1 /* $NetBSD: umac.c,v 1.15 2018/04/06 18:59:00 christos Exp $ */ 2 /* $OpenBSD: umac.c,v 1.16 2017/12/12 15:06:12 naddy 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 #ifndef UMAC_OUTPUT_LEN 57 #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */ 58 #endif 59 /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */ 60 /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */ 61 /* #define SSE2 0 Is SSE2 is available? */ 62 /* #define RUN_TESTS 0 Run basic correctness/speed tests */ 63 /* #define UMAC_AE_SUPPORT 0 Enable auhthenticated encrytion */ 64 65 /* ---------------------------------------------------------------------- */ 66 /* -- Global Includes --------------------------------------------------- */ 67 /* ---------------------------------------------------------------------- */ 68 69 #include "includes.h" 70 __RCSID("$NetBSD: umac.c,v 1.15 2018/04/06 18:59:00 christos Exp $"); 71 #include <sys/types.h> 72 #include <sys/endian.h> 73 #include <string.h> 74 #include <stdio.h> 75 #include <stdlib.h> 76 #include <stddef.h> 77 #include <time.h> 78 79 #include "xmalloc.h" 80 #include "umac.h" 81 #include "misc.h" 82 83 /* ---------------------------------------------------------------------- */ 84 /* --- Primitive Data Types --- */ 85 /* ---------------------------------------------------------------------- */ 86 87 /* The following assumptions may need change on your system */ 88 typedef u_int8_t UINT8; /* 1 byte */ 89 typedef u_int16_t UINT16; /* 2 byte */ 90 typedef u_int32_t UINT32; /* 4 byte */ 91 typedef u_int64_t UINT64; /* 8 bytes */ 92 typedef unsigned int UWORD; /* Register */ 93 94 /* ---------------------------------------------------------------------- */ 95 /* --- Constants -------------------------------------------------------- */ 96 /* ---------------------------------------------------------------------- */ 97 98 #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */ 99 100 /* Message "words" are read from memory in an endian-specific manner. */ 101 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */ 102 /* be set true if the host computer is little-endian. */ 103 104 #if BYTE_ORDER == LITTLE_ENDIAN 105 #define __LITTLE_ENDIAN__ 1 106 #else 107 #define __LITTLE_ENDIAN__ 0 108 #endif 109 110 /* ---------------------------------------------------------------------- */ 111 /* ---------------------------------------------------------------------- */ 112 /* ----- Architecture Specific ------------------------------------------ */ 113 /* ---------------------------------------------------------------------- */ 114 /* ---------------------------------------------------------------------- */ 115 116 117 /* ---------------------------------------------------------------------- */ 118 /* ---------------------------------------------------------------------- */ 119 /* ----- Primitive Routines --------------------------------------------- */ 120 /* ---------------------------------------------------------------------- */ 121 /* ---------------------------------------------------------------------- */ 122 123 124 /* ---------------------------------------------------------------------- */ 125 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */ 126 /* ---------------------------------------------------------------------- */ 127 128 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b))) 129 130 /* ---------------------------------------------------------------------- */ 131 /* --- Endian Conversion --- Forcing assembly on some platforms */ 132 /* ---------------------------------------------------------------------- */ 133 134 /* The following definitions use the above reversal-primitives to do the right 135 * thing on endian specific load and stores. 