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/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 static void nested_table_free(union nested_table *ntbl, unsigned int size)
  69 {
  70         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
  71         const unsigned int len = 1 << shift;
  72         unsigned int i;
  73
  74         ntbl = rcu_dereference_raw(ntbl->table);
  75         if (!ntbl)
  76                 return;
  77
  78         if (size > len) {
  79                 size >>= shift;
  80                 for (i = 0; i < len; i++)
  81                         nested_table_free(ntbl + i, size);
  82         }
  83
  84         kfree(ntbl);
  85 }
  86
  87 static void nested_bucket_table_free(const struct bucket_table *tbl)
  88 {
  89         unsigned int size = tbl->size >> tbl->nest;
  90         unsigned int len = 1 << tbl->nest;
  91         union nested_table *ntbl;
  92         unsigned int i;
  93
  94         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
  95
  96         for (i = 0; i < len; i++)
  97                 nested_table_free(ntbl + i, size);
  98
  99         kfree(ntbl);
 100 }
 101
 102 static void bucket_table_free(const struct bucket_table *tbl)
 103 {
 104         if (tbl->nest)
 105                 nested_bucket_table_free(tbl);
 106
 107         free_bucket_spinlocks(tbl->locks);
 108         kvfree(tbl);
 109 }
 110
 111 static void bucket_table_free_rcu(struct rcu_head *head)
 112 {
 113         bucket_table_free(container_of(head, struct bucket_table, rcu));
 114 }
 115
 116 static union nested_table *nested_table_alloc(struct rhashtable *ht,
 117                                               union nested_table __rcu **prev,
 118                                               unsigned int shifted,
 119                                               unsigned int nhash)
 120 {
 121         union nested_table *ntbl;
 122         int i;
 123
 124         ntbl = rcu_dereference(*prev);
 125         if (ntbl)
 126                 return ntbl;
 127
 128         ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC);
 129
 130         if (ntbl && shifted) {
 131                 for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0].bucket); i++)
 132                         INIT_RHT_NULLS_HEAD(ntbl[i].bucket, ht,
 133                                             (i << shifted) | nhash);
 134         }
 135
 136         rcu_assign_pointer(*prev, ntbl);
 137
 138         return ntbl;
 139 }
 140
 141 static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht,
 142                                                       size_t nbuckets,
 143                                                       gfp_t gfp)
 144 {
 145         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
 146         struct bucket_table *tbl;
 147         size_t size;
 148
 149         if (nbuckets < (1 << (shift + 1)))
 150                 return NULL;
 151
 152         size = sizeof(*tbl) + sizeof(tbl->buckets[0]);
 153
 154         tbl = kzalloc(size, gfp);
 155         if (!tbl)
 156                 return NULL;
 157
 158         if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets,
 159                                 0, 0)) {
 160                 kfree(tbl);
 161                 return NULL;
 162         }
 163
 164         tbl->nest = (ilog2(nbuckets) - 1) % shift + 1;
 165
 166         return tbl;
 167 }
 168
 169 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
 170                                                size_t nbuckets,
 171                                                gfp_t gfp)
 172 {
 173         struct bucket_table *tbl = NULL;
 174         size_t size, max_locks;
 175         int i;
 176
 177         size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
 178         if (gfp != GFP_KERNEL)
 179                 tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY);
 180         else
 181                 tbl = kvzalloc(size, gfp);
 182
 183         size = nbuckets;
 184
 185         if (tbl == NULL && gfp != GFP_KERNEL) {
 186                 tbl = nested_bucket_table_alloc(ht, nbuckets, gfp);
 187                 nbuckets = 0;
 188         }
 189         if (tbl == NULL)
 190                 return NULL;
 191
 192         tbl->size = size;
 193
 194         max_locks = size >> 1;
 195         if (tbl->nest)
 196                 max_locks = min_t(size_t, max_locks, 1U << tbl->nest);
 197
 198         if (alloc_bucket_spinlocks(&tbl->locks, &tbl->locks_mask, max_locks,
 199                                    ht->p.locks_mul, gfp) < 0) {
 200                 bucket_table_free(tbl);
 201                 return NULL;
 202         }
 203
 204         INIT_LIST_HEAD(&tbl->walkers);
 205
 206         tbl->hash_rnd = get_random_u32();
 207
 208         for (i = 0; i < nbuckets; i++)
 209                 INIT_RHT_NULLS_HEAD(tbl->buckets[i], ht, i);
 210
 211         return tbl;
 212 }
 213
 214 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
 215                                                   struct bucket_table *tbl)
 216 {
 217         struct bucket_table *new_tbl;
 218
 219         do {
 220                 new_tbl = tbl;
 221                 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 222         } while (tbl);
 223
 224         return new_tbl;
 225 }
 226
