1 /* $NetBSD: umac.c,v 1.4 2013/11/08 19:18:25 christos Exp $ */ 2 /* $OpenBSD: umac.c,v 1.7.2.1 2013/11/08 01:33:56 djm Exp $ */ 3 /* ----------------------------------------------------------------------- 4 * 5 * umac.c -- C Implementation UMAC Message Authentication 6 * 7 * Version 0.93b of rfc4418.txt -- 2006 July 18 8 * 9 * For a full description of UMAC message authentication see the UMAC 10 * world-wide-web page at http://www.cs.ucdavis.edu/~rogaway/umac 11 * Please report bugs and suggestions to the UMAC webpage. 12 * 13 * Copyright (c) 1999-2006 Ted Krovetz 14 * 15 * Permission to use, copy, modify, and distribute this software and 16 * its documentation for any purpose and with or without fee, is hereby 17 * granted provided that the above copyright notice appears in all copies 18 * and in supporting documentation, and that the name of the copyright 19 * holder not be used in advertising or publicity pertaining to 20 * distribution of the software without specific, written prior permission. 21 * 22 * Comments should be directed to Ted Krovetz (tdk@acm.org) 23 * 24 * ---------------------------------------------------------------------- */ 25 26 /* ////////////////////// IMPORTANT NOTES ///////////////////////////////// 27 * 28 * 1) This version does not work properly on messages larger than 16MB 29 * 30 * 2) If you set the switch to use SSE2, then all data must be 16-byte 31 * aligned 32 * 33 * 3) When calling the function umac(), it is assumed that msg is in 34 * a writable buffer of length divisible by 32 bytes. The message itself 35 * does not have to fill the entire buffer, but bytes beyond msg may be 36 * zeroed. 37 * 38 * 4) Three free AES implementations are supported by this implementation of 39 * UMAC. Paulo Barreto's version is in the public domain and can be found 40 * at http://www.esat.kuleuven.ac.be/~rijmen/rijndael/ (search for 41 * "Barreto"). The only two files needed are rijndael-alg-fst.c and 42 * rijndael-alg-fst.h. Brian Gladman's version is distributed with the GNU 43 * Public lisence at http://fp.gladman.plus.com/AES/index.htm. It 44 * includes a fast IA-32 assembly version. The OpenSSL crypo library is 45 * the third. 46 * 47 * 5) With FORCE_C_ONLY flags set to 0, incorrect results are sometimes 48 * produced under gcc with optimizations set -O3 or higher. Dunno why. 49 * 50 /////////////////////////////////////////////////////////////////////// */ 51 52 /* ---------------------------------------------------------------------- */ 53 /* --- User Switches ---------------------------------------------------- */ 54 /* ---------------------------------------------------------------------- */ 55 56 #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */ 57 /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */ 58 /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */ 59 /* #define SSE2 0 Is SSE2 is available? */ 60 /* #define RUN_TESTS 0 Run basic correctness/speed tests */ 61 /* #define UMAC_AE_SUPPORT 0 Enable auhthenticated encrytion */ 62 63 /* ---------------------------------------------------------------------- */ 64 /* -- Global Includes --------------------------------------------------- */ 65 /* ---------------------------------------------------------------------- */ 66 67 #include "includes.h" 68 __RCSID("$NetBSD: umac.c,v 1.4 2013/11/08 19:18:25 christos Exp $"); 69 #include <sys/types.h> 70 #include <sys/endian.h> 71 72 #include "xmalloc.h" 73 #include "umac.h" 74 #include <string.h> 75 #include <stdlib.h> 76 #include <stddef.h> 77 78 /* ---------------------------------------------------------------------- */ 79 /* --- Primitive Data Types --- */ 80 /* ---------------------------------------------------------------------- */ 81 82 /* The following assumptions may need change on your system */ 83 typedef u_int8_t UINT8; /* 1 byte */ 84 typedef u_int16_t UINT16; /* 2 byte */ 85 typedef u_int32_t UINT32; /* 4 byte */ 86 typedef u_int64_t UINT64; /* 8 bytes */ 87 typedef unsigned int UWORD; /* Register */ 88 89 /* ---------------------------------------------------------------------- */ 90 /* --- Constants -------------------------------------------------------- */ 91 /* ---------------------------------------------------------------------- */ 92 93 #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */ 94 95 /* Message "words" are read from memory in an endian-specific manner. */ 96 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */ 97 /* be set true if the host computer is little-endian. */ 98 99 #if BYTE_ORDER == LITTLE_ENDIAN 100 #define __LITTLE_ENDIAN__ 1 101 #else 102 #define __LITTLE_ENDIAN__ 0 103 #endif 104 105 /* ---------------------------------------------------------------------- */ 106 /* ---------------------------------------------------------------------- */ 107 /* ----- Architecture Specific ------------------------------------------ */ 108 /* ---------------------------------------------------------------------- */ 109 /* ---------------------------------------------------------------------- */ 110 111 112 /* ---------------------------------------------------------------------- */ 113 /* ---------------------------------------------------------------------- */ 114 /* ----- Primitive Routines --------------------------------------------- */ 115 /* ---------------------------------------------------------------------- */ 116 /* ---------------------------------------------------------------------- */ 117 118 119 /* ---------------------------------------------------------------------- */ 120 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */ 121 /* ---------------------------------------------------------------------- */ 122 123 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b))) 124 125 #if defined(__NetBSD__) 126 #include <sys/endian.h> 127 #define LOAD_UINT32_LITTLE(ptr) le32toh(*ptr) 128 #define STORE_UINT32_BIG(ptr,x) (*(UINT32 *)(ptr) = htobe32(x)) 129 #define LOAD_UINT32_REVERSED(p) (bswap32(*(UINT32 *)(p))) 130 #define STORE_UINT32_REVERSED(p,v) (*(UINT32 *)(p) = bswap32(v)) 131 #else /* !NetBSD */ 132 133 /* ---------------------------------------------------------------------- */ 134 /* --- Endian Conversion --- Forcing assembly on some platforms */ 135 136 /* ---------------------------------------------------------------------- */ 137 /* --- Endian Conversion --- Forcing assembly on some platforms */ 138 /* ---------------------------------------------------------------------- */ 139 140 #if !defined(__OpenBSD__) 141 static UINT32 LOAD_UINT32_REVERSED(const void *ptr) 142 { 143 UINT32 temp = *(const UINT32 *)ptr; 144 temp = (temp >> 24) | ((temp & 0x00FF0000) >> 8 ) 145 | ((temp & 0x0000FF00) << 8 ) | (temp << 24); 146 return (UINT32)temp; 147 } 148 149 static void STORE_UINT32_REVERSED(void *ptr, UINT32 x) 150 { 151 UINT32 i = (UINT32)x; 152 *(UINT32 *)ptr = (i >> 24) | ((i & 0x00FF0000) >> 8 ) 153 | ((i & 0x0000FF00) << 8 ) | (i << 24); 154 } 155 #endif 156 157 #else 158 /* The following definitions use the above reversal-primitives to do the right 159 * thing on endian specific load and stores. 160 */ 161 162 #define LOAD_UINT32_REVERSED(p) (swap32(*(const UINT32 *)(p))) 163 #define STORE_UINT32_REVERSED(p,v) (*(UINT32 *)(p) = swap32(v)) 164 #endif 165 166 #if (__LITTLE_ENDIAN__) 167 #define LOAD_UINT32_LITTLE(ptr) (*(const UINT32 *)(ptr)) 168 #define STORE_UINT32_BIG(ptr,x) STORE_UINT32_REVERSED(ptr,x) 169 #else 170 #define LOAD_UINT32_LITTLE(ptr) LOAD_UINT32_REVERSED(ptr) 171 #define STORE_UINT32_BIG(ptr,x) (*(UINT32 *)(ptr) = (UINT32)(x)) 172 #endif 173 #endif /*!NetBSD*/ 174 175 176 177 /* ---------------------------------------------------------------------- */ 178 /* ---------------------------------------------------------------------- */ 179 /* ----- Begin KDF & PDF Section ---------------------------------------- */ 180 /* ---------------------------------------------------------------------- */ 181 /* ---------------------------------------------------------------------- */ 182 183 /* UMAC uses AES with 16 byte block and key lengths */ 184 #define AES_BLOCK_LEN 16 185 186 /* OpenSSL's AES */ 187 #include <openssl/aes.h> 188 typedef AES_KEY aes_int_key[1]; 189 #define aes_encryption(in,out,int_key) \ 190 AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key) 191 #define aes_key_setup(key,int_key) \ 192 AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key) 193 194 /* The user-supplied UMAC key is stretched using AES in a counter 195 * mode to supply all random bits needed by UMAC. The kdf function takes 196 * an AES internal key representation 'key' and writes a stream of 197 * 'nbytes' bytes to the memory pointed at by 'buffer_ptr'. Each distinct 198 * 'ndx' causes a distinct byte stream. 199 */ 200 static void kdf(void *buffer_ptr, aes_int_key key, UINT8 ndx, int nbytes) 201 { 202 UINT8 in_buf[AES_BLOCK_LEN] = {0}; 203 UINT8 out_buf[AES_BLOCK_LEN]; 204 UINT8 *dst_buf = (UINT8 *)buffer_ptr; 205 int i; 206 207 /* Setup the initial value */ 208 in_buf[AES_BLOCK_LEN-9] = ndx; 209 in_buf[AES_BLOCK_LEN-1] = i = 1; 210 211 while (nbytes >= AES_BLOCK_LEN) { 212 aes_encryption(in_buf, out_buf, key); 213 memcpy(dst_buf,out_buf,AES_BLOCK_LEN); 214 in_buf[AES_BLOCK_LEN-1] = ++i; 215 nbytes -= AES_BLOCK_LEN; 216 dst_buf += AES_BLOCK_LEN; 217 } 218 if (nbytes) { 219 aes_encryption(in_buf, out_buf, key); 220 memcpy(dst_buf,out_buf,nbytes); 221 } 222 } 223 224 /* The final UHASH result is XOR'd with the output of a pseudorandom 225 * function. Here, we use AES to generate random output and 226 * xor the appropriate bytes depending on the last bits of nonce. 227 * This scheme is optimized for sequential, increasing big-endian nonces. 228 */ 229 230 typedef struct { 231 UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */ 232 UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */ 233 aes_int_key prf_key; /* Expanded AES key for PDF */ 234 } pdf_ctx; 235 236 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key) 237 { 238 UINT8 buf[UMAC_KEY_LEN]; 239 240 kdf(buf, prf_key, 0, UMAC_KEY_LEN); 241 aes_key_setup(buf, pc->prf_key); 242 243 /* Initialize pdf and cache */ 244 memset(pc->nonce, 0, sizeof(pc->nonce)); 245 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 246 } 247 248 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8]) 249 { 250 /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes 251 * of the AES output. If last time around we returned the ndx-1st 252 * element, then we may have the result in the cache already. 253 */ 254 255 #if (UMAC_OUTPUT_LEN == 4) 256 #define LOW_BIT_MASK 3 257 #elif (UMAC_OUTPUT_LEN == 8) 258 #define LOW_BIT_MASK 1 259 #elif (UMAC_OUTPUT_LEN > 8) 260 #define LOW_BIT_MASK 0 261 #endif 262 union { 263 UINT8 tmp_nonce_lo[4]; 264 UINT32 align; 265 } t; 266 #if LOW_BIT_MASK != 0 267 int ndx = nonce[7] & LOW_BIT_MASK; 268 #endif 269 *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1]; 270 t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */ 271 272 if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) || 273 (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) ) 274 { 275 ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0]; 276 ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0]; 277 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 278 } 279 280 #if (UMAC_OUTPUT_LEN == 4) 281 *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx]; 282 #elif (UMAC_OUTPUT_LEN == 8) 283 *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx]; 284 #elif (UMAC_OUTPUT_LEN == 12) 285 ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; 286 ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2]; 287 #elif (UMAC_OUTPUT_LEN == 16) 288 ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; 289 ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1]; 290 #endif 291 } 292 293 /* ---------------------------------------------------------------------- */ 294 /* ---------------------------------------------------------------------- */ 295 /* ----- Begin NH Hash Section ------------------------------------------ */ 296 /* ---------------------------------------------------------------------- */ 297 /* ---------------------------------------------------------------------- */ 298 299 /* The NH-based hash functions used in UMAC are described in the UMAC paper 300 * and specification, both of which can be found at the UMAC website. 