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 #include <linux/export.h>
  30
  31 #define HASH_DEFAULT_SIZE       64UL
  32 #define HASH_MIN_SIZE           4U
  33 #define BUCKET_LOCKS_PER_CPU    32UL
  34
  35 static u32 head_hashfn(struct rhashtable *ht,
  36                        const struct bucket_table *tbl,
  37                        const struct rhash_head *he)
  38 {
  39         return rht_head_hashfn(ht, tbl, he, ht->p);
  40 }
  41
  42 #ifdef CONFIG_PROVE_LOCKING
  43 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
  44
  45 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
  46 {
  47         return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
  48 }
  49 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
  50
  51 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
  52 {
  53         spinlock_t *lock = rht_bucket_lock(tbl, hash);
  54
  55         return (debug_locks) ? lockdep_is_held(lock) : 1;
  56 }
  57 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
  58 #else
  59 #define ASSERT_RHT_MUTEX(HT)
  60 #endif
  61
  62
  63 static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl,
  64                               gfp_t gfp)
  65 {
  66         unsigned int i, size;
  67 #if defined(CONFIG_PROVE_LOCKING)
  68         unsigned int nr_pcpus = 2;
  69 #else
  70         unsigned int nr_pcpus = num_possible_cpus();
  71 #endif
  72
  73         nr_pcpus = min_t(unsigned int, nr_pcpus, 64UL);
  74         size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
  75
  76         /* Never allocate more than 0.5 locks per bucket */
  77         size = min_t(unsigned int, size, tbl->size >> 1);
  78
  79         if (sizeof(spinlock_t) != 0) {
  80                 tbl->locks = NULL;
  81 #ifdef CONFIG_NUMA
  82                 if (size * sizeof(spinlock_t) > PAGE_SIZE &&
  83                     gfp == GFP_KERNEL)
  84                         tbl->locks = vmalloc(size * sizeof(spinlock_t));
  85 #endif
  86                 if (gfp != GFP_KERNEL)
  87                         gfp |= __GFP_NOWARN | __GFP_NORETRY;
  88
  89                 if (!tbl->locks)
  90                         tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
  91                                                    gfp);
  92                 if (!tbl->locks)
  93                         return -ENOMEM;
  94                 for (i = 0; i < size; i++)
  95                         spin_lock_init(&tbl->locks[i]);
  96         }
  97         tbl->locks_mask = size - 1;
  98
  99         return 0;
 100 }
 101
 102 static void bucket_table_free(const struct bucket_table *tbl)
 103 {
 104         if (tbl)
 105                 kvfree(tbl->locks);
 106
 107         kvfree(tbl);
 108 }
 109
 110 static void bucket_table_free_rcu(struct rcu_head *head)
 111 {
 112         bucket_table_free(container_of(head, struct bucket_table, rcu));
 113 }
 114
 115 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
 116                                                size_t nbuckets,
 117                                                gfp_t gfp)
 118 {
 119         struct bucket_table *tbl = NULL;
 120         size_t size;
 121         int i;
 122
 123         size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
 124         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) ||
 125             gfp != GFP_KERNEL)
 126                 tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
 127         if (tbl == NULL && gfp == GFP_KERNEL)
 128                 tbl = vzalloc(size);
 129         if (tbl == NULL)
 130                 return NULL;
 131
 132         tbl->size = nbuckets;
 133
 134         if (alloc_bucket_locks(ht, tbl, gfp) < 0) {
 135                 bucket_table_free(tbl);
 136                 return NULL;
 137         }
 138
 139         INIT_LIST_HEAD(&tbl->walkers);
 140
 141         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 142
 143         for (i = 0; i < nbuckets; i++)
 144                 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
 145
 146         return tbl;
 147 }
 148
 149 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
 150                                                   struct bucket_table *tbl)
 151 {
 152         struct bucket_table *new_tbl;
 153
 154         do {
 155                 new_tbl = tbl;
 156                 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 157         } while (tbl);
