[thirdparty/openssl.git] / crypto / md32_common.h

/* crypto/md32_common.h */
/* ====================================================================
 * Copyright (c) 1999-2002 The OpenSSL Project.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer. 
 *
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 *
 * 3. All advertising materials mentioning features or use of this
 *    software must display the following acknowledgment:
 *    "This product includes software developed by the OpenSSL Project
 *    for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
 *
 * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
 *    endorse or promote products derived from this software without
 *    prior written permission. For written permission, please contact
 *    licensing@OpenSSL.org.
 *
 * 5. Products derived from this software may not be called "OpenSSL"
 *    nor may "OpenSSL" appear in their names without prior written
 *    permission of the OpenSSL Project.
 *
 * 6. Redistributions of any form whatsoever must retain the following
 *    acknowledgment:
 *    "This product includes software developed by the OpenSSL Project
 *    for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
 *
 * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
 * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
 * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
 * OF THE POSSIBILITY OF SUCH DAMAGE.
 * ====================================================================
 *
 * This product includes cryptographic software written by Eric Young
 * (eay@cryptsoft.com).  This product includes software written by Tim
 * Hudson (tjh@cryptsoft.com).
 *
 */

/*
 * This is a generic 32 bit "collector" for message digest algorithms.
 * Whenever needed it collects input character stream into chunks of
 * 32 bit values and invokes a block function that performs actual hash
 * calculations.
 *
 * Porting guide.
 *
 * Obligatory macros:
 *
 * DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN
 *	this macro defines byte order of input stream.
 * HASH_CBLOCK
 *	size of a unit chunk HASH_BLOCK operates on.
 * HASH_LONG
 *	has to be at lest 32 bit wide, if it's wider, then
 *	HASH_LONG_LOG2 *has to* be defined along
 * HASH_CTX
 *	context structure that at least contains following
 *	members:
 *		typedef struct {
 *			...
 *			HASH_LONG	Nl,Nh;
 *			HASH_LONG	data[HASH_LBLOCK];
 *			int		num;
 *			...
 *			} HASH_CTX;
 * HASH_UPDATE
 *	name of "Update" function, implemented here.
 * HASH_TRANSFORM
 *	name of "Transform" function, implemented here.
 * HASH_FINAL
 *	name of "Final" function, implemented here.
 * HASH_BLOCK_HOST_ORDER
 *	name of "block" function treating *aligned* input message
 *	in host byte order, implemented externally.
 * HASH_BLOCK_DATA_ORDER
 *	name of "block" function treating *unaligned* input message
 *	in original (data) byte order, implemented externally (it
 *	actually is optional if data and host are of the same
 *	"endianess").
 * HASH_MAKE_STRING
 *	macro convering context variables to an ASCII hash string.
 *
 * Optional macros:
 *
 * B_ENDIAN or L_ENDIAN
 *	defines host byte-order.
 * HASH_LONG_LOG2
 *	defaults to 2 if not states otherwise.
 * HASH_LBLOCK
 *	assumed to be HASH_CBLOCK/4 if not stated otherwise.
 * HASH_BLOCK_DATA_ORDER_ALIGNED
 *	alternative "block" function capable of treating
 *	aligned input message in original (data) order,
 *	implemented externally.
 *
 * MD5 example:
 *
 *	#define DATA_ORDER_IS_LITTLE_ENDIAN
 *
 *	#define HASH_LONG		MD5_LONG
 *	#define HASH_LONG_LOG2		MD5_LONG_LOG2
 *	#define HASH_CTX		MD5_CTX
 *	#define HASH_CBLOCK		MD5_CBLOCK
 *	#define HASH_LBLOCK		MD5_LBLOCK
 *	#define HASH_UPDATE		MD5_Update
 *	#define HASH_TRANSFORM		MD5_Transform
 *	#define HASH_FINAL		MD5_Final
 *	#define HASH_BLOCK_HOST_ORDER	md5_block_host_order
 *	#define HASH_BLOCK_DATA_ORDER	md5_block_data_order
 *
 *					<appro@fy.chalmers.se>
 */

#if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN)
#error "DATA_ORDER must be defined!"
#endif

#ifndef HASH_CBLOCK
#error "HASH_CBLOCK must be defined!"
#endif
#ifndef HASH_LONG
#error "HASH_LONG must be defined!"
#endif
#ifndef HASH_CTX
#error "HASH_CTX must be defined!"
#endif

#ifndef HASH_UPDATE
#error "HASH_UPDATE must be defined!"
#endif
#ifndef HASH_TRANSFORM
#error "HASH_TRANSFORM must be defined!"
#endif
#ifndef HASH_FINAL
#error "HASH_FINAL must be defined!"
#endif

#ifndef HASH_BLOCK_HOST_ORDER
#error "HASH_BLOCK_HOST_ORDER must be defined!"
#endif

#if 0
/*
 * Moved below as it's required only if HASH_BLOCK_DATA_ORDER_ALIGNED
 * isn't defined.
 */
#ifndef HASH_BLOCK_DATA_ORDER
#error "HASH_BLOCK_DATA_ORDER must be defined!"
#endif
#endif

#ifndef HASH_LBLOCK
#define HASH_LBLOCK	(HASH_CBLOCK/4)
#endif

#ifndef HASH_LONG_LOG2
#define HASH_LONG_LOG2	2
#endif

/*
 * Engage compiler specific rotate intrinsic function if available.
 */
#undef ROTATE
#ifndef PEDANTIC
# if defined(_MSC_VER) || defined(__ICC)
#  define ROTATE(a,n)	_lrotl(a,n)
# elif defined(__MWERKS__)
#  if defined(__POWERPC__)
#   define ROTATE(a,n)	__rlwinm(a,n,0,31)
#  elif defined(__MC68K__)
    /* Motorola specific tweak. <appro@fy.chalmers.se> */
#   define ROTATE(a,n)	( n<24 ? __rol(a,n) : __ror(a,32-n) )
#  else
#   define ROTATE(a,n)	__rol(a,n)
#  endif
# elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
  /*
   * Some GNU C inline assembler templates. Note that these are
   * rotates by *constant* number of bits! But that's exactly
   * what we need here...
   * 					<appro@fy.chalmers.se>
   */
#  if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
#   define ROTATE(a,n)	({ register unsigned int ret;	\
				asm (			\
				"roll %1,%0"		\
				: "=r"(ret)		\
				: "I"(n), "0"(a)	\
				: "cc");		\
			   ret;				\
			})
#  elif defined(__powerpc) || defined(__ppc)
#   define ROTATE(a,n)	({ register unsigned int ret;	\
				asm (			\
				"rlwinm %0,%1,%2,0,31"	\
				: "=r"(ret)		\
				: "r"(a), "I"(n));	\
			   ret;				\
			})
#  endif
# endif
#endif /* PEDANTIC */

