1 /* $OpenBSD: md32_common.h,v 1.22 2016/11/04 13:56:04 miod Exp $ */ 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 least 32 bit wide. 68 * HASH_CTX 69 * context structure that at least contains following 70 * members: 71 * typedef struct { 72 * ... 73 * HASH_LONG Nl,Nh; 74 * either { 75 * HASH_LONG data[HASH_LBLOCK]; 76 * unsigned char data[HASH_CBLOCK]; 77 * }; 78 * unsigned int num; 79 * ... 80 * } HASH_CTX; 81 * data[] vector is expected to be zeroed upon first call to 82 * HASH_UPDATE. 83 * HASH_UPDATE 84 * name of "Update" function, implemented here. 85 * HASH_TRANSFORM 86 * name of "Transform" function, implemented here. 87 * HASH_FINAL 88 * name of "Final" function, implemented here. 89 * HASH_BLOCK_DATA_ORDER 90 * name of "block" function capable of treating *unaligned* input 91 * message in original (data) byte order, implemented externally. 92 * HASH_MAKE_STRING 93 * macro convering context variables to an ASCII hash string. 94 * 95 * MD5 example: 96 * 97 * #define DATA_ORDER_IS_LITTLE_ENDIAN 98 * 99 * #define HASH_LONG MD5_LONG 100 * #define HASH_CTX MD5_CTX 101 * #define HASH_CBLOCK MD5_CBLOCK 102 * #define HASH_UPDATE MD5_Update 103 * #define HASH_TRANSFORM MD5_Transform 104 * #define HASH_FINAL MD5_Final 105 * #define HASH_BLOCK_DATA_ORDER md5_block_data_order 106 * 107 * <appro@fy.chalmers.se> 108 */ 109 110 #include <stdint.h> 111 112 #include <openssl/opensslconf.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 #if !defined(HASH_FINAL) && !defined(HASH_NO_FINAL) 135 #error "HASH_FINAL or HASH_NO_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 * This common idiom is recognized by the compiler and turned into a 144 * CPU-specific intrinsic as appropriate. 145 * e.g. GCC optimizes to roll on amd64 at -O0 146 */ 147 static inline uint32_t ROTATE(uint32_t a, uint32_t n) 148 { 149 return (a<<n)|(a>>(32-n)); 150 } 151 152 #if defined(DATA_ORDER_IS_BIG_ENDIAN) 153 154 #if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) 155 # if (defined(__i386) || defined(__i386__) || \ 156 defined(__x86_64) || defined(__x86_64__)) 157 /* 158 * This gives ~30-40% performance improvement in SHA-256 compiled 159 * with gcc [on P4]. Well, first macro to be frank. We can pull 160 * this trick on x86* platforms only, because these CPUs can fetch 161 * unaligned data without raising an exception. 162 */ 163 # define HOST_c2l(c,l) ({ unsigned int r=*((const unsigned int *)(c)); \ 164 asm ("bswapl %0":"=r"(r):"0"(r)); \ 165 (c)+=4; (l)=r; }) 166 # define HOST_l2c(l,c) ({ unsigned int r=(l); \ 167 asm ("bswapl %0":"=r"(r):"0"(r)); \ 168 *((unsigned int *)(c))=r; (c)+=4; }) 169 # endif 170 #endif 171 172 #ifndef HOST_c2l 173 #define HOST_c2l(c,l) do {l =(((unsigned long)(*((c)++)))<<24); \ 174 l|=(((unsigned long)(*((c)++)))<<16); \ 175 l|=(((unsigned long)(*((c)++)))<< 8); \ 176 l|=(((unsigned long)(*((c)++))) ); \ 177 } while (0) 178 #endif 179 #ifndef HOST_l2c 180 #define HOST_l2c(l,c) do {*((c)++)=(unsigned char)(((l)>>24)&0xff); \ 181 *((c)++)=(unsigned char)(((l)>>16)&0xff); \ 182 *((c)++)=(unsigned char)(((l)>> 8)&0xff); \ 183 *((c)++)=(unsigned char)(((l) )&0xff); \ 184 } while (0) 185 #endif 186 187 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 188 189 #if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__) 190 # define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4) 191 # define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4) 192 #endif 193 194 #ifndef HOST_c2l 195 #define HOST_c2l(c,l) do {l =(((unsigned long)(*((c)++))) ); \ 196 l|=(((unsigned long)(*((c)++)))<< 8); \ 197 l|=(((unsigned long)(*((c)++)))<<16); \ 198 l|=(((unsigned long)(*((c)++)))<<24); \ 199 } while (0) 200 #endif 201 #ifndef HOST_l2c 202 #define HOST_l2c(l,c) do {*((c)++)=(unsigned char)(((l) )&0xff); \ 203 *((c)++)=(unsigned