xref: /linux/include/linux/rhashtable.h (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Resizable, Scalable, Concurrent Hash Table
4  *
5  * Copyright (c) 2015-2016 Herbert Xu <herbert@gondor.apana.org.au>
6  * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
7  * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
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  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17 
18 #ifndef _LINUX_RHASHTABLE_H
19 #define _LINUX_RHASHTABLE_H
20 
21 #include <linux/err.h>
22 #include <linux/errno.h>
23 #include <linux/jhash.h>
24 #include <linux/list_nulls.h>
25 #include <linux/workqueue.h>
26 #include <linux/rculist.h>
27 
28 #include <linux/rhashtable-types.h>
29 /*
30  * The end of the chain is marked with a special nulls marks which has
31  * the least significant bit set.
32  */
33 
34 /* Maximum chain length before rehash
35  *
36  * The maximum (not average) chain length grows with the size of the hash
37  * table, at a rate of (log N)/(log log N).
38  *
39  * The value of 16 is selected so that even if the hash table grew to
40  * 2^32 you would not expect the maximum chain length to exceed it
41  * unless we are under attack (or extremely unlucky).
42  *
43  * As this limit is only to detect attacks, we don't need to set it to a
44  * lower value as you'd need the chain length to vastly exceed 16 to have
45  * any real effect on the system.
46  */
47 #define RHT_ELASTICITY	16u
48 
49 /**
50  * struct bucket_table - Table of hash buckets
51  * @size: Number of hash buckets
52  * @nest: Number of bits of first-level nested table.
53  * @rehash: Current bucket being rehashed
54  * @hash_rnd: Random seed to fold into hash
55  * @locks_mask: Mask to apply before accessing locks[]
56  * @locks: Array of spinlocks protecting individual buckets
57  * @walkers: List of active walkers
58  * @rcu: RCU structure for freeing the table
59  * @future_tbl: Table under construction during rehashing
60  * @ntbl: Nested table used when out of memory.
61  * @buckets: size * hash buckets
62  */
63 struct bucket_table {
64 	unsigned int		size;
65 	unsigned int		nest;
66 	unsigned int		rehash;
67 	u32			hash_rnd;
68 	unsigned int		locks_mask;
69 	spinlock_t		*locks;
70 	struct list_head	walkers;
71 	struct rcu_head		rcu;
72 
73 	struct bucket_table __rcu *future_tbl;
74 
75 	struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp;
76 };
77 
78 #define INIT_RHT_NULLS_HEAD(ptr)	\
79 	((ptr) = (typeof(ptr)) NULLS_MARKER(0))
80 
81 static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
82 {
83 	return ((unsigned long) ptr & 1);
84 }
85 
86 static inline void *rht_obj(const struct rhashtable *ht,
87 			    const struct rhash_head *he)
88 {
89 	return (char *)he - ht->p.head_offset;
90 }
91 
92 static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
93 					    unsigned int hash)
94 {
95 	return hash & (tbl->size - 1);
96 }
97 
98 static inline unsigned int rht_key_get_hash(struct rhashtable *ht,
99 	const void *key, const struct rhashtable_params params,
100 	unsigned int hash_rnd)
101 {
102 	unsigned int hash;
103 
104 	/* params must be equal to ht->p if it isn't constant. */
105 	if (!__builtin_constant_p(params.key_len))
106 		hash = ht->p.hashfn(key, ht->key_len, hash_rnd);
107 	else if (params.key_len) {
108 		unsigned int key_len = params.key_len;
109 
110 		if (params.hashfn)
111 			hash = params.hashfn(key, key_len, hash_rnd);
112 		else if (key_len & (sizeof(u32) - 1))
113 			hash = jhash(key, key_len, hash_rnd);
114 		else
115 			hash = jhash2(key, key_len / sizeof(u32), hash_rnd);
116 	} else {
117 		unsigned int key_len = ht->p.key_len;
118 
119 		if (params.hashfn)
120 			hash = params.hashfn(key, key_len, hash_rnd);
121 		else
122 			hash = jhash(key, key_len, hash_rnd);
123 	}
124 
125 	return hash;
126 }
127 
128 static inline unsigned int rht_key_hashfn(
129 	struct rhashtable *ht, const struct bucket_table *tbl,
130 	const void *key, const struct rhashtable_params params)
131 {
132 	unsigned int hash = rht_key_get_hash(ht, key, params, tbl->hash_rnd);
133 
134 	return rht_bucket_index(tbl, hash);
135 }
136 
137 static inline unsigned int rht_head_hashfn(
138 	struct rhashtable *ht, const struct bucket_table *tbl,
139 	const struct rhash_head *he, const struct rhashtable_params params)
140 {
141 	const char *ptr = rht_obj(ht, he);
142 
143 	return likely(params.obj_hashfn) ?
