rculfhash.c

   1 /*
   2  * rculfhash.c
   3  *
   4  * Userspace RCU library - Lock-Free Expandable RCU Hash Table
   5  *
   6  * Copyright 2010-2011 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  *
   8  * This library is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * This library is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with this library; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #define _LGPL_SOURCE
  24 #include <stdlib.h>
  25 #include <errno.h>
  26 #include <assert.h>
  27 #include <stdio.h>
  28 #include <stdint.h>
  29 #include <string.h>
  30
  31 #include <urcu.h>
  32 #include <urcu-call-rcu.h>
  33 #include <urcu/arch.h>
  34 #include <urcu/uatomic.h>
  35 #include <urcu/jhash.h>
  36 #include <urcu/compiler.h>
  37 #include <urcu/rculfhash.h>
  38 #include <stdio.h>
  39 #include <pthread.h>
  40
  41 #define DEBUG           /* Test */
  42
  43 #ifdef DEBUG
  44 #define dbg_printf(args...)     printf(args)
  45 #else
  46 #define dbg_printf(args...)
  47 #endif
  48
  49 #define CHAIN_LEN_TARGET                1
  50 #define CHAIN_LEN_RESIZE_THRESHOLD      2
  51
  52 #ifndef max
  53 #define max(a, b)       ((a) > (b) ? (a) : (b))
  54 #endif
  55
  56 struct rcu_table {
  57         unsigned long size;     /* always a power of 2 */
  58         unsigned long resize_target;
  59         int resize_initiated;
  60         struct rcu_head head;
  61         struct rcu_ht_node *tbl[0];
  62 };
  63
  64 struct rcu_ht {
  65         struct rcu_table *t;            /* shared */
  66         ht_hash_fct hash_fct;
  67         ht_compare_fct compare_fct;
  68         unsigned long hash_seed;
  69         pthread_mutex_t resize_mutex;   /* resize mutex: add/del mutex */
  70         void (*ht_call_rcu)(struct rcu_head *head,
  71                       void (*func)(struct rcu_head *head));
  72 };
  73
  74 struct rcu_resize_work {
  75         struct rcu_head head;
  76         struct rcu_ht *ht;
  77 };
  78
  79 /*
  80  * Algorithm to reverse bits in a word by lookup table, extended to
  81  * 64-bit words.
  82  * Source:
  83  * http://graphics.stanford.edu/~seander/bithacks.html#BitReverseTable
  84  * Originally from Public Domain.
  85  */
  86
  87 static const uint8_t BitReverseTable256[256] =
  88 {
  89 #define R2(n) (n),   (n) + 2*64,     (n) + 1*64,     (n) + 3*64
  90 #define R4(n) R2(n), R2((n) + 2*16), R2((n) + 1*16), R2((n) + 3*16)
  91 #define R6(n) R4(n), R4((n) + 2*4 ), R4((n) + 1*4 ), R4((n) + 3*4 )
  92         R6(0), R6(2), R6(1), R6(3)
  93 };
  94 #undef R2
  95 #undef R4
  96 #undef R6
  97
  98 static
  99 uint8_t bit_reverse_u8(uint8_t v)
 100 {
 101         return BitReverseTable256[v];
 102 }
 103
 104 static __attribute__((unused))
 105 uint32_t bit_reverse_u32(uint32_t v)
 106 {
 107         return ((uint32_t) bit_reverse_u8(v) << 24) |
 108                 ((uint32_t) bit_reverse_u8(v >> 8) << 16) |
 109                 ((uint32_t) bit_reverse_u8(v >> 16) << 8) |
 110                 ((uint32_t) bit_reverse_u8(v >> 24));
 111 }
 112
 113 static __attribute__((unused))
 114 uint64_t bit_reverse_u64(uint64_t v)
 115 {
 116         return ((uint64_t) bit_reverse_u8(v) << 56) |
 117                 ((uint64_t) bit_reverse_u8(v >> 8)  << 48) |
 118                 ((uint64_t) bit_reverse_u8(v >> 16) << 40) |
 119                 ((uint64_t) bit_reverse_u8(v >> 24) << 32) |
 120                 ((uint64_t) bit_reverse_u8(v >> 32) << 24) |
 121                 ((uint64_t) bit_reverse_u8(v >> 40) << 16) |
 122                 ((uint64_t) bit_reverse_u8(v >> 48) << 8) |
 123                 ((uint64_t) bit_reverse_u8(v >> 56));
 124 }
 125
 126 static
 127 unsigned long bit_reverse_ulong(unsigned long v)
 128 {
 129 #if (CAA_BITS_PER_LONG == 32)
 130         return bit_reverse_u32(v);
 131 #else
 132         return bit_reverse_u64(v);
 133 #endif
 134 }
 135
 136 /*
 137  * Algorithm to find the log2 of a 32-bit unsigned integer.