 227 static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
 228 {
 229         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 230         struct bucket_table *new_tbl = rhashtable_last_table(ht,
 231                 rht_dereference_rcu(old_tbl->future_tbl, ht));
 232         struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash);
 233         int err = -EAGAIN;
 234         struct rhash_head *head, *next, *entry;
 235         spinlock_t *new_bucket_lock;
 236         unsigned int new_hash;
 237
 238         if (new_tbl->nest)
 239                 goto out;
 240
 241         err = -ENOENT;
 242
 243         rht_for_each(entry, old_tbl, old_hash) {
 244                 err = 0;
 245                 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
 246
 247                 if (rht_is_a_nulls(next))
 248                         break;
 249
 250                 pprev = &entry->next;
 251         }
 252
 253         if (err)
 254                 goto out;
 255
 256         new_hash = head_hashfn(ht, new_tbl, entry);
 257
 258         new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
 259
 260         spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
 261         head = rht_dereference_bucket(new_tbl->buckets[new_hash],
 262                                       new_tbl, new_hash);
 263
 264         RCU_INIT_POINTER(entry->next, head);
 265
 266         rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
 267         spin_unlock(new_bucket_lock);
 268
 269         rcu_assign_pointer(*pprev, next);
 270
 271 out:
 272         return err;
 273 }
 274
 275 static int rhashtable_rehash_chain(struct rhashtable *ht,
 276                                     unsigned int old_hash)
 277 {
 278         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 279         spinlock_t *old_bucket_lock;
 280         int err;
 281
 282         old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
 283
 284         spin_lock_bh(old_bucket_lock);
 285         while (!(err = rhashtable_rehash_one(ht, old_hash)))
 286                 ;
 287
 288         if (err == -ENOENT) {
 289                 old_tbl->rehash++;
 290                 err = 0;
 291         }
 292         spin_unlock_bh(old_bucket_lock);
 293
 294         return err;
 295 }
 296
 297 static int rhashtable_rehash_attach(struct rhashtable *ht,
 298                                     struct bucket_table *old_tbl,
 299                                     struct bucket_table *new_tbl)
 300 {
 301         /* Protect future_tbl using the first bucket lock. */
 302         spin_lock_bh(old_tbl->locks);
 303
 304         /* Did somebody beat us to it? */
 305         if (rcu_access_pointer(old_tbl->future_tbl)) {
 306                 spin_unlock_bh(old_tbl->locks);
 307                 return -EEXIST;
 308         }
 309
 310         /* Make insertions go into the new, empty table right away. Deletions
 311          * and lookups will be attempted in both tables until we synchronize.
 312          */
 313         rcu_assign_pointer(old_tbl->future_tbl, new_tbl);
 314
 315         spin_unlock_bh(old_tbl->locks);
 316
 317         return 0;
 318 }
 319
 320 static int rhashtable_rehash_table(struct rhashtable *ht)
 321 {
 322         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 323         struct bucket_table *new_tbl;
 324         struct rhashtable_walker *walker;
 325         unsigned int old_hash;
 326         int err;
 327
 328         new_tbl = rht_dereference(old_tbl->future_tbl, ht);
 329         if (!new_tbl)
 330                 return 0;
 331
 332         for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
 333                 err = rhashtable_rehash_chain(ht, old_hash);
 334                 if (err)
 335                         return err;
 336                 cond_resched();
 337         }
 338
 339         /* Publish the new table pointer. */
 340         rcu_assign_pointer(ht->tbl, new_tbl);
 341
 342         spin_lock(&ht->lock);
 343         list_for_each_entry(walker, &old_tbl->walkers, list)
 344                 walker->tbl = NULL;
 345         spin_unlock(&ht->lock);
 346
 347         /* Wait for readers. All new readers will see the new
 348          * table, and thus no references to the old table will
 349          * remain.
 350          */
 351         call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
 352
 353         return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
 354 }
 355
 356 static int rhashtable_rehash_alloc(struct rhashtable *ht,
 357                                    struct bucket_table *old_tbl,
 358                                    unsigned int size)
 359 {
 360         struct bucket_table *new_tbl;
 361         int err;
 362
 363         ASSERT_RHT_MUTEX(ht);
 364
 365         new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
 366         if (new_tbl == NULL)
 367                 return -ENOMEM;
 368
 369         err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
 370         if (err)
 371                 bucket_table_free(new_tbl);
 372
 373         return err;
 374 }
 375
 376 /**
 377  * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
 378  * @ht:         the hash table to shrink
 379  *
 380  * This function shrinks the hash table to fit, i.e., the smallest
 381  * size would not cause it to expand right away automatically.
