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                4
  50 #define CHAIN_LEN_RESIZE_THRESHOLD      16
  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         struct rcu_head head;
  60         struct rcu_ht_node *tbl[0];
  61 };
  62
  63 struct rcu_ht {
  64         struct rcu_table *t;            /* shared */
  65         ht_hash_fct hash_fct;
  66         ht_compare_fct compare_fct;
  67         unsigned long hash_seed;
  68         pthread_mutex_t resize_mutex;   /* resize mutex: add/del mutex */
  69         void (*ht_call_rcu)(struct rcu_head *head,
  70                       void (*func)(struct rcu_head *head));
  71 };
  72
  73 struct rcu_resize_work {
  74         struct rcu_head head;
  75         struct rcu_ht *ht;
  76 };
  77
  78 /*
  79  * Algorithm to reverse bits in a word by lookup table, extended to
  80  * 64-bit words.
  81  * Source:
  82  * http://graphics.stanford.edu/~seander/bithacks.html#BitReverseTable
  83  * Originally from Public Domain.
  84  */
  85
  86 static const uint8_t BitReverseTable256[256] =
  87 {
  88 #define R2(n) (n),   (n) + 2*64,     (n) + 1*64,     (n) + 3*64
  89 #define R4(n) R2(n), R2((n) + 2*16), R2((n) + 1*16), R2((n) + 3*16)
  90 #define R6(n) R4(n), R4((n) + 2*4 ), R4((n) + 1*4 ), R4((n) + 3*4 )
  91         R6(0), R6(2), R6(1), R6(3)
  92 };
  93 #undef R2
  94 #undef R4
  95 #undef R6
  96
  97 static
  98 uint8_t bit_reverse_u8(uint8_t v)
  99 {
 100         return BitReverseTable256[v];
 101 }
 102
 103 static __attribute__((unused))
 104 uint32_t bit_reverse_u32(uint32_t v)
 105 {
 106         return ((uint32_t) bit_reverse_u8(v) << 24) |
 107                 ((uint32_t) bit_reverse_u8(v >> 8) << 16) |
 108                 ((uint32_t) bit_reverse_u8(v >> 16) << 8) |
 109                 ((uint32_t) bit_reverse_u8(v >> 24));
 110 }
 111
 112 static __attribute__((unused))
 113 uint64_t bit_reverse_u64(uint64_t v)
 114 {
 115         return ((uint64_t) bit_reverse_u8(v) << 56) |
 116                 ((uint64_t) bit_reverse_u8(v >> 8)  << 48) |
 117                 ((uint64_t) bit_reverse_u8(v >> 16) << 40) |
 118                 ((uint64_t) bit_reverse_u8(v >> 24) << 32) |
 119                 ((uint64_t) bit_reverse_u8(v >> 32) << 24) |
 120                 ((uint64_t) bit_reverse_u8(v >> 40) << 16) |
 121                 ((uint64_t) bit_reverse_u8(v >> 48) << 8) |
 122                 ((uint64_t) bit_reverse_u8(v >> 56));
 123 }
 124
 125 static
 126 unsigned long bit_reverse_ulong(unsigned long v)
 127 {
 128 #if (CAA_BITS_PER_LONG == 32)
 129         return bit_reverse_u32(v);
 130 #else
 131         return bit_reverse_u64(v);
 132 #endif
 133 }
 134
 135 /*
 136  * Algorithm to find the log2 of a 32-bit unsigned integer.
 137  * source: http://graphics.stanford.edu/~seander/bithacks.html#IntegerLogLookup
 138  * Originally from Public Domain.
