1 /* $OpenBSD: umac.c,v 1.4 2011/10/19 10:39:48 djm Exp $ */ 2 /* ----------------------------------------------------------------------- 3 * 4 * umac.c -- C Implementation UMAC Message Authentication 5 * 6 * Version 0.93b of rfc4418.txt -- 2006 July 18 7 * 8 * For a full description of UMAC message authentication see the UMAC 9 * world-wide-web page at http://www.cs.ucdavis.edu/~rogaway/umac 10 * Please report bugs and suggestions to the UMAC webpage. 11 * 12 * Copyright (c) 1999-2006 Ted Krovetz 13 * 14 * Permission to use, copy, modify, and distribute this software and 15 * its documentation for any purpose and with or without fee, is hereby 16 * granted provided that the above copyright notice appears in all copies 17 * and in supporting documentation, and that the name of the copyright 18 * holder not be used in advertising or publicity pertaining to 19 * distribution of the software without specific, written prior permission. 20 * 21 * Comments should be directed to Ted Krovetz (tdk@acm.org) 22 * 23 * ---------------------------------------------------------------------- */ 24 25 /* ////////////////////// IMPORTANT NOTES ///////////////////////////////// 26 * 27 * 1) This version does not work properly on messages larger than 16MB 28 * 29 * 2) If you set the switch to use SSE2, then all data must be 16-byte 30 * aligned 31 * 32 * 3) When calling the function umac(), it is assumed that msg is in 33 * a writable buffer of length divisible by 32 bytes. The message itself 34 * does not have to fill the entire buffer, but bytes beyond msg may be 35 * zeroed. 36 * 37 * 4) Three free AES implementations are supported by this implementation of 38 * UMAC. Paulo Barreto's version is in the public domain and can be found 39 * at http://www.esat.kuleuven.ac.be/~rijmen/rijndael/ (search for 40 * "Barreto"). The only two files needed are rijndael-alg-fst.c and 41 * rijndael-alg-fst.h. Brian Gladman's version is distributed with the GNU 42 * Public lisence at http://fp.gladman.plus.com/AES/index.htm. It 43 * includes a fast IA-32 assembly version. The OpenSSL crypo library is 44 * the third. 45 * 46 * 5) With FORCE_C_ONLY flags set to 0, incorrect results are sometimes 47 * produced under gcc with optimizations set -O3 or higher. Dunno why. 48 * 49 /////////////////////////////////////////////////////////////////////// */ 50 51 /* ---------------------------------------------------------------------- */ 52 /* --- User Switches ---------------------------------------------------- */ 53 /* ---------------------------------------------------------------------- */ 54 55 #ifndef UMAC_OUTPUT_LEN 56 #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */ 57 #endif 58 59 #if UMAC_OUTPUT_LEN != 4 && UMAC_OUTPUT_LEN != 8 && \ 60 UMAC_OUTPUT_LEN != 12 && UMAC_OUTPUT_LEN != 16 61 # error UMAC_OUTPUT_LEN must be defined to 4, 8, 12 or 16 62 #endif 63 64 /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */ 65 /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */ 66 /* #define SSE2 0 Is SSE2 is available? */ 67 /* #define RUN_TESTS 0 Run basic correctness/speed tests */ 68 /* #define UMAC_AE_SUPPORT 0 Enable auhthenticated encrytion */ 69 70 /* ---------------------------------------------------------------------- */ 71 /* -- Global Includes --------------------------------------------------- */ 72 /* ---------------------------------------------------------------------- */ 73 74 #include "includes.h" 75 #include <sys/types.h> 76 77 #include "xmalloc.h" 78 #include "umac.h" 79 #include <string.h> 80 #include <stdlib.h> 81 #include <stddef.h> 82 83 /* ---------------------------------------------------------------------- */ 84 /* --- Primitive Data Types --- */ 85 /* ---------------------------------------------------------------------- */ 86 87 /* The following assumptions may need change on your system */ 88 typedef u_int8_t UINT8; /* 1 byte */ 89 typedef u_int16_t UINT16; /* 2 byte */ 90 typedef u_int32_t UINT32; /* 4 byte */ 91 typedef u_int64_t UINT64; /* 8 bytes */ 92 typedef unsigned int UWORD; /* Register */ 93 94 /* ---------------------------------------------------------------------- */ 95 /* --- Constants -------------------------------------------------------- */ 96 /* ---------------------------------------------------------------------- */ 97 98 #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */ 99 100 /* Message "words" are read from memory in an endian-specific manner. */ 101 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */ 102 /* be set true if the host computer is little-endian. */ 103 104 #if BYTE_ORDER == LITTLE_ENDIAN 105 #define __LITTLE_ENDIAN__ 1 106 #else 107 #define __LITTLE_ENDIAN__ 0 108 #endif 109 110 /* ---------------------------------------------------------------------- */ 111 /* ---------------------------------------------------------------------- */ 112 /* ----- Architecture Specific ------------------------------------------ */ 113 /* ---------------------------------------------------------------------- */ 114 /* ---------------------------------------------------------------------- */ 115 116 117 /* ---------------------------------------------------------------------- */ 118 /* ---------------------------------------------------------------------- */ 119 /* ----- Primitive Routines --------------------------------------------- */ 120 /* ---------------------------------------------------------------------- */ 121 /* ---------------------------------------------------------------------- */ 122 123 124 /* ---------------------------------------------------------------------- */ 125 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */ 126 /* ---------------------------------------------------------------------- */ 127 128 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b))) 129 130 /* ---------------------------------------------------------------------- */ 131 /* --- Endian Conversion --- Forcing assembly on some platforms */ 132 /* ---------------------------------------------------------------------- */ 133 134 #if HAVE_SWAP32 135 #define LOAD_UINT32_REVERSED(p) (swap32(*(UINT32 *)(p))) 136 #define STORE_UINT32_REVERSED(p,v) (*(UINT32 *)(p) = swap32(v)) 137 #else /* HAVE_SWAP32 */ 138 139 static UINT32 LOAD_UINT32_REVERSED(void *ptr) 140 { 141 UINT32 temp = *(UINT32 *)ptr; 142 temp = (temp >> 24) | ((temp & 0x00FF0000) >> 8 ) 143 | ((temp & 0x0000FF00) << 8 ) | (temp << 24); 144 return (UINT32)temp; 145 } 146 147 # if (__LITTLE_ENDIAN__) 148 static void STORE_UINT32_REVERSED(void *ptr, UINT32 x) 149 { 150 UINT32 i = (UINT32)x; 151 *(UINT32 *)ptr = (i >> 24) | ((i & 0x00FF0000) >> 8 ) 152 | ((i & 0x0000FF00) << 8 ) | (i << 24); 153 } 154 # endif /* __LITTLE_ENDIAN */ 155 #endif /* HAVE_SWAP32 */ 156 157 /* The following definitions use the above reversal-primitives to do the right 158 * thing on endian specific load and stores. 159 */ 160 161 #if (__LITTLE_ENDIAN__) 162 #define LOAD_UINT32_LITTLE(ptr) (*(UINT32 *)(ptr)) 163 #define STORE_UINT32_BIG(ptr,x) STORE_UINT32_REVERSED(ptr,x) 164 #else 165 #define LOAD_UINT32_LITTLE(ptr) LOAD_UINT32_REVERSED(ptr) 166 #define STORE_UINT32_BIG(ptr,x) (*(UINT32 *)(ptr) = (UINT32)(x)) 167 #endif 168 169 /* ---------------------------------------------------------------------- */ 170 /* ---------------------------------------------------------------------- */ 171 /* ----- Begin KDF & PDF Section ---------------------------------------- */ 172 /* ---------------------------------------------------------------------- */ 173 /* ---------------------------------------------------------------------- */ 174 175 /* UMAC uses AES with 16 byte block and key lengths */ 176 #define AES_BLOCK_LEN 16 177 178 /* OpenSSL's AES */ 179 #include "openbsd-compat/openssl-compat.h" 180 #ifndef USE_BUILTIN_RIJNDAEL 181 # include <openssl/aes.h> 182 #endif 183 typedef AES_KEY aes_int_key[1]; 184 #define aes_encryption(in,out,int_key) \ 185 AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key) 186 #define aes_key_setup(key,int_key) \ 187 AES_set_encrypt_key((u_char *)(key),UMAC_KEY_LEN*8,int_key) 188 189 /* The user-supplied UMAC key is stretched using AES in a counter 190 * mode to supply all random bits needed by UMAC. The kdf function takes 191 * an AES internal key representation 'key' and writes a stream of 192 * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct 193 * 'ndx' causes a distinct byte stream. 194 */ 195 static void kdf(void *bufp, aes_int_key key, UINT8 ndx, int nbytes) 196 { 197 UINT8 in_buf[AES_BLOCK_LEN] = {0}; 198 UINT8 out_buf[AES_BLOCK_LEN]; 199 UINT8 *dst_buf = (UINT8 *)bufp; 200 int i; 201 202 /* Setup the initial value */ 203 in_buf[AES_BLOCK_LEN-9] = ndx; 204 in_buf[AES_BLOCK_LEN-1] = i = 1; 205 206 while (nbytes >= AES_BLOCK_LEN) { 207 aes_encryption(in_buf, out_buf, key); 208 memcpy(dst_buf,out_buf,AES_BLOCK_LEN); 209 in_buf[AES_BLOCK_LEN-1] = ++i; 210 nbytes -= AES_BLOCK_LEN; 211 dst_buf += AES_BLOCK_LEN; 212 } 213 if (nbytes) { 214 aes_encryption(in_buf, out_buf, key); 215 memcpy(dst_buf,out_buf,nbytes); 216 } 217 } 218 219 /* The final UHASH result is XOR'd with the output of a pseudorandom 220 * function. Here, we use AES to generate random output and 221 * xor the appropriate bytes depending on the last bits of nonce. 222 * This scheme is optimized for sequential, increasing big-endian nonces. 223 */ 224 225 typedef struct { 226 UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */ 227 UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */ 228 aes_int_key prf_key; /* Expanded AES key for PDF */ 229 } pdf_ctx; 230 231 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key) 232 { 233 UINT8 buf[UMAC_KEY_LEN]; 234 235 kdf(buf, prf_key, 0, UMAC_KEY_LEN); 236 aes_key_setup(buf, pc->prf_key); 237 238 /* Initialize pdf and cache */ 239 memset(pc->nonce, 0, sizeof(pc->nonce)); 240 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 241 } 242 243 static void pdf_gen_xor(pdf_ctx *pc, UINT8 nonce[8], UINT8 buf[8]) 244 { 245 /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes 246 * of the AES output. If last time around we returned the ndx-1st 247 * element, then we may have the result in the cache already. 248 */ 249 250 #if (UMAC_OUTPUT_LEN == 4) 251 #define LOW_BIT_MASK 3 252 #elif (UMAC_OUTPUT_LEN == 8) 253 #define LOW_BIT_MASK 1 254 #elif (UMAC_OUTPUT_LEN > 8) 255 #define LOW_BIT_MASK 0 256 #endif 257 258 UINT8 tmp_nonce_lo[4]; 259 #if LOW_BIT_MASK != 0 260 int ndx = nonce[7] & LOW_BIT_MASK; 261 #endif 262 *(UINT32 *)tmp_nonce_lo = ((UINT32 *)nonce)[1]; 263 tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */ 264 265 if ( (((UINT32 *)tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) || 266 (((UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) ) 267 { 268 ((UINT32 *)pc->nonce)[0] = ((UINT32 *)nonce)[0]; 269 ((UINT32 *)pc->nonce)[1] = ((UINT32 *)tmp_nonce_lo)[0]; 270 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 271 } 272 273 #if (UMAC_OUTPUT_LEN == 4) 274 *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx]; 275 #elif (UMAC_OUTPUT_LEN == 8) 276 *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx]; 277 #elif (UMAC_OUTPUT_LEN == 12) 278 ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; 279 ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2]; 280 #elif (UMAC_OUTPUT_LEN == 16) 281 ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; 282 ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1]; 283 #endif 284 } 285 286 /* ---------------------------------------------------------------------- */ 287 /* ---------------------------------------------------------------------- */ 288 /* ----- Begin NH Hash Section ------------------------------------------ */ 289 /* ---------------------------------------------------------------------- */ 290 /* ---------------------------------------------------------------------- */ 291 292 /* The NH-based hash functions used in UMAC are described in the UMAC paper 293 * and specification, both of which can be found at the UMAC website. 