1/* crypto/md5/md5_locl.h */
2/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
3 * All rights reserved.
4 *
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.
8 *
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).
15 *
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.
22 *
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
25 * are met:
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)"
40 *
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
51 * SUCH DAMAGE.
52 *
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.]
57 */
58
59#include <stdlib.h>
60#include <string.h>
61#include <openssl/e_os2.h>
62#include <openssl/md5.h>
63
64#ifndef MD5_LONG_LOG2
65#define MD5_LONG_LOG2 2 /* default to 32 bits */
66#endif
67
68#ifdef MD5_ASM
69# if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) || defined(__x86_64) || defined(__x86_64__)
70#  if !defined(B_ENDIAN)
71#   define md5_block_host_order md5_block_asm_host_order
72#  endif
73# elif defined(__sparc) && defined(OPENSSL_SYS_ULTRASPARC)
74   void md5_block_asm_data_order_aligned (MD5_CTX *c, const MD5_LONG *p,size_t num);
75#  define HASH_BLOCK_DATA_ORDER_ALIGNED md5_block_asm_data_order_aligned
76# endif
77#endif
78
79void md5_block_host_order (MD5_CTX *c, const void *p,size_t num);
80void md5_block_data_order (MD5_CTX *c, const void *p,size_t num);
81
82#if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) || defined(__x86_64) || defined(__x86_64__)
83# if !defined(B_ENDIAN)
84/*
85 * *_block_host_order is expected to handle aligned data while
86 * *_block_data_order - unaligned. As algorithm and host (x86)
87 * are in this case of the same "endianness" these two are
88 * otherwise indistinguishable. But normally you don't want to
89 * call the same function because unaligned access in places
90 * where alignment is expected is usually a "Bad Thing". Indeed,
91 * on RISCs you get punished with BUS ERROR signal or *severe*
92 * performance degradation. Intel CPUs are in turn perfectly
93 * capable of loading unaligned data without such drastic side
94 * effect. Yes, they say it's slower than aligned load, but no
95 * exception is generated and therefore performance degradation
96 * is *incomparable* with RISCs. What we should weight here is
97 * costs of unaligned access against costs of aligning data.
98 * According to my measurements allowing unaligned access results
99 * in ~9% performance improvement on Pentium II operating at
100 * 266MHz. I won't be surprised if the difference will be higher
101 * on faster systems:-)
102 *
103 *				<appro@fy.chalmers.se>
104 */
105# define md5_block_data_order md5_block_host_order
106# endif
107#endif
108
109#define DATA_ORDER_IS_LITTLE_ENDIAN
110
111#define HASH_LONG		MD5_LONG
112#define HASH_LONG_LOG2		MD5_LONG_LOG2
113#define HASH_CTX		MD5_CTX
114#define HASH_CBLOCK		MD5_CBLOCK
115#define HASH_LBLOCK		MD5_LBLOCK
116#define HASH_UPDATE		MD5_Update
117#define HASH_TRANSFORM		MD5_Transform
118#define HASH_FINAL		MD5_Final
119#define	HASH_MAKE_STRING(c,s)	do {	\
120	unsigned long ll;		\
121	ll=(c)->A; HOST_l2c(ll,(s));	\
122	ll=(c)->B; HOST_l2c(ll,(s));	\
123	ll=(c)->C; HOST_l2c(ll,(s));	\
124	ll=(c)->D; HOST_l2c(ll,(s));	\
125	} while (0)
126#define HASH_BLOCK_HOST_ORDER	md5_block_host_order
127#if !defined(L_ENDIAN) || defined(md5_block_data_order)
128#define	HASH_BLOCK_DATA_ORDER	md5_block_data_order
129/*
130 * Little-endians (Intel and Alpha) feel better without this.
131 * It looks like memcpy does better job than generic
132 * md5_block_data_order on copying-n-aligning input data.
133 * But frankly speaking I didn't expect such result on Alpha.
134 * On the other hand I've got this with egcs-1.0.2 and if
135 * program is compiled with another (better?) compiler it
136 * might turn out other way around.
137 *
138 *				<appro@fy.chalmers.se>
139 */
140#endif
141
142#include "md32_common.h"
143
144/*
145#define	F(x,y,z)	(((x) & (y))  |  ((~(x)) & (z)))
146#define	G(x,y,z)	(((x) & (z))  |  ((y) & (~(z))))
147*/
148
149/* As pointed out by Wei Dai <weidai@eskimo.com>, the above can be
150 * simplified to the code below.  Wei attributes these optimizations
151 * to Peter Gutmann's SHS code, and he attributes it to Rich Schroeppel.
152 */
153#define	F(b,c,d)	((((c) ^ (d)) & (b)) ^ (d))
154#define	G(b,c,d)	((((b) ^ (c)) & (d)) ^ (c))
155#define	H(b,c,d)	((b) ^ (c) ^ (d))
156#define	I(b,c,d)	(((~(d)) | (b)) ^ (c))
157
158#define R0(a,b,c,d,k,s,t) { \
159	a+=((k)+(t)+F((b),(c),(d))); \
160	a=ROTATE(a,s); \
161	a+=b; };\
162
163#define R1(a,b,c,d,k,s,t) { \
164	a+=((k)+(t)+G((b),(c),(d))); \
165	a=ROTATE(a,s); \
166	a+=b; };
167
168#define R2(a,b,c,d,k,s,t) { \
169	a+=((k)+(t)+H((b),(c),(d))); \
170	a=ROTATE(a,s); \
171	a+=b; };
172
173#define R3(a,b,c,d,k,s,t) { \
174	a+=((k)+(t)+I((b),(c),(d))); \
175	a=ROTATE(a,s); \
176	a+=b; };
177