blob: 4e1f535c2034e8d292f3d5b5bd569423878bab58 [file] [log] [blame]
Thomas Graf7e1e7762014-08-02 11:47:44 +02001/*
2 * Resizable, Scalable, Concurrent Hash Table
3 *
Herbert Xuca268932016-09-19 19:00:09 +08004 * Copyright (c) 2015-2016 Herbert Xu <herbert@gondor.apana.org.au>
Thomas Grafb5e2c152015-03-24 20:42:19 +00005 * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
Thomas Graf7e1e7762014-08-02 11:47:44 +02006 * 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 Xu07ee0722015-05-15 11:30:47 +080020#include <linux/atomic.h>
Herbert Xuf2dba9c2015-02-04 07:33:23 +110021#include <linux/compiler.h>
Herbert Xu3cf92222015-12-03 20:41:29 +080022#include <linux/err.h>
Herbert Xu6626af62015-03-20 18:18:45 -040023#include <linux/errno.h>
Herbert Xu31ccde22015-03-24 00:50:21 +110024#include <linux/jhash.h>
Thomas Graff89bd6f2015-01-02 23:00:21 +010025#include <linux/list_nulls.h>
Thomas Graf97defe12015-01-02 23:00:20 +010026#include <linux/workqueue.h>
Ying Xue86b35b62015-01-04 15:25:09 +080027#include <linux/mutex.h>
Ingo Molnarb2d09102017-02-04 01:27:20 +010028#include <linux/rculist.h>
Thomas Graf7e1e7762014-08-02 11:47:44 +020029
Thomas Graff89bd6f2015-01-02 23:00:21 +010030/*
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
Herbert Xu02fd97c2015-03-20 21:57:00 +110049/* Base bits plus 1 bit for nulls marker */
50#define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1)
51
Florian Westphal5f8ddea2017-04-16 02:55:09 +020052/* Maximum chain length before rehash
53 *
54 * The maximum (not average) chain length grows with the size of the hash
55 * table, at a rate of (log N)/(log log N).
56 *
57 * The value of 16 is selected so that even if the hash table grew to
58 * 2^32 you would not expect the maximum chain length to exceed it
59 * unless we are under attack (or extremely unlucky).
60 *
61 * As this limit is only to detect attacks, we don't need to set it to a
62 * lower value as you'd need the chain length to vastly exceed 16 to have
63 * any real effect on the system.
64 */
65#define RHT_ELASTICITY 16u
66
Thomas Graf7e1e7762014-08-02 11:47:44 +020067struct rhash_head {
Thomas Graf5300fdc2014-08-13 16:38:29 +020068 struct rhash_head __rcu *next;
Thomas Graf7e1e7762014-08-02 11:47:44 +020069};
70
Herbert Xuca268932016-09-19 19:00:09 +080071struct rhlist_head {
72 struct rhash_head rhead;
73 struct rhlist_head __rcu *next;
74};
75
Thomas Graf97defe12015-01-02 23:00:20 +010076/**
77 * struct bucket_table - Table of hash buckets
78 * @size: Number of hash buckets
Herbert Xuda204202017-02-11 19:26:47 +080079 * @nest: Number of bits of first-level nested table.
Herbert Xu63d512d2015-03-14 13:57:24 +110080 * @rehash: Current bucket being rehashed
Herbert Xu988dfbd2015-03-10 09:27:55 +110081 * @hash_rnd: Random seed to fold into hash
Thomas Graf97defe12015-01-02 23:00:20 +010082 * @locks_mask: Mask to apply before accessing locks[]
83 * @locks: Array of spinlocks protecting individual buckets
Herbert Xueddee5ba2015-03-14 13:57:20 +110084 * @walkers: List of active walkers
Herbert Xu9d901bc2015-03-14 13:57:23 +110085 * @rcu: RCU structure for freeing the table
Herbert Xuc4db8842015-03-14 13:57:25 +110086 * @future_tbl: Table under construction during rehashing
Herbert Xuda204202017-02-11 19:26:47 +080087 * @ntbl: Nested table used when out of memory.
Thomas Graf97defe12015-01-02 23:00:20 +010088 * @buckets: size * hash buckets
89 */
Thomas Graf7e1e7762014-08-02 11:47:44 +020090struct bucket_table {
Herbert Xu63d512d2015-03-14 13:57:24 +110091 unsigned int size;
Herbert Xuda204202017-02-11 19:26:47 +080092 unsigned int nest;
Herbert Xu63d512d2015-03-14 13:57:24 +110093 unsigned int rehash;
Herbert Xu988dfbd2015-03-10 09:27:55 +110094 u32 hash_rnd;
Eric Dumazetb9ebafb2015-02-20 06:48:57 -080095 unsigned int locks_mask;
96 spinlock_t *locks;
Herbert Xueddee5ba2015-03-14 13:57:20 +110097 struct list_head walkers;
Herbert Xu9d901bc2015-03-14 13:57:23 +110098 struct rcu_head rcu;
Eric Dumazetb9ebafb2015-02-20 06:48:57 -080099
Herbert Xuc4db8842015-03-14 13:57:25 +1100100 struct bucket_table __rcu *future_tbl;
101
Herbert Xuda204202017-02-11 19:26:47 +0800102 struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200103};
104
Herbert Xu02fd97c2015-03-20 21:57:00 +1100105/**
106 * struct rhashtable_compare_arg - Key for the function rhashtable_compare
107 * @ht: Hash table
108 * @key: Key to compare against
109 */
110struct rhashtable_compare_arg {
111 struct rhashtable *ht;
112 const void *key;
113};
114
Thomas Graf7e1e7762014-08-02 11:47:44 +0200115typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
Patrick McHardy49f7b332015-03-25 13:07:45 +0000116typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100117typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
118 const void *obj);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200119
120struct rhashtable;
121
122/**
123 * struct rhashtable_params - Hash table construction parameters
124 * @nelem_hint: Hint on number of elements, should be 75% of desired size
125 * @key_len: Length of key
126 * @key_offset: Offset of key in struct to be hashed
127 * @head_offset: Offset of rhash_head in struct to be hashed
Herbert Xuc2e213c2015-03-18 20:01:16 +1100128 * @max_size: Maximum size while expanding
129 * @min_size: Minimum size while shrinking
Davidlohr Bueso895a6072017-09-08 16:15:45 -0700130 * @locks_mul: Number of bucket locks to allocate per cpu (default: 32)
Florian Westphal48e75b432017-05-01 22:18:01 +0200131 * @automatic_shrinking: Enable automatic shrinking of tables
132 * @nulls_base: Base value to generate nulls marker
Herbert Xu31ccde22015-03-24 00:50:21 +1100133 * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200134 * @obj_hashfn: Function to hash object
Herbert Xu02fd97c2015-03-20 21:57:00 +1100135 * @obj_cmpfn: Function to compare key with object
Thomas Graf7e1e7762014-08-02 11:47:44 +0200136 */
137struct rhashtable_params {
Florian Westphal48e75b432017-05-01 22:18:01 +0200138 u16 nelem_hint;
139 u16 key_len;
140 u16 key_offset;
141 u16 head_offset;
Herbert Xuc2e213c2015-03-18 20:01:16 +1100142 unsigned int max_size;
Florian Westphal48e75b432017-05-01 22:18:01 +0200143 u16 min_size;
Thomas Grafb5e2c152015-03-24 20:42:19 +0000144 bool automatic_shrinking;
Florian Westphal48e75b432017-05-01 22:18:01 +0200145 u8 locks_mul;
146 u32 nulls_base;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200147 rht_hashfn_t hashfn;
148 rht_obj_hashfn_t obj_hashfn;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100149 rht_obj_cmpfn_t obj_cmpfn;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200150};
