xref: /linux/include/linux/rhashtable.h (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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 #ifndef _LINUX_RHASHTABLE_H
18 #define _LINUX_RHASHTABLE_H
19 
20 #include <linux/compiler.h>
21 #include <linux/errno.h>
22 #include <linux/jhash.h>
23 #include <linux/list_nulls.h>
24 #include <linux/workqueue.h>
25 #include <linux/mutex.h>
26 #include <linux/rcupdate.h>
27 
28 /*
29  * The end of the chain is marked with a special nulls marks which has
30  * the following format:
31  *
32  * +-------+-----------------------------------------------------+-+
33  * | Base  |                      Hash                           |1|
34  * +-------+-----------------------------------------------------+-+
35  *
36  * Base (4 bits) : Reserved to distinguish between multiple tables.
37  *                 Specified via &struct rhashtable_params.nulls_base.
38  * Hash (27 bits): Full hash (unmasked) of first element added to bucket
39  * 1 (1 bit)     : Nulls marker (always set)
40  *
41  * The remaining bits of the next pointer remain unused for now.
42  */
43 #define RHT_BASE_BITS		4
44 #define RHT_HASH_BITS		27
45 #define RHT_BASE_SHIFT		RHT_HASH_BITS
46 
47 /* Base bits plus 1 bit for nulls marker */
48 #define RHT_HASH_RESERVED_SPACE	(RHT_BASE_BITS + 1)
49 
50 struct rhash_head {
51 	struct rhash_head __rcu		*next;
52 };
53 
54 /**
55  * struct bucket_table - Table of hash buckets
56  * @size: Number of hash buckets
57  * @rehash: Current bucket being rehashed
58  * @hash_rnd: Random seed to fold into hash
59  * @locks_mask: Mask to apply before accessing locks[]
60  * @locks: Array of spinlocks protecting individual buckets
61  * @walkers: List of active walkers
62  * @rcu: RCU structure for freeing the table
63  * @future_tbl: Table under construction during rehashing
64  * @buckets: size * hash buckets
65  */
66 struct bucket_table {
67 	unsigned int		size;
68 	unsigned int		rehash;
69 	u32			hash_rnd;
70 	unsigned int		locks_mask;
71 	spinlock_t		*locks;
72 	struct list_head	walkers;
73 	struct rcu_head		rcu;
74 
75 	struct bucket_table __rcu *future_tbl;
76 
77 	struct rhash_head __rcu	*buckets[] ____cacheline_aligned_in_smp;
78 };
79 
80 /**
81  * struct rhashtable_compare_arg - Key for the function rhashtable_compare
82  * @ht: Hash table
83  * @key: Key to compare against
84  */
85 struct rhashtable_compare_arg {
86 	struct rhashtable *ht;
87 	const void *key;
88 };
89 
90 typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
91 typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed);
92 typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
93 			       const void *obj);
94 
95 struct rhashtable;
96 
97 /**
98  * struct rhashtable_params - Hash table construction parameters
99  * @nelem_hint: Hint on number of elements, should be 75% of desired size
100  * @key_len: Length of key
101  * @key_offset: Offset of key in struct to be hashed
102  * @head_offset: Offset of rhash_head in struct to be hashed
103  * @max_size: Maximum size while expanding
104  * @min_size: Minimum size while shrinking
105  * @nulls_base: Base value to generate nulls marker
106  * @insecure_elasticity: Set to true to disable chain length checks
107  * @automatic_shrinking: Enable automatic shrinking of tables
108  * @locks_mul: Number of bucket locks to allocate per cpu (default: 128)
109  * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
110  * @obj_hashfn: Function to hash object
111  * @obj_cmpfn: Function to compare key with object
112  */
113 struct rhashtable_params {
114 	size_t			nelem_hint;
115 	size_t			key_len;
116 	size_t			key_offset;
117 	size_t			head_offset;
118 	unsigned int		max_size;
119 	unsigned int		min_size;
120 	u32			nulls_base;
