md32_common.h revision 306195
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#include <openssl/crypto.h> 113 114#if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN) 115# error "DATA_ORDER must be defined!" 116#endif 117 118#ifndef HASH_CBLOCK 119# error "HASH_CBLOCK must be defined!" 120#endif 121#ifndef HASH_LONG 122# error "HASH_LONG must be defined!" 123#endif 124#ifndef HASH_CTX 125# error "HASH_CTX must be defined!" 126#endif 127 128#ifndef HASH_UPDATE 129# error "HASH_UPDATE must be defined!" 130#endif 131#ifndef HASH_TRANSFORM 132# error "HASH_TRANSFORM must be defined!" 133#endif 134#ifndef HASH_FINAL 135# error "HASH_FINAL must be defined!" 136#endif 137 138#ifndef HASH_BLOCK_DATA_ORDER 139# error "HASH_BLOCK_DATA_ORDER must be defined!" 140#endif 141 142/* 143 * Engage compiler specific rotate intrinsic function if available. 144 */ 145#undef ROTATE 146#ifndef PEDANTIC 147# if defined(_MSC_VER) 148# define ROTATE(a,n) _lrotl(a,n) 149# elif defined(__ICC) 150# define ROTATE(a,n) _rotl(a,n) 151# elif defined(__MWERKS__) 152# if defined(__POWERPC__) 153# define ROTATE(a,n) __rlwinm(a,n,0,31) 154# elif defined(__MC68K__) 155 /* Motorola specific tweak. <appro@fy.chalmers.se> */ 156# define ROTATE(a,n) ( n<24 ? __rol(a,n) : __ror(a,32-n) ) 157# else 158# define ROTATE(a,n) __rol(a,n) 159# endif 160# elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 161 /* 162 * Some GNU C inline assembler templates. Note that these are 163 * rotates by *constant* number of bits! But that's exactly 164 * what we need here... 165 * <appro@fy.chalmers.se> 166 */ 167# if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 168# define ROTATE(a,n) ({ register unsigned int ret; \ 169 asm ( \ 170 "roll %1,%0" \ 171 : "=r"(ret) \ 172 : "I"(n), "0"((unsigned int)(a)) \ 173 : "cc"); \ 174 ret; \ 175 }) 176# elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \ 177 defined(__powerpc) || defined(__ppc__) || defined(__powerpc64__) 178# define ROTATE(a,n) ({ register unsigned int ret; \ 179 asm ( \ 180 "rlwinm %0,%1,%2,0,31" \ 181 : "=r"(ret) \ 182 : "r"(a), "I"(n)); \ 183 ret; \ 184 }) 185# elif defined(__s390x__) 186# define ROTATE(a,n) ({ register unsigned int ret; \ 187 asm ("rll %0,%1,%2" \ 188 : "=r"(ret) \ 189 : "r"(a), "I"(n)); \ 190 ret; \ 191 }) 192# endif 193# endif 194#endif /* PEDANTIC */ 195 196#ifndef ROTATE 197# define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n)))) 198#endif 199 200#if defined(DATA_ORDER_IS_BIG_ENDIAN) 201 202# ifndef PEDANTIC 203# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 204# if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \ 205 (defined(__x86_64) || defined(__x86_64__)) 206# if !defined(B_ENDIAN) 207 /* 208 * This gives ~30-40% performance improvement in SHA-256 compiled 209 * with gcc [on P4]. Well, first macro to be frank. We can pull 210 * this trick on x86* platforms only, because these CPUs can fetch 211 * unaligned data without raising an exception. 