1 /* 2 * Resizable, Scalable, Concurrent Hash Table 3 * 4 * Copyright (c) 2015-2016 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 #ifndef _LINUX_RHASHTABLE_H 18 #define _LINUX_RHASHTABLE_H 19 20 #include <linux/atomic.h> 21 #include <linux/compiler.h> 22 #include <linux/err.h> 23 #include <linux/errno.h> 24 #include <linux/jhash.h> 25 #include <linux/list_nulls.h> 26 #include <linux/workqueue.h> 27 #include <linux/mutex.h> 28 #include <linux/rculist.h> 29 30 /* 31 * The end of the chain is marked with a special nulls marks which has 32 * the following format: 33 * 34 * +-------+-----------------------------------------------------+-+ 35 * | Base | Hash |1| 36 * +-------+-----------------------------------------------------+-+ 37 * 38 * Base (4 bits) : Reserved to distinguish between multiple tables. 39 * Specified via &struct rhashtable_params.nulls_base. 40 * Hash (27 bits): Full hash (unmasked) of first element added to bucket 41 * 1 (1 bit) : Nulls marker (always set) 42 * 43 * The remaining bits of the next pointer remain unused for now. 44 */ 45 #define RHT_BASE_BITS 4 46 #define RHT_HASH_BITS 27 47 #define RHT_BASE_SHIFT RHT_HASH_BITS 48 49 /* Base bits plus 1 bit for nulls marker */ 50 #define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1) 51 52 struct rhash_head { 53 struct rhash_head __rcu *next; 54 }; 55 56 struct rhlist_head { 57 struct rhash_head rhead; 58 struct rhlist_head __rcu *next; 59 }; 60 61 /** 62 * struct bucket_table - Table of hash buckets 63 * @size: Number of hash buckets 64 * @nest: Number of bits of first-level nested table. 65 * @rehash: Current bucket being rehashed 66 * @hash_rnd: Random seed to fold into hash 67 * @locks_mask: Mask to apply before accessing locks[] 68 * @locks: Array of spinlocks protecting individual buckets 69 * @walkers: List of active walkers 70 * @rcu: RCU structure for freeing the table 71 * @future_tbl: Table under construction during rehashing 72 * @ntbl: Nested table used when out of memory. 73 * @buckets: size * hash buckets 74 */ 75 struct bucket_table { 76 unsigned int size; 77 unsigned int nest; 78 unsigned int rehash; 79 u32 hash_rnd; 80 unsigned int locks_mask; 81 spinlock_t *locks; 82 struct list_head walkers; 83 struct rcu_head rcu; 84 85 struct bucket_table __rcu *future_tbl; 86 87 struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp; 88 }; 89 90 /** 91 * struct rhashtable_compare_arg - Key for the function rhashtable_compare 92 * @ht: Hash table 93 * @key: Key to compare against 94 */ 95 struct rhashtable_compare_arg { 96 struct rhashtable *ht; 97 const void *key; 98 }; 99 100 typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed); 101 typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed); 102 typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg, 103 const void *obj); 104 105 struct rhashtable; 106 107 /** 108 * struct rhashtable_params - Hash table construction parameters 109 * @nelem_hint: Hint on number of elements, should be 75% of desired size 110 * @key_len: Length of key 111 * @key_offset: Offset of key in struct to be hashed 112 * @head_offset: Offset of rhash_head in struct to be hashed 113 * @insecure_max_entries: Maximum number of entries (may be exceeded) 114 * @max_size: Maximum size while expanding 115 * @min_size: Minimum size while shrinking 116 * @nulls_base: Base value to generate nulls marker 117 * @insecure_elasticity: Set to true to disable chain length checks 118 * @automatic_shrinking: Enable automatic shrinking of tables 119 * @locks_mul: Number of bucket locks to allocate per cpu (default: 128) 120 * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash) 121 * @obj_hashfn: Function to hash object 122 * @obj_cmpfn: Function to compare key with object 123 */ 124 struct rhashtable_params { 125 size_t nelem_hint; 126 size_t key_len; 127 size_t key_offset; 128 size_t head_offset; 129 unsigned int insecure_max_entries; 130 unsigned int max_size; 131 unsigned int min_size; 132 u32 nulls_base; 133 bool insecure_elasticity; 134 bool automatic_shrinking; 135 size_t locks_mul; 136 rht_hashfn_t hashfn; 137 rht_obj_hashfn_t obj_hashfn; 138 rht_obj_cmpfn_t obj_cmpfn; 139 }; 140 141 /** 142 * struct rhashtable - Hash table handle 143 * @tbl: Bucket table 144 * @nelems: Number of elements in table 145 * @key_len: Key length for hashfn 146 * @elasticity: Maximum chain length before rehash 147 * @p: Configuration parameters 148 * @rhlist: True if this is an rhltable 149 * @run_work: Deferred worker to expand/shrink asynchronously 150 * @mutex: Mutex to protect current/future table swapping 151 * @lock: Spin lock to protect walker list 152 */ 153 struct rhashtable { 154 struct bucket_table __rcu *tbl; 155 atomic_t nelems; 156 unsigned int key_len; 157 unsigned int elasticity; 158 struct rhashtable_params p; 159 bool rhlist; 160 struct work_struct run_work; 161 struct mutex mutex; 162 spinlock_t lock; 163 }; 164 165 /** 166 * struct rhltable - Hash table with duplicate objects in a list 167 * @ht: Underlying rhtable 168 */ 169 struct rhltable { 170 struct rhashtable ht; 171 }; 172 