| 1 | /* |
| 2 | * Resizable, Scalable, Concurrent Hash Table |
| 3 | * |
| 4 | * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au> |
| 5 | * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch> |
| 6 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> |
| 7 | * |
| 8 | * Code partially derived from nft_hash |
| 9 | * Rewritten with rehash code from br_multicast plus single list |
| 10 | * pointer as suggested by Josh Triplett |
| 11 | * |
| 12 | * This program is free software; you can redistribute it and/or modify |
| 13 | * it under the terms of the GNU General Public License version 2 as |
| 14 | * published by the Free Software Foundation. |
| 15 | */ |
| 16 | |
| 17 | #include <linux/atomic.h> |
| 18 | #include <linux/kernel.h> |
| 19 | #include <linux/init.h> |
| 20 | #include <linux/log2.h> |
| 21 | #include <linux/sched.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/vmalloc.h> |
| 24 | #include <linux/mm.h> |
| 25 | #include <linux/jhash.h> |
| 26 | #include <linux/random.h> |
| 27 | #include <linux/rhashtable.h> |
| 28 | #include <linux/err.h> |
| 29 | |
| 30 | #define HASH_DEFAULT_SIZE 64UL |
| 31 | #define HASH_MIN_SIZE 4U |
| 32 | #define BUCKET_LOCKS_PER_CPU 128UL |
| 33 | |
| 34 | static u32 head_hashfn(struct rhashtable *ht, |
| 35 | const struct bucket_table *tbl, |
| 36 | const struct rhash_head *he) |
| 37 | { |
| 38 | return rht_head_hashfn(ht, tbl, he, ht->p); |
| 39 | } |
| 40 | |
| 41 | #ifdef CONFIG_PROVE_LOCKING |
| 42 | #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT)) |
| 43 | |
| 44 | int lockdep_rht_mutex_is_held(struct rhashtable *ht) |
| 45 | { |
| 46 | return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1; |
| 47 | } |
| 48 | EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held); |
| 49 | |
| 50 | int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash) |
| 51 | { |
| 52 | spinlock_t *lock = rht_bucket_lock(tbl, hash); |
| 53 | |
| 54 | return (debug_locks) ? lockdep_is_held(lock) : 1; |
| 55 | } |
| 56 | EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held); |
| 57 | #else |
| 58 | #define ASSERT_RHT_MUTEX(HT) |
| 59 | #endif |
| 60 | |
| 61 | |
| 62 | static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl, |
| 63 | gfp_t gfp) |
| 64 | { |
| 65 | unsigned int i, size; |
| 66 | #if defined(CONFIG_PROVE_LOCKING) |
| 67 | unsigned int nr_pcpus = 2; |
| 68 | #else |
| 69 | unsigned int nr_pcpus = num_possible_cpus(); |
| 70 | #endif |
| 71 | |
| 72 | nr_pcpus = min_t(unsigned int, nr_pcpus, 32UL); |
| 73 | size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul); |
| 74 | |
| 75 | /* Never allocate more than 0.5 locks per bucket */ |
| 76 | size = min_t(unsigned int, size, tbl->size >> 1); |
| 77 | |
| 78 | if (sizeof(spinlock_t) != 0) { |
| 79 | #ifdef CONFIG_NUMA |
| 80 | if (size * sizeof(spinlock_t) > PAGE_SIZE && |
| 81 | gfp == GFP_KERNEL) |
| 82 | tbl->locks = vmalloc(size * sizeof(spinlock_t)); |
| 83 | else |
| 84 | #endif |
| 85 | tbl->locks = kmalloc_array(size, sizeof(spinlock_t), |
| 86 | gfp); |
| 87 | if (!tbl->locks) |
| 88 | return -ENOMEM; |
| 89 | for (i = 0; i < size; i++) |
| 90 | spin_lock_init(&tbl->locks[i]); |
| 91 | } |
| 92 | tbl->locks_mask = size - 1; |
| 93 | |
| 94 | return 0; |
| 95 | } |
| 96 | |
| 97 | static void bucket_table_free(const struct bucket_table *tbl) |
| 98 | { |
| 99 | if (tbl) |
| 100 | kvfree(tbl->locks); |
| 101 | |
| 102 | kvfree(tbl); |
| 103 | } |
| 104 | |
| 105 | static void bucket_table_free_rcu(struct rcu_head *head) |
