1 /* $OpenBSD: umac.c,v 1.11 2014/07/22 07:13:42 guenther 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 (at) 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 /* In OpenSSH, this file is compiled twice, with different #defines set on the 52 * command line. Since we don't want to stretch the Android build system, in 53 * Android this file is duplicated as umac.c and umac128.c. The latter contains 54 * the #defines (that were set in OpenSSH's Makefile) at the top of the 55 * file. */ 56 57 /* ---------------------------------------------------------------------- */ 58 /* --- User Switches ---------------------------------------------------- */ 59 /* ---------------------------------------------------------------------- */ 60 61 #ifndef UMAC_OUTPUT_LEN 62 #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */ 63 #endif 64 65 #if UMAC_OUTPUT_LEN != 4 && UMAC_OUTPUT_LEN != 8 && \ 66 UMAC_OUTPUT_LEN != 12 && UMAC_OUTPUT_LEN != 16 67 # error UMAC_OUTPUT_LEN must be defined to 4, 8, 12 or 16 68 #endif 69 70 /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */ 71 /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */ 72 /* #define SSE2 0 Is SSE2 is available? */ 73 /* #define RUN_TESTS 0 Run basic correctness/speed tests */ 74 /* #define UMAC_AE_SUPPORT 0 Enable auhthenticated encrytion */ 75 76 /* ---------------------------------------------------------------------- */ 77 /* -- Global Includes --------------------------------------------------- */ 78 /* ---------------------------------------------------------------------- */ 79 80 #include "includes.h" 81 #include <sys/types.h> 82 #include <string.h> 83 #include <stdio.h> 84 #include <stdlib.h> 85 #include <stddef.h> 86 87 #include "xmalloc.h" 88 #include "umac.h" 89 #include "misc.h" 90 91 /* ---------------------------------------------------------------------- */ 92 /* --- Primitive Data Types --- */ 93 /* ---------------------------------------------------------------------- */ 94 95 /* The following assumptions may need change on your system */ 96 typedef u_int8_t UINT8; /* 1 byte */ 97 typedef u_int16_t UINT16; /* 2 byte */ 98 typedef u_int32_t UINT32; /* 4 byte */ 99 typedef u_int64_t UINT64; /* 8 bytes */ 100 typedef unsigned int UWORD; /* Register */ 101 102 /* ---------------------------------------------------------------------- */ 103 /* --- Constants -------------------------------------------------------- */ 104 /* ---------------------------------------------------------------------- */ 105 106 #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */ 107 108 /* Message "words" are read from memory in an endian-specific manner. */ 109 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */ 110 /* be set true if the host computer is little-endian. */ 111 112 #if BYTE_ORDER == LITTLE_ENDIAN 113 #define __LITTLE_ENDIAN__ 1 114 #else 115 #define __LITTLE_ENDIAN__ 0 116 #endif 117 118 /* ---------------------------------------------------------------------- */ 119 /* ---------------------------------------------------------------------- */ 120 /* ----- Architecture Specific ------------------------------------------ */ 121 /* ---------------------------------------------------------------------- */ 122 /* ---------------------------------------------------------------------- */ 123 124 125 /* ---------------------------------------------------------------------- */ 126 /* ---------------------------------------------------------------------- */ 127 /* ----- Primitive Routines --------------------------------------------- */ 128 /* ---------------------------------------------------------------------- */ 129 /* ---------------------------------------------------------------------- */ 130 131 132 /* ---------------------------------------------------------------------- */ 133 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */ 134 /* ---------------------------------------------------------------------- */ 135 136 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b))) 137 138 /* ---------------------------------------------------------------------- */ 139 /* --- Endian Conversion --- Forcing assembly on some platforms */ 140 /* ---------------------------------------------------------------------- */ 141 142 #if (__LITTLE_ENDIAN__) 143 #define LOAD_UINT32_REVERSED(p) get_u32(p) 144 #define STORE_UINT32_REVERSED(p,v) put_u32(p,v) 145 #else 146 #define LOAD_UINT32_REVERSED(p) get_u32_le(p) 147 #define STORE_UINT32_REVERSED(p,v) put_u32_le(p,v) 148 #endif 149 150 #define LOAD_UINT32_LITTLE(p) (get_u32_le(p)) 151 #define STORE_UINT32_BIG(p,v) put_u32(p, v) 152 153 /* ---------------------------------------------------------------------- */ 154 /* ---------------------------------------------------------------------- */ 155 /* ----- Begin KDF & PDF Section ---------------------------------------- */ 156 /* ---------------------------------------------------------------------- */ 157 /* ---------------------------------------------------------------------- */ 158 159 /* UMAC uses AES with 16 byte block and key lengths */ 160 #define AES_BLOCK_LEN 16 161 162 /* OpenSSL's AES */ 163 #ifdef WITH_OPENSSL 164 #include "openbsd-compat/openssl-compat.h" 165 #ifndef USE_BUILTIN_RIJNDAEL 166 # include <openssl/aes.h> 167 #endif 168 typedef AES_KEY aes_int_key[1]; 169 #define aes_encryption(in,out,int_key) \ 170 AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key) 171 #define