121 	bool			insecure_elasticity;
122 	bool			automatic_shrinking;
123 	size_t			locks_mul;
124 	rht_hashfn_t		hashfn;
125 	rht_obj_hashfn_t	obj_hashfn;
126 	rht_obj_cmpfn_t		obj_cmpfn;
127 };
128 
129 /**
130  * struct rhashtable - Hash table handle
131  * @tbl: Bucket table
132  * @nelems: Number of elements in table
133  * @key_len: Key length for hashfn
134  * @elasticity: Maximum chain length before rehash
135  * @p: Configuration parameters
136  * @run_work: Deferred worker to expand/shrink asynchronously
137  * @mutex: Mutex to protect current/future table swapping
138  * @lock: Spin lock to protect walker list
139  */
140 struct rhashtable {
141 	struct bucket_table __rcu	*tbl;
142 	atomic_t			nelems;
143 	unsigned int			key_len;
144 	unsigned int			elasticity;
145 	struct rhashtable_params	p;
146 	struct work_struct		run_work;
147 	struct mutex                    mutex;
148 	spinlock_t			lock;
149 };
150 
151 /**
152  * struct rhashtable_walker - Hash table walker
153  * @list: List entry on list of walkers
154  * @tbl: The table that we were walking over
155  */
156 struct rhashtable_walker {
157 	struct list_head list;
158 	struct bucket_table *tbl;
159 };
160 
161 /**
162  * struct rhashtable_iter - Hash table iterator, fits into netlink cb
163  * @ht: Table to iterate through
164  * @p: Current pointer
165  * @walker: Associated rhashtable walker
166  * @slot: Current slot
167  * @skip: Number of entries to skip in slot
168  */
169 struct rhashtable_iter {
170 	struct rhashtable *ht;
171 	struct rhash_head *p;
172 	struct rhashtable_walker *walker;
173 	unsigned int slot;
174 	unsigned int skip;
175 };
176 
177 static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
178 {
179 	return NULLS_MARKER(ht->p.nulls_base + hash);
180 }
181 
182 #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
183 	((ptr) = (typeof(ptr)) rht_marker(ht, hash))
184 
185 static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
186 {
187 	return ((unsigned long) ptr & 1);
188 }
189 
190 static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
191 {
192 	return ((unsigned long) ptr) >> 1;
193 }
194 
195 static inline void *rht_obj(const struct rhashtable *ht,
196 			    const struct rhash_head *he)
197 {
198 	return (char *)he - ht->p.head_offset;
199 }
200 
201 static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
202 					    unsigned int hash)
203 {
204 	return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
205 }
206 
207 static inline unsigned int rht_key_hashfn(
208 	struct rhashtable *ht, const struct bucket_table *tbl,
209 	const void *key, const struct rhashtable_params params)
210 {
211 	unsigned int hash;
212 
213 	/* params must be equal to ht->p if it isn't constant. */
214 	if (!__builtin_constant_p(params.key_len))
215 		hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd);
216 	else if (params.key_len) {
217 		unsigned int key_len = params.key_len;
218 
219 		if (params.hashfn)
220 			hash = params.hashfn(key, key_len, tbl->hash_rnd);
221 		else if (key_len & (sizeof(u32) - 1))
222 			hash = jhash(key, key_len, tbl->hash_rnd);
223 		else
224 			hash = jhash2(key, key_len / sizeof(u32),
225 				      tbl->hash_rnd);
226 	} else {
227 		unsigned int key_len = ht->p.key_len;
228 
229 		if (params.hashfn)
230 			hash = params.hashfn(key, key_len, tbl->hash_rnd);
231 		else
232 			hash = jhash(key, key_len, tbl->hash_rnd);
233 	}
234 
235 	return rht_bucket_index(tbl, hash);
236 }
237 
238 static inline unsigned int rht_head_hashfn(
239 	struct rhashtable *ht, const struct bucket_table *tbl,
240 	const struct rhash_head *he, const struct rhashtable_params params)
241 {
242 	const char *ptr = rht_obj(ht, he);
243 
244 	return likely(params.obj_hashfn) ?