| 106 | { |
| 107 | bucket_table_free(container_of(head, struct bucket_table, rcu)); |
| 108 | } |
| 109 | |
| 110 | static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, |
| 111 | size_t nbuckets, |
| 112 | gfp_t gfp) |
| 113 | { |
| 114 | struct bucket_table *tbl = NULL; |
| 115 | size_t size; |
| 116 | int i; |
| 117 | |
| 118 | size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); |
| 119 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) || |
| 120 | gfp != GFP_KERNEL) |
| 121 | tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY); |
| 122 | if (tbl == NULL && gfp == GFP_KERNEL) |
| 123 | tbl = vzalloc(size); |
| 124 | if (tbl == NULL) |
| 125 | return NULL; |
| 126 | |
| 127 | tbl->size = nbuckets; |
| 128 | |
| 129 | if (alloc_bucket_locks(ht, tbl, gfp) < 0) { |
| 130 | bucket_table_free(tbl); |
| 131 | return NULL; |
| 132 | } |
| 133 | |
| 134 | INIT_LIST_HEAD(&tbl->walkers); |
| 135 | |
| 136 | get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd)); |
| 137 | |
| 138 | for (i = 0; i < nbuckets; i++) |
| 139 | INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i); |
| 140 | |
| 141 | return tbl; |
| 142 | } |
| 143 | |
| 144 | static struct bucket_table *rhashtable_last_table(struct rhashtable *ht, |
| 145 | struct bucket_table *tbl) |
| 146 | { |
| 147 | struct bucket_table *new_tbl; |
| 148 | |
| 149 | do { |
| 150 | new_tbl = tbl; |
| 151 | tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
| 152 | } while (tbl); |
| 153 | |
| 154 | return new_tbl; |
| 155 | } |
| 156 | |
| 157 | static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash) |
| 158 | { |
| 159 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
| 160 | struct bucket_table *new_tbl = rhashtable_last_table(ht, |
| 161 | rht_dereference_rcu(old_tbl->future_tbl, ht)); |
| 162 | struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash]; |
| 163 | int err = -ENOENT; |
| 164 | struct rhash_head *head, *next, *entry; |
| 165 | spinlock_t *new_bucket_lock; |
| 166 | unsigned int new_hash; |
| 167 | |
| 168 | rht_for_each(entry, old_tbl, old_hash) { |
| 169 | err = 0; |
| 170 | next = rht_dereference_bucket(entry->next, old_tbl, old_hash); |
| 171 | |
| 172 | if (rht_is_a_nulls(next)) |
| 173 | break; |
| 174 | |
| 175 | pprev = &entry->next; |
| 176 | } |
| 177 | |
| 178 | if (err) |
| 179 | goto out; |
| 180 | |
| 181 | new_hash = head_hashfn(ht, new_tbl, entry); |
| 182 | |
| 183 | new_bucket_lock = rht_bucket_lock(new_tbl, new_hash); |
| 184 | |
| 185 | spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING); |
| 186 | head = rht_dereference_bucket(new_tbl->buckets[new_hash], |
| 187 | new_tbl, new_hash); |
| 188 | |
| 189 | if (rht_is_a_nulls(head)) |
| 190 | INIT_RHT_NULLS_HEAD(entry->next, ht, new_hash); |
| 191 | else |
| 192 | RCU_INIT_POINTER(entry->next, head); |
| 193 | |
| 194 | rcu_assign_pointer(new_tbl->buckets[new_hash], entry); |
| 195 | spin_unlock(new_bucket_lock); |
| 196 | |
| 197 | rcu_assign_pointer(*pprev, next); |
| 198 | |
| 199 | out: |
| 200 | return err; |
| 201 | } |
| 202 | |
| 203 | static void rhashtable_rehash_chain(struct rhashtable *ht, |
| 204 | unsigned int old_hash) |
| 205 | { |
| 206 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
| 207 | spinlock_t *old_bucket_lock; |
| 208 | |
| 209 | old_bucket_lock = rht_bucket_lock(old_tbl, old_hash); |
| 210 | |
| 211 | spin_lock_bh(old_bucket_lock); |
| 212 | while (!rhashtable_rehash_one(ht, old_hash)) |
| 213 | ; |
| 214 | old_tbl->rehash++; |