#if HASH_LONG_LOG2==2	/* Engage only if sizeof(HASH_LONG)== 4 */
/* A nice byte order reversal from Wei Dai <weidai@eskimo.com> */
#ifdef ROTATE
/* 5 instructions with rotate instruction, else 9 */
#define REVERSE_FETCH32(a,l)	(					\
		l=*(const HASH_LONG *)(a),				\
		((ROTATE(l,8)&0x00FF00FF)|(ROTATE((l&0x00FF00FF),24)))	\
				)
#else
/* 6 instructions with rotate instruction, else 8 */
#define REVERSE_FETCH32(a,l)	(				\
		l=*(const HASH_LONG *)(a),			\
		l=(((l>>8)&0x00FF00FF)|((l&0x00FF00FF)<<8)),	\
		ROTATE(l,16)					\
				)
/*
 * Originally the middle line started with l=(((l&0xFF00FF00)>>8)|...
 * It's rewritten as above for two reasons:
 *	- RISCs aren't good at long constants and have to explicitely
 *	  compose 'em with several (well, usually 2) instructions in a
 *	  register before performing the actual operation and (as you
 *	  already realized:-) having same constant should inspire the
 *	  compiler to permanently allocate the only register for it;
 *	- most modern CPUs have two ALUs, but usually only one has
 *	  circuitry for shifts:-( this minor tweak inspires compiler
 *	  to schedule shift instructions in a better way...
 *
 *				<appro@fy.chalmers.se>
 */
#endif
#endif

#ifndef ROTATE
#define ROTATE(a,n)     (((a)<<(n))|(((a)&0xffffffff)>>(32-(n))))
#endif

/*
 * Make some obvious choices. E.g., HASH_BLOCK_DATA_ORDER_ALIGNED
 * and HASH_BLOCK_HOST_ORDER ought to be the same if input data
 * and host are of the same "endianess". It's possible to mask
 * this with blank #define HASH_BLOCK_DATA_ORDER though...
 *
 *				<appro@fy.chalmers.se>
 */
#if defined(B_ENDIAN)
#  if defined(DATA_ORDER_IS_BIG_ENDIAN)
#    if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
#      define HASH_BLOCK_DATA_ORDER_ALIGNED	HASH_BLOCK_HOST_ORDER
#    endif
#  endif
#elif defined(L_ENDIAN)
#  if defined(DATA_ORDER_IS_LITTLE_ENDIAN)
#    if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
#      define HASH_BLOCK_DATA_ORDER_ALIGNED	HASH_BLOCK_HOST_ORDER
#    endif
#  endif
#endif

#if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
#ifndef HASH_BLOCK_DATA_ORDER
#error "HASH_BLOCK_DATA_ORDER must be defined!"
#endif
#endif

#if defined(DATA_ORDER_IS_BIG_ENDIAN)

#ifndef PEDANTIC
# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
#  if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
    /*
     * This gives ~30-40% performance improvement in SHA-256 compiled
     * with gcc [on P4]. Well, first macro to be frank. We can pull
     * this trick on x86* platforms only, because these CPUs can fetch
     * unaligned data without raising an exception.
     */
#   define HOST_c2l(c,l)	({ (l)=*((const unsigned int *)(c));	\
				   asm ("bswapl %0":"=r"(l):"0"(l));	\
				   (c)+=4; (l);				})
#   define HOST_l2c(l,c)	({ unsigned int r=(l);			\
				   asm ("bswapl %0":"=r"(r):"0"(r));	\
				   *((unsigned int *)(c))=r; (c)+=4; r;	})
#  endif
# endif
#endif

#ifndef HOST_c2l
#define HOST_c2l(c,l)	(l =(((unsigned long)(*((c)++)))<<24),		\
			 l|=(((unsigned long)(*((c)++)))<<16),		\
			 l|=(((unsigned long)(*((c)++)))<< 8),		\
			 l|=(((unsigned long)(*((c)++)))    ),		\
			 l)
#endif
#define HOST_p_c2l(c,l,n)	{					\
			switch (n) {					\
			case 0: l =((unsigned long)(*((c)++)))<<24;	\
			case 1: l|=((unsigned long)(*((c)++)))<<16;	\
			case 2: l|=((unsigned long)(*((c)++)))<< 8;	\
			case 3: l|=((unsigned long)(*((c)++)));		\
				} }
#define HOST_p_c2l_p(c,l,sc,len) {					\
			switch (sc) {					\
			case 0: l =((unsigned long)(*((c)++)))<<24;	\
				if (--len == 0) break;			\
			case 1: l|=((unsigned long)(*((c)++)))<<16;	\
				if (--len == 0) break;			\
			case 2: l|=((unsigned long)(*((c)++)))<< 8;	\
				} }
/* NOTE the pointer is not incremented at the end of this */
#define HOST_c2l_p(c,l,n)	{					\
			l=0; (c)+=n;					\
			switch (n) {					\
			case 3: l =((unsigned long)(*(--(c))))<< 8;	\
			case 2: l|=((unsigned long)(*(--(c))))<<16;	\
			case 1: l|=((unsigned long)(*(--(c))))<<24;	\
				} }
#ifndef HOST_l2c
#define HOST_l2c(l,c)	(*((c)++)=(unsigned char)(((l)>>24)&0xff),	\
			 *((c)++)=(unsigned char)(((l)>>16)&0xff),	\
			 *((c)++)=(unsigned char)(((l)>> 8)&0xff),	\
			 *((c)++)=(unsigned char)(((l)    )&0xff),	\
			 l)
#endif

