des_locl.h

This a good VPN source

#define	ROTATE(a,n)	(((a)>>(n))+((a)<<(32-(n))))
#endif

/* Don't worry about the LOAD_DATA() stuff, that is used by
 * fcrypt() to add it's little bit to the front */

#ifdef DES_FCRYPT

#define LOAD_DATA_tmp(R,S,u,t,E0,E1) \
	{ DES_LONG tmp; LOAD_DATA(R,S,u,t,E0,E1,tmp); }

#define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
	t=R^(R>>16L); \
	u=t&E0; t&=E1; \
	tmp=(u<<16); u^=R^s[S  ]; u^=tmp; \
	tmp=(t<<16); t^=R^s[S+1]; t^=tmp
#else
#define LOAD_DATA_tmp(a,b,c,d,e,f) LOAD_DATA(a,b,c,d,e,f,g)
#define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
	u=R^s[S  ]; \
	t=R^s[S+1]
#endif

/* The changes to this macro may help or hinder, depending on the
 * compiler and the achitecture.  gcc2 always seems to do well :-).
 * Inspired by Dana How <[e-mail]>
 * DO NOT use the alternative version on machines with 8 byte longs.
 * It does not seem to work on the Alpha, even when DES_LONG is 4
 * bytes, probably an issue of accessing non-word aligned objects :-( */
#ifdef DES_PTR

/* It recently occured to me that 0^0^0^0^0^0^0 == 0, so there
 * is no reason to not xor all the sub items together.  This potentially
 * saves a register since things can be xored directly into L */

#if defined(DES_RISC1) || defined(DES_RISC2)
#ifdef DES_RISC1
#define D_ENCRYPT(LL,R,S) { \
	unsigned int u1,u2,u3; \
	LOAD_DATA(R,S,u,t,E0,E1,u1); \
	u2=(int)u>>8L; \
	u1=(int)u&0xfc; \
	u2&=0xfc; \
	t=ROTATE(t,4); \
	u>>=16L; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP      +u1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x200+u2); \
	u3=(int)(u>>8L); \
	u1=(int)u&0xfc; \
	u3&=0xfc; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x400+u1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x600+u3); \
	u2=(int)t>>8L; \
	u1=(int)t&0xfc; \
	u2&=0xfc; \
	t>>=16L; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x100+u1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x300+u2); \
	u3=(int)t>>8L; \
	u1=(int)t&0xfc; \
	u3&=0xfc; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x500+u1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x700+u3); }
#endif
#ifdef DES_RISC2
#define D_ENCRYPT(LL,R,S) { \
	unsigned int u1,u2,s1,s2; \
	LOAD_DATA(R,S,u,t,E0,E1,u1); \
	u2=(int)u>>8L; \
	u1=(int)u&0xfc; \
	u2&=0xfc; \
	t=ROTATE(t,4); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP      +u1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x200+u2); \
	s1=(int)(u>>16L); \
	s2=(int)(u>>24L); \
	s1&=0xfc; \
	s2&=0xfc; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x400+s1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x600+s2); \
	u2=(int)t>>8L; \
	u1=(int)t&0xfc; \
	u2&=0xfc; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x100+u1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x300+u2); \
	s1=(int)(t>>16L); \
	s2=(int)(t>>24L); \
	s1&=0xfc; \
	s2&=0xfc; \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x500+s1); \
	LL^= *(DES_LONG *)((unsigned char *)des_SP+0x700+s2); }
#endif
#else
#define D_ENCRYPT(LL,R,S) { \
	LOAD_DATA_tmp(R,S,u,t,E0,E1); \
	t=ROTATE(t,4); \
	LL^= \
	*(DES_LONG *)((unsigned char *)des_SP      +((u     )&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x200+((u>> 8L)&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x400+((u>>16L)&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x600+((u>>24L)&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x100+((t     )&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x300+((t>> 8L)&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x500+((t>>16L)&0xfc))^ \
	*(DES_LONG *)((unsigned char *)des_SP+0x700+((t>>24L)&0xfc)); }
#endif