245 	       rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
246 							    ht->p.key_len,
247 						       tbl->hash_rnd)) :
248 	       rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
249 }
250 
251 /**
252  * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
253  * @ht:		hash table
254  * @tbl:	current table
255  */
256 static inline bool rht_grow_above_75(const struct rhashtable *ht,
257 				     const struct bucket_table *tbl)
258 {
259 	/* Expand table when exceeding 75% load */
260 	return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
261 	       (!ht->p.max_size || tbl->size < ht->p.max_size);
262 }
263 
264 /**
265  * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
266  * @ht:		hash table
267  * @tbl:	current table
268  */
269 static inline bool rht_shrink_below_30(const struct rhashtable *ht,
270 				       const struct bucket_table *tbl)
271 {
272 	/* Shrink table beneath 30% load */
273 	return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
274 	       tbl->size > ht->p.min_size;
275 }
276 
277 /**
278  * rht_grow_above_100 - returns true if nelems > table-size
279  * @ht:		hash table
280  * @tbl:	current table
281  */
282 static inline bool rht_grow_above_100(const struct rhashtable *ht,
283 				      const struct bucket_table *tbl)
284 {
285 	return atomic_read(&ht->nelems) > tbl->size &&
286 		(!ht->p.max_size || tbl->size < ht->p.max_size);
287 }
288 
289 /* The bucket lock is selected based on the hash and protects mutations
290  * on a group of hash buckets.
291  *
292  * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
293  * a single lock always covers both buckets which may both contains
294  * entries which link to the same bucket of the old table during resizing.
295  * This allows to simplify the locking as locking the bucket in both
296  * tables during resize always guarantee protection.
297  *
298  * IMPORTANT: When holding the bucket lock of both the old and new table
299  * during expansions and shrinking, the old bucket lock must always be
300  * acquired first.
301  */
302 static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
303 					  unsigned int hash)
304 {
305 	return &tbl->locks[hash & tbl->locks_mask];
306 }
307 
308 #ifdef CONFIG_PROVE_LOCKING
309 int lockdep_rht_mutex_is_held(struct rhashtable *ht);
310 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
311 #else
312 static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
313 {
314 	return 1;
315 }
316 
317 static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
318 					     u32 hash)
319 {
320 	return 1;
321 }
322 #endif /* CONFIG_PROVE_LOCKING */
323 
324 int rhashtable_init(struct rhashtable *ht,
325 		    const struct rhashtable_params *params);
326 
327 int rhashtable_insert_slow(struct rhashtable *ht, const void *key,
328 			   struct rhash_head *obj,
329 			   struct bucket_table *old_tbl);
330 int rhashtable_insert_rehash(struct rhashtable *ht);
331 
332 int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter);
333 void rhashtable_walk_exit(struct rhashtable_iter *iter);
334 int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU);
335 void *rhashtable_walk_next(struct rhashtable_iter *iter);
336 void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
337 
338 void rhashtable_free_and_destroy(struct rhashtable *ht,
339 				 void (*free_fn)(void *ptr, void *arg),
340 				 void *arg);
341 void rhashtable_destroy(struct rhashtable *ht);
342 
343 #define rht_dereference(p, ht) \
344 	rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
345 
346 #define rht_dereference_rcu(p, ht) \
347 	rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
348 
349 #define rht_dereference_bucket(p, tbl, hash) \
350 	rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
351 
352 #define rht_dereference_bucket_rcu(p, tbl, hash) \
353 	rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
354 
355 #define rht_entry(tpos, pos, member) \
356 	({ tpos = container_of(pos, typeof(*tpos), member); 1; })
357 
358 /**
359  * rht_for_each_continue - continue iterating over hash chain
360  * @pos:	the &struct rhash_head to use as a loop cursor.
