xref: /openbmc/linux/lib/rhashtable.c (revision b593bce5)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Resizable, Scalable, Concurrent Hash Table
4  *
5  * Copyright (c) 2015 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>
8  *
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
12  */
13 
14 #include <linux/atomic.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/log2.h>
18 #include <linux/sched.h>
19 #include <linux/rculist.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/mm.h>
23 #include <linux/jhash.h>
24 #include <linux/random.h>
25 #include <linux/rhashtable.h>
26 #include <linux/err.h>
27 #include <linux/export.h>
28 
29 #define HASH_DEFAULT_SIZE	64UL
30 #define HASH_MIN_SIZE		4U
31 
32 union nested_table {
33 	union nested_table __rcu *table;
34 	struct rhash_lock_head *bucket;
35 };
36 
37 static u32 head_hashfn(struct rhashtable *ht,
38 		       const struct bucket_table *tbl,
39 		       const struct rhash_head *he)
40 {
41 	return rht_head_hashfn(ht, tbl, he, ht->p);
42 }
43 
44 #ifdef CONFIG_PROVE_LOCKING
45 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
46 
47 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
48 {
49 	return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
50 }
51 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
52 
53 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
54 {
55 	if (!debug_locks)
56 		return 1;
57 	if (unlikely(tbl->nest))
58 		return 1;
59 	return bit_spin_is_locked(0, (unsigned long *)&tbl->buckets[hash]);
60 }
61 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
62 #else
63 #define ASSERT_RHT_MUTEX(HT)
64 #endif
65 
66 static void nested_table_free(union nested_table *ntbl, unsigned int size)
67 {
68 	const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
69 	const unsigned int len = 1 << shift;
70 	unsigned int i;
71 
72 	ntbl = rcu_dereference_raw(ntbl->table);
73 	if (!ntbl)
74 		return;
75 
76 	if (size > len) {
77 		size >>= shift;
78 		for (i = 0; i < len; i++)
79 			nested_table_free(ntbl + i, size);
80 	}
81 
82 	kfree(ntbl);
83 }
84 
85 static void nested_bucket_table_free(const struct bucket_table *tbl)
86 {
87 	unsigned int size = tbl->size >> tbl->nest;
88 	unsigned int len = 1 << tbl->nest;
89 	union nested_table *ntbl;
90 	unsigned int i;
91 
92 	ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
93 
94 	for (i = 0; i < len; i++)
95 		nested_table_free(ntbl + i, size);
96 
97 	kfree(ntbl);
98 }
99 
100 static void bucket_table_free(const struct bucket_table *tbl)
101 {
102 	if (tbl->nest)
103 		nested_bucket_table_free(tbl);
104 
105 	kvfree(tbl);
106 }
107 
108 static void bucket_table_free_rcu(struct rcu_head *head)
109 {
110 	bucket_table_free(container_of(head, struct bucket_table, rcu));
111 }
112 
113 static union nested_table *nested_table_alloc(struct rhashtable *ht,
114 					      union nested_table __rcu **prev,
115 					      bool leaf)
116 {
117 	union nested_table *ntbl;
118 	int i;
119 
120 	ntbl = rcu_dereference(*prev);
121 	if (ntbl)
122 		return ntbl;
123 
124 	ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC);
125 
126 	if (ntbl && leaf) {
127 		for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0]); i++)
128 			INIT_RHT_NULLS_HEAD(ntbl[i].bucket);
129 	}
130 
131 	if (cmpxchg((union nested_table **)prev, NULL, ntbl) == NULL)
132 		return ntbl;
133 	/* Raced with another thread. */
134 	kfree(ntbl);
135 	return rcu_dereference(*prev);
136 }
137 
138 static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht,
139 						      size_t nbuckets,
140 						      gfp_t gfp)
141 {
142 	const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
143 	struct bucket_table *tbl;
144 	size_t size;
145 
146 	if (nbuckets < (1 << (shift + 1)))
147 		return NULL;
148 
149 	size = sizeof(*tbl) + sizeof(tbl->buckets[0]);
150 
151 	tbl = kzalloc(size, gfp);
152 	if (!tbl)
153 		return NULL;
154 
155 	if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets,
156 				false)) {
157 		kfree(tbl);
158 		return NULL;
159 	}
160 
161 	tbl->nest = (ilog2(nbuckets) - 1) % shift + 1;
162 
163 	return tbl;
164 }
165 
166 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
167 					       size_t nbuckets,
168 					       gfp_t gfp)
169 {
170 	struct bucket_table *tbl = NULL;
171 	size_t size;
172 	int i;
173 	static struct lock_class_key __key;