301 * The interface to this implementation has two 302 * versions, one expects the entire message being hashed to be passed 303 * in a single buffer and returns the hash result immediately. The second 304 * allows the message to be passed in a sequence of buffers. In the 305 * muliple-buffer interface, the client calls the routine nh_update() as 306 * many times as necessary. When there is no more data to be fed to the 307 * hash, the client calls nh_final() which calculates the hash output. 308 * Before beginning another hash calculation the nh_reset() routine 309 * must be called. The single-buffer routine, nh(), is equivalent to 310 * the sequence of calls nh_update() and nh_final(); however it is 311 * optimized and should be prefered whenever the multiple-buffer interface 312 * is not necessary. When using either interface, it is the client's 313 * responsability to pass no more than L1_KEY_LEN bytes per hash result. 314 * 315 * The routine nh_init() initializes the nh_ctx data structure and 316 * must be called once, before any other PDF routine. 317 */ 318 319 /* The "nh_aux" routines do the actual NH hashing work. They 320 * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines 321 * produce output for all STREAMS NH iterations in one call, 322 * allowing the parallel implementation of the streams. 323 */ 324 325 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */ 326 #define L1_KEY_LEN 1024 /* Internal key bytes */ 327 #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */ 328 #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */ 329 #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */ 330 #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */ 331 332 typedef struct { 333 UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */ 334 UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */ 335 int next_data_empty; /* Bookeeping variable for data buffer. */ 336 int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */ 337 UINT64 state[STREAMS]; /* on-line state */ 338 } nh_ctx; 339 340 341 #if (UMAC_OUTPUT_LEN == 4) 342 343 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 344 /* NH hashing primitive. Previous (partial) hash result is loaded and 345 * then stored via hp pointer. The length of the data pointed at by "dp", 346 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key 347 * is expected to be endian compensated in memory at key setup. 348 */ 349 { 350 UINT64 h; 351 UWORD c = dlen / 32; 352 UINT32 *k = (UINT32 *)kp; 353 const UINT32 *d = (const UINT32 *)dp; 354 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 355 UINT32 k0,k1,k2,k3,k4,k5,k6,k7; 356 357 h = *((UINT64 *)hp); 358 do { 359 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 360 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 361 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 362 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 363 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 364 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 365 h += MUL64((k0 + d0), (k4 + d4)); 366 h += MUL64((k1 + d1), (k5 + d5)); 367 h += MUL64((k2 + d2), (k6 + d6)); 368 h += MUL64((k3 + d3), (k7 + d7)); 369 370 d += 8; 371 k += 8; 372 } while (--c); 373 *((UINT64 *)hp) = h; 374 } 375 376 #elif (UMAC_OUTPUT_LEN == 8) 377 378 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 379 /* Same as previous nh_aux, but two streams are handled in one pass, 380 * reading and writing 16 bytes of hash-state per call. 381 */ 382 { 383 UINT64 h1,h2; 384 UWORD c = dlen / 32; 385 UINT32 *k = (UINT32 *)kp; 386 const UINT32 *d = (const UINT32 *)dp; 387 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 388 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 389 k8,k9,k10,k11; 390 391 h1 = *((UINT64 *)hp); 392 h2 = *((UINT64 *)hp + 1); 393 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 394 do { 395 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 396 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 397 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 398 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 399 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 400 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 401 402 h1 += MUL64((k0 + d0), (k4 + d4)); 403 h2 += MUL64((k4 + d0), (k8 + d4)); 404 405 h1 += MUL64((k1 + d1), (k5 + d5)); 406 h2 += MUL64((k5 + d1), (k9 + d5)); 407 408 h1 += MUL64((k2 + d2), (k6 + d6)); 409 h2 += MUL64((k6 + d2), (k10 + d6)); 410 411 h1 += MUL64((k3 + d3), (k7 + d7)); 412 h2 += MUL64((k7 + d3), (k11 + d7)); 413 414 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 415 416 d += 8; 417 k += 8; 418 } while (--c); 419 ((UINT64 *)hp)[0] = h1; 420 ((UINT64 *)hp)[1] = h2; 421 } 422 423 #elif (UMAC_OUTPUT_LEN == 12) 424 425 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 426 /* Same as previous nh_aux, but two streams are handled in one pass, 427 * reading and writing 