md32_common.h revision 279265
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) || 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) ) 153# else 154# define ROTATE(a,n) __rol(a,n) 155# endif 156# elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 157 /* 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> 162 */ 163# if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 164# define ROTATE(a,n) ({ register unsigned int ret; \ 165 asm ( \ 166 "roll %1,%0" \ 167 : "=r"(ret) \ 168 : "I"(n), "0"((unsigned int)(a)) \ 169 : "cc"); \ 170 ret; \ 171 }) 172# elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \ 173 defined(__powerpc) || defined(__ppc__) || defined(__powerpc64__) 174# define ROTATE(a,n) ({ register unsigned int ret; \ 175 asm ( \ 176 "rlwinm %0,%1,%2,0,31" \ 177 : "=r"(ret) \ 178 : "r"(a), "I"(n)); \ 179 ret; \ 180 }) 181# elif defined(__s390x__) 182# define ROTATE(a,n) ({ register unsigned int ret; \ 183 asm ("rll %0,%1,%2" \ 184 : "=r"(ret) \ 185 : "r"(a), "I"(n)); \ 186 ret; \ 187 }) 188# endif 189# endif 190#endif /* PEDANTIC */ 191 192#ifndef ROTATE 193#define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n)))) 194#endif 195 196#if defined(DATA_ORDER_IS_BIG_ENDIAN) 197 198#ifndef PEDANTIC 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) 203 /* 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. 208 */ 209# define HOST_c2l(c,l) ({ unsigned int r=*((const unsigned int *)(c)); \ 210 asm ("bswapl %0":"=r"(r):"0"(r)); \ 211 (c)+=4; (l)=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; }) 215# endif 216# endif 217# endif 218#endif 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)) 222#endif 223 224#ifndef HOST_c2l 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)++))) ) ) 229#endif 230#ifndef HOST_l2c 231#define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \ 232 *((c)++)=(unsigned char)(((l)>>16)&0xff), \ 233 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \ 234 *((c)++)=(unsigned char)(((l) )&0xff), \ 235 l) 236#endif 237 238#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 239 240#ifndef PEDANTIC 241# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 242# if defined(__s390x__) 243# define HOST_c2l(c,l) ({ asm ("lrv %0,%1" \ 244 :"=d"(l) :"m"(*(const unsigned int *)(c)));\ 245 (c)+=4; (l); }) 246# define HOST_l2c(l,c) ({ asm ("strv %1,%0" \ 247 :"=m"(*(unsigned int *)(c)) :"d"(l));\ 248 (c)+=4; (l); }) 249# endif 250# endif 251#endif 252#if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 253# ifndef B_ENDIAN 254 /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */ 255# define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l) 256# define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l) 257# endif 258#endif 259 260#ifndef HOST_c2l 261#define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \ 262 l|=(((unsigned long)(*((c)++)))<< 8), \ 263 l|=(((unsigned long)(*((c)++)))<<16), \ 264 l|=(((unsigned long)(*((c)++)))<<24) ) 265#endif 266#ifndef HOST_l2c 267#define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \ 268 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \ 269 *((c)++)=(unsigned char)(((l)>>16)&0xff), \ 270 *((c)++)=(unsigned char)(((l)>>24)&0xff), \ 271 l) 272#endif 273 274#endif 275 276/* 277 * Time for some action:-) 278 */ 279 280int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len) 281 { 282 const unsigned char *data=data_; 283 unsigned char *p; 284 HASH_LONG l; 285 size_t n; 286 287 if (len==0) return 1; 288 289 l=(c->Nl+(((HASH_LONG)len)<<3))&0xffffffffUL; 290 /* 95-05-24 eay Fixed a bug with the overflow handling, thanks to 291 * Wei Dai <weidai@eskimo.com> for pointing it out. */ 292 if (l < c->Nl) /* overflow */ 293 c->Nh++; 294 c->Nh+=(len>>29); /* might cause compiler warning on 16-bit */ 295 c->Nl=l; 296 297 n = c->num; 298 if (n != 0) 299 { 300 p=(unsigned char *)c->data; 301 302 if (len >= HASH_CBLOCK || len+n >= HASH_CBLOCK) 303 { 304 memcpy (p+n,data,HASH_CBLOCK-n); 305 HASH_BLOCK_DATA_ORDER (c,p,1); 306 n = HASH_CBLOCK-n; 307 data += n; 308 len -= n; 309 c->num = 0; 310 memset (p,0,HASH_CBLOCK); /* keep it zeroed */ 311 } 312 else 313 { 314 memcpy (p+n,data,len); 315 c->num += (unsigned int)len; 316 return 1; 317 } 318 } 319 320 n = len/HASH_CBLOCK; 321 if (n > 0) 322 { 323 HASH_BLOCK_DATA_ORDER (c,data,n); 324 n *= HASH_CBLOCK; 325 data += n; 326 len -= n; 327 } 328 329 if (len != 0) 330 { 331 p = (unsigned char *)c->data; 332 c->num = len; 333 memcpy (p,data,len); 334 } 335 return 1; 336 } 337 338 339void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data) 340 { 341 HASH_BLOCK_DATA_ORDER (c,data,1); 342 } 343 344 345int HASH_FINAL (unsigned char *md, HASH_CTX *c) 346 { 347 unsigned char *p = (unsigned char *)c->data; 348 size_t n = c->num; 349 350 p[n] = 0x80; /* there is always room for one */ 351 n++; 352 353 if (n > (HASH_CBLOCK-8)) 354 { 355 memset (p+n,0,HASH_CBLOCK-n); 356 n=0; 357 HASH_BLOCK_DATA_ORDER (c,p,1); 358 } 359 memset (p+n,0,HASH_CBLOCK-8-n); 360 361 p += HASH_CBLOCK-8; 362#if defined(DATA_ORDER_IS_BIG_ENDIAN) 363 (void)HOST_l2c(c->Nh,p); 364 (void)HOST_l2c(c->Nl,p); 365#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 366 (void)HOST_l2c(c->Nl,p); 367 (void)HOST_l2c(c->Nh,p); 368#endif 369 p -= HASH_CBLOCK; 370 HASH_BLOCK_DATA_ORDER (c,p,1); 371 c->num=0; 372 memset (p,0,HASH_CBLOCK); 373 374#ifndef HASH_MAKE_STRING 375#error "HASH_MAKE_STRING must be defined!" 376#else 377 HASH_MAKE_STRING(c,md); 378#endif 379 380 return 1; 381 } 382 383#ifndef MD32_REG_T 384#if defined(__alpha) || defined(__sparcv9) || defined(__mips) 385#define MD32_REG_T long 386/* 387 * This comment was originaly written for MD5, which is why it 388 * discusses A-D. But it basically applies to all 32-bit digests, 389 * which is why it was moved to common header file. 390 * 391 * In case you wonder why A-D are declared as long and not 392 * as MD5_LONG. Doing so results in slight performance 393 * boost on LP64 architectures. The catch is we don't 394 * really care if 32 MSBs of a 64-bit register get polluted 395 * with eventual overflows as we *save* only 32 LSBs in 396 * *either* case. Now declaring 'em long excuses the compiler 397 * from keeping 32 MSBs zeroed resulting in 13% performance 398 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux. 399 * Well, to be honest it should say that this *prevents* 400 * performance degradation. 401 * <appro@fy.chalmers.se> 402 */ 403#else 404/* 405 * Above is not absolute and there are LP64 compilers that 406 * generate better code if MD32_REG_T is defined int. The above 407 * pre-processor condition reflects the circumstances under which 408 * the conclusion was made and is subject to further extension. 409 * <appro@fy.chalmers.se> 410 */ 411#define MD32_REG_T int 412#endif 413#endif 414