#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)

#if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
  /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
# define HOST_c2l(c,l)	((l)=*((const unsigned int *)(c)), (c)+=4, l)
# define HOST_l2c(l,c)	(*((unsigned int *)(c))=(l), (c)+=4, l)
#endif

#ifndef HOST_c2l
#define HOST_c2l(c,l)	(l =(((unsigned long)(*((c)++)))    ),		\
			 l|=(((unsigned long)(*((c)++)))<< 8),		\
			 l|=(((unsigned long)(*((c)++)))<<16),		\
			 l|=(((unsigned long)(*((c)++)))<<24),		\
			 l)
#endif
#define HOST_p_c2l(c,l,n)	{					\
			switch (n) {					\
			case 0: l =((unsigned long)(*((c)++)));		\
			case 1: l|=((unsigned long)(*((c)++)))<< 8;	\
			case 2: l|=((unsigned long)(*((c)++)))<<16;	\
			case 3: l|=((unsigned long)(*((c)++)))<<24;	\
				} }
#define HOST_p_c2l_p(c,l,sc,len) {					\
			switch (sc) {					\
			case 0: l =((unsigned long)(*((c)++)));		\
				if (--len == 0) break;			\
			case 1: l|=((unsigned long)(*((c)++)))<< 8;	\
				if (--len == 0) break;			\
			case 2: l|=((unsigned long)(*((c)++)))<<16;	\
				} }
/* NOTE the pointer is not incremented at the end of this */
#define HOST_c2l_p(c,l,n)	{					\
			l=0; (c)+=n;					\
			switch (n) {					\
			case 3: l =((unsigned long)(*(--(c))))<<16;	\
			case 2: l|=((unsigned long)(*(--(c))))<< 8;	\
			case 1: l|=((unsigned long)(*(--(c))));		\
				} }
#ifndef HOST_l2c
#define HOST_l2c(l,c)	(*((c)++)=(unsigned char)(((l)    )&0xff),	\
			 *((c)++)=(unsigned char)(((l)>> 8)&0xff),	\
			 *((c)++)=(unsigned char)(((l)>>16)&0xff),	\
			 *((c)++)=(unsigned char)(((l)>>24)&0xff),	\
			 l)
#endif

#endif

/*
 * Time for some action:-)
 */

int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len)
	{
	const unsigned char *data=data_;
	register HASH_LONG * p;
	register HASH_LONG l;
	unsigned int sw,sc,ew,ec;

	if (len==0) return 1;

	l=(c->Nl+(((HASH_LONG)len)<<3))&0xffffffffUL;
	/* 95-05-24 eay Fixed a bug with the overflow handling, thanks to
	 * Wei Dai <weidai@eskimo.com> for pointing it out. */
	if (l < c->Nl) /* overflow */
		c->Nh++;
	c->Nh+=(len>>29);	/* might cause compiler warning on 16-bit */
	c->Nl=l;

	if (c->num != 0)
		{
		p=c->data;
		sw=c->num>>2;
		sc=c->num&0x03;

		if ((c->num+len) >= HASH_CBLOCK)
			{
			l=p[sw]; HOST_p_c2l(data,l,sc); p[sw++]=l;
			for (; sw<HASH_LBLOCK; sw++)
				{
				HOST_c2l(data,l); p[sw]=l;
				}
			HASH_BLOCK_HOST_ORDER (c,p,1);
			len-=(HASH_CBLOCK-c->num);
			c->num=0;
			/* drop through and do the rest */
			}
		else
			{
			c->num+=len;
			if ((sc+len) < 4) /* ugly, add char's to a word */
				{
				l=p[sw]; HOST_p_c2l_p(data,l,sc,len); p[sw]=l;
				}
			else
				{
				ew=(c->num>>2);
				ec=(c->num&0x03);
				if (sc)
					l=p[sw];
				HOST_p_c2l(data,l,sc);
				p[sw++]=l;
				for (; sw < ew; sw++)
					{
					HOST_c2l(data,l); p[sw]=l;
					}
				if (ec)
					{
					HOST_c2l_p(data,l,ec); p[sw]=l;
					}
				}
			return 1;
			}
		}

	sw=len/HASH_CBLOCK;
	if (sw > 0)
		{
#if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
		/*
		 * Note that HASH_BLOCK_DATA_ORDER_ALIGNED gets defined
		 * only if sizeof(HASH_LONG)==4.
		 */
		if ((((size_t)data)%4) == 0)
			{
			/* data is properly aligned so that we can cast it: */
			HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,sw);
			sw*=HASH_CBLOCK;
			data+=sw;
			len-=sw;
			}
		else
#if !defined(HASH_BLOCK_DATA_ORDER)
			while (sw--)
				{
				memcpy (p=c->data,data,HASH_CBLOCK);
				HASH_BLOCK_DATA_ORDER_ALIGNED(c,p,1);
				data+=HASH_CBLOCK;
				len-=HASH_CBLOCK;
				}
#endif
#endif
#if defined(HASH_BLOCK_DATA_ORDER)
			{
			HASH_BLOCK_DATA_ORDER(c,data,sw);
			sw*=HASH_CBLOCK;
			data+=sw;
			len-=sw;
			}
#endif
		}

	if (len!=0)
		{
		p = c->data;
		c->num = len;
		ew=len>>2;	/* words to copy */
		ec=len&0x03;
		for (; ew; ew--,p++)
			{
			HOST_c2l(data,l); *p=l;
			}
		HOST_c2l_p(data,l,ec);
		*p=l;
		}
	return 1;
	}


void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data)
	{
#if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
	if ((((size_t)data)%4) == 0)
		/* data is properly aligned so that we can cast it: */
		HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,1);
	else
#if !defined(HASH_BLOCK_DATA_ORDER)
		{
		memcpy (c->data,data,HASH_CBLOCK);
		HASH_BLOCK_DATA_ORDER_ALIGNED (c,c->data,1);
		}
#endif
#endif
#if defined(HASH_BLOCK_DATA_ORDER)
	HASH_BLOCK_DATA_ORDER (c,data,1);
#endif
	}


int HASH_FINAL (unsigned char *md, HASH_CTX *c)
	{
	register HASH_LONG *p;
	register unsigned long l;
	register int i,j;
	static const unsigned char end[4]={0x80,0x00,0x00,0x00};
	const unsigned char *cp=end;

