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                                               bool leaf)
 119 {
 120         union nested_table *ntbl;
 121         int i;
 122
 123         ntbl = rcu_dereference(*prev);
 124         if (ntbl)
 125                 return ntbl;
 126
 127         ntbl = kzalloc(PAGE_SIZE, GFP_ATOMIC);
 128
 129         if (ntbl && leaf) {
 130                 for (i = 0; i < PAGE_SIZE / sizeof(ntbl[0]); i++)
 131                         INIT_RHT_NULLS_HEAD(ntbl[i].bucket);
 132         }
 133
 134         rcu_assign_pointer(*prev, ntbl);
 135
 136         return ntbl;
 137 }
 138
 139 static struct bucket_table *nested_bucket_table_alloc(struct rhashtable *ht,
 140                                                       size_t nbuckets,
 141                                                       gfp_t gfp)
 142 {
 143         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
 144         struct bucket_table *tbl;
 145         size_t size;
 146
 147         if (nbuckets < (1 << (shift + 1)))
 148                 return NULL;
 149
 150         size = sizeof(*tbl) + sizeof(tbl->buckets[0]);
 151
 152         tbl = kzalloc(size, gfp);
 153         if (!tbl)
 154                 return NULL;
 155
 156         if (!nested_table_alloc(ht, (union nested_table __rcu **)tbl->buckets,
 157                                 false)) {
 158                 kfree(tbl);
 159                 return NULL;
 160         }
 161
 162         tbl->nest = (ilog2(nbuckets) - 1) % shift + 1;
 163
 164         return tbl;
 165 }
 166
 167 static struct bucket_table *bucket_table_alloc(struct rhashtable *ht,
 168                                                size_t nbuckets,
 169                                                gfp_t gfp)
 170 {
 171         struct bucket_table *tbl = NULL;
 172         size_t size, max_locks;
 173         int i;
 174
 175         size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]);
 176         tbl = kvzalloc(size, gfp);
 177
 178         size = nbuckets;
 179
 180         if (tbl == NULL && (gfp & ~__GFP_NOFAIL) != GFP_KERNEL) {
 181                 tbl = nested_bucket_table_alloc(ht, nbuckets, gfp);
 182                 nbuckets = 0;
 183         }
 184
 185         if (tbl == NULL)
 186                 return NULL;
 187
 188         tbl->size = size;
 189
 190         max_locks = size >> 1;
 191         if (tbl->nest)
 192                 max_locks = min_t(size_t, max_locks, 1U << tbl->nest);
 193
 194         if (alloc_bucket_spinlocks(&tbl->locks, &tbl->locks_mask, max_locks,
 195                                    ht->p.locks_mul, gfp) < 0) {
 196                 bucket_table_free(tbl);
 197                 return NULL;
 198         }
 199
 200         INIT_LIST_HEAD(&tbl->walkers);
 201
 202         tbl->hash_rnd = get_random_u32();
 203
 204         for (i = 0; i < nbuckets; i++)
 205                 INIT_RHT_NULLS_HEAD(tbl->buckets[i]);
 206
 207         return tbl;
 208 }
 209
 210 static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
 211                                                   struct bucket_table *tbl)
 212 {
 213         struct bucket_table *new_tbl;
 214
 215         do {
 216                 new_tbl = tbl;
 217                 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 218         } while (tbl);
 219
 220         return new_tbl;
 221 }
 222
 223 static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
 224 {
 225         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 226         struct bucket_table *new_tbl = rhashtable_last_table(ht, old_tbl);
 227         struct rhash_head __rcu **pprev = rht_bucket_var(old_tbl, old_hash);
 228         int err = -EAGAIN;
 229         struct rhash_head *head, *next, *entry;
 230         spinlock_t *new_bucket_lock;
 231         unsigned int new_hash;
 232
 233         if (new_tbl->nest)
 234                 goto out;
 235
 236         err = -ENOENT;
 237
 238         rht_for_each(entry, old_tbl, old_hash) {
 239                 err = 0;
 240                 next = rht_dereference_bucket(entry->next, old_tbl, old_hash);
 241
 242                 if (rht_is_a_nulls(next))
 243                         break;
 244
 245                 pprev = &entry->next;
 246         }
 247
 248         if (err)
 249                 goto out;
 250
 251         new_hash = head_hashfn(ht, new_tbl, entry);
 252
 253         new_bucket_lock = rht_bucket_lock(new_tbl, new_hash);
 254
 255         spin_lock_nested(new_bucket_lock, SINGLE_DEPTH_NESTING);
 256         head = rht_dereference_bucket(new_tbl->buckets[new_hash],
 257                                       new_tbl, new_hash);
 258
 259         RCU_INIT_POINTER(entry->next, head);
 260
 261         rcu_assign_pointer(new_tbl->buckets[new_hash], entry);
 262         spin_unlock(new_bucket_lock);
 263
 264         rcu_assign_pointer(*pprev, next);
 265
 266 out:
 267         return err;
 268 }
 269
 270 static int rhashtable_rehash_chain(struct rhashtable *ht,
 271                                     unsigned int old_hash)
 272 {
 273         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 274         spinlock_t *old_bucket_lock;
 275         int err;
 276
 277         old_bucket_lock = rht_bucket_lock(old_tbl, old_hash);
 278
 279         spin_lock_bh(old_bucket_lock);
 280         while (!(err = rhashtable_rehash_one(ht, old_hash)))
 281                 ;
 282
 283         if (err == -ENOENT) {
 284                 old_tbl->rehash++;
 285                 err = 0;
 286         }
 287         spin_unlock_bh(old_bucket_lock);
 288
 289         return err;
 290 }
 291
 292 static int rhashtable_rehash_attach(struct rhashtable *ht,
 293                                     struct bucket_table *old_tbl,
 294                                     struct bucket_table *new_tbl)
 295 {
 296         /* Make insertions go into the new, empty table right away. Deletions
 297          * and lookups will be attempted in both tables until we synchronize.
