lib/rhashtable.c

   1 /*
   2  * Resizable, Scalable, Concurrent Hash Table
   3  *
   4  * Copyright (c) 2015 Herbert Xu <[email protected]>
   5  * Copyright (c) 2014-2015 Thomas Graf <[email protected]>
   6  * Copyright (c) 2008-2014 Patrick McHardy <[email protected]>
   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/rculist.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/mm.h>
  26 #include <linux/jhash.h>
  27 #include <linux/random.h>
  28 #include <linux/rhashtable.h>
  29 #include <linux/err.h>
  30 #include <linux/export.h>
  31
  32 #define HASH_DEFAULT_SIZE       64UL
  33 #define HASH_MIN_SIZE           4U
  34 #define BUCKET_LOCKS_PER_CPU    32UL
  35
  36 union nested_table {
  37         union nested_table __rcu *table;
  38         struct rhash_head __rcu *bucket;
  39 };
  40
  41 static u32 head_hashfn(struct rhashtable *ht,
  42                        const struct bucket_table *tbl,
  43                        const struct rhash_head *he)
  44 {
  45         return rht_head_hashfn(ht, tbl, he, ht->p);
  46 }
  47
  48 #ifdef CONFIG_PROVE_LOCKING
  49 #define ASSERT_RHT_MUTEX(HT) BUG_ON(!lockdep_rht_mutex_is_held(HT))
  50
  51 int lockdep_rht_mutex_is_held(struct rhashtable *ht)
  52 {
  53         return (debug_locks) ? lockdep_is_held(&ht->mutex) : 1;
  54 }
  55 EXPORT_SYMBOL_GPL(lockdep_rht_mutex_is_held);
  56
  57 int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash)
  58 {
  59         spinlock_t *lock = rht_bucket_lock(tbl, hash);
  60
  61         return (debug_locks) ? lockdep_is_held(lock) : 1;
  62 }
  63 EXPORT_SYMBOL_GPL(lockdep_rht_bucket_is_held);
  64 #else
  65 #define ASSERT_RHT_MUTEX(HT)
  66 #endif
  67
  68
  69 static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table *tbl,
  70                               gfp_t gfp)
  71 {
  72         unsigned int i, size;
  73 #if defined(CONFIG_PROVE_LOCKING)
  74         unsigned int nr_pcpus = 2;
  75 #else
  76         unsigned int nr_pcpus = num_possible_cpus();
  77 #endif
  78
  79         nr_pcpus = min_t(unsigned int, nr_pcpus, 64UL);
  80         size = roundup_pow_of_two(nr_pcpus * ht->p.locks_mul);
  81
  82         /* Never allocate more than 0.5 locks per bucket */
  83         size = min_t(unsigned int, size, tbl->size >> 1);
  84
  85         if (tbl->nest)
  86                 size = min(size, 1U << tbl->nest);
  87
  88         if (sizeof(spinlock_t) != 0) {
  89                 tbl->locks = NULL;
  90 #ifdef CONFIG_NUMA
  91                 if (size * sizeof(spinlock_t) > PAGE_SIZE &&
  92                     gfp == GFP_KERNEL)
  93                         tbl->locks = vmalloc(size * sizeof(spinlock_t));
  94 #endif
  95                 if (gfp != GFP_KERNEL)
  96                         gfp |= __GFP_NOWARN | __GFP_NORETRY;
  97
  98                 if (!tbl->locks)
  99                         tbl->locks = kmalloc_array(size, sizeof(spinlock_t),
 100                                                    gfp);
 101                 if (!tbl->locks)
 102                         return -ENOMEM;
 103                 for (i = 0; i < size; i++)
 104                         spin_lock_init(&tbl->locks[i]);
 105         }
 106         tbl->locks_mask = size - 1;
 107
 108         return 0;
 109 }
 110
 111 static void nested_table_free(union nested_table *ntbl, unsigned int size)
 112 {
 113         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
 114         const unsigned int len = 1 << shift;
 115         unsigned int i;
 116
 117         ntbl = rcu_dereference_raw(ntbl->table);
 118         if (!ntbl)
 119                 return;
 120
 121         if (size > len) {
 122                 size >>= shift;
 123                 for (i = 0; i < len; i++)
 124                         nested_table_free(ntbl + i, size);
 125         }
 126
 127         kfree(ntbl);
 128 }
 129
 130 static void nested_bucket_table_free(const struct bucket_table *tbl)
 131 {
 132         unsigned int size = tbl->size >> tbl->nest;
 133         unsigned int len = 1 << tbl->nest;
 134         union nested_table *ntbl;
 135         unsigned int i;
 136
 137         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
 138
 139         for (i = 0; i < len; i++)
 140                 nested_table_free(ntbl + i, size);
 141
 142         kfree(ntbl);
 143 }
 144
 145 static void bucket_table_free(const struct bucket_table *tbl)
 146 {
 147         if (tbl->nest)
 148                 nested_bucket_table_free(tbl);
 149
 150         kvfree(tbl->locks);
 151         kvfree(tbl);
 152 }
 153
 154 static void bucket_table_free_rcu(struct rcu_head *head)
 155 {
 156         bucket_table_free(container_of(head, struct bucket_table, rcu));
 157 }
 158
 159 static union nested_table *nested_table_alloc(struct rhashtable *ht,
 160                                               union nested_table __rcu **prev,
 161                                               unsigned int shifted,
 162                                               unsigned int nhash)
 163 {
 164         union nested_table *ntbl;
 165         int i;
 166
 167         ntbl = rcu_dereference(*prev);
 168         if (ntbl)
 169                 return ntbl;
 170
 171         ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC);
 172
 173         if (ntbl && shifted) {
 174                 for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0].bucket); i++)
 175                         INIT_RHT_NULLS_HEAD(ntbl[i].bucket, ht,
 176                                             (i << shifted) | nhash);
 177         }
 178
 179         rcu_assign_pointer(*prev, ntbl);
 180
 181         return ntbl;
 182 }
 183
