lib/xxhash.c

   1 /*
   2  * xxHash - Extremely Fast Hash algorithm
   3  * Copyright (C) 2012-2016, Yann Collet.
   4  *
   5  * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions are
   9  * met:
  10  *
  11  *   * Redistributions of source code must retain the above copyright
  12  *     notice, this list of conditions and the following disclaimer.
  13  *   * Redistributions in binary form must reproduce the above
  14  *     copyright notice, this list of conditions and the following disclaimer
  15  *     in the documentation and/or other materials provided with the
  16  *     distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  21  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  22  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  23  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  24  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  28  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29  *
  30  * This program is free software; you can redistribute it and/or modify it under
  31  * the terms of the GNU General Public License version 2 as published by the
  32  * Free Software Foundation. This program is dual-licensed; you may select
  33  * either version 2 of the GNU General Public License ("GPL") or BSD license
  34  * ("BSD").
  35  *
  36  * You can contact the author at:
  37  * - xxHash homepage: https://cyan4973.github.io/xxHash/
  38  * - xxHash source repository: https://github.com/Cyan4973/xxHash
  39  */
  40
  41 #include <asm/unaligned.h>
  42 #include <linux/errno.h>
  43 #include <linux/compiler.h>
  44 #include <linux/kernel.h>
  45 #include <linux/module.h>
  46 #include <linux/string.h>
  47 #include <linux/xxhash.h>
  48
  49 /*-*************************************
  50  * Macros
  51  **************************************/
  52 #define xxh_rotl32(x, r) ((x << r) | (x >> (32 - r)))
  53 #define xxh_rotl64(x, r) ((x << r) | (x >> (64 - r)))
  54
  55 #ifdef __LITTLE_ENDIAN
  56 # define XXH_CPU_LITTLE_ENDIAN 1
  57 #else
  58 # define XXH_CPU_LITTLE_ENDIAN 0
  59 #endif
  60
  61 /*-*************************************
  62  * Constants
  63  **************************************/
  64 static const uint32_t PRIME32_1 = 2654435761U;
  65 static const uint32_t PRIME32_2 = 2246822519U;
  66 static const uint32_t PRIME32_3 = 3266489917U;
  67 static const uint32_t PRIME32_4 =  668265263U;
  68 static const uint32_t PRIME32_5 =  374761393U;
  69
  70 static const uint64_t PRIME64_1 = 11400714785074694791ULL;
  71 static const uint64_t PRIME64_2 = 14029467366897019727ULL;
  72 static const uint64_t PRIME64_3 =  1609587929392839161ULL;
  73 static const uint64_t PRIME64_4 =  9650029242287828579ULL;
  74 static const uint64_t PRIME64_5 =  2870177450012600261ULL;
  75
  76 /*-**************************
  77  *  Utils
  78  ***************************/
  79 void xxh32_copy_state(struct xxh32_state *dst, const struct xxh32_state *src)
  80 {
  81         memcpy(dst, src, sizeof(*dst));
  82 }
  83 EXPORT_SYMBOL(xxh32_copy_state);
  84
  85 void xxh64_copy_state(struct xxh64_state *dst, const struct xxh64_state *src)
  86 {
  87         memcpy(dst, src, sizeof(*dst));
  88 }
  89 EXPORT_SYMBOL(xxh64_copy_state);
  90
  91 /*-***************************
  92  * Simple Hash Functions
  93  ****************************/
  94 static uint32_t xxh32_round(uint32_t seed, const uint32_t input)
  95 {
  96         seed += input * PRIME32_2;
  97         seed = xxh_rotl32(seed, 13);
  98         seed *= PRIME32_1;
  99         return seed;
 100 }
 101
 102 uint32_t xxh32(const void *input, const size_t len, const uint32_t seed)
