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/atomic.h>
21#include <linux/compiler.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/mutex.h>
27#include <linux/rcupdate.h>
28
29/*
30 * The end of the chain is marked with a special nulls marks which has
31 * the following format:
32 *
33 * +-------+-----------------------------------------------------+-+
34 * | Base  |                      Hash                           |1|
35 * +-------+-----------------------------------------------------+-+
36 *
37 * Base (4 bits) : Reserved to distinguish between multiple tables.
38 *                 Specified via &struct rhashtable_params.nulls_base.
39 * Hash (27 bits): Full hash (unmasked) of first element added to bucket
40 * 1 (1 bit)     : Nulls marker (always set)
41 *
42 * The remaining bits of the next pointer remain unused for now.
43 */
44#define RHT_BASE_BITS		4
45#define RHT_HASH_BITS		27
46#define RHT_BASE_SHIFT		RHT_HASH_BITS
47
48/* Base bits plus 1 bit for nulls marker */
49#define RHT_HASH_RESERVED_SPACE	(RHT_BASE_BITS + 1)
50
51struct rhash_head {
52	struct rhash_head __rcu		*next;
53};
54
55/**
56 * struct bucket_table - Table of hash buckets
57 * @size: Number of hash buckets
58 * @rehash: Current bucket being rehashed
59 * @hash_rnd: Random seed to fold into hash
60 * @locks_mask: Mask to apply before accessing locks[]
61 * @locks: Array of spinlocks protecting individual buckets
62 * @walkers: List of active walkers
63 * @rcu: RCU structure for freeing the table
64 * @future_tbl: Table under construction during rehashing
65 * @buckets: size * hash buckets
66 */
67struct bucket_table {
68	unsigned int		size;
69	unsigned int		rehash;
70	u32			hash_rnd;
71	unsigned int		locks_mask;
72	spinlock_t		*locks;
73	struct list_head	walkers;
74	struct rcu_head		rcu;
75
76	struct bucket_table __rcu *future_tbl;
77
78	struct rhash_head __rcu	*buckets[] ____cacheline_aligned_in_smp;
79};
80
81/**
82 * struct rhashtable_compare_arg - Key for the function rhashtable_compare
83 * @ht: Hash table
84 * @key: Key to compare against
85 */
86struct rhashtable_compare_arg {
87	struct rhashtable *ht;
88	const void *key;
89};
90
91typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
92typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed);
93typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
94			       const void *obj);
95
96struct rhashtable;
97
98/**
99 * struct rhashtable_params - Hash table construction parameters
100 * @nelem_hint: Hint on number of elements, should be 75% of desired size
101 * @key_len: Length of key
102 * @key_offset: Offset of key in struct to be hashed
103 * @head_offset: Offset of rhash_head in struct to be hashed
104 * @insecure_max_entries: Maximum number of entries (may be exceeded)
105 * @max_size: Maximum size while expanding
106 * @min_size: Minimum size while shrinking
107 * @nulls_base: Base value to generate nulls marker
108 * @insecure_elasticity: Set to true to disable chain length checks
109 * @automatic_shrinking: Enable automatic shrinking of tables
110 * @locks_mul: Number of bucket locks to allocate per cpu (default: 128)
111 * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
112 * @obj_hashfn: Function to hash object
113 * @obj_cmpfn: Function to compare key with object
114 */
115struct rhashtable_params {
116	size_t			nelem_hint;
117	size_t			key_len;
118	size_t			key_offset;
119	size_t			head_offset;
120	unsigned int		insecure_max_entries;
121	unsigned int		max_size;
122	unsigned int		min_size;
123	u32			nulls_base;
124	bool			insecure_elasticity;
125	bool			automatic_shrinking;
126	size_t			locks_mul;
127	rht_hashfn_t		hashfn;
128	rht_obj_hashfn_t	obj_hashfn;
129	rht_obj_cmpfn_t		obj_cmpfn;
130};
131
132/**
133 * struct rhashtable - Hash table handle
134 * @tbl: Bucket table
135 * @nelems: Number of elements in table
136 * @key_len: Key length for hashfn
137 * @elasticity: Maximum chain length before rehash
138 * @p: Configuration parameters
139 * @run_work: Deferred worker to expand/shrink asynchronously
140 * @mutex: Mutex to protect current/future table swapping
141 * @lock: Spin lock to protect walker list
142 */
143struct rhashtable {
144	struct bucket_table __rcu	*tbl;
145	atomic_t			nelems;
146	unsigned int			key_len;
147	unsigned int			elasticity;
148	struct rhashtable_params	p;
149	struct work_struct		run_work;
150	struct mutex                    mutex;
151	spinlock_t			lock;
152};
153
154/**
155 * struct rhashtable_walker - Hash table walker
