md32_common.h revision 296341
1/* crypto/md32_common.h */ 2/* ==================================================================== 3 * Copyright (c) 1999-2007 The OpenSSL Project. All rights reserved. 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 */ 51 52/*- 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 56 * calculations. 57 * 58 * Porting guide. 59 * 60 * Obligatory macros: 61 * 62 * DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN 63 * this macro defines byte order of input stream. 64 * HASH_CBLOCK 65 * size of a unit chunk HASH_BLOCK operates on. 66 * HASH_LONG 67 * has to be at lest 32 bit wide, if it's wider, then 68 * HASH_LONG_LOG2 *has to* be defined along 69 * HASH_CTX 70 * context structure that at least contains following 71 * members: 72 * typedef struct { 73 * ... 74 * HASH_LONG Nl,Nh; 75 * either { 76 * HASH_LONG data[HASH_LBLOCK]; 77 * unsigned char data[HASH_CBLOCK]; 78 * }; 79 * unsigned int num; 80 * ... 81 * } HASH_CTX; 82 * data[] vector is expected to be zeroed upon first call to 83 * HASH_UPDATE. 84 * HASH_UPDATE 85 * name of "Update" function, implemented here. 86 * HASH_TRANSFORM 87 * name of "Transform" function, implemented here. 88 * HASH_FINAL 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. 93 * HASH_MAKE_STRING 94 * macro convering context variables to an ASCII hash string. 95 * 96 * MD5 example: 97 * 98 * #define DATA_ORDER_IS_LITTLE_ENDIAN 99 * 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 108 * 109 * <appro@fy.chalmers.se> 110 */ 111 112#if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN) 113# error "DATA_ORDER must be defined!" 114#endif 115 116#ifndef HASH_CBLOCK 117# error "HASH_CBLOCK must be defined!" 118#endif 119#ifndef HASH_LONG 120# error "HASH_LONG must be defined!" 121#endif 122#ifndef HASH_CTX 123# error "HASH_CTX must be defined!" 124#endif 125 126#ifndef HASH_UPDATE 127# error "HASH_UPDATE must be defined!" 128#endif 129#ifndef HASH_TRANSFORM 130# error "HASH_TRANSFORM must be defined!" 131#endif 132#ifndef HASH_FINAL 133# error "HASH_FINAL must be defined!" 134#endif 135 136#ifndef HASH_BLOCK_DATA_ORDER 137# error "HASH_BLOCK_DATA_ORDER must be defined!" 138#endif 139 140/* 141 * Engage compiler specific rotate intrinsic function if available. 142 */ 143#undef ROTATE 144#ifndef PEDANTIC 145# if defined(_MSC_VER) 146# define ROTATE(a,n) _lrotl(a,n) 147# elif defined(__ICC) 148# define ROTATE(a,n) _rotl(a,n) 149# elif defined(__MWERKS__) 150# if defined(__POWERPC__) 151# define ROTATE(a,n) __rlwinm(a,n,0,31) 152# elif defined(__MC68K__) 153 /* Motorola specific tweak. <appro@fy.chalmers.se> */ 154# define ROTATE(a,n) ( n<24 ? __rol(a,n) : __ror(a,32-n) ) 155# else 156# define ROTATE(a,n) __rol(a,n) 157# endif 158# elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 159 /* 160 * Some GNU C inline assembler templates. Note that these are 161 * rotates by *constant* number of bits! But that's exactly 162 * what we need here... 163 * <appro@fy.chalmers.se> 164 */ 165# if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 166# define ROTATE(a,n) ({ register unsigned int ret; \ 167 asm ( \ 168 "roll %1,%0" \ 169 : "=r"(ret) \ 170 : "I"(n), "0"((unsigned int)(a)) \ 171 : "cc"); \ 172 ret; \ 173 }) 174# elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \ 175 defined(__powerpc) || defined(__ppc__) || defined(__powerpc64__) 