xref: /linux/include/linux/rhashtable.h (revision 83e007a0c6a3f4bfdf8f3f8d0fc266cda189b3d6)
1 /*
2  * Resizable, Scalable, Concurrent Hash Table
3  *
4  * Copyright (c) 2015-2016 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/err.h>
23 #include <linux/errno.h>
24 #include <linux/jhash.h>
25 #include <linux/list_nulls.h>
26 #include <linux/workqueue.h>
27 #include <linux/mutex.h>
28 #include <linux/rculist.h>
29 
30 /*
31  * The end of the chain is marked with a special nulls marks which has
32  * the following format:
33  *
34  * +-------+-----------------------------------------------------+-+
35  * | Base  |                      Hash                           |1|
36  * +-------+-----------------------------------------------------+-+
37  *
38  * Base (4 bits) : Reserved to distinguish between multiple tables.
39  *                 Specified via &struct rhashtable_params.nulls_base.
40  * Hash (27 bits): Full hash (unmasked) of first element added to bucket
41  * 1 (1 bit)     : Nulls marker (always set)
42  *
43  * The remaining bits of the next pointer remain unused for now.
44  */
45 #define RHT_BASE_BITS		4
46 #define RHT_HASH_BITS		27
47 #define RHT_BASE_SHIFT		RHT_HASH_BITS
48 
49 /* Base bits plus 1 bit for nulls marker */
50 #define RHT_HASH_RESERVED_SPACE	(RHT_BASE_BITS + 1)
51 
52 struct rhash_head {
53 	struct rhash_head __rcu		*next;
54 };
55 
56 struct rhlist_head {
57 	struct rhash_head		rhead;
58 	struct rhlist_head __rcu	*next;
59 };
60 
61 /**
62  * struct bucket_table - Table of hash buckets
63  * @size: Number of hash buckets
64  * @nest: Number of bits of first-level nested table.
65  * @rehash: Current bucket being rehashed
66  * @hash_rnd: Random seed to fold into hash
67  * @locks_mask: Mask to apply before accessing locks[]
68  * @locks: Array of spinlocks protecting individual buckets
69  * @walkers: List of active walkers
70  * @rcu: RCU structure for freeing the table
71  * @future_tbl: Table under construction during rehashing
72  * @ntbl: Nested table used when out of memory.
73  * @buckets: size * hash buckets
74  */
75 struct bucket_table {
76 	unsigned int		size;
77 	unsigned int		nest;
78 	unsigned int		rehash;
79 	u32			hash_rnd;
80 	unsigned int		locks_mask;
81 	spinlock_t		*locks;
82 	struct list_head	walkers;
83 	struct rcu_head		rcu;
84 
85 	struct bucket_table __rcu *future_tbl;
86 
87 	struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp;
88 };
89 
90 /**
91  * struct rhashtable_compare_arg - Key for the function rhashtable_compare
92  * @ht: Hash table
93  * @key: Key to compare against
94  */
95 struct rhashtable_compare_arg {
96 	struct rhashtable *ht;
97 	const void *key;
98 };
99 
100 typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
101 typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed);
102 typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
103 			       const void *obj);
104 
105 struct rhashtable;
106 
107 /**
108  * struct rhashtable_params - Hash table construction parameters
109  * @nelem_hint: Hint on number of elements, should be 75% of desired size
110  * @key_len: Length of key
111  * @key_offset: Offset of key in struct to be hashed
112  * @head_offset: Offset of rhash_head in struct to be hashed
113  * @insecure_max_entries: Maximum number of entries (may be exceeded)
114  * @max_size: Maximum size while expanding
115  * @min_size: Minimum size while shrinking
116  * @nulls_base: Base value to generate nulls marker
117  * @insecure_elasticity: Set to true to disable chain length checks
118  * @automatic_shrinking: Enable automatic shrinking of tables
119  * @locks_mul: Number of bucket locks to allocate per cpu (default: 128)
120  * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
121  * @obj_hashfn: Function to hash object
122  * @obj_cmpfn: Function to compare key with object
123  */
124 struct rhashtable_params {
125 	size_t			nelem_hint;
126 	size_t			key_len;
127 	size_t			key_offset;
128 	size_t			head_offset;
129 	unsigned int		insecure_max_entries;
130 	unsigned int		max_size;
131 	unsigned int		min_size;
132 	u32			nulls_base;
133 	bool			insecure_elasticity;
134 	bool			automatic_shrinking;
135 	size_t			locks_mul;
136 	rht_hashfn_t		hashfn;
137 	rht_obj_hashfn_t	obj_hashfn;
138 	rht_obj_cmpfn_t		obj_cmpfn;
139 };
140 
141 /**
142  * struct rhashtable - Hash table handle
143  * @tbl: Bucket table
144  * @nelems: Number of elements in table
145  * @key_len: Key length for hashfn
146  * @elasticity: Maximum chain length before rehash
147  * @p: Configuration parameters
148  * @rhlist: True if this is an rhltable
149  * @run_work: Deferred worker to expand/shrink asynchronously
150  * @mutex: Mutex to protect current/future table swapping
151  * @lock: Spin lock to protect walker list
152  */
153 struct rhashtable {
154 	struct bucket_table __rcu	*tbl;
155 	atomic_t			nelems;
156 	unsigned int			key_len;
157 	unsigned int			elasticity;
158 	struct rhashtable_params	p;
159 	bool				rhlist;
160 	struct work_struct		run_work;
161 	struct mutex                    mutex;
162 	spinlock_t			lock;
163 };
164 
165 /**
166  * struct rhltable - Hash table with duplicate objects in a list
167  * @ht: Underlying rhtable
168  */
169 struct rhltable {
170 	struct rhashtable ht;
171 };
172 
173 /**