aes_key_setup(key,int_key) \ 172 AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key) 173 #else 174 #include "rijndael.h" 175 #define AES_ROUNDS ((UMAC_KEY_LEN / 4) + 6) 176 typedef UINT8 aes_int_key[AES_ROUNDS+1][4][4]; /* AES internal */ 177 #define aes_encryption(in,out,int_key) \ 178 rijndaelEncrypt((u32 *)(int_key), AES_ROUNDS, (u8 *)(in), (u8 *)(out)) 179 #define aes_key_setup(key,int_key) \ 180 rijndaelKeySetupEnc((u32 *)(int_key), (const unsigned char *)(key), \ 181 UMAC_KEY_LEN*8) 182 #endif 183 184 /* The user-supplied UMAC key is stretched using AES in a counter 185 * mode to supply all random bits needed by UMAC. The kdf function takes 186 * an AES internal key representation 'key' and writes a stream of 187 * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct 188 * 'ndx' causes a distinct byte stream. 189 */ 190 static void kdf(void *bufp, aes_int_key key, UINT8 ndx, int nbytes) 191 { 192 UINT8 in_buf[AES_BLOCK_LEN] = {0}; 193 UINT8 out_buf[AES_BLOCK_LEN]; 194 UINT8 *dst_buf = (UINT8 *)bufp; 195 int i; 196 197 /* Setup the initial value */ 198 in_buf[AES_BLOCK_LEN-9] = ndx; 199 in_buf[AES_BLOCK_LEN-1] = i = 1; 200 201 while (nbytes >= AES_BLOCK_LEN) { 202 aes_encryption(in_buf, out_buf, key); 203 memcpy(dst_buf,out_buf,AES_BLOCK_LEN); 204 in_buf[AES_BLOCK_LEN-1] = ++i; 205 nbytes -= AES_BLOCK_LEN; 206 dst_buf += AES_BLOCK_LEN; 207 } 208 if (nbytes) { 209 aes_encryption(in_buf, out_buf, key); 210 memcpy(dst_buf,out_buf,nbytes); 211 } 212 } 213 214 /* The final UHASH result is XOR'd with the output of a pseudorandom 215 * function. Here, we use AES to generate random output and 216 * xor the appropriate bytes depending on the last bits of nonce. 217 * This scheme is optimized for sequential, increasing big-endian nonces. 218 */ 219 220 typedef struct { 221 UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */ 222 UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */ 223 aes_int_key prf_key; /* Expanded AES key for PDF */ 224 } pdf_ctx; 225 226 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key) 227 { 228 UINT8 buf[UMAC_KEY_LEN]; 229 230 kdf(buf, prf_key, 0, UMAC_KEY_LEN); 231 aes_key_setup(buf, pc->prf_key); 232 233 /* Initialize pdf and cache */ 234 memset(pc->nonce, 0, sizeof(pc->nonce)); 235 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 236 } 237 238 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8]) 239 { 240 /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes 241 * of the AES output. If last time around we returned the ndx-1st 242 * element, then we may have the result in the cache already. 243 */ 244 245 #if (UMAC_OUTPUT_LEN == 4) 246 #define LOW_BIT_MASK 3 247 #elif (UMAC_OUTPUT_LEN == 8) 248 #define LOW_BIT_MASK 1 249 #elif (UMAC_OUTPUT_LEN > 8) 250 #define LOW_BIT_MASK 0 251 #endif 252 union { 253 UINT8 tmp_nonce_lo[4]; 254 UINT32 align; 255 } t; 256 #if LOW_BIT_MASK != 0 257 int ndx = nonce[7] & LOW_BIT_MASK; 258 #endif 259 *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1]; 260 t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */ 261 262 if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) || 263 (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) ) 264 { 265 ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0]; 266 ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0]; 267 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 268 } 269 270 #if (UMAC_OUTPUT_LEN == 4) 271 *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx]; 272 #elif (UMAC_OUTPUT_LEN == 8) 273 *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx]; 274 #elif (UMAC_OUTPUT_LEN == 12) 275 ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; 276 ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2]; 277 #elif (UMAC_OUTPUT_LEN == 16) 278 ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; 279 ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1]; 280 #endif 281 } 282 283 /* ---------------------------------------------------------------------- */ 284 /* ---------------------------------------------------------------------- */ 285 /* ----- Begin NH Hash Section ------------------------------------------ */ 286 /* ---------------------------------------------------------------------- */ 287 /* ---------------------------------------------------------------------- */ 288 289 /* The NH-based hash functions used in UMAC are described in the UMAC paper 290 * and specification, both of which can be found at the UMAC website. 291 * The interface to this implementation has two 292 * versions, one expects the entire message being hashed to be passed 293 * in a single buffer and returns the hash result immediately. The second 294 * allows the message to be passed in a sequence of buffers. In the 295 * muliple-buffer interface, the client calls the routine nh_update() as 296 * many times as necessary. When there is no more data to be fed to the 297 * hash, the client calls nh_final() which calculates the hash output. 298 * Before beginning another hash calculation the nh_reset() routine 299 * must be called. The single-buffer routine, nh(), is equivalent to 300 * the sequence of calls nh_update() and nh_final(); however it is 301 * optimized and should be prefered whenever the multiple-buffer interface 302 * is not necessary. When using either interface, it is the client's 303 * responsability to pass no more than L1_KEY_LEN bytes per hash result. 