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 static
 208 int _ht_add(struct rcu_ht *ht, struct rcu_table *t, struct rcu_ht_node *node,
 209              int unique)
 210 {
 211         struct rcu_ht_node *iter_prev, *iter, *iter_prev_next, *next;
 212
 213         if (!t->size)
 214                 return 0;
 215         for (;;) {
 216                 uint32_t chain_len = 0;
 217
 218                 /*
 219                  * iter_prev points to the non-removed node prior to the
 220                  * insert location.
 221                  * iter iterates until it finds the next non-removed
 222                  * node.
 223                  */
 224                 iter_prev = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 225                 /* We can always skip the dummy node initially */
 226                 iter_prev_next = next = rcu_dereference(iter_prev->next);
 227                 assert(iter_prev);
 228                 assert(iter_prev->reverse_hash <= node->reverse_hash);
 229                 for (;;) {
 230                         iter = next;
 231                         if (unlikely(!clear_flag(iter)))
 232                                 break;
 233                         next = rcu_dereference(clear_flag(iter)->next);
 234                         if (unlikely(is_removed(next)))
 235                                 continue;
 236                         if (unique
 237                             && !clear_flag(iter)->dummy
 238                             && !ht->compare_fct(node->key, node->key_len,
 239                                         clear_flag(iter)->key,
 240                                         clear_flag(iter)->key_len))
 241                                 return -EEXIST;
 242                         if (clear_flag(iter)->reverse_hash > node->reverse_hash)
 243                                 break;
 244                         /* Only account for identical reverse hash once */
 245                         if (iter_prev->reverse_hash != clear_flag(iter)->reverse_hash)
 246                                 check_resize(ht, t, ++chain_len);
 247                         iter_prev = clear_flag(iter);
 248                         iter_prev_next = next;
 249                 }
 250                 assert(node != clear_flag(iter));
 251                 assert(!is_removed(iter_prev));
 252                 assert(iter_prev != node);
 253                 node->next = iter;
 254                 if (uatomic_cmpxchg(&iter_prev->next, iter_prev_next,
 255                                     node) != iter_prev_next)
 256                         continue;
 257                 else
 258                         break;
 259         }
 260         return 0;
 261 }
 262
 263 static
 264 int _ht_remove(struct rcu_ht *ht, struct rcu_table *t, struct rcu_ht_node *node)
 265 {
 266         struct rcu_ht_node *iter_prev, *iter, *iter_prev_next, *next, *old;
 267         int found;
 268         int flagged = 0;
 269
 270         /* logically delete the node */
 271         old = rcu_dereference(node->next);
 272         do {
 273                 next = old;
 274                 if (is_removed(next))
 275                         goto end;
 276                 old = uatomic_cmpxchg(&node->next, next,
 277                                       flag_removed(next));
 278         } while (old != next);
 279
 280         /* We performed the (logical) deletion. */
 281         flagged = 1;
 282
 283         /*
 284          * Ensure that the node is not visible to readers anymore: lookup for
 285          * the node, and remove it if found.
 286          */
 287 retry:
 288         found = 0;
 289         /*
 290          * iter_prev points to the non-removed node prior to the remove
 291          * location.
 292          * next points to the non-removed node following the remove location.
 293          */
 294         iter_prev = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 295         /* We can always skip the dummy node initially */
 296         iter_prev_next = next = rcu_dereference(iter_prev->next);
 297         assert(iter_prev);
 298         assert(iter_prev->reverse_hash <= node->reverse_hash);
 299         for (;;) {
 300                 iter = next;
 301                 if (unlikely(!clear_flag(iter)))
 302                         break;
 303                 next = rcu_dereference(clear_flag(iter)->next);
 304                 if (clear_flag(iter) == node) {
 305                         found = 1;
 306                         assert(is_removed(rcu_dereference(node->next)));
 307                         break;
 308                 }
 309                 if (unlikely(is_removed(next)))
 310                         continue;
 311                 if (clear_flag(iter)->reverse_hash > node->reverse_hash)
 312                         break;
 313                 iter_prev = clear_flag(iter);
 314                 iter_prev_next = next;
 315         }
 316         if (!found)
 317                 goto end;
 318         if (uatomic_cmpxchg(&iter_prev->next, iter_prev_next,
 319                             clear_flag(next)) != iter_prev_next)
 320                 goto retry;
 321 end:
 322         /*
 323          * Only the flagging action indicated that we (and no other)
 324          * removed the node from the hash.