 298          * As cmpxchg() provides strong barriers, we do not need
 299          * rcu_assign_pointer().
 300          */
 301
 302         if (cmpxchg(&old_tbl->future_tbl, NULL, new_tbl) != NULL)
 303                 return -EEXIST;
 304
 305         return 0;
 306 }
 307
 308 static int rhashtable_rehash_table(struct rhashtable *ht)
 309 {
 310         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 311         struct bucket_table *new_tbl;
 312         struct rhashtable_walker *walker;
 313         unsigned int old_hash;
 314         int err;
 315
 316         new_tbl = rht_dereference(old_tbl->future_tbl, ht);
 317         if (!new_tbl)
 318                 return 0;
 319
 320         for (old_hash = 0; old_hash < old_tbl->size; old_hash++) {
 321                 err = rhashtable_rehash_chain(ht, old_hash);
 322                 if (err)
 323                         return err;
 324                 cond_resched();
 325         }
 326
 327         /* Publish the new table pointer. */
 328         rcu_assign_pointer(ht->tbl, new_tbl);
 329
 330         spin_lock(&ht->lock);
 331         list_for_each_entry(walker, &old_tbl->walkers, list)
 332                 walker->tbl = NULL;
 333         spin_unlock(&ht->lock);
 334
 335         /* Wait for readers. All new readers will see the new
 336          * table, and thus no references to the old table will
 337          * remain.
 338          */
 339         call_rcu(&old_tbl->rcu, bucket_table_free_rcu);
 340
 341         return rht_dereference(new_tbl->future_tbl, ht) ? -EAGAIN : 0;
 342 }
 343
 344 static int rhashtable_rehash_alloc(struct rhashtable *ht,
 345                                    struct bucket_table *old_tbl,
 346                                    unsigned int size)
 347 {
 348         struct bucket_table *new_tbl;
 349         int err;
 350
 351         ASSERT_RHT_MUTEX(ht);
 352
 353         new_tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
 354         if (new_tbl == NULL)
 355                 return -ENOMEM;
 356
 357         err = rhashtable_rehash_attach(ht, old_tbl, new_tbl);
 358         if (err)
 359                 bucket_table_free(new_tbl);
 360
 361         return err;
 362 }
 363
 364 /**
 365  * rhashtable_shrink - Shrink hash table while allowing concurrent lookups
 366  * @ht:         the hash table to shrink
 367  *
 368  * This function shrinks the hash table to fit, i.e., the smallest
 369  * size would not cause it to expand right away automatically.
 370  *
 371  * The caller must ensure that no concurrent resizing occurs by holding
 372  * ht->mutex.
 373  *
 374  * The caller must ensure that no concurrent table mutations take place.
 375  * It is however valid to have concurrent lookups if they are RCU protected.
 376  *
 377  * It is valid to have concurrent insertions and deletions protected by per
 378  * bucket locks or concurrent RCU protected lookups and traversals.