24 bytes of hash-state per call. 428 */ 429 { 430 UINT64 h1,h2,h3; 431 UWORD c = dlen / 32; 432 UINT32 *k = (UINT32 *)kp; 433 const UINT32 *d = (const UINT32 *)dp; 434 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 435 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 436 k8,k9,k10,k11,k12,k13,k14,k15; 437 438 h1 = *((UINT64 *)hp); 439 h2 = *((UINT64 *)hp + 1); 440 h3 = *((UINT64 *)hp + 2); 441 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 442 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 443 do { 444 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 445 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 446 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 447 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 448 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 449 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 450 451 h1 += MUL64((k0 + d0), (k4 + d4)); 452 h2 += MUL64((k4 + d0), (k8 + d4)); 453 h3 += MUL64((k8 + d0), (k12 + d4)); 454 455 h1 += MUL64((k1 + d1), (k5 + d5)); 456 h2 += MUL64((k5 + d1), (k9 + d5)); 457 h3 += MUL64((k9 + d1), (k13 + d5)); 458 459 h1 += MUL64((k2 + d2), (k6 + d6)); 460 h2 += MUL64((k6 + d2), (k10 + d6)); 461 h3 += MUL64((k10 + d2), (k14 + d6)); 462 463 h1 += MUL64((k3 + d3), (k7 + d7)); 464 h2 += MUL64((k7 + d3), (k11 + d7)); 465 h3 += MUL64((k11 + d3), (k15 + d7)); 466 467 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 468 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 469 470 d += 8; 471 k += 8; 472 } while (--c); 473 ((UINT64 *)hp)[0] = h1; 474 ((UINT64 *)hp)[1] = h2; 475 ((UINT64 *)hp)[2] = h3; 476 } 477 478 #elif (UMAC_OUTPUT_LEN == 16) 479 480 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 481 /* Same as previous nh_aux, but two streams are handled in one pass, 482 * reading and writing 24 bytes of hash-state per call. 483 */ 484 { 485 UINT64 h1,h2,h3,h4; 486 UWORD c = dlen / 32; 487 UINT32 *k = (UINT32 *)kp; 488 const UINT32 *d = (const UINT32 *)dp; 489 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 490 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 491 k8,k9,k10,k11,k12,k13,k14,k15, 492 k16,k17,k18,k19; 493 494 h1 = *((UINT64 *)hp); 495 h2 = *((UINT64 *)hp + 1); 496 h3 = *((UINT64 *)hp + 2); 497 h4 = *((UINT64 *)hp + 3); 498 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 499 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 500 do { 501 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 502 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 503 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 504 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 505 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 506 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 507 k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19); 508 509 h1 += MUL64((k0 + d0), (k4 + d4)); 510 h2 += MUL64((k4 + d0), (k8 + d4)); 511 h3 += MUL64((k8 + d0), (k12 + d4)); 512 h4 += MUL64((k12 + d0), (k16 + d4)); 513 514 h1 += MUL64((k1 + d1), (k5 + d5)); 515 h2 += MUL64((k5 + d1), (k9 + d5)); 516 h3 += MUL64((k9 + d1), (k13 + d5)); 517 h4 += MUL64((k13 + d1), (k17 + d5)); 518 519 h1 += MUL64((k2 + d2), (k6 + d6)); 520 h2 += MUL64((k6 + d2), (k10 + d6)); 521 h3 += MUL64((k10 + d2), (k14 + d6)); 522 h4 += MUL64((k14 + d2), (k18 + d6)); 523 524 h1 += MUL64((k3 + d3), (k7 + d7)); 525 h2 += MUL64((k7 + d3), (k11 + d7)); 526 h3 += MUL64((k11 + d3), (k15 + d7)); 527 h4 += MUL64((k15 + d3), (k19 + d7)); 528 529 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 530 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 531 k8 = k16; k9 = k17; k10 = k18; k11 = k19; 532 533 d += 8; 534 k += 8; 535 } while (--c); 536 ((UINT64 *)hp)[0] = h1; 537 ((UINT64 *)hp)[1] = h2; 538 ((UINT64 *)hp)[2] = h3; 539 ((UINT64 *)hp)[3] = h4; 540 } 541 542 /* ---------------------------------------------------------------------- */ 543 #endif /* UMAC_OUTPUT_LENGTH */ 544 /* ---------------------------------------------------------------------- */ 545 546 547 /* ---------------------------------------------------------------------- */ 548 549 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes) 550 /* This function is a wrapper for the primitive NH hash functions. It takes 551 * as argument "hc" the current hash context and a buffer which must be a 552 * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset 553 * appropriately according to how much message has been hashed already. 554 */ 555 { 556 UINT8 *key; 557 558 key = hc->nh_key + hc->bytes_hashed; 559 nh_aux(key, buf, hc->state, nbytes); 560 } 561 562 /* ---------------------------------------------------------------------- */ 563 564 #if (__LITTLE_ENDIAN__) 565 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) 566 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes) 567 /* We endian convert the keys on little-endian computers to */ 568 /* compensate for the lack of big-endian memory reads during hashing. */ 569 { 570 UWORD iters = num_bytes / bpw; 571 if (bpw == 4) { 572 UINT32 *p = (UINT32 *)buf; 573 do { 574 *p = LOAD_UINT32_REVERSED(p); 575 p++; 576 } while (--iters); 577 } else if (bpw == 8) { 578 UINT32 *p = (UINT32 *)buf; 579 UINT32 t; 580 do { 581 t = LOAD_UINT32_REVERSED(p+1); 582 p[1] = LOAD_UINT32_REVERSED(p); 583 p[0] = t; 584 p += 2; 585 } while (--iters); 586 } 587 } 588 #if (__LITTLE_ENDIAN__) 589 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) 590 #else 591 #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */ 592 #endif 593 594 /* ---------------------------------------------------------------------- */ 595 596 static void nh_reset(nh_ctx *hc) 597 /* Reset nh_ctx to ready for hashing of new data */ 598 { 599 hc->bytes_hashed = 0; 600 hc->next_data_empty = 0; 601 hc->state[0] = 