174  * struct rhashtable_walker - Hash table walker
175  * @list: List entry on list of walkers
176  * @tbl: The table that we were walking over
177  */
178 struct rhashtable_walker {
179 	struct list_head list;
180 	struct bucket_table *tbl;
181 };
182 
183 /**
184  * struct rhashtable_iter - Hash table iterator
185  * @ht: Table to iterate through
186  * @p: Current pointer
187  * @list: Current hash list pointer
188  * @walker: Associated rhashtable walker
189  * @slot: Current slot
190  * @skip: Number of entries to skip in slot
191  */
192 struct rhashtable_iter {
193 	struct rhashtable *ht;
194 	struct rhash_head *p;
195 	struct rhlist_head *list;
196 	struct rhashtable_walker walker;
197 	unsigned int slot;
198 	unsigned int skip;
199 };
200 
201 static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
202 {
203 	return NULLS_MARKER(ht->p.nulls_base + hash);
204 }
205 
206 #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
207 	((ptr) = (typeof(ptr)) rht_marker(ht, hash))
208 
209 static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
210 {
211 	return ((unsigned long) ptr & 1);
212 }
213 
214 static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
215 {
216 	return ((unsigned long) ptr) >> 1;
217 }
218 
219 static inline void *rht_obj(const struct rhashtable *ht,
220 			    const struct rhash_head *he)
221 {
222 	return (char *)he - ht->p.head_offset;
223 }
224 
225 static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
226 					    unsigned int hash)
227 {
228 	return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
229 }
230 
231 static inline unsigned int rht_key_hashfn(
232 	struct rhashtable *ht, const struct bucket_table *tbl,
233 	const void *key, const struct rhashtable_params params)
234 {
235 	unsigned int hash;
236 
237 	/* params must be equal to ht->p if it isn't constant. */
238 	if (!__builtin_constant_p(params.key_len))
239 		hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd);
240 	else if (params.key_len) {
241 		unsigned int key_len = params.key_len;
242 
243 		if (params.hashfn)
244 			hash = params.hashfn(key, key_len, tbl->hash_rnd);
245 		else if (key_len & (sizeof(u32) - 1))
246 			hash = jhash(key, key_len, tbl->hash_rnd);
247 		else
248 			hash = jhash2(key, key_len / sizeof(u32),
249 				      tbl->hash_rnd);
250 	} else {
251 		unsigned int key_len = ht->p.key_len;
252 
253 		if (params.hashfn)
254 			hash = params.hashfn(key, key_len, tbl->hash_rnd);
255 		else
256 			hash = jhash(key, key_len, tbl->hash_rnd);
257 	}
258 
259 	return rht_bucket_index(tbl, hash);
260 }
261 
262 static inline unsigned int rht_head_hashfn(
263 	struct rhashtable *ht, const struct bucket_table *tbl,
264 	const struct rhash_head *he, const struct rhashtable_params params)
265 {
266 	const char *ptr = rht_obj(ht, he);
267 
268 	return likely(params.obj_hashfn) ?
269 	       rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
270 							    ht->p.key_len,
271 						       tbl->hash_rnd)) :
272 	       rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
273 }
274 
275 /**
276  * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
277  * @ht:		hash table
278  * @tbl:	current table
279  */
280 static inline bool rht_grow_above_75(const struct rhashtable *ht,
281 				     const struct bucket_table *tbl)
282 {
283 	/* Expand table when exceeding 75% load */
284 	return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
285 	       (!ht->p.max_size || tbl->size < ht->p.max_size);
286 }
287 
288 /**
289  * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
290  * @ht:		hash table
291  * @tbl:	current table
292  */
293 static inline bool rht_shrink_below_30(const struct rhashtable *ht,
294 				       const struct bucket_table *tbl)
295 {
296 	/* Shrink table beneath 30% load */
297 	return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
298 	       tbl->size > ht->p.min_size;
299 }
300 
301 /**
302  * rht_grow_above_100 - returns true if nelems > table-size
303  * @ht:		hash table
304  * @tbl:	current table
305  */
306 static inline bool rht_grow_above_100(const struct rhashtable *ht,
307 				      const struct bucket_table *tbl)
308 {
309 	return atomic_read(&ht->nelems) > tbl->size &&
310 		(!ht->p.max_size || tbl->size < ht->p.max_size);
311 }
312 
313 /**
314  * rht_grow_above_max - returns true if table is above maximum
315  * @ht:		hash table
316  * @tbl:	current table
317  */
318 static inline bool rht_grow_above_max(const struct rhashtable *ht,
319 				      const struct bucket_table *tbl)
320 {
321 	return ht->p.insecure_max_entries &&
322 	       atomic_read(&ht->nelems) >= ht->p.insecure_max_entries;
323 }
324 
325 /* The bucket lock is selected based on the hash and protects mutations
326  * on a group of hash buckets.
327  *
328  * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
329  * a single lock always covers both buckets which may both contains
330  * entries which link to the same bucket of the old table during resizing.
331  * This allows to simplify the locking as locking the bucket in both
332  * tables during resize always guarantee protection.
333  *
334  * IMPORTANT: When holding the bucket lock of both the old and new table
335  * during expansions and shrinking, the old bucket lock must always be
336  * acquired first.