 379  */
 380 static int rhashtable_shrink(struct rhashtable *ht)
 381 {
 382         struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 383         unsigned int nelems = atomic_read(&ht->nelems);
 384         unsigned int size = 0;
 385
 386         if (nelems)
 387                 size = roundup_pow_of_two(nelems * 3 / 2);
 388         if (size < ht->p.min_size)
 389                 size = ht->p.min_size;
 390
 391         if (old_tbl->size <= size)
 392                 return 0;
 393
 394         if (rht_dereference(old_tbl->future_tbl, ht))
 395                 return -EEXIST;
 396
 397         return rhashtable_rehash_alloc(ht, old_tbl, size);
 398 }
 399
 400 static void rht_deferred_worker(struct work_struct *work)
 401 {
 402         struct rhashtable *ht;
 403         struct bucket_table *tbl;
 404         int err = 0;
 405
 406         ht = container_of(work, struct rhashtable, run_work);
 407         mutex_lock(&ht->mutex);
 408
 409         tbl = rht_dereference(ht->tbl, ht);
 410         tbl = rhashtable_last_table(ht, tbl);
 411
 412         if (rht_grow_above_75(ht, tbl))
 413                 err = rhashtable_rehash_alloc(ht, tbl, tbl->size * 2);
 414         else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
 415                 err = rhashtable_shrink(ht);
 416         else if (tbl->nest)
 417                 err = rhashtable_rehash_alloc(ht, tbl, tbl->size);
 418
 419         if (!err)
 420                 err = rhashtable_rehash_table(ht);
 421
 422         mutex_unlock(&ht->mutex);
 423
 424         if (err)
 425                 schedule_work(&ht->run_work);
 426 }
 427
 428 static int rhashtable_insert_rehash(struct rhashtable *ht,
 429                                     struct bucket_table *tbl)
 430 {
 431         struct bucket_table *old_tbl;
 432         struct bucket_table *new_tbl;
 433         unsigned int size;
 434         int err;
 435
 436         old_tbl = rht_dereference_rcu(ht->tbl, ht);
 437
 438         size = tbl->size;
 439
 440         err = -EBUSY;
 441
 442         if (rht_grow_above_75(ht, tbl))
 443                 size *= 2;
 444         /* Do not schedule more than one rehash */
 445         else if (old_tbl != tbl)
 446                 goto fail;
 447
 448         err = -ENOMEM;
 449
 450         new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC | __GFP_NOWARN);
 451         if (new_tbl == NULL)
 452                 goto fail;
 453
 454         err = rhashtable_rehash_attach(ht, tbl, new_tbl);
 455         if (err) {
 456                 bucket_table_free(new_tbl);
 457                 if (err == -EEXIST)
 458                         err = 0;
 459         } else
 460                 schedule_work(&ht->run_work);
 461
 462         return err;
 463
 464 fail:
 465         /* Do not fail the insert if someone else did a rehash. */
 466         if (likely(rcu_access_pointer(tbl->future_tbl)))
 467                 return 0;
 468
 469         /* Schedule async rehash to retry allocation in process context. */
 470         if (err == -ENOMEM)
 471                 schedule_work(&ht->run_work);
 472
 473         return err;
 474 }
 475
 476 static void *rhashtable_lookup_one(struct rhashtable *ht,
 477                                    struct bucket_table *tbl, unsigned int hash,
 478                                    const void *key, struct rhash_head *obj)
 479 {
 480         struct rhashtable_compare_arg arg = {
 481                 .ht = ht,
 482                 .key = key,
 483         };
 484         struct rhash_head __rcu **pprev;
 485         struct rhash_head *head;
 486         int elasticity;
 487
 488         elasticity = RHT_ELASTICITY;
 489         pprev = rht_bucket_var(tbl, hash);
 490         rht_for_each_continue(head, *pprev, tbl, hash) {
 491                 struct rhlist_head *list;
 492                 struct rhlist_head *plist;
 493
 494                 elasticity--;
 495                 if (!key ||
 496                     (ht->p.obj_cmpfn ?
 497                      ht->p.obj_cmpfn(&arg, rht_obj(ht, head)) :
 498                      rhashtable_compare(&arg, rht_obj(ht, head)))) {
 499                         pprev = &head->next;
 500                         continue;
 501                 }
 502
 503                 if (!ht->rhlist)
 504                         return rht_obj(ht, head);
 505
 506                 list = container_of(obj, struct rhlist_head, rhead);
 507                 plist = container_of(head, struct rhlist_head, rhead);
 508
 509                 RCU_INIT_POINTER(list->next, plist);
 510                 head = rht_dereference_bucket(head->next, tbl, hash);
 511                 RCU_INIT_POINTER(list->rhead.next, head);
 512                 rcu_assign_pointer(*pprev, obj);
 513
 514                 return NULL;
 515         }
 516
 517         if (elasticity <= 0)
 518                 return ERR_PTR(-EAGAIN);
 519
 520         return ERR_PTR(-ENOENT);
 521 }
 522
 523 static struct bucket_table *rhashtable_insert_one(struct rhashtable *ht,
 524                                                   struct bucket_table *tbl,
 525                                                   unsigned int hash,
 526                                                   struct rhash_head *obj,
 527                                                   void *data)
 528 {
 529         struct rhash_head __rcu **pprev;
 530         struct bucket_table *new_tbl;
 531         struct rhash_head *head;
 532
 533         if (!IS_ERR_OR_NULL(data))
 534                 return ERR_PTR(-EEXIST);
 535
 536         if (PTR_ERR(data) != -EAGAIN && PTR_ERR(data) != -ENOENT)
 537                 return ERR_CAST(data);
 538
 539         new_tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 540         if (new_tbl)
 541                 return new_tbl;
 542
 543         if (PTR_ERR(data) != -ENOENT)
 544                 return ERR_CAST(data);
 545
 546         if (unlikely(rht_grow_above_max(ht, tbl)))
 547                 return ERR_PTR(-E2BIG);
 548
 549         if (unlikely(rht_grow_above_100(ht, tbl)))
 550                 return ERR_PTR(-EAGAIN);
 551
 552         pprev = rht_bucket_insert(ht, tbl, hash);
 553         if (!pprev)
 554                 return ERR_PTR(-ENOMEM);
 555
 556         head = rht_dereference_bucket(*pprev, tbl, hash);
 557
 558         RCU_INIT_POINTER(obj->next, head);
 559         if (ht->rhlist) {
 560                 struct rhlist_head *list;
 561
 562                 list = container_of(obj, struct rhlist_head, rhead);
 563                 RCU_INIT_POINTER(list->next, NULL);
 564         }
 565
 566         rcu_assign_pointer(*pprev, obj);
 567
 568         atomic_inc(&ht->nelems);
 569         if (rht_grow_above_75(ht, tbl))
 570                 schedule_work(&ht->run_work);
 571
 572         return NULL;
 573 }
 574
 575 static void *rhashtable_try_insert(struct rhashtable *ht, const void *key,
 576                                    struct rhash_head *obj)
 577 {
 578         struct bucket_table *new_tbl;
 579         struct bucket_table *tbl;
 580         unsigned int hash;
 581         spinlock_t *lock;
 582         void *data;
 583
 584         tbl = rcu_dereference(ht->tbl);
 585
 586         /* All insertions must grab the oldest table containing
 587          * the hashed bucket that is yet to be rehashed.