0; 602 #if (UMAC_OUTPUT_LEN >= 8) 603 hc->state[1] = 0; 604 #endif 605 #if (UMAC_OUTPUT_LEN >= 12) 606 hc->state[2] = 0; 607 #endif 608 #if (UMAC_OUTPUT_LEN == 16) 609 hc->state[3] = 0; 610 #endif 611 612 } 613 614 /* ---------------------------------------------------------------------- */ 615 616 static void nh_init(nh_ctx *hc, aes_int_key prf_key) 617 /* Generate nh_key, endian convert and reset to be ready for hashing. */ 618 { 619 kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key)); 620 endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key)); 621 nh_reset(hc); 622 } 623 624 /* ---------------------------------------------------------------------- */ 625 626 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes) 627 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */ 628 /* even multiple of HASH_BUF_BYTES. */ 629 { 630 UINT32 i,j; 631 632 j = hc->next_data_empty; 633 if ((j + nbytes) >= HASH_BUF_BYTES) { 634 if (j) { 635 i = HASH_BUF_BYTES - j; 636 memcpy(hc->data+j, buf, i); 637 nh_transform(hc,hc->data,HASH_BUF_BYTES); 638 nbytes -= i; 639 buf += i; 640 hc->bytes_hashed += HASH_BUF_BYTES; 641 } 642 if (nbytes >= HASH_BUF_BYTES) { 643 i = nbytes & ~(HASH_BUF_BYTES - 1); 644 nh_transform(hc, buf, i); 645 nbytes -= i; 646 buf += i; 647 hc->bytes_hashed += i; 648 } 649 j = 0; 650 } 651 memcpy(hc->data + j, buf, nbytes); 652 hc->next_data_empty = j + nbytes; 653 } 654 655 /* ---------------------------------------------------------------------- */ 656 657 static void zero_pad(UINT8 *p, int nbytes) 658 { 659 /* Write "nbytes" of zeroes, beginning at "p" */ 660 if (nbytes >= (int)sizeof(UWORD)) { 661 while ((ptrdiff_t)p % sizeof(UWORD)) { 662 *p = 0; 663 nbytes--; 664 p++; 665 } 666 while (nbytes >= (int)sizeof(UWORD)) { 667 *(UWORD *)p = 0; 668 nbytes -= sizeof(UWORD); 669 p += sizeof(UWORD); 670 } 671 } 672 while (nbytes) { 673 *p = 0; 674 nbytes--; 675 p++; 676 } 677 } 678 679 /* ---------------------------------------------------------------------- */ 680 681 static void nh_final(nh_ctx *hc, UINT8 *result) 682 /* After passing some number of data buffers to nh_update() for integration 683 * into an NH context, nh_final is called to produce a hash result. If any 684 * bytes are in the buffer hc->data, incorporate them into the 685 * NH context. Finally, add into the NH accumulation "state" the total number 686 * of bits hashed. The resulting numbers are written to the buffer "result". 687 * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated. 688 */ 689 { 690 int nh_len, nbits; 691 692 if (hc->next_data_empty != 0) { 693 nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) & 694 ~(L1_PAD_BOUNDARY - 1)); 695 zero_pad(hc->data + hc->next_data_empty, 696 nh_len - hc->next_data_empty); 697 nh_transform(hc, hc->data, nh_len); 698 hc->bytes_hashed += hc->next_data_empty; 699 } else if (hc->bytes_hashed == 0) { 700 nh_len = L1_PAD_BOUNDARY; 701 zero_pad(hc->data, L1_PAD_BOUNDARY); 702 nh_transform(hc, hc->data, nh_len); 703 } 704 705 nbits = (hc->bytes_hashed << 3); 706 ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits; 707 #if (UMAC_OUTPUT_LEN >= 8) 708 ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits; 709 #endif 710 #if (UMAC_OUTPUT_LEN >= 12) 711 ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits; 712 #endif 713 #if (UMAC_OUTPUT_LEN == 16) 714 ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits; 715 #endif 716 nh_reset(hc); 717 } 718 719 /* ---------------------------------------------------------------------- */ 720 721 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len, 722 UINT32 unpadded_len, UINT8 *result) 723 /* All-in-one nh_update() and nh_final() equivalent. 724 * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is 725 * well aligned 726 */ 727 { 728 UINT32 nbits; 729 730 /* Initialize the hash state */ 731 nbits = (unpadded_len << 3); 732 733 ((UINT64 *)result)[0] = nbits; 734 #if (UMAC_OUTPUT_LEN >= 8) 735 ((UINT64 *)result)[1] = nbits; 736 #endif 737 #if (UMAC_OUTPUT_LEN >= 12) 738 ((UINT64 *)result)[2] = nbits; 739 #endif 740 #if (UMAC_OUTPUT_LEN == 16) 741 ((UINT64 *)result)[3] = nbits; 742 #endif 743 744 nh_aux(hc->nh_key, buf, result, padded_len); 745 } 746 747 /* ---------------------------------------------------------------------- */ 748 /* ---------------------------------------------------------------------- */ 749 /* ----- Begin UHASH Section -------------------------------------------- */ 750 /* ---------------------------------------------------------------------- */ 751 /* ---------------------------------------------------------------------- */ 752 753 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first 754 * hashed by NH. The NH output is then hashed by a polynomial-hash layer 755 * unless the initial data to be hashed is short. After the polynomial- 756 * layer, an inner-product hash is used to produce the final UHASH output. 757 * 758 * UHASH provides two interfaces, one all-at-once and another where data 759 * buffers are presented sequentially. In the sequential interface, the 760 * UHASH client calls the routine uhash_update() as many times as necessary. 761 * When there is no more data to be fed to UHASH, the client calls 762 * uhash_final() which 763 * calculates the UHASH output. Before beginning another UHASH calculation 764 * the uhash_reset() routine must be called. The all-at-once UHASH routine, 765 * uhash(), is equivalent to the sequence of calls uhash_update() and 766 * uhash_final(); however it is optimized and should be 767 * used whenever the sequential interface is not necessary. 768 * 769 * The routine uhash_init() initializes the uhash_ctx data structure and 770 * must be called once, before any other UHASH routine. 