#else /* original version */

#if defined(DES_RISC1) || defined(DES_RISC2)
#ifdef DES_RISC1
#define D_ENCRYPT(LL,R,S) {\
	unsigned int u1,u2,u3; \
	LOAD_DATA(R,S,u,t,E0,E1,u1); \
	u>>=2L; \
	t=ROTATE(t,6); \
	u2=(int)u>>8L; \
	u1=(int)u&0x3f; \
	u2&=0x3f; \
	u>>=16L; \
	LL^=des_SPtrans[0][u1]; \
	LL^=des_SPtrans[2][u2]; \
	u3=(int)u>>8L; \
	u1=(int)u&0x3f; \
	u3&=0x3f; \
	LL^=des_SPtrans[4][u1]; \
	LL^=des_SPtrans[6][u3]; \
	u2=(int)t>>8L; \
	u1=(int)t&0x3f; \
	u2&=0x3f; \
	t>>=16L; \
	LL^=des_SPtrans[1][u1]; \
	LL^=des_SPtrans[3][u2]; \
	u3=(int)t>>8L; \
	u1=(int)t&0x3f; \
	u3&=0x3f; \
	LL^=des_SPtrans[5][u1]; \
	LL^=des_SPtrans[7][u3]; }
#endif
#ifdef DES_RISC2
#define D_ENCRYPT(LL,R,S) {\
	unsigned int u1,u2,s1,s2; \
	LOAD_DATA(R,S,u,t,E0,E1,u1); \
	u>>=2L; \
	t=ROTATE(t,6); \
	u2=(int)u>>8L; \
	u1=(int)u&0x3f; \
	u2&=0x3f; \
	LL^=des_SPtrans[0][u1]; \
	LL^=des_SPtrans[2][u2]; \
	s1=(int)u>>16L; \
	s2=(int)u>>24L; \
	s1&=0x3f; \
	s2&=0x3f; \
	LL^=des_SPtrans[4][s1]; \
	LL^=des_SPtrans[6][s2]; \
	u2=(int)t>>8L; \
	u1=(int)t&0x3f; \
	u2&=0x3f; \
	LL^=des_SPtrans[1][u1]; \
	LL^=des_SPtrans[3][u2]; \
	s1=(int)t>>16; \
	s2=(int)t>>24L; \
	s1&=0x3f; \
	s2&=0x3f; \
	LL^=des_SPtrans[5][s1]; \
	LL^=des_SPtrans[7][s2]; }
#endif

#else

#define D_ENCRYPT(LL,R,S) {\
	LOAD_DATA_tmp(R,S,u,t,E0,E1); \
	t=ROTATE(t,4); \
	LL^=\
		des_SPtrans[0][(u>> 2L)&0x3f]^ \
		des_SPtrans[2][(u>>10L)&0x3f]^ \
		des_SPtrans[4][(u>>18L)&0x3f]^ \
		des_SPtrans[6][(u>>26L)&0x3f]^ \
		des_SPtrans[1][(t>> 2L)&0x3f]^ \
		des_SPtrans[3][(t>>10L)&0x3f]^ \
		des_SPtrans[5][(t>>18L)&0x3f]^ \
		des_SPtrans[7][(t>>26L)&0x3f]; }
#endif
#endif

	/* IP and FP
	 * The problem is more of a geometric problem that random bit fiddling.
	 0  1  2  3  4  5  6  7      62 54 46 38 30 22 14  6
	 8  9 10 11 12 13 14 15      60 52 44 36 28 20 12  4
	16 17 18 19 20 21 22 23      58 50 42 34 26 18 10  2
	24 25 26 27 28 29 30 31  to  56 48 40 32 24 16  8  0

	32 33 34 35 36 37 38 39      63 55 47 39 31 23 15  7
	40 41 42 43 44 45 46 47      61 53 45 37 29 21 13  5
	48 49 50 51 52 53 54 55      59 51 43 35 27 19 11  3
	56 57 58 59 60 61 62 63      57 49 41 33 25 17  9  1

	The output has been subject to swaps of the form
	0 1 -> 3 1 but the odd and even bits have been put into
	2 3    2 0
	different words.  The main trick is to remember that
	t=((l>>size)^r)&(mask);
	r^=t;
	l^=(t<<size);
	can be used to swap and move bits between words.

	So l =  0  1  2  3  r = 16 17 18 19
	        4  5  6  7      20 21 22 23
	        8  9 10 11      24 25 26 27
	       12 13 14 15      28 29 30 31
	becomes (for size == 2 and mask == 0x3333)
	   t =   2^16  3^17 -- --   l =  0  1 16 17  r =  2  3 18 19
		 6^20  7^21 -- --        4  5 20 21       6  7 22 23
		10^24 11^25 -- --        8  9 24 25      10 11 24 25
		14^28 15^29 -- --       12 13 28 29      14 15 28 29

	Thanks for hints from Richard Outerbridge - he told me IP&FP
	could be done in 15 xor, 10 shifts and 5 ands.
	When I finally started to think of the problem in 2D
	I first got ~42 operations without xors.  When I remembered
	how to use xors :-) I got it to its final state.
	*/
#define PERM_OP(a,b,t,n,m) ((t)=((((a)>>(n))^(b))&(m)),\
	(b)^=(t),\
	(a)^=((t)<<(n)))

#define IP(l,r) \
	{ \
	register DES_LONG tt; \
	PERM_OP(r,l,tt, 4,0x0f0f0f0fL); \
	PERM_OP(l,r,tt,16,0x0000ffffL); \
	PERM_OP(r,l,tt, 2,0x33333333L); \
	PERM_OP(l,r,tt, 8,0x00ff00ffL); \
	PERM_OP(r,l,tt, 1,0x55555555L); \
	}

#define FP(l,r) \
	{ \
	register DES_LONG tt; \
	PERM_OP(l,r,tt, 1,0x55555555L); \
	PERM_OP(r,l,tt, 8,0x00ff00ffL); \
	PERM_OP(l,r,tt, 2,0x33333333L); \
	PERM_OP(r,l,tt,16,0x0000ffffL); \
	PERM_OP(l,r,tt, 4,0x0f0f0f0fL); \
	}

extern const DES_LONG des_SPtrans[8][64];

#ifndef NOPROTO
void fcrypt_body(DES_LONG *out,des_key_schedule ks,
	DES_LONG Eswap0, DES_LONG Eswap1);
#else
void fcrypt_body();
#endif

#endif