144 	       rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
145 							    ht->p.key_len,
146 						       tbl->hash_rnd)) :
147 	       rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
148 }
149 
150 /**
151  * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
152  * @ht:		hash table
153  * @tbl:	current table
154  */
155 static inline bool rht_grow_above_75(const struct rhashtable *ht,
156 				     const struct bucket_table *tbl)
157 {
158 	/* Expand table when exceeding 75% load */
159 	return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
160 	       (!ht->p.max_size || tbl->size < ht->p.max_size);
161 }
162 
163 /**
164  * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
165  * @ht:		hash table
166  * @tbl:	current table
167  */
168 static inline bool rht_shrink_below_30(const struct rhashtable *ht,
169 				       const struct bucket_table *tbl)
170 {
171 	/* Shrink table beneath 30% load */
172 	return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
173 	       tbl->size > ht->p.min_size;
174 }
175 
176 /**
177  * rht_grow_above_100 - returns true if nelems > table-size
178  * @ht:		hash table
179  * @tbl:	current table
180  */
181 static inline bool rht_grow_above_100(const struct rhashtable *ht,
182 				      const struct bucket_table *tbl)
183 {
184 	return atomic_read(&ht->nelems) > tbl->size &&
185 		(!ht->p.max_size || tbl->size < ht->p.max_size);
186 }
187 
188 /**
189  * rht_grow_above_max - returns true if table is above maximum
190  * @ht:		hash table
191  * @tbl:	current table
192  */
193 static inline bool rht_grow_above_max(const struct rhashtable *ht,
194 				      const struct bucket_table *tbl)
195 {
196 	return atomic_read(&ht->nelems) >= ht->max_elems;
197 }
198 
199 /* The bucket lock is selected based on the hash and protects mutations
200  * on a group of hash buckets.
201  *
202  * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
203  * a single lock always covers both buckets which may both contains
204  * entries which link to the same bucket of the old table during resizing.
205  * This allows to simplify the locking as locking the bucket in both
206  * tables during resize always guarantee protection.
207  *
208  * IMPORTANT: When holding the bucket lock of both the old and new table
209  * during expansions and shrinking, the old bucket lock must always be
210  * acquired first.
211  */
212 static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
213 					  unsigned int hash)
214 {
215 	return &tbl->locks[hash & tbl->locks_mask];
216 }
217 
218 #ifdef CONFIG_PROVE_LOCKING
219 int lockdep_rht_mutex_is_held(struct rhashtable *ht);
220 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
221 #else
222 static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
223 {
224 	return 1;
225 }
226 
227 static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
228 					     u32 hash)
229 {
230 	return 1;
231 }
232 #endif /* CONFIG_PROVE_LOCKING */
233 
234 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
235 			     struct rhash_head *obj);
236 
237 void rhashtable_walk_enter(struct rhashtable *ht,
238 			   struct rhashtable_iter *iter);
239 void rhashtable_walk_exit(struct rhashtable_iter *iter);
240 int rhashtable_walk_start_check(struct rhashtable_iter *iter) __acquires(RCU);
241 
242 static inline void rhashtable_walk_start(struct rhashtable_iter *iter)
243 {
244 	(void)rhashtable_walk_start_check(iter);
245 }
246 
247 void *rhashtable_walk_next(struct rhashtable_iter *iter);
248 void *rhashtable_walk_peek(struct rhashtable_iter *iter);
249 void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
250 
251 void rhashtable_free_and_destroy(struct rhashtable *ht,
252 				 void (*free_fn)(void *ptr, void *arg),
253 				 void *arg);
254 void rhashtable_destroy(struct rhashtable *ht);
255 
256 struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
257 					    unsigned int hash);
258 struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
259 						   struct bucket_table *tbl,
260 						   unsigned int hash);
261 
262 #define rht_dereference(p, ht) \
263 	rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
264 
265 #define rht_dereference_rcu(p, ht) \
266 	rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
267 
268 #define rht_dereference_bucket(p, tbl, hash) \
269 	rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
270 
271 #define rht_dereference_bucket_rcu(p, tbl, hash) \
272 	rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
273 
274 #define rht_entry(tpos, pos, member) \
275 	({ tpos = container_of(pos, typeof(*tpos), member); 1; })
276 
277 static inline struct rhash_head __rcu *const *rht_bucket(
278 	const struct bucket_table *tbl, unsigned int hash)
279 {
280 	return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
281 				     &tbl->buckets[hash];
282 }
283 
284 static inline struct rhash_head __rcu **rht_bucket_var(
285 	struct bucket_table *tbl, unsigned int hash)
286 {
287 	return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
288 				     &tbl->buckets[hash];
289 }
290 
291 static inline struct rhash_head __rcu **rht_bucket_insert(
292 	struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
293 {
294 	return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
295 				     &tbl->buckets[hash];
296 }
297 
298 /**
299  * rht_for_each_continue - continue iterating over hash chain
300  * @pos:	the &struct rhash_head to use as a loop cursor.
301  * @head:	the previous &struct rhash_head to continue from
302  * @tbl:	the &struct bucket_table
303  * @hash:	the hash value / bucket index
304  */
305 #define rht_for_each_continue(pos, head, tbl, hash) \
306 	for (pos = rht_dereference_bucket(head, tbl, hash); \
307 	     !rht_is_a_nulls(pos); \
308 	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
309 
310 /**
311  * rht_for_each - iterate over hash chain
312  * @pos:	the &struct rhash_head to use as a loop cursor.