 184 static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht,
 185                                                       size_t nbuckets,
 186                                                       gfp_t gfp)
 187 {
 188         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
 189         struct bucket_table *tbl;
 190         size_t size;
 191
 192         if (nbuckets < (1 << (shift + 1)))
 193                 return NULL;
 194
 195         size = sizeof(*tbl) + sizeof(tbl->buckets[0]);
 196
 197         tbl = kzalloc(size, gfp);
 198         if (!tbl)
 199                 return NULL;
 200
 201         if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets,
 202                                 0, 0)) {
 203                 kfree(tbl);
 204                 return NULL;
 205         }
 206
 207         tbl->nest = (ilog2(nbuckets) - 1) % shift + 1;
 208
 209         return tbl;
 210 }
 211
 212 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
 213                                                size_t nbuckets,
 214                                                gfp_t gfp)
 215 {
 216         struct bucket_table *tbl = NULL;
 217         size_t size;
 218         int i;
 219
 220         size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
 221         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) ||
 222             gfp != GFP_KERNEL)
 223                 tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
 224         if (tbl == NULL && gfp == GFP_KERNEL)
 225                 tbl = vzalloc(size);
 226
 227         size = nbuckets;
 228
 229         if (tbl == NULL && gfp != GFP_KERNEL) {
 230                 tbl = nested_bucket_table_alloc(ht, nbuckets, gfp);
 231                 nbuckets = 0;
 232         }
 233         if (tbl == NULL)
 234                 return NULL;
 235
 236         tbl->size = size;
 237
 238         if (alloc_bucket_locks(ht, tbl, gfp) < 0) {
 239                 bucket_table_free(tbl);
 240                 return NULL;
 241         }
 242
 243         INIT_LIST_HEAD(&tbl->walkers);
 244
 245         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 246
 247         for (i = 0; i < nbuckets; i++)
 248                 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
 249
 250         return tbl;
 251 }
 252
 253 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
 254                                                   struct bucket_table *tbl)
 255 {
 256         struct bucket_table *new_tbl;
 257
 258         do {
 259                 new_tbl = tbl;
 260                 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 261         } while (tbl);
 262
 263         return new_tbl;
 264 }
 265
 266 static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
 267 {
 268         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 269         struct bucket_table *new_tbl = rhashtable_last_table(ht,
 270                 rht_dereference_rcu(old_tbl->future_tbl, ht));
 271         struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash);
 272         int err = -EAGAIN;
 273         struct rhash_head *head, *next, *entry;
 274         spinlock_t *new_bucket_lock;
 275         unsigned int new_hash;
 276
 277         if (new_tbl->nest)
 278                 goto out;
 279
 280         err = -ENOENT;
 281
 282         rht_for_each(entry, old_tbl, old_hash) {
 283                 err = 0;
 284                 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
 285
 286                 if (rht_is_a_nulls(next))
 287                         break;
 288
 289                 pprev = &entry->next;
 290         }
 291
 292         if (err)
 293                 goto out;
 294
 295         new_hash = head_hashfn(ht, new_tbl, entry);
 296
 297         new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
 298
 299         spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
 300         head = rht_dereference_bucket(new_tbl->buckets[new_hash],
 301                                       new_tbl, new_hash);
 302
 303         RCU_INIT_POINTER(entry->next, head);
 304
 305         rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
 306         spin_unlock(new_bucket_lock);
 307
 308         rcu_assign_pointer(*pprev, next);
 309
 310 out:
 311         return err;
 312 }
 313
 314 static int rhashtable_rehash_chain(struct rhashtable *ht,
 315                                     unsigned int old_hash)
 316 {
 317         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 318         spinlock_t *old_bucket_lock;
 319         int err;
 320
 321         old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
 322
 323         spin_lock_bh(old_bucket_lock);
 324         while (!(err = rhashtable_rehash_one(ht, old_hash)))
 325                 ;
 326
 327         if (err == -ENOENT) {
 328                 old_tbl->rehash++;
 329                 err = 0;
 330         }
 331         spin_unlock_bh(old_bucket_lock);
 332
 333         return err;
 334 }
 335
 336 static int rhashtable_rehash_attach(struct rhashtable *ht,
 337                                     struct bucket_table *old_tbl,
 338                                     struct bucket_table *new_tbl)
 339 {
 340         /* Protect future_tbl using the first bucket lock. */
 341         spin_lock_bh(old_tbl->locks);
 342
 343         /* Did somebody beat us to it? */
 344         if (rcu_access_pointer(old_tbl->future_tbl)) {
 345                 spin_unlock_bh(old_tbl->locks);
 346                 return -EEXIST;
 347         }
 348
 349         /* Make insertions go into the new, empty table right away. Deletions
 350          * and lookups will be attempted in both tables until we synchronize.