174 
175 	tbl = kvzalloc(struct_size(tbl, buckets, nbuckets), gfp);
176 
177 	size = nbuckets;
178 
179 	if (tbl == NULL && (gfp & ~__GFP_NOFAIL) != GFP_KERNEL) {
180 		tbl = nested_bucket_table_alloc(ht, nbuckets, gfp);
181 		nbuckets = 0;
182 	}
183 
184 	if (tbl == NULL)
185 		return NULL;
186 
187 	lockdep_init_map(&tbl->dep_map, "rhashtable_bucket", &__key, 0);
188 
189 	tbl->size = size;
190 
191 	rcu_head_init(&tbl->rcu);
192 	INIT_LIST_HEAD(&tbl->walkers);
193 
194 	tbl->hash_rnd = get_random_u32();
195 
196 	for (i = 0; i < nbuckets; i++)
197 		INIT_RHT_NULLS_HEAD(tbl->buckets[i]);
198 
199 	return tbl;
200 }
201 
202 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
203 						  struct bucket_table *tbl)
204 {
205 	struct bucket_table *new_tbl;
206 
207 	do {
208 		new_tbl = tbl;
209 		tbl = rht_dereference_rcu(tbl->future_tbl, ht);
210 	} while (tbl);
211 
212 	return new_tbl;
213 }
214 
215 static int rhashtable_rehash_one(struct rhashtable *ht,
216 				 struct rhash_lock_head **bkt,
217 				 unsigned int old_hash)
218 {
219 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
220 	struct bucket_table *new_tbl = rhashtable_last_table(ht, old_tbl);
221 	int err = -EAGAIN;
222 	struct rhash_head *head, *next, *entry;
223 	struct rhash_head __rcu **pprev = NULL;
224 	unsigned int new_hash;
225 
226 	if (new_tbl->nest)
227 		goto out;
228 
229 	err = -ENOENT;
230 
231 	rht_for_each_from(entry, rht_ptr(bkt, old_tbl, old_hash),
232 			  old_tbl, old_hash) {
233 		err = 0;
234 		next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
235 
236 		if (rht_is_a_nulls(next))
237 			break;
238 
239 		pprev = &entry->next;
240 	}
241 
242 	if (err)
243 		goto out;
244 
245 	new_hash = head_hashfn(ht, new_tbl, entry);
246 
247 	rht_lock_nested(new_tbl, &new_tbl->buckets[new_hash], SINGLE_DEPTH_NESTING);
248 
249 	head = rht_ptr(new_tbl->buckets + new_hash, new_tbl, new_hash);
250 
251 	RCU_INIT_POINTER(entry->next, head);
252 
253 	rht_assign_unlock(new_tbl, &new_tbl->buckets[new_hash], entry);
254 
255 	if (pprev)
256 		rcu_assign_pointer(*pprev, next);
257 	else
258 		/* Need to preserved the bit lock. */
259 		rht_assign_locked(bkt, next);
260 
261 out:
262 	return err;
263 }
264 
265 static int rhashtable_rehash_chain(struct rhashtable *ht,
266 				    unsigned int old_hash)
267 {
268 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
269 	struct rhash_lock_head **bkt = rht_bucket_var(old_tbl, old_hash);
270 	int err;
271 
272 	if (!bkt)
273 		return 0;
274 	rht_lock(old_tbl, bkt);
275 
276 	while (!(err = rhashtable_rehash_one(ht, bkt, old_hash)))
277 		;
278 
279 	if (err == -ENOENT)
280 		err = 0;
281 	rht_unlock(old_tbl, bkt);
282 
283 	return err;
284 }
285 
286 static int rhashtable_rehash_attach(struct rhashtable *ht,
287 				    struct bucket_table *old_tbl,
288 				    struct bucket_table *new_tbl)
289 {
290 	/* Make insertions go into the new, empty table right away. Deletions
291 	 * and lookups will be attempted in both tables until we synchronize.
292 	 * As cmpxchg() provides strong barriers, we do not need
293 	 * rcu_assign_pointer().
294 	 */
295 
296 	if (cmpxchg((struct bucket_table **)&old_tbl->future_tbl, NULL,
297 		    new_tbl) != NULL)
298 		return -EEXIST;
299 
300 	return 0;
301 }
302 
303 static int rhashtable_rehash_table(struct rhashtable *ht)
304 {
305 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
306 	struct bucket_table *new_tbl;
307 	struct rhashtable_walker *walker;
308 	unsigned int old_hash;
309 	int err;
310 
311 	new_tbl = rht_dereference(old_tbl->future_tbl, ht);
312 	if (!new_tbl)
313 		return 0;
314 
315 	for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
316 		err = rhashtable_rehash_chain(ht, old_hash);
317 		if (err)
318 			return err;
319 		cond_resched();
320 	}
321 
322 	/* Publish the new table pointer. */
323 	rcu_assign_pointer(ht->tbl, new_tbl);
324 
325 	spin_lock(&ht->lock);
326 	list_for_each_entry(walker, &old_tbl->walkers, list)
327 		walker->tbl = NULL;
328 
329 	/* Wait for readers. All new readers will see the new
330 	 * table, and thus no references to the old table will
331 	 * remain.
332 	 * We do this inside the locked region so that
333 	 * rhashtable_walk_stop() can use rcu_head_after_call_rcu()
334 	 * to check if it should not re-link the table.
