source3/lib/jenkins_hash.c

   1  /*
   2    Unix SMB/CIFS implementation.
   3
   4    trivial database library
   5
   6    Copyright (C) Rusty Russell             2010
   7
   8      ** NOTE! The following LGPL license applies to the tdb
   9      ** library. This does NOT imply that all of Samba is released
  10      ** under the LGPL
  11
  12    This library is free software; you can redistribute it and/or
  13    modify it under the terms of the GNU Lesser General Public
  14    License as published by the Free Software Foundation; either
  15    version 3 of the License, or (at your option) any later version.
  16
  17    This library is distributed in the hope that it will be useful,
  18    but WITHOUT ANY WARRANTY; without even the implied warranty of
  19    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20    Lesser General Public License for more details.
  21
  22    You should have received a copy of the GNU Lesser General Public
  23    License along with this library; if not, see <http://www.gnu.org/licenses/>.
  24 */
  25 #include "includes.h"
  26
  27 /* This is based on the hash algorithm from gdbm */
  28
  29
  30 #ifndef WORDS_BIGENDIAN
  31 # define HASH_LITTLE_ENDIAN 1
  32 # define HASH_BIG_ENDIAN 0
  33 #else
  34 # define HASH_LITTLE_ENDIAN 0
  35 # define HASH_BIG_ENDIAN 1
  36 #endif
  37
  38 /*
  39 -------------------------------------------------------------------------------
  40 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
  41
  42 These are functions for producing 32-bit hashes for hash table lookup.
  43 hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
  44 are externally useful functions.  Routines to test the hash are included
  45 if SELF_TEST is defined.  You can use this free for any purpose.  It's in
  46 the public domain.  It has no warranty.
  47
  48 You probably want to use hashlittle().  hashlittle() and hashbig()
  49 hash byte arrays.  hashlittle() is is faster than hashbig() on
  50 little-endian machines.  Intel and AMD are little-endian machines.
  51 On second thought, you probably want hashlittle2(), which is identical to
  52 hashlittle() except it returns two 32-bit hashes for the price of one.
  53 You could implement hashbig2() if you wanted but I haven't bothered here.
  54
  55 If you want to find a hash of, say, exactly 7 integers, do
  56   a = i1;  b = i2;  c = i3;
  57   mix(a,b,c);
  58   a += i4; b += i5; c += i6;
  59   mix(a,b,c);
  60   a += i7;
  61   final(a,b,c);
  62 then use c as the hash value.  If you have a variable length array of
  63 4-byte integers to hash, use hash_word().  If you have a byte array (like
  64 a character string), use hashlittle().  If you have several byte arrays, or
  65 a mix of things, see the comments above hashlittle().
  66
  67 Why is this so big?  I read 12 bytes at a time into 3 4-byte integers,
  68 then mix those integers.  This is fast (you can do a lot more thorough
  69 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
  70 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
  71 */
  72
  73 #define hashsize(n) ((uint32_t)1<<(n))
  74 #define hashmask(n) (hashsize(n)-1)
  75 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
  76
  77 /*
  78 -------------------------------------------------------------------------------
  79 mix -- mix 3 32-bit values reversibly.
  80
  81 This is reversible, so any information in (a,b,c) before mix() is
  82 still in (a,b,c) after mix().
  83
  84 If four pairs of (a,b,c) inputs are run through mix(), or through
  85 mix() in reverse, there are at least 32 bits of the output that
  86 are sometimes the same for one pair and different for another pair.
  87 This was tested for:
  88 * pairs that differed by one bit, by two bits, in any combination
  89   of top bits of (a,b,c), or in any combination of bottom bits of
  90   (a,b,c).
  91 * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
  92   the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
  93   is commonly produced by subtraction) look like a single 1-bit
  94   difference.
  95 * the base values were pseudorandom, all zero but one bit set, or
  96   all zero plus a counter that starts at zero.
  97
  98 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
  99 satisfy this are
 100     4  6  8 16 19  4
 101     9 15  3 18 27 15
 102    14  9  3  7 17  3
 103 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
 104 for "differ" defined as + with a one-bit base and a two-bit delta.  I
 105 used http://burtleburtle.net/bob/hash/avalanche.html to choose
 106 the operations, constants, and arrangements of the variables.
 107
 108 This does not achieve avalanche.  There are input bits of (a,b,c)
 109 that fail to affect some output bits of (a,b,c), especially of a.  The
 110 most thoroughly mixed value is c, but it doesn't really even achieve
 111 avalanche in c.
 112
 113 This allows some parallelism.  Read-after-writes are good at doubling
 114 the number of bits affected, so the goal of mixing pulls in the opposite
 115 direction as the goal of parallelism.  I did what I could.  Rotates
 116 seem to cost as much as shifts on every machine I could lay my hands
 117 on, and rotates are much kinder to the top and bottom bits, so I used
 118 rotates.
