blob: e8ffcdb5e23965f0b6ca2faa84e1d548909f47cc [file] [log] [blame]
Thomas Graf7e1e7762014-08-02 11:47:44 +02001/*
2 * Resizable, Scalable, Concurrent Hash Table
3 *
Herbert Xudc0ee262015-03-20 21:57:06 +11004 * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au>
Thomas Graf7e1e7762014-08-02 11:47:44 +02005 * Copyright (c) 2014 Thomas Graf <tgraf@suug.ch>
6 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
7 *
Thomas Graf7e1e7762014-08-02 11:47:44 +02008 * Code partially derived from nft_hash
Herbert Xudc0ee262015-03-20 21:57:06 +11009 * Rewritten with rehash code from br_multicast plus single list
10 * pointer as suggested by Josh Triplett
Thomas Graf7e1e7762014-08-02 11:47:44 +020011 *
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
Herbert Xuf2dba9c2015-02-04 07:33:23 +110020#include <linux/compiler.h>
Herbert Xu6626af62015-03-20 18:18:45 -040021#include <linux/errno.h>
Herbert Xu31ccde22015-03-24 00:50:21 +110022#include <linux/jhash.h>
Thomas Graff89bd6f2015-01-02 23:00:21 +010023#include <linux/list_nulls.h>
Thomas Graf97defe12015-01-02 23:00:20 +010024#include <linux/workqueue.h>
Ying Xue86b35b62015-01-04 15:25:09 +080025#include <linux/mutex.h>
Herbert Xu02fd97c2015-03-20 21:57:00 +110026#include <linux/rcupdate.h>
Thomas Graf7e1e7762014-08-02 11:47:44 +020027
Thomas Graff89bd6f2015-01-02 23:00:21 +010028/*
29 * The end of the chain is marked with a special nulls marks which has
30 * the following format:
31 *
32 * +-------+-----------------------------------------------------+-+
33 * | Base | Hash |1|
34 * +-------+-----------------------------------------------------+-+
35 *
36 * Base (4 bits) : Reserved to distinguish between multiple tables.
37 * Specified via &struct rhashtable_params.nulls_base.
38 * Hash (27 bits): Full hash (unmasked) of first element added to bucket
39 * 1 (1 bit) : Nulls marker (always set)
40 *
41 * The remaining bits of the next pointer remain unused for now.
42 */
43#define RHT_BASE_BITS 4
44#define RHT_HASH_BITS 27
45#define RHT_BASE_SHIFT RHT_HASH_BITS
46
Herbert Xu02fd97c2015-03-20 21:57:00 +110047/* Base bits plus 1 bit for nulls marker */
48#define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1)
49
Thomas Graf7e1e7762014-08-02 11:47:44 +020050struct rhash_head {
Thomas Graf5300fdc2014-08-13 16:38:29 +020051 struct rhash_head __rcu *next;
Thomas Graf7e1e7762014-08-02 11:47:44 +020052};
53
Thomas Graf97defe12015-01-02 23:00:20 +010054/**
55 * struct bucket_table - Table of hash buckets
56 * @size: Number of hash buckets
Herbert Xu63d512d2015-03-14 13:57:24 +110057 * @rehash: Current bucket being rehashed
Herbert Xu988dfbd2015-03-10 09:27:55 +110058 * @hash_rnd: Random seed to fold into hash
Thomas Graf97defe12015-01-02 23:00:20 +010059 * @locks_mask: Mask to apply before accessing locks[]
60 * @locks: Array of spinlocks protecting individual buckets
Herbert Xueddee5ba2015-03-14 13:57:20 +110061 * @walkers: List of active walkers
Herbert Xu9d901bc2015-03-14 13:57:23 +110062 * @rcu: RCU structure for freeing the table
Herbert Xuc4db8842015-03-14 13:57:25 +110063 * @future_tbl: Table under construction during rehashing
Thomas Graf97defe12015-01-02 23:00:20 +010064 * @buckets: size * hash buckets
65 */
Thomas Graf7e1e7762014-08-02 11:47:44 +020066struct bucket_table {
Herbert Xu63d512d2015-03-14 13:57:24 +110067 unsigned int size;
68 unsigned int rehash;
Herbert Xu988dfbd2015-03-10 09:27:55 +110069 u32 hash_rnd;
Eric Dumazetb9ebafb2015-02-20 06:48:57 -080070 unsigned int locks_mask;
71 spinlock_t *locks;
Herbert Xueddee5ba2015-03-14 13:57:20 +110072 struct list_head walkers;
Herbert Xu9d901bc2015-03-14 13:57:23 +110073 struct rcu_head rcu;
Eric Dumazetb9ebafb2015-02-20 06:48:57 -080074
Herbert Xuc4db8842015-03-14 13:57:25 +110075 struct bucket_table __rcu *future_tbl;
76
Eric Dumazetb9ebafb2015-02-20 06:48:57 -080077 struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp;
Thomas Graf7e1e7762014-08-02 11:47:44 +020078};
79
Herbert Xu02fd97c2015-03-20 21:57:00 +110080/**
