]>
Commit | Line | Data |
---|---|---|
7e1e7763 TG |
1 | /* |
2 | * Resizable, Scalable, Concurrent Hash Table | |
3 | * | |
a5ec68e3 | 4 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> |
7e1e7763 TG |
5 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> |
6 | * | |
7 | * Based on the following paper: | |
8 | * https://www.usenix.org/legacy/event/atc11/tech/final_files/Triplett.pdf | |
9 | * | |
10 | * Code partially derived from nft_hash | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or modify | |
13 | * it under the terms of the GNU General Public License version 2 as | |
14 | * published by the Free Software Foundation. | |
15 | */ | |
16 | ||
17 | #include <linux/kernel.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/log2.h> | |
5beb5c90 | 20 | #include <linux/sched.h> |
7e1e7763 TG |
21 | #include <linux/slab.h> |
22 | #include <linux/vmalloc.h> | |
23 | #include <linux/mm.h> | |
87545899 | 24 | #include <linux/jhash.h> |
7e1e7763 TG |
25 | #include <linux/random.h> |
26 | #include <linux/rhashtable.h> | |
61d7b097 | 27 | #include <linux/err.h> |
7e1e7763 TG |
28 | |
29 | #define HASH_DEFAULT_SIZE 64UL | |
30 | #define HASH_MIN_SIZE 4UL | |
97defe1e TG |
31 | #define BUCKET_LOCKS_PER_CPU 128UL |
32 | ||
f89bd6f8 TG |
33 | /* Base bits plus 1 bit for nulls marker */ |
34 | #define HASH_RESERVED_SPACE (RHT_BASE_BITS + 1) | |
35 | ||
97defe1e TG |
36 | enum { |
37 | RHT_LOCK_NORMAL, | |
38 | RHT_LOCK_NESTED, | |
97defe1e TG |
39 | }; |
40 | ||
41 | /* The bucket lock is selected based on the hash and protects mutations | |
42 | * on a group of hash buckets. | |
43 | * | |
a5ec68e3 TG |
44 | * A maximum of tbl->size/2 bucket locks is allocated. This ensures that |
45 | * a single lock always covers both buckets which may both contains | |
46 | * entries which link to the same bucket of the old table during resizing. | |
47 | * This allows to simplify the locking as locking the bucket in both | |
48 | * tables during resize always guarantee protection. | |
49 | * | |
97defe1e TG |
50 | * IMPORTANT: When holding the bucket lock of both the old and new table |
51 | * during expansions and shrinking, the old bucket lock must always be | |
52 | * acquired first. | |
53 | */ | |
54 | static spinlock_t *bucket_lock(const struct bucket_table *tbl, u32 hash) | |
55 | { | |
56 | return &tbl->locks[hash & tbl->locks_mask]; | |
57 | } | |
7e1e7763 | 58 | |
c91eee56 | 59 | static void *rht_obj(const struct rhashtable *ht, const struct rhash_head *he) |
7e1e7763 TG |
60 | { |
61 | return (void *) he - ht->p.head_offset; | |
62 | } | |
7e1e7763 | 63 | |
8d24c0b4 | 64 | static u32 rht_bucket_index(const struct bucket_table *tbl, u32 hash) |
7e1e7763 | 65 | { |
8d24c0b4 | 66 | return hash & (tbl->size - 1); |
7e1e7763 | 67 | } |
7e1e7763 | 68 | |
8d24c0b4 | 69 | static u32 obj_raw_hashfn(const struct rhashtable *ht, const void *ptr) |
7e1e7763 | 70 | { |
8d24c0b4 | 71 | u32 hash; |
7e1e7763 | 72 | |
8d24c0b4 TG |
73 | if (unlikely(!ht->p.key_len)) |
74 | hash = ht->p.obj_hashfn(ptr, ht->p.hash_rnd); | |
75 | else | |
76 | hash = ht->p.hashfn(ptr + ht->p.key_offset, ht->p.key_len, | |
77 | ht->p.hash_rnd); | |
7e1e7763 | 78 | |
f89bd6f8 | 79 | return hash >> HASH_RESERVED_SPACE; |
7e1e7763 TG |
80 | } |
81 | ||
97defe1e | 82 | static u32 key_hashfn(struct rhashtable *ht, const void *key, u32 len) |
7e1e7763 | 83 | { |
c88455ce | 84 | return ht->p.hashfn(key, len, ht->p.hash_rnd) >> HASH_RESERVED_SPACE; |
7e1e7763 | 85 | } |
7e1e7763 TG |
86 | |
87 | static u32 head_hashfn(const struct rhashtable *ht, | |
8d24c0b4 TG |
88 | const struct bucket_table *tbl, |
89 | const struct rhash_head *he) | |
7e1e7763 | 90 | { |
8d24c0b4 | 91 | return rht_bucket_index(tbl, obj_raw_hashfn(ht, rht_obj(ht, he))); |
7e1e7763 TG |
92 | } |
93 | ||
a03eaec0 TG |
94 | #ifdef CONFIG_PROVE_LOCKING |
95 | static void debug_dump_buckets(const struct rhashtable *ht, | |
96 | const struct bucket_table *tbl) | |
97 | { | |
98 | struct rhash_head *he; | |
99 | unsigned int i, hash; | |
100 | ||
101 | for (i = 0; i < tbl->size; i++) { | |
102 | pr_warn(" [Bucket %d] ", i); | |
103 | rht_for_each_rcu(he, tbl, i) { | |
104 | hash = head_hashfn(ht, tbl, he); | |
105 | pr_cont("[hash = %#x, lock = %p] ", | |
106 | hash, bucket_lock(tbl, hash)); | |
107 | } | |
108 | pr_cont("\n"); | |
109 | } | |
110 | ||
111 | } | |
112 | ||
113 | static void debug_dump_table(struct rhashtable *ht, | |
114 | const struct bucket_table *tbl, | |
115 | unsigned int hash) | |
116 | { | |
117 | struct bucket_table *old_tbl, *future_tbl; | |
118 | ||
119 | pr_emerg("BUG: lock for hash %#x in table %p not held\n", | |
120 | hash, tbl); | |
121 | ||
122 | rcu_read_lock(); | |
123 | future_tbl = rht_dereference_rcu(ht->future_tbl, ht); | |
124 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
125 | if (future_tbl != old_tbl) { | |
126 | pr_warn("Future table %p (size: %zd)\n", | |
127 | future_tbl, future_tbl->size); | |
128 | debug_dump_buckets(ht, future_tbl); | |
129 | } | |
130 | ||
131 | pr_warn("Table %p (size: %zd)\n", old_tbl, old_tbl->size); | |
132 | debug_dump_buckets(ht, old_tbl); | |
133 | ||
134 | rcu_read_unlock(); | |
135 | } | |
136 | ||
137 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) | |
138 | #define ASSERT_BUCKET_LOCK(HT, TBL, HASH) \ | |
139 | do { \ | |
140 | if (unlikely(!lockdep_rht_bucket_is_held(TBL, HASH))) { \ | |
141 | debug_dump_table(HT, TBL, HASH); \ | |
142 | BUG(); \ | |
143 | } \ | |
144 | } while (0) | |
145 | ||
