blob: e5c8586cf7174cfe0526dc8fb3314676601c5e57 [file] [log] [blame]
Thomas Graf7e1e7762014-08-02 11:47:44 +02001/*
2 * Resizable, Scalable, Concurrent Hash Table
3 *
Herbert Xu02fd97c2015-03-20 21:57:00 +11004 * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au>
Thomas Grafa5ec68e2015-02-05 02:03:32 +01005 * 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 Xu02fd97c2015-03-20 21:57:00 +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
Herbert Xu07ee0722015-05-15 11:30:47 +080017#include <linux/atomic.h>
Thomas Graf7e1e7762014-08-02 11:47:44 +020018#include <linux/kernel.h>
19#include <linux/init.h>
20#include <linux/log2.h>
Eric Dumazet5beb5c92015-02-26 07:20:34 -080021#include <linux/sched.h>
Ingo Molnarb2d09102017-02-04 01:27:20 +010022#include <linux/rculist.h>
Thomas Graf7e1e7762014-08-02 11:47:44 +020023#include <linux/slab.h>
24#include <linux/vmalloc.h>
25#include <linux/mm.h>
Daniel Borkmann87545892014-12-10 16:33:11 +010026#include <linux/jhash.h>
Thomas Graf7e1e7762014-08-02 11:47:44 +020027#include <linux/random.h>
28#include <linux/rhashtable.h>
Stephen Rothwell61d7b092015-02-09 14:04:03 +110029#include <linux/err.h>
Hauke Mehrtens6d795412015-06-06 22:07:23 +020030#include <linux/export.h>
Thomas Graf7e1e7762014-08-02 11:47:44 +020031
32#define HASH_DEFAULT_SIZE 64UL
Herbert Xuc2e213c2015-03-18 20:01:16 +110033#define HASH_MIN_SIZE 4U
Florian Westphal4cf0b352016-08-12 12:03:52 +020034#define BUCKET_LOCKS_PER_CPU 32UL
Thomas Graf97defe12015-01-02 23:00:20 +010035
Herbert Xuda204202017-02-11 19:26:47 +080036union nested_table {
37 union nested_table __rcu *table;
38 struct rhash_head __rcu *bucket;
39};
40
Herbert Xu988dfbd2015-03-10 09:27:55 +110041static u32 head_hashfn(struct rhashtable *ht,
Thomas Graf8d24c0b2015-01-02 23:00:14 +010042 const struct bucket_table *tbl,
43 const struct rhash_head *he)
Thomas Graf7e1e7762014-08-02 11:47:44 +020044{
Herbert Xu02fd97c2015-03-20 21:57:00 +110045 return rht_head_hashfn(ht, tbl, he, ht->p);
Thomas Graf7e1e7762014-08-02 11:47:44 +020046}
47
Thomas Grafa03eaec2015-02-05 02:03:34 +010048#ifdef CONFIG_PROVE_LOCKING
Thomas Grafa03eaec2015-02-05 02:03:34 +010049#define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
Thomas Grafa03eaec2015-02-05 02:03:34 +010050
51int lockdep_rht_mutex_is_held(struct rhashtable *ht)
52{
53 return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
54}
55EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
56
57int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
58{
Herbert Xu02fd97c2015-03-20 21:57:00 +110059 spinlock_t *lock = rht_bucket_lock(tbl, hash);
Thomas Grafa03eaec2015-02-05 02:03:34 +010060
61 return (debug_locks) ? lockdep_is_held(lock) : 1;
62}
63EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
64#else
65#define ASSERT_RHT_MUTEX(HT)
Thomas Grafa03eaec2015-02-05 02:03:34 +010066#endif
67
Herbert Xuda204202017-02-11 19:26:47 +080068static void nested_table_free(union nested_table *ntbl, unsigned int size)
69{
70 const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
71 const unsigned int len = 1 << shift;
72 unsigned int i;
73
74 ntbl = rcu_dereference_raw(ntbl->table);
75 if (!ntbl)
76 return;
77
78 if (size > len) {
79 size >>= shift;
80 for (i = 0; i < len; i++)
81 nested_table_free(ntbl + i, size);
82 }
83
84 kfree(ntbl);
85}
86
87static void nested_bucket_table_free(const struct bucket_table *tbl)
88{
89 unsigned int size = tbl->size >> tbl->nest;
90 unsigned int len = 1 << tbl->nest;
91 union nested_table *ntbl;
92 unsigned int i;
93
94 ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
95
96 for (i = 0; i < len; i++)
97 nested_table_free(ntbl + i, size);
98
99 kfree(ntbl);
100}
101
Thomas Graf97defe12015-01-02 23:00:20 +0100102static void bucket_table_free(const struct bucket_table *tbl)
103{
Herbert Xuda204202017-02-11 19:26:47 +0800104 if (tbl->nest)
105 nested_bucket_table_free(tbl);
106
Tom Herbert64e0cd02017-12-04 10:31:45 -0800107 free_bucket_spinlocks(tbl->locks);
Thomas Graf97defe12015-01-02 23:00:20 +0100108 kvfree(tbl);
109}
110
Herbert Xu9d901bc2015-03-14 13:57:23 +1100111static void bucket_table_free_rcu(struct rcu_head *head)
112{
113 bucket_table_free(container_of(head, struct bucket_table, rcu));
114}
115
Herbert Xuda204202017-02-11 19:26:47 +0800116static union nested_table *nested_table_alloc(struct rhashtable *ht,
117 union nested_table __rcu **prev,
118 unsigned int shifted,
119 unsigned int nhash)
120{
121 union nested_table *ntbl;
122 int i;
123
124 ntbl = rcu_dereference(*prev);
125 if (ntbl)
126 return ntbl;
127
128 ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC);
129
130 if (ntbl && shifted) {
131 for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0].bucket); i++)
132 INIT_RHT_NULLS_HEAD(ntbl[i].bucket, ht,
133 (i << shifted) | nhash);
134 }
135
136 rcu_assign_pointer(*prev, ntbl);
137
138 return ntbl;
139}
140
141static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht,
142 size_t nbuckets,
143 gfp_t gfp)
144{
145 const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
146 struct bucket_table *tbl;
147 size_t size;
148
149 if (nbuckets < (1 << (shift + 1)))
150 return NULL;
151
152 size = sizeof(*tbl) + sizeof(tbl->buckets[0]);
153
154 tbl = kzalloc(size, gfp);
155 if (!tbl)
156 return NULL;
157
158 if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets,
159 0, 0)) {
160 kfree(tbl);
161 return NULL;
162 }
163
164 tbl->nest = (ilog2(nbuckets) - 1) % shift + 1;
165
166 return tbl;
167}
168
Thomas Graf97defe12015-01-02 23:00:20 +0100169static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
Herbert Xub9ecfda2015-03-24 00:50:27 +1100170 size_t nbuckets,
171 gfp_t gfp)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200172{
Daniel Borkmanneb6d1ab2015-02-20 00:53:38 +0100173 struct bucket_table *tbl = NULL;
Tom Herbert64e0cd02017-12-04 10:31:45 -0800174 size_t size, max_locks;
Thomas Graff89bd6f2015-01-02 23:00:21 +0100175 int i;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200176
177 size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
Michal Hocko12e8fd62017-07-10 15:51:55 -0700178 if (gfp != GFP_KERNEL)
Herbert Xub9ecfda2015-03-24 00:50:27 +1100179 tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
Michal Hocko12e8fd62017-07-10 15:51:55 -0700180 else
181 tbl = kvzalloc(size, gfp);
Herbert Xuda204202017-02-11 19:26:47 +0800182
183 size = nbuckets;
184
185 if (tbl == NULL && gfp != GFP_KERNEL) {
186 tbl = nested_bucket_table_alloc(ht, nbuckets, gfp);
187 nbuckets = 0;
188 }
Thomas Graf7e1e7762014-08-02 11:47:44 +0200189 if (tbl == NULL)
190 return NULL;
191
Herbert Xuda204202017-02-11 19:26:47 +0800192 tbl->size = size;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200193
Tom Herbert64e0cd02017-12-04 10:31:45 -0800194 max_locks = size >> 1;
195 if (tbl->nest)
196 max_locks = min_t(size_t, max_locks, 1U << tbl->nest);
197
198 if (alloc_bucket_spinlocks(&tbl->locks, &tbl->locks_mask, max_locks,