173 /** 174 * struct rhashtable_walker - Hash table walker 175 * @list: List entry on list of walkers 176 * @tbl: The table that we were walking over 177 */ 178 struct rhashtable_walker { 179 struct list_head list; 180 struct bucket_table *tbl; 181 }; 182 183 /** 184 * struct rhashtable_iter - Hash table iterator 185 * @ht: Table to iterate through 186 * @p: Current pointer 187 * @list: Current hash list pointer 188 * @walker: Associated rhashtable walker 189 * @slot: Current slot 190 * @skip: Number of entries to skip in slot 191 */ 192 struct rhashtable_iter { 193 struct rhashtable *ht; 194 struct rhash_head *p; 195 struct rhlist_head *list; 196 struct rhashtable_walker walker; 197 unsigned int slot; 198 unsigned int skip; 199 }; 200 201 static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash) 202 { 203 return NULLS_MARKER(ht->p.nulls_base + hash); 204 } 205 206 #define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \ 207 ((ptr) = (typeof(ptr)) rht_marker(ht, hash)) 208 209 static inline bool rht_is_a_nulls(const struct rhash_head *ptr) 210 { 211 return ((unsigned long) ptr & 1); 212 } 213 214 static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr) 215 { 216 return ((unsigned long) ptr) >> 1; 217 } 218 219 static inline void *rht_obj(const struct rhashtable *ht, 220 const struct rhash_head *he) 221 { 222 return (char *)he - ht->p.head_offset; 223 } 224 225 static inline unsigned int rht_bucket_index(const struct bucket_table *tbl, 226 unsigned int hash) 227 { 228 return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1); 229 } 230 231 static inline unsigned int rht_key_hashfn( 232 struct rhashtable *ht, const struct bucket_table *tbl, 233 const void *key, const struct rhashtable_params params) 234 { 235 unsigned int hash; 236 237 /* params must be equal to ht->p if it isn't constant. */ 238 if (!__builtin_constant_p(params.key_len)) 239 hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd); 240 else if (params.key_len) { 241 unsigned int key_len = params.key_len; 242 243 if (params.hashfn) 244 hash = params.hashfn(key, key_len, tbl->hash_rnd); 245 else if (key_len & (sizeof(u32) - 1)) 246 hash = jhash(key, key_len, tbl->hash_rnd); 247 else 248 hash = jhash2(key, key_len / sizeof(u32), 249 tbl->hash_rnd); 250 } else { 251 unsigned int key_len = ht->p.key_len; 252 253 if (params.hashfn) 254 hash = params.hashfn(key, key_len, tbl->hash_rnd); 255 else 256 hash = jhash(key, key_len, tbl->hash_rnd); 257 } 258 259 return rht_bucket_index(tbl, hash); 260 } 261 262 static inline unsigned int rht_head_hashfn( 263 struct rhashtable *ht, const struct bucket_table *tbl, 264 const struct rhash_head *he, const struct rhashtable_params params) 265 { 266 const char *ptr = rht_obj(ht, he); 267 268 return likely(params.obj_hashfn) ? 269 rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?: 270 ht->p.key_len, 271 tbl->hash_rnd)) : 272 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params); 273 } 274 275 /** 276 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size 277 * @ht: hash table 278 * @tbl: current table 279 */ 280 static inline bool rht_grow_above_75(const struct rhashtable *ht, 281 const struct bucket_table *tbl) 282 { 283 /* Expand table when exceeding 75% load */ 284 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) && 285 (!ht->p.max_size || tbl->size < ht->p.max_size); 286 } 287 288 /** 289 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size 290 * @ht: hash table 291 * @tbl: current table 292 */ 293 static inline bool rht_shrink_below_30(const struct rhashtable *ht, 294 const struct bucket_table *tbl) 295 { 296 /* Shrink table beneath 30% load */ 297 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) && 298 tbl->size > ht->p.min_size; 299 } 300 301 /** 302 * rht_grow_above_100 - returns true if nelems > table-size 303 * @ht: hash table 304 * @tbl: current table 305 */ 306 static inline bool rht_grow_above_100(const struct rhashtable *ht, 307 const struct bucket_table *tbl) 308 { 309 return atomic_read(&ht->nelems) > tbl->size && 310 (!ht->p.max_size || tbl->size < ht->p.max_size); 311 } 312 313 /** 314 * rht_grow_above_max - returns true if table is above maximum 315 * @ht: hash table 316 * @tbl: current table 317 */ 318 static inline bool rht_grow_above_max(const struct rhashtable *ht, 319 const struct bucket_table *tbl) 320 { 321 return ht->p.insecure_max_entries && 322 atomic_read(&ht->nelems) >= ht->p.insecure_max_entries; 323 } 324 325 /* The bucket lock is selected based on the hash and protects mutations 326 * on a group of hash buckets. 327 * 328 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that 329 * a single lock always covers both buckets which may both contains 330 * entries which link to the same bucket of the old table during resizing. 331 * This allows to simplify the locking as locking the bucket in both 332 * tables during resize always guarantee protection. 333 * 334 * IMPORTANT: When holding the bucket lock of both the old and new table 335 * during expansions and shrinking, the old bucket lock must always be 336 * acquired first. 