1 /* $OpenBSD: des_locl.h,v 1.19 2016/12/21 15:49:29 jsing Exp $ */
2 /* Copyright (C) 1995-1997 Eric Young (eay@cryptsoft.com)
5 * This package is an SSL implementation written
6 * by Eric Young (eay@cryptsoft.com).
7 * The implementation was written so as to conform with Netscapes SSL.
9 * This library is free for commercial and non-commercial use as long as
10 * the following conditions are aheared to. The following conditions
11 * apply to all code found in this distribution, be it the RC4, RSA,
12 * lhash, DES, etc., code; not just the SSL code. The SSL documentation
13 * included with this distribution is covered by the same copyright terms
14 * except that the holder is Tim Hudson (tjh@cryptsoft.com).
16 * Copyright remains Eric Young's, and as such any Copyright notices in
17 * the code are not to be removed.
18 * If this package is used in a product, Eric Young should be given attribution
19 * as the author of the parts of the library used.
20 * This can be in the form of a textual message at program startup or
21 * in documentation (online or textual) provided with the package.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 3. All advertising materials mentioning features or use of this software
32 * must display the following acknowledgement:
33 * "This product includes cryptographic software written by
34 * Eric Young (eay@cryptsoft.com)"
35 * The word 'cryptographic' can be left out if the rouines from the library
36 * being used are not cryptographic related :-).
37 * 4. If you include any Windows specific code (or a derivative thereof) from
38 * the apps directory (application code) you must include an acknowledgement:
39 * "This product includes software written by Tim Hudson (tjh@cryptsoft.com)"
41 * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND
42 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
43 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
44 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
45 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
46 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
47 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
48 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
49 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
50 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53 * The licence and distribution terms for any publically available version or
54 * derivative of this code cannot be changed. i.e. this code cannot simply be
55 * copied and put under another distribution licence
56 * [including the GNU Public Licence.]
59 #ifndef HEADER_DES_LOCL_H
60 #define HEADER_DES_LOCL_H
69 #include <openssl/opensslconf.h>
71 #include <openssl/des.h>
76 #define HALF_ITERATIONS 8
78 /* used in des_read and des_write */
79 #define MAXWRITE (1024*16)
80 #define BSIZE (MAXWRITE+4)
82 #define c2l(c,l) (l =((DES_LONG)(*((c)++))) , \
83 l|=((DES_LONG)(*((c)++)))<< 8L, \
84 l|=((DES_LONG)(*((c)++)))<<16L, \
85 l|=((DES_LONG)(*((c)++)))<<24L)
87 /* NOTE - c is not incremented as per c2l */
88 #define c2ln(c,l1,l2,n) { \
92 case 8: l2 =((DES_LONG)(*(--(c))))<<24L; \
93 case 7: l2|=((DES_LONG)(*(--(c))))<<16L; \
94 case 6: l2|=((DES_LONG)(*(--(c))))<< 8L; \
95 case 5: l2|=((DES_LONG)(*(--(c)))); \
96 case 4: l1 =((DES_LONG)(*(--(c))))<<24L; \
97 case 3: l1|=((DES_LONG)(*(--(c))))<<16L; \
98 case 2: l1|=((DES_LONG)(*(--(c))))<< 8L; \
99 case 1: l1|=((DES_LONG)(*(--(c)))); \
103 #define l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
104 *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
105 *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
106 *((c)++)=(unsigned char)(((l)>>24L)&0xff))
108 /* replacements for htonl and ntohl since I have no idea what to do
109 * when faced with machines with 8 byte longs. */
112 #define n2l(c,l) (l =((DES_LONG)(*((c)++)))<<24L, \
113 l|=((DES_LONG)(*((c)++)))<<16L, \
114 l|=((DES_LONG)(*((c)++)))<< 8L, \
115 l|=((DES_LONG)(*((c)++))))
117 #define l2n(l,c) (*((c)++)=(unsigned char)(((l)>>24L)&0xff), \
118 *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
119 *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
120 *((c)++)=(unsigned char)(((l) )&0xff))
122 /* NOTE - c is not incremented as per l2c */
123 #define l2cn(l1,l2,c,n) { \
126 case 8: *(--(c))=(unsigned char)(((l2)>>24L)&0xff); \
127 case 7: *(--(c))=(unsigned char)(((l2)>>16L)&0xff); \
128 case 6: *(--(c))=(unsigned char)(((l2)>> 8L)&0xff); \
129 case 5: *(--(c))=(unsigned char)(((l2) )&0xff); \
130 case 4: *(--(c))=(unsigned char)(((l1)>>24L)&0xff); \
131 case 3: *(--(c))=(unsigned char)(((l1)>>16L)&0xff); \
132 case 2: *(--(c))=(unsigned char)(((l1)>> 8L)&0xff); \
133 case 1: *(--(c))=(unsigned char)(((l1) )&0xff); \
137 static inline uint32_t ROTATE(uint32_t a
, uint32_t n
)
139 return (a
>>n
)+(a
<<(32-n
));
142 /* Don't worry about the LOAD_DATA() stuff, that is used by
143 * fcrypt() to add it's little bit to the front */
147 #define LOAD_DATA_tmp(R,S,u,t,E0,E1) \
148 { DES_LONG tmp; LOAD_DATA(R,S,u,t,E0,E1,tmp); }
150 #define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
153 tmp=(u<<16); u^=R^s[S ]; u^=tmp; \
154 tmp=(t<<16); t^=R^s[S+1]; t^=tmp
156 #define LOAD_DATA_tmp(a,b,c,d,e,f) LOAD_DATA(a,b,c,d,e,f,g)
157 #define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
162 /* The changes to this macro may help or hinder, depending on the
163 * compiler and the architecture. gcc2 always seems to do well :-).