 351          */
 352         rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
 353
 354         spin_unlock_bh(old_tbl->locks);
 355
 356         return 0;
 357 }
 358
 359 static int rhashtable_rehash_table(struct rhashtable *ht)
 360 {
 361         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 362         struct bucket_table *new_tbl;
 363         struct rhashtable_walker *walker;
 364         unsigned int old_hash;
 365         int err;
 366
 367         new_tbl = rht_dereference(old_tbl->future_tbl, ht);
 368         if (!new_tbl)
 369                 return 0;
 370
 371         for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
 372                 err = rhashtable_rehash_chain(ht, old_hash);
 373                 if (err)
 374                         return err;
 375         }
 376
 377         /* Publish the new table pointer. */
 378         rcu_assign_pointer(ht->tbl, new_tbl);
 379
 380         spin_lock(&ht->lock);
 381         list_for_each_entry(walker, &old_tbl->walkers, list)
 382                 walker->tbl = NULL;
 383         spin_unlock(&ht->lock);
 384
 385         /* Wait for readers. All new readers will see the new
 386          * table, and thus no references to the old table will
 387          * remain.
 388          */
 389         call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
 390
 391         return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
 392 }
 393
 394 static int rhashtable_rehash_alloc(struct rhashtable *ht,
 395                                    struct bucket_table *old_tbl,
 396                                    unsigned int size)
 397 {
 398         struct bucket_table *new_tbl;
 399         int err;
 400
 401         ASSERT_RHT_MUTEX(ht);
 402
 403         new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
 404         if (new_tbl == NULL)
 405                 return -ENOMEM;
 406
 407         err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
 408         if (err)
 409                 bucket_table_free(new_tbl);
 410
 411         return err;
 412 }
 413
 414 /**
 415  * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
 416  * @ht:         the hash table to shrink
 417  *
 418  * This function shrinks the hash table to fit, i.e., the smallest
 419  * size would not cause it to expand right away automatically.
 420  *
 421  * The caller must ensure that no concurrent resizing occurs by holding
 422  * ht->mutex.
 423  *
 424  * The caller must ensure that no concurrent table mutations take place.
 425  * It is however valid to have concurrent lookups if they are RCU protected.
 426  *
 427  * It is valid to have concurrent insertions and deletions protected by per
 428  * bucket locks or concurrent RCU protected lookups and traversals.
