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 #ifdef VALGRIND
 218     const uint8_t  *k8;
 219 #endif
 220
 221     /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
 222     while (length > 12)
 223     {
 224       a += k[0];
 225       b += k[1];
 226       c += k[2];
 227       mix(a,b,c);
 228       length -= 12;
 229       k += 3;
 230     }
 231
 232     /*----------------------------- handle the last (probably partial) block */
 233     /*
 234      * "k[2]&0xffffff" actually reads beyond the end of the string, but
 235      * then masks off the part it's not allowed to read.  Because the
 236      * string is aligned, the masked-off tail is in the same word as the
 237      * rest of the string.  Every machine with memory protection I've seen
 238      * does it on word boundaries, so is OK with this.  But VALGRIND will
 239      * still catch it and complain.  The masking trick does make the hash
 240      * noticably faster for short strings (like English words).
 241      */
 242 #ifndef VALGRIND
 243
 244     switch(length)
 245     {
 246     case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 247     case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
 248     case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
 249     case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
 250     case 8 : b+=k[1]; a+=k[0]; break;
 251     case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
 252     case 6 : b+=k[1]&0xffff; a+=k[0]; break;
 253     case 5 : b+=k[1]&0xff; a+=k[0]; break;
 254     case 4 : a+=k[0]; break;
 255     case 3 : a+=k[0]&0xffffff; break;
 256     case 2 : a+=k[0]&0xffff; break;
 257     case 1 : a+=k[0]&0xff; break;
 258     case 0 : return c;              /* zero length strings require no mixing */
 259     }
 260
 261 #else /* make valgrind happy */
 262
 263     k8 = (const uint8_t *)k;
 264     switch(length)
 265     {
 266     case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 267     case 11: c+=((uint32_t)k8[10])<<16;  /* fall through */
 268     case 10: c+=((uint32_t)k8[9])<<8;    /* fall through */
 269     case 9 : c+=k8[8];                   /* fall through */
 270     case 8 : b+=k[1]; a+=k[0]; break;
 271     case 7 : b+=((uint32_t)k8[6])<<16;   /* fall through */
 272     case 6 : b+=((uint32_t)k8[5])<<8;    /* fall through */
 273     case 5 : b+=k8[4];                   /* fall through */
 274     case 4 : a+=k[0]; break;
 275     case 3 : a+=((uint32_t)k8[2])<<16;   /* fall through */
 276     case 2 : a+=((uint32_t)k8[1])<<8;    /* fall through */
 277     case 1 : a+=k8[0]; break;
 278     case 0 : return c;
 279     }
 280
 281 #endif /* !valgrind */
 282
 283   } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
 284     const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
 285     const uint8_t  *k8;
 286
 287     /*--------------- all but last block: aligned reads and different mixing */
 288     while (length > 12)
 289     {
 290       a += k[0] + (((uint32_t)k[1])<<16);
 291       b += k[2] + (((uint32_t)k[3])<<16);
 292       c += k[4] + (((uint32_t)k[5])<<16);
 293       mix(a,b,c);
 294       length -= 12;
 295       k += 6;
 296     }
 297
 298     /*----------------------------- handle the last (probably partial) block */
 299     k8 = (const uint8_t *)k;
 300     switch(length)
 301     {
 302     case 12: c+=k[4]+(((uint32_t)k[5])<<16);
 303              b+=k[2]+(((uint32_t)k[3])<<16);
 304              a+=k[0]+(((uint32_t)k[1])<<16);
 305              break;
 306     case 11: c+=((uint32_t)k8[10])<<16;     /* fall through */
 307     case 10: c+=k[4];
 308              b+=k[2]+(((uint32_t)k[3])<<16);
 309              a+=k[0]+(((uint32_t)k[1])<<16);
 310              break;
 311     case 9 : c+=k8[8];                      /* fall through */
 312     case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
 313              a+=k[0]+(((uint32_t)k[1])<<16);
 314              break;
 315     case 7 : b+=((uint32_t)k8[6])<<16;      /* fall through */
 316     case 6 : b+=k[2];
 317              a+=k[0]+(((uint32_t)k[1])<<16);
 318              break;
 319     case 5 : b+=k8[4];                      /* fall through */
 320     case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
 321              break;
 322     case 3 : a+=((uint32_t)k8[2])<<16;      /* fall through */
 323     case 2 : a+=k[0];
 324              break;
 325     case 1 : a+=k8[0];
 326              break;
 327     case 0 : return c;                     /* zero length requires no mixing */
 328     }
 329
 330   } else {                        /* need to read the key one byte at a time */
 331     const uint8_t *k = (const uint8_t *)key;
 332
 333     /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
 334     while (length > 12)
 335     {
 336       a += k[0];
 337       a += ((uint32_t)k[1])<<8;
 338       a += ((uint32_t)k[2])<<16;
 339       a += ((uint32_t)k[3])<<24;
 340       b += k[4];
 341       b += ((uint32_t)k[5])<<8;
 342       b += ((uint32_t)k[6])<<16;
 343       b += ((uint32_t)k[7])<<24;
 344       c += k[8];
 345       c += ((uint32_t)k[9])<<8;
 346       c += ((uint32_t)k[10])<<16;
 347       c += ((uint32_t)k[11])<<24;
 348       mix(a,b,c);
 349       length -= 12;
 350       k += 12;
 351     }
 352
 353     /*-------------------------------- last block: affect all 32 bits of (c) */
 354     switch(length)                   /* all the case statements fall through */
 355     {
 356     case 12: c+=((uint32_t)k[11])<<24;
 357     case 11: c+=((uint32_t)k[10])<<16;
 358     case 10: c+=((uint32_t)k[9])<<8;
 359     case 9 : c+=k[8];
 360     case 8 : b+=((uint32_t)k[7])<<24;
 361     case 7 : b+=((uint32_t)k[6])<<16;
 362     case 6 : b+=((uint32_t)k[5])<<8;
 363     case 5 : b+=k[4];
 364     case 4 : a+=((uint32_t)k[3])<<24;
 365     case 3 : a+=((uint32_t)k[2])<<16;
 366     case 2 : a+=((uint32_t)k[1])<<8;
 367     case 1 : a+=k[0];
 368              break;
 369     case 0 : return c;
 370     }
 371   }
 372
 373   final(a,b,c);
 374   return c;
 375 }
 376
 377 _PUBLIC_ unsigned int tdb_jenkins_hash(TDB_DATA *key)
 378 {
 379         return hashlittle(key->dptr, key->dsize);
 380 }