304 * 305 * The routine nh_init() initializes the nh_ctx data structure and 306 * must be called once, before any other PDF routine. 307 */ 308 309 /* The "nh_aux" routines do the actual NH hashing work. They 310 * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines 311 * produce output for all STREAMS NH iterations in one call, 312 * allowing the parallel implementation of the streams. 313 */ 314 315 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */ 316 #define L1_KEY_LEN 1024 /* Internal key bytes */ 317 #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */ 318 #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */ 319 #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */ 320 #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */ 321 322 typedef struct { 323 UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */ 324 UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */ 325 int next_data_empty; /* Bookeeping variable for data buffer. */ 326 int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */ 327 UINT64 state[STREAMS]; /* on-line state */ 328 } nh_ctx; 329 330 331 #if (UMAC_OUTPUT_LEN == 4) 332 333 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 334 /* NH hashing primitive. Previous (partial) hash result is loaded and 335 * then stored via hp pointer. The length of the data pointed at by "dp", 336 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key 337 * is expected to be endian compensated in memory at key setup. 338 */ 339 { 340 UINT64 h; 341 UWORD c = dlen / 32; 342 UINT32 *k = (UINT32 *)kp; 343 const UINT32 *d = (const UINT32 *)dp; 344 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 345 UINT32 k0,k1,k2,k3,k4,k5,k6,k7; 346 347 h = *((UINT64 *)hp); 348 do { 349 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 350 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 351 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 352 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 353 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 354 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 355 h += MUL64((k0 + d0), (k4 + d4)); 356 h += MUL64((k1 + d1), (k5 + d5)); 357 h += MUL64((k2 + d2), (k6 + d6)); 358 h += MUL64((k3 + d3), (k7 + d7)); 359 360 d += 8; 361 k += 8; 362 } while (--c); 363 *((UINT64 *)hp) = h; 364 } 365 366 #elif (UMAC_OUTPUT_LEN == 8) 367 368 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 369 /* Same as previous nh_aux, but two streams are handled in one pass, 370 * reading and writing 16 bytes of hash-state per call. 371 */ 372 { 373 UINT64 h1,h2; 374 UWORD c = dlen / 32; 375 UINT32 *k = (UINT32 *)kp; 376 const UINT32 *d = (const UINT32 *)dp; 377 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 378 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 379 k8,k9,k10,k11; 380 381 h1 = *((UINT64 *)hp); 382 h2 = *((UINT64 *)hp + 1); 383 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 384 do { 385 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 386 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 387 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 388 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 389 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 390 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 391 392 h1 += MUL64((k0 + d0), (k4 + d4)); 393 h2 += MUL64((k4 + d0), (k8 + d4)); 394 395 h1 += MUL64((k1 + d1), (k5 + d5)); 396 h2 += MUL64((k5 + d1), (k9 + d5)); 397 398 h1 += MUL64((k2 + d2), (k6 + d6)); 399 h2 += MUL64((k6 + d2), (k10 + d6)); 400 401 h1 += MUL64((k3 + d3), (k7 + d7)); 402 h2 += MUL64((k7 + d3), (k11 + d7)); 403 404 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 405 406 d += 8; 407 k += 8; 408 } while (--c); 409 ((UINT64 *)hp)[0] = h1; 410 ((UINT64 *)hp)[1] = h2; 411 } 412 413 #elif (UMAC_OUTPUT_LEN == 12) 414 415 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 416 /* Same as previous nh_aux, but two streams are handled in one pass, 417 * reading and writing 24 bytes of hash-state per call. 418 */ 419 { 420 UINT64 h1,h2,h3; 421 UWORD c = dlen / 32; 422 UINT32 *k = (UINT32 *)kp; 423 const UINT32 *d = (const UINT32 *)dp; 424 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 425 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 426 k8,k9,k10,k11,k12,k13,k14,k15; 427 428 h1 = *((UINT64 *)hp); 429 h2 = *((UINT64 *)hp + 1); 430 h3 = *((UINT64 *)hp + 2); 431 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 432 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 433 do { 434 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 435 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 436 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 437 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 438 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 439 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 440 441 h1 += MUL64((k0 + d0), (k4 + d4)); 442 h2 += MUL64((k4 + d0), (k8 + d4)); 443 h3 += MUL64((k8 + d0), (k12 + d4)); 444 445 h1 += MUL64((k1 + d1), (k5 + d5)); 446 h2 += MUL64((k5 + d1), (k9 + d5)); 447 h3 += MUL64((k9 + d1), (k13 + d5)); 448 449 h1 += MUL64((k2 + d2), (k6 + d6)); 450 h2 += MUL64((k6 + d2), (k10 + d6)); 451 h3 += MUL64((k10 + d2), (k14 + d6)); 452 453 h1 += MUL64((k3 + d3), (k7 + d7)); 454 h2 += MUL64((k7 + d3), (k11 + d7)); 455 h3 += MUL64((k11 + d3), (k15 + d7)); 456 457 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 458 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 459 460 d += 8; 461 k += 8; 462 } while (--c); 463 ((UINT64 *)hp)[0] = h1; 464 ((UINT64 *)hp)[1] = h2; 465 ((UINT64 *)hp)[2] = h3; 466 } 467 468 #elif (UMAC_OUTPUT_LEN == 16) 469 470 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 471 /* Same as previous nh_aux, but two streams are handled in one pass, 472 * reading and writing 24 bytes of hash-state per call. 473 */ 474 { 475 UINT64 h1,h2,h3,h4; 476 UWORD c = dlen / 32; 477 UINT32 *k = (UINT32 *)kp; 478 const UINT32 *d = (const UINT32 *)dp; 479 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 480 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 481 k8,k9,k10,k11,k12,k13,k14,k15, 482 k16,k17,k18,k19; 483 484 h1 = *((UINT64 *)hp); 485 h2 = *((UINT64 *)hp + 1); 486 h3 = *((UINT64 *)hp + 2); 487 h4 = *((UINT64 *)hp + 3); 488 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 489 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 490 do { 491 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 492 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 493 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 494 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 495 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 496 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 497 k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19); 498 499 h1 += MUL64((k0 + d0), (k4 + d4)); 500 h2 += MUL64((k4 + d0), (k8 + d4)); 501 h3 += MUL64((k8 + d0), (k12 + d4)); 502 h4 += MUL64((k12 + d0), (k16 + d4)); 503 504 h1 += MUL64((k1 + d1), (k5 + d5)); 505 h2 += MUL64((k5 + d1), (k9 + d5)); 506 h3 += MUL64((k9 + d1), (k13 + d5)); 507 h4 += MUL64((k13 + d1), (k17 + d5)); 508 509 h1 += MUL64((k2 + d2), (k6 + d6)); 510 h2 += MUL64((k6 + d2), (k10 + d6)); 511 h3 += MUL64((k10 + d2), (k14 + d6)); 512 h4 += MUL64((k14 + d2), (k18 + d6)); 513 514 h1 += MUL64((k3 + d3), (k7 + d7)); 515 h2 += MUL64((k7 + d3), (k11 + d7)); 516 h3 += MUL64((k11 + d3), (k15 + d7)); 517 h4 += MUL64((k15 + d3), (k19 + d7)); 518 519 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 520 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 521 k8 = k16; k9 = k17; k10 = k18; k11 = k19; 522 523 d += 8; 524 k += 8; 525 } while (--c); 526 ((UINT64 *)hp)[0] = h1; 527 ((UINT64 *)hp)[1] = h2; 528 ((UINT64 *)hp)[2] = h3; 529 ((UINT64 *)hp)[3] = h4; 530 } 531 532 /* ---------------------------------------------------------------------- */ 533 #endif /* UMAC_OUTPUT_LENGTH */ 534 /* ---------------------------------------------------------------------- */ 535 536 537 /* ---------------------------------------------------------------------- */ 538 539 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes) 540 /* This function is a wrapper for the primitive NH hash functions. It takes 541 * as argument "hc" the current hash context and a buffer which must be a 542 * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset 543 * appropriately according to how much message has been hashed already. 544 */ 545 { 546 UINT8 *key; 547 548 key = hc->nh_key + hc->bytes_hashed; 549 nh_aux(key, buf, hc->state, nbytes); 550 } 551 552 /* ---------------------------------------------------------------------- */ 553 554 #if (__LITTLE_ENDIAN__) 555 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes) 556 /* We endian convert the keys on little-endian computers to */ 557 /* compensate for the lack of big-endian memory reads during hashing. */ 558 { 559 UWORD iters = num_bytes / bpw; 560 if (bpw == 4) { 561 UINT32 *p = (UINT32 *)buf; 562 do { 563 *p = LOAD_UINT32_REVERSED(p); 564 p++; 565 } while (--iters); 566 } else if (bpw == 8) { 567 UINT32 *p = (UINT32 *)buf; 568 UINT32 t; 569 do { 570 t = LOAD_UINT32_REVERSED(p+1); 571 p[1] = LOAD_UINT32_REVERSED(p); 572 p[0] = t; 573 p += 2; 574 } while (--iters); 575 } 576 } 577 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) 578 #else 579 #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */ 580 #endif 581 582 /* ---------------------------------------------------------------------- */ 583 584 static void nh_reset(nh_ctx *hc) 585 /* Reset nh_ctx to ready for hashing of new data */ 586 { 587 hc->bytes_hashed = 0; 588 hc->next_data_empty = 0; 589 hc->state[0] = 0; 590 #if (UMAC_OUTPUT_LEN >= 8) 591 hc->state[1] = 0; 592 #endif 593 #if (UMAC_OUTPUT_LEN >= 12) 594 hc->state[2] = 0; 595 #endif 596 #if (UMAC_OUTPUT_LEN == 16) 597 hc->state[3] = 0; 598 #endif 599 600 } 601 602 /* ---------------------------------------------------------------------- */ 603 604 static void nh_init(nh_ctx *hc, aes_int_key prf_key) 605 /* Generate nh_key, endian convert and reset to be ready for hashing. */ 606 { 607 kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key)); 608 endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key)); 609 nh_reset(hc); 610 } 611 612 /* ---------------------------------------------------------------------- */ 613 614 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes) 615 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */ 616 /* even multiple of HASH_BUF_BYTES. */ 617 { 618 UINT32 i,j; 619 620 j = hc->next_data_empty; 621 if ((j + nbytes) >= HASH_BUF_BYTES) { 622 if (j) { 623 i = HASH_BUF_BYTES - j; 624 memcpy(hc->data+j, buf, i); 625 nh_transform(hc,hc->data,HASH_BUF_BYTES); 626 nbytes -= i; 627 buf += i; 628 hc->bytes_hashed += HASH_BUF_BYTES; 629 } 630 if (nbytes >= HASH_BUF_BYTES) { 631 i = nbytes & ~(HASH_BUF_BYTES - 1); 632 nh_transform(hc, buf, i); 633 nbytes -= i; 634 buf += i; 635 