146 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) | |
147 | { | |
148 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; | |
149 | } | |
150 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); | |
151 | ||
152 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) | |
153 | { | |
154 | spinlock_t *lock = bucket_lock(tbl, hash); | |
155 | ||
156 | return (debug_locks) ? lockdep_is_held(lock) : 1; | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); | |
159 | #else | |
160 | #define ASSERT_RHT_MUTEX(HT) | |
161 | #define ASSERT_BUCKET_LOCK(HT, TBL, HASH) | |
162 | #endif | |
163 | ||
164 | ||
b8e1943e TG |
165 | static struct rhash_head __rcu **bucket_tail(struct bucket_table *tbl, u32 n) |
166 | { | |
167 | struct rhash_head __rcu **pprev; | |
168 | ||
169 | for (pprev = &tbl->buckets[n]; | |
f89bd6f8 | 170 | !rht_is_a_nulls(rht_dereference_bucket(*pprev, tbl, n)); |
b8e1943e TG |
171 | pprev = &rht_dereference_bucket(*pprev, tbl, n)->next) |
172 | ; | |
173 | ||
174 | return pprev; | |
175 | } | |
176 | ||
97defe1e TG |
177 | static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl) |
178 | { | |
179 | unsigned int i, size; | |
180 | #if defined(CONFIG_PROVE_LOCKING) | |
181 | unsigned int nr_pcpus = 2; | |
182 | #else | |
183 | unsigned int nr_pcpus = num_possible_cpus(); | |
184 | #endif | |
185 | ||
186 | nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL); | |
187 | size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul); | |
188 | ||
a5ec68e3 TG |
189 | /* Never allocate more than 0.5 locks per bucket */ |
190 | size = min_t(unsigned int, size, tbl->size >> 1); | |
97defe1e TG |
191 | |
192 | if (sizeof(spinlock_t) != 0) { | |
193 | #ifdef CONFIG_NUMA | |
194 | if (size * sizeof(spinlock_t) > PAGE_SIZE) | |
195 | tbl->locks = vmalloc(size * sizeof(spinlock_t)); | |
196 | else | |
197 | #endif | |
198 | tbl->locks = kmalloc_array(size, sizeof(spinlock_t), | |
199 | GFP_KERNEL); | |
200 | if (!tbl->locks) | |
201 | return -ENOMEM; | |
202 | for (i = 0; i < size; i++) | |
203 | spin_lock_init(&tbl->locks[i]); | |
204 | } | |
205 | tbl->locks_mask = size - 1; | |
206 | ||
207 | return 0; | |
208 | } | |
209 | ||
210 | static void bucket_table_free(const struct bucket_table *tbl) | |
211 | { | |
212 | if (tbl) | |
213 | kvfree(tbl->locks); | |
214 | ||
215 | kvfree(tbl); | |
216 | } | |
217 | ||
218 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, | |
219 | size_t nbuckets) | |
7e1e7763 | 220 | { |
eb6d1abf | 221 | struct bucket_table *tbl = NULL; |
7e1e7763 | 222 | size_t size; |
f89bd6f8 | 223 | int i; |
7e1e7763 TG |
224 | |
225 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); | |
eb6d1abf DB |
226 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) |
227 | tbl = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); | |
7e1e7763 TG |
228 | if (tbl == NULL) |
229 | tbl = vzalloc(size); | |
7e1e7763 TG |
230 | if (tbl == NULL) |
231 | return NULL; | |
232 | ||
233 | tbl->size = nbuckets; | |
234 | ||
97defe1e TG |
235 | if (alloc_bucket_locks(ht, tbl) < 0) { |
236 | bucket_table_free(tbl); | |
237 | return NULL; | |
238 | } | |
7e1e7763 | 239 | |
f89bd6f8 TG |
240 | for (i = 0; i < nbuckets; i++) |
241 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); | |
242 | ||
97defe1e | 243 | return tbl; |
7e1e7763 TG |
244 | } |
245 | ||
246 | /** | |
247 | * rht_grow_above_75 - returns true if nelems > 0.75 * table-size | |
248 | * @ht: hash table | |
249 | * @new_size: new table size | |
250 | */ | |
4c4b52d9 | 251 | static bool rht_grow_above_75(const struct rhashtable *ht, size_t new_size) |
7e1e7763 TG |
252 | { |
253 | /* Expand table when exceeding 75% load */ | |
c0c09bfd | 254 | return atomic_read(&ht->nelems) > (new_size / 4 * 3) && |
8331de75 | 255 | (!ht->p.max_shift || atomic_read(&ht->shift) < ht->p.max_shift); |
7e1e7763 | 256 | } |
7e1e7763 TG |
257 | |
258 | /** | |
259 | * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size | |
260 | * @ht: hash table | |
261 | * @new_size: new table size | |
262 | */ | |
4c4b52d9 | 263 | static bool rht_shrink_below_30(const struct rhashtable *ht, size_t new_size) |
7e1e7763 TG |
264 | { |
265 | /* Shrink table beneath 30% load */ | |
c0c09bfd YX |
266 | return atomic_read(&ht->nelems) < (new_size * 3 / 10) && |
267 | (atomic_read(&ht->shift) > ht->p.min_shift); | |
7e1e7763 | 268 | } |
7e1e7763 | 269 | |
a5ec68e3 TG |
270 | static void lock_buckets(struct bucket_table *new_tbl, |
271 | struct bucket_table *old_tbl, unsigned int hash) | |
272 | __acquires(old_bucket_lock) | |
273 | { | |
274 | spin_lock_bh(bucket_lock(old_tbl, hash)); | |
275 | if (new_tbl != old_tbl) | |
276 | spin_lock_bh_nested(bucket_lock(new_tbl, hash), | |
277 | RHT_LOCK_NESTED); | |
278 | } | |
279 | ||
280 | static void unlock_buckets(struct bucket_table *new_tbl, | |
281 | struct bucket_table *old_tbl, unsigned int hash) | |
282 | __releases(old_bucket_lock) | |
283 | { | |
284 | if (new_tbl != old_tbl) | |
285 | spin_unlock_bh(bucket_lock(new_tbl, hash)); | |
286 | spin_unlock_bh(bucket_lock(old_tbl, hash)); | |
287 | } | |
288 | ||
289 | /** | |
290 | * Unlink entries on bucket which hash to different bucket. | |
291 | * | |
292 | * Returns true if no more work needs to be performed on the bucket. | |
293 | */ | |
a03eaec0 | 294 | static bool hashtable_chain_unzip(struct rhashtable *ht, |
7e1e7763 | 295 | const struct bucket_table *new_tbl, |
97defe1e TG |
296 | struct bucket_table *old_tbl, |
297 | size_t old_hash) | |
7e1e7763 TG |
298 | { |