| 215 | spin_unlock_bh(old_bucket_lock); |
| 216 | } |
| 217 | |
| 218 | static int rhashtable_rehash_attach(struct rhashtable *ht, |
| 219 | struct bucket_table *old_tbl, |
| 220 | struct bucket_table *new_tbl) |
| 221 | { |
| 222 | /* Protect future_tbl using the first bucket lock. */ |
| 223 | spin_lock_bh(old_tbl->locks); |
| 224 | |
| 225 | /* Did somebody beat us to it? */ |
| 226 | if (rcu_access_pointer(old_tbl->future_tbl)) { |
| 227 | spin_unlock_bh(old_tbl->locks); |
| 228 | return -EEXIST; |
| 229 | } |
| 230 | |
| 231 | /* Make insertions go into the new, empty table right away. Deletions |
| 232 | * and lookups will be attempted in both tables until we synchronize. |
| 233 | */ |
| 234 | rcu_assign_pointer(old_tbl->future_tbl, new_tbl); |
| 235 | |
| 236 | /* Ensure the new table is visible to readers. */ |
| 237 | smp_wmb(); |
| 238 | |
| 239 | spin_unlock_bh(old_tbl->locks); |
| 240 | |
| 241 | return 0; |
| 242 | } |
| 243 | |
| 244 | static int rhashtable_rehash_table(struct rhashtable *ht) |
| 245 | { |
| 246 | struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht); |
| 247 | struct bucket_table *new_tbl; |
| 248 | struct rhashtable_walker *walker; |
| 249 | unsigned int old_hash; |
| 250 | |
| 251 | new_tbl = rht_dereference(old_tbl->future_tbl, ht); |
| 252 | if (!new_tbl) |
| 253 | return 0; |
| 254 | |
| 255 | for (old_hash = 0; old_hash < old_tbl->size; old_hash++) |
| 256 | rhashtable_rehash_chain(ht, old_hash); |
| 257 | |
| 258 | /* Publish the new table pointer. */ |
| 259 | rcu_assign_pointer(ht->tbl, new_tbl); |
| 260 | |
| 261 | spin_lock(&ht->lock); |
| 262 | list_for_each_entry(walker, &old_tbl->walkers, list) |
| 263 | walker->tbl = NULL; |
| 264 | spin_unlock(&ht->lock); |
| 265 | |
| 266 | /* Wait for readers. All new readers will see the new |
| 267 | * table, and thus no references to the old table will |
| 268 | * remain. |
| 269 | */ |
| 270 | call_rcu(&old_tbl->rcu, bucket_table_free_rcu); |
| 271 | |
| 272 | return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0; |
| 273 | } |
| 274 | |
| 275 | /** |
| 276 | * rhashtable_expand - Expand hash table while allowing concurrent lookups |
| 277 | * @ht: the hash table to expand |
| 278 | * |
| 279 | * A secondary bucket array is allocated and the hash entries are migrated. |
| 280 | * |
| 281 | * This function may only be called in a context where it is safe to call |
| 282 | * synchronize_rcu(), e.g. not within a rcu_read_lock() section. |
| 283 | * |
| 284 | * The caller must ensure that no concurrent resizing occurs by holding |
| 285 | * ht->mutex. |
| 286 | * |
| 287 | * It is valid to have concurrent insertions and deletions protected by per |
| 288 | * bucket locks or concurrent RCU protected lookups and traversals. |
| 289 | */ |
| 290 | static int rhashtable_expand(struct rhashtable *ht) |
| 291 | { |
| 292 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); |
| 293 | int err; |
| 294 | |
| 295 | ASSERT_RHT_MUTEX(ht); |
| 296 | |
| 297 | old_tbl = rhashtable_last_table(ht, old_tbl); |
| 298 | |
| 299 | new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL); |
| 300 | if (new_tbl == NULL) |
| 301 | return -ENOMEM; |
| 302 | |
| 303 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
| 304 | if (err) |
| 305 | bucket_table_free(new_tbl); |
| 306 | |
| 307 | return err; |
| 308 | } |
| 309 | |
| 310 | /** |
| 311 | * rhashtable_shrink - Shrink hash table while allowing concurrent lookups |
| 312 | * @ht: the hash table to shrink |
| 313 | * |