 103 {
 104         const uint8_t *p = (const uint8_t *)input;
 105         const uint8_t *b_end = p + len;
 106         uint32_t h32;
 107
 108         if (len >= 16) {
 109                 const uint8_t *const limit = b_end - 16;
 110                 uint32_t v1 = seed + PRIME32_1 + PRIME32_2;
 111                 uint32_t v2 = seed + PRIME32_2;
 112                 uint32_t v3 = seed + 0;
 113                 uint32_t v4 = seed - PRIME32_1;
 114
 115                 do {
 116                         v1 = xxh32_round(v1, get_unaligned_le32(p));
 117                         p += 4;
 118                         v2 = xxh32_round(v2, get_unaligned_le32(p));
 119                         p += 4;
 120                         v3 = xxh32_round(v3, get_unaligned_le32(p));
 121                         p += 4;
 122                         v4 = xxh32_round(v4, get_unaligned_le32(p));
 123                         p += 4;
 124                 } while (p <= limit);
 125
 126                 h32 = xxh_rotl32(v1, 1) + xxh_rotl32(v2, 7) +
 127                         xxh_rotl32(v3, 12) + xxh_rotl32(v4, 18);
 128         } else {
 129                 h32 = seed + PRIME32_5;
 130         }
 131
 132         h32 += (uint32_t)len;
 133
 134         while (p + 4 <= b_end) {
 135                 h32 += get_unaligned_le32(p) * PRIME32_3;
 136                 h32 = xxh_rotl32(h32, 17) * PRIME32_4;
 137                 p += 4;
 138         }
 139
 140         while (p < b_end) {
 141                 h32 += (*p) * PRIME32_5;
 142                 h32 = xxh_rotl32(h32, 11) * PRIME32_1;
 143                 p++;
 144         }
 145
 146         h32 ^= h32 >> 15;
 147         h32 *= PRIME32_2;
 148         h32 ^= h32 >> 13;
 149         h32 *= PRIME32_3;
 150         h32 ^= h32 >> 16;
 151
 152         return h32;
 153 }
 154 EXPORT_SYMBOL(xxh32);
 155
 156 static uint64_t xxh64_round(uint64_t acc, const uint64_t input)
 157 {
 158         acc += input * PRIME64_2;
 159         acc = xxh_rotl64(acc, 31);
 160         acc *= PRIME64_1;
 161         return acc;
 162 }
 163
 164 static uint64_t xxh64_merge_round(uint64_t acc, uint64_t val)
 165 {
 166         val = xxh64_round(0, val);
 167         acc ^= val;
 168         acc = acc * PRIME64_1 + PRIME64_4;
 169         return acc;
 170 }
 171
 172 uint64_t xxh64(const void *input, const size_t len, const uint64_t seed)
 173 {
 174         const uint8_t *p = (const uint8_t *)input;
 175         const uint8_t *const b_end = p + len;
 176         uint64_t h64;
 177
 178         if (len >= 32) {
 179                 const uint8_t *const limit = b_end - 32;
 180                 uint64_t v1 = seed + PRIME64_1 + PRIME64_2;
 181                 uint64_t v2 = seed + PRIME64_2;
 182                 uint64_t v3 = seed + 0;
 183                 uint64_t v4 = seed - PRIME64_1;
 184
 185                 do {
 186                         v1 = xxh64_round(v1, get_unaligned_le64(p));
 187                         p += 8;
 188                         v2 = xxh64_round(v2, get_unaligned_le64(p));
 189                         p += 8;
 190                         v3 = xxh64_round(v3, get_unaligned_le64(p));
 191                         p += 8;
 192                         v4 = xxh64_round(v4, get_unaligned_le64(p));
 193                         p += 8;
 194                 } while (p <= limit);
 195
 196                 h64 = xxh_rotl64(v1, 1) + xxh_rotl64(v2, 7) +
 197                         xxh_rotl64(v3, 12) + xxh_rotl64(v4, 18);
 198                 h64 = xxh64_merge_round(h64, v1);
 199                 h64 = xxh64_merge_round(h64, v2);
 200                 h64 = xxh64_merge_round(h64, v3);