337 */ 338 static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl, 339 unsigned int hash) 340 { 341 return &tbl->locks[hash & tbl->locks_mask]; 342 } 343 344 #ifdef CONFIG_PROVE_LOCKING 345 int lockdep_rht_mutex_is_held(struct rhashtable *ht); 346 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash); 347 #else 348 static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht) 349 { 350 return 1; 351 } 352 353 static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, 354 u32 hash) 355 { 356 return 1; 357 } 358 #endif /* CONFIG_PROVE_LOCKING */ 359 360 int rhashtable_init(struct rhashtable *ht, 361 const struct rhashtable_params *params); 362 int rhltable_init(struct rhltable *hlt, 363 const struct rhashtable_params *params); 364 365 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key, 366 struct rhash_head *obj); 367 368 void rhashtable_walk_enter(struct rhashtable *ht, 369 struct rhashtable_iter *iter); 370 void rhashtable_walk_exit(struct rhashtable_iter *iter); 371 int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU); 372 void *rhashtable_walk_next(struct rhashtable_iter *iter); 373 void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU); 374 375 void rhashtable_free_and_destroy(struct rhashtable *ht, 376 void (*free_fn)(void *ptr, void *arg), 377 void *arg); 378 void rhashtable_destroy(struct rhashtable *ht); 379 380 struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl, 381 unsigned int hash); 382 struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht, 383 struct bucket_table *tbl, 384 unsigned int hash); 385 386 #define rht_dereference(p, ht) \ 387 rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht)) 388 389 #define rht_dereference_rcu(p, ht) \ 390 rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht)) 391 392 #define rht_dereference_bucket(p, tbl, hash) \ 393 rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash)) 394 395 #define rht_dereference_bucket_rcu(p, tbl, hash) \ 396 rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash)) 397 398 #define rht_entry(tpos, pos, member) \ 399 ({ tpos = container_of(pos, typeof(*tpos), member); 1; }) 400 401 static inline struct rhash_head __rcu *const *rht_bucket( 402 const struct bucket_table *tbl, unsigned int hash) 403 { 404 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) : 405 &tbl->buckets[hash]; 406 } 407 408 static inline struct rhash_head __rcu **rht_bucket_var( 409 struct bucket_table *tbl, unsigned int hash) 410 { 411 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) : 412 &tbl->buckets[hash]; 413 } 414 415 static inline struct rhash_head __rcu **rht_bucket_insert( 416 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash) 417 { 418 return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) : 419 &tbl->buckets[hash]; 420 } 421 422 /** 423 * rht_for_each_continue - continue iterating over hash chain 424 * @pos: the &struct rhash_head to use as a loop cursor. 425 * @head: the previous &struct rhash_head to continue from 426 * @tbl: the &struct bucket_table 427 * @hash: the hash value / bucket index 428 */ 429 #define rht_for_each_continue(pos, head, tbl, hash) \ 430 for (pos = rht_dereference_bucket(head, tbl, hash); \ 431 !rht_is_a_nulls(pos); \ 432 pos = rht_dereference_bucket((pos)->next, tbl, hash)) 433 434 /** 435 * rht_for_each - iterate over hash chain 436 * @pos: the &struct rhash_head to use as a loop cursor. 437 * @tbl: the &struct bucket_table 438 * @hash: the hash value / bucket index 439 */ 440 #define rht_for_each(pos, tbl, hash) \ 441 rht_for_each_continue(pos, *rht_bucket(tbl, hash), tbl, hash) 442 443 /** 444 * rht_for_each_entry_continue - continue iterating over hash chain 445 * @tpos: the type * to use as a loop cursor. 446 * @pos: the &struct rhash_head to use as a loop cursor. 447 * @head: the previous &struct rhash_head to continue from 448 * @tbl: the &struct bucket_table 449 * @hash: the hash value / bucket index 450 * @member: name of the &struct rhash_head within the hashable struct. 451 */ 452 #define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \ 453 for (pos = rht_dereference_bucket(head, tbl, hash); \ 454 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ 455 pos = rht_dereference_bucket((pos)->next, tbl, hash)) 456 457 /** 458 * rht_for_each_entry - iterate over hash chain of given type 459 * @tpos: the type * to use as a loop cursor. 460 * @pos: the &struct rhash_head to use as a loop cursor. 461 * @tbl: the &struct bucket_table 462 * @hash: the hash value / bucket index 463 * @member: name of the &struct rhash_head within the hashable struct. 464 */ 465 #define rht_for_each_entry(tpos, pos, tbl, hash, member) \ 466 rht_for_each_entry_continue(tpos, pos, *rht_bucket(tbl, hash), \ 467 tbl, hash, member) 468 469 /** 470 * rht_for_each_entry_safe - safely iterate over hash chain of given type 471 * @tpos: the type * to use as a loop cursor. 472 * @pos: the &struct rhash_head to use as a loop cursor. 473 * @next: the &struct rhash_head to use as next in loop cursor. 