81 * struct rhashtable_compare_arg - Key for the function rhashtable_compare
82 * @ht: Hash table
83 * @key: Key to compare against
84 */
85struct rhashtable_compare_arg {
86 struct rhashtable *ht;
87 const void *key;
88};
89
Thomas Graf7e1e7762014-08-02 11:47:44 +020090typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
91typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 seed);
Herbert Xu02fd97c2015-03-20 21:57:00 +110092typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
93 const void *obj);
Thomas Graf7e1e7762014-08-02 11:47:44 +020094
95struct rhashtable;
96
97/**
98 * struct rhashtable_params - Hash table construction parameters
99 * @nelem_hint: Hint on number of elements, should be 75% of desired size
100 * @key_len: Length of key
101 * @key_offset: Offset of key in struct to be hashed
102 * @head_offset: Offset of rhash_head in struct to be hashed
Herbert Xuc2e213c2015-03-18 20:01:16 +1100103 * @max_size: Maximum size while expanding
104 * @min_size: Minimum size while shrinking
Thomas Graff89bd6f2015-01-02 23:00:21 +0100105 * @nulls_base: Base value to generate nulls marker
Thomas Graf97defe12015-01-02 23:00:20 +0100106 * @locks_mul: Number of bucket locks to allocate per cpu (default: 128)
Herbert Xu31ccde22015-03-24 00:50:21 +1100107 * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200108 * @obj_hashfn: Function to hash object
Herbert Xu02fd97c2015-03-20 21:57:00 +1100109 * @obj_cmpfn: Function to compare key with object
Thomas Graf7e1e7762014-08-02 11:47:44 +0200110 */
111struct rhashtable_params {
112 size_t nelem_hint;
113 size_t key_len;
114 size_t key_offset;
115 size_t head_offset;
Herbert Xuc2e213c2015-03-18 20:01:16 +1100116 unsigned int max_size;
117 unsigned int min_size;
Thomas Graff89bd6f2015-01-02 23:00:21 +0100118 u32 nulls_base;
Thomas Graf97defe12015-01-02 23:00:20 +0100119 size_t locks_mul;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200120 rht_hashfn_t hashfn;
121 rht_obj_hashfn_t obj_hashfn;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100122 rht_obj_cmpfn_t obj_cmpfn;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200123};
124
125/**
126 * struct rhashtable - Hash table handle
127 * @tbl: Bucket table
128 * @nelems: Number of elements in table
Herbert Xu31ccde22015-03-24 00:50:21 +1100129 * @key_len: Key length for hashfn
Thomas Graf7e1e7762014-08-02 11:47:44 +0200130 * @p: Configuration parameters
Thomas Graf97defe12015-01-02 23:00:20 +0100131 * @run_work: Deferred worker to expand/shrink asynchronously
132 * @mutex: Mutex to protect current/future table swapping
133 * @being_destroyed: True if table is set up for destruction
Thomas Graf7e1e7762014-08-02 11:47:44 +0200134 */
135struct rhashtable {
136 struct bucket_table __rcu *tbl;
Thomas Graf97defe12015-01-02 23:00:20 +0100137 atomic_t nelems;
Daniel Borkmanna5b68462015-03-12 15:28:40 +0100138 bool being_destroyed;
Herbert Xu31ccde22015-03-24 00:50:21 +1100139 unsigned int key_len;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200140 struct rhashtable_params p;
Ying Xue57699a42015-01-16 11:13:09 +0800141 struct work_struct run_work;
Thomas Graf97defe12015-01-02 23:00:20 +0100142 struct mutex mutex;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200143};
144
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100145/**
146 * struct rhashtable_walker - Hash table walker
147 * @list: List entry on list of walkers
Herbert Xueddee5ba2015-03-14 13:57:20 +1100148 * @tbl: The table that we were walking over
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100149 */
150struct rhashtable_walker {
151 struct list_head list;
Herbert Xueddee5ba2015-03-14 13:57:20 +1100152 struct bucket_table *tbl;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100153};
154
155/**
156 * struct rhashtable_iter - Hash table iterator, fits into netlink cb
157 * @ht: Table to iterate through
158 * @p: Current pointer
159 * @walker: Associated rhashtable walker
160 * @slot: Current slot