299 | struct rhash_head *he, *p, *next; | |
97defe1e TG |
300 | unsigned int new_hash, new_hash2; |
301 | ||
a03eaec0 | 302 | ASSERT_BUCKET_LOCK(ht, old_tbl, old_hash); |
7e1e7763 TG |
303 | |
304 | /* Old bucket empty, no work needed. */ | |
97defe1e TG |
305 | p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl, |
306 | old_hash); | |
f89bd6f8 | 307 | if (rht_is_a_nulls(p)) |
a5ec68e3 | 308 | return false; |
7e1e7763 | 309 | |
a5ec68e3 | 310 | new_hash = head_hashfn(ht, new_tbl, p); |
a03eaec0 | 311 | ASSERT_BUCKET_LOCK(ht, new_tbl, new_hash); |
97defe1e | 312 | |
7e1e7763 TG |
313 | /* Advance the old bucket pointer one or more times until it |
314 | * reaches a node that doesn't hash to the same bucket as the | |
315 | * previous node p. Call the previous node p; | |
316 | */ | |
97defe1e TG |
317 | rht_for_each_continue(he, p->next, old_tbl, old_hash) { |
318 | new_hash2 = head_hashfn(ht, new_tbl, he); | |
a03eaec0 | 319 | ASSERT_BUCKET_LOCK(ht, new_tbl, new_hash2); |
a5ec68e3 | 320 | |
97defe1e | 321 | if (new_hash != new_hash2) |
7e1e7763 TG |
322 | break; |
323 | p = he; | |
324 | } | |
97defe1e TG |
325 | rcu_assign_pointer(old_tbl->buckets[old_hash], p->next); |
326 | ||
7e1e7763 TG |
327 | /* Find the subsequent node which does hash to the same |
328 | * bucket as node P, or NULL if no such node exists. | |
329 | */ | |
f89bd6f8 TG |
330 | INIT_RHT_NULLS_HEAD(next, ht, old_hash); |
331 | if (!rht_is_a_nulls(he)) { | |
97defe1e TG |
332 | rht_for_each_continue(he, he->next, old_tbl, old_hash) { |
333 | if (head_hashfn(ht, new_tbl, he) == new_hash) { | |
7e1e7763 TG |
334 | next = he; |
335 | break; | |
336 | } | |
337 | } | |
338 | } | |
339 | ||
340 | /* Set p's next pointer to that subsequent node pointer, | |
341 | * bypassing the nodes which do not hash to p's bucket | |
342 | */ | |
97defe1e TG |
343 | rcu_assign_pointer(p->next, next); |
344 | ||
a5ec68e3 TG |
345 | p = rht_dereference_bucket(old_tbl->buckets[old_hash], old_tbl, |
346 | old_hash); | |
347 | ||
348 | return !rht_is_a_nulls(p); | |
97defe1e TG |
349 | } |
350 | ||
7cd10db8 | 351 | static void link_old_to_new(struct rhashtable *ht, struct bucket_table *new_tbl, |
97defe1e TG |
352 | unsigned int new_hash, struct rhash_head *entry) |
353 | { | |
7cd10db8 TG |
354 | ASSERT_BUCKET_LOCK(ht, new_tbl, new_hash); |
355 | ||
97defe1e | 356 | rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), entry); |
7e1e7763 TG |
357 | } |
358 | ||
359 | /** | |
360 | * rhashtable_expand - Expand hash table while allowing concurrent lookups | |
361 | * @ht: the hash table to expand | |
7e1e7763 TG |
362 | * |
363 | * A secondary bucket array is allocated and the hash entries are migrated | |
364 | * while keeping them on both lists until the end of the RCU grace period. | |
365 | * | |
366 | * This function may only be called in a context where it is safe to call | |
367 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. | |
368 | * | |
97defe1e TG |
369 | * The caller must ensure that no concurrent resizing occurs by holding |
370 | * ht->mutex. | |
371 | * | |
372 | * It is valid to have concurrent insertions and deletions protected by per | |
373 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 374 | */ |
6eba8224 | 375 | int rhashtable_expand(struct rhashtable *ht) |
7e1e7763 TG |
376 | { |
377 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); | |
378 | struct rhash_head *he; | |
97defe1e TG |
379 | unsigned int new_hash, old_hash; |
380 | bool complete = false; | |
7e1e7763 TG |
381 | |
382 | ASSERT_RHT_MUTEX(ht); | |
383 | ||
97defe1e | 384 | new_tbl = bucket_table_alloc(ht, old_tbl->size * 2); |
7e1e7763 TG |
385 | if (new_tbl == NULL) |
386 | return -ENOMEM; | |
387 | ||
c0c09bfd | 388 | atomic_inc(&ht->shift); |
7e1e7763 | 389 | |
97defe1e TG |
390 | /* Make insertions go into the new, empty table right away. Deletions |
391 | * and lookups will be attempted in both tables until we synchronize. | |
392 | * The synchronize_rcu() guarantees for the new table to be picked up | |
393 | * so no new additions go into the old table while we relink. | |
394 | */ | |
395 | rcu_assign_pointer(ht->future_tbl, new_tbl); | |
396 | synchronize_rcu(); | |
397 | ||
398 | /* For each new bucket, search the corresponding old bucket for the | |
399 | * first entry that hashes to the new bucket, and link the end of | |
400 | * newly formed bucket chain (containing entries added to future | |
401 | * table) to that entry. Since all the entries which will end up in | |
402 | * the new bucket appear in the same old bucket, this constructs an | |
403 | * entirely valid new hash table, but with multiple buckets | |
404 | * "zipped" together into a single imprecise chain. | |
7e1e7763 | 405 | */ |
97defe1e TG |
406 | for (new_hash = 0; new_hash < new_tbl->size; new_hash++) { |
407 | old_hash = rht_bucket_index(old_tbl, new_hash); | |
a5ec68e3 | 408 | lock_buckets(new_tbl, old_tbl, new_hash); |
97defe1e TG |
409 | rht_for_each(he, old_tbl, old_hash) { |
410 | if (head_hashfn(ht, new_tbl, he) == new_hash) { | |
7cd10db8 | 411 | link_old_to_new(ht, new_tbl, new_hash, he); |
7e1e7763 TG |
412 | break; |
413 | } | |
414 | } | |
a5ec68e3 | 415 | unlock_buckets(new_tbl, old_tbl, new_hash); |
5beb5c90 | 416 | cond_resched(); |
7e1e7763 TG |
417 | } |
418 | ||
7e1e7763 | 419 | /* Unzip interleaved hash chains */ |
97defe1e | 420 | while (!complete && !ht->being_destroyed) { |
7e1e7763 TG |
421 | /* Wait for readers. All new readers will see the new |
422 | * table, and thus no references to the old table will | |