335 	 */
336 	call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
337 	spin_unlock(&ht->lock);
338 
339 	return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
340 }
341 
342 static int rhashtable_rehash_alloc(struct rhashtable *ht,
343 				   struct bucket_table *old_tbl,
344 				   unsigned int size)
345 {
346 	struct bucket_table *new_tbl;
347 	int err;
348 
349 	ASSERT_RHT_MUTEX(ht);
350 
351 	new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
352 	if (new_tbl == NULL)
353 		return -ENOMEM;
354 
355 	err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
356 	if (err)
357 		bucket_table_free(new_tbl);
358 
359 	return err;
360 }
361 
362 /**
363  * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
364  * @ht:		the hash table to shrink
365  *
366  * This function shrinks the hash table to fit, i.e., the smallest
367  * size would not cause it to expand right away automatically.
368  *
369  * The caller must ensure that no concurrent resizing occurs by holding
370  * ht->mutex.
371  *
372  * The caller must ensure that no concurrent table mutations take place.
373  * It is however valid to have concurrent lookups if they are RCU protected.
374  *
375  * It is valid to have concurrent insertions and deletions protected by per
376  * bucket locks or concurrent RCU protected lookups and traversals.
377  */
378 static int rhashtable_shrink(struct rhashtable *ht)
379 {
380 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
381 	unsigned int nelems = atomic_read(&ht->nelems);
382 	unsigned int size = 0;
383 
384 	if (nelems)
385 		size = roundup_pow_of_two(nelems * 3 / 2);
386 	if (size < ht->p.min_size)
387 		size = ht->p.min_size;
388 
389 	if (old_tbl->size <= size)
390 		return 0;
391 
392 	if (rht_dereference(old_tbl->future_tbl, ht))
393 		return -EEXIST;
394 
395 	return rhashtable_rehash_alloc(ht, old_tbl, size);
396 }
397 
398 static void rht_deferred_worker(struct work_struct *work)
399 {
400 	struct rhashtable *ht;
401 	struct bucket_table *tbl;
402 	int err = 0;
403 
404 	ht = container_of(work, struct rhashtable, run_work);
405 	mutex_lock(&ht->mutex);
406 
407 	tbl = rht_dereference(ht->tbl, ht);
408 	tbl = rhashtable_last_table(ht, tbl);
409 
410 	if (rht_grow_above_75(ht, tbl))
411 		err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2);
412 	else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
413 		err = rhashtable_shrink(ht);
414 	else if (tbl->nest)
415 		err = rhashtable_rehash_alloc(ht, tbl, tbl->size);
416 
417 	if (!err || err == -EEXIST) {
418 		int nerr;
419 
420 		nerr = rhashtable_rehash_table(ht);
421 		err = err ?: nerr;
422 	}
423 
424 	mutex_unlock(&ht->mutex);
425 
426 	if (err)
427 		schedule_work(&ht->run_work);
428 }
429 
430 static int rhashtable_insert_rehash(struct rhashtable *ht,
431 				    struct bucket_table *tbl)
432 {
433 	struct bucket_table *old_tbl;
434 	struct bucket_table *new_tbl;
435 	unsigned int size;
436 	int err;
437 
438 	old_tbl = rht_dereference_rcu(ht->tbl, ht);
439 
440 	size = tbl->size;
441 
442 	err = -EBUSY;
443 
444 	if (rht_grow_above_75(ht, tbl))
445 		size *= 2;
446 	/* Do not schedule more than one rehash */
447 	else if (old_tbl != tbl)
448 		goto fail;
449 
450 	err = -ENOMEM;
451 
452 	new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC | __GFP_NOWARN);
453 	if (new_tbl == NULL)
454 		goto fail;
455 
456 	err = rhashtable_rehash_attach(ht, tbl, new_tbl);
457 	if (err) {
458 		bucket_table_free(new_tbl);
459 		if (err == -EEXIST)
460 			err = 0;
461 	} else
462 		schedule_work(&ht->run_work);
463 
464 	return err;
465 
466 fail:
467 	/* Do not fail the insert if someone else did a rehash. */
468 	if (likely(rcu_access_pointer(tbl->future_tbl)))
469 		return 0;
470 
471 	/* Schedule async rehash to retry allocation in process context. */
472 	if (err == -ENOMEM)
473 		schedule_work(&ht->run_work);
474 
475 	return err;
476 }
477 
478 static void *rhashtable_lookup_one(struct rhashtable *ht,
479 				   struct rhash_lock_head **bkt,
480 				   struct bucket_table *tbl, unsigned int hash,
481 				   const void *key, struct rhash_head *obj)
482 {
483 	struct rhashtable_compare_arg arg = {
484 		.ht = ht,
485 		.key = key,
486 	};
487 	struct rhash_head __rcu **pprev = NULL;
488 	struct rhash_head *head;
489 	int elasticity;
490 
491 	elasticity = RHT_ELASTICITY;
492 	rht_for_each_from(head, rht_ptr(bkt, tbl, hash), tbl, hash) {
493 		struct rhlist_head *list;
494 		struct rhlist_head *plist;
495 
496 		elasticity--;
497 		if (!key ||
498 		    (ht->p.obj_cmpfn ?