176# define ROTATE(a,n) ({ register unsigned int ret; \ 177 asm ( \ 178 "rlwinm %0,%1,%2,0,31" \ 179 : "=r"(ret) \ 180 : "r"(a), "I"(n)); \ 181 ret; \ 182 }) 183# elif defined(__s390x__) 184# define ROTATE(a,n) ({ register unsigned int ret; \ 185 asm ("rll %0,%1,%2" \ 186 : "=r"(ret) \ 187 : "r"(a), "I"(n)); \ 188 ret; \ 189 }) 190# endif 191# endif 192#endif /* PEDANTIC */ 193 194#ifndef ROTATE 195# define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n)))) 196#endif 197 198#if defined(DATA_ORDER_IS_BIG_ENDIAN) 199 200# ifndef PEDANTIC 201# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 202# if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \ 203 (defined(__x86_64) || defined(__x86_64__)) 204# if !defined(B_ENDIAN) 205 /* 206 * This gives ~30-40% performance improvement in SHA-256 compiled 207 * with gcc [on P4]. Well, first macro to be frank. We can pull 208 * this trick on x86* platforms only, because these CPUs can fetch 209 * unaligned data without raising an exception. 210 */ 211# define HOST_c2l(c,l) ({ unsigned int r=*((const unsigned int *)(c)); \ 212 asm ("bswapl %0":"=r"(r):"0"(r)); \ 213 (c)+=4; (l)=r; }) 214# define HOST_l2c(l,c) ({ unsigned int r=(l); \ 215 asm ("bswapl %0":"=r"(r):"0"(r)); \ 216 *((unsigned int *)(c))=r; (c)+=4; r; }) 217# endif 218# endif 219# endif 220# endif 221# if defined(__s390__) || defined(__s390x__) 222# define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, (l)) 223# define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, (l)) 224# endif 225 226# ifndef HOST_c2l 227# define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++)))<<24), \ 228 l|=(((unsigned long)(*((c)++)))<<16), \ 229 l|=(((unsigned long)(*((c)++)))<< 8), \ 230 l|=(((unsigned long)(*((c)++))) ) ) 231# endif 232# ifndef HOST_l2c 233# define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \ 234 *((c)++)=(unsigned char)(((l)>>16)&0xff), \ 235 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \ 236 *((c)++)=(unsigned char)(((l) )&0xff), \ 237 l) 238# endif 239 240#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 241 242# ifndef PEDANTIC 243# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 244# if defined(__s390x__) 245# define HOST_c2l(c,l) ({ asm ("lrv %0,%1" \ 246 :"=d"(l) :"m"(*(const unsigned int *)(c)));\ 247 (c)+=4; (l); }) 248# define HOST_l2c(l,c) ({ asm ("strv %1,%0" \ 249 :"=m"(*(unsigned int *)(c)) :"d"(l));\ 250 (c)+=4; (l); }) 251# endif 252# endif 253# endif 254# if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 255# ifndef B_ENDIAN 256 /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */ 257# define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l) 258# define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l) 259# endif 260# endif 261 262# ifndef HOST_c2l 263# define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \ 264 l|=(((unsigned long)(*((c)++)))<< 8), \ 265 l|=(((unsigned long)(*((c)++)))<<16), \ 266 l|=(((unsigned long)(*((c)++)))<<24) ) 267# endif 268# ifndef HOST_l2c 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), \ 273 l) 274# endif 275 276#endif 277 278/* 279 * Time for some action:-) 280 */ 281 282int HASH_UPDATE(HASH_CTX *c, const void *data_, size_t len) 283{ 284 const unsigned char *data = data_; 285 unsigned char *p; 286 HASH_LONG l; 287 size_t n; 288 289 if (len == 0) 290 return 1; 291 292 