423 | * remain. | |
424 | */ | |
425 | synchronize_rcu(); | |
426 | ||
427 | /* For each bucket in the old table (each of which | |
428 | * contains items from multiple buckets of the new | |
429 | * table): ... | |
430 | */ | |
431 | complete = true; | |
97defe1e | 432 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) { |
a5ec68e3 | 433 | lock_buckets(new_tbl, old_tbl, old_hash); |
97defe1e | 434 | |
a5ec68e3 TG |
435 | if (hashtable_chain_unzip(ht, new_tbl, old_tbl, |
436 | old_hash)) | |
7e1e7763 | 437 | complete = false; |
97defe1e | 438 | |
a5ec68e3 | 439 | unlock_buckets(new_tbl, old_tbl, old_hash); |
5beb5c90 | 440 | cond_resched(); |
7e1e7763 | 441 | } |
97defe1e | 442 | } |
7e1e7763 | 443 | |
cf52d52f | 444 | rcu_assign_pointer(ht->tbl, new_tbl); |
2af4b529 TG |
445 | synchronize_rcu(); |
446 | ||
7e1e7763 TG |
447 | bucket_table_free(old_tbl); |
448 | return 0; | |
449 | } | |
450 | EXPORT_SYMBOL_GPL(rhashtable_expand); | |
451 | ||
452 | /** | |
453 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups | |
454 | * @ht: the hash table to shrink | |
7e1e7763 TG |
455 | * |
456 | * This function may only be called in a context where it is safe to call | |
457 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. | |
458 | * | |
97defe1e TG |
459 | * The caller must ensure that no concurrent resizing occurs by holding |
460 | * ht->mutex. | |
461 | * | |
7e1e7763 TG |
462 | * The caller must ensure that no concurrent table mutations take place. |
463 | * It is however valid to have concurrent lookups if they are RCU protected. | |
97defe1e TG |
464 | * |
465 | * It is valid to have concurrent insertions and deletions protected by per | |
466 | * bucket locks or concurrent RCU protected lookups and traversals. | |
7e1e7763 | 467 | */ |
6eba8224 | 468 | int rhashtable_shrink(struct rhashtable *ht) |
7e1e7763 | 469 | { |
97defe1e | 470 | struct bucket_table *new_tbl, *tbl = rht_dereference(ht->tbl, ht); |
97defe1e | 471 | unsigned int new_hash; |
7e1e7763 TG |
472 | |
473 | ASSERT_RHT_MUTEX(ht); | |
474 | ||
97defe1e TG |
475 | new_tbl = bucket_table_alloc(ht, tbl->size / 2); |
476 | if (new_tbl == NULL) | |
7e1e7763 TG |
477 | return -ENOMEM; |
478 | ||
97defe1e TG |
479 | rcu_assign_pointer(ht->future_tbl, new_tbl); |
480 | synchronize_rcu(); | |
7e1e7763 | 481 | |
97defe1e TG |
482 | /* Link the first entry in the old bucket to the end of the |
483 | * bucket in the new table. As entries are concurrently being | |
484 | * added to the new table, lock down the new bucket. As we | |
485 | * always divide the size in half when shrinking, each bucket | |
486 | * in the new table maps to exactly two buckets in the old | |
487 | * table. | |
7e1e7763 | 488 | */ |
97defe1e | 489 | for (new_hash = 0; new_hash < new_tbl->size; new_hash++) { |
a5ec68e3 | 490 | lock_buckets(new_tbl, tbl, new_hash); |
97defe1e TG |
491 | |
492 | rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), | |
493 | tbl->buckets[new_hash]); | |
7cd10db8 | 494 | ASSERT_BUCKET_LOCK(ht, tbl, new_hash + new_tbl->size); |
97defe1e TG |
495 | rcu_assign_pointer(*bucket_tail(new_tbl, new_hash), |
496 | tbl->buckets[new_hash + new_tbl->size]); | |
497 | ||
a5ec68e3 | 498 | unlock_buckets(new_tbl, tbl, new_hash); |
5beb5c90 | 499 | cond_resched(); |
7e1e7763 TG |
500 | } |
501 | ||
502 | /* Publish the new, valid hash table */ | |
97defe1e | 503 | rcu_assign_pointer(ht->tbl, new_tbl); |
c0c09bfd | 504 | atomic_dec(&ht->shift); |
7e1e7763 TG |
505 | |
506 | /* Wait for readers. No new readers will have references to the | |
507 | * old hash table. | |
508 | */ | |
509 | synchronize_rcu(); | |
510 | ||
511 | bucket_table_free(tbl); | |
512 | ||
513 | return 0; | |
514 | } | |
515 | EXPORT_SYMBOL_GPL(rhashtable_shrink); | |
516 | ||
97defe1e TG |
517 | static void rht_deferred_worker(struct work_struct *work) |
518 | { | |
519 | struct rhashtable *ht; | |
520 | struct bucket_table *tbl; | |
f2dba9c6 | 521 | struct rhashtable_walker *walker; |
97defe1e | 522 | |
57699a40 | 523 | ht = container_of(work, struct rhashtable, run_work); |
97defe1e | 524 | mutex_lock(&ht->mutex); |
28134a53 HX |
525 | if (ht->being_destroyed) |
526 | goto unlock; | |
527 | ||
97defe1e TG |
528 | tbl = rht_dereference(ht->tbl, ht); |
529 | ||
f2dba9c6 HX |
530 | list_for_each_entry(walker, &ht->walkers, list) |
531 | walker->resize = true; | |
532 | ||
4c4b52d9 | 533 | if (rht_grow_above_75(ht, tbl->size)) |
97defe1e | 534 | rhashtable_expand(ht); |
4c4b52d9 | 535 | else if (rht_shrink_below_30(ht, tbl->size)) |
97defe1e | 536 | rhashtable_shrink(ht); |
28134a53 | 537 | unlock: |
97defe1e TG |
538 | mutex_unlock(&ht->mutex); |
539 | } | |
540 | ||
db304854 | 541 | static void __rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj, |
4c4b52d9 DB |
542 | struct bucket_table *tbl, |
543 | const struct bucket_table *old_tbl, u32 hash) | |
db304854 | 544 | { |
4c4b52d9 | 545 | bool no_resize_running = tbl == old_tbl; |
020219a6 TG |
546 | struct rhash_head *head; |
547 | ||
548 | hash = rht_bucket_index(tbl, hash); | |
549 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); | |
db304854 | 550 | |
7cd10db8 TG |
551 | ASSERT_BUCKET_LOCK(ht, tbl, hash); |
552 | ||
db304854 YX |
553 | if (rht_is_a_nulls(head)) |
554 | INIT_RHT_NULLS_HEAD(obj->next, ht, hash); | |
555 | else | |
556 | RCU_INIT_POINTER(obj->next, head); | |
557 | ||
558 | rcu_assign_pointer(tbl->buckets[hash], obj); | |
559 | ||
560 | atomic_inc(&ht->nelems); | |
4c4b52d9 DB |