 382  *
 383  * The caller must ensure that no concurrent resizing occurs by holding
 384  * ht->mutex.
 385  *
 386  * The caller must ensure that no concurrent table mutations take place.
 387  * It is however valid to have concurrent lookups if they are RCU protected.
 388  *
 389  * It is valid to have concurrent insertions and deletions protected by per
 390  * bucket locks or concurrent RCU protected lookups and traversals.
 391  */
 392 static int rhashtable_shrink(struct rhashtable *ht)
 393 {
 394         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 395         unsigned int nelems = atomic_read(&ht->nelems);
 396         unsigned int size = 0;
 397
 398         if (nelems)
 399                 size = roundup_pow_of_two(nelems * 3 / 2);
 400         if (size < ht->p.min_size)
 401                 size = ht->p.min_size;
 402
 403         if (old_tbl->size <= size)
 404                 return 0;
 405
 406         if (rht_dereference(old_tbl->future_tbl, ht))
 407                 return -EEXIST;
 408
 409         return rhashtable_rehash_alloc(ht, old_tbl, size);
 410 }
 411
 412 static void rht_deferred_worker(struct work_struct *work)
 413 {
 414         struct rhashtable *ht;
 415         struct bucket_table *tbl;
 416         int err = 0;
 417
 418         ht = container_of(work, struct rhashtable, run_work);
 419         mutex_lock(&ht->mutex);
 420
 421         tbl = rht_dereference(ht->tbl, ht);
 422         tbl = rhashtable_last_table(ht, tbl);
 423
 424         if (rht_grow_above_75(ht, tbl))
 425                 err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2);
 426         else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
 427                 err = rhashtable_shrink(ht);
 428         else if (tbl->nest)
 429                 err = rhashtable_rehash_alloc(ht, tbl, tbl->size);
 430
 431         if (!err)
 432                 err = rhashtable_rehash_table(ht);
 433
 434         mutex_unlock(&ht->mutex);
 435
 436         if (err)
 437                 schedule_work(&ht->run_work);
 438 }
 439
 440 static int rhashtable_insert_rehash(struct rhashtable *ht,
 441                                     struct bucket_table *tbl)
 442 {
 443         struct bucket_table *old_tbl;
 444         struct bucket_table *new_tbl;
 445         unsigned int size;
 446         int err;
 447
 448         old_tbl = rht_dereference_rcu(ht->tbl, ht);
 449
 450         size = tbl->size;
 451
 452         err = -EBUSY;
 453
 454         if (rht_grow_above_75(ht, tbl))
 455                 size *= 2;
 456         /* Do not schedule more than one rehash */
 457         else if (old_tbl != tbl)
 458                 goto fail;
 459
 460         err = -ENOMEM;
 461
 462         new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);
 463         if (new_tbl == NULL)
 464                 goto fail;
 465
 466         err = rhashtable_rehash_attach(ht, tbl, new_tbl);
 467         if (err) {
 468                 bucket_table_free(new_tbl);
 469                 if (err == -EEXIST)
 470                         err = 0;
 471         } else
 472                 schedule_work(&ht->run_work);
 473
 474         return err;
 475
 476 fail:
 477         /* Do not fail the insert if someone else did a rehash. */
 478         if (likely(rcu_dereference_raw(tbl->future_tbl)))
 479                 return 0;
 480
 481         /* Schedule async rehash to retry allocation in process context. */
 482         if (err == -ENOMEM)
 483                 schedule_work(&ht->run_work);
 484
 485         return err;
 486 }
 487
 488 static void *rhashtable_lookup_one(struct rhashtable *ht,
 489                                    struct bucket_table *tbl, unsigned int hash,
 490                                    const void *key, struct rhash_head *obj)
 491 {
 492         struct rhashtable_compare_arg arg = {
 493                 .ht = ht,
 494                 .key = key,
 495         };
 496         struct rhash_head __rcu **pprev;
 497         struct rhash_head *head;
 498         int elasticity;
 499
 500         elasticity = RHT_ELASTICITY;
 501         pprev = rht_bucket_var(tbl, hash);
 502         rht_for_each_continue(head, *pprev, tbl, hash) {
 503                 struct rhlist_head *list;
 504                 struct rhlist_head *plist;
 505
 506                 elasticity--;
 507                 if (!key ||
 508                     (ht->p.obj_cmpfn ?