199 ht->p.locks_mul, gfp) < 0) {
Thomas Graf97defe12015-01-02 23:00:20 +0100200 bucket_table_free(tbl);
201 return NULL;
202 }
Thomas Graf7e1e7762014-08-02 11:47:44 +0200203
Herbert Xueddee5ba2015-03-14 13:57:20 +1100204 INIT_LIST_HEAD(&tbl->walkers);
205
Jason A. Donenfeldd48ad082017-06-07 22:47:13 -0400206 tbl->hash_rnd = get_random_u32();
Herbert Xu5269b532015-03-14 13:57:22 +1100207
Thomas Graff89bd6f2015-01-02 23:00:21 +0100208 for (i = 0; i < nbuckets; i++)
209 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
210
Thomas Graf97defe12015-01-02 23:00:20 +0100211 return tbl;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200212}
213
Herbert Xub8244782015-03-24 00:50:26 +1100214static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
215 struct bucket_table *tbl)
216{
217 struct bucket_table *new_tbl;
218
219 do {
220 new_tbl = tbl;
221 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
222 } while (tbl);
223
224 return new_tbl;
225}
226
Thomas Graf299e5c32015-03-24 14:18:17 +0100227static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
Thomas Grafa5ec68e2015-02-05 02:03:32 +0100228{
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100229 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
Herbert Xub8244782015-03-24 00:50:26 +1100230 struct bucket_table *new_tbl = rhashtable_last_table(ht,
231 rht_dereference_rcu(old_tbl->future_tbl, ht));
Herbert Xuda204202017-02-11 19:26:47 +0800232 struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash);
233 int err = -EAGAIN;
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100234 struct rhash_head *head, *next, *entry;
235 spinlock_t *new_bucket_lock;
Thomas Graf299e5c32015-03-24 14:18:17 +0100236 unsigned int new_hash;
Thomas Grafa5ec68e2015-02-05 02:03:32 +0100237
Herbert Xuda204202017-02-11 19:26:47 +0800238 if (new_tbl->nest)
239 goto out;
240
241 err = -ENOENT;
242
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100243 rht_for_each(entry, old_tbl, old_hash) {
244 err = 0;
245 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
Thomas Grafa5ec68e2015-02-05 02:03:32 +0100246
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100247 if (rht_is_a_nulls(next))
Thomas Graf7e1e7762014-08-02 11:47:44 +0200248 break;
Thomas Graf97defe12015-01-02 23:00:20 +0100249
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100250 pprev = &entry->next;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200251 }
252
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100253 if (err)
254 goto out;
Thomas Graf97defe12015-01-02 23:00:20 +0100255
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100256 new_hash = head_hashfn(ht, new_tbl, entry);
Thomas Grafa5ec68e2015-02-05 02:03:32 +0100257
Herbert Xu02fd97c2015-03-20 21:57:00 +1100258 new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100259
Herbert Xu8f2484b2015-03-14 13:57:21 +1100260 spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100261 head = rht_dereference_bucket(new_tbl->buckets[new_hash],
262 new_tbl, new_hash);
263
Dmitriy Vyukov7def0f92015-09-22 10:51:52 +0200264 RCU_INIT_POINTER(entry->next, head);
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100265
266 rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
267 spin_unlock(new_bucket_lock);
268
269 rcu_assign_pointer(*pprev, next);
270
271out:
272 return err;
Thomas Graf97defe12015-01-02 23:00:20 +0100273}
274
Herbert Xuda204202017-02-11 19:26:47 +0800275static int rhashtable_rehash_chain(struct rhashtable *ht,
Thomas Graf299e5c32015-03-24 14:18:17 +0100276 unsigned int old_hash)
Thomas Graf97defe12015-01-02 23:00:20 +0100277{
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100278 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
279 spinlock_t *old_bucket_lock;
Herbert Xuda204202017-02-11 19:26:47 +0800280 int err;
Thomas Graf7cd10db2015-02-05 02:03:35 +0100281
Herbert Xu02fd97c2015-03-20 21:57:00 +1100282 old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100283
284 spin_lock_bh(old_bucket_lock);
Herbert Xuda204202017-02-11 19:26:47 +0800285 while (!(err = rhashtable_rehash_one(ht, old_hash)))
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100286 ;
Herbert Xuda204202017-02-11 19:26:47 +0800287
288 if (err == -ENOENT) {
289 old_tbl->rehash++;
290 err = 0;
291 }
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100292 spin_unlock_bh(old_bucket_lock);
Herbert Xuda204202017-02-11 19:26:47 +0800293
294 return err;
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100295}
296
Herbert Xub8244782015-03-24 00:50:26 +1100297static int rhashtable_rehash_attach(struct rhashtable *ht,
298 struct bucket_table *old_tbl,
299 struct bucket_table *new_tbl)
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100300{
Herbert Xub8244782015-03-24 00:50:26 +1100301 /* Protect future_tbl using the first bucket lock. */
302 spin_lock_bh(old_tbl->locks);
303
304 /* Did somebody beat us to it? */
305 if (rcu_access_pointer(old_tbl->future_tbl)) {
306 spin_unlock_bh(old_tbl->locks);
307 return -EEXIST;
308 }
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100309
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100310 /* Make insertions go into the new, empty table right away. Deletions
311 * and lookups will be attempted in both tables until we synchronize.
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100312 */
Herbert Xuc4db8842015-03-14 13:57:25 +1100313 rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100314
Herbert Xub8244782015-03-24 00:50:26 +1100315 spin_unlock_bh(old_tbl->locks);
316
317 return 0;
318}
319
320static int rhashtable_rehash_table(struct rhashtable *ht)
321{
322 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
323 struct bucket_table *new_tbl;
324 struct rhashtable_walker *walker;
Thomas Graf299e5c32015-03-24 14:18:17 +0100325 unsigned int old_hash;
Herbert Xuda204202017-02-11 19:26:47 +0800326 int err;
Herbert Xub8244782015-03-24 00:50:26 +1100327
328 new_tbl = rht_dereference(old_tbl->future_tbl, ht);
329 if (!new_tbl)
330 return 0;
331
Herbert Xuda204202017-02-11 19:26:47 +0800332 for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
333 err = rhashtable_rehash_chain(ht, old_hash);
334 if (err)
335 return err;
Eric Dumazetae6da1f2018-03-31 12:58:48 -0700336 cond_resched();
Herbert Xuda204202017-02-11 19:26:47 +0800337 }
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100338
339 /* Publish the new table pointer. */
340 rcu_assign_pointer(ht->tbl, new_tbl);
341
Herbert Xuba7c95e2015-03-24 09:53:17 +1100342 spin_lock(&ht->lock);
Herbert Xueddee5ba2015-03-14 13:57:20 +1100343 list_for_each_entry(walker, &old_tbl->walkers, list)
344 walker->tbl = NULL;
Herbert Xuba7c95e2015-03-24 09:53:17 +1100345 spin_unlock(&ht->lock);
Herbert Xueddee5ba2015-03-14 13:57:20 +1100346
Herbert Xuaa34a6cb02015-03-11 09:43:48 +1100347 /* Wait for readers. All new readers will see the new
348 * table, and thus no references to the old table will
349 * remain.