151
152/**
153 * struct rhashtable - Hash table handle
154 * @tbl: Bucket table
Herbert Xu31ccde22015-03-24 00:50:21 +1100155 * @key_len: Key length for hashfn
Herbert Xu6d684e52017-04-27 13:44:51 +0800156 * @max_elems: Maximum number of elements in table
Eric Dumazete5d672a2018-03-31 12:58:56 -0700157 * @p: Configuration parameters
Herbert Xuca268932016-09-19 19:00:09 +0800158 * @rhlist: True if this is an rhltable
Thomas Graf97defe12015-01-02 23:00:20 +0100159 * @run_work: Deferred worker to expand/shrink asynchronously
160 * @mutex: Mutex to protect current/future table swapping
Herbert Xuba7c95e2015-03-24 09:53:17 +1100161 * @lock: Spin lock to protect walker list
Eric Dumazete5d672a2018-03-31 12:58:56 -0700162 * @nelems: Number of elements in table
Thomas Graf7e1e7762014-08-02 11:47:44 +0200163 */
164struct rhashtable {
165 struct bucket_table __rcu *tbl;
Herbert Xu31ccde22015-03-24 00:50:21 +1100166 unsigned int key_len;
Herbert Xu6d684e52017-04-27 13:44:51 +0800167 unsigned int max_elems;
Eric Dumazete5d672a2018-03-31 12:58:56 -0700168 struct rhashtable_params p;
Herbert Xuca268932016-09-19 19:00:09 +0800169 bool rhlist;
Ying Xue57699a42015-01-16 11:13:09 +0800170 struct work_struct run_work;
Thomas Graf97defe12015-01-02 23:00:20 +0100171 struct mutex mutex;
Herbert Xuba7c95e2015-03-24 09:53:17 +1100172 spinlock_t lock;
Eric Dumazete5d672a2018-03-31 12:58:56 -0700173 atomic_t nelems;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200174};
175
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100176/**
Herbert Xuca268932016-09-19 19:00:09 +0800177 * struct rhltable - Hash table with duplicate objects in a list
178 * @ht: Underlying rhtable
179 */
180struct rhltable {
181 struct rhashtable ht;
182};
183
184/**
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100185 * struct rhashtable_walker - Hash table walker
186 * @list: List entry on list of walkers
Herbert Xueddee5ba2015-03-14 13:57:20 +1100187 * @tbl: The table that we were walking over
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100188 */
189struct rhashtable_walker {
190 struct list_head list;
Herbert Xueddee5ba2015-03-14 13:57:20 +1100191 struct bucket_table *tbl;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100192};
193
194/**
Herbert Xuca268932016-09-19 19:00:09 +0800195 * struct rhashtable_iter - Hash table iterator
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100196 * @ht: Table to iterate through
197 * @p: Current pointer
Herbert Xuca268932016-09-19 19:00:09 +0800198 * @list: Current hash list pointer
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100199 * @walker: Associated rhashtable walker
200 * @slot: Current slot
201 * @skip: Number of entries to skip in slot
202 */
203struct rhashtable_iter {
204 struct rhashtable *ht;
205 struct rhash_head *p;
Herbert Xuca268932016-09-19 19:00:09 +0800206 struct rhlist_head *list;
Herbert Xu246779d2016-08-18 16:50:56 +0800207 struct rhashtable_walker walker;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100208 unsigned int slot;
209 unsigned int skip;
Tom Herbert2db54b42017-12-04 10:31:42 -0800210 bool end_of_table;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100211};
212
Thomas Graff89bd6f2015-01-02 23:00:21 +0100213static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
214{
215 return NULLS_MARKER(ht->p.nulls_base + hash);
216}
217
218#define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
219 ((ptr) = (typeof(ptr)) rht_marker(ht, hash))
220
221static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
222{
223 return ((unsigned long) ptr & 1);
224}
225
226static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
227{
228 return ((unsigned long) ptr) >> 1;
229}
230
Herbert Xu02fd97c2015-03-20 21:57:00 +1100231static inline void *rht_obj(const struct rhashtable *ht,
232 const struct rhash_head *he)
233{
234 return (char *)he - ht->p.head_offset;
235}
236
237static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
238 unsigned int hash)
239{
240 return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
241}
242
Tom Herbert2b860932017-12-04 10:31:43 -0800243static inline unsigned int rht_key_get_hash(struct rhashtable *ht,
244 const void *key, const struct rhashtable_params params,
245 unsigned int hash_rnd)
Herbert Xu02fd97c2015-03-20 21:57:00 +1100246{
Thomas Graf299e5c32015-03-24 14:18:17 +0100247 unsigned int hash;
Herbert Xude91b252015-03-24 00:50:20 +1100248
Herbert Xu31ccde22015-03-24 00:50:21 +1100249 /* params must be equal to ht->p if it isn't constant. */
250 if (!__builtin_constant_p(params.key_len))
Tom Herbert2b860932017-12-04 10:31:43 -0800251 hash = ht->p.hashfn(key, ht->key_len, hash_rnd);
Herbert Xu31ccde22015-03-24 00:50:21 +1100252 else if (params.key_len) {
Thomas Graf299e5c32015-03-24 14:18:17 +0100253 unsigned int key_len = params.key_len;
Herbert Xu31ccde22015-03-24 00:50:21 +1100254
255 if (params.hashfn)
Tom Herbert2b860932017-12-04 10:31:43 -0800256 hash = params.hashfn(key, key_len, hash_rnd);
Herbert Xu31ccde22015-03-24 00:50:21 +1100257 else if (key_len & (sizeof(u32) - 1))
Tom Herbert2b860932017-12-04 10:31:43 -0800258 hash = jhash(key, key_len, hash_rnd);
Herbert Xu31ccde22015-03-24 00:50:21 +1100259 else
Tom Herbert2b860932017-12-04 10:31:43 -0800260 hash = jhash2(key, key_len / sizeof(u32), hash_rnd);
Herbert Xu31ccde22015-03-24 00:50:21 +1100261 } else {
Thomas Graf299e5c32015-03-24 14:18:17 +0100262 unsigned int key_len = ht->p.key_len;
Herbert Xu31ccde22015-03-24 00:50:21 +1100263
264 if (params.hashfn)
Tom Herbert2b860932017-12-04 10:31:43 -0800265 hash = params.hashfn(key, key_len, hash_rnd);
Herbert Xu31ccde22015-03-24 00:50:21 +1100266 else
Tom Herbert2b860932017-12-04 10:31:43 -0800267 hash = jhash(key, key_len, hash_rnd);
Herbert Xu31ccde22015-03-24 00:50:21 +1100268 }
269
Tom Herbert2b860932017-12-04 10:31:43 -0800270 return hash;
271}
272
273static inline unsigned int rht_key_hashfn(
274 struct rhashtable *ht, const struct bucket_table *tbl,
275 const void *key, const struct rhashtable_params params)
276{
277 unsigned int hash = rht_key_get_hash(ht, key, params, tbl->hash_rnd);
278
Herbert Xu31ccde22015-03-24 00:50:21 +1100279 return rht_bucket_index(tbl, hash);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100280}
281
282static inline unsigned int rht_head_hashfn(
283 struct rhashtable *ht, const struct bucket_table *tbl,
284 const struct rhash_head *he, const struct rhashtable_params params)
285{
286 const char *ptr = rht_obj(ht, he);
287
288 return likely(params.obj_hashfn) ?