 429  */
 430 static int rhashtable_shrink(struct rhashtable *ht)
 431 {
 432         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 433         unsigned int nelems = atomic_read(&ht->nelems);
 434         unsigned int size = 0;
 435
 436         if (nelems)
 437                 size = roundup_pow_of_two(nelems * 3 / 2);
 438         if (size < ht->p.min_size)
 439                 size = ht->p.min_size;
 440
 441         if (old_tbl->size <= size)
 442                 return 0;
 443
 444         if (rht_dereference(old_tbl->future_tbl, ht))
 445                 return -EEXIST;
 446
 447         return rhashtable_rehash_alloc(ht, old_tbl, size);
 448 }
 449
 450 static void rht_deferred_worker(struct work_struct *work)
 451 {
 452         struct rhashtable *ht;
 453         struct bucket_table *tbl;
 454         int err = 0;
 455
 456         ht = container_of(work, struct rhashtable, run_work);
 457         mutex_lock(&ht->mutex);
 458
 459         tbl = rht_dereference(ht->tbl, ht);
 460         tbl = rhashtable_last_table(ht, tbl);
 461
 462         if (rht_grow_above_75(ht, tbl))
 463                 err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2);
 464         else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
 465                 err = rhashtable_shrink(ht);
 466         else if (tbl->nest)
 467                 err = rhashtable_rehash_alloc(ht, tbl, tbl->size);
 468
 469         if (!err)
 470                 err = rhashtable_rehash_table(ht);
 471
 472         mutex_unlock(&ht->mutex);
 473
 474         if (err)
 475                 schedule_work(&ht->run_work);
 476 }
 477
 478 static int rhashtable_insert_rehash(struct rhashtable *ht,
 479                                     struct bucket_table *tbl)
 480 {
 481         struct bucket_table *old_tbl;
 482         struct bucket_table *new_tbl;
 483         unsigned int size;
 484         int err;
 485
 486         old_tbl = rht_dereference_rcu(ht->tbl, ht);
 487
 488         size = tbl->size;
 489
 490         err = -EBUSY;
 491
 492         if (rht_grow_above_75(ht, tbl))
 493                 size *= 2;
 494         /* Do not schedule more than one rehash */
 495         else if (old_tbl != tbl)
 496                 goto fail;
 497
 498         err = -ENOMEM;
 499
 500         new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
 501         if (new_tbl == NULL)
 502                 goto fail;
 503
 504         err = rhashtable_rehash_attach(ht, tbl, new_tbl);
 505         if (err) {
 506                 bucket_table_free(new_tbl);
 507                 if (err == -EEXIST)
 508                         err = 0;
 509         } else
 510                 schedule_work(&ht->run_work);
 511
 512         return err;
 513
 514 fail:
 515         /* Do not fail the insert if someone else did a rehash. */
 516         if (likely(rcu_dereference_raw(tbl->future_tbl)))
 517                 return 0;
 518
 519         /* Schedule async rehash to retry allocation in process context. */
 520         if (err == -ENOMEM)
 521                 schedule_work(&ht->run_work);
 522
 523         return err;
 524 }
 525
 526 static void *rhashtable_lookup_one(struct rhashtable *ht,
 527                                    struct bucket_table *tbl, unsigned int hash,
 528                                    const void *key, struct rhash_head *obj)
 529 {
 530         struct rhashtable_compare_arg arg = {
 531                 .ht = ht,
 532                 .key = key,
 533         };
 534         struct rhash_head __rcu **pprev;
 535         struct rhash_head *head;
 536         int elasticity;
 537
 538         elasticity = ht->elasticity;
 539         pprev = rht_bucket_var(tbl, hash);
 540         rht_for_each_continue(head, *pprev, tbl, hash) {
 541                 struct rhlist_head *list;
 542                 struct rhlist_head *plist;
 543
 544                 elasticity--;
 545                 if (!key ||
 546                     (ht->p.obj_cmpfn ?
 547                      ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
 548                      rhashtable_compare(&arg, rht_obj(ht, head))))
 549                         continue;
 550
 551                 if (!ht->rhlist)
 552                         return rht_obj(ht, head);
 553
 554                 list = container_of(obj, struct rhlist_head, rhead);
 555                 plist = container_of(head, struct rhlist_head, rhead);
 556
 557                 RCU_INIT_POINTER(list->next, plist);
 558                 head = rht_dereference_bucket(head->next, tbl, hash);
 559                 RCU_INIT_POINTER(list->rhead.next, head);
 560                 rcu_assign_pointer(*pprev, obj);
 561
 562                 return NULL;
 563         }
 564
 565         if (elasticity <= 0)
 566                 return ERR_PTR(-EAGAIN);
 567
 568         return ERR_PTR(-ENOENT);
 569 }
 570
 571 static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
 572                                                   struct bucket_table *tbl,
 573                                                   unsigned int hash,
 574                                                   struct rhash_head *obj,
 575                                                   void *data)
 576 {
 577         struct rhash_head __rcu **pprev;
 578         struct bucket_table *new_tbl;
 579         struct rhash_head *head;
 580
 581         if (!IS_ERR_OR_NULL(data))
 582                 return ERR_PTR(-EEXIST);
 583
 584         if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
 585                 return ERR_CAST(data);
 586
 587         new_tbl = rcu_dereference(tbl->future_tbl);
 588         if (new_tbl)
 589                 return new_tbl;
 590
 591         if (PTR_ERR(data) != -ENOENT)
 592                 return ERR_CAST(data);
 593
 594         if (unlikely(rht_grow_above_max(ht, tbl)))
 595                 return ERR_PTR(-E2BIG);
 596
 597         if (unlikely(rht_grow_above_100(ht, tbl)))
 598                 return ERR_PTR(-EAGAIN);
 599
 600         pprev = rht_bucket_insert(ht, tbl, hash);
 601         if (!pprev)
 602                 return ERR_PTR(-ENOMEM);
 603
 604         head = rht_dereference_bucket(*pprev, tbl, hash);
 605
 606         RCU_INIT_POINTER(obj->next, head);
 607         if (ht->rhlist) {
 608                 struct rhlist_head *list;
 609
 610                 list = container_of(obj, struct rhlist_head, rhead);
 611                 RCU_INIT_POINTER(list->next, NULL);
 612         }
 613
 614         rcu_assign_pointer(*pprev, obj);
 615
 616         atomic_inc(&ht->nelems);
 617         if (rht_grow_above_75(ht, tbl))
 618                 schedule_work(&ht->run_work);
 619
 620         return NULL;
 621 }
 622
 623 static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
 624                                    struct rhash_head *obj)
 625 {
 626         struct bucket_table *new_tbl;
 627         struct bucket_table *tbl;
 628         unsigned int hash;
 629         spinlock_t *lock;
 630         void *data;
 631
 632         tbl = rcu_dereference(ht->tbl);
 633
 634         /* All insertions must grab the oldest table containing
 635          * the hashed bucket that is yet to be rehashed.