161 * @skip: Number of entries to skip in slot
162 */
163struct rhashtable_iter {
164 struct rhashtable *ht;
165 struct rhash_head *p;
166 struct rhashtable_walker *walker;
167 unsigned int slot;
168 unsigned int skip;
169};
170
Thomas Graff89bd6f2015-01-02 23:00:21 +0100171static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
172{
173 return NULLS_MARKER(ht->p.nulls_base + hash);
174}
175
176#define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
177 ((ptr) = (typeof(ptr)) rht_marker(ht, hash))
178
179static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
180{
181 return ((unsigned long) ptr & 1);
182}
183
184static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
185{
186 return ((unsigned long) ptr) >> 1;
187}
188
Herbert Xu02fd97c2015-03-20 21:57:00 +1100189static inline void *rht_obj(const struct rhashtable *ht,
190 const struct rhash_head *he)
191{
192 return (char *)he - ht->p.head_offset;
193}
194
195static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
196 unsigned int hash)
197{
198 return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
199}
200
201static inline unsigned int rht_key_hashfn(
202 struct rhashtable *ht, const struct bucket_table *tbl,
203 const void *key, const struct rhashtable_params params)
204{
Herbert Xu31ccde22015-03-24 00:50:21 +1100205 unsigned hash;
Herbert Xude91b252015-03-24 00:50:20 +1100206
Herbert Xu31ccde22015-03-24 00:50:21 +1100207 /* params must be equal to ht->p if it isn't constant. */
208 if (!__builtin_constant_p(params.key_len))
209 hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd);
210 else if (params.key_len) {
211 unsigned key_len = params.key_len;
212
213 if (params.hashfn)
214 hash = params.hashfn(key, key_len, tbl->hash_rnd);
215 else if (key_len & (sizeof(u32) - 1))
216 hash = jhash(key, key_len, tbl->hash_rnd);
217 else
218 hash = jhash2(key, key_len / sizeof(u32),
219 tbl->hash_rnd);
220 } else {
221 unsigned key_len = ht->p.key_len;
222
223 if (params.hashfn)
224 hash = params.hashfn(key, key_len, tbl->hash_rnd);
225 else
226 hash = jhash(key, key_len, tbl->hash_rnd);
227 }
228
229 return rht_bucket_index(tbl, hash);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100230}
231
232static inline unsigned int rht_head_hashfn(
233 struct rhashtable *ht, const struct bucket_table *tbl,
234 const struct rhash_head *he, const struct rhashtable_params params)
235{
236 const char *ptr = rht_obj(ht, he);
237
238 return likely(params.obj_hashfn) ?
239 rht_bucket_index(tbl, params.obj_hashfn(ptr, tbl->hash_rnd)) :
240 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
241}
242
243/**
244 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
245 * @ht: hash table
246 * @tbl: current table
247 */
248static inline bool rht_grow_above_75(const struct rhashtable *ht,
249 const struct bucket_table *tbl)
250{
251 /* Expand table when exceeding 75% load */
252 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
253 (!ht->p.max_size || tbl->size < ht->p.max_size);
254}
255
256/**
257 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
258 * @ht: hash table
259 * @tbl: current table
260 */
261static inline bool rht_shrink_below_30(const struct rhashtable *ht,
262 const struct bucket_table *tbl)
263{
264 /* Shrink table beneath 30% load */
265 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
266 tbl->size > ht->p.min_size;
267}
268
269/* The bucket lock is selected based on the hash and protects mutations
270 * on a group of hash buckets.
271 *
272 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
273 * a single lock always covers both buckets which may both contains
274 * entries which link to the same bucket of the old table during resizing.
275 * This allows to simplify the locking as locking the bucket in both
276 * tables during resize always guarantee protection.
277 *
278 * IMPORTANT: When holding the bucket lock of both the old and new table
279 * during expansions and shrinking, the old bucket lock must always be
280 * acquired first.