337  */
338 static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
339 					  unsigned int hash)
340 {
341 	return &tbl->locks[hash & tbl->locks_mask];
342 }
343 
344 #ifdef CONFIG_PROVE_LOCKING
345 int lockdep_rht_mutex_is_held(struct rhashtable *ht);
346 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
347 #else
348 static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
349 {
350 	return 1;
351 }
352 
353 static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
354 					     u32 hash)
355 {
356 	return 1;
357 }
358 #endif /* CONFIG_PROVE_LOCKING */
359 
360 int rhashtable_init(struct rhashtable *ht,
361 		    const struct rhashtable_params *params);
362 int rhltable_init(struct rhltable *hlt,
363 		  const struct rhashtable_params *params);
364 
365 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
366 			     struct rhash_head *obj);
367 
368 void rhashtable_walk_enter(struct rhashtable *ht,
369 			   struct rhashtable_iter *iter);
370 void rhashtable_walk_exit(struct rhashtable_iter *iter);
371 int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU);
372 void *rhashtable_walk_next(struct rhashtable_iter *iter);
373 void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
374 
375 void rhashtable_free_and_destroy(struct rhashtable *ht,
376 				 void (*free_fn)(void *ptr, void *arg),
377 				 void *arg);
378 void rhashtable_destroy(struct rhashtable *ht);
379 
380 struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
381 					    unsigned int hash);
382 struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
383 						   struct bucket_table *tbl,
384 						   unsigned int hash);
385 
386 #define rht_dereference(p, ht) \
387 	rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
388 
389 #define rht_dereference_rcu(p, ht) \
390 	rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
391 
392 #define rht_dereference_bucket(p, tbl, hash) \
393 	rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
394 
395 #define rht_dereference_bucket_rcu(p, tbl, hash) \
396 	rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
397 
398 #define rht_entry(tpos, pos, member) \
399 	({ tpos = container_of(pos, typeof(*tpos), member); 1; })
400 
401 static inline struct rhash_head __rcu *const *rht_bucket(
402 	const struct bucket_table *tbl, unsigned int hash)
403 {
404 	return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
405 				     &tbl->buckets[hash];
406 }
407 
408 static inline struct rhash_head __rcu **rht_bucket_var(
409 	struct bucket_table *tbl, unsigned int hash)
410 {
411 	return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
412 				     &tbl->buckets[hash];
413 }
414 
415 static inline struct rhash_head __rcu **rht_bucket_insert(
416 	struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
417 {
418 	return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
419 				     &tbl->buckets[hash];
420 }
421 
422 /**
423  * rht_for_each_continue - continue iterating over hash chain
424  * @pos:	the &struct rhash_head to use as a loop cursor.
425  * @head:	the previous &struct rhash_head to continue from
426  * @tbl:	the &struct bucket_table
427  * @hash:	the hash value / bucket index
428  */
429 #define rht_for_each_continue(pos, head, tbl, hash) \
430 	for (pos = rht_dereference_bucket(head, tbl, hash); \
431 	     !rht_is_a_nulls(pos); \
432 	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
433 
434 /**
435  * rht_for_each - iterate over hash chain
436  * @pos:	the &struct rhash_head to use as a loop cursor.
437  * @tbl:	the &struct bucket_table
438  * @hash:	the hash value / bucket index
439  */
440 #define rht_for_each(pos, tbl, hash) \
441 	rht_for_each_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
442 
443 /**
444  * rht_for_each_entry_continue - continue iterating over hash chain
445  * @tpos:	the type * to use as a loop cursor.
446  * @pos:	the &struct rhash_head to use as a loop cursor.
447  * @head:	the previous &struct rhash_head to continue from
448  * @tbl:	the &struct bucket_table
449  * @hash:	the hash value / bucket index
450  * @member:	name of the &struct rhash_head within the hashable struct.
451  */
452 #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member)	\
453 	for (pos = rht_dereference_bucket(head, tbl, hash);		\
454 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	\
455 	     pos = rht_dereference_bucket((pos)->next, tbl, hash))
456 
457 /**
458  * rht_for_each_entry - iterate over hash chain of given type
459  * @tpos:	the type * to use as a loop cursor.
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  * @member:	name of the &struct rhash_head within the hashable struct.
464  */
465 #define rht_for_each_entry(tpos, pos, tbl, hash, member)		\
466 	rht_for_each_entry_continue(tpos, pos, *rht_bucket(tbl, hash),	\
467 				    tbl, hash, member)
468 
469 /**
470  * rht_for_each_entry_safe - safely iterate over hash chain of given type
471  * @tpos:	the type * to use as a loop cursor.
472  * @pos:	the &struct rhash_head to use as a loop cursor.
473  * @next:	the &struct rhash_head to use as next in loop cursor.
474  * @tbl:	the &struct bucket_table
475  * @hash:	the hash value / bucket index
476  * @member:	name of the &struct rhash_head within the hashable struct.
477  *
478  * This hash chain list-traversal primitive allows for the looped code to
479  * remove the loop cursor from the list.
480  */
481 #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member)	      \
482 	for (pos = rht_dereference_bucket(*rht_bucket(tbl, hash), tbl, hash), \
483 	     next = !rht_is_a_nulls(pos) ?				      \
484 		       rht_dereference_bucket(pos->next, tbl, hash) : NULL;   \
485 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	      \
486 	     pos = next,						      \
487 	     next = !rht_is_a_nulls(pos) ?				      \
488 		       rht_dereference_bucket(pos->next, tbl, hash) : NULL)
489 
490 /**
491  * rht_for_each_rcu_continue - continue iterating over rcu hash chain
492  * @pos:	the &struct rhash_head to use as a loop cursor.