 138  * source: http://graphics.stanford.edu/~seander/bithacks.html#IntegerLogLookup
 139  * Originally from Public Domain.
 140  */
 141 static const char LogTable256[256] =
 142 {
 143 #define LT(n) n, n, n, n, n, n, n, n, n, n, n, n, n, n, n, n
 144         -1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
 145         LT(4), LT(5), LT(5), LT(6), LT(6), LT(6), LT(6),
 146         LT(7), LT(7), LT(7), LT(7), LT(7), LT(7), LT(7), LT(7)
 147 };
 148
 149 uint32_t log2_u32(uint32_t v)
 150 {
 151         uint32_t t, tt;
 152
 153         if ((tt = (v >> 16)))
 154                 return (t = (tt >> 8))
 155                                 ? 24 + LogTable256[t]
 156                                 : 16 + LogTable256[tt];
 157         else
 158                 return (t = (v >> 8))
 159                                 ? 8 + LogTable256[t]
 160                                 : LogTable256[v];
 161 }
 162
 163 static
 164 void ht_resize_lazy(struct rcu_ht *ht, struct rcu_table *t, int growth);
 165
 166 static
 167 void check_resize(struct rcu_ht *ht, struct rcu_table *t,
 168                   uint32_t chain_len)
 169 {
 170         if (chain_len >= CHAIN_LEN_RESIZE_THRESHOLD)
 171                 ht_resize_lazy(ht, t,
 172                         log2_u32(chain_len - CHAIN_LEN_TARGET - 1));
 173 }
 174
 175 static
 176 struct rcu_ht_node *clear_flag(struct rcu_ht_node *node)
 177 {
 178         return (struct rcu_ht_node *) (((unsigned long) node) & ~0x1);
 179 }
 180
 181 static
 182 int is_removed(struct rcu_ht_node *node)
 183 {
 184         return ((unsigned long) node) & 0x1;
 185 }
 186
 187 static
 188 struct rcu_ht_node *flag_removed(struct rcu_ht_node *node)
 189 {
 190         return (struct rcu_ht_node *) (((unsigned long) node) | 0x1);
 191 }
 192
 193 static
 194 unsigned long _uatomic_max(unsigned long *ptr, unsigned long v)
 195 {
 196         unsigned long old1, old2;
 197
 198         old1 = uatomic_read(ptr);
 199         do {
 200                 old2 = old1;
 201                 if (old2 >= v)
 202                         return old2;
 203         } while ((old1 = uatomic_cmpxchg(ptr, old2, v)) != old2);
 204         return v;
 205 }
 206
 207 /*
 208  * Remove all logically deleted nodes from a bucket up to a certain node key.
 209  */
 210 static
 211 void _ht_gc_bucket(struct rcu_ht_node *dummy, struct rcu_ht_node *node)
 212 {
 213         struct rcu_ht_node *iter_prev, *iter, *next;
 214
 215         for (;;) {
 216                 iter_prev = dummy;
 217                 /* We can always skip the dummy node initially */
 218                 iter = rcu_dereference(iter_prev->next);
 219                 assert(iter_prev->reverse_hash <= node->reverse_hash);
 220                 for (;;) {
 221                         if (unlikely(!iter))
 222                                 return;
 223                         if (clear_flag(iter)->reverse_hash > node->reverse_hash)
 224                                 return;
 225                         next = rcu_dereference(clear_flag(iter)->next);
 226                         if (is_removed(next))
 227                                 break;
 228                         iter_prev = iter;
 229                         iter = next;
 230                 }
 231                 assert(!is_removed(iter));
 232                 (void) uatomic_cmpxchg(&iter_prev->next, iter, clear_flag(next));
 233         }
 234 }
 235
 236 static
 237 int _ht_add(struct rcu_ht *ht, struct rcu_table *t, struct rcu_ht_node *node,
 238              int unique)
 239 {
 240         struct rcu_ht_node *iter_prev, *dummy, *iter, *next;
 241
 242         if (!t->size)
 243                 return 0;
 244         for (;;) {
 245                 uint32_t chain_len = 0;
 246
 247                 /*
 248                  * iter_prev points to the non-removed node prior to the
 249                  * insert location.