561 | if (no_resize_running && rht_grow_above_75(ht, tbl->size)) |
562 | schedule_work(&ht->run_work); | |
db304854 YX |
563 | } |
564 | ||
7e1e7763 | 565 | /** |
db304854 | 566 | * rhashtable_insert - insert object into hash table |
7e1e7763 TG |
567 | * @ht: hash table |
568 | * @obj: pointer to hash head inside object | |
7e1e7763 | 569 | * |
97defe1e TG |
570 | * Will take a per bucket spinlock to protect against mutual mutations |
571 | * on the same bucket. Multiple insertions may occur in parallel unless | |
572 | * they map to the same bucket lock. | |
7e1e7763 | 573 | * |
97defe1e TG |
574 | * It is safe to call this function from atomic context. |
575 | * | |
576 | * Will trigger an automatic deferred table resizing if the size grows | |
577 | * beyond the watermark indicated by grow_decision() which can be passed | |
578 | * to rhashtable_init(). | |
7e1e7763 | 579 | */ |
6eba8224 | 580 | void rhashtable_insert(struct rhashtable *ht, struct rhash_head *obj) |
7e1e7763 | 581 | { |
a5ec68e3 | 582 | struct bucket_table *tbl, *old_tbl; |
97defe1e | 583 | unsigned hash; |
7e1e7763 | 584 | |
97defe1e | 585 | rcu_read_lock(); |
7e1e7763 | 586 | |
97defe1e | 587 | tbl = rht_dereference_rcu(ht->future_tbl, ht); |
a5ec68e3 | 588 | old_tbl = rht_dereference_rcu(ht->tbl, ht); |
020219a6 | 589 | hash = obj_raw_hashfn(ht, rht_obj(ht, obj)); |
97defe1e | 590 | |
a5ec68e3 | 591 | lock_buckets(tbl, old_tbl, hash); |
4c4b52d9 | 592 | __rhashtable_insert(ht, obj, tbl, old_tbl, hash); |
a5ec68e3 | 593 | unlock_buckets(tbl, old_tbl, hash); |
7e1e7763 | 594 | |
97defe1e | 595 | rcu_read_unlock(); |
7e1e7763 TG |
596 | } |
597 | EXPORT_SYMBOL_GPL(rhashtable_insert); | |
598 | ||
7e1e7763 TG |
599 | /** |
600 | * rhashtable_remove - remove object from hash table | |
601 | * @ht: hash table | |
602 | * @obj: pointer to hash head inside object | |
7e1e7763 TG |
603 | * |
604 | * Since the hash chain is single linked, the removal operation needs to | |
605 | * walk the bucket chain upon removal. The removal operation is thus | |
606 | * considerable slow if the hash table is not correctly sized. | |
607 | * | |
db304854 | 608 | * Will automatically shrink the table via rhashtable_expand() if the |
7e1e7763 TG |
609 | * shrink_decision function specified at rhashtable_init() returns true. |
610 | * | |
611 | * The caller must ensure that no concurrent table mutations occur. It is | |
612 | * however valid to have concurrent lookups if they are RCU protected. | |
613 | */ | |
6eba8224 | 614 | bool rhashtable_remove(struct rhashtable *ht, struct rhash_head *obj) |
7e1e7763 | 615 | { |
a5ec68e3 | 616 | struct bucket_table *tbl, *new_tbl, *old_tbl; |
7e1e7763 | 617 | struct rhash_head __rcu **pprev; |
cf52d52f | 618 | struct rhash_head *he, *he2; |
a5ec68e3 | 619 | unsigned int hash, new_hash; |
fe6a043c | 620 | bool ret = false; |
7e1e7763 | 621 | |
97defe1e | 622 | rcu_read_lock(); |
020219a6 TG |
623 | old_tbl = rht_dereference_rcu(ht->tbl, ht); |
624 | tbl = new_tbl = rht_dereference_rcu(ht->future_tbl, ht); | |
cf52d52f | 625 | new_hash = obj_raw_hashfn(ht, rht_obj(ht, obj)); |
7e1e7763 | 626 | |
a5ec68e3 | 627 | lock_buckets(new_tbl, old_tbl, new_hash); |
97defe1e | 628 | restart: |
a5ec68e3 | 629 | hash = rht_bucket_index(tbl, new_hash); |
97defe1e TG |
630 | pprev = &tbl->buckets[hash]; |
631 | rht_for_each(he, tbl, hash) { | |
7e1e7763 TG |
632 | if (he != obj) { |
633 | pprev = &he->next; | |
634 | continue; | |
635 | } | |
636 | ||
7cd10db8 | 637 | ASSERT_BUCKET_LOCK(ht, tbl, hash); |
897362e4 | 638 | |
020219a6 TG |
639 | if (old_tbl->size > new_tbl->size && tbl == old_tbl && |
640 | !rht_is_a_nulls(obj->next) && | |
641 | head_hashfn(ht, tbl, obj->next) != hash) { | |
642 | rcu_assign_pointer(*pprev, (struct rhash_head *) rht_marker(ht, hash)); | |
643 | } else if (unlikely(old_tbl->size < new_tbl->size && tbl == new_tbl)) { | |
644 | rht_for_each_continue(he2, obj->next, tbl, hash) { | |
cf52d52f TG |
645 | if (head_hashfn(ht, tbl, he2) == hash) { |
646 | rcu_assign_pointer(*pprev, he2); | |
647 | goto found; | |
648 | } | |
649 | } | |
650 | ||
020219a6 | 651 | rcu_assign_pointer(*pprev, (struct rhash_head *) rht_marker(ht, hash)); |
cf52d52f TG |
652 | } else { |
653 | rcu_assign_pointer(*pprev, obj->next); | |
654 | } | |
655 | ||
656 | found: | |
fe6a043c TG |
657 | ret = true; |
658 | break; | |
7e1e7763 TG |
659 | } |
660 | ||
fe6a043c TG |
661 | /* The entry may be linked in either 'tbl', 'future_tbl', or both. |
662 | * 'future_tbl' only exists for a short period of time during | |
663 | * resizing. Thus traversing both is fine and the added cost is | |
664 | * very rare. | |
665 | */ | |
020219a6 TG |
666 | if (tbl != old_tbl) { |
667 | tbl = old_tbl; | |
97defe1e TG |
668 | goto restart; |
669 | } | |
670 | ||
a5ec68e3 | 671 | unlock_buckets(new_tbl, old_tbl, new_hash); |
fe6a043c TG |
672 | |
673 | if (ret) { | |
4c4b52d9 DB |
674 | bool no_resize_running = new_tbl == old_tbl; |
675 | ||
fe6a043c | 676 | atomic_dec(&ht->nelems); |
4c4b52d9 DB |
677 | if (no_resize_running && rht_shrink_below_30(ht, new_tbl->size)) |
678 | schedule_work(&ht->run_work); | |
fe6a043c TG |
679 | } |
680 | ||
97defe1e TG |
681 | rcu_read_unlock(); |
682 | ||
fe6a043c | 683 | return ret; |
7e1e7763 TG |
684 | } |
685 | EXPORT_SYMBOL_GPL(rhashtable_remove); | |
686 | ||
efb975a6 YX |
687 | struct rhashtable_compare_arg { |
688 | struct rhashtable *ht; | |
689 | const void *key; | |
690 | }; | |
691 | ||
692 | static bool rhashtable_compare(void *ptr, void *arg) | |