350 */
Herbert Xu9d901bc2015-03-14 13:57:23 +1100351 call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
Herbert Xub8244782015-03-24 00:50:26 +1100352
353 return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200354}
355
Herbert Xuda204202017-02-11 19:26:47 +0800356static int rhashtable_rehash_alloc(struct rhashtable *ht,
357 struct bucket_table *old_tbl,
358 unsigned int size)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200359{
Herbert Xuda204202017-02-11 19:26:47 +0800360 struct bucket_table *new_tbl;
Herbert Xub8244782015-03-24 00:50:26 +1100361 int err;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200362
363 ASSERT_RHT_MUTEX(ht);
364
Herbert Xuda204202017-02-11 19:26:47 +0800365 new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200366 if (new_tbl == NULL)
367 return -ENOMEM;
368
Herbert Xub8244782015-03-24 00:50:26 +1100369 err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
370 if (err)
371 bucket_table_free(new_tbl);
372
373 return err;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200374}
Thomas Graf7e1e7762014-08-02 11:47:44 +0200375
376/**
377 * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
378 * @ht: the hash table to shrink
Thomas Graf7e1e7762014-08-02 11:47:44 +0200379 *
Herbert Xu18093d12015-03-24 00:50:25 +1100380 * This function shrinks the hash table to fit, i.e., the smallest
381 * size would not cause it to expand right away automatically.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200382 *
Thomas Graf97defe12015-01-02 23:00:20 +0100383 * The caller must ensure that no concurrent resizing occurs by holding
384 * ht->mutex.
385 *
Thomas Graf7e1e7762014-08-02 11:47:44 +0200386 * The caller must ensure that no concurrent table mutations take place.
387 * It is however valid to have concurrent lookups if they are RCU protected.
Thomas Graf97defe12015-01-02 23:00:20 +0100388 *
389 * It is valid to have concurrent insertions and deletions protected by per
390 * bucket locks or concurrent RCU protected lookups and traversals.
Thomas Graf7e1e7762014-08-02 11:47:44 +0200391 */
Herbert Xub8244782015-03-24 00:50:26 +1100392static int rhashtable_shrink(struct rhashtable *ht)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200393{
Herbert Xuda204202017-02-11 19:26:47 +0800394 struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
Vegard Nossum12311952016-08-12 20:10:44 +0200395 unsigned int nelems = atomic_read(&ht->nelems);
396 unsigned int size = 0;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200397
Vegard Nossum12311952016-08-12 20:10:44 +0200398 if (nelems)
399 size = roundup_pow_of_two(nelems * 3 / 2);
Herbert Xu18093d12015-03-24 00:50:25 +1100400 if (size < ht->p.min_size)
401 size = ht->p.min_size;
402
403 if (old_tbl->size <= size)
404 return 0;
405
Herbert Xub8244782015-03-24 00:50:26 +1100406 if (rht_dereference(old_tbl->future_tbl, ht))
407 return -EEXIST;
408
Herbert Xuda204202017-02-11 19:26:47 +0800409 return rhashtable_rehash_alloc(ht, old_tbl, size);
Thomas Graf7e1e7762014-08-02 11:47:44 +0200410}
Thomas Graf7e1e7762014-08-02 11:47:44 +0200411
Thomas Graf97defe12015-01-02 23:00:20 +0100412static void rht_deferred_worker(struct work_struct *work)
413{
414 struct rhashtable *ht;
415 struct bucket_table *tbl;
Herbert Xub8244782015-03-24 00:50:26 +1100416 int err = 0;
Thomas Graf97defe12015-01-02 23:00:20 +0100417
Ying Xue57699a42015-01-16 11:13:09 +0800418 ht = container_of(work, struct rhashtable, run_work);
Thomas Graf97defe12015-01-02 23:00:20 +0100419 mutex_lock(&ht->mutex);
Herbert Xu28134a52015-02-04 07:33:22 +1100420
Thomas Graf97defe12015-01-02 23:00:20 +0100421 tbl = rht_dereference(ht->tbl, ht);
Herbert Xub8244782015-03-24 00:50:26 +1100422 tbl = rhashtable_last_table(ht, tbl);
Thomas Graf97defe12015-01-02 23:00:20 +0100423
Daniel Borkmanna5b68462015-03-12 15:28:40 +0100424 if (rht_grow_above_75(ht, tbl))
Herbert Xuda204202017-02-11 19:26:47 +0800425 err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2);
Thomas Grafb5e2c152015-03-24 20:42:19 +0000426 else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
Herbert Xuda204202017-02-11 19:26:47 +0800427 err = rhashtable_shrink(ht);
428 else if (tbl->nest)
429 err = rhashtable_rehash_alloc(ht, tbl, tbl->size);
Herbert Xub8244782015-03-24 00:50:26 +1100430
Herbert Xuda204202017-02-11 19:26:47 +0800431 if (!err)
432 err = rhashtable_rehash_table(ht);
Herbert Xub8244782015-03-24 00:50:26 +1100433
Thomas Graf97defe12015-01-02 23:00:20 +0100434 mutex_unlock(&ht->mutex);
Herbert Xub8244782015-03-24 00:50:26 +1100435
436 if (err)
437 schedule_work(&ht->run_work);
Thomas Graf97defe12015-01-02 23:00:20 +0100438}
439
Herbert Xuca268932016-09-19 19:00:09 +0800440static int rhashtable_insert_rehash(struct rhashtable *ht,
441 struct bucket_table *tbl)
Herbert Xuccd57b12015-03-24 00:50:28 +1100442{
443 struct bucket_table *old_tbl;
444 struct bucket_table *new_tbl;
Herbert Xuccd57b12015-03-24 00:50:28 +1100445 unsigned int size;
446 int err;
447
448 old_tbl = rht_dereference_rcu(ht->tbl, ht);
Herbert Xuccd57b12015-03-24 00:50:28 +1100449
450 size = tbl->size;
451
Herbert Xu3cf92222015-12-03 20:41:29 +0800452 err = -EBUSY;
453
Herbert Xuccd57b12015-03-24 00:50:28 +1100454 if (rht_grow_above_75(ht, tbl))
455 size *= 2;
Thomas Grafa87b9eb2015-04-22 09:41:46 +0200456 /* Do not schedule more than one rehash */
457 else if (old_tbl != tbl)