 250                  */
 251                 iter_prev = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 252                 /* We can always skip the dummy node initially */
 253                 iter = rcu_dereference(iter_prev->next);
 254                 assert(iter_prev->reverse_hash <= node->reverse_hash);
 255                 for (;;) {
 256                         if (unlikely(!iter))
 257                                 goto insert;
 258                         if (clear_flag(iter)->reverse_hash > node->reverse_hash)
 259                                 goto insert;
 260                         next = rcu_dereference(clear_flag(iter)->next);
 261                         if (is_removed(next))
 262                                 goto gc;
 263                         /* Only account for identical reverse hash once */
 264                         if (iter_prev->reverse_hash != clear_flag(iter)->reverse_hash)
 265                                 check_resize(ht, t, ++chain_len);
 266                         iter_prev = clear_flag(iter);
 267                         iter = next;
 268                 }
 269         insert:
 270                 assert(node != clear_flag(iter));
 271                 assert(!is_removed(iter_prev));
 272                 assert(iter_prev != node);
 273                 node->next = iter;
 274                 if (uatomic_cmpxchg(&iter_prev->next, iter,
 275                                     node) != iter)
 276                         continue;       /* retry */
 277                 else
 278                         goto gc_end;
 279         gc:
 280                 /* Garbage collect logically removed nodes in the bucket */
 281                 dummy = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 282                 _ht_gc_bucket(dummy, node);
 283                 /* retry */
 284         }
 285 gc_end:
 286         /* Garbage collect logically removed nodes in the bucket */
 287         dummy = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 288         _ht_gc_bucket(dummy, node);
 289         return 0;
 290 }
 291
 292 static
 293 int _ht_remove(struct rcu_ht *ht, struct rcu_table *t, struct rcu_ht_node *node)
 294 {
 295         struct rcu_ht_node *dummy, *next, *old;
 296         int flagged = 0;
 297
 298         /* logically delete the node */
 299         old = rcu_dereference(node->next);
 300         do {
 301                 next = old;
 302                 if (is_removed(next))
 303                         goto end;
 304                 assert(!node->dummy);
 305                 old = uatomic_cmpxchg(&node->next, next,
 306                                       flag_removed(next));
 307         } while (old != next);
 308
 309         /* We performed the (logical) deletion. */
 310         flagged = 1;
 311
 312         /*
 313          * Ensure that the node is not visible to readers anymore: lookup for
 314          * the node, and remove it (along with any other logically removed node)
 315          * if found.
 316          */
 317         dummy = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 318         _ht_gc_bucket(dummy, node);
 319 end:
 320         /*
 321          * Only the flagging action indicated that we (and no other)
 322          * removed the node from the hash.