474 * @tbl: the &struct bucket_table 475 * @hash: the hash value / bucket index 476 * @member: name of the &struct rhash_head within the hashable struct. 477 * 478 * This hash chain list-traversal primitive allows for the looped code to 479 * remove the loop cursor from the list. 480 */ 481 #define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \ 482 for (pos = rht_dereference_bucket(*rht_bucket(tbl, hash), tbl, hash), \ 483 next = !rht_is_a_nulls(pos) ? \ 484 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \ 485 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ 486 pos = next, \ 487 next = !rht_is_a_nulls(pos) ? \ 488 rht_dereference_bucket(pos->next, tbl, hash) : NULL) 489 490 /** 491 * rht_for_each_rcu_continue - continue iterating over rcu hash chain 492 * @pos: the &struct rhash_head to use as a loop cursor. 493 * @head: the previous &struct rhash_head to continue from 494 * @tbl: the &struct bucket_table 495 * @hash: the hash value / bucket index 496 * 497 * This hash chain list-traversal primitive may safely run concurrently with 498 * the _rcu mutation primitives such as rhashtable_insert() as long as the 499 * traversal is guarded by rcu_read_lock(). 500 */ 501 #define rht_for_each_rcu_continue(pos, head, tbl, hash) \ 502 for (({barrier(); }), \ 503 pos = rht_dereference_bucket_rcu(head, tbl, hash); \ 504 !rht_is_a_nulls(pos); \ 505 pos = rcu_dereference_raw(pos->next)) 506 507 /** 508 * rht_for_each_rcu - iterate over rcu hash chain 509 * @pos: the &struct rhash_head to use as a loop cursor. 510 * @tbl: the &struct bucket_table 511 * @hash: the hash value / bucket index 512 * 513 * This hash chain list-traversal primitive may safely run concurrently with 514 * the _rcu mutation primitives such as rhashtable_insert() as long as the 515 * traversal is guarded by rcu_read_lock(). 516 */ 517 #define rht_for_each_rcu(pos, tbl, hash) \ 518 rht_for_each_rcu_continue(pos, *rht_bucket(tbl, hash), tbl, hash) 519 520 /** 521 * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain 522 * @tpos: the type * to use as a loop cursor. 523 * @pos: the &struct rhash_head to use as a loop cursor. 524 * @head: the previous &struct rhash_head to continue from 525 * @tbl: the &struct bucket_table 526 * @hash: the hash value / bucket index 527 * @member: name of the &struct rhash_head within the hashable struct. 528 * 529 * This hash chain list-traversal primitive may safely run concurrently with 530 * the _rcu mutation primitives such as rhashtable_insert() as long as the 531 * traversal is guarded by rcu_read_lock(). 532 */ 533 #define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \ 534 for (({barrier(); }), \ 535 pos = rht_dereference_bucket_rcu(head, tbl, hash); \ 536 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \ 537 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash)) 538 539 /** 540 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type 541 * @tpos: the type * to use as a loop cursor. 542 * @pos: the &struct rhash_head to use as a loop cursor. 543 * @tbl: the &struct bucket_table 544 * @hash: the hash value / bucket index 545 * @member: name of the &struct rhash_head within the hashable struct. 546 * 547 * This hash chain list-traversal primitive may safely run concurrently with 548 * the _rcu mutation primitives such as rhashtable_insert() as long as the 549 * traversal is guarded by rcu_read_lock(). 550 */ 551 #define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \ 552 rht_for_each_entry_rcu_continue(tpos, pos, *rht_bucket(tbl, hash), \ 553 tbl, hash, member) 554 555 /** 556 * rhl_for_each_rcu - iterate over rcu hash table list 557 * @pos: the &struct rlist_head to use as a loop cursor. 558 * @list: the head of the list 559 * 560 * This hash chain list-traversal primitive should be used on the 561 * list returned by rhltable_lookup. 562 */ 563 #define rhl_for_each_rcu(pos, list) \ 564 for (pos = list; pos; pos = rcu_dereference_raw(pos->next)) 565 566 /** 567 * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type 568 * @tpos: the type * to use as a loop cursor. 569 * @pos: the &struct rlist_head to use as a loop cursor. 570 * @list: the head of the list 571 * @member: name of the &struct rlist_head within the hashable struct. 572 * 573 * This hash chain list-traversal primitive should be used on the 574 * list returned by rhltable_lookup. 575 */ 576 #define rhl_for_each_entry_rcu(tpos, pos, list, member) \ 577 for (pos = list; pos && rht_entry(tpos, pos, member); \ 578 pos = rcu_dereference_raw(pos->next)) 579 580 static inline int rhashtable_compare(struct rhashtable_compare_arg *arg, 581 const void *obj) 582 { 583 struct rhashtable *ht = arg->ht; 584 const char *ptr = obj; 585 586 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len); 587 } 588 589 /* Internal function, do not use. */ 590 static inline struct rhash_head *__rhashtable_lookup( 591 struct rhashtable *ht, const void *key, 592 const struct rhashtable_params params) 593 { 594 struct rhashtable_compare_arg arg = { 595 .ht = ht, 596 .key = key, 597 }; 598 struct bucket_table *tbl; 599 struct rhash_head *he; 600 unsigned int hash; 601 602 tbl = rht_dereference_rcu(ht->tbl, ht); 603 restart: 604 hash = rht_key_hashfn(ht, tbl, key, params); 605 rht_for_each_rcu(he, tbl, hash) { 606 if (params.obj_cmpfn ? 