Herbert Xu3cf92222015-12-03 20:41:29 +0800458 goto fail;
459
460 err = -ENOMEM;
Herbert Xuccd57b12015-03-24 00:50:28 +1100461
462 new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
Herbert Xu3cf92222015-12-03 20:41:29 +0800463 if (new_tbl == NULL)
464 goto fail;
Herbert Xuccd57b12015-03-24 00:50:28 +1100465
466 err = rhashtable_rehash_attach(ht, tbl, new_tbl);
467 if (err) {
468 bucket_table_free(new_tbl);
469 if (err == -EEXIST)
470 err = 0;
471 } else
472 schedule_work(&ht->run_work);
473
474 return err;
Herbert Xu3cf92222015-12-03 20:41:29 +0800475
476fail:
477 /* Do not fail the insert if someone else did a rehash. */
478 if (likely(rcu_dereference_raw(tbl->future_tbl)))
479 return 0;
480
481 /* Schedule async rehash to retry allocation in process context. */
482 if (err == -ENOMEM)
483 schedule_work(&ht->run_work);
484
485 return err;
Herbert Xuccd57b12015-03-24 00:50:28 +1100486}
Herbert Xuccd57b12015-03-24 00:50:28 +1100487
Herbert Xuca268932016-09-19 19:00:09 +0800488static void *rhashtable_lookup_one(struct rhashtable *ht,
489 struct bucket_table *tbl, unsigned int hash,
490 const void *key, struct rhash_head *obj)
Herbert Xu02fd97c2015-03-20 21:57:00 +1100491{
Herbert Xuca268932016-09-19 19:00:09 +0800492 struct rhashtable_compare_arg arg = {
493 .ht = ht,
494 .key = key,
495 };
496 struct rhash_head __rcu **pprev;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100497 struct rhash_head *head;
Herbert Xuca268932016-09-19 19:00:09 +0800498 int elasticity;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100499
Florian Westphal5f8ddea2017-04-16 02:55:09 +0200500 elasticity = RHT_ELASTICITY;
Herbert Xuda204202017-02-11 19:26:47 +0800501 pprev = rht_bucket_var(tbl, hash);
502 rht_for_each_continue(head, *pprev, tbl, hash) {
Herbert Xuca268932016-09-19 19:00:09 +0800503 struct rhlist_head *list;
504 struct rhlist_head *plist;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100505
Herbert Xuca268932016-09-19 19:00:09 +0800506 elasticity--;
507 if (!key ||
508 (ht->p.obj_cmpfn ?
509 ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
Paul Blakeyd3dcf8e2018-03-04 17:29:48 +0200510 rhashtable_compare(&arg, rht_obj(ht, head)))) {
511 pprev = &head->next;
Herbert Xuca268932016-09-19 19:00:09 +0800512 continue;
Paul Blakeyd3dcf8e2018-03-04 17:29:48 +0200513 }
Herbert Xuca268932016-09-19 19:00:09 +0800514
515 if (!ht->rhlist)
516 return rht_obj(ht, head);
517
518 list = container_of(obj, struct rhlist_head, rhead);
519 plist = container_of(head, struct rhlist_head, rhead);
520
521 RCU_INIT_POINTER(list->next, plist);
522 head = rht_dereference_bucket(head->next, tbl, hash);
523 RCU_INIT_POINTER(list->rhead.next, head);
524 rcu_assign_pointer(*pprev, obj);
525
526 return NULL;
Pablo Neira Ayuso5ca8cc52016-08-24 12:31:31 +0200527 }
Herbert Xu02fd97c2015-03-20 21:57:00 +1100528
Herbert Xuca268932016-09-19 19:00:09 +0800529 if (elasticity <= 0)
530 return ERR_PTR(-EAGAIN);
531
532 return ERR_PTR(-ENOENT);
533}
534
535static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
536 struct bucket_table *tbl,
537 unsigned int hash,
538 struct rhash_head *obj,
539 void *data)
540{
Herbert Xuda204202017-02-11 19:26:47 +0800541 struct rhash_head __rcu **pprev;
Herbert Xuca268932016-09-19 19:00:09 +0800542 struct bucket_table *new_tbl;
543 struct rhash_head *head;
544
545 if (!IS_ERR_OR_NULL(data))
546 return ERR_PTR(-EEXIST);
547
548 if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
549 return ERR_CAST(data);
550
551 new_tbl = rcu_dereference(tbl->future_tbl);
552 if (new_tbl)
553 return new_tbl;
554
555 if (PTR_ERR(data) != -ENOENT)
556 return ERR_CAST(data);
557
Herbert Xu07ee0722015-05-15 11:30:47 +0800558 if (unlikely(rht_grow_above_max(ht, tbl)))
Herbert Xuca268932016-09-19 19:00:09 +0800559 return ERR_PTR(-E2BIG);
Herbert Xu07ee0722015-05-15 11:30:47 +0800560
Herbert Xuca268932016-09-19 19:00:09 +0800561 if (unlikely(rht_grow_above_100(ht, tbl)))
562 return ERR_PTR(-EAGAIN);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100563
Herbert Xuda204202017-02-11 19:26:47 +0800564 pprev = rht_bucket_insert(ht, tbl, hash);
565 if (!pprev)
566 return ERR_PTR(-ENOMEM);
567
568 head = rht_dereference_bucket(*pprev, tbl, hash);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100569
570 RCU_INIT_POINTER(obj->next, head);
Herbert Xuca268932016-09-19 19:00:09 +0800571 if (ht->rhlist) {
572 struct rhlist_head *list;
573
574 list = container_of(obj, struct rhlist_head, rhead);
575 RCU_INIT_POINTER(list->next, NULL);
576 }
Herbert Xu02fd97c2015-03-20 21:57:00 +1100577
Herbert Xuda204202017-02-11 19:26:47 +0800578 rcu_assign_pointer(*pprev, obj);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100579
580 atomic_inc(&ht->nelems);
Herbert Xuca268932016-09-19 19:00:09 +0800581 if (rht_grow_above_75(ht, tbl))
582 schedule_work(&ht->run_work);
Herbert Xu02fd97c2015-03-20 21:57:00 +1100583
Herbert Xuca268932016-09-19 19:00:09 +0800584 return NULL;
585}
Herbert Xu02fd97c2015-03-20 21:57:00 +1100586
Herbert Xuca268932016-09-19 19:00:09 +0800587static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
588 struct rhash_head *obj)
589{
590 struct bucket_table *new_tbl;
591 struct bucket_table *tbl;
592 unsigned int hash;
593 spinlock_t *lock;
594 void *data;
595
596 tbl = rcu_dereference(ht->tbl);
597
598 /* All insertions must grab the oldest table containing
599 * the hashed bucket that is yet to be rehashed.