 323          */
 324         if (flagged) {
 325                 assert(is_removed(rcu_dereference(node->next)));
 326                 return 0;
 327         } else
 328                 return -ENOENT;
 329 }
 330
 331 static
 332 void init_table(struct rcu_ht *ht, struct rcu_table *t,
 333                 unsigned long first, unsigned long len)
 334 {
 335         unsigned long i, end;
 336
 337         end = first + len;
 338         for (i = first; i < end; i++) {
 339                 /* Update table size when power of two */
 340                 if (i != 0 && !(i & (i - 1)))
 341                         t->size = i;
 342                 t->tbl[i] = calloc(1, sizeof(struct rcu_ht_node));
 343                 t->tbl[i]->dummy = 1;
 344                 t->tbl[i]->hash = i;
 345                 t->tbl[i]->reverse_hash = bit_reverse_ulong(i);
 346                 (void) _ht_add(ht, t, t->tbl[i], 0);
 347         }
 348         t->resize_target = t->size = end;
 349         t->resize_initiated = 0;
 350 }
 351
 352 struct rcu_ht *ht_new(ht_hash_fct hash_fct,
 353                       ht_compare_fct compare_fct,
 354                       unsigned long hash_seed,
 355                       unsigned long init_size,
 356                       void (*ht_call_rcu)(struct rcu_head *head,
 357                                 void (*func)(struct rcu_head *head)))
 358 {
 359         struct rcu_ht *ht;
 360
 361         ht = calloc(1, sizeof(struct rcu_ht));
 362         ht->hash_fct = hash_fct;
 363         ht->compare_fct = compare_fct;
 364         ht->hash_seed = hash_seed;
 365         ht->ht_call_rcu = ht_call_rcu;
 366         /* this mutex should not nest in read-side C.S. */
 367         pthread_mutex_init(&ht->resize_mutex, NULL);
 368         ht->t = calloc(1, sizeof(struct rcu_table)
 369                        + (max(init_size, 1) * sizeof(struct rcu_ht_node *)));
 370         ht->t->size = 0;
 371         pthread_mutex_lock(&ht->resize_mutex);
 372         init_table(ht, ht->t, 0, max(init_size, 1));
 373         pthread_mutex_unlock(&ht->resize_mutex);
 374         return ht;
 375 }
 376
 377 struct rcu_ht_node *ht_lookup(struct rcu_ht *ht, void *key, size_t key_len)
 378 {
 379         struct rcu_table *t;
 380         struct rcu_ht_node *node;
 381         unsigned long hash, reverse_hash;
 382
 383         hash = ht->hash_fct(key, key_len, ht->hash_seed);
 384         reverse_hash = bit_reverse_ulong(hash);
 385
 386         t = rcu_dereference(ht->t);
 387         node = rcu_dereference(t->tbl[hash & (t->size - 1)]);
 388         for (;;) {
 389                 if (unlikely(!node))
 390                         break;
 391                 if (unlikely(node->reverse_hash > reverse_hash)) {
 392                         node = NULL;
 393                         break;
 394                 }
 395                 if (!ht->compare_fct(node->key, node->key_len, key, key_len)) {
 396                         if (likely(!is_removed(rcu_dereference(node->next)))
 397                             && likely(!node->dummy))
 398                                 break;
 399                 }
 400                 node = clear_flag(rcu_dereference(node->next));
 401         }
 402         assert(!node || !node->dummy);
 403         return node;
 404 }
 405
 406 void ht_add(struct rcu_ht *ht, struct rcu_ht_node *node)
 407 {
 408         struct rcu_table *t;
 409
 410         node->hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 411         node->reverse_hash = bit_reverse_ulong((unsigned long) node->hash);
 412
 413         t = rcu_dereference(ht->t);
 414         (void) _ht_add(ht, t, node, 0);
 415 }
 416
 417 int ht_add_unique(struct rcu_ht *ht, struct rcu_ht_node *node)
 418 {
 419         struct rcu_table *t;
 420
 421         node->hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 422         node->reverse_hash = bit_reverse_ulong((unsigned long) node->hash);
 423
 424         t = rcu_dereference(ht->t);
 425         return _ht_add(ht, t, node, 1);
 426 }
 427
 428 int ht_remove(struct rcu_ht *ht, struct rcu_ht_node *node)
 429 {
 430         struct rcu_table *t;
 431
 432         t = rcu_dereference(ht->t);
 433         return _ht_remove(ht, t, node);
 434 }
 435
 436 static
 437 int ht_delete_dummy(struct rcu_ht *ht)
 438 {
 439         struct rcu_table *t;
 440         struct rcu_ht_node *node;
 441         unsigned long i;
 442
 443         t = ht->t;
 444         /* Check that the table is empty */
 445         node = t->tbl[0];
 446         do {
 447                 if (!node->dummy)
 448                         return -EPERM;
 449                 node = node->next;
 450                 assert(!is_removed(node));
 451         } while (node);
 452         /* Internal sanity check: all nodes left should be dummy */
 453         for (i = 0; i < t->size; i++) {
 454                 assert(t->tbl[i]->dummy);
 455                 free(t->tbl[i]);
 456         }
 457         return 0;
 458 }
 459
 460 /*
 461  * Should only be called when no more concurrent readers nor writers can
 462  * possibly access the table.