	/* c->num should definitly have room for at least one more byte. */
	p=c->data;
	i=c->num>>2;
	j=c->num&0x03;

#if 0
	/* purify often complains about the following line as an
	 * Uninitialized Memory Read.  While this can be true, the
	 * following p_c2l macro will reset l when that case is true.
	 * This is because j&0x03 contains the number of 'valid' bytes
	 * already in p[i].  If and only if j&0x03 == 0, the UMR will
	 * occur but this is also the only time p_c2l will do
	 * l= *(cp++) instead of l|= *(cp++)
	 * Many thanks to Alex Tang <altitude@cic.net> for pickup this
	 * 'potential bug' */
#ifdef PURIFY
	if (j==0) p[i]=0; /* Yeah, but that's not the way to fix it:-) */
#endif
	l=p[i];
#else
	l = (j==0) ? 0 : p[i];
#endif
	HOST_p_c2l(cp,l,j); p[i++]=l; /* i is the next 'undefined word' */

	if (i>(HASH_LBLOCK-2)) /* save room for Nl and Nh */
		{
		if (i<HASH_LBLOCK) p[i]=0;
		HASH_BLOCK_HOST_ORDER (c,p,1);
		i=0;
		}
	for (; i<(HASH_LBLOCK-2); i++)
		p[i]=0;

#if   defined(DATA_ORDER_IS_BIG_ENDIAN)
	p[HASH_LBLOCK-2]=c->Nh;
	p[HASH_LBLOCK-1]=c->Nl;
#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
	p[HASH_LBLOCK-2]=c->Nl;
	p[HASH_LBLOCK-1]=c->Nh;
#endif
	HASH_BLOCK_HOST_ORDER (c,p,1);

#ifndef HASH_MAKE_STRING
#error "HASH_MAKE_STRING must be defined!"
#else
	HASH_MAKE_STRING(c,md);
#endif

	c->num=0;
	/* clear stuff, HASH_BLOCK may be leaving some stuff on the stack
	 * but I'm not worried :-)
	OPENSSL_cleanse((void *)c,sizeof(HASH_CTX));
	 */
	return 1;
	}

