1 /* crypto/md32_common.h */
2 /* ====================================================================
3 * Copyright (c) 1999-2007 The OpenSSL Project. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
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
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/)"
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.
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.
31 * 6. Redistributions of any form whatsoever must retain the following
33 * "This product includes software developed by the OpenSSL Project
34 * for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
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 * ====================================================================
53 * This is a generic 32 bit "collector" for message digest algorithms.
54 * Whenever needed it collects input character stream into chunks of
55 * 32 bit values and invokes a block function that performs actual hash
62 * DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN
63 * this macro defines byte order of input stream.
65 * size of a unit chunk HASH_BLOCK operates on.
67 * has to be at lest 32 bit wide, if it's wider, then
68 * HASH_LONG_LOG2 *has to* be defined along
70 * context structure that at least contains following
76 * HASH_LONG data[HASH_LBLOCK];
77 * unsigned char data[HASH_CBLOCK];
82 * data[] vector is expected to be zeroed upon first call to
85 * name of "Update" function, implemented here.
87 * name of "Transform" function, implemented here.
89 * name of "Final" function, implemented here.
90 * HASH_BLOCK_DATA_ORDER
91 * name of "block" function capable of treating *unaligned* input
92 * message in original (data) byte order, implemented externally.
94 * macro convering context variables to an ASCII hash string.
98 * #define DATA_ORDER_IS_LITTLE_ENDIAN
100 * #define HASH_LONG MD5_LONG
101 * #define HASH_LONG_LOG2 MD5_LONG_LOG2
102 * #define HASH_CTX MD5_CTX
103 * #define HASH_CBLOCK MD5_CBLOCK
104 * #define HASH_UPDATE MD5_Update
105 * #define HASH_TRANSFORM MD5_Transform
106 * #define HASH_FINAL MD5_Final
107 * #define HASH_BLOCK_DATA_ORDER md5_block_data_order
109 * <appro@fy.chalmers.se>
112 #if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN)
113 #error "DATA_ORDER must be defined!"
117 #error "HASH_CBLOCK must be defined!"
120 #error "HASH_LONG must be defined!"
123 #error "HASH_CTX must be defined!"
127 #error "HASH_UPDATE must be defined!"
129 #ifndef HASH_TRANSFORM
130 #error "HASH_TRANSFORM must be defined!"
133 #error "HASH_FINAL must be defined!"
136 #ifndef HASH_BLOCK_DATA_ORDER
137 #error "HASH_BLOCK_DATA_ORDER must be defined!"
141 * Engage compiler specific rotate intrinsic function if available.
145 # if defined(_MSC_VER) || defined(__ICC)
146 # define ROTATE(a,n) _lrotl(a,n)
147 # elif defined(__MWERKS__)
148 # if defined(__POWERPC__)
149 # define ROTATE(a,n) __rlwinm(a,n,0,31)
150 # elif defined(__MC68K__)
151 /* Motorola specific tweak. <appro@fy.chalmers.se> */
152 # define ROTATE(a,n) ( n<24 ? __rol(a,n) : __ror(a,32-n) )
154 # define ROTATE(a,n) __rol(a,n)
156 # elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
158 * Some GNU C inline assembler templates. Note that these are
159 * rotates by *constant* number of bits! But that's exactly
160 * what we need here...
161 * <appro@fy.chalmers.se>
163 # if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
164 # define ROTATE(a,n) ({ register unsigned int ret; \
172 # elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \
173 defined(__powerpc) || defined(__ppc__) || defined(__powerpc64__)
174 # define ROTATE(a,n) ({ register unsigned int ret; \
176 "rlwinm %0,%1,%2,0,31" \
181 # elif defined(__s390x__)
182 # define ROTATE(a,n) ({ register unsigned int ret; \
183 asm ("rll %0,%1,%2" \
190 #endif /* PEDANTIC */
193 #define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n))))
196 #if defined(DATA_ORDER_IS_BIG_ENDIAN)
199 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
200 # if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \
201 (defined(__x86_64) || defined(__x86_64__))
202 # if !defined(B_ENDIAN)
204 * This gives ~30-40% performance improvement in SHA-256 compiled
205 * with gcc [on P4]. Well, first macro to be frank. We can pull
206 * this trick on x86* platforms only, because these CPUs can fetch
207 * unaligned data without raising an exception.