294 * The interface to this implementation has two 295 * versions, one expects the entire message being hashed to be passed 296 * in a single buffer and returns the hash result immediately. The second 297 * allows the message to be passed in a sequence of buffers. In the 298 * muliple-buffer interface, the client calls the routine nh_update() as 299 * many times as necessary. When there is no more data to be fed to the 300 * hash, the client calls nh_final() which calculates the hash output. 301 * Before beginning another hash calculation the nh_reset() routine 302 * must be called. The single-buffer routine, nh(), is equivalent to 303 * the sequence of calls nh_update() and nh_final(); however it is 304 * optimized and should be prefered whenever the multiple-buffer interface 305 * is not necessary. When using either interface, it is the client's 306 * responsability to pass no more than L1_KEY_LEN bytes per hash result. 307 * 308 * The routine nh_init() initializes the nh_ctx data structure and 309 * must be called once, before any other PDF routine. 310 */ 311 312 /* The "nh_aux" routines do the actual NH hashing work. They 313 * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines 314 * produce output for all STREAMS NH iterations in one call, 315 * allowing the parallel implementation of the streams. 316 */ 317 318 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */ 319 #define L1_KEY_LEN 1024 /* Internal key bytes */ 320 #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */ 321 #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */ 322 #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */ 323 #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */ 324 325 typedef struct { 326 UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */ 327 UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */ 328 int next_data_empty; /* Bookeeping variable for data buffer. */ 329 int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */ 330 UINT64 state[STREAMS]; /* on-line state */ 331 } nh_ctx; 332 333 334 #if (UMAC_OUTPUT_LEN == 4) 335 336 static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) 337 /* NH hashing primitive. Previous (partial) hash result is loaded and 338 * then stored via hp pointer. The length of the data pointed at by "dp", 339 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key 340 * is expected to be endian compensated in memory at key setup. 341 */ 342 { 343 UINT64 h; 344 UWORD c = dlen / 32; 345 UINT32 *k = (UINT32 *)kp; 346 UINT32 *d = (UINT32 *)dp; 347 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 348 UINT32 k0,k1,k2,k3,k4,k5,k6,k7; 349 350 h = *((UINT64 *)hp); 351 do { 352 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 353 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 354 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 355 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 356 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 357 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 358 h += MUL64((k0 + d0), (k4 + d4)); 359 h += MUL64((k1 + d1), (k5 + d5)); 360 h += MUL64((k2 + d2), (k6 + d6)); 361 h += MUL64((k3 + d3), (k7 + d7)); 362 363 d += 8; 364 k += 8; 365 } while (--c); 366 *((UINT64 *)hp) = h; 367 } 368 369 #elif (UMAC_OUTPUT_LEN == 8) 370 371 static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) 372 /* Same as previous nh_aux, but two streams are handled in one pass, 373 * reading and writing 16 bytes of hash-state per call. 374 */ 375 { 376 UINT64 h1,h2; 377 UWORD c = dlen / 32; 378 UINT32 *k = (UINT32 *)kp; 379 UINT32 *d = (UINT32 *)dp; 380 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 381 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 382 k8,k9,k10,k11; 383 384 h1 = *((UINT64 *)hp); 385 h2 = *((UINT64 *)hp + 1); 386 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 387 do { 388 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 389 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 390 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 391 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 392 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 393 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 394 395 h1 += MUL64((k0 + d0), (k4 + d4)); 396 h2 += MUL64((k4 + d0), (k8 + d4)); 397 398 h1 += MUL64((k1 + d1), (k5 + d5)); 399 h2 += MUL64((k5 + d1), (k9 + d5)); 400 401 h1 += MUL64((k2 + d2), (k6 + d6)); 402 h2 += MUL64((k6 + d2), (k10 + d6)); 403 404 h1 += MUL64((k3 + d3), (k7 + d7)); 405 h2 += MUL64((k7 + d3), (k11 + d7)); 406 407 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 408 409 d += 8; 410 k += 8; 411 } while (--c); 412 ((UINT64 *)hp)[0] = h1; 413 ((UINT64 *)hp)[1] = h2; 414 } 415 416 #elif (UMAC_OUTPUT_LEN == 12) 417 418 static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) 419 /* Same as previous nh_aux, but two streams are handled in one pass, 420 * reading and writing 24 bytes of hash-state per call. 421 */ 422 { 423 UINT64 h1,h2,h3; 424 UWORD c = dlen / 32; 425 UINT32 *k = (UINT32 *)kp; 426 UINT32 *d = (UINT32 *)dp; 427 