#ifndef MD32_REG_T
#define MD32_REG_T long
/*
 * This comment was originaly written for MD5, which is why it
 * discusses A-D. But it basically applies to all 32-bit digests,
 * which is why it was moved to common header file.
 *
 * In case you wonder why A-D are declared as long and not
 * as MD5_LONG. Doing so results in slight performance
 * boost on LP64 architectures. The catch is we don't
 * really care if 32 MSBs of a 64-bit register get polluted
 * with eventual overflows as we *save* only 32 LSBs in
 * *either* case. Now declaring 'em long excuses the compiler
 * from keeping 32 MSBs zeroed resulting in 13% performance
 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
 * Well, to be honest it should say that this *prevents* 
 * performance degradation.
 *				<appro@fy.chalmers.se>
 * Apparently there're LP64 compilers that generate better
 * code if A-D are declared int. Most notably GCC-x86_64
 * generates better code.
 *				<appro@fy.chalmers.se>
 */
#endif
Commit	Line	Data
bd3576d2 UM	1	/* crypto/md32_common.h */
bd3576d2 UM	2	/* ====================================================================
8a09b386	3	* Copyright (c) 1999-2002 The OpenSSL Project. All rights reserved.
bd3576d2 UM	4	*
	5	* Redistribution and use in source and binary forms, with or without
	6	* modification, are permitted provided that the following conditions
	7	* are met:
	8	*
	9	* 1. Redistributions of source code must retain the above copyright
	10	* notice, this list of conditions and the following disclaimer.
	11	*
	12	* 2. Redistributions in binary form must reproduce the above copyright
	13	* notice, this list of conditions and the following disclaimer in
	14	* the documentation and/or other materials provided with the
	15	* distribution.
	16	*
	17	* 3. All advertising materials mentioning features or use of this
	18	* software must display the following acknowledgment:
	19	* "This product includes software developed by the OpenSSL Project
	20	* for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
	21	*
	22	* 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
	23	* endorse or promote products derived from this software without
	24	* prior written permission. For written permission, please contact
	25	* licensing@OpenSSL.org.
	26	*
	27	* 5. Products derived from this software may not be called "OpenSSL"
	28	* nor may "OpenSSL" appear in their names without prior written
	29	* permission of the OpenSSL Project.
	30	*
	31	* 6. Redistributions of any form whatsoever must retain the following
	32	* acknowledgment:
	33	* "This product includes software developed by the OpenSSL Project
	34	* for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
	35	*
	36	* THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
	37	* EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
	38	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
	39	* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR
	40	* ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
	41	* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
	42	* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
	43	* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
	44	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
	45	* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
	46	* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
	47	* OF THE POSSIBILITY OF SUCH DAMAGE.
	48	* ====================================================================
	49	*
	50	* This product includes cryptographic software written by Eric Young
	51	* (eay@cryptsoft.com). This product includes software written by Tim
	52	* Hudson (tjh@cryptsoft.com).
	53	*
	54	*/
	55
	56	/*
	57	* This is a generic 32 bit "collector" for message digest algorithms.
	58	* Whenever needed it collects input character stream into chunks of
	59	* 32 bit values and invokes a block function that performs actual hash
	60	* calculations.
	61	*
	62	* Porting guide.
	63	*
	64	* Obligatory macros:
	65	*
	66	* DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN
	67	* this macro defines byte order of input stream.
68	* HASH_CBLOCK
69	* size of a unit chunk HASH_BLOCK operates on.
70	* HASH_LONG
71	* has to be at lest 32 bit wide, if it's wider, then
72	* HASH_LONG_LOG2 has to be defined along
73	* HASH_CTX
74	* context structure that at least contains following
75	* members:
76	* typedef struct {
77	* ...
78	* HASH_LONG Nl,Nh;
79	* HASH_LONG data[HASH_LBLOCK];
80	* int num;
81	* ...
82	* } HASH_CTX;
83	* HASH_UPDATE
84	* name of "Update" function, implemented here.
85	* HASH_TRANSFORM
86	* name of "Transform" function, implemented here.
87	* HASH_FINAL
88	* name of "Final" function, implemented here.
89	* HASH_BLOCK_HOST_ORDER
90	* name of "block" function treating aligned input message
91	* in host byte order, implemented externally.
92	* HASH_BLOCK_DATA_ORDER
93	* name of "block" function treating unaligned input message
94	* in original (data) byte order, implemented externally (it
95	* actually is optional if data and host are of the same
96	* "endianess").
1cbde6e4 AP	97	* HASH_MAKE_STRING
1cbde6e4 AP	98	* macro convering context variables to an ASCII hash string.
bd3576d2 UM	99	*
	100	* Optional macros:
	101	*
	102	* B_ENDIAN or L_ENDIAN
	103	* defines host byte-order.
	104	* HASH_LONG_LOG2
	105	* defaults to 2 if not states otherwise.
	106	* HASH_LBLOCK
	107	* assumed to be HASH_CBLOCK/4 if not stated otherwise.
	108	* HASH_BLOCK_DATA_ORDER_ALIGNED
