lib/hash.c

   1 /*
   2  * Copyright (C) 2011-2012 Free Software Foundation, Inc.
   3  *
   4  * Author: Nikos Mavrogiannopoulos
   5  *
   6  * This file is part of GnuTLS.
   7  *
   8  * The GnuTLS is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public License
  10  * as published by the Free Software Foundation; either version 3 of
  11  * the License, or (at your option) any later version.
  12  *
  13  * This library is distributed in the hope that it will be useful, but
  14  * 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 License
  19  * along with this program.  If not, see <http://www.gnu.org/licenses/>
  20  *
  21  */
  22
  23 #include <gnutls_int.h>
  24 #include <hash.h>
  25 /*
  26 -------------------------------------------------------------------------------
  27 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
  28
  29 These are functions for producing 32-bit hashes for hash table lookup.
  30 hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
  31 are externally useful functions.  Routines to test the hash are included
  32 if SELF_TEST is defined.  You can use this free for any purpose.  It's in
  33 the public domain.  It has no warranty.
  34
  35 You probably want to use hashlittle().  hashlittle() and hashbig()
  36 hash byte arrays.  hashlittle() is faster than hashbig() on
  37 little-endian machines.  Intel and AMD are little-endian machines.
  38 On second thought, you probably want hashlittle2(), which is identical to
  39 hashlittle() except it returns two 32-bit hashes for the price of one.
  40 You could implement hashbig2() if you wanted but I haven't bothered here.
  41
  42 If you want to find a hash of, say, exactly 7 integers, do
  43   a = i1;  b = i2;  c = i3;
  44   mix(a,b,c);
  45   a += i4; b += i5; c += i6;
  46   mix(a,b,c);
  47   a += i7;
  48   final(a,b,c);
  49 then use c as the hash value.  If you have a variable length array of
  50 4-byte integers to hash, use hashword().  If you have a byte array (like
  51 a character string), use hashlittle().  If you have several byte arrays, or
  52 a mix of things, see the comments above hashlittle().
  53
  54 Why is this so big?  I read 12 bytes at a time into 3 4-byte integers,
  55 then mix those integers.  This is fast (you can do a lot more thorough
  56 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
  57 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
  58 -------------------------------------------------------------------------------
  59 */
  60 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
  61
  62 /*
  63 -------------------------------------------------------------------------------
  64 mix -- mix 3 32-bit values reversibly.
  65
  66 This is reversible, so any information in (a,b,c) before mix() is
  67 still in (a,b,c) after mix().
  68
  69 If four pairs of (a,b,c) inputs are run through mix(), or through
  70 mix() in reverse, there are at least 32 bits of the output that
  71 are sometimes the same for one pair and different for another pair.
  72 This was tested for:
  73 * pairs that differed by one bit, by two bits, in any combination
  74   of top bits of (a,b,c), or in any combination of bottom bits of
  75   (a,b,c).
  76 * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
  77   the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
  78   is commonly produced by subtraction) look like a single 1-bit
  79   difference.
  80 * the base values were pseudorandom, all zero but one bit set, or
  81   all zero plus a counter that starts at zero.
  82
  83 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
  84 satisfy this are
  85     4  6  8 16 19  4
  86     9 15  3 18 27 15
  87    14  9  3  7 17  3
  88 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
  89 for "differ" defined as + with a one-bit base and a two-bit delta.  I
  90 used http://burtleburtle.net/bob/hash/avalanche.html to choose
  91 the operations, constants, and arrangements of the variables.
  92
  93 This does not achieve avalanche.  There are input bits of (a,b,c)
  94 that fail to affect some output bits of (a,b,c), especially of a.  The
  95 most thoroughly mixed value is c, but it doesn't really even achieve
  96 avalanche in c.
  97
  98 This allows some parallelism.  Read-after-writes are good at doubling
  99 the number of bits affected, so the goal of mixing pulls in the opposite
 100 direction as the goal of parallelism.  I did what I could.  Rotates
 101 seem to cost as much as shifts on every machine I could lay my hands
 102 on, and rotates are much kinder to the top and bottom bits, so I used
 103 rotates.
 104 -------------------------------------------------------------------------------
 105 */
 106 #define mix(a,b,c) \
 107 { \
 108   a -= c;  a ^= rot(c, 4);  c += b; \
 109   b -= a;  b ^= rot(a, 6);  a += c; \
 110   c -= b;  c ^= rot(b, 8);  b += a; \
 111   a -= c;  a ^= rot(c,16);  c += b; \
 112   b -= a;  b ^= rot(a,19);  a += c; \
 113   c -= b;  c ^= rot(b, 4);  b += a; \
 114 }
 115
 116 /*
 117 -------------------------------------------------------------------------------
 118 final -- final mixing of 3 32-bit values (a,b,c) into c
 119
 120 Pairs of (a,b,c) values differing in only a few bits will usually
 121 produce values of c that look totally different.  This was tested for
 122 * pairs that differed by one bit, by two bits, in any combination
 123   of top bits of (a,b,c), or in any combination of bottom bits of
 124   (a,b,c).