hc->bytes_hashed += i; 636 } 637 j = 0; 638 } 639 memcpy(hc->data + j, buf, nbytes); 640 hc->next_data_empty = j + nbytes; 641 } 642 643 /* ---------------------------------------------------------------------- */ 644 645 static void zero_pad(UINT8 *p, int nbytes) 646 { 647 /* Write "nbytes" of zeroes, beginning at "p" */ 648 if (nbytes >= (int)sizeof(UWORD)) { 649 while ((ptrdiff_t)p % sizeof(UWORD)) { 650 *p = 0; 651 nbytes--; 652 p++; 653 } 654 while (nbytes >= (int)sizeof(UWORD)) { 655 *(UWORD *)p = 0; 656 nbytes -= sizeof(UWORD); 657 p += sizeof(UWORD); 658 } 659 } 660 while (nbytes) { 661 *p = 0; 662 nbytes--; 663 p++; 664 } 665 } 666 667 /* ---------------------------------------------------------------------- */ 668 669 static void nh_final(nh_ctx *hc, UINT8 *result) 670 /* After passing some number of data buffers to nh_update() for integration 671 * into an NH context, nh_final is called to produce a hash result. If any 672 * bytes are in the buffer hc->data, incorporate them into the 673 * NH context. Finally, add into the NH accumulation "state" the total number 674 * of bits hashed. The resulting numbers are written to the buffer "result". 675 * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated. 676 */ 677 { 678 int nh_len, nbits; 679 680 if (hc->next_data_empty != 0) { 681 nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) & 682 ~(L1_PAD_BOUNDARY - 1)); 683 zero_pad(hc->data + hc->next_data_empty, 684 nh_len - hc->next_data_empty); 685 nh_transform(hc, hc->data, nh_len); 686 hc->bytes_hashed += hc->next_data_empty; 687 } else if (hc->bytes_hashed == 0) { 688 nh_len = L1_PAD_BOUNDARY; 689 zero_pad(hc->data, L1_PAD_BOUNDARY); 690 nh_transform(hc, hc->data, nh_len); 691 } 692 693 nbits = (hc->bytes_hashed << 3); 694 ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits; 695 #if (UMAC_OUTPUT_LEN >= 8) 696 ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits; 697 #endif 698 #if (UMAC_OUTPUT_LEN >= 12) 699 ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits; 700 #endif 701 #if (UMAC_OUTPUT_LEN == 16) 702 ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits; 703 #endif 704 nh_reset(hc); 705 } 706 707 /* ---------------------------------------------------------------------- */ 708 709 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len, 710 UINT32 unpadded_len, UINT8 *result) 711 /* All-in-one nh_update() and nh_final() equivalent. 712 * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is 713 * well aligned 714 */ 715 { 716 UINT32 nbits; 717 718 /* Initialize the hash state */ 719 nbits = (unpadded_len << 3); 720 721 ((UINT64 *)result)[0] = nbits; 722 #if (UMAC_OUTPUT_LEN >= 8) 723 ((UINT64 *)result)[1] = nbits; 724 #endif 725 #if (UMAC_OUTPUT_LEN >= 12) 726 ((UINT64 *)result)[2] = nbits; 727 #endif 728 #if (UMAC_OUTPUT_LEN == 16) 729 ((UINT64 *)result)[3] = nbits; 730 #endif 731 732 nh_aux(hc->nh_key, buf, result, padded_len); 733 } 734 735 /* ---------------------------------------------------------------------- */ 736 /* ---------------------------------------------------------------------- */ 737 /* ----- Begin UHASH Section -------------------------------------------- */ 738 /* ---------------------------------------------------------------------- */ 739 /* ---------------------------------------------------------------------- */ 740 741 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first 742 * hashed by NH. The NH output is then hashed by a polynomial-hash layer 743 * unless the initial data to be hashed is short. After the polynomial- 744 * layer, an inner-product hash is used to produce the final UHASH output. 745 * 746 * UHASH provides two interfaces, one all-at-once and another where data 747 * buffers are presented sequentially. In the sequential interface, the 748 * UHASH client calls the routine uhash_update() as many times as necessary. 749 * When there is no more data to be fed to UHASH, the client calls 750 * uhash_final() which 751 * calculates the UHASH output. Before beginning another UHASH calculation 752 * the uhash_reset() routine must be called. The all-at-once UHASH routine, 753 * uhash(), is equivalent to the sequence of calls uhash_update() and 754 * uhash_final(); however it is optimized and should be 755 * used whenever the sequential interface is not necessary. 756 * 757 * The routine uhash_init() initializes the uhash_ctx data structure and 758 * must be called once, before any other UHASH routine. 759 */ 760 761 /* ---------------------------------------------------------------------- */ 762 /* ----- Constants and uhash_ctx ---------------------------------------- */ 763 /* ---------------------------------------------------------------------- */ 764 765 /* ---------------------------------------------------------------------- */ 766 /* ----- Poly hash and Inner-Product hash Constants --------------------- */ 767 /* ---------------------------------------------------------------------- */ 768 769 /* Primes and masks */ 770 #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */ 771 #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */ 772 #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */ 773 774 775 /* ---------------------------------------------------------------------- */ 776 777 typedef struct uhash_ctx { 778 nh_ctx hash; /* Hash context for L1 NH hash */ 779 UINT64 poly_key_8[STREAMS]; /* p64 poly keys */ 780 UINT64 poly_accum[STREAMS]; /* poly hash result */ 781 UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */ 782 UINT32 ip_trans[STREAMS]; /* Inner-product translation */ 783 UINT32 msg_len; /* Total length of data passed */ 784 /* to uhash */ 785 } uhash_ctx; 786 typedef struct uhash_ctx *uhash_ctx_t; 787 788 /* ---------------------------------------------------------------------- */ 789 790 791 /* The polynomial hashes use Horner's rule to evaluate a polynomial one 792 * word at a time. As described in the specification, poly32 and poly64 793 * require keys from special domains. The following implementations exploit 794 * the special domains to avoid overflow. The results are not guaranteed to 795 * be within Z_p32 and Z_p64, but the Inner-Product hash implementation 796 * patches any errant values. 