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 428 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 429 k8,k9,k10,k11,k12,k13,k14,k15; 430 431 h1 = *((UINT64 *)hp); 432 h2 = *((UINT64 *)hp + 1); 433 h3 = *((UINT64 *)hp + 2); 434 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 435 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 436 do { 437 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 438 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 439 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 440 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 441 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 442 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 443 444 h1 += MUL64((k0 + d0), (k4 + d4)); 445 h2 += MUL64((k4 + d0), (k8 + d4)); 446 h3 += MUL64((k8 + d0), (k12 + d4)); 447 448 h1 += MUL64((k1 + d1), (k5 + d5)); 449 h2 += MUL64((k5 + d1), (k9 + d5)); 450 h3 += MUL64((k9 + d1), (k13 + d5)); 451 452 h1 += MUL64((k2 + d2), (k6 + d6)); 453 h2 += MUL64((k6 + d2), (k10 + d6)); 454 h3 += MUL64((k10 + d2), (k14 + d6)); 455 456 h1 += MUL64((k3 + d3), (k7 + d7)); 457 h2 += MUL64((k7 + d3), (k11 + d7)); 458 h3 += MUL64((k11 + d3), (k15 + d7)); 459 460 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 461 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 462 463 d += 8; 464 k += 8; 465 } while (--c); 466 ((UINT64 *)hp)[0] = h1; 467 ((UINT64 *)hp)[1] = h2; 468 ((UINT64 *)hp)[2] = h3; 469 } 470 471 #elif (UMAC_OUTPUT_LEN == 16) 472 473 static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) 474 /* Same as previous nh_aux, but two streams are handled in one pass, 475 * reading and writing 24 bytes of hash-state per call. 476 */ 477 { 478 UINT64 h1,h2,h3,h4; 479 UWORD c = dlen / 32; 480 UINT32 *k = (UINT32 *)kp; 481 UINT32 *d = (UINT32 *)dp; 482 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 483 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 484 k8,k9,k10,k11,k12,k13,k14,k15, 485 k16,k17,k18,k19; 486 487 h1 = *((UINT64 *)hp); 488 h2 = *((UINT64 *)hp + 1); 489 h3 = *((UINT64 *)hp + 2); 490 h4 = *((UINT64 *)hp + 3); 491 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 492 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 493 do { 494 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 495 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 496 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 497 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 498 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 499 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 500 k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19); 501 502 h1 += MUL64((k0 + d0), (k4 + d4)); 503 h2 += MUL64((k4 + d0), (k8 + d4)); 504 h3 += MUL64((k8 + d0), (k12 + d4)); 505 h4 += MUL64((k12 + d0), (k16 + d4)); 506 507 h1 += MUL64((k1 + d1), (k5 + d5)); 508 h2 += MUL64((k5 + d1), (k9 + d5)); 509 h3 += MUL64((k9 + d1), (k13 + d5)); 510 h4 += MUL64((k13 + d1), (k17 + d5)); 511 512 h1 += MUL64((k2 + d2), (k6 + d6)); 513 h2 += MUL64((k6 + d2), (k10 + d6)); 514 h3 += MUL64((k10 + d2), (k14 + d6)); 515 h4 += MUL64((k14 + d2), (k18 + d6)); 516 517 h1 += MUL64((k3 + d3), (k7 + d7)); 518 h2 += MUL64((k7 + d3), (k11 + d7)); 519 h3 += MUL64((k11 + d3), (k15 + d7)); 520 h4 += MUL64((k15 + d3), (k19 + d7)); 521 522 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 523 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 524 k8 = k16; k9 = k17; k10 = k18; k11 = k19; 525 526 d += 8; 527 k += 8; 528 } while (--c); 529 ((UINT64 *)hp)[0] = h1; 530 ((UINT64 *)hp)[1] = h2; 531 ((UINT64 *)hp)[2] = h3; 532 ((UINT64 *)hp)[3] = h4; 533 } 534 535 /* ---------------------------------------------------------------------- */ 536 #endif /* UMAC_OUTPUT_LENGTH */ 537 /* ---------------------------------------------------------------------- */ 538 539 540 /* ---------------------------------------------------------------------- */ 541 542 static void nh_transform(nh_ctx *hc, UINT8 *buf, UINT32 nbytes) 543 /* This function is a wrapper for the primitive NH hash functions. It takes 544 * as argument "hc" the current hash context and a buffer which must be a 545 * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset 546 * appropriately according to how much message has been hashed already. 547 */ 548 { 549 UINT8 *key; 550 551 key = hc->nh_key + hc->bytes_hashed; 552 nh_aux(key, buf, hc->state, nbytes); 553 } 554 555 /* ---------------------------------------------------------------------- */ 556 557 #if (__LITTLE_ENDIAN__) 558 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes) 559 /* We endian convert the keys on little-endian computers to */ 560 /* compensate for the lack of big-endian memory reads during hashing. */ 561 { 562 UWORD iters = num_bytes / bpw; 563 if (bpw == 4) { 564 UINT32 *p = (UINT32 *)buf; 565 do { 566 *p = LOAD_UINT32_REVERSED(p); 567 p++; 568 } while (--iters); 569 } else if (bpw == 8) { 570 UINT32 *p = (UINT32 *)buf; 571 UINT32 t; 572 do { 573 t = LOAD_UINT32_REVERSED(p+1); 574 p[1] = LOAD_UINT32_REVERSED(p); 575 p[0] = t; 576 p += 2; 577 } while (--iters); 578 } 579 } 580 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) 581 #else 582 #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */ 583 #endif 584 585 /* ---------------------------------------------------------------------- */ 586 587 static void nh_reset(nh_ctx *hc) 588 /* Reset nh_ctx to ready for hashing of new data */ 589 { 590 hc->bytes_hashed = 0; 591 hc->next_data_empty = 0; 592 hc->state[0] = 0; 593 #if (UMAC_OUTPUT_LEN >= 8) 594 hc->state[1] = 0; 595 #endif 596 #if (UMAC_OUTPUT_LEN >= 12) 597 hc->state[2] = 0; 598 #endif 599 #if (UMAC_OUTPUT_LEN == 16) 600 hc->state[3] = 0; 601 #endif 602 603 } 604 605 /* ---------------------------------------------------------------------- */ 606 607 static void nh_init(nh_ctx *hc, aes_int_key prf_key) 608 /* Generate nh_key, endian convert and reset to be ready for hashing. */ 609 { 610 kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key)); 611 endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key)); 612 nh_reset(hc); 613 } 614 615 /* ---------------------------------------------------------------------- */ 616 617 static void nh_update(nh_ctx *hc, UINT8 *buf, UINT32 nbytes) 618 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */ 619 /* even multiple of HASH_BUF_BYTES. */ 620 { 621 UINT32 i,j; 622 623 j = hc->next_data_empty; 624 if ((j + nbytes) >= HASH_BUF_BYTES) { 625 if (j) { 626 i = HASH_BUF_BYTES - j; 627 memcpy(hc->data+j, buf, i); 628 nh_transform(hc,hc->data,HASH_BUF_BYTES); 629 nbytes -= i; 630 buf += i; 631 hc->bytes_hashed += HASH_BUF_BYTES; 632 } 633 if (nbytes >= HASH_BUF_BYTES) { 634 i = nbytes & ~(HASH_BUF_BYTES - 1); 635 nh_transform(hc, buf, i); 636 nbytes -= i; 637 buf += i; 638 hc->bytes_hashed += i; 639 } 640 j = 0; 641 } 642 memcpy(hc->data + j, buf, nbytes); 643 hc->next_data_empty = j + nbytes; 644 } 645 646 /* ---------------------------------------------------------------------- */ 647 648 static void zero_pad(UINT8 *p, int nbytes) 649 { 650 /* Write "nbytes" of zeroes, beginning at "p" */ 651 if (nbytes >= (int)sizeof(UWORD)) { 652 while ((ptrdiff_t)p % sizeof(UWORD)) { 653 *p = 0; 654 nbytes--; 655 p++; 656 } 657 while (nbytes >= (int)sizeof(UWORD)) { 658 *(UWORD *)p = 0; 659 nbytes -= sizeof(UWORD); 660 p += sizeof(UWORD); 661 } 662 } 663 while (nbytes) { 664 *p = 0; 665 nbytes--; 666 p++; 667 } 668 } 669 670 /* ---------------------------------------------------------------------- */ 671 672 static void nh_final(nh_ctx *hc, UINT8 *result) 673 /* After passing some number of data buffers to nh_update() for integration 674 * into an NH context, nh_final is called to produce a hash result. If any 675 * bytes are in the buffer hc->data, incorporate them into the 676 * NH context. Finally, add into the NH accumulation "state" the total number 677 * of bits hashed. The resulting numbers are written to the buffer "result". 678 * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated. 679 */ 680 { 681 int nh_len, nbits; 682 683 if (hc->next_data_empty != 0) { 684 nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) & 685 ~(L1_PAD_BOUNDARY - 1)); 686 zero_pad(hc->data + hc->next_data_empty, 687 nh_len - hc->next_data_empty); 688 nh_transform(hc, hc->data, nh_len); 689 hc->bytes_hashed += hc->next_data_empty; 690 } else if (hc->bytes_hashed == 0) { 691 nh_len = L1_PAD_BOUNDARY; 692 zero_pad(hc->data, L1_PAD_BOUNDARY); 693 nh_transform(hc, hc->data, nh_len); 694 } 695 696 nbits = (hc->bytes_hashed << 3); 697 ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits; 698 #if (UMAC_OUTPUT_LEN >= 8) 699 ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits; 700 #endif 701 #if (UMAC_OUTPUT_LEN >= 12) 702 ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits; 703 #endif 704 #if (UMAC_OUTPUT_LEN == 16) 705 ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits; 706 #endif 707 nh_reset(hc); 708 } 709 710 /* ---------------------------------------------------------------------- */ 711 712 static void nh(nh_ctx *hc, UINT8 *buf, UINT32 padded_len, 713 UINT32 unpadded_len, UINT8 *result) 714 /* All-in-one nh_update() and nh_final() equivalent. 715 * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is 716 * well aligned 717 */ 718 { 719 UINT32 nbits; 720 721 /* Initialize the hash state */ 722 nbits = (unpadded_len << 3); 723 724 ((UINT64 *)result)[0] = nbits; 725 #if (UMAC_OUTPUT_LEN >= 8) 726 ((UINT64 *)result)[1] = nbits; 727 #endif 728 #if (UMAC_OUTPUT_LEN >= 12) 729 ((UINT64 *)result)[2] = nbits; 730 #endif 731 #if (UMAC_OUTPUT_LEN == 16) 732 ((UINT64 *)result)[3] = nbits; 733 #endif 734 735 nh_aux(hc->nh_key, buf, result, padded_len); 736 } 737 738 /* ---------------------------------------------------------------------- */ 739 /* ---------------------------------------------------------------------- */ 740 /* ----- Begin UHASH Section -------------------------------------------- */ 741 /* ---------------------------------------------------------------------- */ 742 /* ---------------------------------------------------------------------- */ 743 744 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first 745 * hashed by NH. The NH output is then hashed by a polynomial-hash layer 746 * unless the initial data to be hashed is short. After the polynomial- 747 * layer, an inner-product hash is used to produce the final UHASH output. 748 * 749 * UHASH provides two interfaces, one all-at-once and another where data 750 * buffers are presented sequentially. In the sequential interface, the 751 * UHASH client calls the routine uhash_update() as many times as necessary. 752 * When there is no more data to be fed to UHASH, the client calls 753 * uhash_final() which 754 * calculates the UHASH output. Before beginning another UHASH calculation 755 * the uhash_reset() routine must be called. The all-at-once UHASH routine, 756 * uhash(), is equivalent to the sequence of calls uhash_update() and 757 * uhash_final(); however it is optimized and should be 758 * used whenever the sequential interface is not necessary. 759 * 760 * The routine uhash_init() initializes the uhash_ctx data structure and 761 * must be called once, before any other UHASH routine. 