493  * @head:	the previous &struct rhash_head to continue from
494  * @tbl:	the &struct bucket_table
495  * @hash:	the hash value / bucket index
496  *
497  * This hash chain list-traversal primitive may safely run concurrently with
498  * the _rcu mutation primitives such as rhashtable_insert() as long as the
499  * traversal is guarded by rcu_read_lock().
500  */
501 #define rht_for_each_rcu_continue(pos, head, tbl, hash)			\
502 	for (({barrier(); }),						\
503 	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		\
504 	     !rht_is_a_nulls(pos);					\
505 	     pos = rcu_dereference_raw(pos->next))
506 
507 /**
508  * rht_for_each_rcu - iterate over rcu hash chain
509  * @pos:	the &struct rhash_head to use as a loop cursor.
510  * @tbl:	the &struct bucket_table
511  * @hash:	the hash value / bucket index
512  *
513  * This hash chain list-traversal primitive may safely run concurrently with
514  * the _rcu mutation primitives such as rhashtable_insert() as long as the
515  * traversal is guarded by rcu_read_lock().
516  */
517 #define rht_for_each_rcu(pos, tbl, hash)				\
518 	rht_for_each_rcu_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
519 
520 /**
521  * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
522  * @tpos:	the type * to use as a loop cursor.
523  * @pos:	the &struct rhash_head to use as a loop cursor.
524  * @head:	the previous &struct rhash_head to continue from
525  * @tbl:	the &struct bucket_table
526  * @hash:	the hash value / bucket index
527  * @member:	name of the &struct rhash_head within the hashable struct.
528  *
529  * This hash chain list-traversal primitive may safely run concurrently with
530  * the _rcu mutation primitives such as rhashtable_insert() as long as the
531  * traversal is guarded by rcu_read_lock().
532  */
533 #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
534 	for (({barrier(); }),						    \
535 	     pos = rht_dereference_bucket_rcu(head, tbl, hash);		    \
536 	     (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member);	    \
537 	     pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
538 
539 /**
540  * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
541  * @tpos:	the type * to use as a loop cursor.
542  * @pos:	the &struct rhash_head to use as a loop cursor.
543  * @tbl:	the &struct bucket_table
544  * @hash:	the hash value / bucket index
545  * @member:	name of the &struct rhash_head within the hashable struct.
546  *
547  * This hash chain list-traversal primitive may safely run concurrently with
548  * the _rcu mutation primitives such as rhashtable_insert() as long as the
549  * traversal is guarded by rcu_read_lock().
550  */
551 #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member)		   \
552 	rht_for_each_entry_rcu_continue(tpos, pos, *rht_bucket(tbl, hash), \
553 					tbl, hash, member)
554 
555 /**
556  * rhl_for_each_rcu - iterate over rcu hash table list
557  * @pos:	the &struct rlist_head to use as a loop cursor.
558  * @list:	the head of the list
559  *
560  * This hash chain list-traversal primitive should be used on the
561  * list returned by rhltable_lookup.
562  */
563 #define rhl_for_each_rcu(pos, list)					\
564 	for (pos = list; pos; pos = rcu_dereference_raw(pos->next))
565 
566 /**
567  * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
568  * @tpos:	the type * to use as a loop cursor.
569  * @pos:	the &struct rlist_head to use as a loop cursor.
570  * @list:	the head of the list
571  * @member:	name of the &struct rlist_head within the hashable struct.
572  *
573  * This hash chain list-traversal primitive should be used on the
574  * list returned by rhltable_lookup.
575  */
576 #define rhl_for_each_entry_rcu(tpos, pos, list, member)			\
577 	for (pos = list; pos && rht_entry(tpos, pos, member);		\
578 	     pos = rcu_dereference_raw(pos->next))
579 
580 static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
581 				     const void *obj)
582 {
583 	struct rhashtable *ht = arg->ht;
584 	const char *ptr = obj;
585 
586 	return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
587 }
588 
589 /* Internal function, do not use. */
590 static inline struct rhash_head *__rhashtable_lookup(
591 	struct rhashtable *ht, const void *key,
592 	const struct rhashtable_params params)
593 {
594 	struct rhashtable_compare_arg arg = {
595 		.ht = ht,
596 		.key = key,
597 	};
598 	struct bucket_table *tbl;
599 	struct rhash_head *he;
600 	unsigned int hash;
601 
602 	tbl = rht_dereference_rcu(ht->tbl, ht);
603 restart:
604 	hash = rht_key_hashfn(ht, tbl, key, params);
605 	rht_for_each_rcu(he, tbl, hash) {
606 		if (params.obj_cmpfn ?
607 		    params.obj_cmpfn(&arg, rht_obj(ht, he)) :
608 		    rhashtable_compare(&arg, rht_obj(ht, he)))
609 			continue;
610 		return he;
611 	}
612 
613 	/* Ensure we see any new tables. */
614 	smp_rmb();
615 
616 	tbl = rht_dereference_rcu(tbl->future_tbl, ht);
617 	if (unlikely(tbl))
618 		goto restart;
619 
620 	return NULL;
621 }
622 
623 /**
624  * rhashtable_lookup - search hash table
625  * @ht:		hash table
626  * @key:	the pointer to the key
627  * @params:	hash table parameters
628  *
629  * Computes the hash value for the key and traverses the bucket chain looking
630  * for a entry with an identical key. The first matching entry is returned.