 158
 159         return new_tbl;
 160 }
 161
 162 static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
 163 {
 164         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 165         struct bucket_table *new_tbl = rhashtable_last_table(ht,
 166                 rht_dereference_rcu(old_tbl->future_tbl, ht));
 167         struct rhash_head __rcu **pprev = &old_tbl->buckets[old_hash];
 168         int err = -ENOENT;
 169         struct rhash_head *head, *next, *entry;
 170         spinlock_t *new_bucket_lock;
 171         unsigned int new_hash;
 172
 173         rht_for_each(entry, old_tbl, old_hash) {
 174                 err = 0;
 175                 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
 176
 177                 if (rht_is_a_nulls(next))
 178                         break;
 179
 180                 pprev = &entry->next;
 181         }
 182
 183         if (err)
 184                 goto out;
 185
 186         new_hash = head_hashfn(ht, new_tbl, entry);
 187
 188         new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
 189
 190         spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
 191         head = rht_dereference_bucket(new_tbl->buckets[new_hash],
 192                                       new_tbl, new_hash);
 193
 194         RCU_INIT_POINTER(entry->next, head);
 195
 196         rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
 197         spin_unlock(new_bucket_lock);
 198
 199         rcu_assign_pointer(*pprev, next);
 200
 201 out:
 202         return err;
 203 }
 204
 205 static void rhashtable_rehash_chain(struct rhashtable *ht,
 206                                     unsigned int old_hash)
 207 {
 208         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 209         spinlock_t *old_bucket_lock;
 210
 211         old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
 212
 213         spin_lock_bh(old_bucket_lock);
 214         while (!rhashtable_rehash_one(ht, old_hash))
 215                 ;
 216         old_tbl->rehash++;
 217         spin_unlock_bh(old_bucket_lock);
 218 }
 219
 220 static int rhashtable_rehash_attach(struct rhashtable *ht,
 221                                     struct bucket_table *old_tbl,
 222                                     struct bucket_table *new_tbl)
 223 {
 224         /* Protect future_tbl using the first bucket lock. */
 225         spin_lock_bh(old_tbl->locks);
 226
 227         /* Did somebody beat us to it? */
 228         if (rcu_access_pointer(old_tbl->future_tbl)) {
 229                 spin_unlock_bh(old_tbl->locks);
 230                 return -EEXIST;
 231         }
 232
 233         /* Make insertions go into the new, empty table right away. Deletions
 234          * and lookups will be attempted in both tables until we synchronize.
 235          */
 236         rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
 237
 238         spin_unlock_bh(old_tbl->locks);
 239
 240         return 0;
 241 }
 242
 243 static int rhashtable_rehash_table(struct rhashtable *ht)
 244 {
 245         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 246         struct bucket_table *new_tbl;
 247         struct rhashtable_walker *walker;
 248         unsigned int old_hash;
 249
 250         new_tbl = rht_dereference(old_tbl->future_tbl, ht);
 251         if (!new_tbl)
 252                 return 0;
 253
 254         for (old_hash = 0; old_hash < old_tbl->size; old_hash++)
 255                 rhashtable_rehash_chain(ht, old_hash);
 256
 257         /* Publish the new table pointer. */
 258         rcu_assign_pointer(ht->tbl, new_tbl);
 259
 260         spin_lock(&ht->lock);
 261         list_for_each_entry(walker, &old_tbl->walkers, list)
 262                 walker->tbl = NULL;
 263         spin_unlock(&ht->lock);
 264
 265         /* Wait for readers. All new readers will see the new
 266          * table, and thus no references to the old table will
 267          * remain.
 268          */
 269         call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
 270
 271         return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
 272 }
 273
 274 /**
 275  * rhashtable_expand - Expand hash table while allowing concurrent lookups
 276  * @ht:         the hash table to expand
 277  *
 278  * A secondary bucket array is allocated and the hash entries are migrated.