 588          */
 589         for (;;) {
 590                 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
 591                 lock = rht_bucket_lock(tbl, hash);
 592                 spin_lock_bh(lock);
 593
 594                 if (tbl->rehash <= hash)
 595                         break;
 596
 597                 spin_unlock_bh(lock);
 598                 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 599         }
 600
 601         data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
 602         new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
 603         if (PTR_ERR(new_tbl) != -EEXIST)
 604                 data = ERR_CAST(new_tbl);
 605
 606         while (!IS_ERR_OR_NULL(new_tbl)) {
 607                 tbl = new_tbl;
 608                 hash = rht_head_hashfn(ht, tbl, obj, ht->p);
 609                 spin_lock_nested(rht_bucket_lock(tbl, hash),
 610                                  SINGLE_DEPTH_NESTING);
 611
 612                 data = rhashtable_lookup_one(ht, tbl, hash, key, obj);
 613                 new_tbl = rhashtable_insert_one(ht, tbl, hash, obj, data);
 614                 if (PTR_ERR(new_tbl) != -EEXIST)
 615                         data = ERR_CAST(new_tbl);
 616
 617                 spin_unlock(rht_bucket_lock(tbl, hash));
 618         }
 619
 620         spin_unlock_bh(lock);
 621
 622         if (PTR_ERR(data) == -EAGAIN)
 623                 data = ERR_PTR(rhashtable_insert_rehash(ht, tbl) ?:
 624                                -EAGAIN);
 625
 626         return data;
 627 }
 628
 629 void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
 630                              struct rhash_head *obj)
 631 {
 632         void *data;
 633
 634         do {
 635                 rcu_read_lock();
 636                 data = rhashtable_try_insert(ht, key, obj);
 637                 rcu_read_unlock();
 638         } while (PTR_ERR(data) == -EAGAIN);
 639
 640         return data;
 641 }
 642 EXPORT_SYMBOL_GPL(rhashtable_insert_slow);
 643
 644 /**
 645  * rhashtable_walk_enter - Initialise an iterator
 646  * @ht:         Table to walk over
 647  * @iter:       Hash table Iterator
 648  *
 649  * This function prepares a hash table walk.
 650  *
 651  * Note that if you restart a walk after rhashtable_walk_stop you
 652  * may see the same object twice.  Also, you may miss objects if
 653  * there are removals in between rhashtable_walk_stop and the next
 654  * call to rhashtable_walk_start.
 655  *
 656  * For a completely stable walk you should construct your own data
 657  * structure outside the hash table.
 658  *
 659  * This function may be called from any process context, including
 660  * non-preemptable context, but cannot be called from softirq or
 661  * hardirq context.
 662  *
 663  * You must call rhashtable_walk_exit after this function returns.
 664  */
 665 void rhashtable_walk_enter(struct rhashtable *ht, struct rhashtable_iter *iter)
 666 {
 667         iter->ht = ht;
 668         iter->p = NULL;
 669         iter->slot = 0;
 670         iter->skip = 0;
 671         iter->end_of_table = 0;
 672
 673         spin_lock(&ht->lock);
 674         iter->walker.tbl =
 675                 rcu_dereference_protected(ht->tbl, lockdep_is_held(&ht->lock));
 676         list_add(&iter->walker.list, &iter->walker.tbl->walkers);
 677         spin_unlock(&ht->lock);
 678 }
 679 EXPORT_SYMBOL_GPL(rhashtable_walk_enter);
 680
 681 /**
 682  * rhashtable_walk_exit - Free an iterator
 683  * @iter:       Hash table Iterator
 684  *
 685  * This function frees resources allocated by rhashtable_walk_init.