| 314 | * This function shrinks the hash table to fit, i.e., the smallest |
| 315 | * size would not cause it to expand right away automatically. |
| 316 | * |
| 317 | * The caller must ensure that no concurrent resizing occurs by holding |
| 318 | * ht->mutex. |
| 319 | * |
| 320 | * The caller must ensure that no concurrent table mutations take place. |
| 321 | * It is however valid to have concurrent lookups if they are RCU protected. |
| 322 | * |
| 323 | * It is valid to have concurrent insertions and deletions protected by per |
| 324 | * bucket locks or concurrent RCU protected lookups and traversals. |
| 325 | */ |
| 326 | static int rhashtable_shrink(struct rhashtable *ht) |
| 327 | { |
| 328 | struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht); |
| 329 | unsigned int size; |
| 330 | int err; |
| 331 | |
| 332 | ASSERT_RHT_MUTEX(ht); |
| 333 | |
| 334 | size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2); |
| 335 | if (size < ht->p.min_size) |
| 336 | size = ht->p.min_size; |
| 337 | |
| 338 | if (old_tbl->size <= size) |
| 339 | return 0; |
| 340 | |
| 341 | if (rht_dereference(old_tbl->future_tbl, ht)) |
| 342 | return -EEXIST; |
| 343 | |
| 344 | new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
| 345 | if (new_tbl == NULL) |
| 346 | return -ENOMEM; |
| 347 | |
| 348 | err = rhashtable_rehash_attach(ht, old_tbl, new_tbl); |
| 349 | if (err) |
| 350 | bucket_table_free(new_tbl); |
| 351 | |
| 352 | return err; |
| 353 | } |
| 354 | |
| 355 | static void rht_deferred_worker(struct work_struct *work) |
| 356 | { |
| 357 | struct rhashtable *ht; |
| 358 | struct bucket_table *tbl; |
| 359 | int err = 0; |
| 360 | |
| 361 | ht = container_of(work, struct rhashtable, run_work); |
| 362 | mutex_lock(&ht->mutex); |
| 363 | |
| 364 | tbl = rht_dereference(ht->tbl, ht); |
| 365 | tbl = rhashtable_last_table(ht, tbl); |
| 366 | |
| 367 | if (rht_grow_above_75(ht, tbl)) |
| 368 | rhashtable_expand(ht); |
| 369 | else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl)) |
| 370 | rhashtable_shrink(ht); |
| 371 | |
| 372 | err = rhashtable_rehash_table(ht); |
| 373 | |
| 374 | mutex_unlock(&ht->mutex); |
| 375 | |
| 376 | if (err) |
| 377 | schedule_work(&ht->run_work); |
| 378 | } |
| 379 | |
| 380 | static bool rhashtable_check_elasticity(struct rhashtable *ht, |
| 381 | struct bucket_table *tbl, |
| 382 | unsigned int hash) |
| 383 | { |
| 384 | unsigned int elasticity = ht->elasticity; |
| 385 | struct rhash_head *head; |
| 386 | |
| 387 | rht_for_each(head, tbl, hash) |
| 388 | if (!--elasticity) |
| 389 | return true; |
| 390 | |
| 391 | return false; |
| 392 | } |
| 393 | |
| 394 | int rhashtable_insert_rehash(struct rhashtable *ht) |
| 395 | { |
| 396 | struct bucket_table *old_tbl; |
| 397 | struct bucket_table *new_tbl; |
| 398 | struct bucket_table *tbl; |
| 399 | unsigned int size; |
| 400 | int err; |
| 401 | |
| 402 | old_tbl = rht_dereference_rcu(ht->tbl, ht); |
| 403 | tbl = rhashtable_last_table(ht, old_tbl); |
| 404 | |
| 405 | size = tbl->size; |
| 406 | |
| 407 | if (rht_grow_above_75(ht, tbl)) |
| 408 | size *= 2; |
| 409 | /* Do not schedule more than one rehash */ |
| 410 | else if (old_tbl != tbl) |
| 411 | return -EBUSY; |
| 412 | |
| 413 | new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC); |
| 414 | if (new_tbl == NULL) { |
| 415 | /* Schedule async resize/rehash to try allocation |
| 416 | * non-atomic context. |
| 417 | */ |
| 418 | schedule_work(&ht->run_work); |
| 419 | return -ENOMEM; |
| 420 | } |
| 421 | |
