1 /* crypto/des/des_locl.org */
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 /* WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING
61 * Always modify des_locl.org since des_locl.h is automatically generated from
62 * it during SSLeay configuration.
64 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING
67 #ifndef HEADER_DES_LOCL_H
68 #define HEADER_DES_LOCL_H
70 #if defined(WIN32) || defined(WIN16)
83 #ifndef DES_DEFAULT_OPTIONS
84 /* the following is tweaked from a config script, that is why it is a
85 * protected undef/define */
90 /* This helps C compiler generate the correct code for multiple functional
91 * units. It reduces register dependancies at the expense of 2 more
101 #if defined(DES_RISC1) && defined(DES_RISC2)
102 YOU SHOULD NOT HAVE BOTH DES_RISC1 AND DES_RISC2 DEFINED!!!!!
105 /* Unroll the inner loop, this sometimes helps, sometimes hinders.
106 * Very mucy CPU dependant */
111 /* These default values were supplied by
112 * Peter Gutman <pgut001@cs.auckland.ac.nz>
113 * They are only used if nothing else has been defined */
114 #if !defined(DES_PTR) && !defined(DES_RISC1) && !defined(DES_RISC2) && !defined(DES_UNROLL)
115 /* Special defines which change the way the code is built depending on the
116 CPU and OS. For SGI machines you can use _MIPS_SZLONG (32 or 64) to find
117 even newer MIPS CPU's, but at the moment one size fits all for
118 optimization options. Older Sparc's work better with only UNROLL, but
119 there's no way to tell at compile time what it is you're running on */
121 #if defined( sun ) /* Newer Sparc's */
125 #elif defined( __ultrix ) /* Older MIPS */
129 #elif defined( __osf1__ ) /* Alpha */
132 #elif defined ( _AIX ) /* RS6000 */
134 #elif defined( __hpux ) /* HP-PA */
136 #elif defined( __aux ) /* 68K */
138 #elif defined( __dgux ) /* 88K (but P6 in latest boxes) */
140 #elif defined( __sgi ) /* Newer MIPS */
144 #elif defined( i386 ) /* x86 boxes, should be gcc */
148 #endif /* Systems-specific speed defines */
151 #endif /* DES_DEFAULT_OPTIONS */
153 #ifdef MSDOS /* Visual C++ 2.1 (Windows NT/95) */
164 #if defined(__STDC__) || defined(VMS) || defined(M_XENIX) || defined(MSDOS)
191 #define srandom(s) srand(s)
195 #define ITERATIONS 16
196 #define HALF_ITERATIONS 8
198 /* used in des_read and des_write */
199 #define MAXWRITE (1024*16)
200 #define BSIZE (MAXWRITE+4)
202 #define c2l(c,l) (l =((DES_LONG)(*((c)++))) , \
203 l|=((DES_LONG)(*((c)++)))<< 8L, \
204 l|=((DES_LONG)(*((c)++)))<<16L, \
205 l|=((DES_LONG)(*((c)++)))<<24L)
207 /* NOTE - c is not incremented as per c2l */
208 #define c2ln(c,l1,l2,n) { \
212 case 8: l2 =((DES_LONG)(*(--(c))))<<24L; \
213 case 7: l2|=((DES_LONG)(*(--(c))))<<16L; \
214 case 6: l2|=((DES_LONG)(*(--(c))))<< 8L; \
215 case 5: l2|=((DES_LONG)(*(--(c)))); \
216 case 4: l1 =((DES_LONG)(*(--(c))))<<24L; \
217 case 3: l1|=((DES_LONG)(*(--(c))))<<16L; \
218 case 2: l1|=((DES_LONG)(*(--(c))))<< 8L; \
219 case 1: l1|=((DES_LONG)(*(--(c)))); \
223 #define l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
224 *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
225 *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
226 *((c)++)=(unsigned char)(((l)>>24L)&0xff))
228 /* replacements for htonl and ntohl since I have no idea what to do
229 * when faced with machines with 8 byte longs. */
232 #define n2l(c,l) (l =((DES_LONG)(*((c)++)))<<24L, \
233 l|=((DES_LONG)(*((c)++)))<<16L, \
234 l|=((DES_LONG)(*((c)++)))<< 8L, \
235 l|=((DES_LONG)(*((c)++))))