 636          */
 637         for (;;) {
 638                 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
 639                 lock = rht_bucket_lock(tbl, hash);
 640                 spin_lock_bh(lock);
 641
 642                 if (tbl->rehash <= hash)
 643                         break;
 644
 645                 spin_unlock_bh(lock);
 646                 tbl = rcu_dereference(tbl->future_tbl);
 647         }
 648
 649         data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
 650         new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
 651         if (PTR_ERR(new_tbl) != -EEXIST)
 652                 data = ERR_CAST(new_tbl);
 653
 654         while (!IS_ERR_OR_NULL(new_tbl)) {
 655                 tbl = new_tbl;
 656                 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
 657                 spin_lock_nested(rht_bucket_lock(tbl, hash),
 658                                  SINGLE_DEPTH_NESTING);
 659
 660                 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
 661                 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
 662                 if (PTR_ERR(new_tbl) != -EEXIST)
 663                         data = ERR_CAST(new_tbl);
 664
 665                 spin_unlock(rht_bucket_lock(tbl, hash));
 666         }
 667
 668         spin_unlock_bh(lock);
 669
 670         if (PTR_ERR(data) == -EAGAIN)
 671                 data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
 672                                -EAGAIN);
 673
 674         return data;
 675 }
 676
 677 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
 678                              struct rhash_head *obj)
 679 {
 680         void *data;
 681
 682         do {
 683                 rcu_read_lock();
 684                 data = rhashtable_try_insert(ht, key, obj);
 685                 rcu_read_unlock();
 686         } while (PTR_ERR(data) == -EAGAIN);
 687
 688         return data;
 689 }
 690 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
 691
 692 /**
 693  * rhashtable_walk_enter - Initialise an iterator
 694  * @ht:         Table to walk over
 695  * @iter:       Hash table Iterator
 696  *
 697  * This function prepares a hash table walk.
 698  *
 699  * Note that if you restart a walk after rhashtable_walk_stop you
 700  * may see the same object twice.  Also, you may miss objects if
 701  * there are removals in between rhashtable_walk_stop and the next
 702  * call to rhashtable_walk_start.
 703  *
 704  * For a completely stable walk you should construct your own data
 705  * structure outside the hash table.
 706  *
 707  * This function may sleep so you must not call it from interrupt
 708  * context or with spin locks held.
 709  *
 710  * You must call rhashtable_walk_exit after this function returns.
 711  */
 712 void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
 713 {
 714         iter->ht = ht;
 715         iter->p = NULL;
 716         iter->slot = 0;
 717         iter->skip = 0;
 718
 719         spin_lock(&ht->lock);
 720         iter->walker.tbl =
 721                 rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
 722         list_add(&iter->walker.list, &iter->walker.tbl->walkers);
 723         spin_unlock(&ht->lock);
 724 }
 725 EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
 726
 727 /**
 728  * rhashtable_walk_exit - Free an iterator
 729  * @iter:       Hash table Iterator
 730  *
 731  * This function frees resources allocated by rhashtable_walk_init.
 732  */
 733 void rhashtable_walk_exit(struct rhashtable_iter *iter)
 734 {
 735         spin_lock(&iter->ht->lock);
 736         if (iter->walker.tbl)
 737                 list_del(&iter->walker.list);
 738         spin_unlock(&iter->ht->lock);
 739 }
 740 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
 741
 742 /**
 743  * rhashtable_walk_start - Start a hash table walk
 744  * @iter:       Hash table iterator
 745  *
 746  * Start a hash table walk.  Note that we take the RCU lock in all
 747  * cases including when we return an error.  So you must always call
 748  * rhashtable_walk_stop to clean up.
 749  *
 750  * Returns zero if successful.
 751  *
 752  * Returns -EAGAIN if resize event occured.  Note that the iterator
 753  * will rewind back to the beginning and you may use it immediately
 754  * by calling rhashtable_walk_next.