771 */ 772 773 /* ---------------------------------------------------------------------- */ 774 /* ----- Constants and uhash_ctx ---------------------------------------- */ 775 /* ---------------------------------------------------------------------- */ 776 777 /* ---------------------------------------------------------------------- */ 778 /* ----- Poly hash and Inner-Product hash Constants --------------------- */ 779 /* ---------------------------------------------------------------------- */ 780 781 /* Primes and masks */ 782 #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */ 783 #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */ 784 #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */ 785 786 787 /* ---------------------------------------------------------------------- */ 788 789 typedef struct uhash_ctx { 790 nh_ctx hash; /* Hash context for L1 NH hash */ 791 UINT64 poly_key_8[STREAMS]; /* p64 poly keys */ 792 UINT64 poly_accum[STREAMS]; /* poly hash result */ 793 UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */ 794 UINT32 ip_trans[STREAMS]; /* Inner-product translation */ 795 UINT32 msg_len; /* Total length of data passed */ 796 /* to uhash */ 797 } uhash_ctx; 798 typedef struct uhash_ctx *uhash_ctx_t; 799 800 /* ---------------------------------------------------------------------- */ 801 802 803 /* The polynomial hashes use Horner's rule to evaluate a polynomial one 804 * word at a time. As described in the specification, poly32 and poly64 805 * require keys from special domains. The following implementations exploit 806 * the special domains to avoid overflow. The results are not guaranteed to 807 * be within Z_p32 and Z_p64, but the Inner-Product hash implementation 808 * patches any errant values. 809 */ 810 811 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data) 812 { 813 UINT32 key_hi = (UINT32)(key >> 32), 814 key_lo = (UINT32)key, 815 cur_hi = (UINT32)(cur >> 32), 816 cur_lo = (UINT32)cur, 817 x_lo, 818 x_hi; 819 UINT64 X,T,res; 820 821 X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo); 822 x_lo = (UINT32)X; 823 x_hi = (UINT32)(X >> 32); 824 825 res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo); 826 827 T = ((UINT64)x_lo << 32); 828 res += T; 829 if (res < T) 830 res += 59; 831 832 res += data; 833 if (res < data) 834 res += 59; 835 836 return res; 837 } 838 839 840 /* Although UMAC is specified to use a ramped polynomial hash scheme, this 841 * implementation does not handle all ramp levels. Because we don't handle 842 * the ramp up to p128 modulus in this implementation, we are limited to 843 * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24 844 * bytes input to UMAC per tag, ie. 16MB). 845 */ 846 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[]) 847 { 848 int i; 849 UINT64 *data=(UINT64*)data_in; 850 851 for (i = 0; i < STREAMS; i++) { 852 if ((UINT32)(data[i] >> 32) == 0xfffffffful) { 853 hc->poly_accum[i] = poly64(hc->poly_accum[i], 854 hc->poly_key_8[i], p64 - 1); 855 hc->poly_accum[i] = poly64(hc->poly_accum[i], 856 hc->poly_key_8[i], (data[i] - 59)); 857 } else { 858 hc->poly_accum[i] = poly64(hc->poly_accum[i], 859 hc->poly_key_8[i], data[i]); 860 } 861 } 862 } 863 864 865 /* ---------------------------------------------------------------------- */ 866 867 868 /* The final step in UHASH is an inner-product hash. The poly hash 869 * produces a result not neccesarily WORD_LEN bytes long. The inner- 870 * product hash breaks the polyhash output into 16-bit chunks and 871 * multiplies each with a 36 bit key. 872 */ 873 874 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data) 875 { 876 t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48); 877 t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32); 878 t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16); 879 t = t + ipkp[3] * (UINT64)(UINT16)(data); 880 881 return t; 882 } 883 884 static UINT32 ip_reduce_p36(UINT64 t) 885 { 886 /* Divisionless modular reduction */ 887 UINT64 ret; 888 889 ret = (t & m36) + 5 * (t >> 36); 890 if (ret >= p36) 891 ret -= p36; 892 893 /* return least significant 32 bits */ 894 return (UINT32)(ret); 895 } 896 897 898 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then 899 * the polyhash stage is skipped and ip_short is applied directly to the 900 * NH output. 901 */ 902 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res) 903 { 904 UINT64 t; 905 UINT64 *nhp = (UINT64 *)nh_res; 906 907 t = ip_aux(0,ahc->ip_keys, nhp[0]); 908 STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]); 909 #if (UMAC_OUTPUT_LEN >= 8) 910 t = ip_aux(0,ahc->ip_keys+4, nhp[1]); 911 STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]); 912 #endif 913 #if (UMAC_OUTPUT_LEN >= 12) 914 t = ip_aux(0,ahc->ip_keys+8, nhp[2]); 915 STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]); 916 #endif 917 #if (UMAC_OUTPUT_LEN == 16) 918 t = ip_aux(0,ahc->ip_keys+12, nhp[3]); 919 STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]); 920 #endif 921 } 922 923 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then 924 * the polyhash stage is not skipped and ip_long is applied to the 925 * polyhash output. 