old_tbl->future_tbl, ht))
337 		return -EEXIST;
338 
339 	new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
340 	if (new_tbl == NULL)
341 		return -ENOMEM;
342 
343 	err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
344 	if (err)
345 		bucket_table_free(new_tbl);
346 
347 	return err;
348 }
349 
350 static void rht_deferred_worker(struct work_struct *work)
351 {
352 	struct rhashtable *ht;
353 	struct bucket_table *tbl;
354 	int err = 0;
355 
356 	ht = container_of(work, struct rhashtable, run_work);
357 	mutex_lock(&ht->mutex);
358 
359 	tbl = rht_dereference(ht->tbl, ht);
360 	tbl = rhashtable_last_table(ht, tbl);
361 
362 	if (rht_grow_above_75(ht, tbl))
363 		rhashtable_expand(ht);
364 	else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
365 		rhashtable_shrink(ht);
366 
367 	err = rhashtable_rehash_table(ht);
368 
369 	mutex_unlock(&ht->mutex);
370 
371 	if (err)
372 		schedule_work(&ht->run_work);
373 }
374 
375 static bool rhashtable_check_elasticity(struct rhashtable *ht,
376 					struct bucket_table *tbl,
377 					unsigned int hash)
378 {
379 	unsigned int elasticity = ht->elasticity;
380 	struct rhash_head *head;
381 
382 	rht_for_each(head, tbl, hash)
383 		if (!--elasticity)
384 			return true;
385 
386 	return false;
387 }
388 
389 int rhashtable_insert_rehash(struct rhashtable *ht,
390 			     struct bucket_table *tbl)
391 {
392 	struct bucket_table *old_tbl;
393 	struct bucket_table *new_tbl;
394 	unsigned int size;
395 	int err;
396 
397 	old_tbl = rht_dereference_rcu(ht->tbl, ht);
398 
399 	size = tbl->size;
400 
401 	err = -EBUSY;
402 
403 	if (rht_grow_above_75(ht, tbl))
404 		size *= 2;
405 	/* Do not schedule more than one rehash */
406 	else if (old_tbl != tbl)
407 		goto fail;
408 
409 	err = -ENOMEM;
410 
411 	new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
412 	if (new_tbl == NULL)
413 		goto fail;
414 
415 	err = rhashtable_rehash_attach(ht, tbl, new_tbl);
416 	if (err) {
417 		bucket_table_free(new_tbl);
418 		if (err == -EEXIST)
419 			err = 0;
420 	} else
421 		schedule_work(&ht->run_work);
422 
423 	return err;
424 
425 fail:
426 	/* Do not fail the insert if someone else did a rehash. */
427 	if (likely(rcu_dereference_raw(tbl->future_tbl)))
428 		return 0;
429 
430 	/* Schedule async rehash to retry allocation in process context. */
431 	if (err == -ENOMEM)
432 		schedule_work(&ht->run_work);
433 
434 	return err;
435 }
436 EXPORT_SYMBOL_GPL(rhashtable_insert_rehash);
437 
438 struct bucket_table *rhashtable_insert_slow(struct rhashtable *ht,
439 					    const void *key,
440 					    struct rhash_head *obj,
441 					    struct bucket_table *tbl)
442 {
443 	struct rhash_head *head;
444 	unsigned int hash;
445 	int err;
446 
447 	tbl = rhashtable_last_table(ht, tbl);
448 	hash = head_hashfn(ht, tbl, obj);
449 	spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING);
450 
451 	err = -EEXIST;
452 	if (key && rhashtable_lookup_fast(ht, key, ht->p))
453 		goto exit;
454 
455 	err = -E2BIG;
456 	if (unlikely(rht_grow_above_max(ht, tbl)))
457 		goto exit;
458 
459 	err = -EAGAIN;
460 	if (rhashtable_check_elasticity(ht, tbl, hash) ||
461 	    rht_grow_above_100(ht, tbl))
462 		goto exit;
463 
464 	err = 0;
465 
466 	head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
467 
468 	RCU_INIT_POINTER(obj->next, head);
469 
470 	rcu_assign_pointer(tbl->buckets[hash], obj);
471 
472 	atomic_inc(&ht->nelems);
473 
474 exit:
475 	spin_unlock(rht_bucket_lock(tbl, hash));
476 
477 	if (err == 0)
478 		return NULL;
479 	else if (err == -EAGAIN)
480 		return tbl;
481 	else
482 		return ERR_PTR(err);
483 }
484 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
485 
486 /**
487  * rhashtable_walk_init - Initialise an iterator
488  * @ht:		Table to walk over
489  * @iter:	Hash table Iterator
490  *
491  * This function prepares a hash table walk.
492  *
493  * Note that if you restart a walk after rhashtable_walk_stop you
494  * may see the same object twice.  Also, you may miss objects if
495  * there are removals in between rhashtable_walk_stop and the next
496  * call to rhashtable_walk_start.
497  *
498  * For a completely stable walk you should construct your own data
499  * structure outside the hash table.
500  *
501  * This function may sleep so you must not call it from interrupt
502  * context or with spin locks held.
503  *
504  * You must call rhashtable_walk_exit if this function returns
505  * successfully.
506  */
507 int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter)
508 {
509 	iter->ht = ht;
510 	iter->p = NULL;
511 	iter->slot = 0;
512 	iter->skip = 0;
513 
514 	iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL);
515 	if (!iter->walker)
516 		return -ENOMEM;
517 
518 	spin_lock(&ht->lock);
519 	iter->walker->tbl =
520 		rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
521 	list_add(&iter->walker->list, &iter->walker->tbl->walkers);
522 	spin_unlock(&ht->lock);
523 
524 	return 0;
525 }
526 EXPORT_SYMBOL_GPL(rhashtable_walk_init);
527 
528 /**
529  * rhashtable_walk_exit - Free an iterator
530  * @iter:	Hash table Iterator
531  *
532  * This function frees resources allocated by rhashtable_walk_init.
533  */
534 void rhashtable_walk_exit(struct rhashtable_iter *iter)
535 {
536 	spin_lock(&iter->ht->lock);
537 	if (iter->walker->tbl)
538 		list_del(&iter->walker->list);
539 	spin_unlock(&iter->ht->lock);
540 	kfree(iter->walker);
541 }
542 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
543 
544 /**
545  * rhashtable_walk_start - Start a hash table walk
546  * @iter:	Hash table iterator
547  *
548  * Start a hash table walk.  Note that we take the RCU lock in all
549  * cases including when we return an error.  So you must always call
550  * rhashtable_walk_stop to clean up.
551  *
552  * Returns zero if successful.
553  *
554  * Returns -EAGAIN if resize event occured.  Note that the iterator
555  * will rewind back to the beginning and you may use it immediately
556  * by calling rhashtable_walk_next.
557  */
558 int rhashtable_walk_start(struct rhashtable_iter *iter)
559 	__acquires(RCU)
560 {
561 	struct rhashtable *ht = iter->ht;
562 
563 	rcu_read_lock();
564 
565 	spin_lock(&ht->lock);
566 	if (iter->walker->tbl)
567 		list_del(&iter->walker->list);
568 	spin_unlock(&ht->lock);
569 
570 	if (!iter->walker->tbl) {
571 		iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht);
572 		return -EAGAIN;
573 	}
574 
575 	return 0;
576 }
577 EXPORT_SYMBOL_GPL(rhashtable_walk_start);
578 
579 /**
580  * rhashtable_walk_next - Return the next object and advance the iterator
581  * @iter:	Hash table iterator
582  *
583  * Note that you must call rhashtable_walk_stop when you are finished
584  * with the walk.
585  *
586  * Returns the next object or NULL when the end of the table is reached.
587  *
588  * Returns -EAGAIN if resize event occured.  Note that the iterator
589  * will rewind back to the beginning and you may continue to use it.
590  */
591 void *rhashtable_walk_next(struct rhashtable_iter *iter)
592 {
593 	struct bucket_table *tbl = iter->walker->tbl;
594 	struct rhashtable *ht = iter->ht;
595 	struct rhash_head *p = iter->p;
596 
597 	if (p) {
598 		p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot);
599 		goto next;
600 	}
601 
602 	for (; iter->slot < tbl->size; iter->slot++) {
603 		int skip = iter->skip;
604 
605 		rht_for_each_rcu(p, tbl, iter->slot) {
606 			if (!skip)
607 				break;
608 			skip--;
609 		}
610 
611 next:
612 		if (!rht_is_a_nulls(p)) {
613 			iter->skip++;
614 			iter->p = p;
615 			return rht_obj(ht, p);
616 		}
617 
618 		iter->skip = 0;
619 	}
620 
621 	iter->p = NULL;
622 
623 	/* Ensure we see any new tables. */
624 	smp_rmb();
625 
626 	iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht);
627 	if (iter->walker->tbl) {
628 		iter->slot = 0;
629 		iter->skip = 0;
630 		return ERR_PTR(-EAGAIN);
631 	}
632 
633 	return NULL;
634 }
635 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
636 
637 /**
638  * rhashtable_walk_stop - Finish a hash table walk
639  * @iter:	Hash table iterator
640  *
641  * Finish a hash table walk.
642  */
643 void rhashtable_walk_stop(struct rhashtable_iter *iter)
644 	__releases(RCU)
645 {
646 	struct rhashtable *ht;
647 	struct bucket_table *tbl = iter->walker->tbl;
648 
649 	if (!tbl)
650 		goto out;
651 
652 	ht = iter->ht;
653 
654 	spin_lock(&ht->lock);
655 	if (tbl->rehash < tbl->size)
656 		list_add(&iter->walker->list, &tbl->walkers);
657 	else
658 		iter->walker->tbl = NULL;
659 	spin_unlock(&ht->lock);
660 
661 	iter->p = NULL;
662 
663 out:
664 	rcu_read_unlock();
665 }
666 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
667 
668 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
669 {
670 	return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
671 		   (unsigned long)params->min_size);
672 }
673 
674 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
675 {
676 	return jhash2(key, length, seed);
677 }
678 
679 /**
680  * rhashtable_init - initialize a new hash table
681  * @ht:		hash table to be initialized
682  * @params:	configuration parameters
683  *
684  * Initializes a new hash table based on the provided configuration
685  * parameters. A table can be configured either with a variable or
686  * fixed length key:
687  *
688  * Configuration Example 1: Fixed length keys
689  * struct test_obj {
690  *	int			key;
691  *	void *			my_member;
692  *	struct rhash_head	node;
693  * };
694  *
695  * struct rhashtable_params params = {
696  *	.head_offset = offsetof(struct test_obj, node),
697  *	.key_offset = offsetof(struct test_obj, key),
698  *	.key_len = sizeof(int),
699  *	.hashfn = jhash,
700  *	.nulls_base = (1U << RHT_BASE_SHIFT),
701  * };
702  *
703  * Configuration Example 2: Variable length keys
704  * struct test_obj {
705  *	[...]
706  *	struct rhash_head	node;
707  * };
708  *
709  * u32 my_hash_fn(const void *data, u32 len, u32 seed)
710  * {
711  *	struct test_obj *obj = data;
712  *
713  *	return [... hash ...];
714  * }
715  *
716  * struct rhashtable_params params = {
717  *	.head_offset = offsetof(struct test_obj, node),
718  *	.hashfn = jhash,
719  *	.obj_hashfn = my_hash_fn,
720  * };
721  */
722 int rhashtable_init(struct rhashtable *ht,
723 		    const struct rhashtable_params *params)
724 {
725 	struct bucket_table *tbl;
726 	size_t size;
727 
728 	size = HASH_DEFAULT_SIZE;
729 
730 	if ((!params->key_len && !params->obj_hashfn) ||
731 	    (params->obj_hashfn && !params->obj_cmpfn))
732 		return -EINVAL;
733 
734 	if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
735 		return -EINVAL;
736 
737 	memset(ht, 0, sizeof(*ht));
738 	mutex_init(&ht->mutex);
739 	spin_lock_init(&ht->lock);
740 	memcpy(&ht->p, params, sizeof(*params));
741 
742 	if (params->min_size)
743 		ht->p.min_size = roundup_pow_of_two(params->min_size);
744 
745 	if (params->max_size)
746 		ht->p.max_size = rounddown_pow_of_two(params->max_size);
747 
748 	if (params->insecure_max_entries)
749 		ht->p.insecure_max_entries =
750 			rounddown_pow_of_two(params->insecure_max_entries);
751 	else
752 		ht->p.insecure_max_entries = ht->p.max_size * 2;
753 
754 	ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE);
755 
756 	if (params->nelem_hint)
757 		size = rounded_hashtable_size(&ht->p);
758 
759 	/* The maximum (not average) chain length grows with the
760 	 * size of the hash table, at a rate of (log N)/(log log N).
761 	 * The value of 16 is selected so that even if the hash
762 	 * table grew to 2^32 you would not expect the maximum
763 	 * chain length to exceed it unless we are under attack
764 	 * (or extremely unlucky).
765 	 *
766 	 * As this limit is only to detect attacks, we don't need
767 	 * to set it to a lower value as you'd need the chain
768 	 * length to vastly exceed 16 to have any real effect
769 	 * on the system.
770 	 */
771 	if (!params->insecure_elasticity)
772 		ht->elasticity = 16;
773 
774 	if (params->locks_mul)
775 		ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
776 	else
777 		ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
778 
779 	ht->key_len = ht->p.key_len;
780 	if (!params->hashfn) {
781 		ht->p.hashfn = jhash;
782 
783 		if (!(ht->key_len & (sizeof(u32) - 1))) {
784 			ht->key_len /= sizeof(u32);
785 			ht->p.hashfn = rhashtable_jhash2;
786 		}
787 	}
788 
789 	tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
790 	if (tbl == NULL)
791 		return -ENOMEM;
792 
793 	atomic_set(&ht->nelems, 0);
794 
795 	RCU_INIT_POINTER(ht->tbl, tbl);
796 
797 	INIT_WORK(&ht->run_work, rht_deferred_worker);
798 
799 	return 0;
800 }
801 EXPORT_SYMBOL_GPL(rhashtable_init);
802 
803 /**
804  * rhashtable_free_and_destroy - free elements and destroy hash table
805  * @ht:		the hash table to destroy
806  * @free_fn:	callback to release resources of element
807  * @arg:	pointer passed to free_fn
808  *
809  * Stops an eventual async resize. If defined, invokes free_fn for each
810  * element to releasal resources. Please note that RCU protected
811  * readers may still be accessing the elements. Releasing of resources
812  * must occur in a compatible manner. Then frees the bucket array.
813  *
814  * This function will eventually sleep to wait for an async resize
815  * to complete. The caller is responsible that no further write operations
816  * occurs in parallel.
817  */
818 void rhashtable_free_and_destroy(struct rhashtable *ht,
819 				 void (*free_fn)(void *ptr, void *arg),
820 				 void *arg)
821 {
822 	const struct bucket_table *tbl;
823 	unsigned int i;
824 
825 	cancel_work_sync(&ht->run_work);
826 
827 	mutex_lock(&ht->mutex);
828 	tbl = rht_dereference(ht->tbl, ht);
829 	if (free_fn) {
830 		for (i = 0; i < tbl->size; i++) {
831 			struct rhash_head *pos, *next;
832 
833 			for (pos = rht_dereference(tbl->buckets[i], ht),
834 			     next = !rht_is_a_nulls(pos) ?
835 					rht_dereference(pos->next, ht) : NULL;
836 			     !rht_is_a_nulls(pos);
837 			     pos = next,
838 			     next = !rht_is_a_nulls(pos) ?
839 					rht_dereference(pos->next, ht) : NULL)
840 				free_fn(rht_obj(ht, pos), arg);
841 		}
842 	}
843 
844 	bucket_table_free(tbl);
845 	mutex_unlock(&ht->mutex);
846 }
847 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
848 
849 void rhashtable_destroy(struct rhashtable *ht)
850 {
851 	return rhashtable_free_and_destroy(ht, NULL, NULL);
852 }
853 EXPORT_SYMBOL_GPL(rhashtable_destroy);
854