797 */ 798 799 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data) 800 { 801 UINT32 key_hi = (UINT32)(key >> 32), 802 key_lo = (UINT32)key, 803 cur_hi = (UINT32)(cur >> 32), 804 cur_lo = (UINT32)cur, 805 x_lo, 806 x_hi; 807 UINT64 X,T,res; 808 809 X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo); 810 x_lo = (UINT32)X; 811 x_hi = (UINT32)(X >> 32); 812 813 res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo); 814 815 T = ((UINT64)x_lo << 32); 816 res += T; 817 if (res < T) 818 res += 59; 819 820 res += data; 821 if (res < data) 822 res += 59; 823 824 return res; 825 } 826 827 828 /* Although UMAC is specified to use a ramped polynomial hash scheme, this 829 * implementation does not handle all ramp levels. Because we don't handle 830 * the ramp up to p128 modulus in this implementation, we are limited to 831 * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24 832 * bytes input to UMAC per tag, ie. 16MB). 833 */ 834 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[]) 835 { 836 int i; 837 UINT64 *data=(UINT64*)data_in; 838 839 for (i = 0; i < STREAMS; i++) { 840 if ((UINT32)(data[i] >> 32) == 0xfffffffful) { 841 hc->poly_accum[i] = poly64(hc->poly_accum[i], 842 hc->poly_key_8[i], p64 - 1); 843 hc->poly_accum[i] = poly64(hc->poly_accum[i], 844 hc->poly_key_8[i], (data[i] - 59)); 845 } else { 846 hc->poly_accum[i] = poly64(hc->poly_accum[i], 847 hc->poly_key_8[i], data[i]); 848 } 849 } 850 } 851 852 853 /* ---------------------------------------------------------------------- */ 854 855 856 /* The final step in UHASH is an inner-product hash. The poly hash 857 * produces a result not neccesarily WORD_LEN bytes long. The inner- 858 * product hash breaks the polyhash output into 16-bit chunks and 859 * multiplies each with a 36 bit key. 860 */ 861 862 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data) 863 { 864 t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48); 865 t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32); 866 t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16); 867 t = t + ipkp[3] * (UINT64)(UINT16)(data); 868 869 return t; 870 } 871 872 static UINT32 ip_reduce_p36(UINT64 t) 873 { 874 /* Divisionless modular reduction */ 875 UINT64 ret; 876 877 ret = (t & m36) + 5 * (t >> 36); 878 if (ret >= p36) 879 ret -= p36; 880 881 /* return least significant 32 bits */ 882 return (UINT32)(ret); 883 } 884 885 886 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then 887 * the polyhash stage is skipped and ip_short is applied directly to the 888 * NH output. 889 */ 890 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res) 891 { 892 UINT64 t; 893 UINT64 *nhp = (UINT64 *)nh_res; 894 895 t = ip_aux(0,ahc->ip_keys, nhp[0]); 896 STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]); 897 #if (UMAC_OUTPUT_LEN >= 8) 898 t = ip_aux(0,ahc->ip_keys+4, nhp[1]); 899 STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]); 900 #endif 901 #if (UMAC_OUTPUT_LEN >= 12) 902 t = ip_aux(0,ahc->ip_keys+8, nhp[2]); 903 STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]); 904 #endif 905 #if (UMAC_OUTPUT_LEN == 16) 906 t = ip_aux(0,ahc->ip_keys+12, nhp[3]); 907 STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]); 908 #endif 909 } 910 911 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then 912 * the polyhash stage is not skipped and ip_long is applied to the 913 * polyhash output. 914 */ 915 static void ip_long(uhash_ctx_t ahc, u_char *res) 916 { 917 int i; 918 UINT64 t; 919 920 for (i = 0; i < STREAMS; i++) { 921 /* fix polyhash output not in Z_p64 */ 922 if (ahc->poly_accum[i] >= p64) 923 ahc->poly_accum[i] -= p64; 924 t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]); 925 STORE_UINT32_BIG((UINT32 *)res+i, 926 ip_reduce_p36(t) ^ ahc->ip_trans[i]); 927 } 928 } 929 930 931 /* ---------------------------------------------------------------------- */ 932 933 /* ---------------------------------------------------------------------- */ 934 935 /* Reset uhash context for next hash session */ 936 static int uhash_reset(uhash_ctx_t pc) 937 { 938 nh_reset(&pc->hash); 939 pc->msg_len = 0; 940 pc->poly_accum[0] = 1; 941 #if (UMAC_OUTPUT_LEN >= 8) 942 pc->poly_accum[1] = 1; 943 #endif 944 #if (UMAC_OUTPUT_LEN >= 12) 945 pc->poly_accum[2] = 1; 946 #endif 947 #if (UMAC_OUTPUT_LEN == 16) 948 pc->poly_accum[3] = 1; 949 #endif 950 return 1; 951 } 952 953 /* ---------------------------------------------------------------------- */ 954 955 /* Given a pointer to the internal key needed by kdf() and a uhash context, 956 * initialize the NH context and generate keys needed for poly and inner- 957 * product hashing. All keys are endian adjusted in memory so that native 958 * loads cause correct keys to be in registers during calculation. 