 755  */
 756 int rhashtable_walk_start(struct rhashtable_iter *iter)
 757         __acquires(RCU)
 758 {
 759         struct rhashtable *ht = iter->ht;
 760
 761         rcu_read_lock();
 762
 763         spin_lock(&ht->lock);
 764         if (iter->walker.tbl)
 765                 list_del(&iter->walker.list);
 766         spin_unlock(&ht->lock);
 767
 768         if (!iter->walker.tbl) {
 769                 iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
 770                 return -EAGAIN;
 771         }
 772
 773         return 0;
 774 }
 775 EXPORT_SYMBOL_GPL(rhashtable_walk_start);
 776
 777 /**
 778  * rhashtable_walk_next - Return the next object and advance the iterator
 779  * @iter:       Hash table iterator
 780  *
 781  * Note that you must call rhashtable_walk_stop when you are finished
 782  * with the walk.
 783  *
 784  * Returns the next object or NULL when the end of the table is reached.
 785  *
 786  * Returns -EAGAIN if resize event occured.  Note that the iterator
 787  * will rewind back to the beginning and you may continue to use it.
 788  */
 789 void *rhashtable_walk_next(struct rhashtable_iter *iter)
 790 {
 791         struct bucket_table *tbl = iter->walker.tbl;
 792         struct rhlist_head *list = iter->list;
 793         struct rhashtable *ht = iter->ht;
 794         struct rhash_head *p = iter->p;
 795         bool rhlist = ht->rhlist;
 796
 797         if (p) {
 798                 if (!rhlist || !(list = rcu_dereference(list->next))) {
 799                         p = rcu_dereference(p->next);
 800                         list = container_of(p, struct rhlist_head, rhead);
 801                 }
 802                 goto next;
 803         }
 804
 805         for (; iter->slot < tbl->size; iter->slot++) {
 806                 int skip = iter->skip;
 807
 808                 rht_for_each_rcu(p, tbl, iter->slot) {
 809                         if (rhlist) {
 810                                 list = container_of(p, struct rhlist_head,
 811                                                     rhead);
 812                                 do {
 813                                         if (!skip)
 814                                                 goto next;
 815                                         skip--;
 816                                         list = rcu_dereference(list->next);
 817                                 } while (list);
 818
 819                                 continue;
 820                         }
 821                         if (!skip)
 822                                 break;
 823                         skip--;
 824                 }
 825
 826 next:
 827                 if (!rht_is_a_nulls(p)) {
 828                         iter->skip++;
 829                         iter->p = p;
 830                         iter->list = list;
 831                         return rht_obj(ht, rhlist ? &list->rhead : p);
 832                 }
 833
 834                 iter->skip = 0;
 835         }
 836
 837         iter->p = NULL;
 838
 839         /* Ensure we see any new tables. */
 840         smp_rmb();
 841
 842         iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 843         if (iter->walker.tbl) {
 844                 iter->slot = 0;
 845                 iter->skip = 0;
 846                 return ERR_PTR(-EAGAIN);
 847         }
 848
 849         return NULL;
 850 }
 851 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
 852
 853 /**
 854  * rhashtable_walk_stop - Finish a hash table walk
 855  * @iter:       Hash table iterator
 856  *
 857  * Finish a hash table walk.
 858  */
 859 void rhashtable_walk_stop(struct rhashtable_iter *iter)
 860         __releases(RCU)
 861 {
 862         struct rhashtable *ht;
 863         struct bucket_table *tbl = iter->walker.tbl;
 864
 865         if (!tbl)
 866                 goto out;
 867
 868         ht = iter->ht;
 869
 870         spin_lock(&ht->lock);
 871         if (tbl->rehash < tbl->size)
 872                 list_add(&iter->walker.list, &tbl->walkers);
 873         else
 874                 iter->walker.tbl = NULL;
 875         spin_unlock(&ht->lock);
 876
 877         iter->p = NULL;
 878
 879 out:
 880         rcu_read_unlock();
 881 }
 882 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
 883
 884 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
 885 {
 886         return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
 887                    (unsigned long)params->min_size);
 888 }
 889
 890 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
 891 {
 892         return jhash2(key, length, seed);
 893 }
 894
 895 /**
 896  * rhashtable_init - initialize a new hash table
 897  * @ht:         hash table to be initialized
 898  * @params:     configuration parameters
 899  *
 900  * Initializes a new hash table based on the provided configuration
 901  * parameters. A table can be configured either with a variable or
 902  * fixed length key:
 903  *
 904  * Configuration Example 1: Fixed length keys
 905  * struct test_obj {
 906  *      int                     key;
 907  *      void *                  my_member;
 908  *      struct rhash_head       node;
 909  * };
 910  *
 911  * struct rhashtable_params params = {
 912  *      .head_offset = offsetof(struct test_obj, node),
 913  *      .key_offset = offsetof(struct test_obj, key),
 914  *      .key_len = sizeof(int),
 915  *      .hashfn = jhash,
 916  *      .nulls_base = (1U << RHT_BASE_SHIFT),
 917  * };
 918  *
 919  * Configuration Example 2: Variable length keys
 920  * struct test_obj {
 921  *      [...]