693 | { | |
694 | struct rhashtable_compare_arg *x = arg; | |
695 | struct rhashtable *ht = x->ht; | |
696 | ||
697 | return !memcmp(ptr + ht->p.key_offset, x->key, ht->p.key_len); | |
698 | } | |
699 | ||
7e1e7763 TG |
700 | /** |
701 | * rhashtable_lookup - lookup key in hash table | |
702 | * @ht: hash table | |
703 | * @key: pointer to key | |
704 | * | |
705 | * Computes the hash value for the key and traverses the bucket chain looking | |
706 | * for a entry with an identical key. The first matching entry is returned. | |
707 | * | |
708 | * This lookup function may only be used for fixed key hash table (key_len | |
db304854 | 709 | * parameter set). It will BUG() if used inappropriately. |
7e1e7763 | 710 | * |
97defe1e | 711 | * Lookups may occur in parallel with hashtable mutations and resizing. |
7e1e7763 | 712 | */ |
97defe1e | 713 | void *rhashtable_lookup(struct rhashtable *ht, const void *key) |
7e1e7763 | 714 | { |
efb975a6 YX |
715 | struct rhashtable_compare_arg arg = { |
716 | .ht = ht, | |
717 | .key = key, | |
718 | }; | |
7e1e7763 TG |
719 | |
720 | BUG_ON(!ht->p.key_len); | |
721 | ||
efb975a6 | 722 | return rhashtable_lookup_compare(ht, key, &rhashtable_compare, &arg); |
7e1e7763 TG |
723 | } |
724 | EXPORT_SYMBOL_GPL(rhashtable_lookup); | |
725 | ||
726 | /** | |
727 | * rhashtable_lookup_compare - search hash table with compare function | |
728 | * @ht: hash table | |
8d24c0b4 | 729 | * @key: the pointer to the key |
7e1e7763 TG |
730 | * @compare: compare function, must return true on match |
731 | * @arg: argument passed on to compare function | |
732 | * | |
733 | * Traverses the bucket chain behind the provided hash value and calls the | |
734 | * specified compare function for each entry. | |
735 | * | |
97defe1e | 736 | * Lookups may occur in parallel with hashtable mutations and resizing. |
7e1e7763 TG |
737 | * |
738 | * Returns the first entry on which the compare function returned true. | |
739 | */ | |
97defe1e | 740 | void *rhashtable_lookup_compare(struct rhashtable *ht, const void *key, |
7e1e7763 TG |
741 | bool (*compare)(void *, void *), void *arg) |
742 | { | |
97defe1e | 743 | const struct bucket_table *tbl, *old_tbl; |
7e1e7763 | 744 | struct rhash_head *he; |
8d24c0b4 | 745 | u32 hash; |
7e1e7763 | 746 | |
97defe1e TG |
747 | rcu_read_lock(); |
748 | ||
749 | old_tbl = rht_dereference_rcu(ht->tbl, ht); | |
750 | tbl = rht_dereference_rcu(ht->future_tbl, ht); | |
8d24c0b4 | 751 | hash = key_hashfn(ht, key, ht->p.key_len); |
97defe1e TG |
752 | restart: |
753 | rht_for_each_rcu(he, tbl, rht_bucket_index(tbl, hash)) { | |
7e1e7763 TG |
754 | if (!compare(rht_obj(ht, he), arg)) |
755 | continue; | |
97defe1e | 756 | rcu_read_unlock(); |
a4b18cda | 757 | return rht_obj(ht, he); |
7e1e7763 TG |
758 | } |
759 | ||
97defe1e TG |
760 | if (unlikely(tbl != old_tbl)) { |
761 | tbl = old_tbl; | |
762 | goto restart; | |
763 | } | |
764 | rcu_read_unlock(); | |
765 | ||
7e1e7763 TG |
766 | return NULL; |
767 | } | |
768 | EXPORT_SYMBOL_GPL(rhashtable_lookup_compare); | |
769 | ||
db304854 YX |
770 | /** |
771 | * rhashtable_lookup_insert - lookup and insert object into hash table | |
772 | * @ht: hash table | |
773 | * @obj: pointer to hash head inside object | |
774 | * | |
775 | * Locks down the bucket chain in both the old and new table if a resize | |
776 | * is in progress to ensure that writers can't remove from the old table | |
777 | * and can't insert to the new table during the atomic operation of search | |
778 | * and insertion. Searches for duplicates in both the old and new table if | |
779 | * a resize is in progress. | |
780 | * | |
781 | * This lookup function may only be used for fixed key hash table (key_len | |
782 | * parameter set). It will BUG() if used inappropriately. | |
783 | * | |
784 | * It is safe to call this function from atomic context. | |
785 | * | |
786 | * Will trigger an automatic deferred table resizing if the size grows | |
787 | * beyond the watermark indicated by grow_decision() which can be passed | |
788 | * to rhashtable_init(). | |
789 | */ | |
790 | bool rhashtable_lookup_insert(struct rhashtable *ht, struct rhash_head *obj) | |
7a868d1e YX |
791 | { |
792 | struct rhashtable_compare_arg arg = { | |
793 | .ht = ht, | |
794 | .key = rht_obj(ht, obj) + ht->p.key_offset, | |
795 | }; | |
796 | ||
797 | BUG_ON(!ht->p.key_len); | |
798 | ||
799 | return rhashtable_lookup_compare_insert(ht, obj, &rhashtable_compare, | |
800 | &arg); | |
801 | } | |
802 | EXPORT_SYMBOL_GPL(rhashtable_lookup_insert); | |
803 | ||
804 | /** | |
805 | * rhashtable_lookup_compare_insert - search and insert object to hash table | |
806 | * with compare function | |
807 | * @ht: hash table | |
808 | * @obj: pointer to hash head inside object | |
809 | * @compare: compare function, must return true on match | |
810 | * @arg: argument passed on to compare function | |
811 | * | |
812 | * Locks down the bucket chain in both the old and new table if a resize | |
813 | * is in progress to ensure that writers can't remove from the old table | |
814 | * and can't insert to the new table during the atomic operation of search | |
815 | * and insertion. Searches for duplicates in both the old and new table if | |
816 | * a resize is in progress. | |
817 | * | |
818 | * Lookups may occur in parallel with hashtable mutations and resizing. | |
819 | * | |
820 | * Will trigger an automatic deferred table resizing if the size grows | |
821 | * beyond the watermark indicated by grow_decision() which can be passed | |