 201                 h64 = xxh64_merge_round(h64, v4);
 202
 203         } else {
 204                 h64  = seed + PRIME64_5;
 205         }
 206
 207         h64 += (uint64_t)len;
 208
 209         while (p + 8 <= b_end) {
 210                 const uint64_t k1 = xxh64_round(0, get_unaligned_le64(p));
 211
 212                 h64 ^= k1;
 213                 h64 = xxh_rotl64(h64, 27) * PRIME64_1 + PRIME64_4;
 214                 p += 8;
 215         }
 216
 217         if (p + 4 <= b_end) {
 218                 h64 ^= (uint64_t)(get_unaligned_le32(p)) * PRIME64_1;
 219                 h64 = xxh_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
 220                 p += 4;
 221         }
 222
 223         while (p < b_end) {
 224                 h64 ^= (*p) * PRIME64_5;
 225                 h64 = xxh_rotl64(h64, 11) * PRIME64_1;
 226                 p++;
 227         }
 228
 229         h64 ^= h64 >> 33;
 230         h64 *= PRIME64_2;
 231         h64 ^= h64 >> 29;
 232         h64 *= PRIME64_3;
 233         h64 ^= h64 >> 32;
 234
 235         return h64;
 236 }
 237 EXPORT_SYMBOL(xxh64);
 238
 239 /*-**************************************************
 240  * Advanced Hash Functions
 241  ***************************************************/
 242 void xxh32_reset(struct xxh32_state *statePtr, const uint32_t seed)
 243 {
 244         /* use a local state for memcpy() to avoid strict-aliasing warnings */
 245         struct xxh32_state state;
 246
 247         memset(&state, 0, sizeof(state));
 248         state.v1 = seed + PRIME32_1 + PRIME32_2;
 249         state.v2 = seed + PRIME32_2;
 250         state.v3 = seed + 0;
 251         state.v4 = seed - PRIME32_1;
 252         memcpy(statePtr, &state, sizeof(state));
 253 }
 254 EXPORT_SYMBOL(xxh32_reset);
 255
 256 void xxh64_reset(struct xxh64_state *statePtr, const uint64_t seed)
 257 {
 258         /* use a local state for memcpy() to avoid strict-aliasing warnings */
 259         struct xxh64_state state;
 260
 261         memset(&state, 0, sizeof(state));
 262         state.v1 = seed + PRIME64_1 + PRIME64_2;
 263         state.v2 = seed + PRIME64_2;
 264         state.v3 = seed + 0;
 265         state.v4 = seed - PRIME64_1;
 266         memcpy(statePtr, &state, sizeof(state));
 267 }
 268 EXPORT_SYMBOL(xxh64_reset);
 269
 270 int xxh32_update(struct xxh32_state *state, const void *input, const size_t len)
 271 {
 272         const uint8_t *p = (const uint8_t *)input;
 273         const uint8_t *const b_end = p + len;
 274
 275         if (input == NULL)
 276                 return -EINVAL;
 277
 278         state->total_len_32 += (uint32_t)len;
 279         state->large_len |= (len >= 16) | (state->total_len_32 >= 16);
 280
 281         if (state->memsize + len < 16) { /* fill in tmp buffer */
 282                 memcpy((uint8_t *)(state->mem32) + state->memsize, input, len);
 283                 state->memsize += (uint32_t)len;
 284                 return 0;
 285         }
 286
 287         if (state->memsize) { /* some data left from previous update */
 288                 const uint32_t *p32 = state->mem32;
 289
 290                 memcpy((uint8_t *)(state->mem32) + state->memsize, input,
 291                         16 - state->memsize);
 292
 293                 state->v1 = xxh32_round(state->v1, get_unaligned_le32(p32));
 294                 p32++;
 295                 state->v2 = xxh32_round(state->v2, get_unaligned_le32(p32));
 296                 p32++;
 297                 state->v3 = xxh32_round(state->v3, get_unaligned_le32(p32));
 298                 p32++;