| 422 | err = rhashtable_rehash_attach(ht, tbl, new_tbl); |
| 423 | if (err) { |
| 424 | bucket_table_free(new_tbl); |
| 425 | if (err == -EEXIST) |
| 426 | err = 0; |
| 427 | } else |
| 428 | schedule_work(&ht->run_work); |
| 429 | |
| 430 | return err; |
| 431 | } |
| 432 | EXPORT_SYMBOL_GPL(rhashtable_insert_rehash); |
| 433 | |
| 434 | int rhashtable_insert_slow(struct rhashtable *ht, const void *key, |
| 435 | struct rhash_head *obj, |
| 436 | struct bucket_table *tbl) |
| 437 | { |
| 438 | struct rhash_head *head; |
| 439 | unsigned int hash; |
| 440 | int err; |
| 441 | |
| 442 | tbl = rhashtable_last_table(ht, tbl); |
| 443 | hash = head_hashfn(ht, tbl, obj); |
| 444 | spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING); |
| 445 | |
| 446 | err = -EEXIST; |
| 447 | if (key && rhashtable_lookup_fast(ht, key, ht->p)) |
| 448 | goto exit; |
| 449 | |
| 450 | err = -E2BIG; |
| 451 | if (unlikely(rht_grow_above_max(ht, tbl))) |
| 452 | goto exit; |
| 453 | |
| 454 | err = -EAGAIN; |
| 455 | if (rhashtable_check_elasticity(ht, tbl, hash) || |
| 456 | rht_grow_above_100(ht, tbl)) |
| 457 | goto exit; |
| 458 | |
| 459 | err = 0; |
| 460 | |
| 461 | head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash); |
| 462 | |
| 463 | RCU_INIT_POINTER(obj->next, head); |
| 464 | |
| 465 | rcu_assign_pointer(tbl->buckets[hash], obj); |
| 466 | |
| 467 | atomic_inc(&ht->nelems); |
| 468 | |
| 469 | exit: |
| 470 | spin_unlock(rht_bucket_lock(tbl, hash)); |
| 471 | |
| 472 | return err; |
| 473 | } |
| 474 | EXPORT_SYMBOL_GPL(rhashtable_insert_slow); |
| 475 | |
| 476 | /** |
| 477 | * rhashtable_walk_init - Initialise an iterator |
| 478 | * @ht: Table to walk over |
| 479 | * @iter: Hash table Iterator |
| 480 | * |
| 481 | * This function prepares a hash table walk. |
| 482 | * |
| 483 | * Note that if you restart a walk after rhashtable_walk_stop you |
| 484 | * may see the same object twice. Also, you may miss objects if |
| 485 | * there are removals in between rhashtable_walk_stop and the next |
| 486 | * call to rhashtable_walk_start. |
| 487 | * |
| 488 | * For a completely stable walk you should construct your own data |
| 489 | * structure outside the hash table. |
| 490 | * |
| 491 | * This function may sleep so you must not call it from interrupt |
| 492 | * context or with spin locks held. |
| 493 | * |
| 494 | * You must call rhashtable_walk_exit if this function returns |
| 495 | * successfully. |
| 496 | */ |
| 497 | int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter) |
| 498 | { |
| 499 | iter->ht = ht; |
| 500 | iter->p = NULL; |
| 501 | iter->slot = 0; |
| 502 | iter->skip = 0; |
| 503 | |
| 504 | iter->walker = kmalloc(sizeof(*iter->walker), GFP_KERNEL); |
| 505 | if (!iter->walker) |
| 506 | return -ENOMEM; |
| 507 | |
| 508 | mutex_lock(&ht->mutex); |
| 509 | iter->walker->tbl = rht_dereference(ht->tbl, ht); |
| 510 | list_add(&iter->walker->list, &iter->walker->tbl->walkers); |
| 511 | mutex_unlock(&ht->mutex); |
| 512 | |
| 513 | return 0; |
| 514 | } |
| 515 | EXPORT_SYMBOL_GPL(rhashtable_walk_init); |
| 516 | |
| 517 | /** |
| 518 | * rhashtable_walk_exit - Free an iterator |
| 519 | * @iter: Hash table Iterator |
| 520 | * |
| 521 | * This function frees resources allocated by rhashtable_walk_init. |
| 522 | */ |
| 523 | void rhashtable_walk_exit(struct rhashtable_iter *iter) |
| 524 | { |
| 525 | mutex_lock(&iter->ht->mutex); |
| 526 | if (iter->walker->tbl) |
| 527 | list_del(&iter->walker->list); |
| 528 | mutex_unlock(&iter->ht->mutex); |