631  *
632  * This must only be called under the RCU read lock.
633  *
634  * Returns the first entry on which the compare function returned true.
635  */
636 static inline void *rhashtable_lookup(
637 	struct rhashtable *ht, const void *key,
638 	const struct rhashtable_params params)
639 {
640 	struct rhash_head *he = __rhashtable_lookup(ht, key, params);
641 
642 	return he ? rht_obj(ht, he) : NULL;
643 }
644 
645 /**
646  * rhashtable_lookup_fast - search hash table, without RCU read lock
647  * @ht:		hash table
648  * @key:	the pointer to the key
649  * @params:	hash table parameters
650  *
651  * Computes the hash value for the key and traverses the bucket chain looking
652  * for a entry with an identical key. The first matching entry is returned.
653  *
654  * Only use this function when you have other mechanisms guaranteeing
655  * that the object won't go away after the RCU read lock is released.
656  *
657  * Returns the first entry on which the compare function returned true.
658  */
659 static inline void *rhashtable_lookup_fast(
660 	struct rhashtable *ht, const void *key,
661 	const struct rhashtable_params params)
662 {
663 	void *obj;
664 
665 	rcu_read_lock();
666 	obj = rhashtable_lookup(ht, key, params);
667 	rcu_read_unlock();
668 
669 	return obj;
670 }
671 
672 /**
673  * rhltable_lookup - search hash list table
674  * @hlt:	hash table
675  * @key:	the pointer to the key
676  * @params:	hash table parameters
677  *
678  * Computes the hash value for the key and traverses the bucket chain looking
679  * for a entry with an identical key.  All matching entries are returned
680  * in a list.
681  *
682  * This must only be called under the RCU read lock.
683  *
684  * Returns the list of entries that match the given key.
685  */
686 static inline struct rhlist_head *rhltable_lookup(
687 	struct rhltable *hlt, const void *key,
688 	const struct rhashtable_params params)
689 {
690 	struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params);
691 
692 	return he ? container_of(he, struct rhlist_head, rhead) : NULL;
693 }
694 
695 /* Internal function, please use rhashtable_insert_fast() instead. This
696  * function returns the existing element already in hashes in there is a clash,
697  * otherwise it returns an error via ERR_PTR().
698  */
699 static inline void *__rhashtable_insert_fast(
700 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
701 	const struct rhashtable_params params, bool rhlist)
702 {
703 	struct rhashtable_compare_arg arg = {
704 		.ht = ht,
705 		.key = key,
706 	};
707 	struct rhash_head __rcu **pprev;
708 	struct bucket_table *tbl;
709 	struct rhash_head *head;
710 	spinlock_t *lock;
711 	unsigned int hash;
712 	int elasticity;
713 	void *data;
714 
715 	rcu_read_lock();
716 
717 	tbl = rht_dereference_rcu(ht->tbl, ht);
718 	hash = rht_head_hashfn(ht, tbl, obj, params);
719 	lock = rht_bucket_lock(tbl, hash);
720 	spin_lock_bh(lock);
721 
722 	if (unlikely(rht_dereference_bucket(tbl->future_tbl, tbl, hash))) {
723 slow_path:
724 		spin_unlock_bh(lock);
725 		rcu_read_unlock();
726 		return rhashtable_insert_slow(ht, key, obj);
727 	}
728 
729 	elasticity = ht->elasticity;
730 	pprev = rht_bucket_insert(ht, tbl, hash);
731 	data = ERR_PTR(-ENOMEM);
732 	if (!pprev)
733 		goto out;
734 
735 	rht_for_each_continue(head, *pprev, tbl, hash) {
736 		struct rhlist_head *plist;
737 		struct rhlist_head *list;
738 
739 		elasticity--;
740 		if (!key ||
741 		    (params.obj_cmpfn ?
742 		     params.obj_cmpfn(&arg, rht_obj(ht, head)) :
743 		     rhashtable_compare(&arg, rht_obj(ht, head))))
744 			continue;
745 
746 		data = rht_obj(ht, head);
747 
748 		if (!rhlist)
749 			goto out;
750 
751 
752 		list = container_of(obj, struct rhlist_head, rhead);
753 		plist = container_of(head, struct rhlist_head, rhead);
754 
755 		RCU_INIT_POINTER(list->next, plist);
756 		head = rht_dereference_bucket(head->next, tbl, hash);
757 		RCU_INIT_POINTER(list->rhead.next, head);
758 		rcu_assign_pointer(*pprev, obj);
759 
760 		goto good;
761 	}
762 
763 	if (elasticity <= 0)
764 		goto slow_path;
765 
766 	data = ERR_PTR(-E2BIG);
767 	if (unlikely(rht_grow_above_max(ht, tbl)))
768 		goto out;
769 
770 	if (unlikely(rht_grow_above_100(ht, tbl)))
771 		goto slow_path;
772 
773 	head = rht_dereference_bucket(*pprev, tbl, hash);
774 
775 	RCU_INIT_POINTER(obj->next, head);
776 	if (rhlist) {
777 		struct rhlist_head *list;
778 
779 		list = container_of(obj, struct rhlist_head, rhead);
780 		RCU_INIT_POINTER(list->next, NULL);
781 	}
782 
783 	rcu_assign_pointer(*pprev, obj);
784 
785 	atomic_inc(&ht->nelems);
786 	if (rht_grow_above_75(ht, tbl))
787 		schedule_work(&ht->run_work);
788 
789 good:
790 	data = NULL;
791 
792 out:
793 	spin_unlock_bh(lock);
794 	rcu_read_unlock();
795 
796 	return data;
797 }
798 
799 /**
800  * rhashtable_insert_fast - insert object into hash table
801  * @ht:		hash table
802  * @obj:	pointer to hash head inside object
803  * @params:	hash table parameters
804  *
805  * Will take a per bucket spinlock to protect against mutual mutations
806  * on the same bucket. Multiple insertions may occur in parallel unless
807  * they map to the same bucket lock.