 119 -------------------------------------------------------------------------------
 120 */
 121 #define mix(a,b,c) \
 122 { \
 123   a -= c;  a ^= rot(c, 4);  c += b; \
 124   b -= a;  b ^= rot(a, 6);  a += c; \
 125   c -= b;  c ^= rot(b, 8);  b += a; \
 126   a -= c;  a ^= rot(c,16);  c += b; \
 127   b -= a;  b ^= rot(a,19);  a += c; \
 128   c -= b;  c ^= rot(b, 4);  b += a; \
 129 }
 130
 131 /*
 132 -------------------------------------------------------------------------------
 133 final -- final mixing of 3 32-bit values (a,b,c) into c
 134
 135 Pairs of (a,b,c) values differing in only a few bits will usually
 136 produce values of c that look totally different.  This was tested for
 137 * pairs that differed by one bit, by two bits, in any combination
 138   of top bits of (a,b,c), or in any combination of bottom bits of
 139   (a,b,c).
 140 * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
 141   the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
 142   is commonly produced by subtraction) look like a single 1-bit
 143   difference.
 144 * the base values were pseudorandom, all zero but one bit set, or
 145   all zero plus a counter that starts at zero.
 146
 147 These constants passed:
 148  14 11 25 16 4 14 24
 149  12 14 25 16 4 14 24
 150 and these came close:
 151   4  8 15 26 3 22 24
 152  10  8 15 26 3 22 24
 153  11  8 15 26 3 22 24
 154 -------------------------------------------------------------------------------
 155 */
 156 #define final(a,b,c) \
 157 { \
 158   c ^= b; c -= rot(b,14); \
 159   a ^= c; a -= rot(c,11); \
 160   b ^= a; b -= rot(a,25); \
 161   c ^= b; c -= rot(b,16); \
 162   a ^= c; a -= rot(c,4);  \
 163   b ^= a; b -= rot(a,14); \
 164   c ^= b; c -= rot(b,24); \
 165 }
 166
 167
 168 /*
 169 -------------------------------------------------------------------------------
 170 hashlittle() -- hash a variable-length key into a 32-bit value
 171   k       : the key (the unaligned variable-length array of bytes)
 172   length  : the length of the key, counting by bytes
 173   val2    : IN: can be any 4-byte value OUT: second 32 bit hash.
 174 Returns a 32-bit value.  Every bit of the key affects every bit of
 175 the return value.  Two keys differing by one or two bits will have
 176 totally different hash values.  Note that the return value is better
 177 mixed than val2, so use that first.
 178
 179 The best hash table sizes are powers of 2.  There is no need to do
 180 mod a prime (mod is sooo slow!).  If you need less than 32 bits,
 181 use a bitmask.  For example, if you need only 10 bits, do
 182   h = (h & hashmask(10));
 183 In which case, the hash table should have hashsize(10) elements.
 184
 185 If you are hashing n strings (uint8_t **)k, do it like this:
 186   for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
 187
 188 By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
 189 code any way you wish, private, educational, or commercial.  It's free.
 190
 191 Use for hash table lookup, or anything where one collision in 2^^32 is
 192 acceptable.  Do NOT use for cryptographic purposes.
 193 -------------------------------------------------------------------------------
 194 */
 195
 196 static uint32_t hashlittle( const void *key, size_t length )
 197 {
 198   uint32_t a,b,c;                                          /* internal state */
 199   union { const void *ptr; size_t i; } u;     /* needed for Mac Powerbook G4 */
 200
 201   /* Set up the internal state */
 202   a = b = c = 0xdeadbeef + ((uint32_t)length);
 203
 204   u.ptr = key;
 205   if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
 206     const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
 207 #ifdef VALGRIND
 208     const uint8_t  *k8;
 209 #endif
 210
 211     /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
 212     while (length > 12)
 213     {
 214       a += k[0];
 215       b += k[1];
 216       c += k[2];
 217       mix(a,b,c);
 218       length -= 12;
 219       k += 3;
 220     }
 221
 222     /*----------------------------- handle the last (probably partial) block */
 223     /*
 224      * "k[2]&0xffffff" actually reads beyond the end of the string, but
 225      * then masks off the part it's not allowed to read.  Because the
 226      * string is aligned, the masked-off tail is in the same word as the
 227      * rest of the string.  Every machine with memory protection I've seen
 228      * does it on word boundaries, so is OK with this.  But VALGRIND will
 229      * still catch it and complain.  The masking trick does make the hash
 230      * noticably faster for short strings (like English words).
 231      */
 232 #ifndef VALGRIND
 233
 234     switch(length)
 235     {
 236     case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 237     case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
 238     case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
 239     case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
 240     case 8 : b+=k[1]; a+=k[0]; break;
 241     case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
 242     case 6 : b+=k[1]&0xffff; a+=k[0]; break;
 243     case 5 : b+=k[1]&0xff; a+=k[0]; break;
 244     case 4 : a+=k[0]; break;
 245     case 3 : a+=k[0]&0xffffff; break;
 246     case 2 : a+=k[0]&0xffff; break;
 247     case 1 : a+=k[0]&0xff; break;
 248     case 0 : return c;              /* zero length strings require no mixing */
 249     }
 250
 251 #else /* make valgrind happy */
 252
 253     k8 = (const uint8_t *)k;
 254     switch(length)
 255     {
 256     case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 257     case 11: c+=((uint32_t)k8[10])<<16;  /* fall through */
 258     case 10: c+=((uint32_t)k8[9])<<8;    /* fall through */
 259     case 9 : c+=k8[8];                   /* fall through */
 260     case 8 : b+=k[1]; a+=k[0]; break;
 261     case 7 : b+=((uint32_t)k8[6])<<16;   /* fall through */
 262     case 6 : b+=((uint32_t)k8[5])<<8;    /* fall through */
 263     case 5 : b+=k8[4];                   /* fall through */
 264     case 4 : a+=k[0]; break;
 265     case 3 : a+=((uint32_t)k8[2])<<16;   /* fall through */
 266     case 2 : a+=((uint32_t)k8[1])<<8;    /* fall through */
 267     case 1 : a+=k8[0]; break;
 268     case 0 : return c;
 269     }
 270
 271 #endif /* !valgrind */
 272
 273   } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
 274     const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
 275     const uint8_t  *k8;
 276
 277     /*--------------- all but last block: aligned reads and different mixing */
 278     while (length > 12)
 279     {
 280       a += k[0] + (((uint32_t)k[1])<<16);
 281       b += k[2] + (((uint32_t)k[3])<<16);
 282       c += k[4] + (((uint32_t)k[5])<<16);
 283       mix(a,b,c);
 284       length -= 12;
 285       k += 6;
 286     }
 287
 288     /*----------------------------- handle the last (probably partial) block */
 289     k8 = (const uint8_t *)k;
 290     switch(length)
 291     {
 292     case 12: c+=k[4]+(((uint32_t)k[5])<<16);
 293              b+=k[2]+(((uint32_t)k[3])<<16);
 294              a+=k[0]+(((uint32_t)k[1])<<16);
 295              break;
 296     case 11: c+=((uint32_t)k8[10])<<16;     /* fall through */
 297     case 10: c+=k[4];
 298              b+=k[2]+(((uint32_t)k[3])<<16);
 299              a+=k[0]+(((uint32_t)k[1])<<16);
 300              break;
 301     case 9 : c+=k8[8];                      /* fall through */
 302     case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
 303              a+=k[0]+(((uint32_t)k[1])<<16);
 304              break;
 305     case 7 : b+=((uint32_t)k8[6])<<16;      /* fall through */
 306     case 6 : b+=k[2];
 307              a+=k[0]+(((uint32_t)k[1])<<16);
 308              break;
 309     case 5 : b+=k8[4];                      /* fall through */
 310     case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
 311              break;
 312     case 3 : a+=((uint32_t)k8[2])<<16;      /* fall through */
 313     case 2 : a+=k[0];
 314              break;
 315     case 1 : a+=k8[0];
 316              break;
 317     case 0 : return c;                     /* zero length requires no mixing */
 318     }
 319
 320   } else {                        /* need to read the key one byte at a time */
 321     const uint8_t *k = (const uint8_t *)key;
 322
 323     /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
 324     while (length > 12)
 325     {
 326       a += k[0];
 327       a += ((uint32_t)k[1])<<8;
 328       a += ((uint32_t)k[2])<<16;
 329       a += ((uint32_t)k[3])<<24;
 330       b += k[4];
 331       b += ((uint32_t)k[5])<<8;
 332       b += ((uint32_t)k[6])<<16;
 333       b += ((uint32_t)k[7])<<24;
 334       c += k[8];
 335       c += ((uint32_t)k[9])<<8;
 336       c += ((uint32_t)k[10])<<16;
 337       c += ((uint32_t)k[11])<<24;
 338       mix(a,b,c);
 339       length -= 12;
 340       k += 12;
 341     }
 342
 343     /*-------------------------------- last block: affect all 32 bits of (c) */
 344     switch(length)                   /* all the case statements fall through */
 345     {
 346     case 12: c+=((uint32_t)k[11])<<24;
 347     case 11: c+=((uint32_t)k[10])<<16;
 348     case 10: c+=((uint32_t)k[9])<<8;
 349     case 9 : c+=k[8];
 350     case 8 : b+=((uint32_t)k[7])<<24;
 351     case 7 : b+=((uint32_t)k[6])<<16;
 352     case 6 : b+=((uint32_t)k[5])<<8;
 353     case 5 : b+=k[4];
 354     case 4 : a+=((uint32_t)k[3])<<24;
 355     case 3 : a+=((uint32_t)k[2])<<16;
 356     case 2 : a+=((uint32_t)k[1])<<8;
 357     case 1 : a+=k[0];
 358              break;
 359     case 0 : return c;
 360     }
 361   }
 362
 363   final(a,b,c);
 364   return c;
 365 }
 366
 367 unsigned int jenkins_hash(TDB_DATA *key)
 368 {
 369         return hashlittle(key->dptr, key->dsize);
 370 }