607 params.obj_cmpfn(&arg, rht_obj(ht, he)) : 608 rhashtable_compare(&arg, rht_obj(ht, he))) 609 continue; 610 return he; 611 } 612 613 /* Ensure we see any new tables. */ 614 smp_rmb(); 615 616 tbl = rht_dereference_rcu(tbl->future_tbl, ht); 617 if (unlikely(tbl)) 618 goto restart; 619 620 return NULL; 621 } 622 623 /** 624 * rhashtable_lookup - search hash table 625 * @ht: hash table 626 * @key: the pointer to the key 627 * @params: hash table parameters 628 * 629 * Computes the hash value for the key and traverses the bucket chain looking 630 * for a entry with an identical key. The first matching entry is returned. 631 * 632 * This must only be called under the RCU read lock. 633 * 634 * Returns the first entry on which the compare function returned true. 635 */ 636 static inline void *rhashtable_lookup( 637 struct rhashtable *ht, const void *key, 638 const struct rhashtable_params params) 639 { 640 struct rhash_head *he = __rhashtable_lookup(ht, key, params); 641 642 return he ? rht_obj(ht, he) : NULL; 643 } 644 645 /** 646 * rhashtable_lookup_fast - search hash table, without RCU read lock 647 * @ht: hash table 648 * @key: the pointer to the key 649 * @params: hash table parameters 650 * 651 * Computes the hash value for the key and traverses the bucket chain looking 652 * for a entry with an identical key. The first matching entry is returned. 653 * 654 * Only use this function when you have other mechanisms guaranteeing 655 * that the object won't go away after the RCU read lock is released. 656 * 657 * Returns the first entry on which the compare function returned true. 658 */ 659 static inline void *rhashtable_lookup_fast( 660 struct rhashtable *ht, const void *key, 661 const struct rhashtable_params params) 662 { 663 void *obj; 664 665 rcu_read_lock(); 666 obj = rhashtable_lookup(ht, key, params); 667 rcu_read_unlock(); 668 669 return obj; 670 } 671 672 /** 673 * rhltable_lookup - search hash list table 674 * @hlt: hash table 675 * @key: the pointer to the key 676 * @params: hash table parameters 677 * 678 * Computes the hash value for the key and traverses the bucket chain looking 679 * for a entry with an identical key. All matching entries are returned 680 * in a list. 681 * 682 * This must only be called under the RCU read lock. 683 * 684 * Returns the list of entries that match the given key. 685 */ 686 static inline struct rhlist_head *rhltable_lookup( 687 struct rhltable *hlt, const void *key, 688 const struct rhashtable_params params) 689 { 690 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params); 691 692 return he ? container_of(he, struct rhlist_head, rhead) : NULL; 693 } 694 695 /* Internal function, please use rhashtable_insert_fast() instead. This 696 * function returns the existing element already in hashes in there is a clash, 697 * otherwise it returns an error via ERR_PTR(). 698 */ 699 static inline void *__rhashtable_insert_fast( 700 struct rhashtable *ht, const void *key, struct rhash_head *obj, 701 const struct rhashtable_params params, bool rhlist) 702 { 703 struct rhashtable_compare_arg arg = { 704 .ht = ht, 705 .key = key, 706 }; 707 struct rhash_head __rcu **pprev; 708 struct bucket_table *tbl; 709 struct rhash_head *head; 710 spinlock_t *lock; 711 unsigned int hash; 712 int elasticity; 713 void *data; 714 715 rcu_read_lock(); 716 717 tbl = rht_dereference_rcu(ht->tbl, ht); 718 hash = rht_head_hashfn(ht, tbl, obj, params); 719 lock = rht_bucket_lock(tbl, hash); 720 spin_lock_bh(lock); 721 722 if (unlikely(rht_dereference_bucket(tbl->future_tbl, tbl, hash))) { 723 slow_path: 724 spin_unlock_bh(lock); 725 rcu_read_unlock(); 726 return rhashtable_insert_slow(ht, key, obj); 727 } 728 729 elasticity = ht->elasticity; 730 pprev = rht_bucket_insert(ht, tbl, hash); 731 data = ERR_PTR(-ENOMEM); 732 if (!pprev) 733 goto out; 734 735 rht_for_each_continue(head, *pprev, tbl, hash) { 736 struct rhlist_head *plist; 737 struct rhlist_head *list; 738 739 elasticity--; 740 if (!key || 741 (params.obj_cmpfn ? 742 params.obj_cmpfn(&arg, rht_obj(ht, head)) : 743 rhashtable_compare(&arg, rht_obj(ht, head)))) 744 continue; 745 746 data = rht_obj(ht, head); 747 748 if (!rhlist) 749 goto out; 750 751 752 list = container_of(obj, struct rhlist_head, rhead); 753 plist = container_of(head, struct rhlist_head, rhead); 754 755 RCU_INIT_POINTER(list->next, plist); 756 head = rht_dereference_bucket(head->next, tbl, hash); 757 RCU_INIT_POINTER(list->rhead.next, head); 758 rcu_assign_pointer(*pprev, obj); 759 760 goto good; 761 } 762 763 if (elasticity <= 0) 764 goto slow_path; 765 766 data = ERR_PTR(-E2BIG); 767 if (unlikely(rht_grow_above_max(ht, tbl))) 768 goto out; 769 770 if (unlikely(rht_grow_above_100(ht, tbl))) 771 goto slow_path; 772 773 head = rht_dereference_bucket(*pprev, tbl, hash); 774 775 RCU_INIT_POINTER(obj->next, head); 776 if (rhlist) { 777 struct rhlist_head *list; 778 779 list = container_of(obj, struct