Patrick McHardy49f7b332015-03-25 13:07:45 +0000289 rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
290 ht->p.key_len,
291 tbl->hash_rnd)) :
Herbert Xu02fd97c2015-03-20 21:57:00 +1100292 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
293}
294
295/**
296 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
297 * @ht: hash table
298 * @tbl: current table
299 */
300static inline bool rht_grow_above_75(const struct rhashtable *ht,
301 const struct bucket_table *tbl)
302{
303 /* Expand table when exceeding 75% load */
304 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
305 (!ht->p.max_size || tbl->size < ht->p.max_size);
306}
307
308/**
309 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
310 * @ht: hash table
311 * @tbl: current table
312 */
313static inline bool rht_shrink_below_30(const struct rhashtable *ht,
314 const struct bucket_table *tbl)
315{
316 /* Shrink table beneath 30% load */
317 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
318 tbl->size > ht->p.min_size;
319}
320
Herbert Xuccd57b12015-03-24 00:50:28 +1100321/**
322 * rht_grow_above_100 - returns true if nelems > table-size
323 * @ht: hash table
324 * @tbl: current table
325 */
326static inline bool rht_grow_above_100(const struct rhashtable *ht,
327 const struct bucket_table *tbl)
328{
Johannes Berg1d8dc3d2015-04-23 16:38:43 +0200329 return atomic_read(&ht->nelems) > tbl->size &&
330 (!ht->p.max_size || tbl->size < ht->p.max_size);
Herbert Xuccd57b12015-03-24 00:50:28 +1100331}
332
Herbert Xu07ee0722015-05-15 11:30:47 +0800333/**
334 * rht_grow_above_max - returns true if table is above maximum
335 * @ht: hash table
336 * @tbl: current table
337 */
338static inline bool rht_grow_above_max(const struct rhashtable *ht,
339 const struct bucket_table *tbl)
340{
Herbert Xu6d684e52017-04-27 13:44:51 +0800341 return atomic_read(&ht->nelems) >= ht->max_elems;
Herbert Xu07ee0722015-05-15 11:30:47 +0800342}
343
Herbert Xu02fd97c2015-03-20 21:57:00 +1100344/* The bucket lock is selected based on the hash and protects mutations
345 * on a group of hash buckets.
346 *
347 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
348 * a single lock always covers both buckets which may both contains
349 * entries which link to the same bucket of the old table during resizing.
350 * This allows to simplify the locking as locking the bucket in both
351 * tables during resize always guarantee protection.
352 *
353 * IMPORTANT: When holding the bucket lock of both the old and new table
354 * during expansions and shrinking, the old bucket lock must always be
355 * acquired first.
356 */
357static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
358 unsigned int hash)
359{
360 return &tbl->locks[hash & tbl->locks_mask];
361}
362
Thomas Graf7e1e7762014-08-02 11:47:44 +0200363#ifdef CONFIG_PROVE_LOCKING
Thomas Graf97defe12015-01-02 23:00:20 +0100364int lockdep_rht_mutex_is_held(struct rhashtable *ht);
Thomas Graf88d6ed12015-01-02 23:00:16 +0100365int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200366#else
Thomas Graf97defe12015-01-02 23:00:20 +0100367static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200368{
369 return 1;
370}
Thomas Graf88d6ed12015-01-02 23:00:16 +0100371
372static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
373 u32 hash)
374{
375 return 1;
376}
Thomas Graf7e1e7762014-08-02 11:47:44 +0200377#endif /* CONFIG_PROVE_LOCKING */
378
Herbert Xu488fb86e2015-03-20 21:56:59 +1100379int rhashtable_init(struct rhashtable *ht,
380 const struct rhashtable_params *params);
Herbert Xuca268932016-09-19 19:00:09 +0800381int rhltable_init(struct rhltable *hlt,
382 const struct rhashtable_params *params);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200383
Herbert Xuca268932016-09-19 19:00:09 +0800384void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
385 struct rhash_head *obj);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200386
Herbert Xu246779d2016-08-18 16:50:56 +0800387void rhashtable_walk_enter(struct rhashtable *ht,
388 struct rhashtable_iter *iter);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100389void rhashtable_walk_exit(struct rhashtable_iter *iter);
Tom Herbert97a6ec42017-12-04 10:31:41 -0800390int rhashtable_walk_start_check(struct rhashtable_iter *iter) __acquires(RCU);
391
392static inline void rhashtable_walk_start(struct rhashtable_iter *iter)
393{
394 (void)rhashtable_walk_start_check(iter);
395}
396
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100397void *rhashtable_walk_next(struct rhashtable_iter *iter);
Tom Herbert2db54b42017-12-04 10:31:42 -0800398void *rhashtable_walk_peek(struct rhashtable_iter *iter);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100399void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
400
Thomas Graf6b6f3022015-03-24 14:18:20 +0100401void rhashtable_free_and_destroy(struct rhashtable *ht,
402 void (*free_fn)(void *ptr, void *arg),
403 void *arg);
Thomas Graf97defe12015-01-02 23:00:20 +0100404void rhashtable_destroy(struct rhashtable *ht);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200405
Herbert Xuda204202017-02-11 19:26:47 +0800406struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
407 unsigned int hash);
408struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
409 struct bucket_table *tbl,
410 unsigned int hash);
411
Thomas Graf7e1e7762014-08-02 11:47:44 +0200412#define rht_dereference(p, ht) \
413 rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
414
415#define rht_dereference_rcu(p, ht) \
416 rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
417
Thomas Graf88d6ed12015-01-02 23:00:16 +0100418#define rht_dereference_bucket(p, tbl, hash) \
419 rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200420
Thomas Graf88d6ed12015-01-02 23:00:16 +0100421#define rht_dereference_bucket_rcu(p, tbl, hash) \
422 rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
423
424#define rht_entry(tpos, pos, member) \
425 ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
426
Herbert Xuda204202017-02-11 19:26:47 +0800427static inline struct rhash_head __rcu *const *rht_bucket(
428 const struct bucket_table *tbl, unsigned int hash)
429{
430 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
431 &tbl->buckets[hash];
432}
433
434static inline struct rhash_head __rcu **rht_bucket_var(
435 struct bucket_table *tbl, unsigned int hash)
436{
437 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
438 &tbl->buckets[hash];
439}
440
441static inline struct rhash_head __rcu **rht_bucket_insert(
442 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
443{
444 return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
445 &tbl->buckets[hash];
446}
447
Thomas Graf88d6ed12015-01-02 23:00:16 +0100448/**
449 * rht_for_each_continue - continue iterating over hash chain
450 * @pos: the &struct rhash_head to use as a loop cursor.