l = (c->Nl + (((HASH_LONG) len) << 3)) & 0xffffffffUL; 293 /* 294 * 95-05-24 eay Fixed a bug with the overflow handling, thanks to Wei Dai 295 * <weidai@eskimo.com> for pointing it out. 296 */ 297 if (l < c->Nl) /* overflow */ 298 c->Nh++; 299 c->Nh += (HASH_LONG) (len >> 29); /* might cause compiler warning on 300 * 16-bit */ 301 c->Nl = l; 302 303 n = c->num; 304 if (n != 0) { 305 p = (unsigned char *)c->data; 306 307 if (len >= HASH_CBLOCK || len + n >= HASH_CBLOCK) { 308 memcpy(p + n, data, HASH_CBLOCK - n); 309 HASH_BLOCK_DATA_ORDER(c, p, 1); 310 n = HASH_CBLOCK - n; 311 data += n; 312 len -= n; 313 c->num = 0; 314 memset(p, 0, HASH_CBLOCK); /* keep it zeroed */ 315 } else { 316 memcpy(p + n, data, len); 317 c->num += (unsigned int)len; 318 return 1; 319 } 320 } 321 322 n = len / HASH_CBLOCK; 323 if (n > 0) { 324 HASH_BLOCK_DATA_ORDER(c, data, n); 325 n *= HASH_CBLOCK; 326 data += n; 327 len -= n; 328 } 329 330 if (len != 0) { 331 p = (unsigned char *)c->data; 332 c->num = (unsigned int)len; 333 memcpy(p, data, len); 334 } 335 return 1; 336} 337 338void HASH_TRANSFORM(HASH_CTX *c, const unsigned char *data) 339{ 340 HASH_BLOCK_DATA_ORDER(c, data, 1); 341} 342 343int HASH_FINAL(unsigned char *md, HASH_CTX *c) 344{ 345 unsigned char *p = (unsigned char *)c->data; 346 size_t n = c->num; 347 348 p[n] = 0x80; /* there is always room for one */ 349 n++; 350 351 if (n > (HASH_CBLOCK - 8)) { 352 memset(p + n, 0, HASH_CBLOCK - n); 353 n = 0; 354 HASH_BLOCK_DATA_ORDER(c, p, 1); 355 } 356 memset(p + n, 0, HASH_CBLOCK - 8 - n); 357 358 p += HASH_CBLOCK - 8; 359#if defined(DATA_ORDER_IS_BIG_ENDIAN) 360 (void)HOST_l2c(c->Nh, p); 361 (void)HOST_l2c(c->Nl, p); 362#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 363 (void)HOST_l2c(c->Nl, p); 364 (void)HOST_l2c(c->Nh, p); 365#endif 366 p -= HASH_CBLOCK; 367 HASH_BLOCK_DATA_ORDER(c, p, 1); 368 c->num = 0; 369 memset(p, 0, HASH_CBLOCK); 370 371#ifndef HASH_MAKE_STRING 372# error "HASH_MAKE_STRING must be defined!" 373#else 374 HASH_MAKE_STRING(c, md); 375#endif 376 377 return 1; 378} 379 380#ifndef MD32_REG_T 381# if defined(__alpha) || defined(__sparcv9) || defined(__mips) 382# define MD32_REG_T long 383/* 384 * This comment was originaly written for MD5, which is why it 385 * discusses A-D. But it basically applies to all 32-bit digests, 386 * which is why it was moved to common header file. 387 * 388 * In case you wonder why A-D are declared as long and not 389 * as MD5_LONG. Doing so results in slight performance 390 * boost on LP64 architectures. The catch is we don't 391 * really care if 32 MSBs of a 64-bit register get polluted 392 * with eventual overflows as we *save* only 32 LSBs in 393 * *either* case. Now declaring 'em long excuses the compiler 394 * from keeping 32 MSBs zeroed resulting in 13% performance 395 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux. 396 * Well, to be honest it should say that this *prevents* 397 * performance degradation. 398 * <appro@fy.chalmers.se> 399 */ 400# else 401/* 402 * Above is not absolute and there are LP64 compilers that 403 * generate better code if MD32_REG_T is defined int. The above 404 * pre-processor condition reflects the circumstances under which 405 * the conclusion was made and is subject to further extension. 406 * <appro@fy.chalmers.se> 407 */ 408# define MD32_REG_T int 409# endif 410#endif 411