136 */ 137 138 #if BYTE_ORDER == LITTLE_ENDIAN 139 #define LOAD_UINT32_REVERSED(p) get_u32(p) 140 #define STORE_UINT32_REVERSED(p,v) put_u32(p,v) 141 #else 142 #define LOAD_UINT32_REVERSED(p) get_u32_le(p) 143 #define STORE_UINT32_REVERSED(p,v) put_u32_le(p,v) 144 #endif 145 146 #define LOAD_UINT32_LITTLE(p) (get_u32_le(p)) 147 #define STORE_UINT32_BIG(p,v) put_u32(p, v) 148 149 150 151 /* ---------------------------------------------------------------------- */ 152 /* ---------------------------------------------------------------------- */ 153 /* ----- Begin KDF & PDF Section ---------------------------------------- */ 154 /* ---------------------------------------------------------------------- */ 155 /* ---------------------------------------------------------------------- */ 156 157 /* UMAC uses AES with 16 byte block and key lengths */ 158 #define AES_BLOCK_LEN 16 159 160 #ifdef WITH_OPENSSL 161 #include <openssl/aes.h> 162 typedef AES_KEY aes_int_key[1]; 163 #define aes_encryption(in,out,int_key) \ 164 AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key) 165 #define aes_key_setup(key,int_key) \ 166 AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key) 167 #else 168 #include "rijndael.h" 169 #define AES_ROUNDS ((UMAC_KEY_LEN / 4) + 6) 170 typedef UINT8 aes_int_key[AES_ROUNDS+1][4][4]; /* AES internal */ 171 #define aes_encryption(in,out,int_key) \ 172 rijndaelEncrypt((u32 *)(int_key), AES_ROUNDS, (u8 *)(in), (u8 *)(out)) 173 #define aes_key_setup(key,int_key) \ 174 rijndaelKeySetupEnc((u32 *)(int_key), (const unsigned char *)(key), \ 175 UMAC_KEY_LEN*8) 176 #endif 177 178 /* The user-supplied UMAC key is stretched using AES in a counter 179 * mode to supply all random bits needed by UMAC. The kdf function takes 180 * an AES internal key representation 'key' and writes a stream of 181 * 'nbytes' bytes to the memory pointed at by 'buffer_ptr'. Each distinct 182 * 'ndx' causes a distinct byte stream. 183 */ 184 static void kdf(void *buffer_ptr, aes_int_key key, UINT8 ndx, int nbytes) 185 { 186 UINT8 in_buf[AES_BLOCK_LEN] = {0}; 187 UINT8 out_buf[AES_BLOCK_LEN]; 188 UINT8 *dst_buf = (UINT8 *)buffer_ptr; 189 int i; 190 191 /* Setup the initial value */ 192 in_buf[AES_BLOCK_LEN-9] = ndx; 193 in_buf[AES_BLOCK_LEN-1] = i = 1; 194 195 while (nbytes >= AES_BLOCK_LEN) { 196 aes_encryption(in_buf, out_buf, key); 197 memcpy(dst_buf,out_buf,AES_BLOCK_LEN); 198 in_buf[AES_BLOCK_LEN-1] = ++i; 199 nbytes -= AES_BLOCK_LEN; 200 dst_buf += AES_BLOCK_LEN; 201 } 202 if (nbytes) { 203 aes_encryption(in_buf, out_buf, key); 204 memcpy(dst_buf,out_buf,nbytes); 205 } 206 explicit_bzero(in_buf, sizeof(in_buf)); 207 explicit_bzero(out_buf, sizeof(out_buf)); 208 } 209 210 /* The final UHASH result is XOR'd with the output of a pseudorandom 211 * function. Here, we use AES to generate random output and 212 * xor the appropriate bytes depending on the last bits of nonce. 213 * This scheme is optimized for sequential, increasing big-endian nonces. 214 */ 215 216 typedef struct { 217 UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */ 218 UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */ 219 aes_int_key prf_key; /* Expanded AES key for PDF */ 220 } pdf_ctx; 221 222 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key) 223 { 224 UINT8 buf[UMAC_KEY_LEN]; 225 226 kdf(buf, prf_key, 0, UMAC_KEY_LEN); 227 aes_key_setup(buf, pc->prf_key); 228 229 /* Initialize pdf and cache */ 230 memset(pc->nonce, 0, sizeof(pc->nonce)); 231 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 232 explicit_bzero(buf, sizeof(buf)); 233 } 234 235 static inline void 236 xor64(uint8_t *dp, int di, uint8_t *sp, int si) 237 { 238 uint64_t dst, src; 239 memcpy(&dst, dp + sizeof(dst) * di, sizeof(dst)); 240 memcpy(&src, sp + sizeof(src) * si, sizeof(src)); 241 dst ^= src; 242 memcpy(dp + sizeof(dst) * di, &dst, sizeof(dst)); 243 } 244 245 __unused static inline void 246 xor32(uint8_t *dp, int di, uint8_t *sp, int si) 247 { 248 uint32_t dst, src; 249 memcpy(&dst, dp + sizeof(dst) * di, sizeof(dst)); 250 memcpy(&src, sp + sizeof(src) * si, sizeof(src)); 251 dst ^= src; 252 memcpy(dp + sizeof(dst) * di, &dst, sizeof(dst)); 253 } 254 255 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8]) 256 { 257 /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes 258 * of the AES output. If last time around we returned the ndx-1st 259 * element, then we may have the result in the cache already. 