 139  */
 140 static const char LogTable256[256] =
 141 {
 142 #define LT(n) n, n, n, n, n, n, n, n, n, n, n, n, n, n, n, n
 143         -1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
 144         LT(4), LT(5), LT(5), LT(6), LT(6), LT(6), LT(6),
 145         LT(7), LT(7), LT(7), LT(7), LT(7), LT(7), LT(7), LT(7)
 146 };
 147
 148 uint32_t log2_u32(uint32_t v)
 149 {
 150         uint32_t t, tt;
 151
 152         if ((tt = (v >> 16)))
 153                 return (t = (tt >> 8))
 154                                 ? 24 + LogTable256[t]
 155                                 : 16 + LogTable256[tt];
 156         else
 157                 return (t = (v >> 8))
 158                                 ? 8 + LogTable256[t]
 159                                 : LogTable256[v];
 160 }
 161
 162 static
 163 void ht_resize_lazy(struct rcu_ht *ht, struct rcu_table *t, int growth);
 164
 165 static
 166 void check_resize(struct rcu_ht *ht, struct rcu_table *t,
 167                   uint32_t chain_len)
 168 {
 169         if (chain_len >= CHAIN_LEN_RESIZE_THRESHOLD)
 170                 ht_resize_lazy(ht, t,
 171                         log2_u32(chain_len - CHAIN_LEN_TARGET));
 172 }
 173
 174 static
 175 struct rcu_ht_node *clear_flag(struct rcu_ht_node *node)
 176 {
 177         return (struct rcu_ht_node *) (((unsigned long) node) & ~0x1);
 178 }
 179
 180 static
 181 int is_removed(struct rcu_ht_node *node)
 182 {
 183         return ((unsigned long) node) & 0x1;
 184 }
 185
 186 static
 187 struct rcu_ht_node *flag_removed(struct rcu_ht_node *node)
 188 {
 189         return (struct rcu_ht_node *) (((unsigned long) node) | 0x1);
 190 }
 191
 192 static
 193 unsigned long _uatomic_max(unsigned long *ptr, unsigned long v)
 194 {
 195         unsigned long old1, old2;
 196
 197         old1 = uatomic_read(ptr);
 198         do {
 199                 old2 = old1;
 200                 if (old2 >= v)
 201                         return old2;
 202         } while ((old1 = uatomic_cmpxchg(ptr, old2, v)) != old2);
 203         return v;
 204 }
 205
 206 static
 207 void _ht_add(struct rcu_ht *ht, struct rcu_table *t, struct rcu_ht_node *node)
 208 {
 209         struct rcu_ht_node *iter_prev = NULL, *iter = NULL;
 210
 211         if (!t->size)
 212                 return;
 213         for (;;) {
 214                 uint32_t chain_len = 0;
 215
 216                 iter_prev = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 217                 assert(iter_prev);
 218                 assert(iter_prev->reverse_hash <= node->reverse_hash);
 219                 for (;;) {
 220                         iter = clear_flag(rcu_dereference(iter_prev->next));
 221                         if (unlikely(!iter))
 222                                 break;
 223                         if (iter->reverse_hash > node->reverse_hash)
 224                                 break;
 225                         iter_prev = iter;
 226                         check_resize(ht, t, ++chain_len);
 227                 }
 228                 /* add in iter_prev->next */
 229                 if (is_removed(iter))
 230                         continue;
 231                 assert(node != iter);
 232                 node->next = iter;
 233                 assert(iter_prev != node);
 234                 if (uatomic_cmpxchg(&iter_prev->next, iter, node) != iter)
 235                         continue;
 236                 break;
 237         }
 238 }
 239
 240 static
 241 int _ht_remove(struct rcu_ht *ht, struct rcu_table *t, struct rcu_ht_node *node)
 242 {
 243         struct rcu_ht_node *iter_prev, *iter, *next, *old;
 244         unsigned long chain_len;
 245         int found, ret = 0;
 246         int flagged = 0;
 247
 248 retry:
 249         chain_len = 0;
 250         found = 0;
 251         iter_prev = rcu_dereference(t->tbl[node->hash & (t->size - 1)]);
 252         assert(iter_prev);
 253         assert(iter_prev->reverse_hash <= node->reverse_hash);
 254         for (;;) {
 255                 iter = clear_flag(rcu_dereference(iter_prev->next));
 256                 if (unlikely(!iter))
 257                         break;
 258                 if (unlikely(iter->reverse_hash > node->reverse_hash))
 259                         break;
 260                 if (iter == node) {
 261                         found = 1;
 262                         break;
 263                 }
 264                 iter_prev = iter;
 265         }
 266         if (!found) {
 267                 ret = -ENOENT;
 268                 goto end;
 269         }
 270         next = rcu_dereference(iter->next);
 271         if (!flagged) {
 272                 if (is_removed(next)) {
 273                         ret = -ENOENT;
 274                         goto end;
 275                 }
 276                 /* set deletion flag */
 277                 if ((old = uatomic_cmpxchg(&iter->next, next,
 278                                            flag_removed(next))) != next) {
 279                         if (old == flag_removed(next)) {
 280                                 ret = -ENOENT;
 281                                 goto end;
 282                         } else {
 283                                 goto retry;
 284                         }
 285                 }
 286                 flagged = 1;
 287         }
 288         /*
 289          * Remove the element from the list. Retry if there has been a
 290          * concurrent add (there cannot be a concurrent delete, because
 291          * we won the deletion flag cmpxchg).