164 * Inspired by Dana How <how@isl.stanford.edu>
165 * DO NOT use the alternative version on machines with 8 byte longs.
166 * It does not seem to work on the Alpha, even when DES_LONG is 4
167 * bytes, probably an issue of accessing non-word aligned objects :-( */
170 /* It recently occurred to me that 0^0^0^0^0^0^0 == 0, so there
171 * is no reason to not xor all the sub items together. This potentially
172 * saves a register since things can be xored directly into L */
174 #if defined(DES_RISC1) || defined(DES_RISC2)
176 #define D_ENCRYPT(LL,R,S) { \
177 unsigned int u1,u2,u3; \
178 LOAD_DATA(R,S,u,t,E0,E1,u1); \
184 LL^= *(const DES_LONG *)(des_SP +u1); \
185 LL^= *(const DES_LONG *)(des_SP+0x200+u2); \
189 LL^= *(const DES_LONG *)(des_SP+0x400+u1); \
190 LL^= *(const DES_LONG *)(des_SP+0x600+u3); \
195 LL^= *(const DES_LONG *)(des_SP+0x100+u1); \
196 LL^= *(const DES_LONG *)(des_SP+0x300+u2); \
200 LL^= *(const DES_LONG *)(des_SP+0x500+u1); \
201 LL^= *(const DES_LONG *)(des_SP+0x700+u3); }
204 #define D_ENCRYPT(LL,R,S) { \
205 unsigned int u1,u2,s1,s2; \
206 LOAD_DATA(R,S,u,t,E0,E1,u1); \
211 LL^= *(const DES_LONG *)(des_SP +u1); \
212 LL^= *(const DES_LONG *)(des_SP+0x200+u2); \
217 LL^= *(const DES_LONG *)(des_SP+0x400+s1); \
218 LL^= *(const DES_LONG *)(des_SP+0x600+s2); \
222 LL^= *(const DES_LONG *)(des_SP+0x100+u1); \
223 LL^= *(const DES_LONG *)(des_SP+0x300+u2); \
228 LL^= *(const DES_LONG *)(des_SP+0x500+s1); \
229 LL^= *(const DES_LONG *)(des_SP+0x700+s2); }
232 #define D_ENCRYPT(LL,R,S) { \
233 LOAD_DATA_tmp(R,S,u,t,E0,E1); \
236 *(const DES_LONG *)(des_SP +((u )&0xfc))^ \
237 *(const DES_LONG *)(des_SP+0x200+((u>> 8L)&0xfc))^ \
238 *(const DES_LONG *)(des_SP+0x400+((u>>16L)&0xfc))^ \
239 *(const DES_LONG *)(des_SP+0x600+((u>>24L)&0xfc))^ \
240 *(const DES_LONG *)(des_SP+0x100+((t )&0xfc))^ \
241 *(const DES_LONG *)(des_SP+0x300+((t>> 8L)&0xfc))^ \
242 *(const DES_LONG *)(des_SP+0x500+((t>>16L)&0xfc))^ \
243 *(const DES_LONG *)(des_SP+0x700+((t>>24L)&0xfc)); }
246 #else /* original version */
248 #if defined(DES_RISC1) || defined(DES_RISC2)
250 #define D_ENCRYPT(LL,R,S) {\
251 unsigned int u1,u2,u3; \
252 LOAD_DATA(R,S,u,t,E0,E1,u1); \
259 LL^=DES_SPtrans[0][u1]; \
260 LL^=DES_SPtrans[2][u2]; \
264 LL^=DES_SPtrans[4][u1]; \
265 LL^=DES_SPtrans[6][u3]; \
270 LL^=DES_SPtrans[1][u1]; \
271 LL^=DES_SPtrans[3][u2]; \
275 LL^=DES_SPtrans[5][u1]; \
276 LL^=DES_SPtrans[7][u3]; }
279 #define D_ENCRYPT(LL,R,S) {\
280 unsigned int u1,u2,s1,s2; \
281 LOAD_DATA(R,S,u,t,E0,E1,u1); \