451 * @head: the previous &struct rhash_head to continue from
452 * @tbl: the &struct bucket_table
453 * @hash: the hash value / bucket index
454 */
455#define rht_for_each_continue(pos, head, tbl, hash) \
456 for (pos = rht_dereference_bucket(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100457 !rht_is_a_nulls(pos); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100458 pos = rht_dereference_bucket((pos)->next, tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200459
460/**
461 * rht_for_each - iterate over hash chain
Thomas Graf88d6ed12015-01-02 23:00:16 +0100462 * @pos: the &struct rhash_head to use as a loop cursor.
463 * @tbl: the &struct bucket_table
464 * @hash: the hash value / bucket index
Thomas Graf7e1e7762014-08-02 11:47:44 +0200465 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100466#define rht_for_each(pos, tbl, hash) \
Herbert Xuda204202017-02-11 19:26:47 +0800467 rht_for_each_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
Thomas Graf88d6ed12015-01-02 23:00:16 +0100468
469/**
470 * rht_for_each_entry_continue - continue iterating over hash chain
471 * @tpos: the type * to use as a loop cursor.
472 * @pos: the &struct rhash_head to use as a loop cursor.
473 * @head: the previous &struct rhash_head to continue from
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#define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \
479 for (pos = rht_dereference_bucket(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100480 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100481 pos = rht_dereference_bucket((pos)->next, tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200482
483/**
484 * rht_for_each_entry - iterate over hash chain of given type
Thomas Graf88d6ed12015-01-02 23:00:16 +0100485 * @tpos: the type * to use as a loop cursor.
486 * @pos: the &struct rhash_head to use as a loop cursor.
487 * @tbl: the &struct bucket_table
488 * @hash: the hash value / bucket index
489 * @member: name of the &struct rhash_head within the hashable struct.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200490 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100491#define rht_for_each_entry(tpos, pos, tbl, hash, member) \
Herbert Xuda204202017-02-11 19:26:47 +0800492 rht_for_each_entry_continue(tpos, pos, *rht_bucket(tbl, hash), \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100493 tbl, hash, member)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200494
495/**
496 * rht_for_each_entry_safe - safely iterate over hash chain of given type
Thomas Graf88d6ed12015-01-02 23:00:16 +0100497 * @tpos: the type * to use as a loop cursor.
498 * @pos: the &struct rhash_head to use as a loop cursor.
499 * @next: the &struct rhash_head to use as next in loop cursor.
500 * @tbl: the &struct bucket_table
501 * @hash: the hash value / bucket index
502 * @member: name of the &struct rhash_head within the hashable struct.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200503 *
504 * This hash chain list-traversal primitive allows for the looped code to
505 * remove the loop cursor from the list.
506 */
Herbert Xuda204202017-02-11 19:26:47 +0800507#define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \
508 for (pos = rht_dereference_bucket(*rht_bucket(tbl, hash), tbl, hash), \
509 next = !rht_is_a_nulls(pos) ? \
510 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
511 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
512 pos = next, \
513 next = !rht_is_a_nulls(pos) ? \
Patrick McHardy607954b2015-01-21 11:12:13 +0000514 rht_dereference_bucket(pos->next, tbl, hash) : NULL)
Thomas Graf88d6ed12015-01-02 23:00:16 +0100515
516/**
517 * rht_for_each_rcu_continue - continue iterating over rcu hash chain
518 * @pos: the &struct rhash_head to use as a loop cursor.
519 * @head: the previous &struct rhash_head to continue from
520 * @tbl: the &struct bucket_table
521 * @hash: the hash value / bucket index
522 *
523 * This hash chain list-traversal primitive may safely run concurrently with
524 * the _rcu mutation primitives such as rhashtable_insert() as long as the
525 * traversal is guarded by rcu_read_lock().
526 */
527#define rht_for_each_rcu_continue(pos, head, tbl, hash) \
528 for (({barrier(); }), \
529 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100530 !rht_is_a_nulls(pos); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100531 pos = rcu_dereference_raw(pos->next))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200532
533/**
534 * rht_for_each_rcu - iterate over rcu hash chain
Thomas Graf88d6ed12015-01-02 23:00:16 +0100535 * @pos: the &struct rhash_head to use as a loop cursor.
536 * @tbl: the &struct bucket_table
537 * @hash: the hash value / bucket index
Thomas Graf7e1e7762014-08-02 11:47:44 +0200538 *
539 * This hash chain list-traversal primitive may safely run concurrently with
Thomas Graf88d6ed12015-01-02 23:00:16 +0100540 * the _rcu mutation primitives such as rhashtable_insert() as long as the
Thomas Graf7e1e7762014-08-02 11:47:44 +0200541 * traversal is guarded by rcu_read_lock().
542 */
Thomas Graf88d6ed12015-01-02 23:00:16 +0100543#define rht_for_each_rcu(pos, tbl, hash) \
Herbert Xuda204202017-02-11 19:26:47 +0800544 rht_for_each_rcu_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
Thomas Graf88d6ed12015-01-02 23:00:16 +0100545
546/**
547 * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
548 * @tpos: the type * to use as a loop cursor.
549 * @pos: the &struct rhash_head to use as a loop cursor.
550 * @head: the previous &struct rhash_head to continue from
551 * @tbl: the &struct bucket_table
552 * @hash: the hash value / bucket index
553 * @member: name of the &struct rhash_head within the hashable struct.
554 *
555 * This hash chain list-traversal primitive may safely run concurrently with
556 * the _rcu mutation primitives such as rhashtable_insert() as long as the
557 * traversal is guarded by rcu_read_lock().
558 */
559#define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
560 for (({barrier(); }), \
561 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
Thomas Graff89bd6f2015-01-02 23:00:21 +0100562 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100563 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200564
565/**
566 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
Thomas Graf88d6ed12015-01-02 23:00:16 +0100567 * @tpos: the type * to use as a loop cursor.
568 * @pos: the &struct rhash_head to use as a loop cursor.
569 * @tbl: the &struct bucket_table
570 * @hash: the hash value / bucket index
571 * @member: name of the &struct rhash_head within the hashable struct.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200572 *
573 * This hash chain list-traversal primitive may safely run concurrently with
Thomas Graf88d6ed12015-01-02 23:00:16 +0100574 * the _rcu mutation primitives such as rhashtable_insert() as long as the
Thomas Graf7e1e7762014-08-02 11:47:44 +0200575 * traversal is guarded by rcu_read_lock().
576 */
Herbert Xuda204202017-02-11 19:26:47 +0800577#define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \
578 rht_for_each_entry_rcu_continue(tpos, pos, *rht_bucket(tbl, hash), \
Thomas Graf88d6ed12015-01-02 23:00:16 +0100579 tbl, hash, member)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200580
Herbert Xuca268932016-09-19 19:00:09 +0800581/**
582 * rhl_for_each_rcu - iterate over rcu hash table list
583 * @pos: the &struct rlist_head to use as a loop cursor.
584 * @list: the head of the list
585 *
586 * This hash chain list-traversal primitive should be used on the
587 * list returned by rhltable_lookup.
588 */
589#define rhl_for_each_rcu(pos, list) \
590 for (pos = list; pos; pos = rcu_dereference_raw(pos->next))
591
592/**
593 * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
594 * @tpos: the type * to use as a loop cursor.
