2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Rusty Russell 2010
8 ** NOTE! The following LGPL license applies to the tdb
9 ** library. This does NOT imply that all of Samba is released
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.
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.
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/>.
25 #include "tdb_private.h"
27 /* This is based on the hash algorithm from gdbm */
28 unsigned int tdb_old_hash(TDB_DATA *key)
30 uint32_t value; /* Used to compute the hash value. */
31 uint32_t i; /* Used to cycle through random values. */
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)));
37 return (1103515243 * value + 12345);
40 #if HAVE_LITTLE_ENDIAN
41 # define HASH_LITTLE_ENDIAN 1
42 # define HASH_BIG_ENDIAN 0
44 # define HASH_LITTLE_ENDIAN 0
45 # define HASH_BIG_ENDIAN 1
47 # error Unknown endian
51 -------------------------------------------------------------------------------
52 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
54 These are functions for producing 32-bit hashes for hash table lookup.
55 hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
56 are externally useful functions. Routines to test the hash are included
57 if SELF_TEST is defined. You can use this free for any purpose. It's in
58 the public domain. It has no warranty.
60 You probably want to use hashlittle(). hashlittle() and hashbig()
61 hash byte arrays. hashlittle() is is faster than hashbig() on
62 little-endian machines. Intel and AMD are little-endian machines.
63 On second thought, you probably want hashlittle2(), which is identical to
64 hashlittle() except it returns two 32-bit hashes for the price of one.
65 You could implement hashbig2() if you wanted but I haven't bothered here.
67 If you want to find a hash of, say, exactly 7 integers, do
68 a = i1; b = i2; c = i3;
70 a += i4; b += i5; c += i6;
74 then use c as the hash value. If you have a variable length array of
75 4-byte integers to hash, use hash_word(). If you have a byte array (like
76 a character string), use hashlittle(). If you have several byte arrays, or
77 a mix of things, see the comments above hashlittle().
79 Why is this so big? I read 12 bytes at a time into 3 4-byte integers,
80 then mix those integers. This is fast (you can do a lot more thorough
81 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
82 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
85 #define hashsize(n) ((uint32_t)1<<(n))
86 #define hashmask(n) (hashsize(n)-1)
87 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
90 -------------------------------------------------------------------------------
91 mix -- mix 3 32-bit values reversibly.
93 This is reversible, so any information in (a,b,c) before mix() is
94 still in (a,b,c) after mix().
96 If four pairs of (a,b,c) inputs are run through mix(), or through
97 mix() in reverse, there are at least 32 bits of the output that
98 are sometimes the same for one pair and different for another pair.
100 * pairs that differed by one bit, by two bits, in any combination
101 of top bits of (a,b,c), or in any combination of bottom bits of
103 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
104 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
105 is commonly produced by subtraction) look like a single 1-bit
107 * the base values were pseudorandom, all zero but one bit set, or
108 all zero plus a counter that starts at zero.
110 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
115 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
116 for "differ" defined as + with a one-bit base and a two-bit delta. I
117 used http://burtleburtle.net/bob/hash/avalanche.html to choose
118 the operations, constants, and arrangements of the variables.
120 This does not achieve avalanche. There are input bits of (a,b,c)
121 that fail to affect some output bits of (a,b,c), especially of a. The
122 most thoroughly mixed value is c, but it doesn't really even achieve
125 This allows some parallelism. Read-after-writes are good at doubling
126 the number of bits affected, so the goal of mixing pulls in the opposite
127 direction as the goal of parallelism. I did what I could. Rotates
128 seem to cost as much as shifts on every machine I could lay my hands
129 on, and rotates are much kinder to the top and bottom bits, so I used
131 -------------------------------------------------------------------------------
135 a -= c; a ^= rot(c, 4); c += b; \
136 b -= a; b ^= rot(a, 6); a += c; \
137 c -= b; c ^= rot(b, 8); b += a; \
138 a -= c; a ^= rot(c,16); c += b; \
139 b -= a; b ^= rot(a,19); a += c; \
140 c -= b; c ^= rot(b, 4); b += a; \
144 -------------------------------------------------------------------------------
145 final -- final mixing of 3 32-bit values (a,b,c) into c
147 Pairs of (a,b,c) values differing in only a few bits will usually
148 produce values of c that look totally different. This was tested for
149 * pairs that differed by one bit, by two bits, in any combination
150 of top bits of (a,b,c), or in any combination of bottom bits of
152 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
153 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
154 is commonly produced by subtraction) look like a single 1-bit
156 * the base values were pseudorandom, all zero but one bit set, or
157 all zero plus a counter that starts at zero.
159 These constants passed:
162 and these came close:
166 -------------------------------------------------------------------------------
168 #define final(a,b,c) \
170 c ^= b; c -= rot(b,14); \
171 a ^= c; a -= rot(c,11); \
172 b ^= a; b -= rot(a,25); \
173 c ^= b; c -= rot(b,16); \
174 a ^= c; a -= rot(c,4); \
175 b ^= a; b -= rot(a,14); \
176 c ^= b; c -= rot(b,24); \
181 -------------------------------------------------------------------------------
182 hashlittle() -- hash a variable-length key into a 32-bit value
183 k : the key (the unaligned variable-length array of bytes)
184 length : the length of the key, counting by bytes
185 val2 : IN: can be any 4-byte value OUT: second 32 bit hash.