rhlist_head, rhead); 780 RCU_INIT_POINTER(list->next, NULL); 781 } 782 783 rcu_assign_pointer(*pprev, obj); 784 785 atomic_inc(&ht->nelems); 786 if (rht_grow_above_75(ht, tbl)) 787 schedule_work(&ht->run_work); 788 789 good: 790 data = NULL; 791 792 out: 793 spin_unlock_bh(lock); 794 rcu_read_unlock(); 795 796 return data; 797 } 798 799 /** 800 * rhashtable_insert_fast - insert object into hash table 801 * @ht: hash table 802 * @obj: pointer to hash head inside object 803 * @params: hash table parameters 804 * 805 * Will take a per bucket spinlock to protect against mutual mutations 806 * on the same bucket. Multiple insertions may occur in parallel unless 807 * they map to the same bucket lock. 808 * 809 * It is safe to call this function from atomic context. 810 * 811 * Will trigger an automatic deferred table resizing if the size grows 812 * beyond the watermark indicated by grow_decision() which can be passed 813 * to rhashtable_init(). 814 */ 815 static inline int rhashtable_insert_fast( 816 struct rhashtable *ht, struct rhash_head *obj, 817 const struct rhashtable_params params) 818 { 819 void *ret; 820 821 ret = __rhashtable_insert_fast(ht, NULL, obj, params, false); 822 if (IS_ERR(ret)) 823 return PTR_ERR(ret); 824 825 return ret == NULL ? 0 : -EEXIST; 826 } 827 828 /** 829 * rhltable_insert_key - insert object into hash list table 830 * @hlt: hash list table 831 * @key: the pointer to the key 832 * @list: pointer to hash list head inside object 833 * @params: hash table parameters 834 * 835 * Will take a per bucket spinlock to protect against mutual mutations 836 * on the same bucket. Multiple insertions may occur in parallel unless 837 * they map to the same bucket lock. 838 * 839 * It is safe to call this function from atomic context. 840 * 841 * Will trigger an automatic deferred table resizing if the size grows 842 * beyond the watermark indicated by grow_decision() which can be passed 843 * to rhashtable_init(). 844 */ 845 static inline int rhltable_insert_key( 846 struct rhltable *hlt, const void *key, struct rhlist_head *list, 847 const struct rhashtable_params params) 848 { 849 return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead, 850 params, true)); 851 } 852 853 /** 854 * rhltable_insert - insert object into hash list table 855 * @hlt: hash list table 856 * @list: pointer to hash list head inside object 857 * @params: hash table parameters 858 * 859 * Will take a per bucket spinlock to protect against mutual mutations 860 * on the same bucket. Multiple insertions may occur in parallel unless 861 * they map to the same bucket lock. 862 * 863 * It is safe to call this function from atomic context. 864 * 865 * Will trigger an automatic deferred table resizing if the size grows 866 * beyond the watermark indicated by grow_decision() which can be passed 867 * to rhashtable_init(). 868 */ 869 static inline int rhltable_insert( 870 struct rhltable *hlt, struct rhlist_head *list, 871 const struct rhashtable_params params) 872 { 873 const char *key = rht_obj(&hlt->ht, &list->rhead); 874 875 key += params.key_offset; 876 877 return rhltable_insert_key(hlt, key, list, params); 878 } 879 880 /** 881 * rhashtable_lookup_insert_fast - lookup and insert object into hash table 882 * @ht: hash table 883 * @obj: pointer to hash head inside object 884 * @params: hash table parameters 885 * 886 * Locks down the bucket chain in both the old and new table if a resize 887 * is in progress to ensure that writers can't remove from the old table 888 * and can't insert to the new table during the atomic operation of search 889 * and insertion. Searches for duplicates in both the old and new table if 890 * a resize is in progress. 891 * 892 * This lookup function may only be used for fixed key hash table (key_len 893 * parameter set). It will BUG() if used inappropriately. 894 * 895 * It is safe to call this function from atomic context. 896 * 897 * Will trigger an automatic deferred table resizing if the size grows 898 * beyond the watermark indicated by grow_decision() which can be passed 899 * to rhashtable_init(). 900 */ 901 static inline int rhashtable_lookup_insert_fast( 902 struct rhashtable *ht, struct rhash_head *obj, 903 const struct rhashtable_params params) 904 { 905 const char *key = rht_obj(ht, obj); 906 void *ret; 907 908 BUG_ON(ht->p.obj_hashfn); 909 910 ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params, 911 false); 912 if (IS_ERR(ret)) 913 return PTR_ERR(ret); 914 915 return ret == NULL ? 0 : -EEXIST; 916 } 917 918 /** 919 * rhashtable_lookup_insert_key - search and insert object to hash table 920 * with explicit key 921 * @ht: hash table 922 * @key: key 923 * @obj: pointer to hash head inside object 924 * @params: hash table parameters 925 * 926 * Locks down the bucket chain in both the old and new table if a resize 927 * is in progress to ensure that writers can't remove from the old table 928 * and can't insert to the new table during the atomic operation of search 929 * and insertion. Searches for duplicates in both the old and new table if 930 * a resize is in progress. 