 325          */
 326         if (flagged) {
 327                 assert(is_removed(rcu_dereference(node->next)));
 328                 return 0;
 329         } else
 330                 return -ENOENT;
 331 }
 332
 333 static
 334 void init_table(struct rcu_ht *ht, struct rcu_table *t,
 335                 unsigned long first, unsigned long len)
 336 {
 337         unsigned long i, end;
 338
 339         end = first + len;
 340         for (i = first; i < end; i++) {
 341                 /* Update table size when power of two */
 342                 if (i != 0 && !(i & (i - 1)))
 343                         t->size = i;
 344                 t->tbl[i] = calloc(1, sizeof(struct rcu_ht_node));
 345                 t->tbl[i]->dummy = 1;
 346                 t->tbl[i]->hash = i;
 347                 t->tbl[i]->reverse_hash = bit_reverse_ulong(i);
 348                 (void) _ht_add(ht, t, t->tbl[i], 0);
 349         }
 350         t->resize_target = t->size = end;
 351         t->resize_initiated = 0;
 352 }
 353
 354 struct rcu_ht *ht_new(ht_hash_fct hash_fct,
 355                       ht_compare_fct compare_fct,
 356                       unsigned long hash_seed,
 357                       unsigned long init_size,
 358                       void (*ht_call_rcu)(struct rcu_head *head,
 359                                 void (*func)(struct rcu_head *head)))
 360 {
 361         struct rcu_ht *ht;
 362
 363         ht = calloc(1, sizeof(struct rcu_ht));
 364         ht->hash_fct = hash_fct;
 365         ht->compare_fct = compare_fct;
 366         ht->hash_seed = hash_seed;
 367         ht->ht_call_rcu = ht_call_rcu;
 368         /* this mutex should not nest in read-side C.S. */
 369         pthread_mutex_init(&ht->resize_mutex, NULL);
 370         ht->t = calloc(1, sizeof(struct rcu_table)
 371                        + (max(init_size, 1) * sizeof(struct rcu_ht_node *)));
 372         ht->t->size = 0;
 373         pthread_mutex_lock(&ht->resize_mutex);
 374         init_table(ht, ht->t, 0, max(init_size, 1));
 375         pthread_mutex_unlock(&ht->resize_mutex);
 376         return ht;
 377 }
 378
 379 struct rcu_ht_node *ht_lookup(struct rcu_ht *ht, void *key, size_t key_len)
 380 {
 381         struct rcu_table *t;
 382         struct rcu_ht_node *node;
 383         unsigned long hash, reverse_hash;
 384
 385         hash = ht->hash_fct(key, key_len, ht->hash_seed);
 386         reverse_hash = bit_reverse_ulong(hash);
 387
 388         t = rcu_dereference(ht->t);
 389         node = rcu_dereference(t->tbl[hash & (t->size - 1)]);
 390         for (;;) {
 391                 if (unlikely(!node))
 392                         break;
 393                 if (unlikely(node->reverse_hash > reverse_hash)) {
 394                         node = NULL;
 395                         break;
 396                 }
 397                 if (!ht->compare_fct(node->key, node->key_len, key, key_len)) {
 398                         if (unlikely(is_removed(rcu_dereference(node->next))))
 399                                 node = NULL;
 400                         break;
 401                 }
 402                 node = clear_flag(rcu_dereference(node->next));
 403         }
 404         return node;
 405 }
 406
 407 void ht_add(struct rcu_ht *ht, struct rcu_ht_node *node)
 408 {
 409         struct rcu_table *t;
 410
 411         node->hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 412         node->reverse_hash = bit_reverse_ulong((unsigned long) node->hash);
 413
 414         t = rcu_dereference(ht->t);
 415         (void) _ht_add(ht, t, node, 0);
 416 }
 417
 418 int ht_add_unique(struct rcu_ht *ht, struct rcu_ht_node *node)
 419 {
 420         struct rcu_table *t;
 421
 422         node->hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 423         node->reverse_hash = bit_reverse_ulong((unsigned long) node->hash);