281 */
282static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
283 unsigned int hash)
284{
285 return &tbl->locks[hash & tbl->locks_mask];
286}
287
Thomas Graf7e1e7762014-08-02 11:47:44 +0200288#ifdef CONFIG_PROVE_LOCKING
Thomas Graf97defe12015-01-02 23:00:20 +0100289int lockdep_rht_mutex_is_held(struct rhashtable *ht);
Thomas Graf88d6ed12015-01-02 23:00:16 +0100290int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200291#else
Thomas Graf97defe12015-01-02 23:00:20 +0100292static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200293{
294 return 1;
295}
Thomas Graf88d6ed12015-01-02 23:00:16 +0100296
297static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
298 u32 hash)
299{
300 return 1;
301}
Thomas Graf7e1e7762014-08-02 11:47:44 +0200302#endif /* CONFIG_PROVE_LOCKING */
303
Herbert Xu488fb86e2015-03-20 21:56:59 +1100304int rhashtable_init(struct rhashtable *ht,
305 const struct rhashtable_params *params);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200306
Herbert Xu02fd97c2015-03-20 21:57:00 +1100307int rhashtable_insert_slow(struct rhashtable *ht, const void *key,
308 struct rhash_head *obj,
309 struct bucket_table *old_tbl);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200310
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100311int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter);
312void rhashtable_walk_exit(struct rhashtable_iter *iter);
313int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU);
314void *rhashtable_walk_next(struct rhashtable_iter *iter);
315void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
316
Thomas Graf97defe12015-01-02 23:00:20 +0100317void rhashtable_destroy(struct rhashtable *ht);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200318
319#define rht_dereference(p, ht) \
320 rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
321
322#define rht_dereference_rcu(p, ht) \
323 rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
324
Thomas Graf88d6ed12015-01-02 23:00:16 +0100325#define rht_dereference_bucket(p, tbl, hash) \
326 rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200327
Thomas Graf88d6ed12015-01-02 23:00:16 +0100328#define rht_dereference_bucket_rcu(p, tbl, hash) \
329 rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
330
331#define rht_entry(tpos, pos, member) \
332 ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
333
334/**
335 * rht_for_each_continue - continue iterating over hash chain
336 * @pos: the &struct rhash_head to use as a loop cursor.
337 * @head: the previous &struct rhash_head to continue from
338 * @tbl: the &struct bucket_table
339 * @hash: the hash value / bucket index
340 */
341#define rht_for_each_continue(pos, head, tbl, hash) \
342 for (pos = rht_dereference_bucket(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100343 !rht_is_a_nulls(pos); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100344 pos = rht_dereference_bucket((pos)->next, tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200345
346/**
347 * rht_for_each - iterate over hash chain
Thomas Graf88d6ed12015-01-02 23:00:16 +0100348 * @pos: the &struct rhash_head to use as a loop cursor.
349 * @tbl: the &struct bucket_table
350 * @hash: the hash value / bucket index
Thomas Graf7e1e7762014-08-02 11:47:44 +0200351 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100352#define rht_for_each(pos, tbl, hash) \
353 rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash)
354
355/**
356 * rht_for_each_entry_continue - continue iterating over hash chain
357 * @tpos: the type * to use as a loop cursor.
358 * @pos: the &struct rhash_head to use as a loop cursor.
359 * @head: the previous &struct rhash_head to continue from
360 * @tbl: the &struct bucket_table
361 * @hash: the hash value / bucket index
362 * @member: name of the &struct rhash_head within the hashable struct.
363 */
364#define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \
365 for (pos = rht_dereference_bucket(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100366 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100367 pos = rht_dereference_bucket((pos)->next, tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200368
369/**
370 * rht_for_each_entry - iterate over hash chain of given type
Thomas Graf88d6ed12015-01-02 23:00:16 +0100371 * @tpos: the type * to use as a loop cursor.
372 * @pos: the &struct rhash_head to use as a loop cursor.
373 * @tbl: the &struct bucket_table
374 * @hash: the hash value / bucket index
375 * @member: name of the &struct rhash_head within the hashable struct.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200376 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100377#define rht_for_each_entry(tpos, pos, tbl, hash, member) \
378 rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \
379 tbl, hash, member)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200380
381/**
382 * rht_for_each_entry_safe - safely iterate over hash chain of given type
Thomas Graf88d6ed12015-01-02 23:00:16 +0100383 * @tpos: the type * to use as a loop cursor.