 292          */
 293         if (uatomic_cmpxchg(&iter_prev->next, iter, clear_flag(next)) != iter)
 294                 goto retry;
 295 end:
 296         return ret;
 297 }
 298
 299 static
 300 void init_table(struct rcu_ht *ht, struct rcu_table *t,
 301                 unsigned long first, unsigned long len)
 302 {
 303         unsigned long i, end;
 304
 305         end = first + len;
 306         for (i = first; i < end; i++) {
 307                 /* Update table size when power of two */
 308                 if (i != 0 && !(i & (i - 1)))
 309                         t->size = i;
 310                 t->tbl[i] = calloc(1, sizeof(struct rcu_ht_node));
 311                 t->tbl[i]->dummy = 1;
 312                 t->tbl[i]->hash = i;
 313                 t->tbl[i]->reverse_hash = bit_reverse_ulong(i);
 314                 _ht_add(ht, t, t->tbl[i]);
 315         }
 316         t->resize_target = t->size = end;
 317 }
 318
 319 struct rcu_ht *ht_new(ht_hash_fct hash_fct,
 320                       ht_compare_fct compare_fct,
 321                       unsigned long hash_seed,
 322                       unsigned long init_size,
 323                       void (*ht_call_rcu)(struct rcu_head *head,
 324                                 void (*func)(struct rcu_head *head)))
 325 {
 326         struct rcu_ht *ht;
 327
 328         ht = calloc(1, sizeof(struct rcu_ht));
 329         ht->hash_fct = hash_fct;
 330         ht->compare_fct = compare_fct;
 331         ht->hash_seed = hash_seed;
 332         ht->ht_call_rcu = ht_call_rcu;
 333         /* this mutex should not nest in read-side C.S. */
 334         pthread_mutex_init(&ht->resize_mutex, NULL);
 335         ht->t = calloc(1, sizeof(struct rcu_table)
 336                        + (max(init_size, 1) * sizeof(struct rcu_ht_node *)));
 337         ht->t->size = 0;
 338         pthread_mutex_lock(&ht->resize_mutex);
 339         init_table(ht, ht->t, 0, max(init_size, 1));
 340         pthread_mutex_unlock(&ht->resize_mutex);
 341         return ht;
 342 }
 343
 344 struct rcu_ht_node *ht_lookup(struct rcu_ht *ht, void *key, size_t key_len)
 345 {
 346         struct rcu_table *t;
 347         struct rcu_ht_node *node;
 348         unsigned long hash, reverse_hash;
 349
 350         hash = ht->hash_fct(key, key_len, ht->hash_seed);
 351         reverse_hash = bit_reverse_ulong(hash);
 352
 353         t = rcu_dereference(ht->t);
 354         node = rcu_dereference(t->tbl[hash & (t->size - 1)]);
 355         for (;;) {
 356                 if (unlikely(!node))
 357                         break;
 358                 if (unlikely(node->reverse_hash > reverse_hash)) {
 359                         node = NULL;
 360                         break;
 361                 }
 362                 if (!ht->compare_fct(node->key, node->key_len, key, key_len)) {
 363                         if (unlikely(is_removed(rcu_dereference(node->next))))
 364                                 node = NULL;
 365                         break;
 366                 }
 367                 node = clear_flag(rcu_dereference(node->next));
 368         }
 369         return node;
 370 }
 371
 372 void ht_add(struct rcu_ht *ht, struct rcu_ht_node *node)
 373 {
 374         struct rcu_table *t;
 375
 376         node->hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
 377         node->reverse_hash = bit_reverse_ulong((unsigned long) node->hash);
 378
 379         t = rcu_dereference(ht->t);
 380         _ht_add(ht, t, node);
 381 }
 382
 383 int ht_remove(struct rcu_ht *ht, struct rcu_ht_node *node)
 384 {
 385         struct rcu_table *t;
 386
 387         t = rcu_dereference(ht->t);
 388         return _ht_remove(ht, t, node);
 389 }
 390
 391 static
 392 int ht_delete_dummy(struct rcu_ht *ht)
 393 {
 394         struct rcu_table *t;
 395         struct rcu_ht_node *node;
 396         unsigned long i;
 397
 398         t = ht->t;
 399         /* Check that the table is empty */
 400         node = t->tbl[0];
 401         do {
 402                 if (!node->dummy)
 403                         return -EPERM;
 404                 node = node->next;
 405         } while (node);
 406         /* Internal sanity check: all nodes left should be dummy */
 407         for (i = 0; i < t->size; i++) {
 408                 assert(t->tbl[i]->dummy);
 409                 free(t->tbl[i]);
 410         }
 411         return 0;
 412 }
 413
 414 /*
 415  * Should only be called when no more concurrent readers nor writers can
 416  * possibly access the table.