959 */ 960 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key) 961 { 962 int i; 963 UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)]; 964 965 /* Zero the entire uhash context */ 966 memset(ahc, 0, sizeof(uhash_ctx)); 967 968 /* Initialize the L1 hash */ 969 nh_init(&ahc->hash, prf_key); 970 971 /* Setup L2 hash variables */ 972 kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */ 973 for (i = 0; i < STREAMS; i++) { 974 /* Fill keys from the buffer, skipping bytes in the buffer not 975 * used by this implementation. Endian reverse the keys if on a 976 * little-endian computer. 977 */ 978 memcpy(ahc->poly_key_8+i, buf+24*i, 8); 979 endian_convert_if_le(ahc->poly_key_8+i, 8, 8); 980 /* Mask the 64-bit keys to their special domain */ 981 ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu; 982 ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */ 983 } 984 985 /* Setup L3-1 hash variables */ 986 kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */ 987 for (i = 0; i < STREAMS; i++) 988 memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64), 989 4*sizeof(UINT64)); 990 endian_convert_if_le(ahc->ip_keys, sizeof(UINT64), 991 sizeof(ahc->ip_keys)); 992 for (i = 0; i < STREAMS*4; i++) 993 ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */ 994 995 /* Setup L3-2 hash variables */ 996 /* Fill buffer with index 4 key */ 997 kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32)); 998 endian_convert_if_le(ahc->ip_trans, sizeof(UINT32), 999 STREAMS * sizeof(UINT32)); 1000 } 1001 1002 /* ---------------------------------------------------------------------- */ 1003 1004 #if 0 1005 static uhash_ctx_t uhash_alloc(u_char key[]) 1006 { 1007 /* Allocate memory and force to a 16-byte boundary. */ 1008 uhash_ctx_t ctx; 1009 u_char bytes_to_add; 1010 aes_int_key prf_key; 1011 1012 ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY); 1013 if (ctx) { 1014 if (ALLOC_BOUNDARY) { 1015 bytes_to_add = ALLOC_BOUNDARY - 1016 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1)); 1017 ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add); 1018 *((u_char *)ctx - 1) = bytes_to_add; 1019 } 1020 aes_key_setup(key,prf_key); 1021 uhash_init(ctx, prf_key); 1022 } 1023 return (ctx); 1024 } 1025 #endif 1026 1027 /* ---------------------------------------------------------------------- */ 1028 1029 #if 0 1030 static int uhash_free(uhash_ctx_t ctx) 1031 { 1032 /* Free memory allocated by uhash_alloc */ 1033 u_char bytes_to_sub; 1034 1035 if (ctx) { 1036 if (ALLOC_BOUNDARY) { 1037 bytes_to_sub = *((u_char *)ctx - 1); 1038 ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub); 1039 } 1040 free(ctx); 1041 } 1042 return (1); 1043 } 1044 #endif 1045 /* ---------------------------------------------------------------------- */ 1046 1047 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len) 1048 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and 1049 * hash each one with NH, calling the polyhash on each NH output. 1050 */ 1051 { 1052 UWORD bytes_hashed, bytes_remaining; 1053 UINT64 result_buf[STREAMS]; 1054 UINT8 *nh_result = (UINT8 *)&result_buf; 1055 1056 if (ctx->msg_len + len <= L1_KEY_LEN) { 1057 nh_update(&ctx->hash, (const UINT8 *)input, len); 1058 ctx->msg_len += len; 1059 } else { 1060 1061 bytes_hashed = ctx->msg_len % L1_KEY_LEN; 1062 if (ctx->msg_len == L1_KEY_LEN) 1063 bytes_hashed = L1_KEY_LEN; 1064 1065 if (bytes_hashed + len >= L1_KEY_LEN) { 1066 1067 /* If some bytes have been passed to the hash function */ 1068 /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */ 1069 /* bytes to complete the current nh_block. */ 1070 if (bytes_hashed) { 1071 bytes_remaining = (L1_KEY_LEN - bytes_hashed); 1072 nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining); 1073 nh_final(&ctx->hash, nh_result); 1074 ctx->msg_len += bytes_remaining; 1075 poly_hash(ctx,(UINT32 *)nh_result); 1076 len -= bytes_remaining; 1077 input += bytes_remaining; 1078 } 1079 1080 /* Hash directly from input stream if enough bytes */ 1081 while (len >= L1_KEY_LEN) { 1082 nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN, 1083 L1_KEY_LEN, nh_result); 1084 ctx->msg_len += L1_KEY_LEN; 1085 len -= L1_KEY_LEN; 1086 input += L1_KEY_LEN; 1087 poly_hash(ctx,(UINT32 *)nh_result); 1088 } 1089 } 1090 1091 /* pass remaining < L1_KEY_LEN bytes of input data to NH */ 1092 if (len) { 1093 nh_update(&ctx->hash, (const UINT8 *)input, len); 1094 ctx->msg_len += len; 1095 } 1096 } 1097 1098 return (1); 1099 } 1100 1101 /* ---------------------------------------------------------------------- */ 1102 1103 static int uhash_final(uhash_ctx_t ctx, u_char *res) 1104 /* Incorporate any pending data, pad, and generate tag */ 1105 { 1106 UINT64 result_buf[STREAMS]; 1107 UINT8 *nh_result = (UINT8 *)&result_buf; 1108 1109 if (ctx->msg_len > L1_KEY_LEN) { 1110 if (ctx->msg_len % L1_KEY_LEN) { 1111 nh_final(&ctx->hash, nh_result); 1112 poly_hash(ctx,(UINT32 *)nh_result); 1113 } 1114 ip_long(ctx, res); 1115 } else { 1116 nh_final(&ctx->hash, nh_result); 1117 ip_short(ctx,nh_result, res); 1118 } 1119 uhash_reset(ctx); 1120 return (1); 1121 } 1122 1123 /* ---------------------------------------------------------------------- */ 1124 1125 #if 0 1126 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res) 1127 /* assumes that msg is in a writable buffer of length divisible by */ 1128 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */ 1129 { 1130 UINT8 nh_result[STREAMS*sizeof(UINT64)]; 1131 UINT32 nh_len; 1132 int extra_zeroes_needed; 1133 1134 /* If the message to be hashed is no longer than L1_HASH_LEN, we skip 1135 * the polyhash. 