384 * @pos: the &struct rhash_head to use as a loop cursor.
385 * @next: the &struct rhash_head to use as next in loop cursor.
386 * @tbl: the &struct bucket_table
387 * @hash: the hash value / bucket index
388 * @member: name of the &struct rhash_head within the hashable struct.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200389 *
390 * This hash chain list-traversal primitive allows for the looped code to
391 * remove the loop cursor from the list.
392 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100393#define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \
394 for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100395 next = !rht_is_a_nulls(pos) ? \
396 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
397 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
Patrick McHardy607954b2015-01-21 11:12:13 +0000398 pos = next, \
399 next = !rht_is_a_nulls(pos) ? \
400 rht_dereference_bucket(pos->next, tbl, hash) : NULL)
Thomas Graf88d6ed12015-01-02 23:00:16 +0100401
402/**
403 * rht_for_each_rcu_continue - continue iterating over rcu hash chain
404 * @pos: the &struct rhash_head to use as a loop cursor.
405 * @head: the previous &struct rhash_head to continue from
406 * @tbl: the &struct bucket_table
407 * @hash: the hash value / bucket index
408 *
409 * This hash chain list-traversal primitive may safely run concurrently with
410 * the _rcu mutation primitives such as rhashtable_insert() as long as the
411 * traversal is guarded by rcu_read_lock().
412 */
413#define rht_for_each_rcu_continue(pos, head, tbl, hash) \
414 for (({barrier(); }), \
415 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100416 !rht_is_a_nulls(pos); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100417 pos = rcu_dereference_raw(pos->next))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200418
419/**
420 * rht_for_each_rcu - iterate over rcu hash chain
Thomas Graf88d6ed12015-01-02 23:00:16 +0100421 * @pos: the &struct rhash_head to use as a loop cursor.
422 * @tbl: the &struct bucket_table
423 * @hash: the hash value / bucket index
Thomas Graf7e1e7762014-08-02 11:47:44 +0200424 *
425 * This hash chain list-traversal primitive may safely run concurrently with
Thomas Graf88d6ed12015-01-02 23:00:16 +0100426 * the _rcu mutation primitives such as rhashtable_insert() as long as the
Thomas Graf7e1e7762014-08-02 11:47:44 +0200427 * traversal is guarded by rcu_read_lock().
428 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100429#define rht_for_each_rcu(pos, tbl, hash) \
430 rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash)
431
432/**
433 * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
434 * @tpos: the type * to use as a loop cursor.
435 * @pos: the &struct rhash_head to use as a loop cursor.
436 * @head: the previous &struct rhash_head to continue from
437 * @tbl: the &struct bucket_table
438 * @hash: the hash value / bucket index
439 * @member: name of the &struct rhash_head within the hashable struct.
440 *
441 * This hash chain list-traversal primitive may safely run concurrently with
442 * the _rcu mutation primitives such as rhashtable_insert() as long as the
443 * traversal is guarded by rcu_read_lock().
444 */
445#define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
446 for (({barrier(); }), \
447 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100448 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100449 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200450
451/**
452 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
Thomas Graf88d6ed12015-01-02 23:00:16 +0100453 * @tpos: the type * to use as a loop cursor.
454 * @pos: the &struct rhash_head to use as a loop cursor.
455 * @tbl: the &struct bucket_table
456 * @hash: the hash value / bucket index
457 * @member: name of the &struct rhash_head within the hashable struct.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200458 *
459 * This hash chain list-traversal primitive may safely run concurrently with
Thomas Graf88d6ed12015-01-02 23:00:16 +0100460 * the _rcu mutation primitives such as rhashtable_insert() as long as the
Thomas Graf7e1e7762014-08-02 11:47:44 +0200461 * traversal is guarded by rcu_read_lock().
462 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100463#define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \
464 rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\
465 tbl, hash, member)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200466
Herbert Xu02fd97c2015-03-20 21:57:00 +1100467static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
468 const void *obj)
469{
470 struct rhashtable *ht = arg->ht;
471 const char *ptr = obj;
472
473 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
474}
475
476/**
477 * rhashtable_lookup_fast - search hash table, inlined version
478 * @ht: hash table
479 * @key: the pointer to the key
480 * @params: hash table parameters
481 *
482 * Computes the hash value for the key and traverses the bucket chain looking
483 * for a entry with an identical key. The first matching entry is returned.
484 *
485 * Returns the first entry on which the compare function returned true.