 417  */
 418 int ht_destroy(struct rcu_ht *ht)
 419 {
 420         int ret;
 421
 422         ret = ht_delete_dummy(ht);
 423         if (ret)
 424                 return ret;
 425         free(ht->t);
 426         free(ht);
 427         return ret;
 428 }
 429
 430 static
 431 void ht_free_table_cb(struct rcu_head *head)
 432 {
 433         struct rcu_table *t =
 434                 caa_container_of(head, struct rcu_table, head);
 435         free(t);
 436 }
 437
 438 /* called with resize mutex held */
 439 static
 440 void _do_ht_resize(struct rcu_ht *ht)
 441 {
 442         unsigned long new_size, old_size;
 443         struct rcu_table *new_t, *old_t;
 444
 445         old_t = ht->t;
 446         old_size = old_t->size;
 447
 448         new_size = CMM_LOAD_SHARED(old_t->resize_target);
 449         dbg_printf("rculfhash: resize from %lu to %lu buckets\n",
 450                    old_size, new_size);
 451         if (old_size == new_size)
 452                 return;
 453         new_t = malloc(sizeof(struct rcu_table)
 454                         + (new_size * sizeof(struct rcu_ht_node *)));
 455         assert(new_size > old_size);
 456         memcpy(&new_t->tbl, &old_t->tbl,
 457                old_size * sizeof(struct rcu_ht_node *));
 458         init_table(ht, new_t, old_size, new_size - old_size);
 459         /* Changing table and size atomically wrt lookups */
 460         rcu_assign_pointer(ht->t, new_t);
 461         ht->ht_call_rcu(&old_t->head, ht_free_table_cb);
 462 }
 463
 464 static
 465 unsigned long resize_target_update(struct rcu_table *t,
 466                                    int growth_order)
 467 {
 468         return _uatomic_max(&t->resize_target,
 469                             t->size << growth_order);
 470 }
 471
 472 void ht_resize(struct rcu_ht *ht, int growth)
 473 {
 474         struct rcu_table *t = rcu_dereference(ht->t);
 475         unsigned long target_size;
 476
 477         target_size = resize_target_update(t, growth);
 478         if (t->size < target_size) {
 479                 pthread_mutex_lock(&ht->resize_mutex);
 480                 _do_ht_resize(ht);
 481                 pthread_mutex_unlock(&ht->resize_mutex);
 482         }
 483 }
 484
 485 static
 486 void do_resize_cb(struct rcu_head *head)
 487 {
 488         struct rcu_resize_work *work =
 489                 caa_container_of(head, struct rcu_resize_work, head);
 490         struct rcu_ht *ht = work->ht;
 491
 492         pthread_mutex_lock(&ht->resize_mutex);
 493         _do_ht_resize(ht);
 494         pthread_mutex_unlock(&ht->resize_mutex);
 495         free(work);
 496 }
 497
 498 static
 499 void ht_resize_lazy(struct rcu_ht *ht, struct rcu_table *t, int growth)
 500 {
 501         struct rcu_resize_work *work;
 502         unsigned long target_size;
 503
 504         target_size = resize_target_update(t, growth);
 505         if (t->size < target_size) {
 506                 work = malloc(sizeof(*work));
 507                 work->ht = ht;
 508                 ht->ht_call_rcu(&work->head, do_resize_cb);
 509         }
 510 }