808  *
809  * It is safe to call this function from atomic context.
810  *
811  * Will trigger an automatic deferred table resizing if the size grows
812  * beyond the watermark indicated by grow_decision() which can be passed
813  * to rhashtable_init().
814  */
815 static inline int rhashtable_insert_fast(
816 	struct rhashtable *ht, struct rhash_head *obj,
817 	const struct rhashtable_params params)
818 {
819 	void *ret;
820 
821 	ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
822 	if (IS_ERR(ret))
823 		return PTR_ERR(ret);
824 
825 	return ret == NULL ? 0 : -EEXIST;
826 }
827 
828 /**
829  * rhltable_insert_key - insert object into hash list table
830  * @hlt:	hash list table
831  * @key:	the pointer to the key
832  * @list:	pointer to hash list head inside object
833  * @params:	hash table parameters
834  *
835  * Will take a per bucket spinlock to protect against mutual mutations
836  * on the same bucket. Multiple insertions may occur in parallel unless
837  * they map to the same bucket lock.
838  *
839  * It is safe to call this function from atomic context.
840  *
841  * Will trigger an automatic deferred table resizing if the size grows
842  * beyond the watermark indicated by grow_decision() which can be passed
843  * to rhashtable_init().
844  */
845 static inline int rhltable_insert_key(
846 	struct rhltable *hlt, const void *key, struct rhlist_head *list,
847 	const struct rhashtable_params params)
848 {
849 	return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
850 						params, true));
851 }
852 
853 /**
854  * rhltable_insert - insert object into hash list table
855  * @hlt:	hash list table
856  * @list:	pointer to hash list head inside object
857  * @params:	hash table parameters
858  *
859  * Will take a per bucket spinlock to protect against mutual mutations
860  * on the same bucket. Multiple insertions may occur in parallel unless
861  * they map to the same bucket lock.
862  *
863  * It is safe to call this function from atomic context.
864  *
865  * Will trigger an automatic deferred table resizing if the size grows
866  * beyond the watermark indicated by grow_decision() which can be passed
867  * to rhashtable_init().
868  */
869 static inline int rhltable_insert(
870 	struct rhltable *hlt, struct rhlist_head *list,
871 	const struct rhashtable_params params)
872 {
873 	const char *key = rht_obj(&hlt->ht, &list->rhead);
874 
875 	key += params.key_offset;
876 
877 	return rhltable_insert_key(hlt, key, list, params);
878 }
879 
880 /**
881  * rhashtable_lookup_insert_fast - lookup and insert object into hash table
882  * @ht:		hash table
883  * @obj:	pointer to hash head inside object
884  * @params:	hash table parameters
885  *
886  * Locks down the bucket chain in both the old and new table if a resize
887  * is in progress to ensure that writers can't remove from the old table
888  * and can't insert to the new table during the atomic operation of search
889  * and insertion. Searches for duplicates in both the old and new table if
890  * a resize is in progress.
891  *
892  * This lookup function may only be used for fixed key hash table (key_len
893  * parameter set). It will BUG() if used inappropriately.
894  *
895  * It is safe to call this function from atomic context.
896  *
897  * Will trigger an automatic deferred table resizing if the size grows
898  * beyond the watermark indicated by grow_decision() which can be passed
899  * to rhashtable_init().
900  */
901 static inline int rhashtable_lookup_insert_fast(
902 	struct rhashtable *ht, struct rhash_head *obj,
903 	const struct rhashtable_params params)
904 {
905 	const char *key = rht_obj(ht, obj);
906 	void *ret;
907 
908 	BUG_ON(ht->p.obj_hashfn);
909 
910 	ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
911 				       false);
912 	if (IS_ERR(ret))
913 		return PTR_ERR(ret);
914 
915 	return ret == NULL ? 0 : -EEXIST;
916 }
917 
918 /**
919  * rhashtable_lookup_insert_key - search and insert object to hash table
920  *				  with explicit key
921  * @ht:		hash table
922  * @key:	key
923  * @obj:	pointer to hash head inside object
924  * @params:	hash table parameters
925  *
926  * Locks down the bucket chain in both the old and new table if a resize
927  * is in progress to ensure that writers can't remove from the old table
928  * and can't insert to the new table during the atomic operation of search
929  * and insertion. Searches for duplicates in both the old and new table if
930  * a resize is in progress.
931  *
932  * Lookups may occur in parallel with hashtable mutations and resizing.
933  *
934  * Will trigger an automatic deferred table resizing if the size grows
935  * beyond the watermark indicated by grow_decision() which can be passed
936  * to rhashtable_init().
937  *
938  * Returns zero on success.