822 | * to rhashtable_init(). | |
823 | */ | |
824 | bool rhashtable_lookup_compare_insert(struct rhashtable *ht, | |
825 | struct rhash_head *obj, | |
826 | bool (*compare)(void *, void *), | |
827 | void *arg) | |
db304854 YX |
828 | { |
829 | struct bucket_table *new_tbl, *old_tbl; | |
a5ec68e3 | 830 | u32 new_hash; |
db304854 YX |
831 | bool success = true; |
832 | ||
833 | BUG_ON(!ht->p.key_len); | |
834 | ||
835 | rcu_read_lock(); | |
db304854 | 836 | old_tbl = rht_dereference_rcu(ht->tbl, ht); |
db304854 | 837 | new_tbl = rht_dereference_rcu(ht->future_tbl, ht); |
020219a6 | 838 | new_hash = obj_raw_hashfn(ht, rht_obj(ht, obj)); |
a5ec68e3 TG |
839 | |
840 | lock_buckets(new_tbl, old_tbl, new_hash); | |
db304854 | 841 | |
7a868d1e YX |
842 | if (rhashtable_lookup_compare(ht, rht_obj(ht, obj) + ht->p.key_offset, |
843 | compare, arg)) { | |
db304854 YX |
844 | success = false; |
845 | goto exit; | |
846 | } | |
847 | ||
4c4b52d9 | 848 | __rhashtable_insert(ht, obj, new_tbl, old_tbl, new_hash); |
db304854 YX |
849 | |
850 | exit: | |
a5ec68e3 | 851 | unlock_buckets(new_tbl, old_tbl, new_hash); |
db304854 YX |
852 | rcu_read_unlock(); |
853 | ||
854 | return success; | |
855 | } | |
7a868d1e | 856 | EXPORT_SYMBOL_GPL(rhashtable_lookup_compare_insert); |
db304854 | 857 | |
f2dba9c6 HX |
858 | /** |
859 | * rhashtable_walk_init - Initialise an iterator | |
860 | * @ht: Table to walk over | |
861 | * @iter: Hash table Iterator | |
862 | * | |
863 | * This function prepares a hash table walk. | |
864 | * | |
865 | * Note that if you restart a walk after rhashtable_walk_stop you | |
866 | * may see the same object twice. Also, you may miss objects if | |
867 | * there are removals in between rhashtable_walk_stop and the next | |
868 | * call to rhashtable_walk_start. | |
869 | * | |
870 | * For a completely stable walk you should construct your own data | |
871 | * structure outside the hash table. | |
872 | * | |
873 | * This function may sleep so you must not call it from interrupt | |
874 | * context or with spin locks held. | |
875 | * | |
876 | * You must call rhashtable_walk_exit if this function returns | |
877 | * successfully. | |
878 | */ | |
879 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter) | |
880 | { | |
881 | iter->ht = ht; | |
882 | iter->p = NULL; | |
883 | iter->slot = 0; | |
884 | iter->skip = 0; | |
885 | ||
886 | iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL); | |
887 | if (!iter->walker) | |
888 | return -ENOMEM; | |
889 | ||
71bb0012 SL |
890 | INIT_LIST_HEAD(&iter->walker->list); |
891 | iter->walker->resize = false; | |
892 | ||
f2dba9c6 HX |
893 | mutex_lock(&ht->mutex); |
894 | list_add(&iter->walker->list, &ht->walkers); | |
895 | mutex_unlock(&ht->mutex); | |
896 | ||
897 | return 0; | |
898 | } | |
899 | EXPORT_SYMBOL_GPL(rhashtable_walk_init); | |
900 | ||
901 | /** | |
902 | * rhashtable_walk_exit - Free an iterator | |
903 | * @iter: Hash table Iterator | |
904 | * | |
905 | * This function frees resources allocated by rhashtable_walk_init. | |
906 | */ | |
907 | void rhashtable_walk_exit(struct rhashtable_iter *iter) | |
908 | { | |
909 | mutex_lock(&iter->ht->mutex); | |
910 | list_del(&iter->walker->list); | |
911 | mutex_unlock(&iter->ht->mutex); | |
912 | kfree(iter->walker); | |
913 | } | |
914 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); | |
915 | ||
916 | /** | |
917 | * rhashtable_walk_start - Start a hash table walk | |
918 | * @iter: Hash table iterator | |
919 | * | |
920 | * Start a hash table walk. Note that we take the RCU lock in all | |
921 | * cases including when we return an error. So you must always call | |
922 | * rhashtable_walk_stop to clean up. | |
923 | * | |
924 | * Returns zero if successful. | |
925 | * | |
926 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
927 | * will rewind back to the beginning and you may use it immediately | |
928 | * by calling rhashtable_walk_next. | |
929 | */ | |
930 | int rhashtable_walk_start(struct rhashtable_iter *iter) | |
931 | { | |
932 | rcu_read_lock(); | |
933 | ||
934 | if (iter->walker->resize) { | |
935 | iter->slot = 0; | |
936 | iter->skip = 0; | |
937 | iter->walker->resize = false; | |
938 | return -EAGAIN; | |
939 | } | |
940 | ||
941 | return 0; | |
942 | } | |
943 | EXPORT_SYMBOL_GPL(rhashtable_walk_start); | |
944 | ||
945 | /** | |
946 | * rhashtable_walk_next - Return the next object and advance the iterator | |
947 | * @iter: Hash table iterator | |
948 | * | |
949 | * Note that you must call rhashtable_walk_stop when you are finished | |
950 | * with the walk. | |
951 | * | |
952 | * Returns the next object or NULL when the end of the table is reached. | |
953 | * | |
954 | * Returns -EAGAIN if resize event occured. Note that the iterator | |
955 | * will rewind back to the beginning and you may continue to use it. | |
956 | */ | |
957 | void *rhashtable_walk_next(struct rhashtable_iter *iter) | |
958 | { | |
959 | const struct bucket_table *tbl; | |
960 | struct rhashtable *ht = iter->ht; | |
961 | struct rhash_head *p = iter->p; | |
962 | void *obj = NULL; | |
963 | ||
964 | tbl = rht_dereference_rcu(ht->tbl, ht); | |
965 | ||
966 | if (p) { | |
967 | p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot); | |
968 | goto next; | |
969 | } | |
970 | ||
971 | for (; iter->slot < tbl->size; iter->slot++) { | |
972 | int skip = iter->skip; | |
973 | ||
974 | rht_for_each_rcu(p, tbl, iter->slot) { | |
975 | if (!skip) | |
976 | break; | |