 125 * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
 126   the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
 127   is commonly produced by subtraction) look like a single 1-bit
 128   difference.
 129 * the base values were pseudorandom, all zero but one bit set, or
 130   all zero plus a counter that starts at zero.
 131
 132 These constants passed:
 133  14 11 25 16 4 14 24
 134  12 14 25 16 4 14 24
 135 and these came close:
 136   4  8 15 26 3 22 24
 137  10  8 15 26 3 22 24
 138  11  8 15 26 3 22 24
 139 -------------------------------------------------------------------------------
 140 */
 141 #define final(a,b,c) \
 142 { \
 143   c ^= b; c -= rot(b,14); \
 144   a ^= c; a -= rot(c,11); \
 145   b ^= a; b -= rot(a,25); \
 146   c ^= b; c -= rot(b,16); \
 147   a ^= c; a -= rot(c,4);  \
 148   b ^= a; b -= rot(a,14); \
 149   c ^= b; c -= rot(b,24); \
 150 }
 151
 152
 153 /*
 154 -------------------------------------------------------------------------------
 155 hashlittle() -- hash a variable-length key into a 32-bit value
 156   k       : the key (the unaligned variable-length array of bytes)
 157   length  : the length of the key, counting by bytes
 158   initval : can be any 4-byte value
 159 Returns a 32-bit value.  Every bit of the key affects every bit of
 160 the return value.  Two keys differing by one or two bits will have
 161 totally different hash values.
 162
 163 The best hash table sizes are powers of 2.  There is no need to do
 164 mod a prime (mod is sooo slow!).  If you need less than 32 bits,
 165 use a bitmask.  For example, if you need only 10 bits, do
 166   h = (h & hashmask(10));
 167 In which case, the hash table should have hashsize(10) elements.
 168
 169 If you are hashing n strings (uint8_t **)k, do it like this:
 170   for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
 171
 172 By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
 173 code any way you wish, private, educational, or commercial.  It's free.
 174
 175 Use for hash table lookup, or anything where one collision in 2^^32 is
 176 acceptable.  Do NOT use for cryptographic purposes.
 177 -------------------------------------------------------------------------------
 178 */
 179
 180 uint32_t _gnutls_bhash( const void *key, size_t length, uint32_t initval)
 181 {
 182   uint32_t a,b,c;                                          /* internal state */
 183   const uint8_t *k;
 184
 185   /* Set up the internal state */
 186   a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
 187
 188   k = (const uint8_t *)key;
 189
 190   /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
 191   while (length > 12)
 192   {
 193     a += k[0];
 194     a += ((uint32_t)k[1])<<8;
 195     a += ((uint32_t)k[2])<<16;
 196     a += ((uint32_t)k[3])<<24;
 197     b += k[4];
 198     b += ((uint32_t)k[5])<<8;
 199     b += ((uint32_t)k[6])<<16;
 200     b += ((uint32_t)k[7])<<24;
 201     c += k[8];
 202     c += ((uint32_t)k[9])<<8;
 203     c += ((uint32_t)k[10])<<16;
 204     c += ((uint32_t)k[11])<<24;
 205     mix(a,b,c);
 206     length -= 12;
 207     k += 12;
 208   }
 209
 210   /*-------------------------------- last block: affect all 32 bits of (c) */
 211   switch(length)                   /* all the case statements fall through */
 212   {
 213     case 12: c+=((uint32_t)k[11])<<24;
 214     case 11: c+=((uint32_t)k[10])<<16;
 215     case 10: c+=((uint32_t)k[9])<<8;
 216     case 9 : c+=k[8];
 217     case 8 : b+=((uint32_t)k[7])<<24;
 218     case 7 : b+=((uint32_t)k[6])<<16;
 219     case 6 : b+=((uint32_t)k[5])<<8;
 220     case 5 : b+=k[4];
 221     case 4 : a+=((uint32_t)k[3])<<24;
 222     case 3 : a+=((uint32_t)k[2])<<16;
 223     case 2 : a+=((uint32_t)k[1])<<8;
 224     case 1 : a+=k[0];
 225              break;
 226     case 0 : return c;
 227   }
 228
 229   final(a,b,c);
 230   return c;
 231 }
 232
 233
 234