186 Returns a 32-bit value. Every bit of the key affects every bit of
187 the return value. Two keys differing by one or two bits will have
188 totally different hash values. Note that the return value is better
189 mixed than val2, so use that first.
191 The best hash table sizes are powers of 2. There is no need to do
192 mod a prime (mod is sooo slow!). If you need less than 32 bits,
193 use a bitmask. For example, if you need only 10 bits, do
194 h = (h & hashmask(10));
195 In which case, the hash table should have hashsize(10) elements.
197 If you are hashing n strings (uint8_t **)k, do it like this:
198 for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
200 By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
201 code any way you wish, private, educational, or commercial. It's free.
203 Use for hash table lookup, or anything where one collision in 2^^32 is
204 acceptable. Do NOT use for cryptographic purposes.
205 -------------------------------------------------------------------------------
208 static uint32_t hashlittle( const void *key, size_t length )
210 uint32_t a,b,c; /* internal state */
211 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
213 /* Set up the internal state */
214 a = b = c = 0xdeadbeef + ((uint32_t)length);
217 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
218 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
223 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
234 /*----------------------------- handle the last (probably partial) block */
236 * "k[2]&0xffffff" actually reads beyond the end of the string, but
237 * then masks off the part it's not allowed to read. Because the
238 * string is aligned, the masked-off tail is in the same word as the
239 * rest of the string. Every machine with memory protection I've seen
240 * does it on word boundaries, so is OK with this. But VALGRIND will
241 * still catch it and complain. The masking trick does make the hash
242 * noticably faster for short strings (like English words).
248 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
249 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
250 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
251 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
252 case 8 : b+=k[1]; a+=k[0]; break;
253 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
254 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
255 case 5 : b+=k[1]&0xff; a+=k[0]; break;
256 case 4 : a+=k[0]; break;
257 case 3 : a+=k[0]&0xffffff; break;
258 case 2 : a+=k[0]&0xffff; break;
259 case 1 : a+=k[0]&0xff; break;
260 case 0 : return c; /* zero length strings require no mixing */
263 #else /* make valgrind happy */
265 k8 = (const uint8_t *)k;
268 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
269 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
270 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
271 case 9 : c+=k8[8]; /* fall through */
272 case 8 : b+=k[1]; a+=k[0]; break;
273 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
274 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
275 case 5 : b+=k8[4]; /* fall through */
276 case 4 : a+=k[0]; break;
277 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
278 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
279 case 1 : a+=k8[0]; break;
283 #endif /* !valgrind */
285 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
286 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
289 /*--------------- all but last block: aligned reads and different mixing */
292 a += k[0] + (((uint32_t)k[1])<<16);
293 b += k[2] + (((uint32_t)k[3])<<16);
294 c += k[4] + (((uint32_t)k[5])<<16);
300 /*----------------------------- handle the last (probably partial) block */
301 k8 = (const uint8_t *)k;
304 case 12: c+=k[4]+(((uint32_t)k[5])<<16);
305 b+=k[2]+(((uint32_t)k[3])<<16);
306 a+=k[0]+(((uint32_t)k[1])<<16);
308 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
310 b+=k[2]+(((uint32_t)k[3])<<16);
311 a+=k[0]+(((uint32_t)k[1])<<16);
313 case 9 : c+=k8[8]; /* fall through */
314 case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
315 a+=k[0]+(((uint32_t)k[1])<<16);
317 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
319 a+=k[0]+(((uint32_t)k[1])<<16);
321 case 5 : b+=k8[4]; /* fall through */
322 case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
324 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
329 case 0 : return c; /* zero length requires no mixing */
332 } else { /* need to read the key one byte at a time */
333 const uint8_t *k = (const uint8_t *)key;
335 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
339 a += ((uint32_t)k[1])<<8;
340 a += ((uint32_t)k[2])<<16;
341 a += ((uint32_t)k[3])<<24;
343 b += ((uint32_t)k[5])<<8;
344 b += ((uint32_t)k[6])<<16;
345 b += ((uint32_t)k[7])<<24;
347 c += ((uint32_t)k[9])<<8;
348 c += ((uint32_t)k[10])<<16;
349 c += ((uint32_t)k[11])<<24;
355 /*-------------------------------- last block: affect all 32 bits of (c) */
356 switch(length) /* all the case statements fall through */
358 case 12: c+=((uint32_t)k[11])<<24;
359 case 11: c+=((uint32_t)k[10])<<16;
360 case 10: c+=((uint32_t)k[9])<<8;
362 case 8 : b+=((uint32_t)k[7])<<24;
363 case 7 : b+=((uint32_t)k[6])<<16;
364 case 6 : b+=((uint32_t)k[5])<<8;
366 case 4 : a+=((uint32_t)k[3])<<24;
367 case 3 : a+=((uint32_t)k[2])<<16;
368 case 2 : a+=((uint32_t)k[1])<<8;
379 unsigned int tdb_jenkins_hash(TDB_DATA *key)
381 return hashlittle(key->dptr, key->dsize);