212 */ 213# define HOST_c2l(c,l) ({ unsigned int r=*((const unsigned int *)(c)); \ 214 asm ("bswapl %0":"=r"(r):"0"(r)); \ 215 (c)+=4; (l)=r; }) 216# define HOST_l2c(l,c) ({ unsigned int r=(l); \ 217 asm ("bswapl %0":"=r"(r):"0"(r)); \ 218 *((unsigned int *)(c))=r; (c)+=4; r; }) 219# endif 220# elif defined(__aarch64__) 221# if defined(__BYTE_ORDER__) 222# if defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__==__ORDER_LITTLE_ENDIAN__ 223# define HOST_c2l(c,l) ({ unsigned int r; \ 224 asm ("rev %w0,%w1" \ 225 :"=r"(r) \ 226 :"r"(*((const unsigned int *)(c))));\ 227 (c)+=4; (l)=r; }) 228# define HOST_l2c(l,c) ({ unsigned int r; \ 229 asm ("rev %w0,%w1" \ 230 :"=r"(r) \ 231 :"r"((unsigned int)(l)));\ 232 *((unsigned int *)(c))=r; (c)+=4; r; }) 233# elif defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__==__ORDER_BIG_ENDIAN__ 234# define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, (l)) 235# define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, (l)) 236# endif 237# endif 238# endif 239# endif 240# if defined(__s390__) || defined(__s390x__) 241# define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, (l)) 242# define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, (l)) 243# endif 244# endif 245 246# ifndef HOST_c2l 247# define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++)))<<24), \ 248 l|=(((unsigned long)(*((c)++)))<<16), \ 249 l|=(((unsigned long)(*((c)++)))<< 8), \ 250 l|=(((unsigned long)(*((c)++))) ) ) 251# endif 252# ifndef HOST_l2c 253# define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \ 254 *((c)++)=(unsigned char)(((l)>>16)&0xff), \ 255 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \ 256 *((c)++)=(unsigned char)(((l) )&0xff), \ 257 l) 258# endif 259 260#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 261 262# ifndef PEDANTIC 263# if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 264# if defined(__s390x__) 265# define HOST_c2l(c,l) ({ asm ("lrv %0,%1" \ 266 :"=d"(l) :"m"(*(const unsigned int *)(c)));\ 267 (c)+=4; (l); }) 268# define HOST_l2c(l,c) ({ asm ("strv %1,%0" \ 269 :"=m"(*(unsigned int *)(c)) :"d"(l));\ 270 (c)+=4; (l); }) 271# endif 272# endif 273# if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 274# ifndef B_ENDIAN 275 /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */ 276# define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l) 277# define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l) 278# endif 279# endif 280# endif 281 282# ifndef HOST_c2l 283# define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \ 284 l|=(((unsigned long)(*((c)++)))<< 8), \ 285 l|=(((unsigned long)(*((c)++)))<<16), \ 286 l|=(((unsigned long)(*((c)++)))<<24) ) 287# endif 288# ifndef HOST_l2c 289# define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \ 290 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \ 291 *((c)++)=(unsigned char)(((l)>>16)&0xff), \ 292 *((c)++)=(unsigned char)(((l)>>24)&0xff), \ 293 l) 294# endif 295 296#endif 297 298/* 299 * Time for some action:-) 300 */ 301 302int HASH_UPDATE(HASH_CTX *c, const void *data_, size_t len) 303{ 304 const unsigned char *data = data_; 305 unsigned char *p; 306 HASH_LONG l; 307 size_t n; 308 309 if (len == 0) 310 return 1; 311 312 l = (c->Nl + (((HASH_LONG) len) << 3)) & 0xffffffffUL; 313 /* 314 * 95-05-24 eay Fixed a bug with the overflow handling, thanks to Wei Dai 315 * <weidai@eskimo.com> for pointing it out. 316 */ 317 if (l < c->Nl) /* overflow */ 318 c->Nh++; 319 c->Nh += (HASH_LONG) (len >> 29); /* might cause compiler warning on 320 * 16-bit */ 321 c->Nl = l; 322 323 n = c->num; 324 if (n != 0) { 325 p = (unsigned char *)c->data; 326 327 if (len >= HASH_CBLOCK || len + n >= HASH_CBLOCK) { 328 memcpy(p + n, data, HASH_CBLOCK - n); 329 HASH_BLOCK_DATA_ORDER(c, p, 1); 330 n = HASH_CBLOCK - n; 331 data += n; 332 len -= n; 333 c->num = 0; 334 /* 335 * We use memset rather than OPENSSL_cleanse() here deliberately. 