 686  */
 687 void rhashtable_walk_exit(struct rhashtable_iter *iter)
 688 {
 689         spin_lock(&iter->ht->lock);
 690         if (iter->walker.tbl)
 691                 list_del(&iter->walker.list);
 692         spin_unlock(&iter->ht->lock);
 693 }
 694 EXPORT_SYMBOL_GPL(rhashtable_walk_exit);
 695
 696 /**
 697  * rhashtable_walk_start_check - Start a hash table walk
 698  * @iter:       Hash table iterator
 699  *
 700  * Start a hash table walk at the current iterator position.  Note that we take
 701  * the RCU lock in all cases including when we return an error.  So you must
 702  * always call rhashtable_walk_stop to clean up.
 703  *
 704  * Returns zero if successful.
 705  *
 706  * Returns -EAGAIN if resize event occured.  Note that the iterator
 707  * will rewind back to the beginning and you may use it immediately
 708  * by calling rhashtable_walk_next.
 709  *
 710  * rhashtable_walk_start is defined as an inline variant that returns
 711  * void. This is preferred in cases where the caller would ignore
 712  * resize events and always continue.
 713  */
 714 int rhashtable_walk_start_check(struct rhashtable_iter *iter)
 715         __acquires(RCU)
 716 {
 717         struct rhashtable *ht = iter->ht;
 718         bool rhlist = ht->rhlist;
 719
 720         rcu_read_lock();
 721
 722         spin_lock(&ht->lock);
 723         if (iter->walker.tbl)
 724                 list_del(&iter->walker.list);
 725         spin_unlock(&ht->lock);
 726
 727         if (iter->end_of_table)
 728                 return 0;
 729         if (!iter->walker.tbl) {
 730                 iter->walker.tbl = rht_dereference_rcu(ht->tbl, ht);
 731                 iter->slot = 0;
 732                 iter->skip = 0;
 733                 return -EAGAIN;
 734         }
 735
 736         if (iter->p && !rhlist) {
 737                 /*
 738                  * We need to validate that 'p' is still in the table, and
 739                  * if so, update 'skip'
 740                  */
 741                 struct rhash_head *p;
 742                 int skip = 0;
 743                 rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
 744                         skip++;
 745                         if (p == iter->p) {
 746                                 iter->skip = skip;
 747                                 goto found;
 748                         }
 749                 }
 750                 iter->p = NULL;
 751         } else if (iter->p && rhlist) {
 752                 /* Need to validate that 'list' is still in the table, and
 753                  * if so, update 'skip' and 'p'.
 754                  */
 755                 struct rhash_head *p;
 756                 struct rhlist_head *list;
 757                 int skip = 0;
 758                 rht_for_each_rcu(p, iter->walker.tbl, iter->slot) {
 759                         for (list = container_of(p, struct rhlist_head, rhead);
 760                              list;
 761                              list = rcu_dereference(list->next)) {
 762                                 skip++;
 763                                 if (list == iter->list) {
 764                                         iter->p = p;
 765                                         iter->skip = skip;
 766                                         goto found;
 767                                 }
 768                         }
 769                 }
 770                 iter->p = NULL;
 771         }
 772 found:
 773         return 0;
 774 }
 775 EXPORT_SYMBOL_GPL(rhashtable_walk_start_check);
 776
 777 /**
 778  * __rhashtable_walk_find_next - Find the next element in a table (or the first
 779  * one in case of a new walk).
 780  *
 781  * @iter:       Hash table iterator
 782  *
 783  * Returns the found object or NULL when the end of the table is reached.
 784  *
 785  * Returns -EAGAIN if resize event occurred.
 786  */
 787 static void *__rhashtable_walk_find_next(struct rhashtable_iter *iter)
 788 {
 789         struct bucket_table *tbl = iter->walker.tbl;
 790         struct rhlist_head *list = iter->list;
 791         struct rhashtable *ht = iter->ht;
 792         struct rhash_head *p = iter->p;
 793         bool rhlist = ht->rhlist;
 794
 795         if (!tbl)
 796                 return NULL;
 797
 798         for (; iter->slot < tbl->size; iter->slot++) {
 799                 int skip = iter->skip;
 800
 801                 rht_for_each_rcu(p, tbl, iter->slot) {
 802                         if (rhlist) {
 803                                 list = container_of(p, struct rhlist_head,
 804                                                     rhead);
 805                                 do {
 806                                         if (!skip)
 807                                                 goto next;
 808                                         skip--;
 809                                         list = rcu_dereference(list->next);
 810                                 } while (list);
 811
 812                                 continue;
 813                         }
 814                         if (!skip)
 815                                 break;
 816                         skip--;
 817                 }
 818
 819 next:
 820                 if (!rht_is_a_nulls(p)) {
 821                         iter->skip++;
 822                         iter->p = p;
 823                         iter->list = list;
 824                         return rht_obj(ht, rhlist ? &list->rhead : p);
 825                 }
 826
 827                 iter->skip = 0;
 828         }
 829
 830         iter->p = NULL;
 831
 832         /* Ensure we see any new tables. */
 833         smp_rmb();
 834
 835         iter->walker.tbl = rht_dereference_rcu(tbl->future_tbl, ht);
 836         if (iter->walker.tbl) {
 837                 iter->slot = 0;
 838                 iter->skip = 0;
 839                 return ERR_PTR(-EAGAIN);
 840         } else {
 841                 iter->end_of_table = true;
 842         }
 843
 844         return NULL;
 845 }
 846
 847 /**
 848  * rhashtable_walk_next - Return the next object and advance the iterator
 849  * @iter:       Hash table iterator
 850  *
 851  * Note that you must call rhashtable_walk_stop when you are finished
 852  * with the walk.