595 * @pos: the &struct rlist_head to use as a loop cursor.
596 * @list: the head of the list
597 * @member: name of the &struct rlist_head within the hashable struct.
598 *
599 * This hash chain list-traversal primitive should be used on the
600 * list returned by rhltable_lookup.
601 */
602#define rhl_for_each_entry_rcu(tpos, pos, list, member) \
603 for (pos = list; pos && rht_entry(tpos, pos, member); \
604 pos = rcu_dereference_raw(pos->next))
605
Herbert Xu02fd97c2015-03-20 21:57:00 +1100606static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
607 const void *obj)
608{
609 struct rhashtable *ht = arg->ht;
610 const char *ptr = obj;
611
612 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
613}
614
Herbert Xuca268932016-09-19 19:00:09 +0800615/* Internal function, do not use. */
616static inline struct rhash_head *__rhashtable_lookup(
Herbert Xu02fd97c2015-03-20 21:57:00 +1100617 struct rhashtable *ht, const void *key,
618 const struct rhashtable_params params)
619{
620 struct rhashtable_compare_arg arg = {
621 .ht = ht,
622 .key = key,
623 };
Herbert Xuda204202017-02-11 19:26:47 +0800624 struct bucket_table *tbl;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100625 struct rhash_head *he;
Thomas Graf299e5c32015-03-24 14:18:17 +0100626 unsigned int hash;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100627
Herbert Xu02fd97c2015-03-20 21:57:00 +1100628 tbl = rht_dereference_rcu(ht->tbl, ht);
629restart:
630 hash = rht_key_hashfn(ht, tbl, key, params);
631 rht_for_each_rcu(he, tbl, hash) {
632 if (params.obj_cmpfn ?
633 params.obj_cmpfn(&arg, rht_obj(ht, he)) :
634 rhashtable_compare(&arg, rht_obj(ht, he)))
635 continue;
Herbert Xuca268932016-09-19 19:00:09 +0800636 return he;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100637 }
638
639 /* Ensure we see any new tables. */
640 smp_rmb();
641
642 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
643 if (unlikely(tbl))
644 goto restart;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100645
646 return NULL;
647}
648
Herbert Xuca268932016-09-19 19:00:09 +0800649/**
650 * rhashtable_lookup - search hash table
651 * @ht: hash table
652 * @key: the pointer to the key
653 * @params: hash table parameters
654 *
655 * Computes the hash value for the key and traverses the bucket chain looking
656 * for a entry with an identical key. The first matching entry is returned.
657 *
658 * This must only be called under the RCU read lock.
659 *
660 * Returns the first entry on which the compare function returned true.
661 */
662static inline void *rhashtable_lookup(
663 struct rhashtable *ht, const void *key,
664 const struct rhashtable_params params)
665{
666 struct rhash_head *he = __rhashtable_lookup(ht, key, params);
667
668 return he ? rht_obj(ht, he) : NULL;
669}
670
671/**
672 * rhashtable_lookup_fast - search hash table, without RCU read lock
673 * @ht: hash table
674 * @key: the pointer to the key
675 * @params: hash table parameters
676 *
677 * Computes the hash value for the key and traverses the bucket chain looking
678 * for a entry with an identical key. The first matching entry is returned.
679 *
680 * Only use this function when you have other mechanisms guaranteeing
681 * that the object won't go away after the RCU read lock is released.
682 *
683 * Returns the first entry on which the compare function returned true.
684 */
685static inline void *rhashtable_lookup_fast(
686 struct rhashtable *ht, const void *key,
687 const struct rhashtable_params params)
688{
689 void *obj;
690
691 rcu_read_lock();
692 obj = rhashtable_lookup(ht, key, params);
693 rcu_read_unlock();
694
695 return obj;
696}
697
698/**
699 * rhltable_lookup - search hash list table
700 * @hlt: hash table
701 * @key: the pointer to the key
702 * @params: hash table parameters
703 *
704 * Computes the hash value for the key and traverses the bucket chain looking
705 * for a entry with an identical key. All matching entries are returned
706 * in a list.
707 *
708 * This must only be called under the RCU read lock.
709 *
710 * Returns the list of entries that match the given key.
711 */
712static inline struct rhlist_head *rhltable_lookup(
713 struct rhltable *hlt, const void *key,
714 const struct rhashtable_params params)
715{
716 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params);
717
718 return he ? container_of(he, struct rhlist_head, rhead) : NULL;
719}
720
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200721/* Internal function, please use rhashtable_insert_fast() instead. This
722 * function returns the existing element already in hashes in there is a clash,
723 * otherwise it returns an error via ERR_PTR().
724 */
725static inline void *__rhashtable_insert_fast(
Herbert Xu02fd97c2015-03-20 21:57:00 +1100726 struct rhashtable *ht, const void *key, struct rhash_head *obj,
Herbert Xuca268932016-09-19 19:00:09 +0800727 const struct rhashtable_params params, bool rhlist)
Herbert Xu02fd97c2015-03-20 21:57:00 +1100728{
729 struct rhashtable_compare_arg arg = {
730 .ht = ht,
731 .key = key,
732 };
Herbert Xuca268932016-09-19 19:00:09 +0800733 struct rhash_head __rcu **pprev;
734 struct bucket_table *tbl;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100735 struct rhash_head *head;
736 spinlock_t *lock;
Thomas Graf299e5c32015-03-24 14:18:17 +0100737 unsigned int hash;
Herbert Xuca268932016-09-19 19:00:09 +0800738 int elasticity;
739 void *data;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100740
741 rcu_read_lock();
742
743 tbl = rht_dereference_rcu(ht->tbl, ht);
Herbert Xuca268932016-09-19 19:00:09 +0800744 hash = rht_head_hashfn(ht, tbl, obj, params);
745 lock = rht_bucket_lock(tbl, hash);
746 spin_lock_bh(lock);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100747
Herbert Xuca268932016-09-19 19:00:09 +0800748 if (unlikely(rht_dereference_bucket(tbl->future_tbl, tbl, hash))) {
749slow_path:
Herbert Xub8244782015-03-24 00:50:26 +1100750 spin_unlock_bh(lock);
Herbert Xuca268932016-09-19 19:00:09 +0800751 rcu_read_unlock();
752 return rhashtable_insert_slow(ht, key, obj);
Herbert Xub8244782015-03-24 00:50:26 +1100753 }
754
Florian Westphal5f8ddea2017-04-16 02:55:09 +0200755 elasticity = RHT_ELASTICITY;
Herbert Xuda204202017-02-11 19:26:47 +0800756 pprev = rht_bucket_insert(ht, tbl, hash);
757 data = ERR_PTR(-ENOMEM);
758 if (!pprev)
759 goto out;
760
761 rht_for_each_continue(head, *pprev, tbl, hash) {
Herbert Xuca268932016-09-19 19:00:09 +0800762 struct rhlist_head *plist;
763 struct rhlist_head *list;
Herbert Xu3cf92222015-12-03 20:41:29 +0800764
Herbert Xuca268932016-09-19 19:00:09 +0800765 elasticity--;
766 if (!key ||
767 (params.obj_cmpfn ?