600 */
601 for (;;) {
602 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
603 lock = rht_bucket_lock(tbl, hash);
604 spin_lock_bh(lock);
605
606 if (tbl->rehash <= hash)
607 break;
608
609 spin_unlock_bh(lock);
610 tbl = rcu_dereference(tbl->future_tbl);
611 }
612
613 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
614 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
615 if (PTR_ERR(new_tbl) != -EEXIST)
616 data = ERR_CAST(new_tbl);
617
618 while (!IS_ERR_OR_NULL(new_tbl)) {
619 tbl = new_tbl;
620 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
621 spin_lock_nested(rht_bucket_lock(tbl, hash),
622 SINGLE_DEPTH_NESTING);
623
624 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
625 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
626 if (PTR_ERR(new_tbl) != -EEXIST)
627 data = ERR_CAST(new_tbl);
628
629 spin_unlock(rht_bucket_lock(tbl, hash));
630 }
631
632 spin_unlock_bh(lock);
633
634 if (PTR_ERR(data) == -EAGAIN)
635 data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
636 -EAGAIN);
637
638 return data;
639}
640
641void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
642 struct rhash_head *obj)
643{
644 void *data;
645
646 do {
647 rcu_read_lock();
648 data = rhashtable_try_insert(ht, key, obj);
649 rcu_read_unlock();
650 } while (PTR_ERR(data) == -EAGAIN);
651
652 return data;
Herbert Xu02fd97c2015-03-20 21:57:00 +1100653}
654EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
655
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100656/**
Herbert Xu246779d2016-08-18 16:50:56 +0800657 * rhashtable_walk_enter - Initialise an iterator
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100658 * @ht: Table to walk over
659 * @iter: Hash table Iterator
660 *
661 * This function prepares a hash table walk.
662 *
663 * Note that if you restart a walk after rhashtable_walk_stop you
664 * may see the same object twice. Also, you may miss objects if
665 * there are removals in between rhashtable_walk_stop and the next
666 * call to rhashtable_walk_start.
667 *
668 * For a completely stable walk you should construct your own data
669 * structure outside the hash table.
670 *
NeilBrown82266e92018-04-24 08:29:13 +1000671 * This function may be called from any process context, including
672 * non-preemptable context, but cannot be called from softirq or
673 * hardirq context.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100674 *
Herbert Xu246779d2016-08-18 16:50:56 +0800675 * You must call rhashtable_walk_exit after this function returns.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100676 */
Herbert Xu246779d2016-08-18 16:50:56 +0800677void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100678{
679 iter->ht = ht;
680 iter->p = NULL;
681 iter->slot = 0;
682 iter->skip = 0;
Tom Herbert2db54b42017-12-04 10:31:42 -0800683 iter->end_of_table = 0;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100684
Herbert Xuc6ff5262015-12-16 16:45:54 +0800685 spin_lock(&ht->lock);
Herbert Xu246779d2016-08-18 16:50:56 +0800686 iter->walker.tbl =
Herbert Xu179ccc02015-12-19 10:45:28 +0800687 rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
Herbert Xu246779d2016-08-18 16:50:56 +0800688 list_add(&iter->walker.list, &iter->walker.tbl->walkers);
Herbert Xuc6ff5262015-12-16 16:45:54 +0800689 spin_unlock(&ht->lock);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100690}
Herbert Xu246779d2016-08-18 16:50:56 +0800691EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100692
693/**
694 * rhashtable_walk_exit - Free an iterator
695 * @iter: Hash table Iterator
696 *
697 * This function frees resources allocated by rhashtable_walk_init.
698 */
699void rhashtable_walk_exit(struct rhashtable_iter *iter)
700{
Herbert Xuc6ff5262015-12-16 16:45:54 +0800701 spin_lock(&iter->ht->lock);
Herbert Xu246779d2016-08-18 16:50:56 +0800702 if (iter->walker.tbl)
703 list_del(&iter->walker.list);
Herbert Xuc6ff5262015-12-16 16:45:54 +0800704 spin_unlock(&iter->ht->lock);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100705}
706EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
707
708/**
Tom Herbert97a6ec42017-12-04 10:31:41 -0800709 * rhashtable_walk_start_check - Start a hash table walk
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100710 * @iter: Hash table iterator
711 *
Andreas Gruenbacher06471692017-09-19 12:41:37 +0200712 * Start a hash table walk at the current iterator position. Note that we take
713 * the RCU lock in all cases including when we return an error. So you must
714 * always call rhashtable_walk_stop to clean up.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100715 *
716 * Returns zero if successful.
717 *
718 * Returns -EAGAIN if resize event occured. Note that the iterator
719 * will rewind back to the beginning and you may use it immediately
720 * by calling rhashtable_walk_next.
Tom Herbert97a6ec42017-12-04 10:31:41 -0800721 *
722 * rhashtable_walk_start is defined as an inline variant that returns
723 * void. This is preferred in cases where the caller would ignore
724 * resize events and always continue.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100725 */
Tom Herbert97a6ec42017-12-04 10:31:41 -0800726int rhashtable_walk_start_check(struct rhashtable_iter *iter)
Thomas Grafdb4374f2015-03-16 10:42:27 +0100727 __acquires(RCU)
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100728{
Herbert Xueddee5ba2015-03-14 13:57:20 +1100729 struct rhashtable *ht = iter->ht;
NeilBrown5d240a82018-04-24 08:29:13 +1000730 bool rhlist = ht->rhlist;
Herbert Xueddee5ba2015-03-14 13:57:20 +1100731
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100732 rcu_read_lock();
733
Herbert Xuc6ff5262015-12-16 16:45:54 +0800734 spin_lock(&ht->lock);
Herbert Xu246779d2016-08-18 16:50:56 +0800735 if (iter->walker.tbl)
736 list_del(&iter->walker.list);
Herbert Xuc6ff5262015-12-16 16:45:54 +0800737 spin_unlock(&ht->lock);
Herbert Xueddee5ba2015-03-14 13:57:20 +1100738
NeilBrown5d240a82018-04-24 08:29:13 +1000739 if (iter->end_of_table)
740 return 0;
741 if (!iter->walker.tbl) {
Herbert Xu246779d2016-08-18 16:50:56 +0800742 iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
NeilBrownb41cc042018-04-24 08:29:13 +1000743 iter->slot = 0;
744 iter->skip = 0;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100745 return -EAGAIN;
746 }
747
NeilBrown5d240a82018-04-24 08:29:13 +1000748 if (iter->p && !rhlist) {
749 /*
750 * We need to validate that 'p' is still in the table, and
751 * if so, update 'skip'
752 */
753 struct rhash_head *p;
754 int skip = 0;
755 rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
756 skip++;
757 if (p == iter->p) {
758 iter->skip = skip;
759 goto found;
760 }
761 }
762 iter->p = NULL;
763 } else if (iter->p && rhlist) {
764 /* Need to validate that 'list' is still in the table, and
765 * if so, update 'skip' and 'p'.
766 */
767 struct rhash_head *p;
768 struct rhlist_head *list;
769 int skip = 0;
770 rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
771 for (list = container_of(p, struct rhlist_head, rhead);
772 list;
773 list = rcu_dereference(list->next)) {
774 skip++;
775 if (list == iter->list) {
776 iter->p = p;
Rishabh Bhatnagarc643ecf2018-07-02 09:35:34 -0700777 iter->skip = skip;
NeilBrown5d240a82018-04-24 08:29:13 +1000778 goto found;
779 }
780 }
781 }
782 iter->p = NULL;
783 }
784found:
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100785 return 0;
786}
Tom Herbert97a6ec42017-12-04 10:31:41 -0800787EXPORT_SYMBOL_GPL(rhashtable_walk_start_check);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100788
789/**
Tom Herbert2db54b42017-12-04 10:31:42 -0800790 * __rhashtable_walk_find_next - Find the next element in a table (or the first
791 * one in case of a new walk).