287 LL^=DES_SPtrans[0][u1]; \
288 LL^=DES_SPtrans[2][u2]; \
293 LL^=DES_SPtrans[4][s1]; \
294 LL^=DES_SPtrans[6][s2]; \
298 LL^=DES_SPtrans[1][u1]; \
299 LL^=DES_SPtrans[3][u2]; \
304 LL^=DES_SPtrans[5][s1]; \
305 LL^=DES_SPtrans[7][s2]; }
310 #define D_ENCRYPT(LL,R,S) {\
311 LOAD_DATA_tmp(R,S,u,t,E0,E1); \
314 DES_SPtrans[0][(u>> 2L)&0x3f]^ \
315 DES_SPtrans[2][(u>>10L)&0x3f]^ \
316 DES_SPtrans[4][(u>>18L)&0x3f]^ \
317 DES_SPtrans[6][(u>>26L)&0x3f]^ \
318 DES_SPtrans[1][(t>> 2L)&0x3f]^ \
319 DES_SPtrans[3][(t>>10L)&0x3f]^ \
320 DES_SPtrans[5][(t>>18L)&0x3f]^ \
321 DES_SPtrans[7][(t>>26L)&0x3f]; }
326 * The problem is more of a geometric problem that random bit fiddling.
327 0 1 2 3 4 5 6 7 62 54 46 38 30 22 14 6
328 8 9 10 11 12 13 14 15 60 52 44 36 28 20 12 4
329 16 17 18 19 20 21 22 23 58 50 42 34 26 18 10 2
330 24 25 26 27 28 29 30 31 to 56 48 40 32 24 16 8 0
332 32 33 34 35 36 37 38 39 63 55 47 39 31 23 15 7
333 40 41 42 43 44 45 46 47 61 53 45 37 29 21 13 5
334 48 49 50 51 52 53 54 55 59 51 43 35 27 19 11 3
335 56 57 58 59 60 61 62 63 57 49 41 33 25 17 9 1
337 The output has been subject to swaps of the form
338 0 1 -> 3 1 but the odd and even bits have been put into
340 different words. The main trick is to remember that
341 t=((l>>size)^r)&(mask);
344 can be used to swap and move bits between words.
346 So l = 0 1 2 3 r = 16 17 18 19
348 8 9 10 11 24 25 26 27
349 12 13 14 15 28 29 30 31
350 becomes (for size == 2 and mask == 0x3333)
351 t = 2^16 3^17 -- -- l = 0 1 16 17 r = 2 3 18 19
352 6^20 7^21 -- -- 4 5 20 21 6 7 22 23
353 10^24 11^25 -- -- 8 9 24 25 10 11 24 25
354 14^28 15^29 -- -- 12 13 28 29 14 15 28 29
356 Thanks for hints from Richard Outerbridge - he told me IP&FP
357 could be done in 15 xor, 10 shifts and 5 ands.
358 When I finally started to think of the problem in 2D
359 I first got ~42 operations without xors. When I remembered
360 how to use xors :-) I got it to its final state.
362 #define PERM_OP(a,b,t,n,m) ((t)=((((a)>>(n))^(b))&(m)),\
369 PERM_OP(r,l,tt, 4,0x0f0f0f0fL); \
370 PERM_OP(l,r,tt,16,0x0000ffffL); \
371 PERM_OP(r,l,tt, 2,0x33333333L); \
372 PERM_OP(l,r,tt, 8,0x00ff00ffL); \
373 PERM_OP(r,l,tt, 1,0x55555555L); \
379 PERM_OP(l,r,tt, 1,0x55555555L); \
380 PERM_OP(r,l,tt, 8,0x00ff00ffL); \
381 PERM_OP(l,r,tt, 2,0x33333333L); \
382 PERM_OP(r,l,tt,16,0x0000ffffL); \
383 PERM_OP(l,r,tt, 4,0x0f0f0f0fL); \
386 extern const DES_LONG DES_SPtrans
[8][64];
388 void fcrypt_body(DES_LONG
*out
,DES_key_schedule
*ks
,
389 DES_LONG Eswap0
, DES_LONG Eswap1
);
391 #ifdef OPENSSL_SMALL_FOOTPRINT