twofish2.c

Assemble source code for twofish algorithm

	for (i=0;i<256;i++)
		{
		m1[0]=P8x8[0][i];		/* compute all the matrix elements */
		mX[0]=(BYTE) Mul_X(m1[0]);
		mY[0]=(BYTE) Mul_Y(m1[0]);

		m1[1]=P8x8[1][i];
		mX[1]=(BYTE) Mul_X(m1[1]);
		mY[1]=(BYTE) Mul_Y(m1[1]);

#undef	Mul_1					/* change what the pre-processor does with Mij */
#undef	Mul_X
#undef	Mul_Y
#define	Mul_1	m1				/* It will now access m01[], m5B[], and mEF[] */
#define	Mul_X	mX				
#define	Mul_Y	mY

#define	SetMDS(N)					\
		b0(d) = M0##N[P_##N##0];	\
		b1(d) = M1##N[P_##N##0];	\
		b2(d) = M2##N[P_##N##0];	\
		b3(d) = M3##N[P_##N##0];	\
		MDStab[N][i] = d;

		SetMDS(0);				/* fill in the matrix with elements computed above */
		SetMDS(1);
		SetMDS(2);
		SetMDS(3);
		}
#undef	Mul_1
#undef	Mul_X
#undef	Mul_Y
#define	Mul_1	Mx_1			/* re-enable true multiply */
#define	Mul_X	Mx_X
#define	Mul_Y	Mx_Y
	
#if BIG_TAB
	{
	int j,k;
	BYTE *q0,*q1;

	for (i=0;i<4;i++)
		{
		switch (i)
			{
			case 0:	q0=p8(01); q1=p8(02);	break;
			case 1:	q0=p8(11); q1=p8(12);	break;
			case 2:	q0=p8(21); q1=p8(22);	break;
			case 3:	q0=p8(31); q1=p8(32);	break;
			}
		for (j=0;j<256;j++)
			for (k=0;k<256;k++)
				bigTab[i][j][k]=q0[q1[k]^j];
		}
	}
#endif

	needToBuildMDS=0;			/* NEVER modify the table again! */
	}

/*
+*****************************************************************************
*
* Function Name:	ReverseRoundSubkeys
*
* Function:			Reverse order of round subkeys to switch between encrypt/decrypt
*
* Arguments:		key		=	ptr to keyInstance to be reversed
*					newDir	=	new direction value
*
* Return:			None.
*
* Notes:
*	This optimization allows both blockEncrypt and blockDecrypt to use the same
*	"fallthru" switch statement based on the number of rounds.
*	Note that key->numRounds must be even and >= 2 here.
*
-****************************************************************************/
void ReverseRoundSubkeys(keyInstance *key,BYTE newDir)
	{
	DWORD t0,t1;
	register DWORD *r0=key->subKeys+ROUND_SUBKEYS;
	register DWORD *r1=r0 + 2*key->numRounds - 2;

	for (;r0 < r1;r0+=2,r1-=2)
		{
		t0=r0[0];			/* swap the order */
		t1=r0[1];
		r0[0]=r1[0];		/* but keep relative order within pairs */
		r0[1]=r1[1];
		r1[0]=t0;
		r1[1]=t1;
		}

	key->direction=newDir;
	}

/*
+*****************************************************************************
*
* Function Name:	Xor256
*
* Function:			Copy an 8-bit permutation (256 bytes), xoring with a byte
*
* Arguments:		dst		=	where to put result
*					src		=	where to get data (can be same asa dst)
*					b		=	byte to xor
*
* Return:			None
*
* Notes:
* 	BorlandC's optimization is terrible!  When we put the code inline,
*	it generates fairly good code in the *following* segment (not in the Xor256
*	code itself).  If the call is made, the code following the call is awful!
*	The penalty is nearly 50%!  So we take the code size hit for inlining for
*	Borland, while Microsoft happily works with a call.
*
-****************************************************************************/
#if defined(__BORLANDC__)	/* do it inline */
#define Xor32(dst,src,i) { ((DWORD *)dst)[i] = ((DWORD *)src)[i] ^ tmpX; } 
#define	Xor256(dst,src,b)				\
	{									\
	register DWORD tmpX=0x01010101u * b;\
	for (i=0;i<64;i+=4)					\
		{ Xor32(dst,src,i  ); Xor32(dst,src,i+1); Xor32(dst,src,i+2); Xor32(dst,src,i+3); }	\
	}
#else						/* do it as a function call */
void Xor256(void *dst,void *src,BYTE b)
	{
	register DWORD	x=b*0x01010101u;	/* replicate byte to all four bytes */
	register DWORD *d=(DWORD *)dst;
	register DWORD *s=(DWORD *)src;
#define X_8(N)	{ d[N]=s[N] ^ x; d[N+1]=s[N+1] ^ x; }
#define X_32(N)	{ X_8(N); X_8(N+2); X_8(N+4); X_8(N+6); }
	X_32(0 ); X_32( 8); X_32(16); X_32(24);	/* all inline */
	d+=32;	/* keep offsets small! */
	s+=32;
	X_32(0 ); X_32( 8); X_32(16); X_32(24);	/* all inline */
	}
#endif