486 */
487static inline void *rhashtable_lookup_fast(
488 struct rhashtable *ht, const void *key,
489 const struct rhashtable_params params)
490{
491 struct rhashtable_compare_arg arg = {
492 .ht = ht,
493 .key = key,
494 };
495 const struct bucket_table *tbl;
496 struct rhash_head *he;
497 unsigned hash;
498
499 rcu_read_lock();
500
501 tbl = rht_dereference_rcu(ht->tbl, ht);
502restart:
503 hash = rht_key_hashfn(ht, tbl, key, params);
504 rht_for_each_rcu(he, tbl, hash) {
505 if (params.obj_cmpfn ?
506 params.obj_cmpfn(&arg, rht_obj(ht, he)) :
507 rhashtable_compare(&arg, rht_obj(ht, he)))
508 continue;
509 rcu_read_unlock();
510 return rht_obj(ht, he);
511 }
512
513 /* Ensure we see any new tables. */
514 smp_rmb();
515
516 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
517 if (unlikely(tbl))
518 goto restart;
519 rcu_read_unlock();
520
521 return NULL;
522}
523
524static inline int __rhashtable_insert_fast(
525 struct rhashtable *ht, const void *key, struct rhash_head *obj,
526 const struct rhashtable_params params)
527{
528 struct rhashtable_compare_arg arg = {
529 .ht = ht,
530 .key = key,
531 };
532 int err = -EEXIST;
533 struct bucket_table *tbl, *new_tbl;
534 struct rhash_head *head;
535 spinlock_t *lock;
536 unsigned hash;
537
538 rcu_read_lock();
539
540 tbl = rht_dereference_rcu(ht->tbl, ht);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100541
Herbert Xub8244782015-03-24 00:50:26 +1100542 /* All insertions must grab the oldest table containing
543 * the hashed bucket that is yet to be rehashed.
Herbert Xu02fd97c2015-03-20 21:57:00 +1100544 */
Herbert Xub8244782015-03-24 00:50:26 +1100545 for (;;) {
546 hash = rht_head_hashfn(ht, tbl, obj, params);
547 lock = rht_bucket_lock(tbl, hash);
548 spin_lock_bh(lock);
549
550 if (tbl->rehash <= hash)
551 break;
552
553 spin_unlock_bh(lock);
554 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
555 }
556
Herbert Xu02fd97c2015-03-20 21:57:00 +1100557 new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
558 if (unlikely(new_tbl)) {
559 err = rhashtable_insert_slow(ht, key, obj, new_tbl);
560 goto out;
561 }
562
563 if (!key)
564 goto skip_lookup;
565
566 rht_for_each(head, tbl, hash) {
567 if (unlikely(!(params.obj_cmpfn ?
568 params.obj_cmpfn(&arg, rht_obj(ht, head)) :
569 rhashtable_compare(&arg, rht_obj(ht, head)))))
570 goto out;
571 }
572
573skip_lookup:
574 err = 0;
575
576 head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
577
578 RCU_INIT_POINTER(obj->next, head);
579
580 rcu_assign_pointer(tbl->buckets[hash], obj);
581
582 atomic_inc(&ht->nelems);
583 if (rht_grow_above_75(ht, tbl))
584 schedule_work(&ht->run_work);
585
586out:
587 spin_unlock_bh(lock);
588 rcu_read_unlock();
589
590 return err;
591}
592
593/**
594 * rhashtable_insert_fast - insert object into hash table
595 * @ht: hash table
596 * @obj: pointer to hash head inside object
597 * @params: hash table parameters
598 *
599 * Will take a per bucket spinlock to protect against mutual mutations
600 * on the same bucket. Multiple insertions may occur in parallel unless
601 * they map to the same bucket lock.
602 *
603 * It is safe to call this function from atomic context.
604 *
605 * Will trigger an automatic deferred table resizing if the size grows
606 * beyond the watermark indicated by grow_decision() which can be passed
607 * to rhashtable_init().
608 */
609static inline int rhashtable_insert_fast(
610 struct rhashtable *ht, struct rhash_head *obj,
611 const struct rhashtable_params params)
612{
613 return __rhashtable_insert_fast(ht, NULL, obj, params);
614}
615
616/**
617 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
618 * @ht: hash table
619 * @obj: pointer to hash head inside object
620 * @params: hash table parameters
621 *
622 * Locks down the bucket chain in both the old and new table if a resize
623 * is in progress to ensure that writers can't remove from the old table
624 * and can't insert to the new table during the atomic operation of search
625 * and insertion. Searches for duplicates in both the old and new table if
626 * a resize is in progress.
