2 * Copyright 1995-2023 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the Apache License 2.0 (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
10 #ifndef OSSL_CRYPTO_DES_LOCAL_H
11 # define OSSL_CRYPTO_DES_LOCAL_H
13 # include <openssl/e_os2.h>
19 # include <openssl/des.h>
21 # ifdef OPENSSL_BUILD_SHLIBCRYPTO
22 # undef OPENSSL_EXTERN
23 # define OPENSSL_EXTERN OPENSSL_EXPORT
26 # define ITERATIONS 16
27 # define HALF_ITERATIONS 8
29 # define c2l(c,l) (l =((DES_LONG)(*((c)++))) , \
30 l|=((DES_LONG)(*((c)++)))<< 8L, \
31 l|=((DES_LONG)(*((c)++)))<<16L, \
32 l|=((DES_LONG)(*((c)++)))<<24L)
34 /* NOTE - c is not incremented as per c2l */
35 # define c2ln(c,l1,l2,n) { \
39 case 8: l2 =((DES_LONG)(*(--(c))))<<24L; \
41 case 7: l2|=((DES_LONG)(*(--(c))))<<16L; \
43 case 6: l2|=((DES_LONG)(*(--(c))))<< 8L; \
45 case 5: l2|=((DES_LONG)(*(--(c)))); \
47 case 4: l1 =((DES_LONG)(*(--(c))))<<24L; \
49 case 3: l1|=((DES_LONG)(*(--(c))))<<16L; \
51 case 2: l1|=((DES_LONG)(*(--(c))))<< 8L; \
53 case 1: l1|=((DES_LONG)(*(--(c)))); \
57 # define l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
58 *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
59 *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
60 *((c)++)=(unsigned char)(((l)>>24L)&0xff))
62 /* NOTE - c is not incremented as per l2c */
63 # define l2cn(l1,l2,c,n) { \
66 case 8: *(--(c))=(unsigned char)(((l2)>>24L)&0xff); \
68 case 7: *(--(c))=(unsigned char)(((l2)>>16L)&0xff); \
70 case 6: *(--(c))=(unsigned char)(((l2)>> 8L)&0xff); \
72 case 5: *(--(c))=(unsigned char)(((l2) )&0xff); \
74 case 4: *(--(c))=(unsigned char)(((l1)>>24L)&0xff); \
76 case 3: *(--(c))=(unsigned char)(((l1)>>16L)&0xff); \
78 case 2: *(--(c))=(unsigned char)(((l1)>> 8L)&0xff); \
80 case 1: *(--(c))=(unsigned char)(((l1) )&0xff); \
84 # if defined(_MSC_VER)
85 # define ROTATE(a,n) (_lrotr(a,n))
87 # define ROTATE(a,n) (_rotr(a,n))
88 # elif defined(__GNUC__) && __GNUC__>=2 && !defined(__STRICT_ANSI__) && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC)
89 # if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
90 # define ROTATE(a,n) ({ register unsigned int ret; \
97 # elif defined(__riscv_zbb) || defined(__riscv_zbkb)
98 # if __riscv_xlen == 64
99 # define ROTATE(x, n) ({ register unsigned int ret; \
100 asm ("roriw %0, %1, %2" \
102 : "r"(x), "i"(n)); ret; })
104 # if __riscv_xlen == 32
105 # define ROTATE(x, n) ({ register unsigned int ret; \
106 asm ("rori %0, %1, %2" \
108 : "r"(x), "i"(n)); ret; })
113 # define ROTATE(a,n) (((a)>>(n))+((a)<<(32-(n))))
117 * Don't worry about the LOAD_DATA() stuff, that is used by fcrypt() to add
118 * it's little bit to the front
123 # define LOAD_DATA_tmp(R,S,u,t,E0,E1) \
124 { DES_LONG tmp; LOAD_DATA(R,S,u,t,E0,E1,tmp); }
126 # define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
129 tmp=(u<<16); u^=R^s[S ]; u^=tmp; \
130 tmp=(t<<16); t^=R^s[S+1]; t^=tmp
132 # define LOAD_DATA_tmp(a,b,c,d,e,f) LOAD_DATA(a,b,c,d,e,f,g)
133 # define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
139 * It recently occurred to me that 0^0^0^0^0^0^0 == 0, so there is no reason
140 * to not xor all the sub items together. This potentially saves a register
141 * since things can be xored directly into L
144 # define D_ENCRYPT(LL,R,S) { \
145 LOAD_DATA_tmp(R,S,u,t,E0,E1); \
148 DES_SPtrans[0][(u>> 2L)&0x3f]^ \
149 DES_SPtrans[2][(u>>10L)&0x3f]^ \
150 DES_SPtrans[4][(u>>18L)&0x3f]^ \
151 DES_SPtrans[6][(u>>26L)&0x3f]^ \
152 DES_SPtrans[1][(t>> 2L)&0x3f]^ \
153 DES_SPtrans[3][(t>>10L)&0x3f]^ \
154 DES_SPtrans[5][(t>>18L)&0x3f]^ \
155 DES_SPtrans[7][(t>>26L)&0x3f]; }
159 * The problem is more of a geometric problem that random bit fiddling.