361  * @head:	the previous &struct rhash_head to continue from
362  * @tbl:	the &struct bucket_table
363  * @hash:	the hash value / bucket index
364  */
365 #define rht_for_each_continue(pos, head, tbl, hash) \
366 	for (pos = rht_dereference_bucket(head, tbl, hash); \
367 	     !rht_is_a_nulls(pos); \
368 	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
369 
370 /**
371  * rht_for_each - iterate over hash chain
372  * @pos:	the &struct rhash_head to use as a loop cursor.
373  * @tbl:	the &struct bucket_table
374  * @hash:	the hash value / bucket index
375  */
376 #define rht_for_each(pos, tbl, hash) \
377 	rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash)
378 
379 /**
380  * rht_for_each_entry_continue - continue iterating over hash chain
381  * @tpos:	the type * to use as a loop cursor.
382  * @pos:	the &struct rhash_head to use as a loop cursor.
383  * @head:	the previous &struct rhash_head to continue from
384  * @tbl:	the &struct bucket_table
385  * @hash:	the hash value / bucket index
386  * @member:	name of the &struct rhash_head within the hashable struct.
387  */
388 #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member)	\
389 	for (pos = rht_dereference_bucket(head, tbl, hash);		\
390 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	\
391 	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
392 
393 /**
394  * rht_for_each_entry - iterate over hash chain of given type
395  * @tpos:	the type * to use as a loop cursor.
396  * @pos:	the &struct rhash_head to use as a loop cursor.
397  * @tbl:	the &struct bucket_table
398  * @hash:	the hash value / bucket index
399  * @member:	name of the &struct rhash_head within the hashable struct.
400  */
401 #define rht_for_each_entry(tpos, pos, tbl, hash, member)		\
402 	rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash],	\
403 				    tbl, hash, member)
404 
405 /**
406  * rht_for_each_entry_safe - safely iterate over hash chain of given type
407  * @tpos:	the type * to use as a loop cursor.
408  * @pos:	the &struct rhash_head to use as a loop cursor.
409  * @next:	the &struct rhash_head to use as next in loop cursor.
410  * @tbl:	the &struct bucket_table
411  * @hash:	the hash value / bucket index
412  * @member:	name of the &struct rhash_head within the hashable struct.
413  *
414  * This hash chain list-traversal primitive allows for the looped code to
415  * remove the loop cursor from the list.
416  */
417 #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member)	    \
418 	for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \
419 	     next = !rht_is_a_nulls(pos) ?				    \
420 		       rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
421 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	    \
422 	     pos = next,						    \
423 	     next = !rht_is_a_nulls(pos) ?				    \
424 		       rht_dereference_bucket(pos->next, tbl, hash) : NULL)
425 
426 /**
427  * rht_for_each_rcu_continue - continue iterating over rcu hash chain
428  * @pos:	the &struct rhash_head to use as a loop cursor.
429  * @head:	the previous &struct rhash_head to continue from
430  * @tbl:	the &struct bucket_table
431  * @hash:	the hash value / bucket index
432  *
433  * This hash chain list-traversal primitive may safely run concurrently with
434  * the _rcu mutation primitives such as rhashtable_insert() as long as the
435  * traversal is guarded by rcu_read_lock().
436  */
437 #define rht_for_each_rcu_continue(pos, head, tbl, hash)			\
438 	for (({barrier(); }),						\
439 	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		\
440 	     !rht_is_a_nulls(pos);					\
441 	     pos = rcu_dereference_raw(pos->next))
442 
443 /**
444  * rht_for_each_rcu - iterate over rcu hash chain
445  * @pos:	the &struct rhash_head to use as a loop cursor.
446  * @tbl:	the &struct bucket_table
447  * @hash:	the hash value / bucket index
448  *
449  * This hash chain list-traversal primitive may safely run concurrently with
450  * the _rcu mutation primitives such as rhashtable_insert() as long as the
451  * traversal is guarded by rcu_read_lock().