237 #define l2n(l,c) (*((c)++)=(unsigned char)(((l)>>24L)&0xff), \
238 *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
239 *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
240 *((c)++)=(unsigned char)(((l) )&0xff))
242 /* NOTE - c is not incremented as per l2c */
243 #define l2cn(l1,l2,c,n) { \
246 case 8: *(--(c))=(unsigned char)(((l2)>>24L)&0xff); \
247 case 7: *(--(c))=(unsigned char)(((l2)>>16L)&0xff); \
248 case 6: *(--(c))=(unsigned char)(((l2)>> 8L)&0xff); \
249 case 5: *(--(c))=(unsigned char)(((l2) )&0xff); \
250 case 4: *(--(c))=(unsigned char)(((l1)>>24L)&0xff); \
251 case 3: *(--(c))=(unsigned char)(((l1)>>16L)&0xff); \
252 case 2: *(--(c))=(unsigned char)(((l1)>> 8L)&0xff); \
253 case 1: *(--(c))=(unsigned char)(((l1) )&0xff); \
258 #define ROTATE(a,n) (_lrotr(a,n))
260 #define ROTATE(a,n) (((a)>>(n))+((a)<<(32-(n))))
263 /* Don't worry about the LOAD_DATA() stuff, that is used by
264 * fcrypt() to add it's little bit to the front */
268 #define LOAD_DATA_tmp(R,S,u,t,E0,E1) \
269 { DES_LONG tmp; LOAD_DATA(R,S,u,t,E0,E1,tmp); }
271 #define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
274 tmp=(u<<16); u^=R^s[S ]; u^=tmp; \
275 tmp=(t<<16); t^=R^s[S+1]; t^=tmp
277 #define LOAD_DATA_tmp(a,b,c,d,e,f) LOAD_DATA(a,b,c,d,e,f,g)
278 #define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
283 /* The changes to this macro may help or hinder, depending on the
284 * compiler and the achitecture. gcc2 always seems to do well :-).
285 * Inspired by Dana How <how@isl.stanford.edu>
286 * DO NOT use the alternative version on machines with 8 byte longs.
287 * It does not seem to work on the Alpha, even when DES_LONG is 4
288 * bytes, probably an issue of accessing non-word aligned objects :-( */
291 /* It recently occured to me that 0^0^0^0^0^0^0 == 0, so there
292 * is no reason to not xor all the sub items together. This potentially
293 * saves a register since things can be xored directly into L */
295 #if defined(DES_RISC1) || defined(DES_RISC2)
297 #define D_ENCRYPT(LL,R,S) { \
298 unsigned int u1,u2,u3; \
299 LOAD_DATA(R,S,u,t,E0,E1,u1); \
305 LL^= *(DES_LONG *)((unsigned char *)des_SP +u1); \
306 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x200+u2); \
310 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x400+u1); \
311 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x600+u3); \
316 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x100+u1); \
317 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x300+u2); \
321 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x500+u1); \
322 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x700+u3); }
325 #define D_ENCRYPT(LL,R,S) { \
326 unsigned int u1,u2,s1,s2; \
327 LOAD_DATA(R,S,u,t,E0,E1,u1); \
332 LL^= *(DES_LONG *)((unsigned char *)des_SP +u1); \
333 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x200+u2); \
338 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x400+s1); \
339 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x600+s2); \
343 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x100+u1); \
344 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x300+u2); \
349 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x500+s1); \
350 LL^= *(DES_LONG *)((unsigned char *)des_SP+0x700+s2); }
353 #define D_ENCRYPT(LL,R,S) { \
354 LOAD_DATA_tmp(R,S,u,t,E0,E1); \
357 *(DES_LONG *)((unsigned char *)des_SP +((u )&0xfc))^ \
358 *(DES_LONG *)((unsigned char *)des_SP+0x200+((u>> 8L)&0xfc))^ \