313  * @tbl:	the &struct bucket_table
314  * @hash:	the hash value / bucket index
315  */
316 #define rht_for_each(pos, tbl, hash) \
317 	rht_for_each_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
318 
319 /**
320  * rht_for_each_entry_continue - continue iterating over hash chain
321  * @tpos:	the type * to use as a loop cursor.
322  * @pos:	the &struct rhash_head to use as a loop cursor.
323  * @head:	the previous &struct rhash_head to continue from
324  * @tbl:	the &struct bucket_table
325  * @hash:	the hash value / bucket index
326  * @member:	name of the &struct rhash_head within the hashable struct.
327  */
328 #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member)	\
329 	for (pos = rht_dereference_bucket(head, tbl, hash);		\
330 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	\
331 	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
332 
333 /**
334  * rht_for_each_entry - iterate over hash chain of given type
335  * @tpos:	the type * to use as a loop cursor.
336  * @pos:	the &struct rhash_head to use as a loop cursor.
337  * @tbl:	the &struct bucket_table
338  * @hash:	the hash value / bucket index
339  * @member:	name of the &struct rhash_head within the hashable struct.
340  */
341 #define rht_for_each_entry(tpos, pos, tbl, hash, member)		\
342 	rht_for_each_entry_continue(tpos, pos, *rht_bucket(tbl, hash),	\
343 				    tbl, hash, member)
344 
345 /**
346  * rht_for_each_entry_safe - safely iterate over hash chain of given type
347  * @tpos:	the type * to use as a loop cursor.
348  * @pos:	the &struct rhash_head to use as a loop cursor.
349  * @next:	the &struct rhash_head to use as next in loop cursor.
350  * @tbl:	the &struct bucket_table
351  * @hash:	the hash value / bucket index
352  * @member:	name of the &struct rhash_head within the hashable struct.
353  *
354  * This hash chain list-traversal primitive allows for the looped code to
355  * remove the loop cursor from the list.
356  */
357 #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member)	      \
358 	for (pos = rht_dereference_bucket(*rht_bucket(tbl, hash), tbl, hash), \
359 	     next = !rht_is_a_nulls(pos) ?				      \
360 		       rht_dereference_bucket(pos->next, tbl, hash) : NULL;   \
361 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	      \
362 	     pos = next,						      \
363 	     next = !rht_is_a_nulls(pos) ?				      \
364 		       rht_dereference_bucket(pos->next, tbl, hash) : NULL)
365 
366 /**
367  * rht_for_each_rcu_continue - continue iterating over rcu hash chain
368  * @pos:	the &struct rhash_head to use as a loop cursor.
369  * @head:	the previous &struct rhash_head to continue from
370  * @tbl:	the &struct bucket_table
371  * @hash:	the hash value / bucket index
372  *
373  * This hash chain list-traversal primitive may safely run concurrently with
374  * the _rcu mutation primitives such as rhashtable_insert() as long as the
375  * traversal is guarded by rcu_read_lock().
376  */
377 #define rht_for_each_rcu_continue(pos, head, tbl, hash)			\
378 	for (({barrier(); }),						\
379 	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		\
380 	     !rht_is_a_nulls(pos);					\
381 	     pos = rcu_dereference_raw(pos->next))
382 
383 /**
384  * rht_for_each_rcu - iterate over rcu hash chain
385  * @pos:	the &struct rhash_head to use as a loop cursor.
386  * @tbl:	the &struct bucket_table
387  * @hash:	the hash value / bucket index
388  *
389  * This hash chain list-traversal primitive may safely run concurrently with
390  * the _rcu mutation primitives such as rhashtable_insert() as long as the
391  * traversal is guarded by rcu_read_lock().
392  */
393 #define rht_for_each_rcu(pos, tbl, hash)				\
394 	rht_for_each_rcu_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
395 
396 /**
397  * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
398  * @tpos:	the type * to use as a loop cursor.
399  * @pos:	the &struct rhash_head to use as a loop cursor.
400  * @head:	the previous &struct rhash_head to continue from
401  * @tbl:	the &struct bucket_table
402  * @hash:	the hash value / bucket index
403  * @member:	name of the &struct rhash_head within the hashable struct.
404  *
405  * This hash chain list-traversal primitive may safely run concurrently with
406  * the _rcu mutation primitives such as rhashtable_insert() as long as the
407  * traversal is guarded by rcu_read_lock().
408  */
409 #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
410 	for (({barrier(); }),						    \
411 	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		    \
412 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	    \
413 	     pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
414 
415 /**
416  * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
417  * @tpos:	the type * to use as a loop cursor.
418  * @pos:	the &struct rhash_head to use as a loop cursor.
419  * @tbl:	the &struct bucket_table
420  * @hash:	the hash value / bucket index
421  * @member:	name of the &struct rhash_head within the hashable struct.
