lib/rhashtable.c

   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);