452  */
453 #define rht_for_each_rcu(pos, tbl, hash)				\
454 	rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash)
455 
456 /**
457  * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
458  * @tpos:	the type * to use as a loop cursor.
459  * @pos:	the &struct rhash_head to use as a loop cursor.
460  * @head:	the previous &struct rhash_head to continue from
461  * @tbl:	the &struct bucket_table
462  * @hash:	the hash value / bucket index
463  * @member:	name of the &struct rhash_head within the hashable struct.
464  *
465  * This hash chain list-traversal primitive may safely run concurrently with
466  * the _rcu mutation primitives such as rhashtable_insert() as long as the
467  * traversal is guarded by rcu_read_lock().
468  */
469 #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
470 	for (({barrier(); }),						    \
471 	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		    \
472 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	    \
473 	     pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
474 
475 /**
476  * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
477  * @tpos:	the type * to use as a loop cursor.
478  * @pos:	the &struct rhash_head to use as a loop cursor.
479  * @tbl:	the &struct bucket_table
480  * @hash:	the hash value / bucket index
481  * @member:	name of the &struct rhash_head within the hashable struct.
482  *
483  * This hash chain list-traversal primitive may safely run concurrently with
484  * the _rcu mutation primitives such as rhashtable_insert() as long as the
485  * traversal is guarded by rcu_read_lock().
486  */
487 #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member)		\
488 	rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\
489 					tbl, hash, member)
490 
491 static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
492 				     const void *obj)
493 {
494 	struct rhashtable *ht = arg->ht;
495 	const char *ptr = obj;
496 
497 	return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
498 }
499 
500 /**
501  * rhashtable_lookup_fast - search hash table, inlined version
502  * @ht:		hash table
503  * @key:	the pointer to the key
504  * @params:	hash table parameters
505  *
506  * Computes the hash value for the key and traverses the bucket chain looking
507  * for a entry with an identical key. The first matching entry is returned.
508  *
509  * Returns the first entry on which the compare function returned true.
510  */
511 static inline void *rhashtable_lookup_fast(
512 	struct rhashtable *ht, const void *key,
513 	const struct rhashtable_params params)
514 {
515 	struct rhashtable_compare_arg arg = {
516 		.ht = ht,
517 		.key = key,
518 	};
519 	const struct bucket_table *tbl;
520 	struct rhash_head *he;
521 	unsigned int hash;
522 
523 	rcu_read_lock();
524 
525 	tbl = rht_dereference_rcu(ht->tbl, ht);
526 restart:
527 	hash = rht_key_hashfn(ht, tbl, key, params);
528 	rht_for_each_rcu(he, tbl, hash) {
529 		if (params.obj_cmpfn ?
530 		    params.obj_cmpfn(&arg, rht_obj(ht, he)) :
531 		    rhashtable_compare(&arg, rht_obj(ht, he)))
532 			continue;
533 		rcu_read_unlock();
534 		return rht_obj(ht, he);
535 	}
536 
537 	/* Ensure we see any new tables. */
538 	smp_rmb();
539 
540 	tbl = rht_dereference_rcu(tbl->future_tbl, ht);
541 	if (unlikely(tbl))
542 		goto restart;
543 	rcu_read_unlock();
544 
545 	return NULL;
546 }
547 
548 /* Internal function, please use rhashtable_insert_fast() instead */
549 static inline int __rhashtable_insert_fast(
550 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
551 	const struct rhashtable_params params)
552 {
553 	struct rhashtable_compare_arg arg = {
554 		.ht = ht,
555 		.key = key,
556 	};
557 	struct bucket_table *tbl, *new_tbl;
558 	struct rhash_head *head;
559 	spinlock_t *lock;
560 	unsigned int elasticity;
561 	unsigned int hash;
562 	int err;
563 
564 restart:
565 	rcu_read_lock();
566 
567 	tbl = rht_dereference_rcu(ht->tbl, ht);
568 
569 	/* All insertions must grab the oldest table containing
570 	 * the hashed bucket that is yet to be rehashed.