/*
+*****************************************************************************
*
* Function Name:	reKey
*
* Function:			Initialize the Twofish key schedule from key32
*
* Arguments:		key			=	ptr to keyInstance to be initialized
*
* Return:			TRUE on success
*
* Notes:
*	Here we precompute all the round subkeys, although that is not actually
*	required.  For example, on a smartcard, the round subkeys can 
*	be generated on-the-fly	using f32()
*
-****************************************************************************/
int reKey(keyInstance *key)
	{
	int		i,j,k64Cnt,keyLen;
	int		subkeyCnt;
	DWORD	A=0,B=0,q;
	DWORD	sKey[MAX_KEY_BITS/64],k32e[MAX_KEY_BITS/64],k32o[MAX_KEY_BITS/64];
	BYTE	L0[256],L1[256];	/* small local 8-bit permutations */

#if VALIDATE_PARMS
  #if ALIGN32
	if (((int)key) & 3)
		return BAD_ALIGN32;
	if ((key->keyLen % 64) || (key->keyLen < MIN_KEY_BITS))
		return BAD_KEY_INSTANCE;
  #endif
#endif

	if (needToBuildMDS)			/* do this one time only */
		BuildMDS();

#define	F32(res,x,k32)	\
	{															\
	DWORD t=x;													\
	switch (k64Cnt & 3)											\
	    {														\
		case 0:  /* same as 4 */								\
					b0(t)   = p8(04)[b0(t)] ^ b0(k32[3]);		\
					b1(t)   = p8(14)[b1(t)] ^ b1(k32[3]);		\
					b2(t)   = p8(24)[b2(t)] ^ b2(k32[3]);		\
					b3(t)   = p8(34)[b3(t)] ^ b3(k32[3]);		\
				 /* fall thru, having pre-processed t */		\
		case 3:		b0(t)   = p8(03)[b0(t)] ^ b0(k32[2]);		\
					b1(t)   = p8(13)[b1(t)] ^ b1(k32[2]);		\
					b2(t)   = p8(23)[b2(t)] ^ b2(k32[2]);		\
					b3(t)   = p8(33)[b3(t)] ^ b3(k32[2]);		\
				 /* fall thru, having pre-processed t */		\
		case 2:	 /* 128-bit keys (optimize for this case) */	\
			res=	MDStab[0][p8(01)[p8(02)[b0(t)] ^ b0(k32[1])] ^ b0(k32[0])] ^	\
					MDStab[1][p8(11)[p8(12)[b1(t)] ^ b1(k32[1])] ^ b1(k32[0])] ^	\
					MDStab[2][p8(21)[p8(22)[b2(t)] ^ b2(k32[1])] ^ b2(k32[0])] ^	\
					MDStab[3][p8(31)[p8(32)[b3(t)] ^ b3(k32[1])] ^ b3(k32[0])] ;	\
		}														\
	}