762 */ 763 764 /* ---------------------------------------------------------------------- */ 765 /* ----- Constants and uhash_ctx ---------------------------------------- */ 766 /* ---------------------------------------------------------------------- */ 767 768 /* ---------------------------------------------------------------------- */ 769 /* ----- Poly hash and Inner-Product hash Constants --------------------- */ 770 /* ---------------------------------------------------------------------- */ 771 772 /* Primes and masks */ 773 #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */ 774 #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */ 775 #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */ 776 777 778 /* ---------------------------------------------------------------------- */ 779 780 typedef struct uhash_ctx { 781 nh_ctx hash; /* Hash context for L1 NH hash */ 782 UINT64 poly_key_8[STREAMS]; /* p64 poly keys */ 783 UINT64 poly_accum[STREAMS]; /* poly hash result */ 784 UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */ 785 UINT32 ip_trans[STREAMS]; /* Inner-product translation */ 786 UINT32 msg_len; /* Total length of data passed */ 787 /* to uhash */ 788 } uhash_ctx; 789 typedef struct uhash_ctx *uhash_ctx_t; 790 791 /* ---------------------------------------------------------------------- */ 792 793 794 /* The polynomial hashes use Horner's rule to evaluate a polynomial one 795 * word at a time. As described in the specification, poly32 and poly64 796 * require keys from special domains. The following implementations exploit 797 * the special domains to avoid overflow. The results are not guaranteed to 798 * be within Z_p32 and Z_p64, but the Inner-Product hash implementation 799 * patches any errant values. 800 */ 801 802 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data) 803 { 804 UINT32 key_hi = (UINT32)(key >> 32), 805 key_lo = (UINT32)key, 806 cur_hi = (UINT32)(cur >> 32), 807 cur_lo = (UINT32)cur, 808 x_lo, 809 x_hi; 810 UINT64 X,T,res; 811 812 X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo); 813 x_lo = (UINT32)X; 814 x_hi = (UINT32)(X >> 32); 815 816 res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo); 817 818 T = ((UINT64)x_lo << 32); 819 res += T; 820 if (res < T) 821 res += 59; 822 823 res += data; 824 if (res < data) 825 res += 59; 826 827 return res; 828 } 829 830 831 /* Although UMAC is specified to use a ramped polynomial hash scheme, this 832 * implementation does not handle all ramp levels. Because we don't handle 833 * the ramp up to p128 modulus in this implementation, we are limited to 834 * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24 835 * bytes input to UMAC per tag, ie. 16MB). 836 */ 837 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[]) 838 { 839 int i; 840 UINT64 *data=(UINT64*)data_in; 841 842 for (i = 0; i < STREAMS; i++) { 843 if ((UINT32)(data[i] >> 32) == 0xfffffffful) { 844 hc->poly_accum[i] = poly64(hc->poly_accum[i], 845 hc->poly_key_8[i], p64 - 1); 846 hc->poly_accum[i] = poly64(hc->poly_accum[i], 847 hc->poly_key_8[i], (data[i] - 59)); 848 } else { 849 hc->poly_accum[i] = poly64(hc->poly_accum[i], 850 hc->poly_key_8[i], data[i]); 851 } 852 } 853 } 854 855 856 /* ---------------------------------------------------------------------- */ 857 858 859 /* The final step in UHASH is an inner-product hash. The poly hash 860 * produces a result not neccesarily WORD_LEN bytes long. The inner- 861 * product hash breaks the polyhash output into 16-bit chunks and 862 * multiplies each with a 36 bit key. 863 */ 864 865 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data) 866 { 867 t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48); 868 t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32); 869 t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16); 870 t = t + ipkp[3] * (UINT64)(UINT16)(data); 871 872 return t; 873 } 874 875 static UINT32 ip_reduce_p36(UINT64 t) 876 { 877 /* Divisionless modular reduction */ 878 UINT64 ret; 879 880 ret = (t & m36) + 5 * (t >> 36); 881 if (ret >= p36) 882 ret -= p36; 883 884 /* return least significant 32 bits */ 885 return (UINT32)(ret); 886 } 887 888 889 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then 890 * the polyhash stage is skipped and ip_short is applied directly to the 891 * NH output. 892 */ 893 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res) 894 { 895 UINT64 t; 896 UINT64 *nhp = (UINT64 *)nh_res; 897 898 t = ip_aux(0,ahc->ip_keys, nhp[0]); 899 STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]); 900 #if (UMAC_OUTPUT_LEN >= 8) 901 t = ip_aux(0,ahc->ip_keys+4, nhp[1]); 902 STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]); 903 #endif 904 #if (UMAC_OUTPUT_LEN >= 12) 905 t = ip_aux(0,ahc->ip_keys+8, nhp[2]); 906 STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]); 907 #endif 908 #if (UMAC_OUTPUT_LEN == 16) 909 t = ip_aux(0,ahc->ip_keys+12, nhp[3]); 910 STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]); 911 #endif 912 } 913 914 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then 915 * the polyhash stage is not skipped and ip_long is applied to the 916 * polyhash output. 917 */ 918 static void ip_long(uhash_ctx_t ahc, u_char *res) 919 { 920 int i; 921 UINT64 t; 922 923 for (i = 0; i < STREAMS; i++) { 924 /* fix polyhash output not in Z_p64 */ 925 if (ahc->poly_accum[i] >= p64) 926 ahc->poly_accum[i] -= p64; 927 t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]); 928 STORE_UINT32_BIG((UINT32 *)res+i, 929 ip_reduce_p36(t) ^ ahc->ip_trans[i]); 930 } 931 } 932 933 934 /* ---------------------------------------------------------------------- */ 935 936 /* ---------------------------------------------------------------------- */ 937 938 /* Reset uhash context for next hash session */ 939 static int uhash_reset(uhash_ctx_t pc) 940 { 941 nh_reset(&pc->hash); 942 pc->msg_len = 0; 943 pc->poly_accum[0] = 1; 944 #if (UMAC_OUTPUT_LEN >= 8) 945 pc->poly_accum[1] = 1; 946 #endif 947 #if (UMAC_OUTPUT_LEN >= 12) 948 pc->poly_accum[2] = 1; 949 #endif 950 #if (UMAC_OUTPUT_LEN == 16) 951 pc->poly_accum[3] = 1; 952 #endif 953 return 1; 954 } 955 956 /* ---------------------------------------------------------------------- */ 957 958 /* Given a pointer to the internal key needed by kdf() and a uhash context, 959 * initialize the NH context and generate keys needed for poly and inner- 960 * product hashing. All keys are endian adjusted in memory so that native 961 * loads cause correct keys to be in registers during calculation. 