 299                 state->v4 = xxh32_round(state->v4, get_unaligned_le32(p32));
 300                 p32++;
 301
 302                 p += 16-state->memsize;
 303                 state->memsize = 0;
 304         }
 305
 306         if (p <= b_end - 16) {
 307                 const uint8_t *const limit = b_end - 16;
 308                 uint32_t v1 = state->v1;
 309                 uint32_t v2 = state->v2;
 310                 uint32_t v3 = state->v3;
 311                 uint32_t v4 = state->v4;
 312
 313                 do {
 314                         v1 = xxh32_round(v1, get_unaligned_le32(p));
 315                         p += 4;
 316                         v2 = xxh32_round(v2, get_unaligned_le32(p));
 317                         p += 4;
 318                         v3 = xxh32_round(v3, get_unaligned_le32(p));
 319                         p += 4;
 320                         v4 = xxh32_round(v4, get_unaligned_le32(p));
 321                         p += 4;
 322                 } while (p <= limit);
 323
 324                 state->v1 = v1;
 325                 state->v2 = v2;
 326                 state->v3 = v3;
 327                 state->v4 = v4;
 328         }
 329
 330         if (p < b_end) {
 331                 memcpy(state->mem32, p, (size_t)(b_end-p));
 332                 state->memsize = (uint32_t)(b_end-p);
 333         }
 334
 335         return 0;
 336 }
 337 EXPORT_SYMBOL(xxh32_update);
 338
 339 uint32_t xxh32_digest(const struct xxh32_state *state)
 340 {
 341         const uint8_t *p = (const uint8_t *)state->mem32;
 342         const uint8_t *const b_end = (const uint8_t *)(state->mem32) +
 343                 state->memsize;
 344         uint32_t h32;
 345
 346         if (state->large_len) {
 347                 h32 = xxh_rotl32(state->v1, 1) + xxh_rotl32(state->v2, 7) +
 348                         xxh_rotl32(state->v3, 12) + xxh_rotl32(state->v4, 18);
 349         } else {
 350                 h32 = state->v3 /* == seed */ + PRIME32_5;
 351         }
 352
 353         h32 += state->total_len_32;
 354
 355         while (p + 4 <= b_end) {
 356                 h32 += get_unaligned_le32(p) * PRIME32_3;
 357                 h32 = xxh_rotl32(h32, 17) * PRIME32_4;
 358                 p += 4;
 359         }
 360
 361         while (p < b_end) {
 362                 h32 += (*p) * PRIME32_5;
 363                 h32 = xxh_rotl32(h32, 11) * PRIME32_1;
 364                 p++;
 365         }
 366
 367         h32 ^= h32 >> 15;
 368         h32 *= PRIME32_2;
 369         h32 ^= h32 >> 13;
 370         h32 *= PRIME32_3;
 371         h32 ^= h32 >> 16;
 372
 373         return h32;
 374 }
 375 EXPORT_SYMBOL(xxh32_digest);
 376
 377 int xxh64_update(struct xxh64_state *state, const void *input, const size_t len)
 378 {
 379         const uint8_t *p = (const uint8_t *)input;
 380         const uint8_t *const b_end = p + len;
 381
 382         if (input == NULL)
 383                 return -EINVAL;
 384
 385         state->total_len += len;
 386
 387         if (state->memsize + len < 32) { /* fill in tmp buffer */
 388                 memcpy(((uint8_t *)state->mem64) + state->memsize, input, len);
 389                 state->memsize += (uint32_t)len;
 390                 return 0;
 391         }
 392
 393         if (state->memsize) { /* tmp buffer is full */
 394                 uint64_t *p64 = state->mem64;
 395
 396                 memcpy(((uint8_t *)p64) + state->memsize, input,
 397                         32 - state->memsize);
 398
 399                 state->v1 = xxh64_round(state->v1, get_unaligned_le64(p64));
 400                 p64++;