499 		     ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
500 		     rhashtable_compare(&arg, rht_obj(ht, head)))) {
501 			pprev = &head->next;
502 			continue;
503 		}
504 
505 		if (!ht->rhlist)
506 			return rht_obj(ht, head);
507 
508 		list = container_of(obj, struct rhlist_head, rhead);
509 		plist = container_of(head, struct rhlist_head, rhead);
510 
511 		RCU_INIT_POINTER(list->next, plist);
512 		head = rht_dereference_bucket(head->next, tbl, hash);
513 		RCU_INIT_POINTER(list->rhead.next, head);
514 		if (pprev)
515 			rcu_assign_pointer(*pprev, obj);
516 		else
517 			/* Need to preserve the bit lock */
518 			rht_assign_locked(bkt, obj);
519 
520 		return NULL;
521 	}
522 
523 	if (elasticity <= 0)
524 		return ERR_PTR(-EAGAIN);
525 
526 	return ERR_PTR(-ENOENT);
527 }
528 
529 static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
530 						  struct rhash_lock_head **bkt,
531 						  struct bucket_table *tbl,
532 						  unsigned int hash,
533 						  struct rhash_head *obj,
534 						  void *data)
535 {
536 	struct bucket_table *new_tbl;
537 	struct rhash_head *head;
538 
539 	if (!IS_ERR_OR_NULL(data))
540 		return ERR_PTR(-EEXIST);
541 
542 	if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
543 		return ERR_CAST(data);
544 
545 	new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
546 	if (new_tbl)
547 		return new_tbl;
548 
549 	if (PTR_ERR(data) != -ENOENT)
550 		return ERR_CAST(data);
551 
552 	if (unlikely(rht_grow_above_max(ht, tbl)))
553 		return ERR_PTR(-E2BIG);
554 
555 	if (unlikely(rht_grow_above_100(ht, tbl)))
556 		return ERR_PTR(-EAGAIN);
557 
558 	head = rht_ptr(bkt, tbl, hash);
559 
560 	RCU_INIT_POINTER(obj->next, head);
561 	if (ht->rhlist) {
562 		struct rhlist_head *list;
563 
564 		list = container_of(obj, struct rhlist_head, rhead);
565 		RCU_INIT_POINTER(list->next, NULL);
566 	}
567 
568 	/* bkt is always the head of the list, so it holds
569 	 * the lock, which we need to preserve
570 	 */
571 	rht_assign_locked(bkt, obj);
572 
573 	atomic_inc(&ht->nelems);
574 	if (rht_grow_above_75(ht, tbl))
575 		schedule_work(&ht->run_work);
576 
577 	return NULL;
578 }
579 
580 static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
581 				   struct rhash_head *obj)
582 {
583 	struct bucket_table *new_tbl;
584 	struct bucket_table *tbl;
585 	struct rhash_lock_head **bkt;
586 	unsigned int hash;
587 	void *data;
588 
589 	new_tbl = rcu_dereference(ht->tbl);
590 
591 	do {
592 		tbl = new_tbl;
593 		hash = rht_head_hashfn(ht, tbl, obj, ht->p);
594 		if (rcu_access_pointer(tbl->future_tbl))
595 			/* Failure is OK */
596 			bkt = rht_bucket_var(tbl, hash);
597 		else
598 			bkt = rht_bucket_insert(ht, tbl, hash);
599 		if (bkt == NULL) {
600 			new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
601 			data = ERR_PTR(-EAGAIN);
602 		} else {
603 			rht_lock(tbl, bkt);
604 			data = rhashtable_lookup_one(ht, bkt, tbl,
605 						     hash, key, obj);
606 			new_tbl = rhashtable_insert_one(ht, bkt, tbl,
607 							hash, obj, data);
608 			if (PTR_ERR(new_tbl) != -EEXIST)
609 				data = ERR_CAST(new_tbl);
610 
611 			rht_unlock(tbl, bkt);
612 		}
613 	} while (!IS_ERR_OR_NULL(new_tbl));
614 
615 	if (PTR_ERR(data) == -EAGAIN)
616 		data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
617 			       -EAGAIN);
618 
619 	return data;
620 }
621 
622 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
623 			     struct rhash_head *obj)
624 {
625 	void *data;
626 
627 	do {
628 		rcu_read_lock();
629 		data = rhashtable_try_insert(ht, key, obj);
630 		rcu_read_unlock();
631 	} while (PTR_ERR(data) == -EAGAIN);
632 
633 	return data;
634 }
635 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
636 
637 /**
638  * rhashtable_walk_enter - Initialise an iterator
639  * @ht:		Table to walk over
640  * @iter:	Hash table Iterator
641  *
642  * This function prepares a hash table walk.