792 *
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100793 * @iter: Hash table iterator
794 *
Tom Herbert2db54b42017-12-04 10:31:42 -0800795 * Returns the found object or NULL when the end of the table is reached.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100796 *
Tom Herbert2db54b42017-12-04 10:31:42 -0800797 * Returns -EAGAIN if resize event occurred.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100798 */
Tom Herbert2db54b42017-12-04 10:31:42 -0800799static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter)
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100800{
Herbert Xu246779d2016-08-18 16:50:56 +0800801 struct bucket_table *tbl = iter->walker.tbl;
Herbert Xuca268932016-09-19 19:00:09 +0800802 struct rhlist_head *list = iter->list;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100803 struct rhashtable *ht = iter->ht;
804 struct rhash_head *p = iter->p;
Herbert Xuca268932016-09-19 19:00:09 +0800805 bool rhlist = ht->rhlist;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100806
Tom Herbert2db54b42017-12-04 10:31:42 -0800807 if (!tbl)
808 return NULL;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100809
810 for (; iter->slot < tbl->size; iter->slot++) {
811 int skip = iter->skip;
812
813 rht_for_each_rcu(p, tbl, iter->slot) {
Herbert Xuca268932016-09-19 19:00:09 +0800814 if (rhlist) {
815 list = container_of(p, struct rhlist_head,
816 rhead);
817 do {
818 if (!skip)
819 goto next;
820 skip--;
821 list = rcu_dereference(list->next);
822 } while (list);
823
824 continue;
825 }
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100826 if (!skip)
827 break;
828 skip--;
829 }
830
831next:
832 if (!rht_is_a_nulls(p)) {
833 iter->skip++;
834 iter->p = p;
Herbert Xuca268932016-09-19 19:00:09 +0800835 iter->list = list;
836 return rht_obj(ht, rhlist ? &list->rhead : p);
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100837 }
838
839 iter->skip = 0;
840 }
841
Phil Sutter142b9422015-07-06 15:51:20 +0200842 iter->p = NULL;
843
Herbert Xud88252f2015-03-24 00:50:19 +1100844 /* Ensure we see any new tables. */
845 smp_rmb();
846
Herbert Xu246779d2016-08-18 16:50:56 +0800847 iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht);
848 if (iter->walker.tbl) {
Herbert Xueddee5ba2015-03-14 13:57:20 +1100849 iter->slot = 0;
850 iter->skip = 0;
851 return ERR_PTR(-EAGAIN);
Tom Herbert2db54b42017-12-04 10:31:42 -0800852 } else {
853 iter->end_of_table = true;
Herbert Xueddee5ba2015-03-14 13:57:20 +1100854 }
855
Thomas Grafc936a792015-05-05 02:22:53 +0200856 return NULL;
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100857}
Tom Herbert2db54b42017-12-04 10:31:42 -0800858
859/**
860 * rhashtable_walk_next - Return the next object and advance the iterator
861 * @iter: Hash table iterator
862 *
863 * Note that you must call rhashtable_walk_stop when you are finished
864 * with the walk.
865 *
866 * Returns the next object or NULL when the end of the table is reached.
867 *
868 * Returns -EAGAIN if resize event occurred. Note that the iterator
869 * will rewind back to the beginning and you may continue to use it.
870 */
871void *rhashtable_walk_next(struct rhashtable_iter *iter)
872{
873 struct rhlist_head *list = iter->list;
874 struct rhashtable *ht = iter->ht;
875 struct rhash_head *p = iter->p;
876 bool rhlist = ht->rhlist;
877
878 if (p) {
879 if (!rhlist || !(list = rcu_dereference(list->next))) {
880 p = rcu_dereference(p->next);
881 list = container_of(p, struct rhlist_head, rhead);
882 }
883 if (!rht_is_a_nulls(p)) {
884 iter->skip++;
885 iter->p = p;
886 iter->list = list;
887 return rht_obj(ht, rhlist ? &list->rhead : p);
888 }
889
890 /* At the end of this slot, switch to next one and then find
891 * next entry from that point.
892 */
893 iter->skip = 0;
894 iter->slot++;
895 }
896
897 return __rhashtable_walk_find_next(iter);
898}
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100899EXPORT_SYMBOL_GPL(rhashtable_walk_next);
900
901/**
Tom Herbert2db54b42017-12-04 10:31:42 -0800902 * rhashtable_walk_peek - Return the next object but don't advance the iterator
903 * @iter: Hash table iterator
904 *
905 * Returns the next object or NULL when the end of the table is reached.
906 *
907 * Returns -EAGAIN if resize event occurred. Note that the iterator
908 * will rewind back to the beginning and you may continue to use it.
909 */
910void *rhashtable_walk_peek(struct rhashtable_iter *iter)
911{
912 struct rhlist_head *list = iter->list;
913 struct rhashtable *ht = iter->ht;
914 struct rhash_head *p = iter->p;
915
916 if (p)
917 return rht_obj(ht, ht->rhlist ? &list->rhead : p);
918
919 /* No object found in current iter, find next one in the table. */
920
921 if (iter->skip) {
922 /* A nonzero skip value points to the next entry in the table
923 * beyond that last one that was found. Decrement skip so
924 * we find the current value. __rhashtable_walk_find_next
925 * will restore the original value of skip assuming that
926 * the table hasn't changed.
927 */
928 iter->skip--;
929 }
930
931 return __rhashtable_walk_find_next(iter);
932}
933EXPORT_SYMBOL_GPL(rhashtable_walk_peek);
934
935/**
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100936 * rhashtable_walk_stop - Finish a hash table walk
937 * @iter: Hash table iterator
938 *
Andreas Gruenbacher06471692017-09-19 12:41:37 +0200939 * Finish a hash table walk. Does not reset the iterator to the start of the
940 * hash table.