768 params.obj_cmpfn(&arg, rht_obj(ht, head)) :
Paul Blakeyd3dcf8e2018-03-04 17:29:48 +0200769 rhashtable_compare(&arg, rht_obj(ht, head)))) {
770 pprev = &head->next;
Herbert Xuca268932016-09-19 19:00:09 +0800771 continue;
Paul Blakeyd3dcf8e2018-03-04 17:29:48 +0200772 }
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200773
Herbert Xuca268932016-09-19 19:00:09 +0800774 data = rht_obj(ht, head);
775
776 if (!rhlist)
777 goto out;
778
779
780 list = container_of(obj, struct rhlist_head, rhead);
781 plist = container_of(head, struct rhlist_head, rhead);
782
783 RCU_INIT_POINTER(list->next, plist);
784 head = rht_dereference_bucket(head->next, tbl, hash);
785 RCU_INIT_POINTER(list->rhead.next, head);
786 rcu_assign_pointer(*pprev, obj);
787
788 goto good;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100789 }
790
Herbert Xuca268932016-09-19 19:00:09 +0800791 if (elasticity <= 0)
792 goto slow_path;
793
794 data = ERR_PTR(-E2BIG);
Herbert Xu07ee0722015-05-15 11:30:47 +0800795 if (unlikely(rht_grow_above_max(ht, tbl)))
796 goto out;
797
Herbert Xuca268932016-09-19 19:00:09 +0800798 if (unlikely(rht_grow_above_100(ht, tbl)))
799 goto slow_path;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100800
Herbert Xuda204202017-02-11 19:26:47 +0800801 head = rht_dereference_bucket(*pprev, tbl, hash);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100802
803 RCU_INIT_POINTER(obj->next, head);
Herbert Xuca268932016-09-19 19:00:09 +0800804 if (rhlist) {
805 struct rhlist_head *list;
806
807 list = container_of(obj, struct rhlist_head, rhead);
808 RCU_INIT_POINTER(list->next, NULL);
809 }
Herbert Xu02fd97c2015-03-20 21:57:00 +1100810
Herbert Xuda204202017-02-11 19:26:47 +0800811 rcu_assign_pointer(*pprev, obj);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100812
813 atomic_inc(&ht->nelems);
814 if (rht_grow_above_75(ht, tbl))
815 schedule_work(&ht->run_work);
816
Herbert Xuca268932016-09-19 19:00:09 +0800817good:
818 data = NULL;
819
Herbert Xu02fd97c2015-03-20 21:57:00 +1100820out:
821 spin_unlock_bh(lock);
822 rcu_read_unlock();
823
Herbert Xuca268932016-09-19 19:00:09 +0800824 return data;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100825}
826
827/**
828 * rhashtable_insert_fast - insert object into hash table
829 * @ht: hash table
830 * @obj: pointer to hash head inside object
831 * @params: hash table parameters
832 *
833 * Will take a per bucket spinlock to protect against mutual mutations
834 * on the same bucket. Multiple insertions may occur in parallel unless
835 * they map to the same bucket lock.
836 *
837 * It is safe to call this function from atomic context.
838 *
NeilBrown0c6f69a2018-04-24 08:29:13 +1000839 * Will trigger an automatic deferred table resizing if residency in the
840 * table grows beyond 70%.
Herbert Xu02fd97c2015-03-20 21:57:00 +1100841 */
842static inline int rhashtable_insert_fast(
843 struct rhashtable *ht, struct rhash_head *obj,
844 const struct rhashtable_params params)
845{
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200846 void *ret;
847
Herbert Xuca268932016-09-19 19:00:09 +0800848 ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200849 if (IS_ERR(ret))
850 return PTR_ERR(ret);
851
852 return ret == NULL ? 0 : -EEXIST;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100853}
854
855/**
Herbert Xuca268932016-09-19 19:00:09 +0800856 * rhltable_insert_key - insert object into hash list table
857 * @hlt: hash list table
858 * @key: the pointer to the key
859 * @list: pointer to hash list head inside object
860 * @params: hash table parameters
861 *
862 * Will take a per bucket spinlock to protect against mutual mutations
863 * on the same bucket. Multiple insertions may occur in parallel unless
864 * they map to the same bucket lock.
865 *
866 * It is safe to call this function from atomic context.
867 *
NeilBrown0c6f69a2018-04-24 08:29:13 +1000868 * Will trigger an automatic deferred table resizing if residency in the
869 * table grows beyond 70%.
Herbert Xuca268932016-09-19 19:00:09 +0800870 */
871static inline int rhltable_insert_key(
872 struct rhltable *hlt, const void *key, struct rhlist_head *list,
873 const struct rhashtable_params params)
874{
875 return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
876 params, true));
877}
878
879/**
880 * rhltable_insert - insert object into hash list table
881 * @hlt: hash list table
882 * @list: pointer to hash list head inside object
883 * @params: hash table parameters
884 *
885 * Will take a per bucket spinlock to protect against mutual mutations
886 * on the same bucket. Multiple insertions may occur in parallel unless
887 * they map to the same bucket lock.
888 *
889 * It is safe to call this function from atomic context.
890 *
NeilBrown0c6f69a2018-04-24 08:29:13 +1000891 * Will trigger an automatic deferred table resizing if residency in the
892 * table grows beyond 70%.
Herbert Xuca268932016-09-19 19:00:09 +0800893 */
894static inline int rhltable_insert(
895 struct rhltable *hlt, struct rhlist_head *list,
896 const struct rhashtable_params params)
897{
898 const char *key = rht_obj(&hlt->ht, &list->rhead);
899
900 key += params.key_offset;
901
902 return rhltable_insert_key(hlt, key, list, params);
903}
904
905/**
Herbert Xu02fd97c2015-03-20 21:57:00 +1100906 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
907 * @ht: hash table
908 * @obj: pointer to hash head inside object
909 * @params: hash table parameters
910 *
911 * Locks down the bucket chain in both the old and new table if a resize
912 * is in progress to ensure that writers can't remove from the old table
913 * and can't insert to the new table during the atomic operation of search
914 * and insertion. Searches for duplicates in both the old and new table if
915 * a resize is in progress.
916 *
917 * This lookup function may only be used for fixed key hash table (key_len
918 * parameter set). It will BUG() if used inappropriately.
919 *
920 * It is safe to call this function from atomic context.
921 *
NeilBrown0c6f69a2018-04-24 08:29:13 +1000922 * Will trigger an automatic deferred table resizing if residency in the
923 * table grows beyond 70%.
Herbert Xu02fd97c2015-03-20 21:57:00 +1100924 */
925static inline int rhashtable_lookup_insert_fast(
926 struct rhashtable *ht, struct rhash_head *obj,
927 const struct rhashtable_params params)
928{
929 const char *key = rht_obj(ht, obj);
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200930 void *ret;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100931
932 BUG_ON(ht->p.obj_hashfn);
933
Herbert Xuca268932016-09-19 19:00:09 +0800934 ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
935 false);
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200936 if (IS_ERR(ret))
937 return PTR_ERR(ret);
938
939 return ret == NULL ? 0 : -EEXIST;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100940}
941
942/**
Andreas Gruenbacherf9fe1c12017-03-18 00:36:15 +0100943 * rhashtable_lookup_get_insert_fast - lookup and insert object into hash table
944 * @ht: hash table
945 * @obj: pointer to hash head inside object
946 * @params: hash table parameters
947 *
948 * Just like rhashtable_lookup_insert_fast(), but this function returns the
949 * object if it exists, NULL if it did not and the insertion was successful,
950 * and an ERR_PTR otherwise.