 279  *
 280  * This function may only be called in a context where it is safe to call
 281  * synchronize_rcu(), e.g. not within a rcu_read_lock() section.
 282  *
 283  * The caller must ensure that no concurrent resizing occurs by holding
 284  * ht->mutex.
 285  *
 286  * It is valid to have concurrent insertions and deletions protected by per
 287  * bucket locks or concurrent RCU protected lookups and traversals.
 288  */
 289 static int rhashtable_expand(struct rhashtable *ht)
 290 {
 291         struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
 292         int err;
 293
 294         ASSERT_RHT_MUTEX(ht);
 295
 296         old_tbl = rhashtable_last_table(ht, old_tbl);
 297
 298         new_tbl = bucket_table_alloc(ht, old_tbl->size * 2, GFP_KERNEL);
 299         if (new_tbl == NULL)
 300                 return -ENOMEM;
 301
 302         err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
 303         if (err)
 304                 bucket_table_free(new_tbl);
 305
 306         return err;
 307 }
 308
 309 /**
 310  * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
 311  * @ht:         the hash table to shrink
 312  *
 313  * This function shrinks the hash table to fit, i.e., the smallest
 314  * size would not cause it to expand right away automatically.
 315  *
 316  * The caller must ensure that no concurrent resizing occurs by holding
 317  * ht->mutex.
 318  *
 319  * The caller must ensure that no concurrent table mutations take place.
 320  * It is however valid to have concurrent lookups if they are RCU protected.
 321  *
 322  * It is valid to have concurrent insertions and deletions protected by per
 323  * bucket locks or concurrent RCU protected lookups and traversals.
 324  */
 325 static int rhashtable_shrink(struct rhashtable *ht)
 326 {
 327         struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
 328         unsigned int nelems = atomic_read(&ht->nelems);
 329         unsigned int size = 0;
 330         int err;
 331
 332         ASSERT_RHT_MUTEX(ht);
 333
 334         if (nelems)
 335                 size = roundup_pow_of_two(nelems * 3 / 2);
 336         if (size < ht->p.min_size)
 337                 size = ht->p.min_size;
 338
 339         if (old_tbl->size <= size)
 340                 return 0;
 341
 342         if (rht_dereference(old_tbl->future_tbl, ht))
 343                 return -EEXIST;
 344
 345         new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
 346         if (new_tbl == NULL)
 347                 return -ENOMEM;
 348
 349         err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
 350         if (err)
 351                 bucket_table_free(new_tbl);
 352
 353         return err;
 354 }
 355
 356 static void rht_deferred_worker(struct work_struct *work)
 357 {
 358         struct rhashtable *ht;
 359         struct bucket_table *tbl;
 360         int err = 0;
 361
 362         ht = container_of(work, struct rhashtable, run_work);
 363         mutex_lock(&ht->mutex);
 364
 365         tbl = rht_dereference(ht->tbl, ht);
 366         tbl = rhashtable_last_table(ht, tbl);
 367
 368         if (rht_grow_above_75(ht, tbl))
 369                 rhashtable_expand(ht);
 370         else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
 371                 rhashtable_shrink(ht);
 372
 373         err = rhashtable_rehash_table(ht);
 374
 375         mutex_unlock(&ht->mutex);
 376
 377         if (err)
 378                 schedule_work(&ht->run_work);
 379 }
 380
 381 static bool rhashtable_check_elasticity(struct rhashtable *ht,
 382                                         struct bucket_table *tbl,
 383                                         unsigned int hash)
 384 {
 385         unsigned int elasticity = ht->elasticity;
 386         struct rhash_head *head;
 387
 388         rht_for_each(head, tbl, hash)
 389                 if (!--elasticity)
 390                         return true;
 391
 392         return false;
 393 }
 394