 509                      ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
 510                      rhashtable_compare(&arg, rht_obj(ht, head)))) {
 511                         pprev = &head->next;
 512                         continue;
 513                 }
 514
 515                 if (!ht->rhlist)
 516                         return rht_obj(ht, head);
 517
 518                 list = container_of(obj, struct rhlist_head, rhead);
 519                 plist = container_of(head, struct rhlist_head, rhead);
 520
 521                 RCU_INIT_POINTER(list->next, plist);
 522                 head = rht_dereference_bucket(head->next, tbl, hash);
 523                 RCU_INIT_POINTER(list->rhead.next, head);
 524                 rcu_assign_pointer(*pprev, obj);
 525
 526                 return NULL;
 527         }
 528
 529         if (elasticity <= 0)
 530                 return ERR_PTR(-EAGAIN);
 531
 532         return ERR_PTR(-ENOENT);
 533 }
 534
 535 static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
 536                                                   struct bucket_table *tbl,
 537                                                   unsigned int hash,
 538                                                   struct rhash_head *obj,
 539                                                   void *data)
 540 {
 541         struct rhash_head __rcu **pprev;
 542         struct bucket_table *new_tbl;
 543         struct rhash_head *head;
 544
 545         if (!IS_ERR_OR_NULL(data))
 546                 return ERR_PTR(-EEXIST);
 547
 548         if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
 549                 return ERR_CAST(data);
 550
 551         new_tbl = rcu_dereference(tbl->future_tbl);
 552         if (new_tbl)
 553                 return new_tbl;
 554
 555         if (PTR_ERR(data) != -ENOENT)
 556                 return ERR_CAST(data);
 557
 558         if (unlikely(rht_grow_above_max(ht, tbl)))
 559                 return ERR_PTR(-E2BIG);
 560
 561         if (unlikely(rht_grow_above_100(ht, tbl)))
 562                 return ERR_PTR(-EAGAIN);
 563
 564         pprev = rht_bucket_insert(ht, tbl, hash);
 565         if (!pprev)
 566                 return ERR_PTR(-ENOMEM);
 567
 568         head = rht_dereference_bucket(*pprev, tbl, hash);
 569
 570         RCU_INIT_POINTER(obj->next, head);
 571         if (ht->rhlist) {
 572                 struct rhlist_head *list;
 573
 574                 list = container_of(obj, struct rhlist_head, rhead);
 575                 RCU_INIT_POINTER(list->next, NULL);
 576         }
 577
 578         rcu_assign_pointer(*pprev, obj);
 579
 580         atomic_inc(&ht->nelems);
 581         if (rht_grow_above_75(ht, tbl))
 582                 schedule_work(&ht->run_work);
 583
 584         return NULL;
 585 }
 586
 587 static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
 588                                    struct rhash_head *obj)
 589 {
 590         struct bucket_table *new_tbl;
 591         struct bucket_table *tbl;
 592         unsigned int hash;
 593         spinlock_t *lock;
 594         void *data;
 595
 596         tbl = rcu_dereference(ht->tbl);
 597
 598         /* All insertions must grab the oldest table containing
 599          * the hashed bucket that is yet to be rehashed.
 600          */
 601         for (;;) {
 602                 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
 603                 lock = rht_bucket_lock(tbl, hash);
 604                 spin_lock_bh(lock);
 605
 606                 if (tbl->rehash <= hash)
 607                         break;
 608
 609                 spin_unlock_bh(lock);
 610                 tbl = rcu_dereference(tbl->future_tbl);
 611         }
 612
 613         data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
 614         new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
 615         if (PTR_ERR(new_tbl) != -EEXIST)
 616                 data = ERR_CAST(new_tbl);
 617
 618         while (!IS_ERR_OR_NULL(new_tbl)) {
 619                 tbl = new_tbl;
 620                 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
 621                 spin_lock_nested(rht_bucket_lock(tbl, hash),
 622                                  SINGLE_DEPTH_NESTING);
 623
 624                 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
 625                 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
 626                 if (PTR_ERR(new_tbl) != -EEXIST)
 627                         data = ERR_CAST(new_tbl);
 628
 629                 spin_unlock(rht_bucket_lock(tbl, hash));
 630         }
 631
 632         spin_unlock_bh(lock);
 633
 634         if (PTR_ERR(data) == -EAGAIN)
 635                 data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
 636                                -EAGAIN);
 637
 638         return data;
 639 }
 640
 641 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
 642                              struct rhash_head *obj)
 643 {
 644         void *data;
 645
 646         do {
 647                 rcu_read_lock();
 648                 data = rhashtable_try_insert(ht, key, obj);
 649                 rcu_read_unlock();
 650         } while (PTR_ERR(data) == -EAGAIN);
 651
 652         return data;
 653 }
 654 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
 655
 656 /**
 657  * rhashtable_walk_enter - Initialise an iterator
 658  * @ht:         Table to walk over
 659  * @iter:       Hash table Iterator
 660  *
 661  * This function prepares a hash table walk.