1136 */ 1137 if (len <= L1_KEY_LEN) { 1138 if (len == 0) /* If zero length messages will not */ 1139 nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */ 1140 else 1141 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1142 extra_zeroes_needed = nh_len - len; 1143 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1144 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1145 ip_short(ahc,nh_result, res); 1146 } else { 1147 /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH 1148 * output to poly_hash(). 1149 */ 1150 do { 1151 nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result); 1152 poly_hash(ahc,(UINT32 *)nh_result); 1153 len -= L1_KEY_LEN; 1154 msg += L1_KEY_LEN; 1155 } while (len >= L1_KEY_LEN); 1156 if (len) { 1157 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1158 extra_zeroes_needed = nh_len - len; 1159 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1160 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1161 poly_hash(ahc,(UINT32 *)nh_result); 1162 } 1163 1164 ip_long(ahc, res); 1165 } 1166 1167 uhash_reset(ahc); 1168 return 1; 1169 } 1170 #endif 1171 1172 /* ---------------------------------------------------------------------- */ 1173 /* ---------------------------------------------------------------------- */ 1174 /* ----- Begin UMAC Section --------------------------------------------- */ 1175 /* ---------------------------------------------------------------------- */ 1176 /* ---------------------------------------------------------------------- */ 1177 1178 /* The UMAC interface has two interfaces, an all-at-once interface where 1179 * the entire message to be authenticated is passed to UMAC in one buffer, 1180 * and a sequential interface where the message is presented a little at a 1181 * time. The all-at-once is more optimaized than the sequential version and 1182 * should be preferred when the sequential interface is not required. 1183 */ 1184 struct umac_ctx { 1185 uhash_ctx hash; /* Hash function for message compression */ 1186 pdf_ctx pdf; /* PDF for hashed output */ 1187 void *free_ptr; /* Address to free this struct via */ 1188 } umac_ctx; 1189 1190 /* ---------------------------------------------------------------------- */ 1191 1192 #if 0 1193 int umac_reset(struct umac_ctx *ctx) 1194 /* Reset the hash function to begin a new authentication. */ 1195 { 1196 uhash_reset(&ctx->hash); 1197 return (1); 1198 } 1199 #endif 1200 1201 /* ---------------------------------------------------------------------- */ 1202 1203 int umac_delete(struct umac_ctx *ctx) 1204 /* Deallocate the ctx structure */ 1205 { 1206 if (ctx) { 1207 if (ALLOC_BOUNDARY) 1208 ctx = (struct umac_ctx *)ctx->free_ptr; 1209 free(ctx); 1210 } 1211 return (1); 1212 } 1213 1214 /* ---------------------------------------------------------------------- */ 1215 1216 struct umac_ctx *umac_new(const u_char key[]) 1217 /* Dynamically allocate a umac_ctx struct, initialize variables, 1218 * generate subkeys from key. Align to 16-byte boundary. 1219 */ 1220 { 1221 struct umac_ctx *ctx, *octx; 1222 size_t bytes_to_add; 1223 aes_int_key prf_key; 1224 1225 octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY); 1226 if (ctx) { 1227 if (ALLOC_BOUNDARY) { 1228 bytes_to_add = ALLOC_BOUNDARY - 1229 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1)); 1230 ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add); 1231 } 1232 ctx->free_ptr = octx; 1233 aes_key_setup(key, prf_key); 1234 pdf_init(&ctx->pdf, prf_key); 1235 uhash_init(&ctx->hash, prf_key); 1236 } 1237 1238 return (ctx); 1239 } 1240 1241 /* ---------------------------------------------------------------------- */ 1242 1243 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8]) 1244 /* Incorporate any pending data, pad, and generate tag */ 1245 { 1246 uhash_final(&ctx->hash, (u_char *)tag); 1247 pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag); 1248 1249 return (1); 1250 } 1251 1252 /* ---------------------------------------------------------------------- */ 1253 1254 int umac_update(struct umac_ctx *ctx, const u_char *input, long len) 1255 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */ 1256 /* hash each one, calling the PDF on the hashed output whenever the hash- */ 1257 /* output buffer is full. */ 1258 { 1259 uhash_update(&ctx->hash, input, len); 1260 return (1); 1261 } 1262 1263 /* ---------------------------------------------------------------------- */ 1264 1265 #if 0 1266 int umac(struct umac_ctx *ctx, u_char *input, 1267 long len, u_char tag[], 1268 u_char nonce[8]) 1269 /* All-in-one version simply calls umac_update() and umac_final(). */ 1270 { 1271 uhash(&ctx->hash, input, len, (u_char *)tag); 1272 pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); 1273 1274 return (1); 1275 } 1276 #endif 1277 1278 /* ---------------------------------------------------------------------- */ 1279 /* ---------------------------------------------------------------------- */ 1280 /* ----- End UMAC Section ----------------------------------------------- */ 1281 /* ---------------------------------------------------------------------- */ 1282 /* ---------------------------------------------------------------------- */ 1283