char)(((l)>> 8)&0xff); \ 204 *((c)++)=(unsigned char)(((l)>>16)&0xff); \ 205 *((c)++)=(unsigned char)(((l)>>24)&0xff); \ 206 } while (0) 207 #endif 208 209 #endif 210 211 /* 212 * Time for some action:-) 213 */ 214 215 int 216 HASH_UPDATE(HASH_CTX *c, const void *data_, size_t len) 217 { 218 const unsigned char *data = data_; 219 unsigned char *p; 220 HASH_LONG l; 221 size_t n; 222 223 if (len == 0) 224 return 1; 225 226 l = (c->Nl + (((HASH_LONG)len) << 3))&0xffffffffUL; 227 /* 95-05-24 eay Fixed a bug with the overflow handling, thanks to 228 * Wei Dai <weidai@eskimo.com> for pointing it out. */ 229 if (l < c->Nl) /* overflow */ 230 c->Nh++; 231 c->Nh+=(HASH_LONG)(len>>29); /* might cause compiler warning on 16-bit */ 232 c->Nl = l; 233 234 n = c->num; 235 if (n != 0) { 236 p = (unsigned char *)c->data; 237 238 if (len >= HASH_CBLOCK || len + n >= HASH_CBLOCK) { 239 memcpy (p + n, data, HASH_CBLOCK - n); 240 HASH_BLOCK_DATA_ORDER (c, p, 1); 241 n = HASH_CBLOCK - n; 242 data += n; 243 len -= n; 244 c->num = 0; 245 memset (p,0,HASH_CBLOCK); /* keep it zeroed */ 246 } else { 247 memcpy (p + n, data, len); 248 c->num += (unsigned int)len; 249 return 1; 250 } 251 } 252 253 n = len/HASH_CBLOCK; 254 if (n > 0) { 255 HASH_BLOCK_DATA_ORDER (c, data, n); 256 n *= HASH_CBLOCK; 257 data += n; 258 len -= n; 259 } 260 261 if (len != 0) { 262 p = (unsigned char *)c->data; 263 c->num = (unsigned int)len; 264 memcpy (p, data, len); 265 } 266 return 1; 267 } 268 269 270 void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data) 271 { 272 HASH_BLOCK_DATA_ORDER (c, data, 1); 273 } 274 275 276 #ifndef HASH_NO_FINAL 277 int HASH_FINAL (unsigned char *md, HASH_CTX *c) 278 { 279 unsigned char *p = (unsigned char *)c->data; 280 size_t n = c->num; 281 282 p[n] = 0x80; /* there is always room for one */ 283 n++; 284 285 if (n > (HASH_CBLOCK - 8)) { 286 memset (p + n, 0, HASH_CBLOCK - n); 287 n = 0; 288 HASH_BLOCK_DATA_ORDER (c, p, 1); 289 } 290 memset (p + n, 0, HASH_CBLOCK - 8 - n); 291 292 p += HASH_CBLOCK - 8; 293 #if defined(DATA_ORDER_IS_BIG_ENDIAN) 294 HOST_l2c(c->Nh, p); 295 HOST_l2c(c->Nl, p); 296 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) 297 HOST_l2c(c->Nl, p); 298 HOST_l2c(c->Nh, p); 299 #endif 300 p -= HASH_CBLOCK; 301 HASH_BLOCK_DATA_ORDER (c, p, 1); 302 c->num = 0; 303 memset (p, 0, HASH_CBLOCK); 304 305 #ifndef HASH_MAKE_STRING 306 #error "HASH_MAKE_STRING must be defined!" 307 #else 308 HASH_MAKE_STRING(c, md); 309 #endif 310 311 return 1; 312 } 313 #endif 314 315 #ifndef MD32_REG_T 316 #if defined(__alpha) || defined(__sparcv9) || defined(__mips) 317 #define MD32_REG_T long 318 /* 319 * This comment was originaly written for MD5, which is why it 320 * discusses A-D. But it basically applies to all 32-bit digests, 321 * which is why it was moved to common header file. 322 * 323 * In case you wonder why A-D are declared as long and not 324 * as MD5_LONG. Doing so results in slight performance 325 * boost on LP64 architectures. The catch is we don't 326 * really care if 32 MSBs of a 64-bit register get polluted 327 * with eventual overflows as we *save* only 32 LSBs in 328 * *either* case. Now declaring 'em long excuses the compiler 329 * from keeping 32 MSBs zeroed resulting in 13% performance 330 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux. 331 * Well, to be honest it should say that this *prevents* 332 * performance degradation. 333 * <appro@fy.chalmers.se> 334 */ 335 #else 336 /* 337 * Above is not absolute and there are LP64 compilers that 338 * generate better code if MD32_REG_T is defined int. The above 339 * pre-processor condition reflects the circumstances under which 340 * the conclusion was made and is subject to further extension. 341 * <appro@fy.chalmers.se> 342 */ 343 #define MD32_REG_T int 344 #endif 345 #endif 346