 662  *
 663  * Note that if you restart a walk after rhashtable_walk_stop you
 664  * may see the same object twice.  Also, you may miss objects if
 665  * there are removals in between rhashtable_walk_stop and the next
 666  * call to rhashtable_walk_start.
 667  *
 668  * For a completely stable walk you should construct your own data
 669  * structure outside the hash table.
 670  *
 671  * This function may sleep so you must not call it from interrupt
 672  * context or with spin locks held.
 673  *
 674  * You must call rhashtable_walk_exit after this function returns.
 675  */
 676 void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
 677 {
 678         iter->ht = ht;
 679         iter->p = NULL;
 680         iter->slot = 0;
 681         iter->skip = 0;
 682         iter->end_of_table = 0;
 683
 684         spin_lock(&ht->lock);
 685         iter->walker.tbl =
 686                 rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
 687         list_add(&iter->walker.list, &iter->walker.tbl->walkers);
 688         spin_unlock(&ht->lock);
 689 }
 690 EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
 691
 692 /**
 693  * rhashtable_walk_exit - Free an iterator
 694  * @iter:       Hash table Iterator
 695  *
 696  * This function frees resources allocated by rhashtable_walk_init.
 697  */
 698 void rhashtable_walk_exit(struct rhashtable_iter *iter)
 699 {
 700         spin_lock(&iter->ht->lock);
 701         if (iter->walker.tbl)
 702                 list_del(&iter->walker.list);
 703         spin_unlock(&iter->ht->lock);
 704 }
 705 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
 706
 707 /**
 708  * rhashtable_walk_start_check - Start a hash table walk
 709  * @iter:       Hash table iterator
 710  *
 711  * Start a hash table walk at the current iterator position.  Note that we take
 712  * the RCU lock in all cases including when we return an error.  So you must
 713  * always call rhashtable_walk_stop to clean up.
 714  *
 715  * Returns zero if successful.
 716  *
 717  * Returns -EAGAIN if resize event occured.  Note that the iterator
 718  * will rewind back to the beginning and you may use it immediately
 719  * by calling rhashtable_walk_next.
 720  *
 721  * rhashtable_walk_start is defined as an inline variant that returns
 722  * void. This is preferred in cases where the caller would ignore
 723  * resize events and always continue.
 724  */
 725 int rhashtable_walk_start_check(struct rhashtable_iter *iter)
 726         __acquires(RCU)
 727 {
 728         struct rhashtable *ht = iter->ht;
 729
 730         rcu_read_lock();
 731
 732         spin_lock(&ht->lock);
 733         if (iter->walker.tbl)
 734                 list_del(&iter->walker.list);
 735         spin_unlock(&ht->lock);
 736
 737         if (!iter->walker.tbl && !iter->end_of_table) {
 738                 iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
 739                 return -EAGAIN;
 740         }
 741
 742         return 0;
 743 }
 744 EXPORT_SYMBOL_GPL(rhashtable_walk_start_check);
 745
 746 /**
 747  * __rhashtable_walk_find_next - Find the next element in a table (or the first
 748  * one in case of a new walk).
 749  *
 750  * @iter:       Hash table iterator
 751  *
 752  * Returns the found object or NULL when the end of the table is reached.
 753  *
 754  * Returns -EAGAIN if resize event occurred.
 755  */
 756 static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter)
 757 {
 758         struct bucket_table *tbl = iter->walker.tbl;
 759         struct rhlist_head *list = iter->list;
 760         struct rhashtable *ht = iter->ht;
 761         struct rhash_head *p = iter->p;
 762         bool rhlist = ht->rhlist;
 763
 764         if (!tbl)
 765                 return NULL;
 766
 767         for (; iter->slot < tbl->size; iter->slot++) {
 768                 int skip = iter->skip;
 769
 770                 rht_for_each_rcu(p, tbl, iter->slot) {
 771                         if (rhlist) {
 772                                 list = container_of(p, struct rhlist_head,
 773                                                     rhead);
 774                                 do {
 775                                         if (!skip)
 776                                                 goto next;
 777                                         skip--;
 778                                         list = rcu_dereference(list->next);
 779                                 } while (list);
 780
 781                                 continue;
 782                         }
 783                         if (!skip)
 784                                 break;
 785                         skip--;
 786                 }
 787
 788 next:
 789                 if (!rht_is_a_nulls(p)) {
 790                         iter->skip++;
 791                         iter->p = p;
 792                         iter->list = list;
 793                         return rht_obj(ht, rhlist ? &list->rhead : p);
 794                 }
 795
 796                 iter->skip = 0;
 797         }
 798
 799         iter->p = NULL;
 800
 801         /* Ensure we see any new tables. */
 802         smp_rmb();
 803
 804         iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 805         if (iter->walker.tbl) {
 806                 iter->slot = 0;
 807                 iter->skip = 0;
 808                 return ERR_PTR(-EAGAIN);
 809         } else {
 810                 iter->end_of_table = true;
 811         }
 812
 813         return NULL;
 814 }
 815
 816 /**
 817  * rhashtable_walk_next - Return the next object and advance the iterator
 818  * @iter:       Hash table iterator
 819  *
 820  * Note that you must call rhashtable_walk_stop when you are finished
 821  * with the walk.