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100941 */
942void rhashtable_walk_stop(struct rhashtable_iter *iter)
Thomas Grafdb4374f2015-03-16 10:42:27 +0100943 __releases(RCU)
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100944{
Herbert Xueddee5ba2015-03-14 13:57:20 +1100945 struct rhashtable *ht;
Herbert Xu246779d2016-08-18 16:50:56 +0800946 struct bucket_table *tbl = iter->walker.tbl;
Herbert Xueddee5ba2015-03-14 13:57:20 +1100947
Herbert Xueddee5ba2015-03-14 13:57:20 +1100948 if (!tbl)
Herbert Xu963ecbd2015-03-15 21:12:04 +1100949 goto out;
Herbert Xueddee5ba2015-03-14 13:57:20 +1100950
951 ht = iter->ht;
952
Herbert Xuba7c95e2015-03-24 09:53:17 +1100953 spin_lock(&ht->lock);
Herbert Xuc4db8842015-03-14 13:57:25 +1100954 if (tbl->rehash < tbl->size)
Herbert Xu246779d2016-08-18 16:50:56 +0800955 list_add(&iter->walker.list, &tbl->walkers);
Herbert Xueddee5ba2015-03-14 13:57:20 +1100956 else
Herbert Xu246779d2016-08-18 16:50:56 +0800957 iter->walker.tbl = NULL;
Herbert Xuba7c95e2015-03-24 09:53:17 +1100958 spin_unlock(&ht->lock);
Herbert Xueddee5ba2015-03-14 13:57:20 +1100959
Herbert Xu963ecbd2015-03-15 21:12:04 +1100960out:
961 rcu_read_unlock();
Herbert Xuf2dba9c2015-02-04 07:33:23 +1100962}
963EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
964
Herbert Xu488fb86e2015-03-20 21:56:59 +1100965static size_t rounded_hashtable_size(const struct rhashtable_params *params)
Thomas Graf7e1e7762014-08-02 11:47:44 +0200966{
Davidlohr Bueso107d01f2018-07-16 13:26:13 -0700967 size_t retsize;
968
969 if (params->nelem_hint)
970 retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
971 (unsigned long)params->min_size);
972 else
973 retsize = max(HASH_DEFAULT_SIZE,
974 (unsigned long)params->min_size);
975
976 return retsize;
Thomas Graf7e1e7762014-08-02 11:47:44 +0200977}
978
Herbert Xu31ccde22015-03-24 00:50:21 +1100979static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
980{
981 return jhash2(key, length, seed);
982}
983
Thomas Graf7e1e7762014-08-02 11:47:44 +0200984/**
985 * rhashtable_init - initialize a new hash table
986 * @ht: hash table to be initialized
987 * @params: configuration parameters
988 *
989 * Initializes a new hash table based on the provided configuration
990 * parameters. A table can be configured either with a variable or
991 * fixed length key:
992 *
993 * Configuration Example 1: Fixed length keys
994 * struct test_obj {
995 * int key;
996 * void * my_member;
997 * struct rhash_head node;
998 * };
999 *
1000 * struct rhashtable_params params = {
1001 * .head_offset = offsetof(struct test_obj, node),
1002 * .key_offset = offsetof(struct test_obj, key),
1003 * .key_len = sizeof(int),
Daniel Borkmann87545892014-12-10 16:33:11 +01001004 * .hashfn = jhash,
Thomas Graff89bd6f2015-01-02 23:00:21 +01001005 * .nulls_base = (1U << RHT_BASE_SHIFT),
Thomas Graf7e1e7762014-08-02 11:47:44 +02001006 * };
1007 *
1008 * Configuration Example 2: Variable length keys
1009 * struct test_obj {
1010 * [...]
1011 * struct rhash_head node;
1012 * };
1013 *
Patrick McHardy49f7b332015-03-25 13:07:45 +00001014 * u32 my_hash_fn(const void *data, u32 len, u32 seed)
Thomas Graf7e1e7762014-08-02 11:47:44 +02001015 * {
1016 * struct test_obj *obj = data;
1017 *
1018 * return [... hash ...];
1019 * }
1020 *
1021 * struct rhashtable_params params = {
1022 * .head_offset = offsetof(struct test_obj, node),
Daniel Borkmann87545892014-12-10 16:33:11 +01001023 * .hashfn = jhash,
Thomas Graf7e1e7762014-08-02 11:47:44 +02001024 * .obj_hashfn = my_hash_fn,
Thomas Graf7e1e7762014-08-02 11:47:44 +02001025 * };
1026 */
Herbert Xu488fb86e2015-03-20 21:56:59 +11001027int rhashtable_init(struct rhashtable *ht,
1028 const struct rhashtable_params *params)
Thomas Graf7e1e7762014-08-02 11:47:44 +02001029{
1030 struct bucket_table *tbl;
1031 size_t size;
1032
Herbert Xu31ccde22015-03-24 00:50:21 +11001033 if ((!params->key_len && !params->obj_hashfn) ||
Herbert Xu02fd97c2015-03-20 21:57:00 +11001034 (params->obj_hashfn && !params->obj_cmpfn))
Thomas Graf7e1e7762014-08-02 11:47:44 +02001035 return -EINVAL;
1036
Thomas Graff89bd6f2015-01-02 23:00:21 +01001037 if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
1038 return -EINVAL;
1039
Thomas Graf97defe12015-01-02 23:00:20 +01001040 memset(ht, 0, sizeof(*ht));
1041 mutex_init(&ht->mutex);
Herbert Xuba7c95e2015-03-24 09:53:17 +11001042 spin_lock_init(&ht->lock);
Thomas Graf97defe12015-01-02 23:00:20 +01001043 memcpy(&ht->p, params, sizeof(*params));
1044
Thomas Grafa998f712015-03-19 22:31:13 +00001045 if (params->min_size)
1046 ht->p.min_size = roundup_pow_of_two(params->min_size);
1047
Herbert Xu6d684e52017-04-27 13:44:51 +08001048 /* Cap total entries at 2^31 to avoid nelems overflow. */
1049 ht->max_elems = 1u << 31;
Herbert Xu2d2ab652017-04-28 14:10:48 +08001050
1051 if (params->max_size) {
1052 ht->p.max_size = rounddown_pow_of_two(params->max_size);
1053 if (ht->p.max_size < ht->max_elems / 2)
1054 ht->max_elems = ht->p.max_size * 2;
1055 }
Herbert Xu6d684e52017-04-27 13:44:51 +08001056
Florian Westphal48e75b432017-05-01 22:18:01 +02001057 ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
Thomas Grafa998f712015-03-19 22:31:13 +00001058
Davidlohr Bueso107d01f2018-07-16 13:26:13 -07001059 size = rounded_hashtable_size(&ht->p);
Herbert Xu3a324602015-12-16 18:13:14 +08001060
Thomas Graf97defe12015-01-02 23:00:20 +01001061 if (params->locks_mul)
1062 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
1063 else
1064 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
1065
Herbert Xu31ccde22015-03-24 00:50:21 +11001066 ht->key_len = ht->p.key_len;
1067 if (!params->hashfn) {
1068 ht->p.hashfn = jhash;
1069
1070 if (!(ht->key_len & (sizeof(u32) - 1))) {
1071 ht->key_len /= sizeof(u32);
1072 ht->p.hashfn = rhashtable_jhash2;
1073 }
1074 }
1075
Herbert Xub9ecfda2015-03-24 00:50:27 +11001076 tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
Thomas Graf7e1e7762014-08-02 11:47:44 +02001077 if (tbl == NULL)
1078 return -ENOMEM;
1079
Ying Xue545a1482015-01-07 13:41:57 +08001080 atomic_set(&ht->nelems, 0);
Daniel Borkmanna5b68462015-03-12 15:28:40 +01001081
Thomas Graf7e1e7762014-08-02 11:47:44 +02001082 RCU_INIT_POINTER(ht->tbl, tbl);
1083
Daniel Borkmann4c4b52d2015-02-25 16:31:54 +01001084 INIT_WORK(&ht->run_work, rht_deferred_worker);
Thomas Graf97defe12015-01-02 23:00:20 +01001085
Thomas Graf7e1e7762014-08-02 11:47:44 +02001086 return 0;
1087}
1088EXPORT_SYMBOL_GPL(rhashtable_init);
1089
1090/**
Herbert Xuca268932016-09-19 19:00:09 +08001091 * rhltable_init - initialize a new hash list table
1092 * @hlt: hash list table to be initialized
1093 * @params: configuration parameters
1094 *
1095 * Initializes a new hash list table.
1096 *
1097 * See documentation for rhashtable_init.