931 * 932 * Lookups may occur in parallel with hashtable mutations and resizing. 933 * 934 * Will trigger an automatic deferred table resizing if the size grows 935 * beyond the watermark indicated by grow_decision() which can be passed 936 * to rhashtable_init(). 937 * 938 * Returns zero on success. 939 */ 940 static inline int rhashtable_lookup_insert_key( 941 struct rhashtable *ht, const void *key, struct rhash_head *obj, 942 const struct rhashtable_params params) 943 { 944 void *ret; 945 946 BUG_ON(!ht->p.obj_hashfn || !key); 947 948 ret = __rhashtable_insert_fast(ht, key, obj, params, false); 949 if (IS_ERR(ret)) 950 return PTR_ERR(ret); 951 952 return ret == NULL ? 0 : -EEXIST; 953 } 954 955 /** 956 * rhashtable_lookup_get_insert_key - lookup and insert object into hash table 957 * @ht: hash table 958 * @obj: pointer to hash head inside object 959 * @params: hash table parameters 960 * @data: pointer to element data already in hashes 961 * 962 * Just like rhashtable_lookup_insert_key(), but this function returns the 963 * object if it exists, NULL if it does not and the insertion was successful, 964 * and an ERR_PTR otherwise. 965 */ 966 static inline void *rhashtable_lookup_get_insert_key( 967 struct rhashtable *ht, const void *key, struct rhash_head *obj, 968 const struct rhashtable_params params) 969 { 970 BUG_ON(!ht->p.obj_hashfn || !key); 971 972 return __rhashtable_insert_fast(ht, key, obj, params, false); 973 } 974 975 /* Internal function, please use rhashtable_remove_fast() instead */ 976 static inline int __rhashtable_remove_fast_one( 977 struct rhashtable *ht, struct bucket_table *tbl, 978 struct rhash_head *obj, const struct rhashtable_params params, 979 bool rhlist) 980 { 981 struct rhash_head __rcu **pprev; 982 struct rhash_head *he; 983 spinlock_t * lock; 984 unsigned int hash; 985 int err = -ENOENT; 986 987 hash = rht_head_hashfn(ht, tbl, obj, params); 988 lock = rht_bucket_lock(tbl, hash); 989 990 spin_lock_bh(lock); 991 992 pprev = rht_bucket_var(tbl, hash); 993 rht_for_each_continue(he, *pprev, tbl, hash) { 994 struct rhlist_head *list; 995 996 list = container_of(he, struct rhlist_head, rhead); 997 998 if (he != obj) { 999 struct rhlist_head __rcu **lpprev; 1000 1001 pprev = &he->next; 1002 1003 if (!rhlist) 1004 continue; 1005 1006 do { 1007 lpprev = &list->next; 1008 list = rht_dereference_bucket(list->next, 1009 tbl, hash); 1010 } while (list && obj != &list->rhead); 1011 1012 if (!list) 1013 continue; 1014 1015 list = rht_dereference_bucket(list->next, tbl, hash); 1016 RCU_INIT_POINTER(*lpprev, list); 1017 err = 0; 1018 break; 1019 } 1020 1021 obj = rht_dereference_bucket(obj->next, tbl, hash); 1022 err = 1; 1023 1024 if (rhlist) { 1025 list = rht_dereference_bucket(list->next, tbl, hash); 1026 if (list) { 1027 RCU_INIT_POINTER(list->rhead.next, obj); 1028 obj = &list->rhead; 1029 err = 0; 1030 } 1031 } 1032 1033 rcu_assign_pointer(*pprev, obj); 1034 break; 1035 } 1036 1037 spin_unlock_bh(lock); 1038 1039 if (err > 0) { 1040 atomic_dec(&ht->nelems); 1041 if (unlikely(ht->p.automatic_shrinking && 1042 rht_shrink_below_30(ht, tbl))) 1043 schedule_work(&ht->run_work); 1044 err = 0; 1045 } 1046 1047 return err; 1048 } 1049 1050 /* Internal function, please use rhashtable_remove_fast() instead */ 1051 static inline int __rhashtable_remove_fast( 1052 struct rhashtable *ht, struct rhash_head *obj, 1053 const struct rhashtable_params params, bool rhlist) 1054 { 1055 struct bucket_table *tbl; 1056 int err; 1057 1058 rcu_read_lock(); 1059 1060 tbl = rht_dereference_rcu(ht->tbl, ht); 1061 1062 /* Because we have already taken (and released) the bucket 1063 * lock in old_tbl, if we find that future_tbl is not yet 1064 * visible then that guarantees the entry to still be in 1065 * the old tbl if it exists. 1066 */ 1067 while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params, 1068 rhlist)) && 1069 (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) 1070 ; 1071 1072 rcu_read_unlock(); 1073 1074 return err; 1075 } 1076 1077 /** 1078 * rhashtable_remove_fast - remove object from hash table 1079 * @ht: hash table 1080 * @obj: pointer to hash head inside object 1081 * @params: hash table parameters 1082 * 1083 * Since the hash chain is single linked, the removal operation needs to 1084 * walk the bucket chain upon removal. The removal operation is thus 1085 * considerable slow if the hash table is not correctly sized. 1086 * 1087 * Will automatically shrink the table via rhashtable_expand() if the 1088 * shrink_decision function specified at rhashtable_init() returns true. 1089 * 1090 * Returns zero on success, -ENOENT if the entry could not be found. 1091 */ 1092 static inline int rhashtable_remove_fast( 1093 struct rhashtable *ht, struct rhash_head *obj, 1094 const struct rhashtable_params params) 1095 { 1096 return __rhashtable_remove_fast(ht, obj, params, false); 1097 } 1098 1099 /** 1100 * rhltable_remove - remove object from hash list table 1101 * @hlt: hash list table 1102 * @list: pointer to hash list head inside object 1103 * @params: hash table parameters 1104 * 1105 * Since the hash chain is single linked, the removal operation needs to 1106 * walk the bucket chain upon removal. The removal operation is thus 1107 * considerable slow if the hash table is not correctly sized. 