571 	 */
572 	for (;;) {
573 		hash = rht_head_hashfn(ht, tbl, obj, params);
574 		lock = rht_bucket_lock(tbl, hash);
575 		spin_lock_bh(lock);
576 
577 		if (tbl->rehash <= hash)
578 			break;
579 
580 		spin_unlock_bh(lock);
581 		tbl = rht_dereference_rcu(tbl->future_tbl, ht);
582 	}
583 
584 	new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
585 	if (unlikely(new_tbl)) {
586 		err = rhashtable_insert_slow(ht, key, obj, new_tbl);
587 		if (err == -EAGAIN)
588 			goto slow_path;
589 		goto out;
590 	}
591 
592 	if (unlikely(rht_grow_above_100(ht, tbl))) {
593 slow_path:
594 		spin_unlock_bh(lock);
595 		err = rhashtable_insert_rehash(ht);
596 		rcu_read_unlock();
597 		if (err)
598 			return err;
599 
600 		goto restart;
601 	}
602 
603 	err = -EEXIST;
604 	elasticity = ht->elasticity;
605 	rht_for_each(head, tbl, hash) {
606 		if (key &&
607 		    unlikely(!(params.obj_cmpfn ?
608 			       params.obj_cmpfn(&arg, rht_obj(ht, head)) :
609 			       rhashtable_compare(&arg, rht_obj(ht, head)))))
610 			goto out;
611 		if (!--elasticity)
612 			goto slow_path;
613 	}
614 
615 	err = 0;
616 
617 	head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
618 
619 	RCU_INIT_POINTER(obj->next, head);
620 
621 	rcu_assign_pointer(tbl->buckets[hash], obj);
622 
623 	atomic_inc(&ht->nelems);
624 	if (rht_grow_above_75(ht, tbl))
625 		schedule_work(&ht->run_work);
626 
627 out:
628 	spin_unlock_bh(lock);
629 	rcu_read_unlock();
630 
631 	return err;
632 }
633 
634 /**
635  * rhashtable_insert_fast - insert object into hash table
636  * @ht:		hash table
637  * @obj:	pointer to hash head inside object
638  * @params:	hash table parameters
639  *
640  * Will take a per bucket spinlock to protect against mutual mutations
641  * on the same bucket. Multiple insertions may occur in parallel unless
642  * they map to the same bucket lock.
643  *
644  * It is safe to call this function from atomic context.
645  *
646  * Will trigger an automatic deferred table resizing if the size grows
647  * beyond the watermark indicated by grow_decision() which can be passed
648  * to rhashtable_init().
649  */
650 static inline int rhashtable_insert_fast(
651 	struct rhashtable *ht, struct rhash_head *obj,
652 	const struct rhashtable_params params)
653 {
654 	return __rhashtable_insert_fast(ht, NULL, obj, params);
655 }
656 
657 /**
658  * rhashtable_lookup_insert_fast - lookup and insert object into hash table
659  * @ht:		hash table
660  * @obj:	pointer to hash head inside object
661  * @params:	hash table parameters
662  *
663  * Locks down the bucket chain in both the old and new table if a resize
664  * is in progress to ensure that writers can't remove from the old table
665  * and can't insert to the new table during the atomic operation of search
666  * and insertion. Searches for duplicates in both the old and new table if
667  * a resize is in progress.
668  *
669  * This lookup function may only be used for fixed key hash table (key_len
670  * parameter set). It will BUG() if used inappropriately.
671  *
672  * It is safe to call this function from atomic context.
673  *
674  * Will trigger an automatic deferred table resizing if the size grows
675  * beyond the watermark indicated by grow_decision() which can be passed
676  * to rhashtable_init().