209 # define HOST_c2l(c,l) ({ unsigned int r=*((const unsigned int *)(c)); \
210 asm ("bswapl %0":"=r"(r):"0"(r)); \
212 # define HOST_l2c(l,c) ({ unsigned int r=(l); \
213 asm ("bswapl %0":"=r"(r):"0"(r)); \
214 *((unsigned int *)(c))=r; (c)+=4; r; })
219 #if defined(__s390__) || defined(__s390x__)
220 # define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, (l))
221 # define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, (l))
225 #define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++)))<<24), \
226 l|=(((unsigned long)(*((c)++)))<<16), \
227 l|=(((unsigned long)(*((c)++)))<< 8), \
228 l|=(((unsigned long)(*((c)++))) ), \
232 #define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \
233 *((c)++)=(unsigned char)(((l)>>16)&0xff), \
234 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \
235 *((c)++)=(unsigned char)(((l) )&0xff), \
239 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
242 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
243 # if defined(__s390x__)
244 # define HOST_c2l(c,l) ({ asm ("lrv %0,0(%1)" \
245 :"=r"(l) : "r"(c)); \
247 # define HOST_l2c(l,c) ({ asm ("strv %0,0(%1)" \
248 : : "r"(l),"r"(c) : "memory"); \
253 #if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
255 /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
256 # define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l)
257 # define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l)
262 #define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \
263 l|=(((unsigned long)(*((c)++)))<< 8), \
264 l|=(((unsigned long)(*((c)++)))<<16), \
265 l|=(((unsigned long)(*((c)++)))<<24), \
269 #define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
270 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \
271 *((c)++)=(unsigned char)(((l)>>16)&0xff), \
272 *((c)++)=(unsigned char)(((l)>>24)&0xff), \
279 * Time for some action:-)
282 int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len)
284 const unsigned char *data=data_;
289 if (len==0) return 1;
291 l=(c->Nl+(((HASH_LONG)len)<<3))&0xffffffffUL;
292 /* 95-05-24 eay Fixed a bug with the overflow handling, thanks to
293 * Wei Dai <weidai@eskimo.com> for pointing it out. */
294 if (l < c->Nl) /* overflow */
296 c->Nh+=(HASH_LONG)(len>>29); /* might cause compiler warning on 16-bit */
302 p=(unsigned char *)c->data;
304 if (len >= HASH_CBLOCK || len+n >= HASH_CBLOCK)
306 memcpy (p+n,data,HASH_CBLOCK-n);
307 HASH_BLOCK_DATA_ORDER (c,p,1);
312 memset (p,0,HASH_CBLOCK); /* keep it zeroed */
316 memcpy (p+n,data,len);
317 c->num += (unsigned int)len;
325 HASH_BLOCK_DATA_ORDER (c,data,n);
333 p = (unsigned char *)c->data;
334 c->num = (unsigned int)len;
341 void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data)
343 HASH_BLOCK_DATA_ORDER (c,data,1);
347 int HASH_FINAL (unsigned char *md, HASH_CTX *c)
349 unsigned char *p = (unsigned char *)c->data;
352 p[n] = 0x80; /* there is always room for one */
355 if (n > (HASH_CBLOCK-8))
357 memset (p+n,0,HASH_CBLOCK-n);
359 HASH_BLOCK_DATA_ORDER (c,p,1);
361 memset (p+n,0,HASH_CBLOCK-8-n);
364 #if defined(DATA_ORDER_IS_BIG_ENDIAN)
365 (void)HOST_l2c(c->Nh,p);
366 (void)HOST_l2c(c->Nl,p);
367 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
368 (void)HOST_l2c(c->Nl,p);
369 (void)HOST_l2c(c->Nh,p);
372 HASH_BLOCK_DATA_ORDER (c,p,1);
374 memset (p,0,HASH_CBLOCK);
376 #ifndef HASH_MAKE_STRING
377 #error "HASH_MAKE_STRING must be defined!"
379 HASH_MAKE_STRING(c,md);
386 #define MD32_REG_T long
388 * This comment was originaly written for MD5, which is why it
389 * discusses A-D. But it basically applies to all 32-bit digests,
390 * which is why it was moved to common header file.
392 * In case you wonder why A-D are declared as long and not
393 * as MD5_LONG. Doing so results in slight performance
394 * boost on LP64 architectures. The catch is we don't
395 * really care if 32 MSBs of a 64-bit register get polluted
396 * with eventual overflows as we *save* only 32 LSBs in
397 * *either* case. Now declaring 'em long excuses the compiler
398 * from keeping 32 MSBs zeroed resulting in 13% performance
399 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
400 * Well, to be honest it should say that this *prevents*
401 * performance degradation.
402 * <appro@fy.chalmers.se>
403 * Apparently there're LP64 compilers that generate better
404 * code if A-D are declared int. Most notably GCC-x86_64
405 * generates better code.
406 * <appro@fy.chalmers.se>