422  *
423  * This hash chain list-traversal primitive may safely run concurrently with
424  * the _rcu mutation primitives such as rhashtable_insert() as long as the
425  * traversal is guarded by rcu_read_lock().
426  */
427 #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member)		   \
428 	rht_for_each_entry_rcu_continue(tpos, pos, *rht_bucket(tbl, hash), \
429 					tbl, hash, member)
430 
431 /**
432  * rhl_for_each_rcu - iterate over rcu hash table list
433  * @pos:	the &struct rlist_head to use as a loop cursor.
434  * @list:	the head of the list
435  *
436  * This hash chain list-traversal primitive should be used on the
437  * list returned by rhltable_lookup.
438  */
439 #define rhl_for_each_rcu(pos, list)					\
440 	for (pos = list; pos; pos = rcu_dereference_raw(pos->next))
441 
442 /**
443  * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
444  * @tpos:	the type * to use as a loop cursor.
445  * @pos:	the &struct rlist_head to use as a loop cursor.
446  * @list:	the head of the list
447  * @member:	name of the &struct rlist_head within the hashable struct.
448  *
449  * This hash chain list-traversal primitive should be used on the
450  * list returned by rhltable_lookup.
451  */
452 #define rhl_for_each_entry_rcu(tpos, pos, list, member)			\
453 	for (pos = list; pos && rht_entry(tpos, pos, member);		\
454 	     pos = rcu_dereference_raw(pos->next))
455 
456 static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
457 				     const void *obj)
458 {
459 	struct rhashtable *ht = arg->ht;
460 	const char *ptr = obj;
461 
462 	return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
463 }
464 
465 /* Internal function, do not use. */
466 static inline struct rhash_head *__rhashtable_lookup(
467 	struct rhashtable *ht, const void *key,
468 	const struct rhashtable_params params)
469 {
470 	struct rhashtable_compare_arg arg = {
471 		.ht = ht,
472 		.key = key,
473 	};
474 	struct bucket_table *tbl;
475 	struct rhash_head *he;
476 	unsigned int hash;
477 
478 	tbl = rht_dereference_rcu(ht->tbl, ht);
479 restart:
480 	hash = rht_key_hashfn(ht, tbl, key, params);
481 	rht_for_each_rcu(he, tbl, hash) {
482 		if (params.obj_cmpfn ?
483 		    params.obj_cmpfn(&arg, rht_obj(ht, he)) :
484 		    rhashtable_compare(&arg, rht_obj(ht, he)))
485 			continue;
486 		return he;
487 	}
488 
489 	/* Ensure we see any new tables. */
490 	smp_rmb();
491 
492 	tbl = rht_dereference_rcu(tbl->future_tbl, ht);
493 	if (unlikely(tbl))
494 		goto restart;
495 
496 	return NULL;
497 }
498 
499 /**
500  * rhashtable_lookup - search hash table
501  * @ht:		hash table
502  * @key:	the pointer to the key
503  * @params:	hash table parameters
504  *
505  * Computes the hash value for the key and traverses the bucket chain looking
506  * for a entry with an identical key. The first matching entry is returned.
507  *
508  * This must only be called under the RCU read lock.
509  *
510  * Returns the first entry on which the compare function returned true.
511  */
512 static inline void *rhashtable_lookup(
513 	struct rhashtable *ht, const void *key,
514 	const struct rhashtable_params params)
515 {
516 	struct rhash_head *he = __rhashtable_lookup(ht, key, params);
517 
518 	return he ? rht_obj(ht, he) : NULL;
519 }
520 
521 /**
522  * rhashtable_lookup_fast - search hash table, without RCU read lock
523  * @ht:		hash table
524  * @key:	the pointer to the key
525  * @params:	hash table parameters
526  *
527  * Computes the hash value for the key and traverses the bucket chain looking
528  * for a entry with an identical key. The first matching entry is returned.
529  *
530  * Only use this function when you have other mechanisms guaranteeing
531  * that the object won't go away after the RCU read lock is released.
532  *
533  * Returns the first entry on which the compare function returned true.
534  */
535 static inline void *rhashtable_lookup_fast(
536 	struct rhashtable *ht, const void *key,
537 	const struct rhashtable_params params)
538 {
539 	void *obj;
540 
541 	rcu_read_lock();
542 	obj = rhashtable_lookup(ht, key, params);
543 	rcu_read_unlock();
544 
545 	return obj;
546 }
547 
548 /**
549  * rhltable_lookup - search hash list table
550  * @hlt:	hash table
551  * @key:	the pointer to the key
552  * @params:	hash table parameters
553  *
554  * Computes the hash value for the key and traverses the bucket chain looking
555  * for a entry with an identical key.  All matching entries are returned
556  * in a list.
557  *
558  * This must only be called under the RCU read lock.
559  *
560  * Returns the list of entries that match the given key.