	109	* alternative "block" function capable of treating
	110	* aligned input message in original (data) order,
	111	* implemented externally.
	112	*
	113	* MD5 example:
	114	*
	115	* #define DATA_ORDER_IS_LITTLE_ENDIAN
	116	*
	117	* #define HASH_LONG MD5_LONG
	118	* #define HASH_LONG_LOG2 MD5_LONG_LOG2
	119	* #define HASH_CTX MD5_CTX
	120	* #define HASH_CBLOCK MD5_CBLOCK
	121	* #define HASH_LBLOCK MD5_LBLOCK
	122	* #define HASH_UPDATE MD5_Update
	123	* #define HASH_TRANSFORM MD5_Transform
	124	* #define HASH_FINAL MD5_Final
	125	* #define HASH_BLOCK_HOST_ORDER md5_block_host_order
	126	* #define HASH_BLOCK_DATA_ORDER md5_block_data_order
	127	*
	128	* <appro@fy.chalmers.se>
	129	*/
	130
	131	#if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN)
	132	#error "DATA_ORDER must be defined!"
	133	#endif
	134
	135	#ifndef HASH_CBLOCK
	136	#error "HASH_CBLOCK must be defined!"
	137	#endif
	138	#ifndef HASH_LONG
	139	#error "HASH_LONG must be defined!"
	140	#endif
	141	#ifndef HASH_CTX
	142	#error "HASH_CTX must be defined!"
	143	#endif
	144
	145	#ifndef HASH_UPDATE
	146	#error "HASH_UPDATE must be defined!"
	147	#endif
	148	#ifndef HASH_TRANSFORM
	149	#error "HASH_TRANSFORM must be defined!"
	150	#endif
	151	#ifndef HASH_FINAL
	152	#error "HASH_FINAL must be defined!"
	153	#endif
	154
	155	#ifndef HASH_BLOCK_HOST_ORDER
	156	#error "HASH_BLOCK_HOST_ORDER must be defined!"
	157	#endif
	158
	159	#if 0
	160	/*
	161	* Moved below as it's required only if HASH_BLOCK_DATA_ORDER_ALIGNED
	162	* isn't defined.
163	*/
164	#ifndef HASH_BLOCK_DATA_ORDER
165	#error "HASH_BLOCK_DATA_ORDER must be defined!"
166	#endif
167	#endif
168
169	#ifndef HASH_LBLOCK
170	#define HASH_LBLOCK (HASH_CBLOCK/4)
171	#endif
172
173	#ifndef HASH_LONG_LOG2
174	#define HASH_LONG_LOG2 2
175	#endif
176
177	/*
178	* Engage compiler specific rotate intrinsic function if available.
179	*/
180	#undef ROTATE
181	#ifndef PEDANTIC
1a979201	182	# if defined(_MSC_VER) \|\| defined(__ICC)
1cbde6e4 AP	183	# define ROTATE(a,n) _lrotl(a,n)
1cbde6e4 AP	184	# elif defined(__MWERKS__)
1eab9a1f	185	# if defined(__POWERPC__)
a7c5241f	186	# define ROTATE(a,n) __rlwinm(a,n,0,31)
1eab9a1f AP	187	# elif defined(__MC68K__)
	188	/* Motorola specific tweak. <appro@fy.chalmers.se> */
	189	# define ROTATE(a,n) ( n<24 ? __rol(a,n) : __ror(a,32-n) )
9a1e34e5 AP	190	# else
	191	# define ROTATE(a,n) __rol(a,n)
	192	# endif
cf1b7d96	193	# elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
bd3576d2 UM	194	/*
	195	* Some GNU C inline assembler templates. Note that these are
	196	* rotates by constant number of bits! But that's exactly
	197	* what we need here...
bd3576d2 UM	198	* <appro@fy.chalmers.se>
bd3576d2 UM	199	*/
2f98abbc	200	# if defined(__i386) \|\| defined(__i386__) \|\| defined(__x86_64) \|\| defined(__x86_64__)
bd3576d2	201	# define ROTATE(a,n) ({ register unsigned int ret; \
0fad6cb7	202	asm ( \
bd3576d2 UM	203	"roll %1,%0" \
	204	: "=r"(ret) \
	205	: "I"(n), "0"(a) \
	206	: "cc"); \
	207	ret; \
	208	})
0fad6cb7	209	# elif defined(__powerpc) \|\| defined(__ppc)
bd3576d2	210	# define ROTATE(a,n) ({ register unsigned int ret; \
0fad6cb7	211	asm ( \
bd3576d2 UM	212	"rlwinm %0,%1,%2,0,31" \
	213	: "=r"(ret) \
	214	: "r"(a), "I"(n)); \
	215	ret; \
	216	})
	217	# endif
	218	# endif
bd3576d2 UM	219	#endif /* PEDANTIC */
	220
	221	#if HASH_LONG_LOG2==2 /* Engage only if sizeof(HASH_LONG)== 4 */
	222	/* A nice byte order reversal from Wei Dai <weidai@eskimo.com> */
	223	#ifdef ROTATE
	224	/* 5 instructions with rotate instruction, else 9 */
	225	#define REVERSE_FETCH32(a,l) ( \
	226	l=(const HASH_LONG )(a), \
	227	((ROTATE(l,8)&0x00FF00FF)\|(ROTATE((l&0x00FF00FF),24))) \
	228	)
	229	#else
	230	/* 6 instructions with rotate instruction, else 8 */
	231	#define REVERSE_FETCH32(a,l) ( \
	232	l=(const HASH_LONG )(a), \
	233	l=(((l>>8)&0x00FF00FF)\|((l&0x00FF00FF)<<8)), \
	234	ROTATE(l,16) \
	235	)
	236	/*
	237	* Originally the middle line started with l=(((l&0xFF00FF00)>>8)\|...
	238	* It's rewritten as above for two reasons:
	239	* - RISCs aren't good at long constants and have to explicitely
	240	* compose 'em with several (well, usually 2) instructions in a
	241	* register before performing the actual operation and (as you
	242	* already realized:-) having same constant should inspire the
	243	* compiler to permanently allocate the only register for it;
	244	* - most modern CPUs have two ALUs, but usually only one has
	245	* circuitry for shifts:-( this minor tweak inspires compiler
	246	* to schedule shift instructions in a better way...
	247	*
	248	* <appro@fy.chalmers.se>
	249	*/
	250	#endif
	251	#endif
	252
	253	#ifndef ROTATE
	254	#define ROTATE(a,n) (((a)<<(n))\|(((a)&0xffffffff)>>(32-(n))))
	255	#endif
	256
	257	/*
	258	* Make some obvious choices. E.g., HASH_BLOCK_DATA_ORDER_ALIGNED
	259	* and HASH_BLOCK_HOST_ORDER ought to be the same if input data
	260	* and host are of the same "endianess". It's possible to mask
	261	* this with blank #define HASH_BLOCK_DATA_ORDER though...
	262	*
	263	* <appro@fy.chalmers.se>
	264	*/
	265	#if defined(B_ENDIAN)
	266	# if defined(DATA_ORDER_IS_BIG_ENDIAN)
	267	# if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
	268	# define HASH_BLOCK_DATA_ORDER_ALIGNED HASH_BLOCK_HOST_ORDER
	269	# endif
bd3576d2 UM	270	# endif
	271	#elif defined(L_ENDIAN)
	272	# if defined(DATA_ORDER_IS_LITTLE_ENDIAN)
	273	# if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
	274	# define HASH_BLOCK_DATA_ORDER_ALIGNED HASH_BLOCK_HOST_ORDER