 822  *
 823  * Returns the next object or NULL when the end of the table is reached.
 824  *
 825  * Returns -EAGAIN if resize event occurred.  Note that the iterator
 826  * will rewind back to the beginning and you may continue to use it.
 827  */
 828 void *rhashtable_walk_next(struct rhashtable_iter *iter)
 829 {
 830         struct rhlist_head *list = iter->list;
 831         struct rhashtable *ht = iter->ht;
 832         struct rhash_head *p = iter->p;
 833         bool rhlist = ht->rhlist;
 834
 835         if (p) {
 836                 if (!rhlist || !(list = rcu_dereference(list->next))) {
 837                         p = rcu_dereference(p->next);
 838                         list = container_of(p, struct rhlist_head, rhead);
 839                 }
 840                 if (!rht_is_a_nulls(p)) {
 841                         iter->skip++;
 842                         iter->p = p;
 843                         iter->list = list;
 844                         return rht_obj(ht, rhlist ? &list->rhead : p);
 845                 }
 846
 847                 /* At the end of this slot, switch to next one and then find
 848                  * next entry from that point.
 849                  */
 850                 iter->skip = 0;
 851                 iter->slot++;
 852         }
 853
 854         return __rhashtable_walk_find_next(iter);
 855 }
 856 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
 857
 858 /**
 859  * rhashtable_walk_peek - Return the next object but don't advance the iterator
 860  * @iter:       Hash table iterator
 861  *
 862  * Returns the next object or NULL when the end of the table is reached.
 863  *
 864  * Returns -EAGAIN if resize event occurred.  Note that the iterator
 865  * will rewind back to the beginning and you may continue to use it.
 866  */
 867 void *rhashtable_walk_peek(struct rhashtable_iter *iter)
 868 {
 869         struct rhlist_head *list = iter->list;
 870         struct rhashtable *ht = iter->ht;
 871         struct rhash_head *p = iter->p;
 872
 873         if (p)
 874                 return rht_obj(ht, ht->rhlist ? &list->rhead : p);
 875
 876         /* No object found in current iter, find next one in the table. */
 877
 878         if (iter->skip) {
 879                 /* A nonzero skip value points to the next entry in the table
 880                  * beyond that last one that was found. Decrement skip so
 881                  * we find the current value. __rhashtable_walk_find_next
 882                  * will restore the original value of skip assuming that
 883                  * the table hasn't changed.
 884                  */
 885                 iter->skip--;
 886         }
 887
 888         return __rhashtable_walk_find_next(iter);
 889 }
 890 EXPORT_SYMBOL_GPL(rhashtable_walk_peek);
 891
 892 /**
 893  * rhashtable_walk_stop - Finish a hash table walk
 894  * @iter:       Hash table iterator
 895  *
 896  * Finish a hash table walk.  Does not reset the iterator to the start of the
 897  * hash table.
 898  */
 899 void rhashtable_walk_stop(struct rhashtable_iter *iter)
 900         __releases(RCU)
 901 {
 902         struct rhashtable *ht;
 903         struct bucket_table *tbl = iter->walker.tbl;
 904
 905         if (!tbl)
 906                 goto out;
 907
 908         ht = iter->ht;
 909
 910         spin_lock(&ht->lock);
 911         if (tbl->rehash < tbl->size)
 912                 list_add(&iter->walker.list, &tbl->walkers);
 913         else
 914                 iter->walker.tbl = NULL;
 915         spin_unlock(&ht->lock);
 916
 917         iter->p = NULL;
 918
 919 out:
 920         rcu_read_unlock();
 921 }
 922 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
 923
 924 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
 925 {
 926         return max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
 927                    (unsigned long)params->min_size);
 928 }
 929
 930 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
 931 {
 932         return jhash2(key, length, seed);
 933 }
 934
 935 /**
 936  * rhashtable_init - initialize a new hash table
 937  * @ht:         hash table to be initialized
 938  * @params:     configuration parameters
 939  *
 940  * Initializes a new hash table based on the provided configuration
 941  * parameters. A table can be configured either with a variable or
 942  * fixed length key:
 943  *
 944  * Configuration Example 1: Fixed length keys
 945  * struct test_obj {
 946  *      int                     key;
 947  *      void *                  my_member;
 948  *      struct rhash_head       node;
 949  * };
 950  *
 951  * struct rhashtable_params params = {
 952  *      .head_offset = offsetof(struct test_obj, node),
 953  *      .key_offset = offsetof(struct test_obj, key),
 954  *      .key_len = sizeof(int),
 955  *      .hashfn = jhash,
 956  *      .nulls_base = (1U << RHT_BASE_SHIFT),
 957  * };
 958  *
 959  * Configuration Example 2: Variable length keys
 960  * struct test_obj {
 961  *      [...]