962 */ 963 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key) 964 { 965 int i; 966 UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)]; 967 968 /* Zero the entire uhash context */ 969 memset(ahc, 0, sizeof(uhash_ctx)); 970 971 /* Initialize the L1 hash */ 972 nh_init(&ahc->hash, prf_key); 973 974 /* Setup L2 hash variables */ 975 kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */ 976 for (i = 0; i < STREAMS; i++) { 977 /* Fill keys from the buffer, skipping bytes in the buffer not 978 * used by this implementation. Endian reverse the keys if on a 979 * little-endian computer. 980 */ 981 memcpy(ahc->poly_key_8+i, buf+24*i, 8); 982 endian_convert_if_le(ahc->poly_key_8+i, 8, 8); 983 /* Mask the 64-bit keys to their special domain */ 984 ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu; 985 ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */ 986 } 987 988 /* Setup L3-1 hash variables */ 989 kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */ 990 for (i = 0; i < STREAMS; i++) 991 memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64), 992 4*sizeof(UINT64)); 993 endian_convert_if_le(ahc->ip_keys, sizeof(UINT64), 994 sizeof(ahc->ip_keys)); 995 for (i = 0; i < STREAMS*4; i++) 996 ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */ 997 998 /* Setup L3-2 hash variables */ 999 /* Fill buffer with index 4 key */ 1000 kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32)); 1001 endian_convert_if_le(ahc->ip_trans, sizeof(UINT32), 1002 STREAMS * sizeof(UINT32)); 1003 } 1004 1005 /* ---------------------------------------------------------------------- */ 1006 1007 #if 0 1008 static uhash_ctx_t uhash_alloc(u_char key[]) 1009 { 1010 /* Allocate memory and force to a 16-byte boundary. */ 1011 uhash_ctx_t ctx; 1012 u_char bytes_to_add; 1013 aes_int_key prf_key; 1014 1015 ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY); 1016 if (ctx) { 1017 if (ALLOC_BOUNDARY) { 1018 bytes_to_add = ALLOC_BOUNDARY - 1019 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1)); 1020 ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add); 1021 *((u_char *)ctx - 1) = bytes_to_add; 1022 } 1023 aes_key_setup(key,prf_key); 1024 uhash_init(ctx, prf_key); 1025 } 1026 return (ctx); 1027 } 1028 #endif 1029 1030 /* ---------------------------------------------------------------------- */ 1031 1032 #if 0 1033 static int uhash_free(uhash_ctx_t ctx) 1034 { 1035 /* Free memory allocated by uhash_alloc */ 1036 u_char bytes_to_sub; 1037 1038 if (ctx) { 1039 if (ALLOC_BOUNDARY) { 1040 bytes_to_sub = *((u_char *)ctx - 1); 1041 ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub); 1042 } 1043 free(ctx); 1044 } 1045 return (1); 1046 } 1047 #endif 1048 /* ---------------------------------------------------------------------- */ 1049 1050 static int uhash_update(uhash_ctx_t ctx, u_char *input, long len) 1051 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and 1052 * hash each one with NH, calling the polyhash on each NH output. 1053 */ 1054 { 1055 UWORD bytes_hashed, bytes_remaining; 1056 UINT64 result_buf[STREAMS]; 1057 UINT8 *nh_result = (UINT8 *)&result_buf; 1058 1059 if (ctx->msg_len + len <= L1_KEY_LEN) { 1060 nh_update(&ctx->hash, (UINT8 *)input, len); 1061 ctx->msg_len += len; 1062 } else { 1063 1064 bytes_hashed = ctx->msg_len % L1_KEY_LEN; 1065 if (ctx->msg_len == L1_KEY_LEN) 1066 bytes_hashed = L1_KEY_LEN; 1067 1068 if (bytes_hashed + len >= L1_KEY_LEN) { 1069 1070 /* If some bytes have been passed to the hash function */ 1071 /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */ 1072 /* bytes to complete the current nh_block. */ 1073 if (bytes_hashed) { 1074 bytes_remaining = (L1_KEY_LEN - bytes_hashed); 1075 nh_update(&ctx->hash, (UINT8 *)input, bytes_remaining); 1076 nh_final(&ctx->hash, nh_result); 1077 ctx->msg_len += bytes_remaining; 1078 poly_hash(ctx,(UINT32 *)nh_result); 1079 len -= bytes_remaining; 1080 input += bytes_remaining; 1081 } 1082 1083 /* Hash directly from input stream if enough bytes */ 1084 while (len >= L1_KEY_LEN) { 1085 nh(&ctx->hash, (UINT8 *)input, L1_KEY_LEN, 1086 L1_KEY_LEN, nh_result); 1087 ctx->msg_len += L1_KEY_LEN; 1088 len -= L1_KEY_LEN; 1089 input += L1_KEY_LEN; 1090 poly_hash(ctx,(UINT32 *)nh_result); 1091 } 1092 } 1093 1094 /* pass remaining < L1_KEY_LEN bytes of input data to NH */ 1095 if (len) { 1096 nh_update(&ctx->hash, (UINT8 *)input, len); 1097 ctx->msg_len += len; 1098 } 1099 } 1100 1101 return (1); 1102 } 1103 1104 /* ---------------------------------------------------------------------- */ 1105 1106 static int uhash_final(uhash_ctx_t ctx, u_char *res) 1107 /* Incorporate any pending data, pad, and generate tag */ 1108 { 1109 UINT64 result_buf[STREAMS]; 1110 UINT8 *nh_result = (UINT8 *)&result_buf; 1111 1112 if (ctx->msg_len > L1_KEY_LEN) { 1113 if (ctx->msg_len % L1_KEY_LEN) { 1114 nh_final(&ctx->hash, nh_result); 1115 poly_hash(ctx,(UINT32 *)nh_result); 1116 } 1117 ip_long(ctx, res); 1118 } else { 1119 nh_final(&ctx->hash, nh_result); 1120 ip_short(ctx,nh_result, res); 1121 } 1122 uhash_reset(ctx); 1123 return (1); 1124 } 1125 1126 /* ---------------------------------------------------------------------- */ 1127 1128 #if 0 1129 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res) 1130 /* assumes that msg is in a writable buffer of length divisible by */ 1131 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */ 1132 { 1133 UINT8 nh_result[STREAMS*sizeof(UINT64)]; 1134 UINT32 nh_len; 1135 int extra_zeroes_needed; 1136 1137 /* If the message to be hashed is no longer than L1_HASH_LEN, we skip 1138 * the polyhash. 