643  *
644  * Note that if you restart a walk after rhashtable_walk_stop you
645  * may see the same object twice.  Also, you may miss objects if
646  * there are removals in between rhashtable_walk_stop and the next
647  * call to rhashtable_walk_start.
648  *
649  * For a completely stable walk you should construct your own data
650  * structure outside the hash table.
651  *
652  * This function may be called from any process context, including
653  * non-preemptable context, but cannot be called from softirq or
654  * hardirq context.
655  *
656  * You must call rhashtable_walk_exit after this function returns.
657  */
658 void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
659 {
660 	iter->ht = ht;
661 	iter->p = NULL;
662 	iter->slot = 0;
663 	iter->skip = 0;
664 	iter->end_of_table = 0;
665 
666 	spin_lock(&ht->lock);
667 	iter->walker.tbl =
668 		rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
669 	list_add(&iter->walker.list, &iter->walker.tbl->walkers);
670 	spin_unlock(&ht->lock);
671 }
672 EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
673 
674 /**
675  * rhashtable_walk_exit - Free an iterator
676  * @iter:	Hash table Iterator
677  *
678  * This function frees resources allocated by rhashtable_walk_enter.
679  */
680 void rhashtable_walk_exit(struct rhashtable_iter *iter)
681 {
682 	spin_lock(&iter->ht->lock);
683 	if (iter->walker.tbl)
684 		list_del(&iter->walker.list);
685 	spin_unlock(&iter->ht->lock);
686 }
687 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
688 
689 /**
690  * rhashtable_walk_start_check - Start a hash table walk
691  * @iter:	Hash table iterator
692  *
693  * Start a hash table walk at the current iterator position.  Note that we take
694  * the RCU lock in all cases including when we return an error.  So you must
695  * always call rhashtable_walk_stop to clean up.
696  *
697  * Returns zero if successful.
698  *
699  * Returns -EAGAIN if resize event occured.  Note that the iterator
700  * will rewind back to the beginning and you may use it immediately
701  * by calling rhashtable_walk_next.
702  *
703  * rhashtable_walk_start is defined as an inline variant that returns
704  * void. This is preferred in cases where the caller would ignore
705  * resize events and always continue.
706  */
707 int rhashtable_walk_start_check(struct rhashtable_iter *iter)
708 	__acquires(RCU)
709 {
710 	struct rhashtable *ht = iter->ht;
711 	bool rhlist = ht->rhlist;
712 
713 	rcu_read_lock();
714 
715 	spin_lock(&ht->lock);
716 	if (iter->walker.tbl)
717 		list_del(&iter->walker.list);
718 	spin_unlock(&ht->lock);
719 
720 	if (iter->end_of_table)
721 		return 0;
722 	if (!iter->walker.tbl) {
723 		iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
724 		iter->slot = 0;
725 		iter->skip = 0;
726 		return -EAGAIN;
727 	}
728 
729 	if (iter->p && !rhlist) {
730 		/*
731 		 * We need to validate that 'p' is still in the table, and
732 		 * if so, update 'skip'
733 		 */
734 		struct rhash_head *p;
735 		int skip = 0;
736 		rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
737 			skip++;
738 			if (p == iter->p) {
739 				iter->skip = skip;
740 				goto found;
741 			}
742 		}
743 		iter->p = NULL;
744 	} else if (iter->p && rhlist) {
745 		/* Need to validate that 'list' is still in the table, and
746 		 * if so, update 'skip' and 'p'.
747 		 */
748 		struct rhash_head *p;
749 		struct rhlist_head *list;
750 		int skip = 0;
751 		rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
752 			for (list = container_of(p, struct rhlist_head, rhead);
753 			     list;
754 			     list = rcu_dereference(list->next)) {
755 				skip++;
756 				if (list == iter->list) {
757 					iter->p = p;
758 					iter->skip = skip;
759 					goto found;
760 				}
761 			}
762 		}
763 		iter->p = NULL;
764 	}
765 found:
766 	return 0;
767 }
768 EXPORT_SYMBOL_GPL(rhashtable_walk_start_check);
769 
770 /**
771  * __rhashtable_walk_find_next - Find the next element in a table (or the first
772  * one in case of a new walk).
773  *
774  * @iter:	Hash table iterator
775  *
776  * Returns the found object or NULL when the end of the table is reached.
777  *
778  * Returns -EAGAIN if resize event occurred.