 962  *      struct rhash_head       node;
 963  * };
 964  *
 965  * u32 my_hash_fn(const void *data, u32 len, u32 seed)
 966  * {
 967  *      struct test_obj *obj = data;
 968  *
 969  *      return [... hash ...];
 970  * }
 971  *
 972  * struct rhashtable_params params = {
 973  *      .head_offset = offsetof(struct test_obj, node),
 974  *      .hashfn = jhash,
 975  *      .obj_hashfn = my_hash_fn,
 976  * };
 977  */
 978 int rhashtable_init(struct rhashtable *ht,
 979                     const struct rhashtable_params *params)
 980 {
 981         struct bucket_table *tbl;
 982         size_t size;
 983
 984         size = HASH_DEFAULT_SIZE;
 985
 986         if ((!params->key_len && !params->obj_hashfn) ||
 987             (params->obj_hashfn && !params->obj_cmpfn))
 988                 return -EINVAL;
 989
 990         if (params->nulls_base && params->nulls_base < (1U << RHT_BASE_SHIFT))
 991                 return -EINVAL;
 992
 993         memset(ht, 0, sizeof(*ht));
 994         mutex_init(&ht->mutex);
 995         spin_lock_init(&ht->lock);
 996         memcpy(&ht->p, params, sizeof(*params));
 997
 998         if (params->min_size)
 999                 ht->p.min_size = roundup_pow_of_two(params->min_size);
1000
1001         /* Cap total entries at 2^31 to avoid nelems overflow. */
1002         ht->max_elems = 1u << 31;
1003
1004         if (params->max_size) {
1005                 ht->p.max_size = rounddown_pow_of_two(params->max_size);
1006                 if (ht->p.max_size < ht->max_elems / 2)
1007                         ht->max_elems = ht->p.max_size * 2;
1008         }
1009
1010         ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
1011
1012         if (params->nelem_hint)
1013                 size = rounded_hashtable_size(&ht->p);
1014
1015         if (params->locks_mul)
1016                 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
1017         else
1018                 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
1019
1020         ht->key_len = ht->p.key_len;
1021         if (!params->hashfn) {
1022                 ht->p.hashfn = jhash;
1023
1024                 if (!(ht->key_len & (sizeof(u32) - 1))) {
1025                         ht->key_len /= sizeof(u32);
1026                         ht->p.hashfn = rhashtable_jhash2;
1027                 }
1028         }
1029
1030         tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
1031         if (tbl == NULL)
1032                 return -ENOMEM;
1033
1034         atomic_set(&ht->nelems, 0);
1035
1036         RCU_INIT_POINTER(ht->tbl, tbl);
1037
1038         INIT_WORK(&ht->run_work, rht_deferred_worker);
1039
1040         return 0;
1041 }
1042 EXPORT_SYMBOL_GPL(rhashtable_init);
1043
1044 /**
1045  * rhltable_init - initialize a new hash list table
1046  * @hlt:        hash list table to be initialized
1047  * @params:     configuration parameters
1048  *
1049  * Initializes a new hash list table.
1050  *
1051  * See documentation for rhashtable_init.
1052  */
1053 int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
1054 {
1055         int err;
1056
1057         /* No rhlist NULLs marking for now. */
1058         if (params->nulls_base)
1059                 return -EINVAL;
1060
1061         err = rhashtable_init(&hlt->ht, params);
1062         hlt->ht.rhlist = true;
1063         return err;
1064 }
1065 EXPORT_SYMBOL_GPL(rhltable_init);
1066
1067 static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
1068                                 void (*free_fn)(void *ptr, void *arg),
1069                                 void *arg)
1070 {
1071         struct rhlist_head *list;
1072
1073         if (!ht->rhlist) {
1074                 free_fn(rht_obj(ht, obj), arg);
1075                 return;
1076         }
1077
1078         list = container_of(obj, struct rhlist_head, rhead);
1079         do {
1080                 obj = &list->rhead;
1081                 list = rht_dereference(list->next, ht);
1082                 free_fn(rht_obj(ht, obj), arg);
1083         } while (list);
1084 }
1085
1086 /**
1087  * rhashtable_free_and_destroy - free elements and destroy hash table
1088  * @ht:         the hash table to destroy
1089  * @free_fn:    callback to release resources of element
1090  * @arg:        pointer passed to free_fn
1091  *
1092  * Stops an eventual async resize. If defined, invokes free_fn for each
1093  * element to releasal resources. Please note that RCU protected
1094  * readers may still be accessing the elements. Releasing of resources
1095  * must occur in a compatible manner. Then frees the bucket array.