336 * Using OPENSSL_cleanse() here could be a performance issue. It 337 * will get properly cleansed on finalisation so this isn't a 338 * security problem. 339 */ 340 memset(p, 0, HASH_CBLOCK); /* keep it zeroed */ 341 } else { 342 memcpy(p + n, data, len); 343 c->num += (unsigned int)len; 344 return 1; 345 } 346 } 347 348 n = len / HASH_CBLOCK; 349 if (n > 0) { 350 HASH_BLOCK_DATA_ORDER(c, data, n); 351 n *= HASH_CBLOCK; 352 data += n; 353 len -= n; 354 } 355 356 if (len != 0) { 357 p = (unsigned char *)c->data; 358 c->num = (unsigned int)len; 359 memcpy(p, data, len); 360 } 361 return 1; 362} 363 364void HASH_TRANSFORM(HASH_CTX *c, const unsigned char *data) 365{ 366 HASH_BLOCK_DATA_ORDER(c, data, 1); 367} 368 369int HASH_FINAL(unsigned char *md, HASH_CTX *c) 370{ 371 unsigned char *p = (unsigned char *)c->data; 372 size_t n = c->num; 373 374 p[n] = 0x80; /* there is always room for one */ 375 n++; 376 377 if (n > (HASH_CBLOCK - 8)) { 378 memset(p + n, 0, HASH_CBLOCK - n); 379 n = 0; 380 HASH_BLOCK_DATA_ORDER(c, p, 1); 381 } 382 memset(p + n, 0, HASH_CBLOCK - 8 - n); 383 384 p += HASH_CBLOCK - 8; 385#if defined(DATA_ORDER_IS_BIG_ENDIAN) 386 (void)HOST_l2c(c->Nh, p); 387 (void)HOST_l2c(c->Nl, p); 388#elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 389 (void)HOST_l2c(c->Nl, p); 390 (void)HOST_l2c(c->Nh, p); 391#endif 392 p -= HASH_CBLOCK; 393 HASH_BLOCK_DATA_ORDER(c, p, 1); 394 c->num = 0; 395 OPENSSL_cleanse(p, HASH_CBLOCK); 396 397#ifndef HASH_MAKE_STRING 398# error "HASH_MAKE_STRING must be defined!" 399#else 400 HASH_MAKE_STRING(c, md); 401#endif 402 403 return 1; 404} 405 406#ifndef MD32_REG_T 407# if defined(__alpha) || defined(__sparcv9) || defined(__mips) 408# define MD32_REG_T long 409/* 410 * This comment was originaly written for MD5, which is why it 411 * discusses A-D. But it basically applies to all 32-bit digests, 412 * which is why it was moved to common header file. 413 * 414 * In case you wonder why A-D are declared as long and not 415 * as MD5_LONG. Doing so results in slight performance 416 * boost on LP64 architectures. The catch is we don't 417 * really care if 32 MSBs of a 64-bit register get polluted 418 * with eventual overflows as we *save* only 32 LSBs in 419 * *either* case. Now declaring 'em long excuses the compiler 420 * from keeping 32 MSBs zeroed resulting in 13% performance 421 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux. 422 * Well, to be honest it should say that this *prevents* 423 * performance degradation. 424 * <appro@fy.chalmers.se> 425 */ 426# else 427/* 428 * Above is not absolute and there are LP64 compilers that 429 * generate better code if MD32_REG_T is defined int. The above 430 * pre-processor condition reflects the circumstances under which 431 * the conclusion was made and is subject to further extension. 432 * <appro@fy.chalmers.se> 433 */ 434# define MD32_REG_T int 435# endif 436#endif 437