779  */
780 static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter)
781 {
782 	struct bucket_table *tbl = iter->walker.tbl;
783 	struct rhlist_head *list = iter->list;
784 	struct rhashtable *ht = iter->ht;
785 	struct rhash_head *p = iter->p;
786 	bool rhlist = ht->rhlist;
787 
788 	if (!tbl)
789 		return NULL;
790 
791 	for (; iter->slot < tbl->size; iter->slot++) {
792 		int skip = iter->skip;
793 
794 		rht_for_each_rcu(p, tbl, iter->slot) {
795 			if (rhlist) {
796 				list = container_of(p, struct rhlist_head,
797 						    rhead);
798 				do {
799 					if (!skip)
800 						goto next;
801 					skip--;
802 					list = rcu_dereference(list->next);
803 				} while (list);
804 
805 				continue;
806 			}
807 			if (!skip)
808 				break;
809 			skip--;
810 		}
811 
812 next:
813 		if (!rht_is_a_nulls(p)) {
814 			iter->skip++;
815 			iter->p = p;
816 			iter->list = list;
817 			return rht_obj(ht, rhlist ? &list->rhead : p);
818 		}
819 
820 		iter->skip = 0;
821 	}
822 
823 	iter->p = NULL;
824 
825 	/* Ensure we see any new tables. */
826 	smp_rmb();
827 
828 	iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht);
829 	if (iter->walker.tbl) {
830 		iter->slot = 0;
831 		iter->skip = 0;
832 		return ERR_PTR(-EAGAIN);
833 	} else {
834 		iter->end_of_table = true;
835 	}
836 
837 	return NULL;
838 }
839 
840 /**
841  * rhashtable_walk_next - Return the next object and advance the iterator
842  * @iter:	Hash table iterator
843  *
844  * Note that you must call rhashtable_walk_stop when you are finished
845  * with the walk.
846  *
847  * Returns the next object or NULL when the end of the table is reached.
848  *
849  * Returns -EAGAIN if resize event occurred.  Note that the iterator
850  * will rewind back to the beginning and you may continue to use it.
851  */
852 void *rhashtable_walk_next(struct rhashtable_iter *iter)
853 {
854 	struct rhlist_head *list = iter->list;
855 	struct rhashtable *ht = iter->ht;
856 	struct rhash_head *p = iter->p;
857 	bool rhlist = ht->rhlist;
858 
859 	if (p) {
860 		if (!rhlist || !(list = rcu_dereference(list->next))) {
861 			p = rcu_dereference(p->next);
862 			list = container_of(p, struct rhlist_head, rhead);
863 		}
864 		if (!rht_is_a_nulls(p)) {
865 			iter->skip++;
866 			iter->p = p;
867 			iter->list = list;
868 			return rht_obj(ht, rhlist ? &list->rhead : p);
869 		}
870 
871 		/* At the end of this slot, switch to next one and then find
872 		 * next entry from that point.
873 		 */
874 		iter->skip = 0;
875 		iter->slot++;
876 	}
877 
878 	return __rhashtable_walk_find_next(iter);
879 }
880 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
881 
882 /**
883  * rhashtable_walk_peek - Return the next object but don't advance the iterator
884  * @iter:	Hash table iterator
885  *
886  * Returns the next object or NULL when the end of the table is reached.
887  *
888  * Returns -EAGAIN if resize event occurred.  Note that the iterator
889  * will rewind back to the beginning and you may continue to use it.
890  */
891 void *rhashtable_walk_peek(struct rhashtable_iter *iter)
892 {
893 	struct rhlist_head *list = iter->list;
894 	struct rhashtable *ht = iter->ht;
895 	struct rhash_head *p = iter->p;
896 
897 	if (p)
898 		return rht_obj(ht, ht->rhlist ? &list->rhead : p);
899 
900 	/* No object found in current iter, find next one in the table. */
901 
902 	if (iter->skip) {
903 		/* A nonzero skip value points to the next entry in the table
904 		 * beyond that last one that was found. Decrement skip so
905 		 * we find the current value. __rhashtable_walk_find_next
906 		 * will restore the original value of skip assuming that
907 		 * the table hasn't changed.
908 		 */
909 		iter->skip--;
910 	}
911 
912 	return __rhashtable_walk_find_next(iter);
913 }
914 EXPORT_SYMBOL_GPL(rhashtable_walk_peek);
915 
916 /**
917  * rhashtable_walk_stop - Finish a hash table walk
918  * @iter:	Hash table iterator
919  *
920  * Finish a hash table walk.  Does not reset the iterator to the start of the
921  * hash table.
922  */
923 void rhashtable_walk_stop(struct rhashtable_iter *iter)
924 	__releases(RCU)
925 {
926 	struct rhashtable *ht;
927 	struct bucket_table *tbl = iter->walker.tbl;
928 
929 	if (!tbl)
930 		goto out;
931 
932 	ht = iter->ht;
933 
934 	spin_lock(&ht->lock);
935 	if (rcu_head_after_call_rcu(&tbl->rcu, bucket_table_free_rcu))
936 		/* This bucket table is being freed, don't re-link it. */
937 		iter->walker.tbl = NULL;
938 	else
939 		list_add(&iter->walker.list, &tbl->walkers);
940 	spin_unlock(&ht->lock);
941 
942 out:
943 	rcu_read_unlock();
944 }
945 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
946 
947 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
948 {
949 	size_t retsize;
950 
951 	if (params->nelem_hint)
952 		retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
953 			      (unsigned long)params->min_size);
954 	else
955 		retsize = max(HASH_DEFAULT_SIZE,
956 			      (unsigned long)params->min_size);
957 
958 	return retsize;
959 }
960 
961 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
962 {
963 	return jhash2(key, length, seed);
964 }
965 
966 /**
967  * rhashtable_init - initialize a new hash table
968  * @ht:		hash table to be initialized
969  * @params:	configuration parameters
970  *
971  * Initializes a new hash table based on the provided configuration
972  * parameters. A table can be configured either with a variable or
973  * fixed length key:
974  *
975  * Configuration Example 1: Fixed length keys
976  * struct test_obj {
977  *	int			key;
978  *	void *			my_member;
979  *	struct rhash_head	node;
980  * };
981  *
982  * struct rhashtable_params params = {
983  *	.head_offset = offsetof(struct test_obj, node),
984  *	.key_offset = offsetof(struct test_obj, key),
985  *	.key_len = sizeof(int),
986  *	.hashfn = jhash,
987  * };
988  *
989  * Configuration Example 2: Variable length keys
990  * struct test_obj {
991  *	[...]