627 *
628 * This lookup function may only be used for fixed key hash table (key_len
629 * parameter set). It will BUG() if used inappropriately.
630 *
631 * It is safe to call this function from atomic context.
632 *
633 * Will trigger an automatic deferred table resizing if the size grows
634 * beyond the watermark indicated by grow_decision() which can be passed
635 * to rhashtable_init().
636 */
637static inline int rhashtable_lookup_insert_fast(
638 struct rhashtable *ht, struct rhash_head *obj,
639 const struct rhashtable_params params)
640{
641 const char *key = rht_obj(ht, obj);
642
643 BUG_ON(ht->p.obj_hashfn);
644
645 return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj,
646 params);
647}
648
649/**
650 * rhashtable_lookup_insert_key - search and insert object to hash table
651 * with explicit key
652 * @ht: hash table
653 * @key: key
654 * @obj: pointer to hash head inside object
655 * @params: hash table parameters
656 *
657 * Locks down the bucket chain in both the old and new table if a resize
658 * is in progress to ensure that writers can't remove from the old table
659 * and can't insert to the new table during the atomic operation of search
660 * and insertion. Searches for duplicates in both the old and new table if
661 * a resize is in progress.
662 *
663 * Lookups may occur in parallel with hashtable mutations and resizing.
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 *
669 * Returns zero on success.
670 */
671static inline int rhashtable_lookup_insert_key(
672 struct rhashtable *ht, const void *key, struct rhash_head *obj,
673 const struct rhashtable_params params)
674{
675 BUG_ON(!ht->p.obj_hashfn || !key);
676
677 return __rhashtable_insert_fast(ht, key, obj, params);
678}
679
680static inline int __rhashtable_remove_fast(
681 struct rhashtable *ht, struct bucket_table *tbl,
682 struct rhash_head *obj, const struct rhashtable_params params)
683{
684 struct rhash_head __rcu **pprev;
685 struct rhash_head *he;
686 spinlock_t * lock;
687 unsigned hash;
688 int err = -ENOENT;
689
690 hash = rht_head_hashfn(ht, tbl, obj, params);
691 lock = rht_bucket_lock(tbl, hash);
692
693 spin_lock_bh(lock);
694
695 pprev = &tbl->buckets[hash];
696 rht_for_each(he, tbl, hash) {
697 if (he != obj) {
698 pprev = &he->next;
699 continue;
700 }
701
702 rcu_assign_pointer(*pprev, obj->next);
703 err = 0;
704 break;
705 }
706
707 spin_unlock_bh(lock);
708
709 return err;
710}
711
712/**
713 * rhashtable_remove_fast - remove object from hash table
714 * @ht: hash table
715 * @obj: pointer to hash head inside object
716 * @params: hash table parameters
717 *
718 * Since the hash chain is single linked, the removal operation needs to
719 * walk the bucket chain upon removal. The removal operation is thus
720 * considerable slow if the hash table is not correctly sized.
721 *
722 * Will automatically shrink the table via rhashtable_expand() if the
723 * shrink_decision function specified at rhashtable_init() returns true.
724 *
725 * Returns zero on success, -ENOENT if the entry could not be found.
726 */
727static inline int rhashtable_remove_fast(
728 struct rhashtable *ht, struct rhash_head *obj,
729 const struct rhashtable_params params)
730{
731 struct bucket_table *tbl;
732 int err;
733
734 rcu_read_lock();
735
736 tbl = rht_dereference_rcu(ht->tbl, ht);
737
738 /* Because we have already taken (and released) the bucket
739 * lock in old_tbl, if we find that future_tbl is not yet
740 * visible then that guarantees the entry to still be in
741 * the old tbl if it exists.
742 */
743 while ((err = __rhashtable_remove_fast(ht, tbl, obj, params)) &&
744 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
745 ;
746
747 if (err)
748 goto out;
749
750 atomic_dec(&ht->nelems);
751 if (rht_shrink_below_30(ht, tbl))
752 schedule_work(&ht->run_work);
753
754out:
755 rcu_read_unlock();
756
757 return err;
758}
759
Thomas Graf7e1e7762014-08-02 11:47:44 +0200760#endif /* _LINUX_RHASHTABLE_H */