 401                 state->v2 = xxh64_round(state->v2, get_unaligned_le64(p64));
 402                 p64++;
 403                 state->v3 = xxh64_round(state->v3, get_unaligned_le64(p64));
 404                 p64++;
 405                 state->v4 = xxh64_round(state->v4, get_unaligned_le64(p64));
 406
 407                 p += 32 - state->memsize;
 408                 state->memsize = 0;
 409         }
 410
 411         if (p + 32 <= b_end) {
 412                 const uint8_t *const limit = b_end - 32;
 413                 uint64_t v1 = state->v1;
 414                 uint64_t v2 = state->v2;
 415                 uint64_t v3 = state->v3;
 416                 uint64_t v4 = state->v4;
 417
 418                 do {
 419                         v1 = xxh64_round(v1, get_unaligned_le64(p));
 420                         p += 8;
 421                         v2 = xxh64_round(v2, get_unaligned_le64(p));
 422                         p += 8;
 423                         v3 = xxh64_round(v3, get_unaligned_le64(p));
 424                         p += 8;
 425                         v4 = xxh64_round(v4, get_unaligned_le64(p));
 426                         p += 8;
 427                 } while (p <= limit);
 428
 429                 state->v1 = v1;
 430                 state->v2 = v2;
 431                 state->v3 = v3;
 432                 state->v4 = v4;
 433         }
 434
 435         if (p < b_end) {
 436                 memcpy(state->mem64, p, (size_t)(b_end-p));
 437                 state->memsize = (uint32_t)(b_end - p);
 438         }
 439
 440         return 0;
 441 }
 442 EXPORT_SYMBOL(xxh64_update);
 443
 444 uint64_t xxh64_digest(const struct xxh64_state *state)
 445 {
 446         const uint8_t *p = (const uint8_t *)state->mem64;
 447         const uint8_t *const b_end = (const uint8_t *)state->mem64 +
 448                 state->memsize;
 449         uint64_t h64;
 450
 451         if (state->total_len >= 32) {
 452                 const uint64_t v1 = state->v1;
 453                 const uint64_t v2 = state->v2;
 454                 const uint64_t v3 = state->v3;
 455                 const uint64_t v4 = state->v4;
 456
 457                 h64 = xxh_rotl64(v1, 1) + xxh_rotl64(v2, 7) +
 458                         xxh_rotl64(v3, 12) + xxh_rotl64(v4, 18);
 459                 h64 = xxh64_merge_round(h64, v1);
 460                 h64 = xxh64_merge_round(h64, v2);
 461                 h64 = xxh64_merge_round(h64, v3);
 462                 h64 = xxh64_merge_round(h64, v4);
 463         } else {
 464                 h64  = state->v3 + PRIME64_5;
 465         }
 466
 467         h64 += (uint64_t)state->total_len;
 468
 469         while (p + 8 <= b_end) {
 470                 const uint64_t k1 = xxh64_round(0, get_unaligned_le64(p));
 471
 472                 h64 ^= k1;
 473                 h64 = xxh_rotl64(h64, 27) * PRIME64_1 + PRIME64_4;
 474                 p += 8;
 475         }
 476
 477         if (p + 4 <= b_end) {
 478                 h64 ^= (uint64_t)(get_unaligned_le32(p)) * PRIME64_1;
 479                 h64 = xxh_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
 480                 p += 4;
 481         }
 482
 483         while (p < b_end) {
 484                 h64 ^= (*p) * PRIME64_5;
 485                 h64 = xxh_rotl64(h64, 11) * PRIME64_1;
 486                 p++;
 487         }
 488
 489         h64 ^= h64 >> 33;
 490         h64 *= PRIME64_2;
 491         h64 ^= h64 >> 29;
 492         h64 *= PRIME64_3;
 493         h64 ^= h64 >> 32;
 494
 495         return h64;
 496 }
 497 EXPORT_SYMBOL(xxh64_digest);
 498
 499 MODULE_LICENSE("Dual BSD/GPL");
 500 MODULE_DESCRIPTION("xxHash");