| 529 | kfree(iter->walker); |
| 530 | } |
| 531 | EXPORT_SYMBOL_GPL(rhashtable_walk_exit); |
| 532 | |
| 533 | /** |
| 534 | * rhashtable_walk_start - Start a hash table walk |
| 535 | * @iter: Hash table iterator |
| 536 | * |
| 537 | * Start a hash table walk. Note that we take the RCU lock in all |
| 538 | * cases including when we return an error. So you must always call |
| 539 | * rhashtable_walk_stop to clean up. |
| 540 | * |
| 541 | * Returns zero if successful. |
| 542 | * |
| 543 | * Returns -EAGAIN if resize event occured. Note that the iterator |
| 544 | * will rewind back to the beginning and you may use it immediately |
| 545 | * by calling rhashtable_walk_next. |
| 546 | */ |
| 547 | int rhashtable_walk_start(struct rhashtable_iter *iter) |
| 548 | __acquires(RCU) |
| 549 | { |
| 550 | struct rhashtable *ht = iter->ht; |
| 551 | |
| 552 | mutex_lock(&ht->mutex); |
| 553 | |
| 554 | if (iter->walker->tbl) |
| 555 | list_del(&iter->walker->list); |
| 556 | |
| 557 | rcu_read_lock(); |
| 558 | |
| 559 | mutex_unlock(&ht->mutex); |
| 560 | |
| 561 | if (!iter->walker->tbl) { |
| 562 | iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht); |
| 563 | return -EAGAIN; |
| 564 | } |
| 565 | |
| 566 | return 0; |
| 567 | } |
| 568 | EXPORT_SYMBOL_GPL(rhashtable_walk_start); |
| 569 | |
| 570 | /** |
| 571 | * rhashtable_walk_next - Return the next object and advance the iterator |
| 572 | * @iter: Hash table iterator |
| 573 | * |
| 574 | * Note that you must call rhashtable_walk_stop when you are finished |
| 575 | * with the walk. |
| 576 | * |
| 577 | * Returns the next object or NULL when the end of the table is reached. |
| 578 | * |
| 579 | * Returns -EAGAIN if resize event occured. Note that the iterator |
| 580 | * will rewind back to the beginning and you may continue to use it. |
| 581 | */ |
| 582 | void *rhashtable_walk_next(struct rhashtable_iter *iter) |
| 583 | { |
| 584 | struct bucket_table *tbl = iter->walker->tbl; |
| 585 | struct rhashtable *ht = iter->ht; |
| 586 | struct rhash_head *p = iter->p; |
| 587 | void *obj = NULL; |
| 588 | |
| 589 | if (p) { |
| 590 | p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot); |
| 591 | goto next; |
| 592 | } |
| 593 | |
| 594 | for (; iter->slot < tbl->size; iter->slot++) { |
| 595 | int skip = iter->skip; |
| 596 | |
| 597 | rht_for_each_rcu(p, tbl, iter->slot) { |
| 598 | if (!skip) |
| 599 | break; |
| 600 | skip--; |
| 601 | } |
| 602 | |
| 603 | next: |
| 604 | if (!rht_is_a_nulls(p)) { |
| 605 | iter->skip++; |
| 606 | iter->p = p; |
| 607 | obj = rht_obj(ht, p); |
| 608 | goto out; |
| 609 | } |
| 610 | |
| 611 | iter->skip = 0; |
| 612 | } |
| 613 | |
| 614 | /* Ensure we see any new tables. */ |
| 615 | smp_rmb(); |
| 616 | |
| 617 | iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht); |
| 618 | if (iter->walker->tbl) { |
| 619 | iter->slot = 0; |
| 620 | iter->skip = 0; |
| 621 | return ERR_PTR(-EAGAIN); |
| 622 | } |
| 623 | |
| 624 | iter->p = NULL; |
| 625 | |
| 626 | out: |
| 627 | |
| 628 | return obj; |
| 629 | } |
| 630 | EXPORT_SYMBOL_GPL(rhashtable_walk_next); |
| 631 | |
| 632 | /** |
| 633 | * rhashtable_walk_stop - Finish a hash table walk |
| 634 | * @iter: Hash table iterator |
| 635 | * |
| 636 | * Finish a hash table walk. |
| 637 | */ |
| 638 | void rhashtable_walk_stop(struct rhashtable_iter *iter) |
| 639 | __releases(RCU) |
| 640 | { |
| 641 | struct rhashtable *ht; |
| 642 | struct bucket_table *tbl = iter->walker->tbl; |
| 643 | |
| 644 | if (!tbl) |