977 | skip--; | |
978 | } | |
979 | ||
980 | next: | |
981 | if (!rht_is_a_nulls(p)) { | |
982 | iter->skip++; | |
983 | iter->p = p; | |
984 | obj = rht_obj(ht, p); | |
985 | goto out; | |
986 | } | |
987 | ||
988 | iter->skip = 0; | |
989 | } | |
990 | ||
991 | iter->p = NULL; | |
992 | ||
993 | out: | |
994 | if (iter->walker->resize) { | |
995 | iter->p = NULL; | |
996 | iter->slot = 0; | |
997 | iter->skip = 0; | |
998 | iter->walker->resize = false; | |
999 | return ERR_PTR(-EAGAIN); | |
1000 | } | |
1001 | ||
1002 | return obj; | |
1003 | } | |
1004 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); | |
1005 | ||
1006 | /** | |
1007 | * rhashtable_walk_stop - Finish a hash table walk | |
1008 | * @iter: Hash table iterator | |
1009 | * | |
1010 | * Finish a hash table walk. | |
1011 | */ | |
1012 | void rhashtable_walk_stop(struct rhashtable_iter *iter) | |
1013 | { | |
1014 | rcu_read_unlock(); | |
1015 | iter->p = NULL; | |
1016 | } | |
1017 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); | |
1018 | ||
94000176 | 1019 | static size_t rounded_hashtable_size(struct rhashtable_params *params) |
7e1e7763 | 1020 | { |
94000176 YX |
1021 | return max(roundup_pow_of_two(params->nelem_hint * 4 / 3), |
1022 | 1UL << params->min_shift); | |
7e1e7763 TG |
1023 | } |
1024 | ||
1025 | /** | |
1026 | * rhashtable_init - initialize a new hash table | |
1027 | * @ht: hash table to be initialized | |
1028 | * @params: configuration parameters | |
1029 | * | |
1030 | * Initializes a new hash table based on the provided configuration | |
1031 | * parameters. A table can be configured either with a variable or | |
1032 | * fixed length key: | |
1033 | * | |
1034 | * Configuration Example 1: Fixed length keys | |
1035 | * struct test_obj { | |
1036 | * int key; | |
1037 | * void * my_member; | |
1038 | * struct rhash_head node; | |
1039 | * }; | |
1040 | * | |
1041 | * struct rhashtable_params params = { | |
1042 | * .head_offset = offsetof(struct test_obj, node), | |
1043 | * .key_offset = offsetof(struct test_obj, key), | |
1044 | * .key_len = sizeof(int), | |
87545899 | 1045 | * .hashfn = jhash, |
f89bd6f8 | 1046 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
7e1e7763 TG |
1047 | * }; |
1048 | * | |
1049 | * Configuration Example 2: Variable length keys | |
1050 | * struct test_obj { | |
1051 | * [...] | |
1052 | * struct rhash_head node; | |
1053 | * }; | |
1054 | * | |
1055 | * u32 my_hash_fn(const void *data, u32 seed) | |
1056 | * { | |
1057 | * struct test_obj *obj = data; | |
1058 | * | |
1059 | * return [... hash ...]; | |
1060 | * } | |
1061 | * | |
1062 | * struct rhashtable_params params = { | |
1063 | * .head_offset = offsetof(struct test_obj, node), | |
87545899 | 1064 | * .hashfn = jhash, |
7e1e7763 | 1065 | * .obj_hashfn = my_hash_fn, |
7e1e7763 TG |
1066 | * }; |
1067 | */ | |
1068 | int rhashtable_init(struct rhashtable *ht, struct rhashtable_params *params) | |
1069 | { | |
1070 | struct bucket_table *tbl; | |
1071 | size_t size; | |
1072 | ||
1073 | size = HASH_DEFAULT_SIZE; | |
1074 | ||
1075 | if ((params->key_len && !params->hashfn) || | |
1076 | (!params->key_len && !params->obj_hashfn)) | |
1077 | return -EINVAL; | |
1078 | ||
f89bd6f8 TG |
1079 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
1080 | return -EINVAL; | |
1081 | ||
94000176 YX |
1082 | params->min_shift = max_t(size_t, params->min_shift, |
1083 | ilog2(HASH_MIN_SIZE)); | |
1084 | ||
7e1e7763 | 1085 | if (params->nelem_hint) |
94000176 | 1086 | size = rounded_hashtable_size(params); |
7e1e7763 | 1087 | |
97defe1e TG |
1088 | memset(ht, 0, sizeof(*ht)); |
1089 | mutex_init(&ht->mutex); | |
1090 | memcpy(&ht->p, params, sizeof(*params)); | |
f2dba9c6 | 1091 | INIT_LIST_HEAD(&ht->walkers); |
97defe1e TG |
1092 | |
1093 | if (params->locks_mul) | |
1094 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); | |
1095 | else | |
1096 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; | |
1097 | ||
1098 | tbl = bucket_table_alloc(ht, size); | |
7e1e7763 TG |
1099 | if (tbl == NULL) |
1100 | return -ENOMEM; | |
1101 | ||
545a148e | 1102 | atomic_set(&ht->nelems, 0); |
c0c09bfd | 1103 | atomic_set(&ht->shift, ilog2(tbl->size)); |
7e1e7763 | 1104 | RCU_INIT_POINTER(ht->tbl, tbl); |
97defe1e | 1105 | RCU_INIT_POINTER(ht->future_tbl, tbl); |
7e1e7763 TG |
1106 | |
1107 | if (!ht->p.hash_rnd) | |
1108 | get_random_bytes(&ht->p.hash_rnd, sizeof(ht->p.hash_rnd)); | |
1109 | ||
4c4b52d9 | 1110 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
97defe1e | 1111 | |
7e1e7763 TG |
1112 | return 0; |
1113 | } | |
1114 | EXPORT_SYMBOL_GPL(rhashtable_init); | |
1115 | ||
1116 | /** | |
1117 | * rhashtable_destroy - destroy hash table | |
1118 | * @ht: the hash table to destroy | |
1119 | * | |
ae82ddcf PNA |
1120 | * Frees the bucket array. This function is not rcu safe, therefore the caller |
1121 | * has to make sure that no resizing may happen by unpublishing the hashtable | |
1122 | * and waiting for the quiescent cycle before releasing the bucket array. | |
7e1e7763 | 1123 | */ |
97defe1e | 1124 | void rhashtable_destroy(struct rhashtable *ht) |
7e1e7763 | 1125 | { |
97defe1e TG |
1126 | ht->being_destroyed = true; |
1127 | ||
4c4b52d9 | 1128 | cancel_work_sync(&ht->run_work); |
97defe1e | 1129 | |
57699a40 | 1130 | mutex_lock(&ht->mutex); |
97defe1e | 1131 | bucket_table_free(rht_dereference(ht->tbl, ht)); |
97defe1e | 1132 | mutex_unlock(&ht->mutex); |
7e1e7763 TG |
1133 | } |
1134 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |