lib/tdb/common/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 "tdb_private.h"
  26
  27 /* This is based on the hash algorithm from gdbm */
  28 unsigned int tdb_old_hash(TDB_DATA *key)
  29 {
  30         uint32_t value; /* Used to compute the hash value.  */
  31         uint32_t   i;   /* Used to cycle through random values. */
  32
  33         /* Set the initial value from the key size. */
  34         for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
  35                 value = (value + (key->dptr[i] << (i*5 % 24)));
  36
  37         return (1103515243 * value + 12345);
  38 }
  39
  40 #ifndef WORDS_BIGENDIAN
  41 # define HASH_LITTLE_ENDIAN 1
  42 # define HASH_BIG_ENDIAN 0
  43 #else
  44 # define HASH_LITTLE_ENDIAN 0
  45 # define HASH_BIG_ENDIAN 1
  46 #endif
  47
  48 /*
  49 -------------------------------------------------------------------------------
  50 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
  51
  52 These are functions for producing 32-bit hashes for hash table lookup.
  53 hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
  54 are externally useful functions.  Routines to test the hash are included
  55 if SELF_TEST is defined.  You can use this free for any purpose.  It's in
  56 the public domain.  It has no warranty.
  57
  58 You probably want to use hashlittle().  hashlittle() and hashbig()
  59 hash byte arrays.  hashlittle() is is faster than hashbig() on
  60 little-endian machines.  Intel and AMD are little-endian machines.
  61 On second thought, you probably want hashlittle2(), which is identical to
  62 hashlittle() except it returns two 32-bit hashes for the price of one.
  63 You could implement hashbig2() if you wanted but I haven't bothered here.
  64
  65 If you want to find a hash of, say, exactly 7 integers, do
  66   a = i1;  b = i2;  c = i3;
  67   mix(a,b,c);
  68   a += i4; b += i5; c += i6;
  69   mix(a,b,c);
  70   a += i7;
  71   final(a,b,c);
  72 then use c as the hash value.  If you have a variable length array of
  73 4-byte integers to hash, use hash_word().  If you have a byte array (like
  74 a character string), use hashlittle().  If you have several byte arrays, or
  75 a mix of things, see the comments above hashlittle().
  76
  77 Why is this so big?  I read 12 bytes at a time into 3 4-byte integers,
  78 then mix those integers.  This is fast (you can do a lot more thorough
  79 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
  80 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
  81 */
  82
  83 #define hashsize(n) ((uint32_t)1<<(n))
  84 #define hashmask(n) (hashsize(n)-1)
  85 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
  86
  87 /*
  88 -------------------------------------------------------------------------------
  89 mix -- mix 3 32-bit values reversibly.
  90
  91 This is reversible, so any information in (a,b,c) before mix() is
  92 still in (a,b,c) after mix().
  93
  94 If four pairs of (a,b,c) inputs are run through mix(), or through
  95 mix() in reverse, there are at least 32 bits of the output that
  96 are sometimes the same for one pair and different for another pair.
  97 This was tested for:
  98 * pairs that differed by one bit, by two bits, in any combination
  99   of top bits of (a,b,c), or in any combination of bottom bits of
 100   (a,b,c).
 101 * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
 102   the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
 103   is commonly produced by subtraction) look like a single 1-bit
 104   difference.
 105 * the base values were pseudorandom, all zero but one bit set, or
 106   all zero plus a counter that starts at zero.
 107
 108 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
 109 satisfy this are
 110     4  6  8 16 19  4
 111     9 15  3 18 27 15
 112    14  9  3  7 17  3
 113 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
 114 for "differ" defined as + with a one-bit base and a two-bit delta.  I
 115 used http://burtleburtle.net/bob/hash/avalanche.html to choose
 116 the operations, constants, and arrangements of the variables.
 117
 118 This does not achieve avalanche.  There are input bits of (a,b,c)
 119 that fail to affect some output bits of (a,b,c), especially of a.  The
 120 most thoroughly mixed value is c, but it doesn't really even achieve
 121 avalanche in c.
 122
 123 This allows some parallelism.  Read-after-writes are good at doubling
 124 the number of bits affected, so the goal of mixing pulls in the opposite
 125 direction as the goal of parallelism.  I did what I could.  Rotates
 126 seem to cost as much as shifts on every machine I could lay my hands
 127 on, and rotates are much kinder to the top and bottom bits, so I used
 128 rotates.
 129 -------------------------------------------------------------------------------
 130 */
 131 #define mix(a,b,c) \
 132 { \
 133   a -= c;  a ^= rot(c, 4);  c += b; \
 134   b -= a;  b ^= rot(a, 6);  a += c; \
 135   c -= b;  c ^= rot(b, 8);  b += a; \
 136   a -= c;  a ^= rot(c,16);  c += b; \
 137   b -= a;  b ^= rot(a,19);  a += c; \
 138   c -= b;  c ^= rot(b, 4);  b += a; \
 139 }
 140
 141 /*
 142 -------------------------------------------------------------------------------
 143 final -- final mixing of 3 32-bit values (a,b,c) into c
 144
 145 Pairs of (a,b,c) values differing in only a few bits will usually
 146 produce values of c that look totally different.  This was tested for
 147 * pairs that differed by one bit, by two bits, in any combination
 148   of top bits of (a,b,c), or in any combination of bottom bits of
 149   (a,b,c).
 150 * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
 151   the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
 152   is commonly produced by subtraction) look like a single 1-bit
 153   difference.
 154 * the base values were pseudorandom, all zero but one bit set, or
 155   all zero plus a counter that starts at zero.
 156
 157 These constants passed:
 158  14 11 25 16 4 14 24
 159  12 14 25 16 4 14 24
 160 and these came close:
 161   4  8 15 26 3 22 24
 162  10  8 15 26 3 22 24
 163  11  8 15 26 3 22 24
 164 -------------------------------------------------------------------------------
 165 */
 166 #define final(a,b,c) \
 167 { \
 168   c ^= b; c -= rot(b,14); \
 169   a ^= c; a -= rot(c,11); \
 170   b ^= a; b -= rot(a,25); \
 171   c ^= b; c -= rot(b,16); \
 172   a ^= c; a -= rot(c,4);  \
 173   b ^= a; b -= rot(a,14); \
 174   c ^= b; c -= rot(b,24); \
 175 }
 176
 177
 178 /*
 179 -------------------------------------------------------------------------------
 180 hashlittle() -- hash a variable-length key into a 32-bit value
 181   k       : the key (the unaligned variable-length array of bytes)
 182   length  : the length of the key, counting by bytes
 183   val2    : IN: can be any 4-byte value OUT: second 32 bit hash.
 184 Returns a 32-bit value.  Every bit of the key affects every bit of
 185 the return value.  Two keys differing by one or two bits will have
 186 totally different hash values.  Note that the return value is better
 187 mixed than val2, so use that first.
 188
 189 The best hash table sizes are powers of 2.  There is no need to do
 190 mod a prime (mod is sooo slow!).  If you need less than 32 bits,
 191 use a bitmask.  For example, if you need only 10 bits, do
 192   h = (h & hashmask(10));
 193 In which case, the hash table should have hashsize(10) elements.
 194
 195 If you are hashing n strings (uint8_t **)k, do it like this:
 196   for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
 197
 198 By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
 199 code any way you wish, private, educational, or commercial.  It's free.
 200
 201 Use for hash table lookup, or anything where one collision in 2^^32 is
 202 acceptable.  Do NOT use for cryptographic purposes.
 203 -------------------------------------------------------------------------------
 204 */
 205
 206 static uint32_t hashlittle( const void *key, size_t length )
 207 {
 208   uint32_t a,b,c;                                          /* internal state */
 209   union { const void *ptr; size_t i; } u;     /* needed for Mac Powerbook G4 */
 210
 211   /* Set up the internal state */
 212   a = b = c = 0xdeadbeef + ((uint32_t)length);
 213
 214   u.ptr = key;
 215   if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
 216     const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
 217     const uint8_t  *k8;
 218
 219     /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
 220     while (length > 12)
 221     {
 222       a += k[0];
 223       b += k[1];
 224       c += k[2];
 225       mix(a,b,c);
 226       length -= 12;
 227       k += 3;
 228     }
 229
 230     /*----------------------------- handle the last (probably partial) block */
 231     k8 = (const uint8_t *)k;
 232     switch(length)
 233     {
 234     case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 235     case 11: c+=((uint32_t)k8[10])<<16;  /* fall through */
 236     case 10: c+=((uint32_t)k8[9])<<8;    /* fall through */
 237     case 9 : c+=k8[8];                   /* fall through */
 238     case 8 : b+=k[1]; a+=k[0]; break;
 239     case 7 : b+=((uint32_t)k8[6])<<16;   /* fall through */
 240     case 6 : b+=((uint32_t)k8[5])<<8;    /* fall through */
 241     case 5 : b+=k8[4];                   /* fall through */
 242     case 4 : a+=k[0]; break;
 243     case 3 : a+=((uint32_t)k8[2])<<16;   /* fall through */
 244     case 2 : a+=((uint32_t)k8[1])<<8;    /* fall through */
 245     case 1 : a+=k8[0]; break;
 246     case 0 : return c;
 247     }
 248   } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
 249     const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
 250     const uint8_t  *k8;
 251
 252     /*--------------- all but last block: aligned reads and different mixing */
 253     while (length > 12)
 254     {
 255       a += k[0] + (((uint32_t)k[1])<<16);
 256       b += k[2] + (((uint32_t)k[3])<<16);
 257       c += k[4] + (((uint32_t)k[5])<<16);
 258       mix(a,b,c);
 259       length -= 12;
 260       k += 6;
 261     }
 262
 263     /*----------------------------- handle the last (probably partial) block */
 264     k8 = (const uint8_t *)k;
 265     switch(length)
 266     {
 267     case 12: c+=k[4]+(((uint32_t)k[5])<<16);
 268              b+=k[2]+(((uint32_t)k[3])<<16);
 269              a+=k[0]+(((uint32_t)k[1])<<16);
 270              break;
 271     case 11: c+=((uint32_t)k8[10])<<16;     /* fall through */
 272     case 10: c+=k[4];
 273              b+=k[2]+(((uint32_t)k[3])<<16);
 274              a+=k[0]+(((uint32_t)k[1])<<16);
 275              break;
 276     case 9 : c+=k8[8];                      /* fall through */
 277     case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
 278              a+=k[0]+(((uint32_t)k[1])<<16);
 279              break;
 280     case 7 : b+=((uint32_t)k8[6])<<16;      /* fall through */
 281     case 6 : b+=k[2];
 282              a+=k[0]+(((uint32_t)k[1])<<16);
 283              break;
 284     case 5 : b+=k8[4];                      /* fall through */
 285     case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
 286              break;
 287     case 3 : a+=((uint32_t)k8[2])<<16;      /* fall through */
 288     case 2 : a+=k[0];
 289              break;
 290     case 1 : a+=k8[0];
 291              break;
 292     case 0 : return c;                     /* zero length requires no mixing */
 293     }
 294
 295   } else {                        /* need to read the key one byte at a time */
 296     const uint8_t *k = (const uint8_t *)key;
 297
 298     /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
 299     while (length > 12)
 300     {
 301       a += k[0];
 302       a += ((uint32_t)k[1])<<8;
 303       a += ((uint32_t)k[2])<<16;
 304       a += ((uint32_t)k[3])<<24;
 305       b += k[4];
 306       b += ((uint32_t)k[5])<<8;
 307       b += ((uint32_t)k[6])<<16;
 308       b += ((uint32_t)k[7])<<24;
 309       c += k[8];
 310       c += ((uint32_t)k[9])<<8;
 311       c += ((uint32_t)k[10])<<16;
 312       c += ((uint32_t)k[11])<<24;
 313       mix(a,b,c);
 314       length -= 12;
 315       k += 12;
 316     }
 317
 318     /*-------------------------------- last block: affect all 32 bits of (c) */
 319     switch(length)                   /* all the case statements fall through */
 320     {
 321     case 12: c+=((uint32_t)k[11])<<24;
 322     case 11: c+=((uint32_t)k[10])<<16;
 323     case 10: c+=((uint32_t)k[9])<<8;
 324     case 9 : c+=k[8];
 325     case 8 : b+=((uint32_t)k[7])<<24;
 326     case 7 : b+=((uint32_t)k[6])<<16;
 327     case 6 : b+=((uint32_t)k[5])<<8;
 328     case 5 : b+=k[4];
 329     case 4 : a+=((uint32_t)k[3])<<24;
 330     case 3 : a+=((uint32_t)k[2])<<16;
 331     case 2 : a+=((uint32_t)k[1])<<8;
 332     case 1 : a+=k[0];
 333              break;
 334     case 0 : return c;
 335     }
 336   }
 337
 338   final(a,b,c);
 339   return c;
 340 }
 341
 342 _PUBLIC_ unsigned int tdb_jenkins_hash(TDB_DATA *key)
 343 {
 344         return hashlittle(key->dptr, key->dsize);
 345 }