926 */ 927 static void ip_long(uhash_ctx_t ahc, u_char *res) 928 { 929 int i; 930 UINT64 t; 931 932 for (i = 0; i < STREAMS; i++) { 933 /* fix polyhash output not in Z_p64 */ 934 if (ahc->poly_accum[i] >= p64) 935 ahc->poly_accum[i] -= p64; 936 t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]); 937 STORE_UINT32_BIG((UINT32 *)res+i, 938 ip_reduce_p36(t) ^ ahc->ip_trans[i]); 939 } 940 } 941 942 943 /* ---------------------------------------------------------------------- */ 944 945 /* ---------------------------------------------------------------------- */ 946 947 /* Reset uhash context for next hash session */ 948 static int uhash_reset(uhash_ctx_t pc) 949 { 950 nh_reset(&pc->hash); 951 pc->msg_len = 0; 952 pc->poly_accum[0] = 1; 953 #if (UMAC_OUTPUT_LEN >= 8) 954 pc->poly_accum[1] = 1; 955 #endif 956 #if (UMAC_OUTPUT_LEN >= 12) 957 pc->poly_accum[2] = 1; 958 #endif 959 #if (UMAC_OUTPUT_LEN == 16) 960 pc->poly_accum[3] = 1; 961 #endif 962 return 1; 963 } 964 965 /* ---------------------------------------------------------------------- */ 966 967 /* Given a pointer to the internal key needed by kdf() and a uhash context, 968 * initialize the NH context and generate keys needed for poly and inner- 969 * product hashing. All keys are endian adjusted in memory so that native 970 * loads cause correct keys to be in registers during calculation. 971 */ 972 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key) 973 { 974 int i; 975 UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)]; 976 977 /* Zero the entire uhash context */ 978 memset(ahc, 0, sizeof(uhash_ctx)); 979 980 /* Initialize the L1 hash */ 981 nh_init(&ahc->hash, prf_key); 982 983 /* Setup L2 hash variables */ 984 kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */ 985 for (i = 0; i < STREAMS; i++) { 986 /* Fill keys from the buffer, skipping bytes in the buffer not 987 * used by this implementation. Endian reverse the keys if on a 988 * little-endian computer. 989 */ 990 memcpy(ahc->poly_key_8+i, buf+24*i, 8); 991 endian_convert_if_le(ahc->poly_key_8+i, 8, 8); 992 /* Mask the 64-bit keys to their special domain */ 993 ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu; 994 ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */ 995 } 996 997 /* Setup L3-1 hash variables */ 998 kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */ 999 for (i = 0; i < STREAMS; i++) 1000 memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64), 1001 4*sizeof(UINT64)); 1002 endian_convert_if_le(ahc->ip_keys, sizeof(UINT64), 1003 sizeof(ahc->ip_keys)); 1004 for (i = 0; i < STREAMS*4; i++) 1005 ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */ 1006 1007 /* Setup L3-2 hash variables */ 1008 /* Fill buffer with index 4 key */ 1009 kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32)); 1010 endian_convert_if_le(ahc->ip_trans, sizeof(UINT32), 1011 STREAMS * sizeof(UINT32)); 1012 } 1013 1014 /* ---------------------------------------------------------------------- */ 1015 1016 #if 0 1017 static uhash_ctx_t uhash_alloc(u_char key[]) 1018 { 1019 /* Allocate memory and force to a 16-byte boundary. */ 1020 uhash_ctx_t ctx; 1021 u_char bytes_to_add; 1022 aes_int_key prf_key; 1023 1024 ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY); 1025 if (ctx) { 1026 if (ALLOC_BOUNDARY) { 1027 bytes_to_add = ALLOC_BOUNDARY - 1028 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1)); 1029 ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add); 1030 *((u_char *)ctx - 1) = bytes_to_add; 1031 } 1032 aes_key_setup(key,prf_key); 1033 uhash_init(ctx, prf_key); 1034 } 1035 return (ctx); 1036 } 1037 #endif 1038 1039 /* ---------------------------------------------------------------------- */ 1040 1041 #if 0 1042 static int uhash_free(uhash_ctx_t ctx) 1043 { 1044 /* Free memory allocated by uhash_alloc */ 1045 u_char bytes_to_sub; 1046 1047 if (ctx) { 1048 if (ALLOC_BOUNDARY) { 1049 bytes_to_sub = *((u_char *)ctx - 1); 1050 ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub); 1051 } 1052 free(ctx); 1053 } 1054 return (1); 1055 } 1056 #endif 1057 /* ---------------------------------------------------------------------- */ 1058 1059 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len) 1060 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and 1061 * hash each one with NH, calling the polyhash on each NH output. 1062 */ 1063 { 1064 UWORD bytes_hashed, bytes_remaining; 1065 UINT64 result_buf[STREAMS]; 1066 UINT8 *nh_result = (UINT8 *)&result_buf; 1067 1068 if (ctx->msg_len + len <= L1_KEY_LEN) { 1069 nh_update(&ctx->hash, (const UINT8 *)input, len); 1070 ctx->msg_len += len; 1071 } else { 1072 1073 bytes_hashed = ctx->msg_len % L1_KEY_LEN; 1074 if (ctx->msg_len == L1_KEY_LEN) 1075 bytes_hashed = L1_KEY_LEN; 1076 1077 if (bytes_hashed + len >= L1_KEY_LEN) { 1078 1079 /* If some bytes have been passed to the hash function */ 1080 /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */ 1081 /* bytes to complete the current nh_block. */ 1082 if (bytes_hashed) { 1083 bytes_remaining = (L1_KEY_LEN - bytes_hashed); 1084 nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining); 1085 nh_final(&ctx->hash, nh_result); 1086 ctx->msg_len += bytes_remaining; 1087 poly_hash(ctx,(UINT32 *)nh_result); 1088 len -= bytes_remaining; 1089 input += bytes_remaining; 1090 } 1091 1092 /* Hash directly from input stream if enough bytes */ 1093 while (len >= L1_KEY_LEN) { 1094 nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN, 1095 L1_KEY_LEN, nh_result); 1096 ctx->msg_len += L1_KEY_LEN; 1097 len -= L1_KEY_LEN; 1098 input += L1_KEY_LEN; 1099 poly_hash(ctx,(UINT32 *)nh_result); 1100 } 1101 } 1102 1103 /* pass remaining < L1_KEY_LEN bytes of input data to NH */ 1104 if (len) { 1105 nh_update(&ctx->hash, (const UINT8 *)input, len); 1106 ctx->msg_len += len; 1107 } 1108 } 1109 1110 return (1); 1111 } 1112 1113 /* ---------------------------------------------------------------------- */ 1114 1115 static int uhash_final(uhash_ctx_t ctx, u_char *res) 1116 /* Incorporate any pending data, pad, and generate tag */ 1117 { 1118 UINT64 result_buf[STREAMS]; 1119 UINT8 *nh_result = (UINT8 *)&result_buf; 1120 1121 if (ctx->msg_len > L1_KEY_LEN) { 1122 if (ctx->msg_len % L1_KEY_LEN) { 1123 nh_final(&ctx->hash, nh_result); 1124 poly_hash(ctx,(UINT32 *)nh_result); 1125 } 1126 ip_long(ctx, res); 1127 } else { 1128 nh_final(&ctx->hash, nh_result); 1129 ip_short(ctx,nh_result, res); 1130 } 1131 uhash_reset(ctx); 1132 return (1); 1133 } 1134 1135 /* ---------------------------------------------------------------------- */ 1136 1137 #if 0 1138 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res) 1139 /* assumes that msg is in a writable buffer of length divisible by */ 1140 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */ 1141 { 1142 UINT8 nh_result[STREAMS*sizeof(UINT64)]; 1143 UINT32 nh_len; 1144 int extra_zeroes_needed; 1145 1146 /* If the message to be hashed is no longer than L1_HASH_LEN, we skip 1147 * the polyhash. 