#if !CHECK_TABLE
#if defined(USE_ASM)				/* only do this if not using assember */
if (!(useAsm & 4))
#endif
#endif
	{
	subkeyCnt = ROUND_SUBKEYS + 2*key->numRounds;
	keyLen=key->keyLen;
	k64Cnt=(keyLen+63)/64;			/* number of 64-bit key words */
	for (i=0,j=k64Cnt-1;i<k64Cnt;i++,j--)
		{							/* split into even/odd key dwords */
		k32e[i]=key->key32[2*i  ];
		k32o[i]=key->key32[2*i+1];
		/* compute S-box keys using (12,8) Reed-Solomon code over GF(256) */
		sKey[j]=key->sboxKeys[j]=RS_MDS_Encode(k32e[i],k32o[i]);	/* reverse order */
		}
	}

#ifdef USE_ASM
if (useAsm & 4)
	{
	#if defined(COMPILE_KEY) && defined(USE_ASM)
		key->keySig		= VALID_SIG;			/* show that we are initialized */
		key->codeSize	= sizeof(key->compiledCode);	/* set size */
	#endif
	reKey_86(key);
	}
else
#endif
	{
	for (i=q=0;i<subkeyCnt/2;i++,q+=SK_STEP)	
		{							/* compute round subkeys for PHT */
		F32(A,q        ,k32e);		/* A uses even key dwords */
		F32(B,q+SK_BUMP,k32o);		/* B uses odd  key dwords */
		B = ROL(B,8);
		key->subKeys[2*i  ] = A+B;	/* combine with a PHT */
		B = A + 2*B;
		key->subKeys[2*i+1] = ROL(B,SK_ROTL);
		}
#if !defined(ZERO_KEY)
	switch (keyLen)	/* case out key length for speed in generating S-boxes */
		{
		case 128:
		#if defined(FULL_KEY) || defined(PART_KEY)
#if BIG_TAB
			#define	one128(N,J)	sbSet(N,i,J,L0[i+J])
			#define	sb128(N) {						\
				BYTE *qq=bigTab[N][b##N(sKey[1])];	\
				Xor256(L0,qq,b##N(sKey[0]));		\
				for (i=0;i<256;i+=2) { one128(N,0); one128(N,1); } }
#else
			#define	one128(N,J)	sbSet(N,i,J,p8(N##1)[L0[i+J]]^k0)
			#define	sb128(N) {					\
				Xor256(L0,p8(N##2),b##N(sKey[1]));	\
				{ register DWORD k0=b##N(sKey[0]);	\
				for (i=0;i<256;i+=2) { one128(N,0); one128(N,1); } } }
#endif
		#elif defined(MIN_KEY)
			#define	sb128(N) Xor256(_sBox8_(N),p8(N##2),b##N(sKey[1]))
		#endif
			sb128(0); sb128(1); sb128(2); sb128(3);
			break;
		case 192:
		#if defined(FULL_KEY) || defined(PART_KEY)
			#define one192(N,J) sbSet(N,i,J,p8(N##1)[p8(N##2)[L0[i+J]]^k1]^k0)
			#define	sb192(N) {						\
				Xor256(L0,p8(N##3),b##N(sKey[2]));	\
				{ register DWORD k0=b##N(sKey[0]);	\
				  register DWORD k1=b##N(sKey[1]);	\
				  for (i=0;i<256;i+=2) { one192(N,0); one192(N,1); } } }
		#elif defined(MIN_KEY)
			#define one192(N,J) sbSet(N,i,J,p8(N##2)[L0[i+J]]^k1)
			#define	sb192(N) {						\
				Xor256(L0,p8(N##3),b##N(sKey[2]));	\
				{ register DWORD k1=b##N(sKey[1]);	\
				  for (i=0;i<256;i+=2) { one192(N,0); one192(N,1); } } }
		#endif
			sb192(0); sb192(1); sb192(2); sb192(3);
			break;
		case 256:
		#if defined(FULL_KEY) || defined(PART_KEY)
			#define one256(N,J) sbSet(N,i,J,p8(N##1)[p8(N##2)[L0[i+J]]^k1]^k0)
			#define	sb256(N) {										\
				Xor256(L1,p8(N##4),b##N(sKey[3]));					\
				for (i=0;i<256;i+=2) {L0[i  ]=p8(N##3)[L1[i]];		\
									  L0[i+1]=p8(N##3)[L1[i+1]]; }	\
				Xor256(L0,L0,b##N(sKey[2]));						\
				{ register DWORD k0=b##N(sKey[0]);					\
				  register DWORD k1=b##N(sKey[1]);					\
				  for (i=0;i<256;i+=2) { one256(N,0); one256(N,1); } } }
		#elif defined(MIN_KEY)
			#define one256(N,J) sbSet(N,i,J,p8(N##2)[L0[i+J]]^k1)
			#define	sb256(N) {										\
				Xor256(L1,p8(N##4),b##N(sKey[3]));					\
				for (i=0;i<256;i+=2) {L0[i  ]=p8(N##3)[L1[i]];		\
									  L0[i+1]=p8(N##3)[L1[i+1]]; }	\
				Xor256(L0,L0,b##N(sKey[2]));						\
				{ register DWORD k1=b##N(sKey[1]);					\
				  for (i=0;i<256;i+=2) { one256(N,0); one256(N,1); } } }
		#endif
			sb256(0); sb256(1);	sb256(2); sb256(3);
			break;
		}
#endif
	}