 922  *      struct rhash_head       node;
 923  * };
 924  *
 925  * u32 my_hash_fn(const void *data, u32 len, u32 seed)
 926  * {
 927  *      struct test_obj *obj = data;
 928  *
 929  *      return [... hash ...];
 930  * }
 931  *
 932  * struct rhashtable_params params = {
 933  *      .head_offset = offsetof(struct test_obj, node),
 934  *      .hashfn = jhash,
 935  *      .obj_hashfn = my_hash_fn,
 936  * };
 937  */
 938 int rhashtable_init(struct rhashtable *ht,
 939                     const struct rhashtable_params *params)
 940 {
 941         struct bucket_table *tbl;
 942         size_t size;
 943
 944         size = HASH_DEFAULT_SIZE;
 945
 946         if ((!params->key_len && !params->obj_hashfn) ||
 947             (params->obj_hashfn && !params->obj_cmpfn))
 948                 return -EINVAL;
 949
 950         if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
 951                 return -EINVAL;
 952
 953         memset(ht, 0, sizeof(*ht));
 954         mutex_init(&ht->mutex);
 955         spin_lock_init(&ht->lock);
 956         memcpy(&ht->p, params, sizeof(*params));
 957
 958         if (params->min_size)
 959                 ht->p.min_size = roundup_pow_of_two(params->min_size);
 960
 961         if (params->max_size)
 962                 ht->p.max_size = rounddown_pow_of_two(params->max_size);
 963
 964         if (params->insecure_max_entries)
 965                 ht->p.insecure_max_entries =
 966                         rounddown_pow_of_two(params->insecure_max_entries);
 967         else
 968                 ht->p.insecure_max_entries = ht->p.max_size * 2;
 969
 970         ht->p.min_size = max(ht->p.min_size, HASH_MIN_SIZE);
 971
 972         if (params->nelem_hint)
 973                 size = rounded_hashtable_size(&ht->p);
 974
 975         /* The maximum (not average) chain length grows with the
 976          * size of the hash table, at a rate of (log N)/(log log N).
 977          * The value of 16 is selected so that even if the hash
 978          * table grew to 2^32 you would not expect the maximum
 979          * chain length to exceed it unless we are under attack
 980          * (or extremely unlucky).
 981          *
 982          * As this limit is only to detect attacks, we don't need
 983          * to set it to a lower value as you'd need the chain
 984          * length to vastly exceed 16 to have any real effect
 985          * on the system.
 986          */
 987         if (!params->insecure_elasticity)
 988                 ht->elasticity = 16;
 989
 990         if (params->locks_mul)
 991                 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
 992         else
 993                 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
 994
 995         ht->key_len = ht->p.key_len;
 996         if (!params->hashfn) {
 997                 ht->p.hashfn = jhash;
 998
 999                 if (!(ht->key_len & (sizeof(u32) - 1))) {
1000                         ht->key_len /= sizeof(u32);
1001                         ht->p.hashfn = rhashtable_jhash2;
1002                 }
1003         }
1004
1005         tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
1006         if (tbl == NULL)
1007                 return -ENOMEM;
1008
1009         atomic_set(&ht->nelems, 0);
1010
1011         RCU_INIT_POINTER(ht->tbl, tbl);
1012
1013         INIT_WORK(&ht->run_work, rht_deferred_worker);
1014
1015         return 0;
1016 }
1017 EXPORT_SYMBOL_GPL(rhashtable_init);
1018
1019 /**
1020  * rhltable_init - initialize a new hash list table
1021  * @hlt:        hash list table to be initialized
1022  * @params:     configuration parameters
1023  *
1024  * Initializes a new hash list table.
1025  *
1026  * See documentation for rhashtable_init.