| 645 | goto out; |
| 646 | |
| 647 | ht = iter->ht; |
| 648 | |
| 649 | spin_lock(&ht->lock); |
| 650 | if (tbl->rehash < tbl->size) |
| 651 | list_add(&iter->walker->list, &tbl->walkers); |
| 652 | else |
| 653 | iter->walker->tbl = NULL; |
| 654 | spin_unlock(&ht->lock); |
| 655 | |
| 656 | iter->p = NULL; |
| 657 | |
| 658 | out: |
| 659 | rcu_read_unlock(); |
| 660 | } |
| 661 | EXPORT_SYMBOL_GPL(rhashtable_walk_stop); |
| 662 | |
| 663 | static size_t rounded_hashtable_size(const struct rhashtable_params *params) |
| 664 | { |
| 665 | return max(roundup_pow_of_two(params->nelem_hint * 4 / 3), |
| 666 | (unsigned long)params->min_size); |
| 667 | } |
| 668 | |
| 669 | static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed) |
| 670 | { |
| 671 | return jhash2(key, length, seed); |
| 672 | } |
| 673 | |
| 674 | /** |
| 675 | * rhashtable_init - initialize a new hash table |
| 676 | * @ht: hash table to be initialized |
| 677 | * @params: configuration parameters |
| 678 | * |
| 679 | * Initializes a new hash table based on the provided configuration |
| 680 | * parameters. A table can be configured either with a variable or |
| 681 | * fixed length key: |
| 682 | * |
| 683 | * Configuration Example 1: Fixed length keys |
| 684 | * struct test_obj { |
| 685 | * int key; |
| 686 | * void * my_member; |
| 687 | * struct rhash_head node; |
| 688 | * }; |
| 689 | * |
| 690 | * struct rhashtable_params params = { |
| 691 | * .head_offset = offsetof(struct test_obj, node), |
| 692 | * .key_offset = offsetof(struct test_obj, key), |
| 693 | * .key_len = sizeof(int), |
| 694 | * .hashfn = jhash, |
| 695 | * .nulls_base = (1U << RHT_BASE_SHIFT), |
| 696 | * }; |
| 697 | * |
| 698 | * Configuration Example 2: Variable length keys |
| 699 | * struct test_obj { |
| 700 | * [...] |
| 701 | * struct rhash_head node; |
| 702 | * }; |
| 703 | * |
| 704 | * u32 my_hash_fn(const void *data, u32 len, u32 seed) |
| 705 | * { |
| 706 | * struct test_obj *obj = data; |
| 707 | * |
| 708 | * return [... hash ...]; |
| 709 | * } |
| 710 | * |
| 711 | * struct rhashtable_params params = { |
| 712 | * .head_offset = offsetof(struct test_obj, node), |
| 713 | * .hashfn = jhash, |
| 714 | * .obj_hashfn = my_hash_fn, |
| 715 | * }; |
| 716 | */ |
| 717 | int rhashtable_init(struct rhashtable *ht, |
| 718 | const struct rhashtable_params *params) |
| 719 | { |
| 720 | struct bucket_table *tbl; |
| 721 | size_t size; |
| 722 | |
| 723 | size = HASH_DEFAULT_SIZE; |
| 724 | |
| 725 | if ((!params->key_len && !params->obj_hashfn) || |
| 726 | (params->obj_hashfn && !params->obj_cmpfn)) |
| 727 | return -EINVAL; |
| 728 | |
| 729 | if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT)) |
| 730 | return -EINVAL; |
| 731 | |
| 732 | if (params->nelem_hint) |
| 733 | size = rounded_hashtable_size(params); |
| 734 | |
| 735 | memset(ht, 0, sizeof(*ht)); |
| 736 | mutex_init(&ht->mutex); |
| 737 | spin_lock_init(&ht->lock); |
| 738 | memcpy(&ht->p, params, sizeof(*params)); |
| 739 | |
| 740 | if (params->min_size) |
| 741 | ht->p.min_size = roundup_pow_of_two(params->min_size); |
| 742 | |
| 743 | if (params->max_size) |
| 744 | ht->p.max_size = rounddown_pow_of_two(params->max_size); |
| 745 | |
| 746 | if (params->insecure_max_entries) |
| 747 | ht->p.insecure_max_entries = |
| 748 | rounddown_pow_of_two(params->insecure_max_entries); |
| 749 | else |
| 750 | ht->p.insecure_max_entries = ht->p.max_size * 2; |
| 751 | |
| 752 | ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE); |