992  *	struct rhash_head	node;
993  * };
994  *
995  * u32 my_hash_fn(const void *data, u32 len, u32 seed)
996  * {
997  *	struct test_obj *obj = data;
998  *
999  *	return [... hash ...];
1000  * }
1001  *
1002  * struct rhashtable_params params = {
1003  *	.head_offset = offsetof(struct test_obj, node),
1004  *	.hashfn = jhash,
1005  *	.obj_hashfn = my_hash_fn,
1006  * };
1007  */
1008 int rhashtable_init(struct rhashtable *ht,
1009 		    const struct rhashtable_params *params)
1010 {
1011 	struct bucket_table *tbl;
1012 	size_t size;
1013 
1014 	if ((!params->key_len && !params->obj_hashfn) ||
1015 	    (params->obj_hashfn && !params->obj_cmpfn))
1016 		return -EINVAL;
1017 
1018 	memset(ht, 0, sizeof(*ht));
1019 	mutex_init(&ht->mutex);
1020 	spin_lock_init(&ht->lock);
1021 	memcpy(&ht->p, params, sizeof(*params));
1022 
1023 	if (params->min_size)
1024 		ht->p.min_size = roundup_pow_of_two(params->min_size);
1025 
1026 	/* Cap total entries at 2^31 to avoid nelems overflow. */
1027 	ht->max_elems = 1u << 31;
1028 
1029 	if (params->max_size) {
1030 		ht->p.max_size = rounddown_pow_of_two(params->max_size);
1031 		if (ht->p.max_size < ht->max_elems / 2)
1032 			ht->max_elems = ht->p.max_size * 2;
1033 	}
1034 
1035 	ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
1036 
1037 	size = rounded_hashtable_size(&ht->p);
1038 
1039 	ht->key_len = ht->p.key_len;
1040 	if (!params->hashfn) {
1041 		ht->p.hashfn = jhash;
1042 
1043 		if (!(ht->key_len & (sizeof(u32) - 1))) {
1044 			ht->key_len /= sizeof(u32);
1045 			ht->p.hashfn = rhashtable_jhash2;
1046 		}
1047 	}
1048 
1049 	/*
1050 	 * This is api initialization and thus we need to guarantee the
1051 	 * initial rhashtable allocation. Upon failure, retry with the
1052 	 * smallest possible size with __GFP_NOFAIL semantics.
1053 	 */
1054 	tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
1055 	if (unlikely(tbl == NULL)) {
1056 		size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
1057 		tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL);
1058 	}
1059 
1060 	atomic_set(&ht->nelems, 0);
1061 
1062 	RCU_INIT_POINTER(ht->tbl, tbl);
1063 
1064 	INIT_WORK(&ht->run_work, rht_deferred_worker);
1065 
1066 	return 0;
1067 }
1068 EXPORT_SYMBOL_GPL(rhashtable_init);
1069 
1070 /**
1071  * rhltable_init - initialize a new hash list table
1072  * @hlt:	hash list table to be initialized
1073  * @params:	configuration parameters
1074  *
1075  * Initializes a new hash list table.
1076  *
1077  * See documentation for rhashtable_init.
1078  */
1079 int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
1080 {
1081 	int err;
1082 
1083 	err = rhashtable_init(&hlt->ht, params);
1084 	hlt->ht.rhlist = true;
1085 	return err;
1086 }
1087 EXPORT_SYMBOL_GPL(rhltable_init);
1088 
1089 static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
1090 				void (*free_fn)(void *ptr, void *arg),
1091 				void *arg)
1092 {
1093 	struct rhlist_head *list;
1094 
1095 	if (!ht->rhlist) {
1096 		free_fn(rht_obj(ht, obj), arg);
1097 		return;
1098 	}
1099 
1100 	list = container_of(obj, struct rhlist_head, rhead);
1101 	do {
1102 		obj = &list->rhead;
1103 		list = rht_dereference(list->next, ht);
1104 		free_fn(rht_obj(ht, obj), arg);
1105 	} while (list);
1106 }
1107 
1108 /**
1109  * rhashtable_free_and_destroy - free elements and destroy hash table
1110  * @ht:		the hash table to destroy
1111  * @free_fn:	callback to release resources of element
1112  * @arg:	pointer passed to free_fn
1113  *
1114  * Stops an eventual async resize. If defined, invokes free_fn for each
1115  * element to releasal resources. Please note that RCU protected
1116  * readers may still be accessing the elements. Releasing of resources
1117  * must occur in a compatible manner. Then frees the bucket array.