 395 int rhashtable_insert_rehash(struct rhashtable *ht,
 396                              struct bucket_table *tbl)
 397 {
 398         struct bucket_table *old_tbl;
 399         struct bucket_table *new_tbl;
 400         unsigned int size;
 401         int err;
 402
 403         old_tbl = rht_dereference_rcu(ht->tbl, ht);
 404
 405         size = tbl->size;
 406
 407         err = -EBUSY;
 408
 409         if (rht_grow_above_75(ht, tbl))
 410                 size *= 2;
 411         /* Do not schedule more than one rehash */
 412         else if (old_tbl != tbl)
 413                 goto fail;
 414
 415         err = -ENOMEM;
 416
 417         new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
 418         if (new_tbl == NULL)
 419                 goto fail;
 420
 421         err = rhashtable_rehash_attach(ht, tbl, new_tbl);
 422         if (err) {
 423                 bucket_table_free(new_tbl);
 424                 if (err == -EEXIST)
 425                         err = 0;
 426         } else
 427                 schedule_work(&ht->run_work);
 428
 429         return err;
 430
 431 fail:
 432         /* Do not fail the insert if someone else did a rehash. */
 433         if (likely(rcu_dereference_raw(tbl->future_tbl)))
 434                 return 0;
 435
 436         /* Schedule async rehash to retry allocation in process context. */
 437         if (err == -ENOMEM)
 438                 schedule_work(&ht->run_work);
 439
 440         return err;
 441 }
 442 EXPORT_SYMBOL_GPL(rhashtable_insert_rehash);
 443
 444 struct bucket_table *rhashtable_insert_slow(struct rhashtable *ht,
 445                                             const void *key,
 446                                             struct rhash_head *obj,
 447                                             struct bucket_table *tbl)
 448 {
 449         struct rhash_head *head;
 450         unsigned int hash;
 451         int err;
 452
 453         tbl = rhashtable_last_table(ht, tbl);
 454         hash = head_hashfn(ht, tbl, obj);
 455         spin_lock_nested(rht_bucket_lock(tbl, hash), SINGLE_DEPTH_NESTING);
 456
 457         err = -EEXIST;
 458         if (key && rhashtable_lookup_fast(ht, key, ht->p))
 459                 goto exit;
 460
 461         err = -E2BIG;
 462         if (unlikely(rht_grow_above_max(ht, tbl)))
 463                 goto exit;
 464
 465         err = -EAGAIN;
 466         if (rhashtable_check_elasticity(ht, tbl, hash) ||
 467             rht_grow_above_100(ht, tbl))
 468                 goto exit;
 469
 470         err = 0;
 471
 472         head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
 473
 474         RCU_INIT_POINTER(obj->next, head);
 475
 476         rcu_assign_pointer(tbl->buckets[hash], obj);
 477
 478         atomic_inc(&ht->nelems);
 479
 480 exit:
 481         spin_unlock(rht_bucket_lock(tbl, hash));
 482
 483         if (err == 0)
 484                 return NULL;
 485         else if (err == -EAGAIN)
 486                 return tbl;
 487         else
 488                 return ERR_PTR(err);
 489 }
 490 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
 491
 492 /**
 493  * rhashtable_walk_init - Initialise an iterator
 494  * @ht:         Table to walk over
 495  * @iter:       Hash table Iterator
 496  * @gfp:        GFP flags for allocations
 497  *
 498  * This function prepares a hash table walk.
 499  *
 500  * Note that if you restart a walk after rhashtable_walk_stop you
 501  * may see the same object twice.  Also, you may miss objects if
 502  * there are removals in between rhashtable_walk_stop and the next
 503  * call to rhashtable_walk_start.
 504  *
 505  * For a completely stable walk you should construct your own data
 506  * structure outside the hash table.
 507  *
 508  * This function may sleep so you must not call it from interrupt
 509  * context or with spin locks held.
 510  *
 511  * You must call rhashtable_walk_exit if this function returns
 512  * successfully.