156 * @list: List entry on list of walkers
157 * @tbl: The table that we were walking over
158 */
159struct rhashtable_walker {
160	struct list_head list;
161	struct bucket_table *tbl;
162};
163
164/**
165 * struct rhashtable_iter - Hash table iterator, fits into netlink cb
166 * @ht: Table to iterate through
167 * @p: Current pointer
168 * @walker: Associated rhashtable walker
169 * @slot: Current slot
170 * @skip: Number of entries to skip in slot
171 */
172struct rhashtable_iter {
173	struct rhashtable *ht;
174	struct rhash_head *p;
175	struct rhashtable_walker *walker;
176	unsigned int slot;
177	unsigned int skip;
178};
179
180static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
181{
182	return NULLS_MARKER(ht->p.nulls_base + hash);
183}
184
185#define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
186	((ptr) = (typeof(ptr)) rht_marker(ht, hash))
187
188static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
189{
190	return ((unsigned long) ptr & 1);
191}
192
193static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
194{
195	return ((unsigned long) ptr) >> 1;
196}
197
198static inline void *rht_obj(const struct rhashtable *ht,
199			    const struct rhash_head *he)
200{
201	return (char *)he - ht->p.head_offset;
202}
203
204static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
205					    unsigned int hash)
206{
207	return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
208}
209
210static inline unsigned int rht_key_hashfn(
211	struct rhashtable *ht, const struct bucket_table *tbl,
212	const void *key, const struct rhashtable_params params)
213{
214	unsigned int hash;
215
216	/* params must be equal to ht->p if it isn't constant. */
217	if (!__builtin_constant_p(params.key_len))
218		hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd);
219	else if (params.key_len) {
220		unsigned int key_len = params.key_len;
221
222		if (params.hashfn)
223			hash = params.hashfn(key, key_len, tbl->hash_rnd);
224		else if (key_len & (sizeof(u32) - 1))
225			hash = jhash(key, key_len, tbl->hash_rnd);
226		else
227			hash = jhash2(key, key_len / sizeof(u32),
228				      tbl->hash_rnd);
229	} else {
230		unsigned int key_len = ht->p.key_len;
231
232		if (params.hashfn)
233			hash = params.hashfn(key, key_len, tbl->hash_rnd);
234		else
235			hash = jhash(key, key_len, tbl->hash_rnd);
236	}
237
238	return rht_bucket_index(tbl, hash);
239}
240
241static inline unsigned int rht_head_hashfn(
242	struct rhashtable *ht, const struct bucket_table *tbl,
243	const struct rhash_head *he, const struct rhashtable_params params)
244{
245	const char *ptr = rht_obj(ht, he);
246
247	return likely(params.obj_hashfn) ?
248	       rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
249							    ht->p.key_len,
250						       tbl->hash_rnd)) :
251	       rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
252}
253
254/**
255 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
256 * @ht:		hash table
257 * @tbl:	current table
258 */
259static inline bool rht_grow_above_75(const struct rhashtable *ht,
260				     const struct bucket_table *tbl)
261{
262	/* Expand table when exceeding 75% load */
263	return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
264	       (!ht->p.max_size || tbl->size < ht->p.max_size);
265}
266
267/**
268 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
269 * @ht:		hash table
270 * @tbl:	current table
271 */
272static inline bool rht_shrink_below_30(const struct rhashtable *ht,
273				       const struct bucket_table *tbl)
274{
275	/* Shrink table beneath 30% load */
276	return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
277	       tbl->size > ht->p.min_size;
278}
279
280/**
281 * rht_grow_above_100 - returns true if nelems > table-size
282 * @ht:		hash table
283 * @tbl:	current table
284 */
285static inline bool rht_grow_above_100(const struct rhashtable *ht,
286				      const struct bucket_table *tbl)
287{
288	return atomic_read(&ht->nelems) > tbl->size &&
289		(!ht->p.max_size || tbl->size < ht->p.max_size);
290}
291
292/**
293 * rht_grow_above_max - returns true if table is above maximum
294 * @ht:		hash table
295 * @tbl:	current table
296 */
297static inline bool rht_grow_above_max(const struct rhashtable *ht,
298				      const struct bucket_table *tbl)
299{
300	return ht->p.insecure_max_entries &&
301	       atomic_read(&ht->nelems) >= ht->p.insecure_max_entries;
302}
303
304/* The bucket lock is selected based on the hash and protects mutations
305 * on a group of hash buckets.
306 *
307 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
308 * a single lock always covers both buckets which may both contains
309 * entries which link to the same bucket of the old table during resizing.
310 * This allows to simplify the locking as locking the bucket in both
311 * tables during resize always guarantee protection.
312 *
313 * IMPORTANT: When holding the bucket lock of both the old and new table
314 * during expansions and shrinking, the old bucket lock must always be
315 * acquired first.
316 */
317static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
318					  unsigned int hash)
319{
320	return &tbl->locks[hash & tbl->locks_mask];
321}
322
323#ifdef CONFIG_PROVE_LOCKING
324int lockdep_rht_mutex_is_held(struct rhashtable *ht);
325int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
326#else
327static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
328{
329	return 1;
330}
331
332static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
333					     u32 hash)
334{
335	return 1;
336}
337#endif /* CONFIG_PROVE_LOCKING */
338
339int rhashtable_init(struct rhashtable *ht,
340		    const struct rhashtable_params *params);
341
342int rhashtable_insert_slow(struct rhashtable *ht, const void *key,
343			   struct rhash_head *obj,
344			   struct bucket_table *old_tbl);
345int rhashtable_insert_rehash(struct rhashtable *ht);
346
347int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter);
348void rhashtable_walk_exit(struct rhashtable_iter *iter);
349int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU);
350void *rhashtable_walk_next(struct rhashtable_iter *iter);
351void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
352
353void rhashtable_free_and_destroy(struct rhashtable *ht,
354				 void (*free_fn)(void *ptr, void *arg),
355				 void *arg);
356void rhashtable_destroy(struct rhashtable *ht);
357
358#define rht_dereference(p, ht) \
359	rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
360
361#define rht_dereference_rcu(p, ht) \
362	rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
363
364#define rht_dereference_bucket(p, tbl, hash) \
365	rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
366
367#define rht_dereference_bucket_rcu(p, tbl, hash) \
368	rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
369
370#define rht_entry(tpos, pos, member) \
371	({ tpos = container_of(pos, typeof(*tpos), member); 1; })
372
373/**
374 * rht_for_each_continue - continue iterating over hash chain
375 * @pos:	the &struct rhash_head to use as a loop cursor.
376 * @head:	the previous &struct rhash_head to continue from
377 * @tbl:	the &struct bucket_table
378 * @hash:	the hash value / bucket index
379 */
380#define rht_for_each_continue(pos, head, tbl, hash) \
381	for (pos = rht_dereference_bucket(head, tbl, hash); \
382	     !rht_is_a_nulls(pos); \
383	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
384
385/**
386 * rht_for_each - iterate over hash chain
387 * @pos:	the &struct rhash_head to use as a loop cursor.
388 * @tbl:	the &struct bucket_table
389 * @hash:	the hash value / bucket index
390 */
391#define rht_for_each(pos, tbl, hash) \
392	rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash)
393
394/**
395 * rht_for_each_entry_continue - continue iterating over hash chain
396 * @tpos:	the type * to use as a loop cursor.
397 * @pos:	the &struct rhash_head to use as a loop cursor.
398 * @head:	the previous &struct rhash_head to continue from
399 * @tbl:	the &struct bucket_table
400 * @hash:	the hash value / bucket index
401 * @member:	name of the &struct rhash_head within the hashable struct.
402 */
403#define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member)	\
404	for (pos = rht_dereference_bucket(head, tbl, hash);		\
405	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	\
406	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
407
408/**
409 * rht_for_each_entry - iterate over hash chain of given type
410 * @tpos:	the type * to use as a loop cursor.
411 * @pos:	the &struct rhash_head to use as a loop cursor.
412 * @tbl:	the &struct bucket_table
413 * @hash:	the hash value / bucket index
414 * @member:	name of the &struct rhash_head within the hashable struct.
415 */
416#define rht_for_each_entry(tpos, pos, tbl, hash, member)		\
417	rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash],	\
418				    tbl, hash, member)
419
420/**
421 * rht_for_each_entry_safe - safely iterate over hash chain of given type
422 * @tpos:	the type * to use as a loop cursor.
423 * @pos:	the &struct rhash_head to use as a loop cursor.
424 * @next:	the &struct rhash_head to use as next in loop cursor.
425 * @tbl:	the &struct bucket_table
426 * @hash:	the hash value / bucket index
427 * @member:	name of the &struct rhash_head within the hashable struct.
428 *
429 * This hash chain list-traversal primitive allows for the looped code to
430 * remove the loop cursor from the list.
431 */
432#define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member)	    \
433	for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \
434	     next = !rht_is_a_nulls(pos) ?				    \
435		       rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
436	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	    \
437	     pos = next,						    \
438	     next = !rht_is_a_nulls(pos) ?				    \
439		       rht_dereference_bucket(pos->next, tbl, hash) : NULL)
440
441/**
442 * rht_for_each_rcu_continue - continue iterating over rcu hash chain
443 * @pos:	the &struct rhash_head to use as a loop cursor.
444 * @head:	the previous &struct rhash_head to continue from
445 * @tbl:	the &struct bucket_table
446 * @hash:	the hash value / bucket index
447 *
448 * This hash chain list-traversal primitive may safely run concurrently with
449 * the _rcu mutation primitives such as rhashtable_insert() as long as the
450 * traversal is guarded by rcu_read_lock().
451 */
452#define rht_for_each_rcu_continue(pos, head, tbl, hash)			\
453	for (({barrier(); }),						\
454	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		\
455	     !rht_is_a_nulls(pos);					\
456	     pos = rcu_dereference_raw(pos->next))
457
458/**
459 * rht_for_each_rcu - iterate over rcu hash chain
460 * @pos:	the &struct rhash_head to use as a loop cursor.
461 * @tbl:	the &struct bucket_table
462 * @hash:	the hash value / bucket index
463 *
464 * This hash chain list-traversal primitive may safely run concurrently with
465 * the _rcu mutation primitives such as rhashtable_insert() as long as the
466 * traversal is guarded by rcu_read_lock().
467 */
468#define rht_for_each_rcu(pos, tbl, hash)				\
469	rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash)
470
471/**
472 * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
473 * @tpos:	the type * to use as a loop cursor.
474 * @pos:	the &struct rhash_head to use as a loop cursor.
475 * @head:	the previous &struct rhash_head to continue from
476 * @tbl:	the &struct bucket_table
477 * @hash:	the hash value / bucket index
478 * @member:	name of the &struct rhash_head within the hashable struct.
479 *
480 * This hash chain list-traversal primitive may safely run concurrently with
481 * the _rcu mutation primitives such as rhashtable_insert() as long as the
482 * traversal is guarded by rcu_read_lock().
483 */
484#define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
485	for (({barrier(); }),						    \
486	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		    \
487	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	    \
488	     pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
489
490/**
491 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
492 * @tpos:	the type * to use as a loop cursor.
493 * @pos:	the &struct rhash_head to use as a loop cursor.
494 * @tbl:	the &struct bucket_table
495 * @hash:	the hash value / bucket index
496 * @member:	name of the &struct rhash_head within the hashable struct.
497 *
498 * This hash chain list-traversal primitive may safely run concurrently with
499 * the _rcu mutation primitives such as rhashtable_insert() as long as the
500 * traversal is guarded by rcu_read_lock().
501 */
502#define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member)		\
503	rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\
504					tbl, hash, member)
505
506static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
507				     const void *obj)
508{
509	struct rhashtable *ht = arg->ht;
510	const char *ptr = obj;
511
512	return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
513}
514
515/**
516 * rhashtable_lookup_fast - search hash table, inlined version
517 * @ht:		hash table
518 * @key:	the pointer to the key
519 * @params:	hash table parameters
520 *
521 * Computes the hash value for the key and traverses the bucket chain looking
522 * for a entry with an identical key. The first matching entry is returned.
523 *
524 * Returns the first entry on which the compare function returned true.
525 */
526static inline void *rhashtable_lookup_fast(
527	struct rhashtable *ht, const void *key,
528	const struct rhashtable_params params)
529{
530	struct rhashtable_compare_arg arg = {
531		.ht = ht,
532		.key = key,
533	};
534	const struct bucket_table *tbl;
535	struct rhash_head *he;
536	unsigned int hash;
537
538	rcu_read_lock();
539
540	tbl = rht_dereference_rcu(ht->tbl, ht);
541restart:
542	hash = rht_key_hashfn(ht, tbl, key, params);
543	rht_for_each_rcu(he, tbl, hash) {
544		if (params.obj_cmpfn ?
545		    params.obj_cmpfn(&arg, rht_obj(ht, he)) :
546		    rhashtable_compare(&arg, rht_obj(ht, he)))
547			continue;
548		rcu_read_unlock();
549		return rht_obj(ht, he);
550	}
551
552	/* Ensure we see any new tables. */
553	smp_rmb();
554
555	tbl = rht_dereference_rcu(tbl->future_tbl, ht);
556	if (unlikely(tbl))
557		goto restart;
558	rcu_read_unlock();
559
560	return NULL;
561}
562
563/* Internal function, please use rhashtable_insert_fast() instead */
564static inline int __rhashtable_insert_fast(
565	struct rhashtable *ht, const void *key, struct rhash_head *obj,
566	const struct rhashtable_params params)
567{
568	struct rhashtable_compare_arg arg = {
569		.ht = ht,
570		.key = key,
571	};
572	struct bucket_table *tbl, *new_tbl;
573	struct rhash_head *head;
574	spinlock_t *lock;
575	unsigned int elasticity;
576	unsigned int hash;
577	int err;
578
579restart:
580	rcu_read_lock();
581
582	tbl = rht_dereference_rcu(ht->tbl, ht);
583
584	/* All insertions must grab the oldest table containing
585	 * the hashed bucket that is yet to be rehashed.
586	 */
587	for (;;) {
588		hash = rht_head_hashfn(ht, tbl, obj, params);
589		lock = rht_bucket_lock(tbl, hash);
590		spin_lock_bh(lock);
591
592		if (tbl->rehash <= hash)
593			break;
594
595		spin_unlock_bh(lock);
596		tbl = rht_dereference_rcu(tbl->future_tbl, ht);
597	}
598
599	new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
600	if (unlikely(new_tbl)) {
601		err = rhashtable_insert_slow(ht, key, obj, new_tbl);
602		if (err == -EAGAIN)
603			goto slow_path;
604		goto out;
605	}
606
607	err = -E2BIG;
608	if (unlikely(rht_grow_above_max(ht, tbl)))
609		goto out;
610
611	if (unlikely(rht_grow_above_100(ht, tbl))) {
612slow_path:
613		spin_unlock_bh(lock);
614		err = rhashtable_insert_rehash(ht);
615		rcu_read_unlock();
616		if (err)
617			return err;
618
619		goto restart;
620	}
621
622	err = -EEXIST;
623	elasticity = ht->elasticity;
624	rht_for_each(head, tbl, hash) {
625		if (key &&
626		    unlikely(!(params.obj_cmpfn ?
627			       params.obj_cmpfn(&arg, rht_obj(ht, head)) :
628			       rhashtable_compare(&arg, rht_obj(ht, head)))))
629			goto out;
630		if (!--elasticity)
631			goto slow_path;
632	}
633
634	err = 0;
635
636	head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
637
638	RCU_INIT_POINTER(obj->next, head);
639
640	rcu_assign_pointer(tbl->buckets[hash], obj);
641
642	atomic_inc(&ht->nelems);
643	if (rht_grow_above_75(ht, tbl))
644		schedule_work(&ht->run_work);
645
646out:
647	spin_unlock_bh(lock);
648	rcu_read_unlock();
649
650	return err;
651}
652
653/**
654 * rhashtable_insert_fast - insert object into hash table
655 * @ht:		hash table
656 * @obj:	pointer to hash head inside object
657 * @params:	hash table parameters
658 *
659 * Will take a per bucket spinlock to protect against mutual mutations
660 * on the same bucket. Multiple insertions may occur in parallel unless
661 * they map to the same bucket lock.
662 *
663 * It is safe to call this function from atomic context.
664 *
665 * Will trigger an automatic deferred table resizing if the size grows
666 * beyond the watermark indicated by grow_decision() which can be passed
667 * to rhashtable_init().
668 */
669static inline int rhashtable_insert_fast(
670	struct rhashtable *ht, struct rhash_head *obj,
671	const struct rhashtable_params params)
672{
673	return __rhashtable_insert_fast(ht, NULL, obj, params);
674}
675
676/**
677 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
678 * @ht:		hash table
679 * @obj:	pointer to hash head inside object
680 * @params:	hash table parameters
681 *
682 * Locks down the bucket chain in both the old and new table if a resize
683 * is in progress to ensure that writers can't remove from the old table
684 * and can't insert to the new table during the atomic operation of search
685 * and insertion. Searches for duplicates in both the old and new table if
686 * a resize is in progress.
687 *
688 * This lookup function may only be used for fixed key hash table (key_len
689 * parameter set). It will BUG() if used inappropriately.
690 *
691 * It is safe to call this function from atomic context.
692 *
693 * Will trigger an automatic deferred table resizing if the size grows
694 * beyond the watermark indicated by grow_decision() which can be passed
695 * to rhashtable_init().
696 */
697static inline int rhashtable_lookup_insert_fast(
698	struct rhashtable *ht, struct rhash_head *obj,
699	const struct rhashtable_params params)
700{
701	const char *key = rht_obj(ht, obj);
702
703	BUG_ON(ht->p.obj_hashfn);
704
705	return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj,
706					params);
707}
708
709/**
710 * rhashtable_lookup_insert_key - search and insert object to hash table
711 *				  with explicit key
712 * @ht:		hash table
713 * @key:	key
714 * @obj:	pointer to hash head inside object
715 * @params:	hash table parameters
716 *
717 * Locks down the bucket chain in both the old and new table if a resize
718 * is in progress to ensure that writers can't remove from the old table
719 * and can't insert to the new table during the atomic operation of search
720 * and insertion. Searches for duplicates in both the old and new table if
721 * a resize is in progress.
722 *
723 * Lookups may occur in parallel with hashtable mutations and resizing.
724 *
725 * Will trigger an automatic deferred table resizing if the size grows
726 * beyond the watermark indicated by grow_decision() which can be passed
727 * to rhashtable_init().
728 *
729 * Returns zero on success.
730 */
731static inline int rhashtable_lookup_insert_key(
732	struct rhashtable *ht, const void *key, struct rhash_head *obj,
733	const struct rhashtable_params params)
734{
735	BUG_ON(!ht->p.obj_hashfn || !key);
736
737	return __rhashtable_insert_fast(ht, key, obj, params);
738}
739
740/* Internal function, please use rhashtable_remove_fast() instead */
741static inline int __rhashtable_remove_fast(
742	struct rhashtable *ht, struct bucket_table *tbl,
743	struct rhash_head *obj, const struct rhashtable_params params)
744{
745	struct rhash_head __rcu **pprev;
746	struct rhash_head *he;
747	spinlock_t * lock;
748	unsigned int hash;
749	int err = -ENOENT;
750
751	hash = rht_head_hashfn(ht, tbl, obj, params);
752	lock = rht_bucket_lock(tbl, hash);
753
754	spin_lock_bh(lock);
755
756	pprev = &tbl->buckets[hash];
757	rht_for_each(he, tbl, hash) {
758		if (he != obj) {
759			pprev = &he->next;
760			continue;
761		}
762
763		rcu_assign_pointer(*pprev, obj->next);
764		err = 0;
765		break;
766	}
767
768	spin_unlock_bh(lock);
769
770	return err;
771}
772
773/**
774 * rhashtable_remove_fast - remove object from hash table
775 * @ht:		hash table
776 * @obj:	pointer to hash head inside object
777 * @params:	hash table parameters
778 *
779 * Since the hash chain is single linked, the removal operation needs to
780 * walk the bucket chain upon removal. The removal operation is thus
781 * considerable slow if the hash table is not correctly sized.
782 *
783 * Will automatically shrink the table via rhashtable_expand() if the
784 * shrink_decision function specified at rhashtable_init() returns true.
785 *
786 * Returns zero on success, -ENOENT if the entry could not be found.
787 */
788static inline int rhashtable_remove_fast(
789	struct rhashtable *ht, struct rhash_head *obj,
790	const struct rhashtable_params params)
791{
792	struct bucket_table *tbl;
793	int err;
794
795	rcu_read_lock();
796
797	tbl = rht_dereference_rcu(ht->tbl, ht);
798
799	/* Because we have already taken (and released) the bucket
800	 * lock in old_tbl, if we find that future_tbl is not yet
801	 * visible then that guarantees the entry to still be in
802	 * the old tbl if it exists.
803	 */
804	while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) &&
805	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
806		;
807
808	if (err)
809		goto out;
810
811	atomic_dec(&ht->nelems);
812	if (unlikely(ht->p.automatic_shrinking &&
813		     rht_shrink_below_30(ht, tbl)))
814		schedule_work(&ht->run_work);
815
816out:
817	rcu_read_unlock();
818
819	return err;
820}
821
822#endif /* _LINUX_RHASHTABLE_H */
823