#if CHECK_TABLE						/* sanity check  vs. pedagogical code*/
	{
	GetSboxKey;
	for (i=0;i<subkeyCnt/2;i++)
		{
		A = f32(i*SK_STEP        ,k32e,keyLen);	/* A uses even key dwords */
		B = f32(i*SK_STEP+SK_BUMP,k32o,keyLen);	/* B uses odd  key dwords */
		B = ROL(B,8);
		assert(key->subKeys[2*i  ] == A+  B);
		assert(key->subKeys[2*i+1] == ROL(A+2*B,SK_ROTL));
		}
  #if !defined(ZERO_KEY)			/* any S-boxes to check? */
	for (i=q=0;i<256;i++,q+=0x01010101)
		assert(f32(q,key->sboxKeys,keyLen) == Fe32_(q,0));
  #endif
	}
#endif /* CHECK_TABLE */

	DebugDumpKey(key);

	if (key->direction == DIR_ENCRYPT)	
		ReverseRoundSubkeys(key,DIR_ENCRYPT);	/* reverse the round subkey order */

	return TRUE;
	}
/*
+*****************************************************************************
*
* Function Name:	makeKey
*
* Function:			Initialize the Twofish key schedule
*
* Arguments:		key			=	ptr to keyInstance to be initialized
*					direction	=	DIR_ENCRYPT or DIR_DECRYPT
*					keyLen		=	# bits of key text at *keyMaterial
*					keyMaterial	=	ptr to hex ASCII chars representing key bits
*
* Return:			TRUE on success
*					else error code (e.g., BAD_KEY_DIR)
*
* Notes:	This parses the key bits from keyMaterial.  Zeroes out unused key bits
*
-****************************************************************************/
int makeKey(keyInstance *key, BYTE direction, int keyLen,CONST char *keyMaterial)
	{
#if VALIDATE_PARMS				/* first, sanity check on parameters */
	if (key == NULL)			
		return BAD_KEY_INSTANCE;/* must have a keyInstance to initialize */
	if ((direction != DIR_ENCRYPT) && (direction != DIR_DECRYPT))
		return BAD_KEY_DIR;		/* must have valid direction */
	if ((keyLen > MAX_KEY_BITS) || (keyLen < 8) || (keyLen & 0x3F))
		return BAD_KEY_MAT;		/* length must be valid */
	key->keySig = VALID_SIG;	/* show that we are initialized */
  #if ALIGN32
	if ((((int)key) & 3) || (((int)key->key32) & 3))
		return BAD_ALIGN32;
  #endif
#endif

	key->direction	= direction;/* set our cipher direction */
	key->keyLen		= (keyLen+63) & ~63;		/* round up to multiple of 64 */
	key->numRounds	= numRounds[(keyLen-1)/64];
	memset(key->key32,0,sizeof(key->key32));	/* zero unused bits */
	key->keyMaterial[MAX_KEY_SIZE]=0;	/* terminate ASCII string */

	if ((keyMaterial == NULL) || (keyMaterial[0]==0))
		return TRUE;			/* allow a "dummy" call */
		
	if (ParseHexDword(keyLen,keyMaterial,key->key32,key->keyMaterial))
		return BAD_KEY_MAT;	

	return reKey(key);			/* generate round subkeys */
	}


/*
+*****************************************************************************
*
* Function Name:	cipherInit
*