 513  */
 514 int rhashtable_walk_init(struct rhashtable *ht, struct rhashtable_iter *iter,
 515                          gfp_t gfp)
 516 {
 517         iter->ht = ht;
 518         iter->p = NULL;
 519         iter->slot = 0;
 520         iter->skip = 0;
 521
 522         iter->walker = kmalloc(sizeof(*iter->walker), gfp);
 523         if (!iter->walker)
 524                 return -ENOMEM;
 525
 526         spin_lock(&ht->lock);
 527         iter->walker->tbl =
 528                 rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
 529         list_add(&iter->walker->list, &iter->walker->tbl->walkers);
 530         spin_unlock(&ht->lock);
 531
 532         return 0;
 533 }
 534 EXPORT_SYMBOL_GPL(rhashtable_walk_init);
 535
 536 /**
 537  * rhashtable_walk_exit - Free an iterator
 538  * @iter:       Hash table Iterator
 539  *
 540  * This function frees resources allocated by rhashtable_walk_init.
 541  */
 542 void rhashtable_walk_exit(struct rhashtable_iter *iter)
 543 {
 544         spin_lock(&iter->ht->lock);
 545         if (iter->walker->tbl)
 546                 list_del(&iter->walker->list);
 547         spin_unlock(&iter->ht->lock);
 548         kfree(iter->walker);
 549 }
 550 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
 551
 552 /**
 553  * rhashtable_walk_start - Start a hash table walk
 554  * @iter:       Hash table iterator
 555  *
 556  * Start a hash table walk.  Note that we take the RCU lock in all
 557  * cases including when we return an error.  So you must always call
 558  * rhashtable_walk_stop to clean up.
 559  *
 560  * Returns zero if successful.
 561  *
 562  * Returns -EAGAIN if resize event occured.  Note that the iterator
 563  * will rewind back to the beginning and you may use it immediately
 564  * by calling rhashtable_walk_next.
 565  */
 566 int rhashtable_walk_start(struct rhashtable_iter *iter)
 567         __acquires(RCU)
 568 {
 569         struct rhashtable *ht = iter->ht;
 570
 571         rcu_read_lock();
 572
 573         spin_lock(&ht->lock);
 574         if (iter->walker->tbl)
 575                 list_del(&iter->walker->list);
 576         spin_unlock(&ht->lock);
 577
 578         if (!iter->walker->tbl) {
 579                 iter->walker->tbl = rht_dereference_rcu(ht->tbl, ht);
 580                 return -EAGAIN;
 581         }
 582
 583         return 0;
 584 }
 585 EXPORT_SYMBOL_GPL(rhashtable_walk_start);
 586
 587 /**
 588  * rhashtable_walk_next - Return the next object and advance the iterator
 589  * @iter:       Hash table iterator
 590  *
 591  * Note that you must call rhashtable_walk_stop when you are finished
 592  * with the walk.
 593  *
 594  * Returns the next object or NULL when the end of the table is reached.
 595  *
 596  * Returns -EAGAIN if resize event occured.  Note that the iterator
 597  * will rewind back to the beginning and you may continue to use it.
 598  */
 599 void *rhashtable_walk_next(struct rhashtable_iter *iter)
 600 {
 601         struct bucket_table *tbl = iter->walker->tbl;
 602         struct rhashtable *ht = iter->ht;
 603         struct rhash_head *p = iter->p;
 604
 605         if (p) {
 606                 p = rht_dereference_bucket_rcu(p->next, tbl, iter->slot);
 607                 goto next;
 608         }
 609
 610         for (; iter->slot < tbl->size; iter->slot++) {
 611                 int skip = iter->skip;
 612
 613                 rht_for_each_rcu(p, tbl, iter->slot) {
 614                         if (!skip)
 615                                 break;
 616                         skip--;
 617                 }
 618
 619 next:
 620                 if (!rht_is_a_nulls(p)) {
 621                         iter->skip++;
 622                         iter->p = p;
 623                         return rht_obj(ht, p);
 624                 }
 625
 626                 iter->skip = 0;
 627         }
 628
 629         iter->p = NULL;
 630
 631         /* Ensure we see any new tables. */
 632         smp_rmb();
 633
 634         iter->walker->tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 635         if (iter->walker->tbl) {
 636                 iter->slot = 0;
 637                 iter->skip = 0;
 638                 return ERR_PTR(-EAGAIN);
 639         }
 640
 641         return NULL;
 642 }
 643 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
 644
 645 /**
 646  * rhashtable_walk_stop - Finish a hash table walk
 647  * @iter:       Hash table iterator
 648  *
 649  * Finish a hash table walk.