951 */
952static inline void *rhashtable_lookup_get_insert_fast(
953 struct rhashtable *ht, struct rhash_head *obj,
954 const struct rhashtable_params params)
955{
956 const char *key = rht_obj(ht, obj);
957
958 BUG_ON(ht->p.obj_hashfn);
959
960 return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
961 false);
962}
963
964/**
Herbert Xu02fd97c2015-03-20 21:57:00 +1100965 * rhashtable_lookup_insert_key - search and insert object to hash table
966 * with explicit key
967 * @ht: hash table
968 * @key: key
969 * @obj: pointer to hash head inside object
970 * @params: hash table parameters
971 *
972 * Locks down the bucket chain in both the old and new table if a resize
973 * is in progress to ensure that writers can't remove from the old table
974 * and can't insert to the new table during the atomic operation of search
975 * and insertion. Searches for duplicates in both the old and new table if
976 * a resize is in progress.
977 *
978 * Lookups may occur in parallel with hashtable mutations and resizing.
979 *
NeilBrown0c6f69a2018-04-24 08:29:13 +1000980 * Will trigger an automatic deferred table resizing if residency in the
981 * table grows beyond 70%.
Herbert Xu02fd97c2015-03-20 21:57:00 +1100982 *
983 * Returns zero on success.
984 */
985static inline int rhashtable_lookup_insert_key(
986 struct rhashtable *ht, const void *key, struct rhash_head *obj,
987 const struct rhashtable_params params)
988{
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200989 void *ret;
990
991 BUG_ON(!ht->p.obj_hashfn || !key);
992
Herbert Xuca268932016-09-19 19:00:09 +0800993 ret = __rhashtable_insert_fast(ht, key, obj, params, false);
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200994 if (IS_ERR(ret))
995 return PTR_ERR(ret);
996
997 return ret == NULL ? 0 : -EEXIST;
998}
999
1000/**
1001 * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
1002 * @ht: hash table
1003 * @obj: pointer to hash head inside object
1004 * @params: hash table parameters
1005 * @data: pointer to element data already in hashes
1006 *
1007 * Just like rhashtable_lookup_insert_key(), but this function returns the
1008 * object if it exists, NULL if it does not and the insertion was successful,
1009 * and an ERR_PTR otherwise.
1010 */
1011static inline void *rhashtable_lookup_get_insert_key(
1012 struct rhashtable *ht, const void *key, struct rhash_head *obj,
1013 const struct rhashtable_params params)
1014{
Herbert Xu02fd97c2015-03-20 21:57:00 +11001015 BUG_ON(!ht->p.obj_hashfn || !key);
1016
Herbert Xuca268932016-09-19 19:00:09 +08001017 return __rhashtable_insert_fast(ht, key, obj, params, false);
Herbert Xu02fd97c2015-03-20 21:57:00 +11001018}
1019
Thomas Grafac833bd2015-03-24 14:18:18 +01001020/* Internal function, please use rhashtable_remove_fast() instead */
Herbert Xuca268932016-09-19 19:00:09 +08001021static inline int __rhashtable_remove_fast_one(
Herbert Xu02fd97c2015-03-20 21:57:00 +11001022 struct rhashtable *ht, struct bucket_table *tbl,
Herbert Xuca268932016-09-19 19:00:09 +08001023 struct rhash_head *obj, const struct rhashtable_params params,
1024 bool rhlist)
Herbert Xu02fd97c2015-03-20 21:57:00 +11001025{
1026 struct rhash_head __rcu **pprev;
1027 struct rhash_head *he;
1028 spinlock_t * lock;
Thomas Graf299e5c32015-03-24 14:18:17 +01001029 unsigned int hash;
Herbert Xu02fd97c2015-03-20 21:57:00 +11001030 int err = -ENOENT;
1031
1032 hash = rht_head_hashfn(ht, tbl, obj, params);
1033 lock = rht_bucket_lock(tbl, hash);
1034
1035 spin_lock_bh(lock);
1036
Herbert Xuda204202017-02-11 19:26:47 +08001037 pprev = rht_bucket_var(tbl, hash);
1038 rht_for_each_continue(he, *pprev, tbl, hash) {
Herbert Xuca268932016-09-19 19:00:09 +08001039 struct rhlist_head *list;
1040
1041 list = container_of(he, struct rhlist_head, rhead);
1042
Herbert Xu02fd97c2015-03-20 21:57:00 +11001043 if (he != obj) {
Herbert Xuca268932016-09-19 19:00:09 +08001044 struct rhlist_head __rcu **lpprev;
1045
Herbert Xu02fd97c2015-03-20 21:57:00 +11001046 pprev = &he->next;
Herbert Xuca268932016-09-19 19:00:09 +08001047
1048 if (!rhlist)
1049 continue;
1050
1051 do {
1052 lpprev = &list->next;
1053 list = rht_dereference_bucket(list->next,
1054 tbl, hash);
1055 } while (list && obj != &list->rhead);
1056
1057 if (!list)
1058 continue;
1059
1060 list = rht_dereference_bucket(list->next, tbl, hash);
1061 RCU_INIT_POINTER(*lpprev, list);
1062 err = 0;
1063 break;
Herbert Xu02fd97c2015-03-20 21:57:00 +11001064 }
1065
Herbert Xuca268932016-09-19 19:00:09 +08001066 obj = rht_dereference_bucket(obj->next, tbl, hash);
1067 err = 1;
1068
1069 if (rhlist) {
1070 list = rht_dereference_bucket(list->next, tbl, hash);
1071 if (list) {
1072 RCU_INIT_POINTER(list->rhead.next, obj);
1073 obj = &list->rhead;
1074 err = 0;
1075 }
1076 }
1077
1078 rcu_assign_pointer(*pprev, obj);
Herbert Xu02fd97c2015-03-20 21:57:00 +11001079 break;
1080 }
1081
1082 spin_unlock_bh(lock);
1083
Herbert Xuca268932016-09-19 19:00:09 +08001084 if (err > 0) {
1085 atomic_dec(&ht->nelems);
1086 if (unlikely(ht->p.automatic_shrinking &&
1087 rht_shrink_below_30(ht, tbl)))
1088 schedule_work(&ht->run_work);
1089 err = 0;
1090 }
1091
1092 return err;
1093}
1094
1095/* Internal function, please use rhashtable_remove_fast() instead */
1096static inline int __rhashtable_remove_fast(
1097 struct rhashtable *ht, struct rhash_head *obj,
1098 const struct rhashtable_params params, bool rhlist)
1099{
1100 struct bucket_table *tbl;
1101 int err;
1102
1103 rcu_read_lock();
1104
1105 tbl = rht_dereference_rcu(ht->tbl, ht);
1106
1107 /* Because we have already taken (and released) the bucket
1108 * lock in old_tbl, if we find that future_tbl is not yet
1109 * visible then that guarantees the entry to still be in
1110 * the old tbl if it exists.