 853  *
 854  * Returns the next object or NULL when the end of the table is reached.
 855  *
 856  * Returns -EAGAIN if resize event occurred.  Note that the iterator
 857  * will rewind back to the beginning and you may continue to use it.
 858  */
 859 void *rhashtable_walk_next(struct rhashtable_iter *iter)
 860 {
 861         struct rhlist_head *list = iter->list;
 862         struct rhashtable *ht = iter->ht;
 863         struct rhash_head *p = iter->p;
 864         bool rhlist = ht->rhlist;
 865
 866         if (p) {
 867                 if (!rhlist || !(list = rcu_dereference(list->next))) {
 868                         p = rcu_dereference(p->next);
 869                         list = container_of(p, struct rhlist_head, rhead);
 870                 }
 871                 if (!rht_is_a_nulls(p)) {
 872                         iter->skip++;
 873                         iter->p = p;
 874                         iter->list = list;
 875                         return rht_obj(ht, rhlist ? &list->rhead : p);
 876                 }
 877
 878                 /* At the end of this slot, switch to next one and then find
 879                  * next entry from that point.
 880                  */
 881                 iter->skip = 0;
 882                 iter->slot++;
 883         }
 884
 885         return __rhashtable_walk_find_next(iter);
 886 }
 887 EXPORT_SYMBOL_GPL(rhashtable_walk_next);
 888
 889 /**
 890  * rhashtable_walk_peek - Return the next object but don't advance the iterator
 891  * @iter:       Hash table iterator
 892  *
 893  * Returns the next object or NULL when the end of the table is reached.
 894  *
 895  * Returns -EAGAIN if resize event occurred.  Note that the iterator
 896  * will rewind back to the beginning and you may continue to use it.
 897  */
 898 void *rhashtable_walk_peek(struct rhashtable_iter *iter)
 899 {
 900         struct rhlist_head *list = iter->list;
 901         struct rhashtable *ht = iter->ht;
 902         struct rhash_head *p = iter->p;
 903
 904         if (p)
 905                 return rht_obj(ht, ht->rhlist ? &list->rhead : p);
 906
 907         /* No object found in current iter, find next one in the table. */
 908
 909         if (iter->skip) {
 910                 /* A nonzero skip value points to the next entry in the table
 911                  * beyond that last one that was found. Decrement skip so
 912                  * we find the current value. __rhashtable_walk_find_next
 913                  * will restore the original value of skip assuming that
 914                  * the table hasn't changed.
 915                  */
 916                 iter->skip--;
 917         }
 918
 919         return __rhashtable_walk_find_next(iter);
 920 }
 921 EXPORT_SYMBOL_GPL(rhashtable_walk_peek);
 922
 923 /**
 924  * rhashtable_walk_stop - Finish a hash table walk
 925  * @iter:       Hash table iterator
 926  *
 927  * Finish a hash table walk.  Does not reset the iterator to the start of the
 928  * hash table.
 929  */
 930 void rhashtable_walk_stop(struct rhashtable_iter *iter)
 931         __releases(RCU)
 932 {
 933         struct rhashtable *ht;
 934         struct bucket_table *tbl = iter->walker.tbl;
 935
 936         if (!tbl)
 937                 goto out;
 938
 939         ht = iter->ht;
 940
 941         spin_lock(&ht->lock);
 942         if (tbl->rehash < tbl->size)
 943                 list_add(&iter->walker.list, &tbl->walkers);
 944         else
 945                 iter->walker.tbl = NULL;
 946         spin_unlock(&ht->lock);
 947
 948 out:
 949         rcu_read_unlock();
 950 }
 951 EXPORT_SYMBOL_GPL(rhashtable_walk_stop);
 952
 953 static size_t rounded_hashtable_size(const struct rhashtable_params *params)
 954 {
 955         size_t retsize;
 956
 957         if (params->nelem_hint)
 958                 retsize = max(roundup_pow_of_two(params->nelem_hint * 4 / 3),
 959                               (unsigned long)params->min_size);
 960         else
 961                 retsize = max(HASH_DEFAULT_SIZE,
 962                               (unsigned long)params->min_size);
 963
 964         return retsize;
 965 }
 966
 967 static u32 rhashtable_jhash2(const void *key, u32 length, u32 seed)
 968 {
 969         return jhash2(key, length, seed);
 970 }
 971
 972 /**
 973  * rhashtable_init - initialize a new hash table
 974  * @ht:         hash table to be initialized
 975  * @params:     configuration parameters
 976  *
 977  * Initializes a new hash table based on the provided configuration
 978  * parameters. A table can be configured either with a variable or
 979  * fixed length key:
 980  *
 981  * Configuration Example 1: Fixed length keys
 982  * struct test_obj {
 983  *      int                     key;
 984  *      void *                  my_member;
 985  *      struct rhash_head       node;
 986  * };
 987  *
 988  * struct rhashtable_params params = {
 989  *      .head_offset = offsetof(struct test_obj, node),
 990  *      .key_offset = offsetof(struct test_obj, key),
 991  *      .key_len = sizeof(int),
 992  *      .hashfn = jhash,
 993  * };
 994  *
 995  * Configuration Example 2: Variable length keys
 996  * struct test_obj {
 997  *      [...]