 463  */
 464 int ht_destroy(struct rcu_ht *ht)
 465 {
 466         int ret;
 467
 468         ret = ht_delete_dummy(ht);
 469         if (ret)
 470                 return ret;
 471         free(ht->t);
 472         free(ht);
 473         return ret;
 474 }
 475
 476 void ht_count_nodes(struct rcu_ht *ht,
 477                 unsigned long *count,
 478                 unsigned long *removed)
 479 {
 480         struct rcu_table *t;
 481         struct rcu_ht_node *node, *next;
 482
 483         *count = 0;
 484         *removed = 0;
 485
 486         t = rcu_dereference(ht->t);
 487         /* Check that the table is empty */
 488         node = rcu_dereference(t->tbl[0]);
 489         do {
 490                 next = rcu_dereference(node->next);
 491                 if (is_removed(next)) {
 492                         assert(!node->dummy);
 493                         (*removed)++;
 494                 } else if (!node->dummy)
 495                         (*count)++;
 496                 node = clear_flag(next);
 497         } while (node);
 498 }
 499
 500 static
 501 void ht_free_table_cb(struct rcu_head *head)
 502 {
 503         struct rcu_table *t =
 504                 caa_container_of(head, struct rcu_table, head);
 505         free(t);
 506 }
 507
 508 /* called with resize mutex held */
 509 static
 510 void _do_ht_resize(struct rcu_ht *ht)
 511 {
 512         unsigned long new_size, old_size;
 513         struct rcu_table *new_t, *old_t;
 514
 515         old_t = ht->t;
 516         old_size = old_t->size;
 517
 518         new_size = CMM_LOAD_SHARED(old_t->resize_target);
 519         dbg_printf("rculfhash: resize from %lu to %lu buckets\n",
 520                    old_size, new_size);
 521         if (old_size == new_size)
 522                 return;
 523         new_t = malloc(sizeof(struct rcu_table)
 524                         + (new_size * sizeof(struct rcu_ht_node *)));
 525         assert(new_size > old_size);
 526         memcpy(&new_t->tbl, &old_t->tbl,
 527                old_size * sizeof(struct rcu_ht_node *));
 528         init_table(ht, new_t, old_size, new_size - old_size);
 529         /* Changing table and size atomically wrt lookups */
 530         rcu_assign_pointer(ht->t, new_t);
 531         ht->ht_call_rcu(&old_t->head, ht_free_table_cb);
 532 }
 533
 534 static
 535 unsigned long resize_target_update(struct rcu_table *t,
 536                                    int growth_order)
 537 {
 538         return _uatomic_max(&t->resize_target,
 539                             t->size << growth_order);
 540 }
 541
 542 void ht_resize(struct rcu_ht *ht, int growth)
 543 {
 544         struct rcu_table *t = rcu_dereference(ht->t);
 545         unsigned long target_size;
 546
 547         target_size = resize_target_update(t, growth);
 548         if (t->size < target_size) {
 549                 CMM_STORE_SHARED(t->resize_initiated, 1);
 550                 pthread_mutex_lock(&ht->resize_mutex);
 551                 _do_ht_resize(ht);
 552                 pthread_mutex_unlock(&ht->resize_mutex);
 553         }
 554 }
 555
 556 static
 557 void do_resize_cb(struct rcu_head *head)
 558 {
 559         struct rcu_resize_work *work =
 560                 caa_container_of(head, struct rcu_resize_work, head);
 561         struct rcu_ht *ht = work->ht;
 562
 563         pthread_mutex_lock(&ht->resize_mutex);
 564         _do_ht_resize(ht);
 565         pthread_mutex_unlock(&ht->resize_mutex);
 566         free(work);
 567 }
 568
 569 static
 570 void ht_resize_lazy(struct rcu_ht *ht, struct rcu_table *t, int growth)
 571 {
 572         struct rcu_resize_work *work;
 573         unsigned long target_size;
 574
 575         target_size = resize_target_update(t, growth);
 576         if (!CMM_LOAD_SHARED(t->resize_initiated) && t->size < target_size) {
 577                 work = malloc(sizeof(*work));
 578                 work->ht = ht;
 579                 ht->ht_call_rcu(&work->head, do_resize_cb);
 580                 CMM_STORE_SHARED(t->resize_initiated, 1);
 581         }
 582 }