1098 */
1099int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
1100{
1101 int err;
1102
1103 /* No rhlist NULLs marking for now. */
1104 if (params->nulls_base)
1105 return -EINVAL;
1106
1107 err = rhashtable_init(&hlt->ht, params);
1108 hlt->ht.rhlist = true;
1109 return err;
1110}
1111EXPORT_SYMBOL_GPL(rhltable_init);
1112
1113static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
1114 void (*free_fn)(void *ptr, void *arg),
1115 void *arg)
1116{
1117 struct rhlist_head *list;
1118
1119 if (!ht->rhlist) {
1120 free_fn(rht_obj(ht, obj), arg);
1121 return;
1122 }
1123
1124 list = container_of(obj, struct rhlist_head, rhead);
1125 do {
1126 obj = &list->rhead;
1127 list = rht_dereference(list->next, ht);
1128 free_fn(rht_obj(ht, obj), arg);
1129 } while (list);
1130}
1131
1132/**
Thomas Graf6b6f3022015-03-24 14:18:20 +01001133 * rhashtable_free_and_destroy - free elements and destroy hash table
Thomas Graf7e1e7762014-08-02 11:47:44 +02001134 * @ht: the hash table to destroy
Thomas Graf6b6f3022015-03-24 14:18:20 +01001135 * @free_fn: callback to release resources of element
1136 * @arg: pointer passed to free_fn
Thomas Graf7e1e7762014-08-02 11:47:44 +02001137 *
Thomas Graf6b6f3022015-03-24 14:18:20 +01001138 * Stops an eventual async resize. If defined, invokes free_fn for each
1139 * element to releasal resources. Please note that RCU protected
1140 * readers may still be accessing the elements. Releasing of resources
1141 * must occur in a compatible manner. Then frees the bucket array.
1142 *
1143 * This function will eventually sleep to wait for an async resize
1144 * to complete. The caller is responsible that no further write operations
1145 * occurs in parallel.
Thomas Graf7e1e7762014-08-02 11:47:44 +02001146 */
Thomas Graf6b6f3022015-03-24 14:18:20 +01001147void rhashtable_free_and_destroy(struct rhashtable *ht,
1148 void (*free_fn)(void *ptr, void *arg),
1149 void *arg)
Thomas Graf7e1e7762014-08-02 11:47:44 +02001150{
Taehee Yoo00261292018-07-08 11:55:51 +09001151 struct bucket_table *tbl, *next_tbl;
Thomas Graf6b6f3022015-03-24 14:18:20 +01001152 unsigned int i;
Thomas Graf97defe12015-01-02 23:00:20 +01001153
Daniel Borkmann4c4b52d2015-02-25 16:31:54 +01001154 cancel_work_sync(&ht->run_work);
Ying Xue57699a42015-01-16 11:13:09 +08001155
Thomas Graf97defe12015-01-02 23:00:20 +01001156 mutex_lock(&ht->mutex);
Thomas Graf6b6f3022015-03-24 14:18:20 +01001157 tbl = rht_dereference(ht->tbl, ht);
Taehee Yoo00261292018-07-08 11:55:51 +09001158restart:
Thomas Graf6b6f3022015-03-24 14:18:20 +01001159 if (free_fn) {
1160 for (i = 0; i < tbl->size; i++) {
1161 struct rhash_head *pos, *next;
1162
Eric Dumazetae6da1f2018-03-31 12:58:48 -07001163 cond_resched();
Herbert Xuda204202017-02-11 19:26:47 +08001164 for (pos = rht_dereference(*rht_bucket(tbl, i), ht),
Thomas Graf6b6f3022015-03-24 14:18:20 +01001165 next = !rht_is_a_nulls(pos) ?
1166 rht_dereference(pos->next, ht) : NULL;
1167 !rht_is_a_nulls(pos);
1168 pos = next,
1169 next = !rht_is_a_nulls(pos) ?
1170 rht_dereference(pos->next, ht) : NULL)
Herbert Xuca268932016-09-19 19:00:09 +08001171 rhashtable_free_one(ht, pos, free_fn, arg);
Thomas Graf6b6f3022015-03-24 14:18:20 +01001172 }
1173 }
1174
Taehee Yoo00261292018-07-08 11:55:51 +09001175 next_tbl = rht_dereference(tbl->future_tbl, ht);
Thomas Graf6b6f3022015-03-24 14:18:20 +01001176 bucket_table_free(tbl);
Taehee Yoo00261292018-07-08 11:55:51 +09001177 if (next_tbl) {
1178 tbl = next_tbl;
1179 goto restart;
1180 }
Thomas Graf97defe12015-01-02 23:00:20 +01001181 mutex_unlock(&ht->mutex);
Thomas Graf7e1e7762014-08-02 11:47:44 +02001182}
Thomas Graf6b6f3022015-03-24 14:18:20 +01001183EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
1184
1185void rhashtable_destroy(struct rhashtable *ht)
1186{
1187 return rhashtable_free_and_destroy(ht, NULL, NULL);
1188}
Thomas Graf7e1e7762014-08-02 11:47:44 +02001189EXPORT_SYMBOL_GPL(rhashtable_destroy);
Herbert Xuda204202017-02-11 19:26:47 +08001190
1191struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
1192 unsigned int hash)
1193{
1194 const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1195 static struct rhash_head __rcu *rhnull =
1196 (struct rhash_head __rcu *)NULLS_MARKER(0);
1197 unsigned int index = hash & ((1 << tbl->nest) - 1);
1198 unsigned int size = tbl->size >> tbl->nest;
1199 unsigned int subhash = hash;
1200 union nested_table *ntbl;
1201
1202 ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
Herbert Xuc4d26032017-02-25 22:39:50 +08001203 ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash);
Herbert Xuda204202017-02-11 19:26:47 +08001204 subhash >>= tbl->nest;
1205
1206 while (ntbl && size > (1 << shift)) {
1207 index = subhash & ((1 << shift) - 1);
Herbert Xuc4d26032017-02-25 22:39:50 +08001208 ntbl = rht_dereference_bucket_rcu(ntbl[index].table,
1209 tbl, hash);
Herbert Xuda204202017-02-11 19:26:47 +08001210 size >>= shift;
1211 subhash >>= shift;
1212 }
1213
1214 if (!ntbl)
1215 return &rhnull;
1216
1217 return &ntbl[subhash].bucket;
1218
1219}
1220EXPORT_SYMBOL_GPL(rht_bucket_nested);
1221
1222struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
1223 struct bucket_table *tbl,
1224 unsigned int hash)
1225{
1226 const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1227 unsigned int index = hash & ((1 << tbl->nest) - 1);
1228 unsigned int size = tbl->size >> tbl->nest;
1229 union nested_table *ntbl;
1230 unsigned int shifted;
1231 unsigned int nhash;
1232
1233 ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1234 hash >>= tbl->nest;
1235 nhash = index;
1236 shifted = tbl->nest;
1237 ntbl = nested_table_alloc(ht, &ntbl[index].table,
1238 size <= (1 << shift) ? shifted : 0, nhash);
1239
1240 while (ntbl && size > (1 << shift)) {
1241 index = hash & ((1 << shift) - 1);
1242 size >>= shift;
1243 hash >>= shift;
1244 nhash |= index << shifted;
1245 shifted += shift;
1246 ntbl = nested_table_alloc(ht, &ntbl[index].table,
1247 size <= (1 << shift) ? shifted : 0,
1248 nhash);
1249 }
1250
1251 if (!ntbl)
1252 return NULL;
1253
1254 return &ntbl[hash].bucket;
1255
1256}
1257EXPORT_SYMBOL_GPL(rht_bucket_nested_insert);