 998  *      struct rhash_head       node;
 999  * };
1000  *
1001  * u32 my_hash_fn(const void *data, u32 len, u32 seed)
1002  * {
1003  *      struct test_obj *obj = data;
1004  *
1005  *      return [... hash ...];
1006  * }
1007  *
1008  * struct rhashtable_params params = {
1009  *      .head_offset = offsetof(struct test_obj, node),
1010  *      .hashfn = jhash,
1011  *      .obj_hashfn = my_hash_fn,
1012  * };
1013  */
1014 int rhashtable_init(struct rhashtable *ht,
1015                     const struct rhashtable_params *params)
1016 {
1017         struct bucket_table *tbl;
1018         size_t size;
1019
1020         if ((!params->key_len && !params->obj_hashfn) ||
1021             (params->obj_hashfn && !params->obj_cmpfn))
1022                 return -EINVAL;
1023
1024         memset(ht, 0, sizeof(*ht));
1025         mutex_init(&ht->mutex);
1026         spin_lock_init(&ht->lock);
1027         memcpy(&ht->p, params, sizeof(*params));
1028
1029         if (params->min_size)
1030                 ht->p.min_size = roundup_pow_of_two(params->min_size);
1031
1032         /* Cap total entries at 2^31 to avoid nelems overflow. */
1033         ht->max_elems = 1u << 31;
1034
1035         if (params->max_size) {
1036                 ht->p.max_size = rounddown_pow_of_two(params->max_size);
1037                 if (ht->p.max_size < ht->max_elems / 2)
1038                         ht->max_elems = ht->p.max_size * 2;
1039         }
1040
1041         ht->p.min_size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
1042
1043         size = rounded_hashtable_size(&ht->p);
1044
1045         if (params->locks_mul)
1046                 ht->p.locks_mul = roundup_pow_of_two(params->locks_mul);
1047         else
1048                 ht->p.locks_mul = BUCKET_LOCKS_PER_CPU;
1049
1050         ht->key_len = ht->p.key_len;
1051         if (!params->hashfn) {
1052                 ht->p.hashfn = jhash;
1053
1054                 if (!(ht->key_len & (sizeof(u32) - 1))) {
1055                         ht->key_len /= sizeof(u32);
1056                         ht->p.hashfn = rhashtable_jhash2;
1057                 }
1058         }
1059
1060         /*
1061          * This is api initialization and thus we need to guarantee the
1062          * initial rhashtable allocation. Upon failure, retry with the
1063          * smallest possible size with __GFP_NOFAIL semantics.
1064          */
1065         tbl = bucket_table_alloc(ht, size, GFP_KERNEL);
1066         if (unlikely(tbl == NULL)) {
1067                 size = max_t(u16, ht->p.min_size, HASH_MIN_SIZE);
1068                 tbl = bucket_table_alloc(ht, size, GFP_KERNEL | __GFP_NOFAIL);
1069         }
1070
1071         atomic_set(&ht->nelems, 0);
1072
1073         RCU_INIT_POINTER(ht->tbl, tbl);
1074
1075         INIT_WORK(&ht->run_work, rht_deferred_worker);
1076
1077         return 0;
1078 }
1079 EXPORT_SYMBOL_GPL(rhashtable_init);
1080
1081 /**
1082  * rhltable_init - initialize a new hash list table
1083  * @hlt:        hash list table to be initialized
1084  * @params:     configuration parameters
1085  *
1086  * Initializes a new hash list table.
1087  *
1088  * See documentation for rhashtable_init.