359 *(DES_LONG *)((unsigned char *)des_SP+0x400+((u>>16L)&0xfc))^ \
360 *(DES_LONG *)((unsigned char *)des_SP+0x600+((u>>24L)&0xfc))^ \
361 *(DES_LONG *)((unsigned char *)des_SP+0x100+((t )&0xfc))^ \
362 *(DES_LONG *)((unsigned char *)des_SP+0x300+((t>> 8L)&0xfc))^ \
363 *(DES_LONG *)((unsigned char *)des_SP+0x500+((t>>16L)&0xfc))^ \
364 *(DES_LONG *)((unsigned char *)des_SP+0x700+((t>>24L)&0xfc)); }
367 #else /* original version */
369 #if defined(DES_RISC1) || defined(DES_RISC2)
371 #define D_ENCRYPT(LL,R,S) {\
372 unsigned int u1,u2,u3; \
373 LOAD_DATA(R,S,u,t,E0,E1,u1); \
380 LL^=des_SPtrans[0][u1]; \
381 LL^=des_SPtrans[2][u2]; \
385 LL^=des_SPtrans[4][u1]; \
386 LL^=des_SPtrans[6][u3]; \
391 LL^=des_SPtrans[1][u1]; \
392 LL^=des_SPtrans[3][u2]; \
396 LL^=des_SPtrans[5][u1]; \
397 LL^=des_SPtrans[7][u3]; }
400 #define D_ENCRYPT(LL,R,S) {\
401 unsigned int u1,u2,s1,s2; \
402 LOAD_DATA(R,S,u,t,E0,E1,u1); \
408 LL^=des_SPtrans[0][u1]; \
409 LL^=des_SPtrans[2][u2]; \
414 LL^=des_SPtrans[4][s1]; \
415 LL^=des_SPtrans[6][s2]; \
419 LL^=des_SPtrans[1][u1]; \
420 LL^=des_SPtrans[3][u2]; \
425 LL^=des_SPtrans[5][s1]; \
426 LL^=des_SPtrans[7][s2]; }
431 #define D_ENCRYPT(LL,R,S) {\
432 LOAD_DATA_tmp(R,S,u,t,E0,E1); \
435 des_SPtrans[0][(u>> 2L)&0x3f]^ \
436 des_SPtrans[2][(u>>10L)&0x3f]^ \
437 des_SPtrans[4][(u>>18L)&0x3f]^ \
438 des_SPtrans[6][(u>>26L)&0x3f]^ \
439 des_SPtrans[1][(t>> 2L)&0x3f]^ \
440 des_SPtrans[3][(t>>10L)&0x3f]^ \
441 des_SPtrans[5][(t>>18L)&0x3f]^ \
442 des_SPtrans[7][(t>>26L)&0x3f]; }
447 * The problem is more of a geometric problem that random bit fiddling.
448 0 1 2 3 4 5 6 7 62 54 46 38 30 22 14 6
449 8 9 10 11 12 13 14 15 60 52 44 36 28 20 12 4
450 16 17 18 19 20 21 22 23 58 50 42 34 26 18 10 2
451 24 25 26 27 28 29 30 31 to 56 48 40 32 24 16 8 0
453 32 33 34 35 36 37 38 39 63 55 47 39 31 23 15 7
454 40 41 42 43 44 45 46 47 61 53 45 37 29 21 13 5
455 48 49 50 51 52 53 54 55 59 51 43 35 27 19 11 3
456 56 57 58 59 60 61 62 63 57 49 41 33 25 17 9 1
458 The output has been subject to swaps of the form
459 0 1 -> 3 1 but the odd and even bits have been put into
461 different words. The main trick is to remember that
462 t=((l>>size)^r)&(mask);
465 can be used to swap and move bits between words.
467 So l = 0 1 2 3 r = 16 17 18 19
469 8 9 10 11 24 25 26 27
470 12 13 14 15 28 29 30 31
471 becomes (for size == 2 and mask == 0x3333)
472 t = 2^16 3^17 -- -- l = 0 1 16 17 r = 2 3 18 19
473 6^20 7^21 -- -- 4 5 20 21 6 7 22 23
474 10^24 11^25 -- -- 8 9 24 25 10 11 24 25
475 14^28 15^29 -- -- 12 13 28 29 14 15 28 29
477 Thanks for hints from Richard Outerbridge - he told me IP&FP
478 could be done in 15 xor, 10 shifts and 5 ands.
479 When I finally started to think of the problem in 2D
480 I first got ~42 operations without xors. When I remembered
481 how to use xors :-) I got it to its final state.
483 #define PERM_OP(a,b,t,n,m) ((t)=((((a)>>(n))^(b))&(m)),\
489 register DES_LONG tt; \
490 PERM_OP(r,l,tt, 4,0x0f0f0f0fL); \
491 PERM_OP(l,r,tt,16,0x0000ffffL); \
492 PERM_OP(r,l,tt, 2,0x33333333L); \
493 PERM_OP(l,r,tt, 8,0x00ff00ffL); \
494 PERM_OP(r,l,tt, 1,0x55555555L); \
499 register DES_LONG tt; \
500 PERM_OP(l,r,tt, 1,0x55555555L); \
501 PERM_OP(r,l,tt, 8,0x00ff00ffL); \
502 PERM_OP(l,r,tt, 2,0x33333333L); \
503 PERM_OP(r,l,tt,16,0x0000ffffL); \
504 PERM_OP(l,r,tt, 4,0x0f0f0f0fL); \
507 extern const DES_LONG des_SPtrans[8][64];
510 void fcrypt_body(DES_LONG *out,des_key_schedule ks,
511 DES_LONG Eswap0, DES_LONG Eswap1);