1148 */ 1149 if (len <= L1_KEY_LEN) { 1150 if (len == 0) /* If zero length messages will not */ 1151 nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */ 1152 else 1153 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1154 extra_zeroes_needed = nh_len - len; 1155 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1156 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1157 ip_short(ahc,nh_result, res); 1158 } else { 1159 /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH 1160 * output to poly_hash(). 1161 */ 1162 do { 1163 nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result); 1164 poly_hash(ahc,(UINT32 *)nh_result); 1165 len -= L1_KEY_LEN; 1166 msg += L1_KEY_LEN; 1167 } while (len >= L1_KEY_LEN); 1168 if (len) { 1169 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1170 extra_zeroes_needed = nh_len - len; 1171 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1172 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1173 poly_hash(ahc,(UINT32 *)nh_result); 1174 } 1175 1176 ip_long(ahc, res); 1177 } 1178 1179 uhash_reset(ahc); 1180 return 1; 1181 } 1182 #endif 1183 1184 /* ---------------------------------------------------------------------- */ 1185 /* ---------------------------------------------------------------------- */ 1186 /* ----- Begin UMAC Section --------------------------------------------- */ 1187 /* ---------------------------------------------------------------------- */ 1188 /* ---------------------------------------------------------------------- */ 1189 1190 /* The UMAC interface has two interfaces, an all-at-once interface where 1191 * the entire message to be authenticated is passed to UMAC in one buffer, 1192 * and a sequential interface where the message is presented a little at a 1193 * time. The all-at-once is more optimaized than the sequential version and 1194 * should be preferred when the sequential interface is not required. 1195 */ 1196 struct umac_ctx { 1197 uhash_ctx hash; /* Hash function for message compression */ 1198 pdf_ctx pdf; /* PDF for hashed output */ 1199 void *free_ptr; /* Address to free this struct via */ 1200 } umac_ctx; 1201 1202 /* ---------------------------------------------------------------------- */ 1203 1204 #if 0 1205 int umac_reset(struct umac_ctx *ctx) 1206 /* Reset the hash function to begin a new authentication. */ 1207 { 1208 uhash_reset(&ctx->hash); 1209 return (1); 1210 } 1211 #endif 1212 1213 /* ---------------------------------------------------------------------- */ 1214 1215 int umac_delete(struct umac_ctx *ctx) 1216 /* Deallocate the ctx structure */ 1217 { 1218 if (ctx) { 1219 if (ALLOC_BOUNDARY) 1220 ctx = (struct umac_ctx *)ctx->free_ptr; 1221 free(ctx); 1222 } 1223 return (1); 1224 } 1225 1226 /* ---------------------------------------------------------------------- */ 1227 1228 struct umac_ctx *umac_new(const u_char key[]) 1229 /* Dynamically allocate a umac_ctx struct, initialize variables, 1230 * generate subkeys from key. Align to 16-byte boundary. 1231 */ 1232 { 1233 struct umac_ctx *ctx, *octx; 1234 size_t bytes_to_add; 1235 aes_int_key prf_key; 1236 1237 octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY); 1238 if (ctx) { 1239 if (ALLOC_BOUNDARY) { 1240 bytes_to_add = ALLOC_BOUNDARY - 1241 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1)); 1242 ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add); 1243 } 1244 ctx->free_ptr = octx; 1245 aes_key_setup(key, prf_key); 1246 pdf_init(&ctx->pdf, prf_key); 1247 uhash_init(&ctx->hash, prf_key); 1248 } 1249 1250 return (ctx); 1251 } 1252 1253 /* ---------------------------------------------------------------------- */ 1254 1255 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8]) 1256 /* Incorporate any pending data, pad, and generate tag */ 1257 { 1258 uhash_final(&ctx->hash, (u_char *)tag); 1259 pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag); 1260 1261 return (1); 1262 } 1263 1264 /* ---------------------------------------------------------------------- */ 1265 1266 int umac_update(struct umac_ctx *ctx, const u_char *input, long len) 1267 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */ 1268 /* hash each one, calling the PDF on the hashed output whenever the hash- */ 1269 /* output buffer is full. */ 1270 { 1271 uhash_update(&ctx->hash, input, len); 1272 return (1); 1273 } 1274 1275 /* ---------------------------------------------------------------------- */ 1276 1277 #if 0 1278 int umac(struct umac_ctx *ctx, u_char *input, 1279 long len, u_char tag[], 1280 u_char nonce[8]) 1281 /* All-in-one version simply calls umac_update() and umac_final(). */ 1282 { 1283 uhash(&ctx->hash, input, len, (u_char *)tag); 1284 pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); 1285 1286 return (1); 1287 } 1288 #endif 1289 1290 /* ---------------------------------------------------------------------- */ 1291 /* ---------------------------------------------------------------------- */ 1292 /* ----- End UMAC Section ----------------------------------------------- */ 1293 /* ---------------------------------------------------------------------- */ 1294 /* ---------------------------------------------------------------------- */ 1295