1118  *
1119  * This function will eventually sleep to wait for an async resize
1120  * to complete. The caller is responsible that no further write operations
1121  * occurs in parallel.
1122  */
1123 void rhashtable_free_and_destroy(struct rhashtable *ht,
1124 				 void (*free_fn)(void *ptr, void *arg),
1125 				 void *arg)
1126 {
1127 	struct bucket_table *tbl, *next_tbl;
1128 	unsigned int i;
1129 
1130 	cancel_work_sync(&ht->run_work);
1131 
1132 	mutex_lock(&ht->mutex);
1133 	tbl = rht_dereference(ht->tbl, ht);
1134 restart:
1135 	if (free_fn) {
1136 		for (i = 0; i < tbl->size; i++) {
1137 			struct rhash_head *pos, *next;
1138 
1139 			cond_resched();
1140 			for (pos = rht_ptr_exclusive(rht_bucket(tbl, i)),
1141 			     next = !rht_is_a_nulls(pos) ?
1142 					rht_dereference(pos->next, ht) : NULL;
1143 			     !rht_is_a_nulls(pos);
1144 			     pos = next,
1145 			     next = !rht_is_a_nulls(pos) ?
1146 					rht_dereference(pos->next, ht) : NULL)
1147 				rhashtable_free_one(ht, pos, free_fn, arg);
1148 		}
1149 	}
1150 
1151 	next_tbl = rht_dereference(tbl->future_tbl, ht);
1152 	bucket_table_free(tbl);
1153 	if (next_tbl) {
1154 		tbl = next_tbl;
1155 		goto restart;
1156 	}
1157 	mutex_unlock(&ht->mutex);
1158 }
1159 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
1160 
1161 void rhashtable_destroy(struct rhashtable *ht)
1162 {
1163 	return rhashtable_free_and_destroy(ht, NULL, NULL);
1164 }
1165 EXPORT_SYMBOL_GPL(rhashtable_destroy);
1166 
1167 struct rhash_lock_head **__rht_bucket_nested(const struct bucket_table *tbl,
1168 					     unsigned int hash)
1169 {
1170 	const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1171 	unsigned int index = hash & ((1 << tbl->nest) - 1);
1172 	unsigned int size = tbl->size >> tbl->nest;
1173 	unsigned int subhash = hash;
1174 	union nested_table *ntbl;
1175 
1176 	ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1177 	ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash);
1178 	subhash >>= tbl->nest;
1179 
1180 	while (ntbl && size > (1 << shift)) {
1181 		index = subhash & ((1 << shift) - 1);
1182 		ntbl = rht_dereference_bucket_rcu(ntbl[index].table,
1183 						  tbl, hash);
1184 		size >>= shift;
1185 		subhash >>= shift;
1186 	}
1187 
1188 	if (!ntbl)
1189 		return NULL;
1190 
1191 	return &ntbl[subhash].bucket;
1192 
1193 }
1194 EXPORT_SYMBOL_GPL(__rht_bucket_nested);
1195 
1196 struct rhash_lock_head **rht_bucket_nested(const struct bucket_table *tbl,
1197 					   unsigned int hash)
1198 {
1199 	static struct rhash_lock_head *rhnull;
1200 
1201 	if (!rhnull)
1202 		INIT_RHT_NULLS_HEAD(rhnull);
1203 	return __rht_bucket_nested(tbl, hash) ?: &rhnull;
1204 }
1205 EXPORT_SYMBOL_GPL(rht_bucket_nested);
1206 
1207 struct rhash_lock_head **rht_bucket_nested_insert(struct rhashtable *ht,
1208 						  struct bucket_table *tbl,
1209 						  unsigned int hash)
1210 {
1211 	const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1212 	unsigned int index = hash & ((1 << tbl->nest) - 1);
1213 	unsigned int size = tbl->size >> tbl->nest;
1214 	union nested_table *ntbl;
1215 
1216 	ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1217 	hash >>= tbl->nest;
1218 	ntbl = nested_table_alloc(ht, &ntbl[index].table,
1219 				  size <= (1 << shift));
1220 
1221 	while (ntbl && size > (1 << shift)) {
1222 		index = hash & ((1 << shift) - 1);
1223 		size >>= shift;
1224 		hash >>= shift;
1225 		ntbl = nested_table_alloc(ht, &ntbl[index].table,
1226 					  size <= (1 << shift));
1227 	}
1228 
1229 	if (!ntbl)
1230 		return NULL;
1231 
1232 	return &ntbl[hash].bucket;
1233 
1234 }
1235 EXPORT_SYMBOL_GPL(rht_bucket_nested_insert);
1236