1108 * 1109 * Will automatically shrink the table via rhashtable_expand() if the 1110 * shrink_decision function specified at rhashtable_init() returns true. 1111 * 1112 * Returns zero on success, -ENOENT if the entry could not be found. 1113 */ 1114 static inline int rhltable_remove( 1115 struct rhltable *hlt, struct rhlist_head *list, 1116 const struct rhashtable_params params) 1117 { 1118 return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true); 1119 } 1120 1121 /* Internal function, please use rhashtable_replace_fast() instead */ 1122 static inline int __rhashtable_replace_fast( 1123 struct rhashtable *ht, struct bucket_table *tbl, 1124 struct rhash_head *obj_old, struct rhash_head *obj_new, 1125 const struct rhashtable_params params) 1126 { 1127 struct rhash_head __rcu **pprev; 1128 struct rhash_head *he; 1129 spinlock_t *lock; 1130 unsigned int hash; 1131 int err = -ENOENT; 1132 1133 /* Minimally, the old and new objects must have same hash 1134 * (which should mean identifiers are the same). 1135 */ 1136 hash = rht_head_hashfn(ht, tbl, obj_old, params); 1137 if (hash != rht_head_hashfn(ht, tbl, obj_new, params)) 1138 return -EINVAL; 1139 1140 lock = rht_bucket_lock(tbl, hash); 1141 1142 spin_lock_bh(lock); 1143 1144 pprev = rht_bucket_var(tbl, hash); 1145 rht_for_each_continue(he, *pprev, tbl, hash) { 1146 if (he != obj_old) { 1147 pprev = &he->next; 1148 continue; 1149 } 1150 1151 rcu_assign_pointer(obj_new->next, obj_old->next); 1152 rcu_assign_pointer(*pprev, obj_new); 1153 err = 0; 1154 break; 1155 } 1156 1157 spin_unlock_bh(lock); 1158 1159 return err; 1160 } 1161 1162 /** 1163 * rhashtable_replace_fast - replace an object in hash table 1164 * @ht: hash table 1165 * @obj_old: pointer to hash head inside object being replaced 1166 * @obj_new: pointer to hash head inside object which is new 1167 * @params: hash table parameters 1168 * 1169 * Replacing an object doesn't affect the number of elements in the hash table 1170 * or bucket, so we don't need to worry about shrinking or expanding the 1171 * table here. 1172 * 1173 * Returns zero on success, -ENOENT if the entry could not be found, 1174 * -EINVAL if hash is not the same for the old and new objects. 1175 */ 1176 static inline int rhashtable_replace_fast( 1177 struct rhashtable *ht, struct rhash_head *obj_old, 1178 struct rhash_head *obj_new, 1179 const struct rhashtable_params params) 1180 { 1181 struct bucket_table *tbl; 1182 int err; 1183 1184 rcu_read_lock(); 1185 1186 tbl = rht_dereference_rcu(ht->tbl, ht); 1187 1188 /* Because we have already taken (and released) the bucket 1189 * lock in old_tbl, if we find that future_tbl is not yet 1190 * visible then that guarantees the entry to still be in 1191 * the old tbl if it exists. 1192 */ 1193 while ((err = __rhashtable_replace_fast(ht, tbl, obj_old, 1194 obj_new, params)) && 1195 (tbl = rht_dereference_rcu(tbl->future_tbl, ht))) 1196 ; 1197 1198 rcu_read_unlock(); 1199 1200 return err; 1201 } 1202 1203 /* Obsolete function, do not use in new code. */ 1204 static inline int rhashtable_walk_init(struct rhashtable *ht, 1205 struct rhashtable_iter *iter, gfp_t gfp) 1206 { 1207 rhashtable_walk_enter(ht, iter); 1208 return 0; 1209 } 1210 1211 /** 1212 * rhltable_walk_enter - Initialise an iterator 1213 * @hlt: Table to walk over 1214 * @iter: Hash table Iterator 1215 * 1216 * This function prepares a hash table walk. 1217 * 1218 * Note that if you restart a walk after rhashtable_walk_stop you 1219 * may see the same object twice. Also, you may miss objects if 1220 * there are removals in between rhashtable_walk_stop and the next 1221 * call to rhashtable_walk_start. 1222 * 1223 * For a completely stable walk you should construct your own data 1224 * structure outside the hash table. 1225 * 1226 * This function may sleep so you must not call it from interrupt 1227 * context or with spin locks held. 1228 * 1229 * You must call rhashtable_walk_exit after this function returns. 1230 */ 1231 static inline void rhltable_walk_enter(struct rhltable *hlt, 1232 struct rhashtable_iter *iter) 1233 { 1234 return rhashtable_walk_enter(&hlt->ht, iter); 1235 } 1236 1237 /** 1238 * rhltable_free_and_destroy - free elements and destroy hash list table 1239 * @hlt: the hash list table to destroy 1240 * @free_fn: callback to release resources of element 1241 * @arg: pointer passed to free_fn 1242 * 1243 * See documentation for rhashtable_free_and_destroy. 1244 */ 1245 static inline void rhltable_free_and_destroy(struct rhltable *hlt, 1246 void (*free_fn)(void *ptr, 1247 void *arg), 1248 void *arg) 1249 { 1250 return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg); 1251 } 1252 1253 static inline void rhltable_destroy(struct rhltable *hlt) 1254 { 1255 return rhltable_free_and_destroy(hlt, NULL, NULL); 1256 } 1257 1258 #endif /* _LINUX_RHASHTABLE_H */ 1259