	275	# endif
bd3576d2 UM	276	# endif
	277	#endif
	278
0a78c297	279	#if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
bd3576d2 UM	280	#ifndef HASH_BLOCK_DATA_ORDER
	281	#error "HASH_BLOCK_DATA_ORDER must be defined!"
	282	#endif
	283	#endif
	284
	285	#if defined(DATA_ORDER_IS_BIG_ENDIAN)
	286
a2eb9688 AP	287	#ifndef PEDANTIC
	288	# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
	289	# if defined(__i386) \|\| defined(__i386__) \|\| defined(__x86_64) \|\| defined(__x86_64__)
	290	/*
	291	* This gives ~30-40% performance improvement in SHA-256 compiled
	292	* with gcc [on P4]. Well, first macro to be frank. We can pull
	293	* this trick on x86* platforms only, because these CPUs can fetch
	294	* unaligned data without raising an exception.
	295	*/
	296	# define HOST_c2l(c,l) ({ (l)=((const unsigned int )(c)); \
	297	asm ("bswapl %0":"=r"(l):"0"(l)); \
	298	(c)+=4; (l); })
	299	# define HOST_l2c(l,c) ({ unsigned int r=(l); \
	300	asm ("bswapl %0":"=r"(r):"0"(r)); \
	301	((unsigned int )(c))=r; (c)+=4; r; })
	302	# endif
	303	# endif
	304	#endif
	305
	306	#ifndef HOST_c2l
bd3576d2 UM	307	#define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++)))<<24), \
	308	l\|=(((unsigned long)(*((c)++)))<<16), \
	309	l\|=(((unsigned long)(*((c)++)))<< 8), \
	310	l\|=(((unsigned long)(*((c)++))) ), \
	311	l)
a2eb9688	312	#endif
bd3576d2 UM	313	#define HOST_p_c2l(c,l,n) { \
	314	switch (n) { \
	315	case 0: l =((unsigned long)(*((c)++)))<<24; \
	316	case 1: l\|=((unsigned long)(*((c)++)))<<16; \
	317	case 2: l\|=((unsigned long)(*((c)++)))<< 8; \
	318	case 3: l\|=((unsigned long)(*((c)++))); \
	319	} }
	320	#define HOST_p_c2l_p(c,l,sc,len) { \
	321	switch (sc) { \
	322	case 0: l =((unsigned long)(*((c)++)))<<24; \
	323	if (--len == 0) break; \
	324	case 1: l\|=((unsigned long)(*((c)++)))<<16; \
	325	if (--len == 0) break; \
	326	case 2: l\|=((unsigned long)(*((c)++)))<< 8; \
	327	} }
	328	/* NOTE the pointer is not incremented at the end of this */
	329	#define HOST_c2l_p(c,l,n) { \
	330	l=0; (c)+=n; \
	331	switch (n) { \
	332	case 3: l =((unsigned long)(*(--(c))))<< 8; \
	333	case 2: l\|=((unsigned long)(*(--(c))))<<16; \
	334	case 1: l\|=((unsigned long)(*(--(c))))<<24; \
	335	} }
a2eb9688	336	#ifndef HOST_l2c
bd3576d2 UM	337	#define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \
	338	*((c)++)=(unsigned char)(((l)>>16)&0xff), \
	339	*((c)++)=(unsigned char)(((l)>> 8)&0xff), \
	340	*((c)++)=(unsigned char)(((l) )&0xff), \
	341	l)
a2eb9688	342	#endif
bd3576d2 UM	343
	344	#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
	345
a2eb9688 AP	346	#if defined(__i386) \|\| defined(__i386__) \|\| defined(__x86_64) \|\| defined(__x86_64__)
	347	/* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
	348	# define HOST_c2l(c,l) ((l)=((const unsigned int )(c)), (c)+=4, l)
	349	# define HOST_l2c(l,c) (((unsigned int )(c))=(l), (c)+=4, l)
	350	#endif
	351
	352	#ifndef HOST_c2l
bd3576d2 UM	353	#define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \
	354	l\|=(((unsigned long)(*((c)++)))<< 8), \
	355	l\|=(((unsigned long)(*((c)++)))<<16), \
	356	l\|=(((unsigned long)(*((c)++)))<<24), \
	357	l)
a2eb9688	358	#endif
bd3576d2 UM	359	#define HOST_p_c2l(c,l,n) { \
	360	switch (n) { \
	361	case 0: l =((unsigned long)(*((c)++))); \
	362	case 1: l\|=((unsigned long)(*((c)++)))<< 8; \
	363	case 2: l\|=((unsigned long)(*((c)++)))<<16; \
	364	case 3: l\|=((unsigned long)(*((c)++)))<<24; \
	365	} }
	366	#define HOST_p_c2l_p(c,l,sc,len) { \
	367	switch (sc) { \
	368	case 0: l =((unsigned long)(*((c)++))); \
	369	if (--len == 0) break; \
	370	case 1: l\|=((unsigned long)(*((c)++)))<< 8; \
	371	if (--len == 0) break; \
	372	case 2: l\|=((unsigned long)(*((c)++)))<<16; \
	373	} }
	374	/* NOTE the pointer is not incremented at the end of this */
	375	#define HOST_c2l_p(c,l,n) { \
	376	l=0; (c)+=n; \
	377	switch (n) { \
	378	case 3: l =((unsigned long)(*(--(c))))<<16; \
	379	case 2: l\|=((unsigned long)(*(--(c))))<< 8; \
	380	case 1: l\|=((unsigned long)(*(--(c)))); \
	381	} }
a2eb9688	382	#ifndef HOST_l2c
bd3576d2 UM	383	#define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
	384	*((c)++)=(unsigned char)(((l)>> 8)&0xff), \
	385	*((c)++)=(unsigned char)(((l)>>16)&0xff), \
	386	*((c)++)=(unsigned char)(((l)>>24)&0xff), \
	387	l)
a2eb9688	388	#endif
bd3576d2 UM	389
	390	#endif
	391
	392	/*
	393	* Time for some action:-)
	394	*/
	395
9e0aad9f	396	int HASH_UPDATE (HASH_CTX c, const void data_, size_t len)
bd3576d2	397	{
bd445703	398	const unsigned char *data=data_;
bd3576d2	399	register HASH_LONG * p;
9e0aad9f	400	register HASH_LONG l;
a2eb9688	401	unsigned int sw,sc,ew,ec;
bd3576d2	402
2dc769a1	403	if (len==0) return 1;
bd3576d2	404
9e0aad9f	405	l=(c->Nl+(((HASH_LONG)len)<<3))&0xffffffffUL;
bd3576d2 UM	406	/* 95-05-24 eay Fixed a bug with the overflow handling, thanks to
	407	* Wei Dai <weidai@eskimo.com> for pointing it out. */
	408	if (l < c->Nl) /* overflow */
	409	c->Nh++;
9e0aad9f	410	c->Nh+=(len>>29); /* might cause compiler warning on 16-bit */
bd3576d2 UM	411	c->Nl=l;
	412
	413	if (c->num != 0)
	414	{
	415	p=c->data;
	416	sw=c->num>>2;
	417	sc=c->num&0x03;
	418
	419	if ((c->num+len) >= HASH_CBLOCK)
	420	{
	421	l=p[sw]; HOST_p_c2l(data,l,sc); p[sw++]=l;
	422	for (; sw<HASH_LBLOCK; sw++)
	423	{
	424	HOST_c2l(data,l); p[sw]=l;
	425	}
	426	HASH_BLOCK_HOST_ORDER (c,p,1);
	427	len-=(HASH_CBLOCK-c->num);
	428	c->num=0;
	429	/* drop through and do the rest */
	430	}
	431	else
	432	{