160 0 1 2 3 4 5 6 7 62 54 46 38 30 22 14 6
161 8 9 10 11 12 13 14 15 60 52 44 36 28 20 12 4
162 16 17 18 19 20 21 22 23 58 50 42 34 26 18 10 2
163 24 25 26 27 28 29 30 31 to 56 48 40 32 24 16 8 0
165 32 33 34 35 36 37 38 39 63 55 47 39 31 23 15 7
166 40 41 42 43 44 45 46 47 61 53 45 37 29 21 13 5
167 48 49 50 51 52 53 54 55 59 51 43 35 27 19 11 3
168 56 57 58 59 60 61 62 63 57 49 41 33 25 17 9 1
170 The output has been subject to swaps of the form
171 0 1 -> 3 1 but the odd and even bits have been put into
173 different words. The main trick is to remember that
174 t=((l>>size)^r)&(mask);
177 can be used to swap and move bits between words.
179 So l = 0 1 2 3 r = 16 17 18 19
181 8 9 10 11 24 25 26 27
182 12 13 14 15 28 29 30 31
183 becomes (for size == 2 and mask == 0x3333)
184 t = 2^16 3^17 -- -- l = 0 1 16 17 r = 2 3 18 19
185 6^20 7^21 -- -- 4 5 20 21 6 7 22 23
186 10^24 11^25 -- -- 8 9 24 25 10 11 24 25
187 14^28 15^29 -- -- 12 13 28 29 14 15 28 29
189 Thanks for hints from Richard Outerbridge - he told me IP&FP
190 could be done in 15 xor, 10 shifts and 5 ands.
191 When I finally started to think of the problem in 2D
192 I first got ~42 operations without xors. When I remembered
193 how to use xors :-) I got it to its final state.
195 # define PERM_OP(a,b,t,n,m) ((t)=((((a)>>(n))^(b))&(m)),\
201 register DES_LONG tt; \
202 PERM_OP(r,l,tt, 4,0x0f0f0f0fL); \
203 PERM_OP(l,r,tt,16,0x0000ffffL); \
204 PERM_OP(r,l,tt, 2,0x33333333L); \
205 PERM_OP(l,r,tt, 8,0x00ff00ffL); \
206 PERM_OP(r,l,tt, 1,0x55555555L); \
211 register DES_LONG tt; \
212 PERM_OP(l,r,tt, 1,0x55555555L); \
213 PERM_OP(r,l,tt, 8,0x00ff00ffL); \
214 PERM_OP(l,r,tt, 2,0x33333333L); \
215 PERM_OP(r,l,tt,16,0x0000ffffL); \
216 PERM_OP(l,r,tt, 4,0x0f0f0f0fL); \
219 extern const DES_LONG DES_SPtrans[8][64];
221 void fcrypt_body(DES_LONG *out, DES_key_schedule *ks,
222 DES_LONG Eswap0, DES_LONG Eswap1);