1027  */
1028 int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
1029 {
1030         int err;
1031
1032         /* No rhlist NULLs marking for now. */
1033         if (params->nulls_base)
1034                 return -EINVAL;
1035
1036         err = rhashtable_init(&hlt->ht, params);
1037         hlt->ht.rhlist = true;
1038         return err;
1039 }
1040 EXPORT_SYMBOL_GPL(rhltable_init);
1041
1042 static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
1043                                 void (*free_fn)(void *ptr, void *arg),
1044                                 void *arg)
1045 {
1046         struct rhlist_head *list;
1047
1048         if (!ht->rhlist) {
1049                 free_fn(rht_obj(ht, obj), arg);
1050                 return;
1051         }
1052
1053         list = container_of(obj, struct rhlist_head, rhead);
1054         do {
1055                 obj = &list->rhead;
1056                 list = rht_dereference(list->next, ht);
1057                 free_fn(rht_obj(ht, obj), arg);
1058         } while (list);
1059 }
1060
1061 /**
1062  * rhashtable_free_and_destroy - free elements and destroy hash table
1063  * @ht:         the hash table to destroy
1064  * @free_fn:    callback to release resources of element
1065  * @arg:        pointer passed to free_fn
1066  *
1067  * Stops an eventual async resize. If defined, invokes free_fn for each
1068  * element to releasal resources. Please note that RCU protected
1069  * readers may still be accessing the elements. Releasing of resources
1070  * must occur in a compatible manner. Then frees the bucket array.
1071  *
1072  * This function will eventually sleep to wait for an async resize
1073  * to complete. The caller is responsible that no further write operations
1074  * occurs in parallel.
1075  */
1076 void rhashtable_free_and_destroy(struct rhashtable *ht,
1077                                  void (*free_fn)(void *ptr, void *arg),
1078                                  void *arg)
1079 {
1080         struct bucket_table *tbl;
1081         unsigned int i;
1082
1083         cancel_work_sync(&ht->run_work);
1084
1085         mutex_lock(&ht->mutex);
1086         tbl = rht_dereference(ht->tbl, ht);
1087         if (free_fn) {
1088                 for (i = 0; i < tbl->size; i++) {
1089                         struct rhash_head *pos, *next;
1090
1091                         for (pos = rht_dereference(*rht_bucket(tbl, i), ht),
1092                              next = !rht_is_a_nulls(pos) ?
1093                                         rht_dereference(pos->next, ht) : NULL;
1094                              !rht_is_a_nulls(pos);
1095                              pos = next,
1096                              next = !rht_is_a_nulls(pos) ?
1097                                         rht_dereference(pos->next, ht) : NULL)
1098                                 rhashtable_free_one(ht, pos, free_fn, arg);
1099                 }
1100         }
1101
1102         bucket_table_free(tbl);
1103         mutex_unlock(&ht->mutex);
1104 }
1105 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
1106
1107 void rhashtable_destroy(struct rhashtable *ht)
1108 {
1109         return rhashtable_free_and_destroy(ht, NULL, NULL);
1110 }
1111 EXPORT_SYMBOL_GPL(rhashtable_destroy);
1112
1113 struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
1114                                             unsigned int hash)
1115 {
1116         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1117         static struct rhash_head __rcu *rhnull =
1118                 (struct rhash_head __rcu *)NULLS_MARKER(0);
1119         unsigned int index = hash & ((1 << tbl->nest) - 1);
1120         unsigned int size = tbl->size >> tbl->nest;
1121         unsigned int subhash = hash;
1122         union nested_table *ntbl;
1123
1124         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1125         ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash);
1126         subhash >>= tbl->nest;
1127
1128         while (ntbl && size > (1 << shift)) {
1129                 index = subhash & ((1 << shift) - 1);
1130                 ntbl = rht_dereference_bucket_rcu(ntbl[index].table,
1131                                                   tbl, hash);
1132                 size >>= shift;
1133                 subhash >>= shift;
1134         }
1135
1136         if (!ntbl)
1137                 return &rhnull;
1138
1139         return &ntbl[subhash].bucket;
1140
1141 }
1142 EXPORT_SYMBOL_GPL(rht_bucket_nested);
1143
1144 struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
1145                                                    struct bucket_table *tbl,
1146                                                    unsigned int hash)
1147 {
1148         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1149         unsigned int index = hash & ((1 << tbl->nest) - 1);
1150         unsigned int size = tbl->size >> tbl->nest;
1151         union nested_table *ntbl;
1152         unsigned int shifted;
1153         unsigned int nhash;
1154
1155         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1156         hash >>= tbl->nest;
1157         nhash = index;
1158         shifted = tbl->nest;
1159         ntbl = nested_table_alloc(ht, &ntbl[index].table,
1160                                   size <= (1 << shift) ? shifted : 0, nhash);
1161
1162         while (ntbl && size > (1 << shift)) {
1163                 index = hash & ((1 << shift) - 1);
1164                 size >>= shift;
1165                 hash >>= shift;
1166                 nhash |= index << shifted;
1167                 shifted += shift;
1168                 ntbl = nested_table_alloc(ht, &ntbl[index].table,
1169                                           size <= (1 << shift) ? shifted : 0,
1170                                           nhash);
1171         }
1172
1173         if (!ntbl)
1174                 return NULL;
1175
1176         return &ntbl[hash].bucket;
1177
1178 }
1179 EXPORT_SYMBOL_GPL(rht_bucket_nested_insert);