	433	c->num+=len;
	434	if ((sc+len) < 4) /* ugly, add char's to a word */
	435	{
	436	l=p[sw]; HOST_p_c2l_p(data,l,sc,len); p[sw]=l;
	437	}
	438	else
	439	{
	440	ew=(c->num>>2);
	441	ec=(c->num&0x03);
8a09b386 BM	442	if (sc)
	443	l=p[sw];
	444	HOST_p_c2l(data,l,sc);
	445	p[sw++]=l;
bd3576d2 UM	446	for (; sw < ew; sw++)
	447	{
	448	HOST_c2l(data,l); p[sw]=l;
	449	}
	450	if (ec)
	451	{
	452	HOST_c2l_p(data,l,ec); p[sw]=l;
	453	}
	454	}
2dc769a1	455	return 1;
bd3576d2 UM	456	}
	457	}
	458
	459	sw=len/HASH_CBLOCK;
	460	if (sw > 0)
	461	{
0a78c297	462	#if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
bd3576d2 UM	463	/*
	464	* Note that HASH_BLOCK_DATA_ORDER_ALIGNED gets defined
	465	* only if sizeof(HASH_LONG)==4.
	466	*/
a2eb9688	467	if ((((size_t)data)%4) == 0)
bd3576d2	468	{
4d5d543e	469	/* data is properly aligned so that we can cast it: */
8087d8f7	470	HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,sw);
bd3576d2 UM	471	sw*=HASH_CBLOCK;
	472	data+=sw;
	473	len-=sw;
	474	}
	475	else
	476	#if !defined(HASH_BLOCK_DATA_ORDER)
	477	while (sw--)
	478	{
	479	memcpy (p=c->data,data,HASH_CBLOCK);
	480	HASH_BLOCK_DATA_ORDER_ALIGNED(c,p,1);
	481	data+=HASH_CBLOCK;
	482	len-=HASH_CBLOCK;
	483	}
	484	#endif
	485	#endif
	486	#if defined(HASH_BLOCK_DATA_ORDER)
	487	{
531b2cf7	488	HASH_BLOCK_DATA_ORDER(c,data,sw);
bd3576d2 UM	489	sw*=HASH_CBLOCK;
	490	data+=sw;
	491	len-=sw;
	492	}
	493	#endif
	494	}
	495
	496	if (len!=0)
	497	{
	498	p = c->data;
	499	c->num = len;
	500	ew=len>>2; /* words to copy */
	501	ec=len&0x03;
	502	for (; ew; ew--,p++)
	503	{
	504	HOST_c2l(data,l); *p=l;
	505	}
	506	HOST_c2l_p(data,l,ec);
	507	*p=l;
	508	}
2dc769a1	509	return 1;
bd3576d2 UM	510	}
	511
	512
ac7d0785	513	void HASH_TRANSFORM (HASH_CTX c, const unsigned char data)
bd3576d2	514	{
0a78c297	515	#if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
a2eb9688	516	if ((((size_t)data)%4) == 0)
4d5d543e	517	/* data is properly aligned so that we can cast it: */
8087d8f7	518	HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,1);
bd3576d2 UM	519	else
	520	#if !defined(HASH_BLOCK_DATA_ORDER)
	521	{
	522	memcpy (c->data,data,HASH_CBLOCK);
	523	HASH_BLOCK_DATA_ORDER_ALIGNED (c,c->data,1);
	524	}
	525	#endif
	526	#endif
	527	#if defined(HASH_BLOCK_DATA_ORDER)
cdfb093f	528	HASH_BLOCK_DATA_ORDER (c,data,1);
bd3576d2 UM	529	#endif
	530	}
	531
	532
2dc769a1	533	int HASH_FINAL (unsigned char md, HASH_CTX c)
bd3576d2 UM	534	{
	535	register HASH_LONG *p;
	536	register unsigned long l;
	537	register int i,j;
	538	static const unsigned char end[4]={0x80,0x00,0x00,0x00};
	539	const unsigned char *cp=end;
	540
	541	/* c->num should definitly have room for at least one more byte. */
	542	p=c->data;
	543	i=c->num>>2;
	544	j=c->num&0x03;
	545
	546	#if 0
	547	/* purify often complains about the following line as an
	548	* Uninitialized Memory Read. While this can be true, the
	549	* following p_c2l macro will reset l when that case is true.
	550	* This is because j&0x03 contains the number of 'valid' bytes
	551	* already in p[i]. If and only if j&0x03 == 0, the UMR will
	552	* occur but this is also the only time p_c2l will do
	553	* l= (cp++) instead of l\|= (cp++)
	554	* Many thanks to Alex Tang <altitude@cic.net> for pickup this
	555	* 'potential bug' */
	556	#ifdef PURIFY
	557	if (j==0) p[i]=0; /* Yeah, but that's not the way to fix it:-) */
	558	#endif
	559	l=p[i];
	560	#else
	561	l = (j==0) ? 0 : p[i];
	562	#endif
	563	HOST_p_c2l(cp,l,j); p[i++]=l; /* i is the next 'undefined word' */
	564
	565	if (i>(HASH_LBLOCK-2)) /* save room for Nl and Nh */
	566	{
	567	if (i<HASH_LBLOCK) p[i]=0;
	568	HASH_BLOCK_HOST_ORDER (c,p,1);
	569	i=0;
	570	}
	571	for (; i<(HASH_LBLOCK-2); i++)
	572	p[i]=0;
	573
	574	#if defined(DATA_ORDER_IS_BIG_ENDIAN)
	575	p[HASH_LBLOCK-2]=c->Nh;
	576	p[HASH_LBLOCK-1]=c->Nl;
	577	#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
	578	p[HASH_LBLOCK-2]=c->Nl;
	579	p[HASH_LBLOCK-1]=c->Nh;
	580	#endif
	581	HASH_BLOCK_HOST_ORDER (c,p,1);
	582
1cbde6e4 AP	583	#ifndef HASH_MAKE_STRING
	584	#error "HASH_MAKE_STRING must be defined!"
	585	#else
	586	HASH_MAKE_STRING(c,md);
	587	#endif
bd3576d2 UM	588
	589	c->num=0;
	590	/* clear stuff, HASH_BLOCK may be leaving some stuff on the stack
	591	* but I'm not worried :-)
4579924b	592	OPENSSL_cleanse((void *)c,sizeof(HASH_CTX));
bd3576d2	593	*/
2dc769a1	594	return 1;
bd3576d2	595	}
2f98abbc AP	596
	597	#ifndef MD32_REG_T
	598	#define MD32_REG_T long
	599	/*
	600	* This comment was originaly written for MD5, which is why it
	601	* discusses A-D. But it basically applies to all 32-bit digests,
	602	* which is why it was moved to common header file.
	603	*
	604	* In case you wonder why A-D are declared as long and not
	605	* as MD5_LONG. Doing so results in slight performance
	606	* boost on LP64 architectures. The catch is we don't
	607	* really care if 32 MSBs of a 64-bit register get polluted
	608	* with eventual overflows as we save only 32 LSBs in
	609	* either case. Now declaring 'em long excuses the compiler
	610	* from keeping 32 MSBs zeroed resulting in 13% performance
	611	* improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
	612	* Well, to be honest it should say that this prevents
	613	* performance degradation.
	614	* <appro@fy.chalmers.se>
	615	* Apparently there're LP64 compilers that generate better
	616	* code if A-D are declared int. Most notably GCC-x86_64
	617	* generates better code.
	618	* <appro@fy.chalmers.se>
	619	*/
	620	#endif