677  */
678 static inline int rhashtable_lookup_insert_fast(
679 	struct rhashtable *ht, struct rhash_head *obj,
680 	const struct rhashtable_params params)
681 {
682 	const char *key = rht_obj(ht, obj);
683 
684 	BUG_ON(ht->p.obj_hashfn);
685 
686 	return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj,
687 					params);
688 }
689 
690 /**
691  * rhashtable_lookup_insert_key - search and insert object to hash table
692  *				  with explicit key
693  * @ht:		hash table
694  * @key:	key
695  * @obj:	pointer to hash head inside object
696  * @params:	hash table parameters
697  *
698  * Locks down the bucket chain in both the old and new table if a resize
699  * is in progress to ensure that writers can't remove from the old table
700  * and can't insert to the new table during the atomic operation of search
701  * and insertion. Searches for duplicates in both the old and new table if
702  * a resize is in progress.
703  *
704  * Lookups may occur in parallel with hashtable mutations and resizing.
705  *
706  * Will trigger an automatic deferred table resizing if the size grows
707  * beyond the watermark indicated by grow_decision() which can be passed
708  * to rhashtable_init().
709  *
710  * Returns zero on success.
711  */
712 static inline int rhashtable_lookup_insert_key(
713 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
714 	const struct rhashtable_params params)
715 {
716 	BUG_ON(!ht->p.obj_hashfn || !key);
717 
718 	return __rhashtable_insert_fast(ht, key, obj, params);
719 }
720 
721 /* Internal function, please use rhashtable_remove_fast() instead */
722 static inline int __rhashtable_remove_fast(
723 	struct rhashtable *ht, struct bucket_table *tbl,
724 	struct rhash_head *obj, const struct rhashtable_params params)
725 {
726 	struct rhash_head __rcu **pprev;
727 	struct rhash_head *he;
728 	spinlock_t * lock;
729 	unsigned int hash;
730 	int err = -ENOENT;
731 
732 	hash = rht_head_hashfn(ht, tbl, obj, params);
733 	lock = rht_bucket_lock(tbl, hash);
734 
735 	spin_lock_bh(lock);
736 
737 	pprev = &tbl->buckets[hash];
738 	rht_for_each(he, tbl, hash) {
739 		if (he != obj) {
740 			pprev = &he->next;
741 			continue;
742 		}
743 
744 		rcu_assign_pointer(*pprev, obj->next);
745 		err = 0;
746 		break;
747 	}
748 
749 	spin_unlock_bh(lock);
750 
751 	return err;
752 }
753 
754 /**
755  * rhashtable_remove_fast - remove object from hash table
756  * @ht:		hash table
757  * @obj:	pointer to hash head inside object
758  * @params:	hash table parameters
759  *
760  * Since the hash chain is single linked, the removal operation needs to
761  * walk the bucket chain upon removal. The removal operation is thus
762  * considerable slow if the hash table is not correctly sized.
763  *
764  * Will automatically shrink the table via rhashtable_expand() if the
765  * shrink_decision function specified at rhashtable_init() returns true.
766  *
767  * Returns zero on success, -ENOENT if the entry could not be found.
768  */
769 static inline int rhashtable_remove_fast(
770 	struct rhashtable *ht, struct rhash_head *obj,
771 	const struct rhashtable_params params)
772 {
773 	struct bucket_table *tbl;
774 	int err;
775 
776 	rcu_read_lock();
777 
778 	tbl = rht_dereference_rcu(ht->tbl, ht);
779 
780 	/* Because we have already taken (and released) the bucket
781 	 * lock in old_tbl, if we find that future_tbl is not yet
782 	 * visible then that guarantees the entry to still be in
783 	 * the old tbl if it exists.
784 	 */
785 	while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) &&
786 	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
787 		;
788 
789 	if (err)
790 		goto out;
791 
792 	atomic_dec(&ht->nelems);
793 	if (unlikely(ht->p.automatic_shrinking &&
794 		     rht_shrink_below_30(ht, tbl)))
795 		schedule_work(&ht->run_work);
796 
797 out:
798 	rcu_read_unlock();
799 
800 	return err;
801 }
802 
803 #endif /* _LINUX_RHASHTABLE_H */
804