561  */
562 static inline struct rhlist_head *rhltable_lookup(
563 	struct rhltable *hlt, const void *key,
564 	const struct rhashtable_params params)
565 {
566 	struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params);
567 
568 	return he ? container_of(he, struct rhlist_head, rhead) : NULL;
569 }
570 
571 /* Internal function, please use rhashtable_insert_fast() instead. This
572  * function returns the existing element already in hashes in there is a clash,
573  * otherwise it returns an error via ERR_PTR().
574  */
575 static inline void *__rhashtable_insert_fast(
576 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
577 	const struct rhashtable_params params, bool rhlist)
578 {
579 	struct rhashtable_compare_arg arg = {
580 		.ht = ht,
581 		.key = key,
582 	};
583 	struct rhash_head __rcu **pprev;
584 	struct bucket_table *tbl;
585 	struct rhash_head *head;
586 	spinlock_t *lock;
587 	unsigned int hash;
588 	int elasticity;
589 	void *data;
590 
591 	rcu_read_lock();
592 
593 	tbl = rht_dereference_rcu(ht->tbl, ht);
594 	hash = rht_head_hashfn(ht, tbl, obj, params);
595 	lock = rht_bucket_lock(tbl, hash);
596 	spin_lock_bh(lock);
597 
598 	if (unlikely(rcu_access_pointer(tbl->future_tbl))) {
599 slow_path:
600 		spin_unlock_bh(lock);
601 		rcu_read_unlock();
602 		return rhashtable_insert_slow(ht, key, obj);
603 	}
604 
605 	elasticity = RHT_ELASTICITY;
606 	pprev = rht_bucket_insert(ht, tbl, hash);
607 	data = ERR_PTR(-ENOMEM);
608 	if (!pprev)
609 		goto out;
610 
611 	rht_for_each_continue(head, *pprev, tbl, hash) {
612 		struct rhlist_head *plist;
613 		struct rhlist_head *list;
614 
615 		elasticity--;
616 		if (!key ||
617 		    (params.obj_cmpfn ?
618 		     params.obj_cmpfn(&arg, rht_obj(ht, head)) :
619 		     rhashtable_compare(&arg, rht_obj(ht, head)))) {
620 			pprev = &head->next;
621 			continue;
622 		}
623 
624 		data = rht_obj(ht, head);
625 
626 		if (!rhlist)
627 			goto out;
628 
629 
630 		list = container_of(obj, struct rhlist_head, rhead);
631 		plist = container_of(head, struct rhlist_head, rhead);
632 
633 		RCU_INIT_POINTER(list->next, plist);
634 		head = rht_dereference_bucket(head->next, tbl, hash);
635 		RCU_INIT_POINTER(list->rhead.next, head);
636 		rcu_assign_pointer(*pprev, obj);
637 
638 		goto good;
639 	}
640 
641 	if (elasticity <= 0)
642 		goto slow_path;
643 
644 	data = ERR_PTR(-E2BIG);
645 	if (unlikely(rht_grow_above_max(ht, tbl)))
646 		goto out;
647 
648 	if (unlikely(rht_grow_above_100(ht, tbl)))
649 		goto slow_path;
650 
651 	head = rht_dereference_bucket(*pprev, tbl, hash);
652 
653 	RCU_INIT_POINTER(obj->next, head);
654 	if (rhlist) {
655 		struct rhlist_head *list;
656 
657 		list = container_of(obj, struct rhlist_head, rhead);
658 		RCU_INIT_POINTER(list->next, NULL);
659 	}
660 
661 	rcu_assign_pointer(*pprev, obj);
662 
663 	atomic_inc(&ht->nelems);
664 	if (rht_grow_above_75(ht, tbl))
665 		schedule_work(&ht->run_work);
666 
667 good:
668 	data = NULL;
669 
670 out:
671 	spin_unlock_bh(lock);
672 	rcu_read_unlock();
673 
674 	return data;
675 }
676 
677 /**
678  * rhashtable_insert_fast - insert object into hash table
679  * @ht:		hash table
680  * @obj:	pointer to hash head inside object
681  * @params:	hash table parameters
682  *
683  * Will take a per bucket spinlock to protect against mutual mutations
684  * on the same bucket. Multiple insertions may occur in parallel unless
685  * they map to the same bucket lock.
686  *
687  * It is safe to call this function from atomic context.
688  *
689  * Will trigger an automatic deferred table resizing if residency in the
690  * table grows beyond 70%.
691  */
692 static inline int rhashtable_insert_fast(
693 	struct rhashtable *ht, struct rhash_head *obj,
694 	const struct rhashtable_params params)
695 {
696 	void *ret;
697 
698 	ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
699 	if (IS_ERR(ret))
700 		return PTR_ERR(ret);
701 
702 	return ret == NULL ? 0 : -EEXIST;
703 }
704 
705 /**
706  * rhltable_insert_key - insert object into hash list table
707  * @hlt:	hash list table
708  * @key:	the pointer to the key
709  * @list:	pointer to hash list head inside object
710  * @params:	hash table parameters
711  *
712  * Will take a per bucket spinlock to protect against mutual mutations
713  * on the same bucket. Multiple insertions may occur in parallel unless
714  * they map to the same bucket lock.