| 753 | |
| 754 | /* The maximum (not average) chain length grows with the |
| 755 | * size of the hash table, at a rate of (log N)/(log log N). |
| 756 | * The value of 16 is selected so that even if the hash |
| 757 | * table grew to 2^32 you would not expect the maximum |
| 758 | * chain length to exceed it unless we are under attack |
| 759 | * (or extremely unlucky). |
| 760 | * |
| 761 | * As this limit is only to detect attacks, we don't need |
| 762 | * to set it to a lower value as you'd need the chain |
| 763 | * length to vastly exceed 16 to have any real effect |
| 764 | * on the system. |
| 765 | */ |
| 766 | if (!params->insecure_elasticity) |
| 767 | ht->elasticity = 16; |
| 768 | |
| 769 | if (params->locks_mul) |
| 770 | ht->p.locks_mul = roundup_pow_of_two(params->locks_mul); |
| 771 | else |
| 772 | ht->p.locks_mul = BUCKET_LOCKS_PER_CPU; |
| 773 | |
| 774 | ht->key_len = ht->p.key_len; |
| 775 | if (!params->hashfn) { |
| 776 | ht->p.hashfn = jhash; |
| 777 | |
| 778 | if (!(ht->key_len & (sizeof(u32) - 1))) { |
| 779 | ht->key_len /= sizeof(u32); |
| 780 | ht->p.hashfn = rhashtable_jhash2; |
| 781 | } |
| 782 | } |
| 783 | |
| 784 | tbl = bucket_table_alloc(ht, size, GFP_KERNEL); |
| 785 | if (tbl == NULL) |
| 786 | return -ENOMEM; |
| 787 | |
| 788 | atomic_set(&ht->nelems, 0); |
| 789 | |
| 790 | RCU_INIT_POINTER(ht->tbl, tbl); |
| 791 | |
| 792 | INIT_WORK(&ht->run_work, rht_deferred_worker); |
| 793 | |
| 794 | return 0; |
| 795 | } |
| 796 | EXPORT_SYMBOL_GPL(rhashtable_init); |
| 797 | |
| 798 | /** |
| 799 | * rhashtable_free_and_destroy - free elements and destroy hash table |
| 800 | * @ht: the hash table to destroy |
| 801 | * @free_fn: callback to release resources of element |
| 802 | * @arg: pointer passed to free_fn |
| 803 | * |
| 804 | * Stops an eventual async resize. If defined, invokes free_fn for each |
| 805 | * element to releasal resources. Please note that RCU protected |
| 806 | * readers may still be accessing the elements. Releasing of resources |
| 807 | * must occur in a compatible manner. Then frees the bucket array. |
| 808 | * |
| 809 | * This function will eventually sleep to wait for an async resize |
| 810 | * to complete. The caller is responsible that no further write operations |
| 811 | * occurs in parallel. |
| 812 | */ |
| 813 | void rhashtable_free_and_destroy(struct rhashtable *ht, |
| 814 | void (*free_fn)(void *ptr, void *arg), |
| 815 | void *arg) |
| 816 | { |
| 817 | const struct bucket_table *tbl; |
| 818 | unsigned int i; |
| 819 | |
| 820 | cancel_work_sync(&ht->run_work); |
| 821 | |
| 822 | mutex_lock(&ht->mutex); |
| 823 | tbl = rht_dereference(ht->tbl, ht); |
| 824 | if (free_fn) { |
| 825 | for (i = 0; i < tbl->size; i++) { |
| 826 | struct rhash_head *pos, *next; |
| 827 | |
| 828 | for (pos = rht_dereference(tbl->buckets[i], ht), |
| 829 | next = !rht_is_a_nulls(pos) ? |
| 830 | rht_dereference(pos->next, ht) : NULL; |
| 831 | !rht_is_a_nulls(pos); |
| 832 | pos = next, |
| 833 | next = !rht_is_a_nulls(pos) ? |
| 834 | rht_dereference(pos->next, ht) : NULL) |
| 835 | free_fn(rht_obj(ht, pos), arg); |
| 836 | } |
| 837 | } |
| 838 | |
| 839 | bucket_table_free(tbl); |
| 840 | mutex_unlock(&ht->mutex); |
| 841 | } |
| 842 | EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy); |
| 843 | |
| 844 | void rhashtable_destroy(struct rhashtable *ht) |
| 845 | { |
| 846 | return rhashtable_free_and_destroy(ht, NULL, NULL); |
| 847 | } |
| 848 | EXPORT_SYMBOL_GPL(rhashtable_destroy); |