939  */
940 static inline int rhashtable_lookup_insert_key(
941 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
942 	const struct rhashtable_params params)
943 {
944 	void *ret;
945 
946 	BUG_ON(!ht->p.obj_hashfn || !key);
947 
948 	ret = __rhashtable_insert_fast(ht, key, obj, params, false);
949 	if (IS_ERR(ret))
950 		return PTR_ERR(ret);
951 
952 	return ret == NULL ? 0 : -EEXIST;
953 }
954 
955 /**
956  * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
957  * @ht:		hash table
958  * @obj:	pointer to hash head inside object
959  * @params:	hash table parameters
960  * @data:	pointer to element data already in hashes
961  *
962  * Just like rhashtable_lookup_insert_key(), but this function returns the
963  * object if it exists, NULL if it does not and the insertion was successful,
964  * and an ERR_PTR otherwise.
965  */
966 static inline void *rhashtable_lookup_get_insert_key(
967 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
968 	const struct rhashtable_params params)
969 {
970 	BUG_ON(!ht->p.obj_hashfn || !key);
971 
972 	return __rhashtable_insert_fast(ht, key, obj, params, false);
973 }
974 
975 /* Internal function, please use rhashtable_remove_fast() instead */
976 static inline int __rhashtable_remove_fast_one(
977 	struct rhashtable *ht, struct bucket_table *tbl,
978 	struct rhash_head *obj, const struct rhashtable_params params,
979 	bool rhlist)
980 {
981 	struct rhash_head __rcu **pprev;
982 	struct rhash_head *he;
983 	spinlock_t * lock;
984 	unsigned int hash;
985 	int err = -ENOENT;
986 
987 	hash = rht_head_hashfn(ht, tbl, obj, params);
988 	lock = rht_bucket_lock(tbl, hash);
989 
990 	spin_lock_bh(lock);
991 
992 	pprev = rht_bucket_var(tbl, hash);
993 	rht_for_each_continue(he, *pprev, tbl, hash) {
994 		struct rhlist_head *list;
995 
996 		list = container_of(he, struct rhlist_head, rhead);
997 
998 		if (he != obj) {
999 			struct rhlist_head __rcu **lpprev;
1000 
1001 			pprev = &he->next;
1002 
1003 			if (!rhlist)
1004 				continue;
1005 
1006 			do {
1007 				lpprev = &list->next;
1008 				list = rht_dereference_bucket(list->next,
1009 							      tbl, hash);
1010 			} while (list && obj != &list->rhead);
1011 
1012 			if (!list)
1013 				continue;
1014 
1015 			list = rht_dereference_bucket(list->next, tbl, hash);
1016 			RCU_INIT_POINTER(*lpprev, list);
1017 			err = 0;
1018 			break;
1019 		}
1020 
1021 		obj = rht_dereference_bucket(obj->next, tbl, hash);
1022 		err = 1;
1023 
1024 		if (rhlist) {
1025 			list = rht_dereference_bucket(list->next, tbl, hash);
1026 			if (list) {
1027 				RCU_INIT_POINTER(list->rhead.next, obj);
1028 				obj = &list->rhead;
1029 				err = 0;
1030 			}
1031 		}
1032 
1033 		rcu_assign_pointer(*pprev, obj);
1034 		break;
1035 	}
1036 
1037 	spin_unlock_bh(lock);
1038 
1039 	if (err > 0) {
1040 		atomic_dec(&ht->nelems);
1041 		if (unlikely(ht->p.automatic_shrinking &&
1042 			     rht_shrink_below_30(ht, tbl)))
1043 			schedule_work(&ht->run_work);
1044 		err = 0;
1045 	}
1046 
1047 	return err;
1048 }
1049 
1050 /* Internal function, please use rhashtable_remove_fast() instead */
1051 static inline int __rhashtable_remove_fast(
1052 	struct rhashtable *ht, struct rhash_head *obj,
1053 	const struct rhashtable_params params, bool rhlist)
1054 {
1055 	struct bucket_table *tbl;
1056 	int err;
1057 
1058 	rcu_read_lock();
1059 
1060 	tbl = rht_dereference_rcu(ht->tbl, ht);
1061 
1062 	/* Because we have already taken (and released) the bucket
1063 	 * lock in old_tbl, if we find that future_tbl is not yet
1064 	 * visible then that guarantees the entry to still be in
1065 	 * the old tbl if it exists.
1066 	 */
1067 	while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
1068 						   rhlist)) &&
1069 	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1070 		;
1071 
1072 	rcu_read_unlock();
1073 
1074 	return err;
1075 }
1076 
1077 /**
1078  * rhashtable_remove_fast - remove object from hash table
1079  * @ht:		hash table
1080  * @obj:	pointer to hash head inside object
1081  * @params:	hash table parameters
1082  *
1083  * Since the hash chain is single linked, the removal operation needs to
1084  * walk the bucket chain upon removal. The removal operation is thus
1085  * considerable slow if the hash table is not correctly sized.
1086  *
1087  * Will automatically shrink the table via rhashtable_expand() if the
1088  * shrink_decision function specified at rhashtable_init() returns true.
1089  *
1090  * Returns zero on success, -ENOENT if the entry could not be found.
1091  */
1092 static inline int rhashtable_remove_fast(
1093 	struct rhashtable *ht, struct rhash_head *obj,
1094 	const struct rhashtable_params params)
1095 {
1096 	return __rhashtable_remove_fast(ht, obj, params, false);
1097 }
1098 
1099 /**
1100  * rhltable_remove - remove object from hash list table
1101  * @hlt:	hash list table
1102  * @list:	pointer to hash list head inside object
1103  * @params:	hash table parameters
1104  *
1105  * Since the hash chain is single linked, the removal operation needs to
1106  * walk the bucket chain upon removal. The removal operation is thus
1107  * considerable slow if the hash table is not correctly sized.