 650  */
 651 void rhashtable_walk_stop(struct rhashtable_iter *iter)
 652         __releases(RCU)
 653 {
 654         struct rhashtable *ht;
 655         struct bucket_table *tbl = iter->walker->tbl;
 656
 657         if (!tbl)
 658                 goto out;
 659
 660         ht = iter->ht;
 661
 662         spin_lock(&ht->lock);
 663         if (tbl->rehash < tbl->size)
 664                 list_add(&iter->walker->list, &tbl->walkers);
 665         else
 666                 iter->walker->tbl = NULL;
 667         spin_unlock(&ht->lock);
 668
 669         iter->p = NULL;
 670
 671 out:
 672         rcu_read_unlock();
 673 }
 674 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
 675
 676 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
 677 {
 678         return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
 679                    (unsigned long)params->min_size);
 680 }
 681
 682 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
 683 {
 684         return jhash2(key, length, seed);
 685 }
 686
 687 /**
 688  * rhashtable_init - initialize a new hash table
 689  * @ht:         hash table to be initialized
 690  * @params:     configuration parameters
 691  *
 692  * Initializes a new hash table based on the provided configuration
 693  * parameters. A table can be configured either with a variable or
 694  * fixed length key:
 695  *
 696  * Configuration Example 1: Fixed length keys
 697  * struct test_obj {
 698  *      int                     key;
 699  *      void *                  my_member;
 700  *      struct rhash_head       node;
 701  * };
 702  *
 703  * struct rhashtable_params params = {
 704  *      .head_offset = offsetof(struct test_obj, node),
 705  *      .key_offset = offsetof(struct test_obj, key),
 706  *      .key_len = sizeof(int),
 707  *      .hashfn = jhash,
 708  *      .nulls_base = (1U << RHT_BASE_SHIFT),
 709  * };
 710  *
 711  * Configuration Example 2: Variable length keys
 712  * struct test_obj {
 713  *      [...]
 714  *      struct rhash_head       node;
 715  * };
 716  *
 717  * u32 my_hash_fn(const void *data, u32 len, u32 seed)
 718  * {
 719  *      struct test_obj *obj = data;
 720  *
 721  *      return [... hash ...];
 722  * }
 723  *
 724  * struct rhashtable_params params = {
 725  *      .head_offset = offsetof(struct test_obj, node),
 726  *      .hashfn = jhash,
 727  *      .obj_hashfn = my_hash_fn,
 728  * };
 729  */
 730 int rhashtable_init(struct rhashtable *ht,
 731                     const struct rhashtable_params *params)
 732 {
 733         struct bucket_table *tbl;
 734         size_t size;
 735
 736         size = HASH_DEFAULT_SIZE;
 737
 738         if ((!params->key_len && !params->obj_hashfn) ||
 739             (params->obj_hashfn && !params->obj_cmpfn))
 740                 return -EINVAL;
 741
 742         if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
 743                 return -EINVAL;
 744
 745         memset(ht, 0, sizeof(*ht));
 746         mutex_init(&ht->mutex);
 747         spin_lock_init(&ht->lock);
 748         memcpy(&ht->p, params, sizeof(*params));
 749
 750         if (params->min_size)
 751                 ht->p.min_size = roundup_pow_of_two(params->min_size);
 752
 753         if (params->max_size)
 754                 ht->p.max_size = rounddown_pow_of_two(params->max_size);
 755
 756         if (params->insecure_max_entries)
 757                 ht->p.insecure_max_entries =
 758                         rounddown_pow_of_two(params->insecure_max_entries);
 759         else
 760                 ht->p.insecure_max_entries = ht->p.max_size * 2;
 761
 762         ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE);
 763
 764         if (params->nelem_hint)
 765                 size = rounded_hashtable_size(&ht->p);
 766
 767         /* The maximum (not average) chain length grows with the
 768          * size of the hash table, at a rate of (log N)/(log log N).