1111 */
1112 while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
1113 rhlist)) &&
1114 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1115 ;
1116
1117 rcu_read_unlock();
1118
Herbert Xu02fd97c2015-03-20 21:57:00 +11001119 return err;
1120}
1121
1122/**
1123 * rhashtable_remove_fast - remove object from hash table
1124 * @ht: hash table
1125 * @obj: pointer to hash head inside object
1126 * @params: hash table parameters
1127 *
1128 * Since the hash chain is single linked, the removal operation needs to
1129 * walk the bucket chain upon removal. The removal operation is thus
1130 * considerable slow if the hash table is not correctly sized.
1131 *
NeilBrown0c6f69a2018-04-24 08:29:13 +10001132 * Will automatically shrink the table if permitted when residency drops
1133 * below 30%.
Herbert Xu02fd97c2015-03-20 21:57:00 +11001134 *
1135 * Returns zero on success, -ENOENT if the entry could not be found.
1136 */
1137static inline int rhashtable_remove_fast(
1138 struct rhashtable *ht, struct rhash_head *obj,
1139 const struct rhashtable_params params)
1140{
Herbert Xuca268932016-09-19 19:00:09 +08001141 return __rhashtable_remove_fast(ht, obj, params, false);
1142}
Herbert Xu02fd97c2015-03-20 21:57:00 +11001143
Herbert Xuca268932016-09-19 19:00:09 +08001144/**
1145 * rhltable_remove - remove object from hash list table
1146 * @hlt: hash list table
1147 * @list: pointer to hash list head inside object
1148 * @params: hash table parameters
1149 *
1150 * Since the hash chain is single linked, the removal operation needs to
1151 * walk the bucket chain upon removal. The removal operation is thus
1152 * considerable slow if the hash table is not correctly sized.
1153 *
NeilBrown0c6f69a2018-04-24 08:29:13 +10001154 * Will automatically shrink the table if permitted when residency drops
1155 * below 30%
Herbert Xuca268932016-09-19 19:00:09 +08001156 *
1157 * Returns zero on success, -ENOENT if the entry could not be found.
1158 */
1159static inline int rhltable_remove(
1160 struct rhltable *hlt, struct rhlist_head *list,
1161 const struct rhashtable_params params)
1162{
1163 return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
Herbert Xu02fd97c2015-03-20 21:57:00 +11001164}
1165
Tom Herbert3502cad2015-12-15 15:41:36 -08001166/* Internal function, please use rhashtable_replace_fast() instead */
1167static inline int __rhashtable_replace_fast(
1168 struct rhashtable *ht, struct bucket_table *tbl,
1169 struct rhash_head *obj_old, struct rhash_head *obj_new,
1170 const struct rhashtable_params params)
1171{
1172 struct rhash_head __rcu **pprev;
1173 struct rhash_head *he;
1174 spinlock_t *lock;
1175 unsigned int hash;
1176 int err = -ENOENT;
1177
1178 /* Minimally, the old and new objects must have same hash
1179 * (which should mean identifiers are the same).
1180 */
1181 hash = rht_head_hashfn(ht, tbl, obj_old, params);
1182 if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1183 return -EINVAL;
1184
1185 lock = rht_bucket_lock(tbl, hash);
1186
1187 spin_lock_bh(lock);
1188
Herbert Xuda204202017-02-11 19:26:47 +08001189 pprev = rht_bucket_var(tbl, hash);
1190 rht_for_each_continue(he, *pprev, tbl, hash) {
Tom Herbert3502cad2015-12-15 15:41:36 -08001191 if (he != obj_old) {
1192 pprev = &he->next;
1193 continue;
1194 }
1195
1196 rcu_assign_pointer(obj_new->next, obj_old->next);
1197 rcu_assign_pointer(*pprev, obj_new);
1198 err = 0;
1199 break;
1200 }
1201
1202 spin_unlock_bh(lock);
1203
1204 return err;
1205}
1206
1207/**
1208 * rhashtable_replace_fast - replace an object in hash table
1209 * @ht: hash table
1210 * @obj_old: pointer to hash head inside object being replaced
1211 * @obj_new: pointer to hash head inside object which is new
1212 * @params: hash table parameters
1213 *
1214 * Replacing an object doesn't affect the number of elements in the hash table
1215 * or bucket, so we don't need to worry about shrinking or expanding the
1216 * table here.
1217 *
1218 * Returns zero on success, -ENOENT if the entry could not be found,
1219 * -EINVAL if hash is not the same for the old and new objects.
1220 */
1221static inline int rhashtable_replace_fast(
1222 struct rhashtable *ht, struct rhash_head *obj_old,
1223 struct rhash_head *obj_new,
1224 const struct rhashtable_params params)
1225{
1226 struct bucket_table *tbl;
1227 int err;
1228
1229 rcu_read_lock();
1230
1231 tbl = rht_dereference_rcu(ht->tbl, ht);
1232
1233 /* Because we have already taken (and released) the bucket
1234 * lock in old_tbl, if we find that future_tbl is not yet
1235 * visible then that guarantees the entry to still be in
1236 * the old tbl if it exists.
1237 */
1238 while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1239 obj_new, params)) &&
1240 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1241 ;
1242
1243 rcu_read_unlock();
1244
1245 return err;
1246}
1247
Herbert Xu246779d2016-08-18 16:50:56 +08001248/* Obsolete function, do not use in new code. */
1249static inline int rhashtable_walk_init(struct rhashtable *ht,
1250 struct rhashtable_iter *iter, gfp_t gfp)
1251{
1252 rhashtable_walk_enter(ht, iter);
1253 return 0;
1254}
1255
Herbert Xuca268932016-09-19 19:00:09 +08001256/**
1257 * rhltable_walk_enter - Initialise an iterator
1258 * @hlt: Table to walk over
1259 * @iter: Hash table Iterator
1260 *
1261 * This function prepares a hash table walk.
1262 *
1263 * Note that if you restart a walk after rhashtable_walk_stop you
1264 * may see the same object twice. Also, you may miss objects if
1265 * there are removals in between rhashtable_walk_stop and the next
1266 * call to rhashtable_walk_start.
1267 *
1268 * For a completely stable walk you should construct your own data
1269 * structure outside the hash table.
1270 *
NeilBrown82266e92018-04-24 08:29:13 +10001271 * This function may be called from any process context, including
1272 * non-preemptable context, but cannot be called from softirq or
1273 * hardirq context.
Herbert Xuca268932016-09-19 19:00:09 +08001274 *
1275 * You must call rhashtable_walk_exit after this function returns.
1276 */
1277static inline void rhltable_walk_enter(struct rhltable *hlt,
1278 struct rhashtable_iter *iter)
1279{
1280 return rhashtable_walk_enter(&hlt->ht, iter);
1281}
1282
1283/**
1284 * rhltable_free_and_destroy - free elements and destroy hash list table
1285 * @hlt: the hash list table to destroy
1286 * @free_fn: callback to release resources of element
1287 * @arg: pointer passed to free_fn
1288 *
1289 * See documentation for rhashtable_free_and_destroy.
1290 */
1291static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1292 void (*free_fn)(void *ptr,
1293 void *arg),
1294 void *arg)
1295{
1296 return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1297}
1298
1299static inline void rhltable_destroy(struct rhltable *hlt)
1300{
1301 return rhltable_free_and_destroy(hlt, NULL, NULL);
1302}
1303
Thomas Graf7e1e7762014-08-02 11:47:44 +02001304#endif /* _LINUX_RHASHTABLE_H */