1139 */ 1140 if (len <= L1_KEY_LEN) { 1141 if (len == 0) /* If zero length messages will not */ 1142 nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */ 1143 else 1144 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1145 extra_zeroes_needed = nh_len - len; 1146 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1147 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1148 ip_short(ahc,nh_result, res); 1149 } else { 1150 /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH 1151 * output to poly_hash(). 1152 */ 1153 do { 1154 nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result); 1155 poly_hash(ahc,(UINT32 *)nh_result); 1156 len -= L1_KEY_LEN; 1157 msg += L1_KEY_LEN; 1158 } while (len >= L1_KEY_LEN); 1159 if (len) { 1160 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1161 extra_zeroes_needed = nh_len - len; 1162 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1163 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1164 poly_hash(ahc,(UINT32 *)nh_result); 1165 } 1166 1167 ip_long(ahc, res); 1168 } 1169 1170 uhash_reset(ahc); 1171 return 1; 1172 } 1173 #endif 1174 1175 /* ---------------------------------------------------------------------- */ 1176 /* ---------------------------------------------------------------------- */ 1177 /* ----- Begin UMAC Section --------------------------------------------- */ 1178 /* ---------------------------------------------------------------------- */ 1179 /* ---------------------------------------------------------------------- */ 1180 1181 /* The UMAC interface has two interfaces, an all-at-once interface where 1182 * the entire message to be authenticated is passed to UMAC in one buffer, 1183 * and a sequential interface where the message is presented a little at a 1184 * time. The all-at-once is more optimaized than the sequential version and 1185 * should be preferred when the sequential interface is not required. 1186 */ 1187 struct umac_ctx { 1188 uhash_ctx hash; /* Hash function for message compression */ 1189 pdf_ctx pdf; /* PDF for hashed output */ 1190 void *free_ptr; /* Address to free this struct via */ 1191 } umac_ctx; 1192 1193 /* ---------------------------------------------------------------------- */ 1194 1195 #if 0 1196 int umac_reset(struct umac_ctx *ctx) 1197 /* Reset the hash function to begin a new authentication. */ 1198 { 1199 uhash_reset(&ctx->hash); 1200 return (1); 1201 } 1202 #endif 1203 1204 /* ---------------------------------------------------------------------- */ 1205 1206 int umac_delete(struct umac_ctx *ctx) 1207 /* Deallocate the ctx structure */ 1208 { 1209 if (ctx) { 1210 if (ALLOC_BOUNDARY) 1211 ctx = (struct umac_ctx *)ctx->free_ptr; 1212 xfree(ctx); 1213 } 1214 return (1); 1215 } 1216 1217 /* ---------------------------------------------------------------------- */ 1218 1219 struct umac_ctx *umac_new(u_char key[]) 1220 /* Dynamically allocate a umac_ctx struct, initialize variables, 1221 * generate subkeys from key. Align to 16-byte boundary. 1222 */ 1223 { 1224 struct umac_ctx *ctx, *octx; 1225 size_t bytes_to_add; 1226 aes_int_key prf_key; 1227 1228 octx = ctx = xmalloc(sizeof(*ctx) + ALLOC_BOUNDARY); 1229 if (ctx) { 1230 if (ALLOC_BOUNDARY) { 1231 bytes_to_add = ALLOC_BOUNDARY - 1232 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1)); 1233 ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add); 1234 } 1235 ctx->free_ptr = octx; 1236 aes_key_setup(key,prf_key); 1237 pdf_init(&ctx->pdf, prf_key); 1238 uhash_init(&ctx->hash, prf_key); 1239 } 1240 1241 return (ctx); 1242 } 1243 1244 /* ---------------------------------------------------------------------- */ 1245 1246 int umac_final(struct umac_ctx *ctx, u_char tag[], u_char nonce[8]) 1247 /* Incorporate any pending data, pad, and generate tag */ 1248 { 1249 uhash_final(&ctx->hash, (u_char *)tag); 1250 pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); 1251 1252 return (1); 1253 } 1254 1255 /* ---------------------------------------------------------------------- */ 1256 1257 int umac_update(struct umac_ctx *ctx, u_char *input, long len) 1258 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */ 1259 /* hash each one, calling the PDF on the hashed output whenever the hash- */ 1260 /* output buffer is full. */ 1261 { 1262 uhash_update(&ctx->hash, input, len); 1263 return (1); 1264 } 1265 1266 /* ---------------------------------------------------------------------- */ 1267 1268 #if 0 1269 int umac(struct umac_ctx *ctx, u_char *input, 1270 long len, u_char tag[], 1271 u_char nonce[8]) 1272 /* All-in-one version simply calls umac_update() and umac_final(). */ 1273 { 1274 uhash(&ctx->hash, input, len, (u_char *)tag); 1275 pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); 1276 1277 return (1); 1278 } 1279 #endif 1280 1281 /* ---------------------------------------------------------------------- */ 1282 /* ---------------------------------------------------------------------- */ 1283 /* ----- End UMAC Section ----------------------------------------------- */ 1284 /* ---------------------------------------------------------------------- */ 1285 /* ---------------------------------------------------------------------- */ 1286