1089  */
1090 int rhltable_init(struct rhltable *hlt, const struct rhashtable_params *params)
1091 {
1092         int err;
1093
1094         err = rhashtable_init(&hlt->ht, params);
1095         hlt->ht.rhlist = true;
1096         return err;
1097 }
1098 EXPORT_SYMBOL_GPL(rhltable_init);
1099
1100 static void rhashtable_free_one(struct rhashtable *ht, struct rhash_head *obj,
1101                                 void (*free_fn)(void *ptr, void *arg),
1102                                 void *arg)
1103 {
1104         struct rhlist_head *list;
1105
1106         if (!ht->rhlist) {
1107                 free_fn(rht_obj(ht, obj), arg);
1108                 return;
1109         }
1110
1111         list = container_of(obj, struct rhlist_head, rhead);
1112         do {
1113                 obj = &list->rhead;
1114                 list = rht_dereference(list->next, ht);
1115                 free_fn(rht_obj(ht, obj), arg);
1116         } while (list);
1117 }
1118
1119 /**
1120  * rhashtable_free_and_destroy - free elements and destroy hash table
1121  * @ht:         the hash table to destroy
1122  * @free_fn:    callback to release resources of element
1123  * @arg:        pointer passed to free_fn
1124  *
1125  * Stops an eventual async resize. If defined, invokes free_fn for each
1126  * element to releasal resources. Please note that RCU protected
1127  * readers may still be accessing the elements. Releasing of resources
1128  * must occur in a compatible manner. Then frees the bucket array.
1129  *
1130  * This function will eventually sleep to wait for an async resize
1131  * to complete. The caller is responsible that no further write operations
1132  * occurs in parallel.
1133  */
1134 void rhashtable_free_and_destroy(struct rhashtable *ht,
1135                                  void (*free_fn)(void *ptr, void *arg),
1136                                  void *arg)
1137 {
1138         struct bucket_table *tbl, *next_tbl;
1139         unsigned int i;
1140
1141         cancel_work_sync(&ht->run_work);
1142
1143         mutex_lock(&ht->mutex);
1144         tbl = rht_dereference(ht->tbl, ht);
1145 restart:
1146         if (free_fn) {
1147                 for (i = 0; i < tbl->size; i++) {
1148                         struct rhash_head *pos, *next;
1149
1150                         cond_resched();
1151                         for (pos = rht_dereference(*rht_bucket(tbl, i), ht),
1152                              next = !rht_is_a_nulls(pos) ?
1153                                         rht_dereference(pos->next, ht) : NULL;
1154                              !rht_is_a_nulls(pos);
1155                              pos = next,
1156                              next = !rht_is_a_nulls(pos) ?
1157                                         rht_dereference(pos->next, ht) : NULL)
1158                                 rhashtable_free_one(ht, pos, free_fn, arg);
1159                 }
1160         }
1161
1162         next_tbl = rht_dereference(tbl->future_tbl, ht);
1163         bucket_table_free(tbl);
1164         if (next_tbl) {
1165                 tbl = next_tbl;
1166                 goto restart;
1167         }
1168         mutex_unlock(&ht->mutex);
1169 }
1170 EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
1171
1172 void rhashtable_destroy(struct rhashtable *ht)
1173 {
1174         return rhashtable_free_and_destroy(ht, NULL, NULL);
1175 }
1176 EXPORT_SYMBOL_GPL(rhashtable_destroy);
1177
1178 struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
1179                                             unsigned int hash)
1180 {
1181         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1182         static struct rhash_head __rcu *rhnull;
1183         unsigned int index = hash & ((1 << tbl->nest) - 1);
1184         unsigned int size = tbl->size >> tbl->nest;
1185         unsigned int subhash = hash;
1186         union nested_table *ntbl;
1187
1188         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1189         ntbl = rht_dereference_bucket_rcu(ntbl[index].table, tbl, hash);
1190         subhash >>= tbl->nest;
1191
1192         while (ntbl && size > (1 << shift)) {
1193                 index = subhash & ((1 << shift) - 1);
1194                 ntbl = rht_dereference_bucket_rcu(ntbl[index].table,
1195                                                   tbl, hash);
1196                 size >>= shift;
1197                 subhash >>= shift;
1198         }
1199
1200         if (!ntbl) {
1201                 if (!rhnull)
1202                         INIT_RHT_NULLS_HEAD(rhnull);
1203                 return &rhnull;
1204         }
1205
1206         return &ntbl[subhash].bucket;
1207
1208 }
1209 EXPORT_SYMBOL_GPL(rht_bucket_nested);
1210
1211 struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
1212                                                    struct bucket_table *tbl,
1213                                                    unsigned int hash)
1214 {
1215         const unsigned int shift = PAGE_SHIFT - ilog2(sizeof(void *));
1216         unsigned int index = hash & ((1 << tbl->nest) - 1);
1217         unsigned int size = tbl->size >> tbl->nest;
1218         union nested_table *ntbl;
1219
1220         ntbl = (union nested_table *)rcu_dereference_raw(tbl->buckets[0]);
1221         hash >>= tbl->nest;
1222         ntbl = nested_table_alloc(ht, &ntbl[index].table,
1223                                   size <= (1 << shift));
1224
1225         while (ntbl && size > (1 << shift)) {
1226                 index = hash & ((1 << shift) - 1);
1227                 size >>= shift;
1228                 hash >>= shift;
1229                 ntbl = nested_table_alloc(ht, &ntbl[index].table,
1230                                           size <= (1 << shift));
1231         }
1232
1233         if (!ntbl)
1234                 return NULL;
1235
1236         return &ntbl[hash].bucket;
1237
1238 }
1239 EXPORT_SYMBOL_GPL(rht_bucket_nested_insert);