715  *
716  * It is safe to call this function from atomic context.
717  *
718  * Will trigger an automatic deferred table resizing if residency in the
719  * table grows beyond 70%.
720  */
721 static inline int rhltable_insert_key(
722 	struct rhltable *hlt, const void *key, struct rhlist_head *list,
723 	const struct rhashtable_params params)
724 {
725 	return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
726 						params, true));
727 }
728 
729 /**
730  * rhltable_insert - insert object into hash list table
731  * @hlt:	hash list table
732  * @list:	pointer to hash list head inside object
733  * @params:	hash table parameters
734  *
735  * Will take a per bucket spinlock to protect against mutual mutations
736  * on the same bucket. Multiple insertions may occur in parallel unless
737  * they map to the same bucket lock.
738  *
739  * It is safe to call this function from atomic context.
740  *
741  * Will trigger an automatic deferred table resizing if residency in the
742  * table grows beyond 70%.
743  */
744 static inline int rhltable_insert(
745 	struct rhltable *hlt, struct rhlist_head *list,
746 	const struct rhashtable_params params)
747 {
748 	const char *key = rht_obj(&hlt->ht, &list->rhead);
749 
750 	key += params.key_offset;
751 
752 	return rhltable_insert_key(hlt, key, list, params);
753 }
754 
755 /**
756  * rhashtable_lookup_insert_fast - lookup and insert object into hash table
757  * @ht:		hash table
758  * @obj:	pointer to hash head inside object
759  * @params:	hash table parameters
760  *
761  * Locks down the bucket chain in both the old and new table if a resize
762  * is in progress to ensure that writers can't remove from the old table
763  * and can't insert to the new table during the atomic operation of search
764  * and insertion. Searches for duplicates in both the old and new table if
765  * a resize is in progress.
766  *
767  * This lookup function may only be used for fixed key hash table (key_len
768  * parameter set). It will BUG() if used inappropriately.
769  *
770  * It is safe to call this function from atomic context.
771  *
772  * Will trigger an automatic deferred table resizing if residency in the
773  * table grows beyond 70%.
774  */
775 static inline int rhashtable_lookup_insert_fast(
776 	struct rhashtable *ht, struct rhash_head *obj,
777 	const struct rhashtable_params params)
778 {
779 	const char *key = rht_obj(ht, obj);
780 	void *ret;
781 
782 	BUG_ON(ht->p.obj_hashfn);
783 
784 	ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
785 				       false);
786 	if (IS_ERR(ret))
787 		return PTR_ERR(ret);
788 
789 	return ret == NULL ? 0 : -EEXIST;
790 }
791 
792 /**
793  * rhashtable_lookup_get_insert_fast - lookup and insert object into hash table
794  * @ht:		hash table
795  * @obj:	pointer to hash head inside object
796  * @params:	hash table parameters
797  *
798  * Just like rhashtable_lookup_insert_fast(), but this function returns the
799  * object if it exists, NULL if it did not and the insertion was successful,
800  * and an ERR_PTR otherwise.
801  */
802 static inline void *rhashtable_lookup_get_insert_fast(
803 	struct rhashtable *ht, struct rhash_head *obj,
804 	const struct rhashtable_params params)
805 {
806 	const char *key = rht_obj(ht, obj);
807 
808 	BUG_ON(ht->p.obj_hashfn);
809 
810 	return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
811 					false);
812 }
813 
814 /**
815  * rhashtable_lookup_insert_key - search and insert object to hash table
816  *				  with explicit key
817  * @ht:		hash table
818  * @key:	key
819  * @obj:	pointer to hash head inside object
820  * @params:	hash table parameters
821  *
822  * Locks down the bucket chain in both the old and new table if a resize
823  * is in progress to ensure that writers can't remove from the old table
824  * and can't insert to the new table during the atomic operation of search
825  * and insertion. Searches for duplicates in both the old and new table if
826  * a resize is in progress.
827  *
828  * Lookups may occur in parallel with hashtable mutations and resizing.
829  *
830  * Will trigger an automatic deferred table resizing if residency in the
831  * table grows beyond 70%.
832  *
833  * Returns zero on success.
834  */
835 static inline int rhashtable_lookup_insert_key(
836 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
837 	const struct rhashtable_params params)
838 {
839 	void *ret;
840 
841 	BUG_ON(!ht->p.obj_hashfn || !key);
842 
843 	ret = __rhashtable_insert_fast(ht, key, obj, params, false);
844 	if (IS_ERR(ret))
845 		return PTR_ERR(ret);
846 
847 	return ret == NULL ? 0 : -EEXIST;
848 }
849 
850 /**
851  * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
852  * @ht:		hash table
853  * @obj:	pointer to hash head inside object
854  * @params:	hash table parameters
855  * @data:	pointer to element data already in hashes
856  *
857  * Just like rhashtable_lookup_insert_key(), but this function returns the
858  * object if it exists, NULL if it does not and the insertion was successful,
859  * and an ERR_PTR otherwise.
860  */
861 static inline void *rhashtable_lookup_get_insert_key(
862 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
863 	const struct rhashtable_params params)
864 {
865 	BUG_ON(!ht->p.obj_hashfn || !key);
866 
867 	return __rhashtable_insert_fast(ht, key, obj, params, false);
868 }
869 
870 /* Internal function, please use rhashtable_remove_fast() instead */
871 static inline int __rhashtable_remove_fast_one(
872 	struct rhashtable *ht, struct bucket_table *tbl,
873 	struct rhash_head *obj, const struct rhashtable_params params,
874 	bool rhlist)
875 {
876 	struct rhash_head __rcu **pprev;
877 	struct rhash_head *he;
878 	spinlock_t * lock;
879 	unsigned int hash;
880 	int err = -ENOENT;
881 
882 	hash = rht_head_hashfn(ht, tbl, obj, params);
883 	lock = rht_bucket_lock(tbl, hash);
884 
885 	spin_lock_bh(lock);
886 
887 	pprev = rht_bucket_var(tbl, hash);
888 	rht_for_each_continue(he, *pprev, tbl, hash) {
889 		struct rhlist_head *list;
890 
891 		list = container_of(he, struct rhlist_head, rhead);
892 
893 		if (he != obj) {
894 			struct rhlist_head __rcu **lpprev;
895 
896 			pprev = &he->next;
897 
898 			if (!rhlist)
899 				continue;
900 
901 			do {
902 				lpprev = &list->next;
903 				list = rht_dereference_bucket(list->next,
904 							      tbl, hash);
905 			} while (list && obj != &list->rhead);
906 
907 			if (!list)
908 				continue;
909 
910 			list = rht_dereference_bucket(list->next, tbl, hash);
911 			RCU_INIT_POINTER(*lpprev, list);
912 			err = 0;
913 			break;
914 		}
915 
916 		obj = rht_dereference_bucket(obj->next, tbl, hash);
917 		err = 1;
918 
919 		if (rhlist) {
920 			list = rht_dereference_bucket(list->next, tbl, hash);
921 			if (list) {
922 				RCU_INIT_POINTER(list->rhead.next, obj);
923 				obj = &list->rhead;
924 				err = 0;
925 			}
926 		}
927 
928 		rcu_assign_pointer(*pprev, obj);
929 		break;
930 	}
931 
932 	spin_unlock_bh(lock);
933 
934 	if (err > 0) {
935 		atomic_dec(&ht->nelems);
936 		if (unlikely(ht->p.automatic_shrinking &&
937 			     rht_shrink_below_30(ht, tbl)))
938 			schedule_work(&ht->run_work);
939 		err = 0;
940 	}
941 
942 	return err;
943 }
944 
945 /* Internal function, please use rhashtable_remove_fast() instead */
946 static inline int __rhashtable_remove_fast(
947 	struct rhashtable *ht, struct rhash_head *obj,
948 	const struct rhashtable_params params, bool rhlist)
949 {
950 	struct bucket_table *tbl;
951 	int err;
952 
953 	rcu_read_lock();
954 
955 	tbl = rht_dereference_rcu(ht->tbl, ht);
956 
957 	/* Because we have already taken (and released) the bucket
958 	 * lock in old_tbl, if we find that future_tbl is not yet
959 	 * visible then that guarantees the entry to still be in
960 	 * the old tbl if it exists.
961 	 */
962 	while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
963 						   rhlist)) &&
964 	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
965 		;
966 
967 	rcu_read_unlock();
968 
969 	return err;
970 }
971 
972 /**
973  * rhashtable_remove_fast - remove object from hash table
974  * @ht:		hash table
975  * @obj:	pointer to hash head inside object
976  * @params:	hash table parameters
977  *
978  * Since the hash chain is single linked, the removal operation needs to
979  * walk the bucket chain upon removal. The removal operation is thus
980  * considerable slow if the hash table is not correctly sized.
981  *
982  * Will automatically shrink the table if permitted when residency drops
983  * below 30%.
984  *
985  * Returns zero on success, -ENOENT if the entry could not be found.
986  */
987 static inline int rhashtable_remove_fast(
988 	struct rhashtable *ht, struct rhash_head *obj,
989 	const struct rhashtable_params params)
990 {
991 	return __rhashtable_remove_fast(ht, obj, params, false);
992 }
993 
994 /**
995  * rhltable_remove - remove object from hash list table
996  * @hlt:	hash list table
997  * @list:	pointer to hash list head inside object
998  * @params:	hash table parameters
999  *
1000  * Since the hash chain is single linked, the removal operation needs to
1001  * walk the bucket chain upon removal. The removal operation is thus
1002  * considerable slow if the hash table is not correctly sized.
1003  *
1004  * Will automatically shrink the table if permitted when residency drops
1005  * below 30%
1006  *
1007  * Returns zero on success, -ENOENT if the entry could not be found.
1008  */
1009 static inline int rhltable_remove(
1010 	struct rhltable *hlt, struct rhlist_head *list,
1011 	const struct rhashtable_params params)
1012 {
1013 	return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
1014 }
1015 
1016 /* Internal function, please use rhashtable_replace_fast() instead */
1017 static inline int __rhashtable_replace_fast(
1018 	struct rhashtable *ht, struct bucket_table *tbl,
1019 	struct rhash_head *obj_old, struct rhash_head *obj_new,
1020 	const struct rhashtable_params params)
1021 {
1022 	struct rhash_head __rcu **pprev;
1023 	struct rhash_head *he;
1024 	spinlock_t *lock;
1025 	unsigned int hash;
1026 	int err = -ENOENT;
1027 
1028 	/* Minimally, the old and new objects must have same hash
1029 	 * (which should mean identifiers are the same).
1030 	 */
1031 	hash = rht_head_hashfn(ht, tbl, obj_old, params);
1032 	if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1033 		return -EINVAL;
1034 
1035 	lock = rht_bucket_lock(tbl, hash);
1036 
1037 	spin_lock_bh(lock);
1038 
1039 	pprev = rht_bucket_var(tbl, hash);
1040 	rht_for_each_continue(he, *pprev, tbl, hash) {
1041 		if (he != obj_old) {
1042 			pprev = &he->next;
1043 			continue;
1044 		}
1045 
1046 		rcu_assign_pointer(obj_new->next, obj_old->next);
1047 		rcu_assign_pointer(*pprev, obj_new);
1048 		err = 0;
1049 		break;
1050 	}
1051 
1052 	spin_unlock_bh(lock);
1053 
1054 	return err;
1055 }
1056 
1057 /**
1058  * rhashtable_replace_fast - replace an object in hash table
1059  * @ht:		hash table
1060  * @obj_old:	pointer to hash head inside object being replaced
1061  * @obj_new:	pointer to hash head inside object which is new
1062  * @params:	hash table parameters
1063  *
1064  * Replacing an object doesn't affect the number of elements in the hash table
1065  * or bucket, so we don't need to worry about shrinking or expanding the
1066  * table here.
1067  *
1068  * Returns zero on success, -ENOENT if the entry could not be found,
1069  * -EINVAL if hash is not the same for the old and new objects.
1070  */
1071 static inline int rhashtable_replace_fast(
1072 	struct rhashtable *ht, struct rhash_head *obj_old,
1073 	struct rhash_head *obj_new,
1074 	const struct rhashtable_params params)
1075 {
1076 	struct bucket_table *tbl;
1077 	int err;
1078 
1079 	rcu_read_lock();
1080 
1081 	tbl = rht_dereference_rcu(ht->tbl, ht);
1082 
1083 	/* Because we have already taken (and released) the bucket
1084 	 * lock in old_tbl, if we find that future_tbl is not yet
1085 	 * visible then that guarantees the entry to still be in
1086 	 * the old tbl if it exists.
1087 	 */
1088 	while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1089 						obj_new, params)) &&
1090 	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1091 		;
1092 
1093 	rcu_read_unlock();
1094 
1095 	return err;
1096 }
1097 
1098 /* Obsolete function, do not use in new code. */
1099 static inline int rhashtable_walk_init(struct rhashtable *ht,
1100 				       struct rhashtable_iter *iter, gfp_t gfp)
1101 {
1102 	rhashtable_walk_enter(ht, iter);
1103 	return 0;
1104 }
1105 
1106 /**
1107  * rhltable_walk_enter - Initialise an iterator
1108  * @hlt:	Table to walk over
1109  * @iter:	Hash table Iterator
1110  *
1111  * This function prepares a hash table walk.
1112  *
1113  * Note that if you restart a walk after rhashtable_walk_stop you
1114  * may see the same object twice.  Also, you may miss objects if
1115  * there are removals in between rhashtable_walk_stop and the next
1116  * call to rhashtable_walk_start.
1117  *
1118  * For a completely stable walk you should construct your own data
1119  * structure outside the hash table.
1120  *
1121  * This function may be called from any process context, including
1122  * non-preemptable context, but cannot be called from softirq or
1123  * hardirq context.
1124  *
1125  * You must call rhashtable_walk_exit after this function returns.
1126  */
1127 static inline void rhltable_walk_enter(struct rhltable *hlt,
1128 				       struct rhashtable_iter *iter)
1129 {
1130 	return rhashtable_walk_enter(&hlt->ht, iter);
1131 }
1132 
1133 /**
1134  * rhltable_free_and_destroy - free elements and destroy hash list table
1135  * @hlt:	the hash list table to destroy
1136  * @free_fn:	callback to release resources of element
1137  * @arg:	pointer passed to free_fn
1138  *
1139  * See documentation for rhashtable_free_and_destroy.
1140  */
1141 static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1142 					     void (*free_fn)(void *ptr,
1143 							     void *arg),
1144 					     void *arg)
1145 {
1146 	return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1147 }
1148 
1149 static inline void rhltable_destroy(struct rhltable *hlt)
1150 {
1151 	return rhltable_free_and_destroy(hlt, NULL, NULL);
1152 }
1153 
1154 #endif /* _LINUX_RHASHTABLE_H */
1155