1096  *
1097  * This function will eventually sleep to wait for an async resize
1098  * to complete. The caller is responsible that no further write operations
1099  * occurs in parallel.
1100  */
1101 void rhashtable_free_and_destroy(struct rhashtable *ht,
1102                                  void (*free_fn)(void *ptr, void *arg),
1103                                  void *arg)
1104 {
1105         struct bucket_table *tbl;
1106         unsigned int i;
1107
1108         cancel_work_sync(&ht->run_work);
1109
1110         mutex_lock(&ht->mutex);
1111         tbl = rht_dereference(ht->tbl, ht);
1112         if (free_fn) {
1113                 for (i = 0; i < tbl->size; i++) {
1114                         struct rhash_head *pos, *next;
1115
1116                         cond_resched();
1117                         for (pos = rht_dereference(*rht_bucket(tbl, i), ht),
1118                              next = !rht_is_a_nulls(pos) ?
1119                                         rht_dereference(pos->next, ht) : NULL;
1120                              !rht_is_a_nulls(pos);
1121                              pos = next,
1122                              next = !rht_is_a_nulls(pos) ?
1123                                         rht_dereference(pos->next, ht) : NULL)
1124                                 rhashtable_free_one(ht, pos, free_fn, arg);
1125                 }
1126         }
1127
1128         bucket_table_free(tbl);
1129         mutex_unlock(&ht->mutex);
1130 }
1131 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
1132
1133 void rhashtable_destroy(struct rhashtable *ht)
1134 {
1135         return rhashtable_free_and_destroy(ht, NULL, NULL);
1136 }
1137 EXPORT_SYMBOL_GPL(rhashtable_destroy);
1138
1139 struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
1140                                             unsigned int hash)
1141 {
1142         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1143         static struct rhash_head __rcu *rhnull =
1144                 (struct rhash_head __rcu *)NULLS_MARKER(0);
1145         unsigned int index = hash & ((1 << tbl->nest) - 1);
1146         unsigned int size = tbl->size >> tbl->nest;
1147         unsigned int subhash = hash;
1148         union nested_table *ntbl;
1149
1150         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1151         ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash);
1152         subhash >>= tbl->nest;
1153
1154         while (ntbl && size > (1 << shift)) {
1155                 index = subhash & ((1 << shift) - 1);
1156                 ntbl = rht_dereference_bucket_rcu(ntbl[index].table,
1157                                                   tbl, hash);
1158                 size >>= shift;
1159                 subhash >>= shift;
1160         }
1161
1162         if (!ntbl)
1163                 return &rhnull;
1164
1165         return &ntbl[subhash].bucket;
1166
1167 }
1168 EXPORT_SYMBOL_GPL(rht_bucket_nested);
1169
1170 struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
1171                                                    struct bucket_table *tbl,
1172                                                    unsigned int hash)
1173 {
1174         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1175         unsigned int index = hash & ((1 << tbl->nest) - 1);
1176         unsigned int size = tbl->size >> tbl->nest;
1177         union nested_table *ntbl;
1178         unsigned int shifted;
1179         unsigned int nhash;
1180
1181         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1182         hash >>= tbl->nest;
1183         nhash = index;
1184         shifted = tbl->nest;
1185         ntbl = nested_table_alloc(ht, &ntbl[index].table,
1186                                   size <= (1 << shift) ? shifted : 0, nhash);
1187
1188         while (ntbl && size > (1 << shift)) {
1189                 index = hash & ((1 << shift) - 1);
1190                 size >>= shift;
1191                 hash >>= shift;
1192                 nhash |= index << shifted;
1193                 shifted += shift;
1194                 ntbl = nested_table_alloc(ht, &ntbl[index].table,
1195                                           size <= (1 << shift) ? shifted : 0,
1196                                           nhash);
1197         }
1198
1199         if (!ntbl)
1200                 return NULL;
1201
1202         return &ntbl[hash].bucket;
1203
1204 }
1205 EXPORT_SYMBOL_GPL(rht_bucket_nested_insert);