 769          * The value of 16 is selected so that even if the hash
 770          * table grew to 2^32 you would not expect the maximum
 771          * chain length to exceed it unless we are under attack
 772          * (or extremely unlucky).
 773          *
 774          * As this limit is only to detect attacks, we don't need
 775          * to set it to a lower value as you'd need the chain
 776          * length to vastly exceed 16 to have any real effect
 777          * on the system.
 778          */
 779         if (!params->insecure_elasticity)
 780                 ht->elasticity = 16;
 781
 782         if (params->locks_mul)
 783                 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
 784         else
 785                 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
 786
 787         ht->key_len = ht->p.key_len;
 788         if (!params->hashfn) {
 789                 ht->p.hashfn = jhash;
 790
 791                 if (!(ht->key_len & (sizeof(u32) - 1))) {
 792                         ht->key_len /= sizeof(u32);
 793                         ht->p.hashfn = rhashtable_jhash2;
 794                 }
 795         }
 796
 797         tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
 798         if (tbl == NULL)
 799                 return -ENOMEM;
 800
 801         atomic_set(&ht->nelems, 0);
 802
 803         RCU_INIT_POINTER(ht->tbl, tbl);
 804
 805         INIT_WORK(&ht->run_work, rht_deferred_worker);
 806
 807         return 0;
 808 }
 809 EXPORT_SYMBOL_GPL(rhashtable_init);
 810
 811 /**
 812  * rhashtable_free_and_destroy - free elements and destroy hash table
 813  * @ht:         the hash table to destroy
 814  * @free_fn:    callback to release resources of element
 815  * @arg:        pointer passed to free_fn
 816  *
 817  * Stops an eventual async resize. If defined, invokes free_fn for each
 818  * element to releasal resources. Please note that RCU protected
 819  * readers may still be accessing the elements. Releasing of resources
 820  * must occur in a compatible manner. Then frees the bucket array.
 821  *
 822  * This function will eventually sleep to wait for an async resize
 823  * to complete. The caller is responsible that no further write operations
 824  * occurs in parallel.
 825  */
 826 void rhashtable_free_and_destroy(struct rhashtable *ht,
 827                                  void (*free_fn)(void *ptr, void *arg),
 828                                  void *arg)
 829 {
 830         const struct bucket_table *tbl;
 831         unsigned int i;
 832
 833         cancel_work_sync(&ht->run_work);
 834
 835         mutex_lock(&ht->mutex);
 836         tbl = rht_dereference(ht->tbl, ht);
 837         if (free_fn) {
 838                 for (i = 0; i < tbl->size; i++) {
 839                         struct rhash_head *pos, *next;
 840
 841                         for (pos = rht_dereference(tbl->buckets[i], ht),
 842                              next = !rht_is_a_nulls(pos) ?
 843                                         rht_dereference(pos->next, ht) : NULL;
 844                              !rht_is_a_nulls(pos);
 845                              pos = next,
 846                              next = !rht_is_a_nulls(pos) ?
 847                                         rht_dereference(pos->next, ht) : NULL)
 848                                 free_fn(rht_obj(ht, pos), arg);
 849                 }
 850         }
 851
 852         bucket_table_free(tbl);
 853         mutex_unlock(&ht->mutex);
 854 }
 855 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
 856
 857 void rhashtable_destroy(struct rhashtable *ht)
 858 {
 859         return rhashtable_free_and_destroy(ht, NULL, NULL);
 860 }
 861 EXPORT_SYMBOL_GPL(rhashtable_destroy);