1108  *
1109  * Will automatically shrink the table via rhashtable_expand() if the
1110  * shrink_decision function specified at rhashtable_init() returns true.
1111  *
1112  * Returns zero on success, -ENOENT if the entry could not be found.
1113  */
1114 static inline int rhltable_remove(
1115 	struct rhltable *hlt, struct rhlist_head *list,
1116 	const struct rhashtable_params params)
1117 {
1118 	return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
1119 }
1120 
1121 /* Internal function, please use rhashtable_replace_fast() instead */
1122 static inline int __rhashtable_replace_fast(
1123 	struct rhashtable *ht, struct bucket_table *tbl,
1124 	struct rhash_head *obj_old, struct rhash_head *obj_new,
1125 	const struct rhashtable_params params)
1126 {
1127 	struct rhash_head __rcu **pprev;
1128 	struct rhash_head *he;
1129 	spinlock_t *lock;
1130 	unsigned int hash;
1131 	int err = -ENOENT;
1132 
1133 	/* Minimally, the old and new objects must have same hash
1134 	 * (which should mean identifiers are the same).
1135 	 */
1136 	hash = rht_head_hashfn(ht, tbl, obj_old, params);
1137 	if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1138 		return -EINVAL;
1139 
1140 	lock = rht_bucket_lock(tbl, hash);
1141 
1142 	spin_lock_bh(lock);
1143 
1144 	pprev = rht_bucket_var(tbl, hash);
1145 	rht_for_each_continue(he, *pprev, tbl, hash) {
1146 		if (he != obj_old) {
1147 			pprev = &he->next;
1148 			continue;
1149 		}
1150 
1151 		rcu_assign_pointer(obj_new->next, obj_old->next);
1152 		rcu_assign_pointer(*pprev, obj_new);
1153 		err = 0;
1154 		break;
1155 	}
1156 
1157 	spin_unlock_bh(lock);
1158 
1159 	return err;
1160 }
1161 
1162 /**
1163  * rhashtable_replace_fast - replace an object in hash table
1164  * @ht:		hash table
1165  * @obj_old:	pointer to hash head inside object being replaced
1166  * @obj_new:	pointer to hash head inside object which is new
1167  * @params:	hash table parameters
1168  *
1169  * Replacing an object doesn't affect the number of elements in the hash table
1170  * or bucket, so we don't need to worry about shrinking or expanding the
1171  * table here.
1172  *
1173  * Returns zero on success, -ENOENT if the entry could not be found,
1174  * -EINVAL if hash is not the same for the old and new objects.
1175  */
1176 static inline int rhashtable_replace_fast(
1177 	struct rhashtable *ht, struct rhash_head *obj_old,
1178 	struct rhash_head *obj_new,
1179 	const struct rhashtable_params params)
1180 {
1181 	struct bucket_table *tbl;
1182 	int err;
1183 
1184 	rcu_read_lock();
1185 
1186 	tbl = rht_dereference_rcu(ht->tbl, ht);
1187 
1188 	/* Because we have already taken (and released) the bucket
1189 	 * lock in old_tbl, if we find that future_tbl is not yet
1190 	 * visible then that guarantees the entry to still be in
1191 	 * the old tbl if it exists.
1192 	 */
1193 	while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1194 						obj_new, params)) &&
1195 	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1196 		;
1197 
1198 	rcu_read_unlock();
1199 
1200 	return err;
1201 }
1202 
1203 /* Obsolete function, do not use in new code. */
1204 static inline int rhashtable_walk_init(struct rhashtable *ht,
1205 				       struct rhashtable_iter *iter, gfp_t gfp)
1206 {
1207 	rhashtable_walk_enter(ht, iter);
1208 	return 0;
1209 }
1210 
1211 /**
1212  * rhltable_walk_enter - Initialise an iterator
1213  * @hlt:	Table to walk over
1214  * @iter:	Hash table Iterator
1215  *
1216  * This function prepares a hash table walk.
1217  *
1218  * Note that if you restart a walk after rhashtable_walk_stop you
1219  * may see the same object twice.  Also, you may miss objects if
1220  * there are removals in between rhashtable_walk_stop and the next
1221  * call to rhashtable_walk_start.
1222  *
1223  * For a completely stable walk you should construct your own data
1224  * structure outside the hash table.
1225  *
1226  * This function may sleep so you must not call it from interrupt
1227  * context or with spin locks held.
1228  *
1229  * You must call rhashtable_walk_exit after this function returns.
1230  */
1231 static inline void rhltable_walk_enter(struct rhltable *hlt,
1232 				       struct rhashtable_iter *iter)
1233 {
1234 	return rhashtable_walk_enter(&hlt->ht, iter);
1235 }
1236 
1237 /**
1238  * rhltable_free_and_destroy - free elements and destroy hash list table
1239  * @hlt:	the hash list table to destroy
1240  * @free_fn:	callback to release resources of element
1241  * @arg:	pointer passed to free_fn
1242  *
1243  * See documentation for rhashtable_free_and_destroy.
1244  */
1245 static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1246 					     void (*free_fn)(void *ptr,
1247 							     void *arg),
1248 					     void *arg)
1249 {
1250 	return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1251 }
1252 
1253 static inline void rhltable_destroy(struct rhltable *hlt)
1254 {
1255 	return rhltable_free_and_destroy(hlt, NULL, NULL);
1256 }
1257 
1258 #endif /* _LINUX_RHASHTABLE_H */
1259