 424
 425         t = rcu_dereference(ht->t);
 426         return _ht_add(ht, t, node, 1);
 427 }
 428
 429 int ht_remove(struct rcu_ht *ht, struct rcu_ht_node *node)
 430 {
 431         struct rcu_table *t;
 432
 433         t = rcu_dereference(ht->t);
 434         return _ht_remove(ht, t, node);
 435 }
 436
 437 static
 438 int ht_delete_dummy(struct rcu_ht *ht)
 439 {
 440         struct rcu_table *t;
 441         struct rcu_ht_node *node;
 442         unsigned long i;
 443
 444         t = ht->t;
 445         /* Check that the table is empty */
 446         node = t->tbl[0];
 447         do {
 448                 if (!node->dummy)
 449                         return -EPERM;
 450                 node = node->next;
 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 static
 477 void ht_free_table_cb(struct rcu_head *head)
 478 {
 479         struct rcu_table *t =
 480                 caa_container_of(head, struct rcu_table, head);
 481         free(t);
 482 }
 483
 484 /* called with resize mutex held */
 485 static
 486 void _do_ht_resize(struct rcu_ht *ht)
 487 {
 488         unsigned long new_size, old_size;
 489         struct rcu_table *new_t, *old_t;
 490
 491         old_t = ht->t;
 492         old_size = old_t->size;
 493
 494         new_size = CMM_LOAD_SHARED(old_t->resize_target);
 495         dbg_printf("rculfhash: resize from %lu to %lu buckets\n",
 496                    old_size, new_size);
 497         if (old_size == new_size)
 498                 return;
 499         new_t = malloc(sizeof(struct rcu_table)
 500                         + (new_size * sizeof(struct rcu_ht_node *)));
 501         assert(new_size > old_size);
 502         memcpy(&new_t->tbl, &old_t->tbl,
 503                old_size * sizeof(struct rcu_ht_node *));
 504         init_table(ht, new_t, old_size, new_size - old_size);
 505         /* Changing table and size atomically wrt lookups */
 506         rcu_assign_pointer(ht->t, new_t);
 507         ht->ht_call_rcu(&old_t->head, ht_free_table_cb);
 508 }
 509
 510 static
 511 unsigned long resize_target_update(struct rcu_table *t,
 512                                    int growth_order)
 513 {
 514         return _uatomic_max(&t->resize_target,
 515                             t->size << growth_order);
 516 }
 517
 518 void ht_resize(struct rcu_ht *ht, int growth)
 519 {
 520         struct rcu_table *t = rcu_dereference(ht->t);
 521         unsigned long target_size;
 522
 523         target_size = resize_target_update(t, growth);
 524         if (t->size < target_size) {
 525                 CMM_STORE_SHARED(t->resize_initiated, 1);
 526                 pthread_mutex_lock(&ht->resize_mutex);
 527                 _do_ht_resize(ht);
 528                 pthread_mutex_unlock(&ht->resize_mutex);
 529         }
 530 }
 531
 532 static
 533 void do_resize_cb(struct rcu_head *head)
 534 {
 535         struct rcu_resize_work *work =
 536                 caa_container_of(head, struct rcu_resize_work, head);
 537         struct rcu_ht *ht = work->ht;
 538
 539         pthread_mutex_lock(&ht->resize_mutex);
 540         _do_ht_resize(ht);
 541         pthread_mutex_unlock(&ht->resize_mutex);
 542         free(work);
 543 }
 544
 545 static
 546 void ht_resize_lazy(struct rcu_ht *ht, struct rcu_table *t, int growth)
 547 {
 548         struct rcu_resize_work *work;
 549         unsigned long target_size;
 550
 551         target_size = resize_target_update(t, growth);
 552         if (!CMM_LOAD_SHARED(t->resize_initiated) && t->size < target_size) {
 553                 work = malloc(sizeof(*work));
 554                 work->ht = ht;
 555                 ht->ht_call_rcu(&work->head, do_resize_cb);
 556                 CMM_STORE_SHARED(t->resize_initiated, 1);
 557         }
 558 }