260 */ 261 262 #if (UMAC_OUTPUT_LEN == 4) 263 #define LOW_BIT_MASK 3 264 #elif (UMAC_OUTPUT_LEN == 8) 265 #define LOW_BIT_MASK 1 266 #elif (UMAC_OUTPUT_LEN > 8) 267 #define LOW_BIT_MASK 0 268 #endif 269 union { 270 UINT8 tmp_nonce_lo[4]; 271 UINT32 align; 272 } t; 273 #if LOW_BIT_MASK != 0 274 int ndx = nonce[7] & LOW_BIT_MASK; 275 #endif 276 memcpy(t.tmp_nonce_lo, nonce + 4, sizeof(t.tmp_nonce_lo)); 277 t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */ 278 279 if (memcmp(t.tmp_nonce_lo, pc->nonce + 1, sizeof(t.tmp_nonce_lo)) != 0 || 280 memcmp(nonce, pc->nonce, sizeof(t.tmp_nonce_lo)) != 0) 281 { 282 memcpy(pc->nonce, nonce, sizeof(t.tmp_nonce_lo)); 283 memcpy(pc->nonce + 4, t.tmp_nonce_lo, sizeof(t.tmp_nonce_lo)); 284 aes_encryption(pc->nonce, pc->cache, pc->prf_key); 285 } 286 287 #if (UMAC_OUTPUT_LEN == 4) 288 xor32(buf, 0, pc->cache, ndx); 289 #elif (UMAC_OUTPUT_LEN == 8) 290 xor64(buf, 0, pc->cache, ndx); 291 #elif (UMAC_OUTPUT_LEN == 12) 292 xor64(buf, 0, pc->cache, 0); 293 xor32(buf, 2, pc->cache, 2); 294 #elif (UMAC_OUTPUT_LEN == 16) 295 xor64(buf, 0, pc->cache, 0); 296 xor64(buf, 1, pc->cache, 1); 297 #endif 298 } 299 300 /* ---------------------------------------------------------------------- */ 301 /* ---------------------------------------------------------------------- */ 302 /* ----- Begin NH Hash Section ------------------------------------------ */ 303 /* ---------------------------------------------------------------------- */ 304 /* ---------------------------------------------------------------------- */ 305 306 /* The NH-based hash functions used in UMAC are described in the UMAC paper 307 * and specification, both of which can be found at the UMAC website. 308 * The interface to this implementation has two 309 * versions, one expects the entire message being hashed to be passed 310 * in a single buffer and returns the hash result immediately. The second 311 * allows the message to be passed in a sequence of buffers. In the 312 * muliple-buffer interface, the client calls the routine nh_update() as 313 * many times as necessary. When there is no more data to be fed to the 314 * hash, the client calls nh_final() which calculates the hash output. 315 * Before beginning another hash calculation the nh_reset() routine 316 * must be called. The single-buffer routine, nh(), is equivalent to 317 * the sequence of calls nh_update() and nh_final(); however it is 318 * optimized and should be prefered whenever the multiple-buffer interface 319 * is not necessary. When using either interface, it is the client's 320 * responsability to pass no more than L1_KEY_LEN bytes per hash result. 321 * 322 * The routine nh_init() initializes the nh_ctx data structure and 323 * must be called once, before any other PDF routine. 324 */ 325 326 /* The "nh_aux" routines do the actual NH hashing work. They 327 * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines 328 * produce output for all STREAMS NH iterations in one call, 329 * allowing the parallel implementation of the streams. 330 */ 331 332 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */ 333 #define L1_KEY_LEN 1024 /* Internal key bytes */ 334 #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */ 335 #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */ 336 #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */ 337 #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */ 338 339 typedef struct { 340 UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */ 341 UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */ 342 int next_data_empty; /* Bookeeping variable for data buffer. */ 343 int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */ 344 UINT64 state[STREAMS]; /* on-line state */ 345 } nh_ctx; 346 347 348 #if (UMAC_OUTPUT_LEN == 4) 349 350 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 351 /* NH hashing primitive. Previous (partial) hash result is loaded and 352 * then stored via hp pointer. The length of the data pointed at by "dp", 353 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key 354 * is expected to be endian compensated in memory at key setup. 355 */ 356 { 357 UINT64 h; 358 UWORD c = dlen / 32; 359 UINT32 *k = (UINT32 *)kp; 360 const UINT32 *d = (const UINT32 *)dp; 361 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 362 UINT32 k0,k1,k2,k3,k4,k5,k6,k7; 363 364 h = *((UINT64 *)hp); 365 do { 366 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 367 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 368 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 369 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 370 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 371 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 372 h += MUL64((k0 + d0), (k4 + d4)); 373 h += MUL64((k1 + d1), (k5 + d5)); 374 h += MUL64((k2 + d2), (k6 + d6)); 375 h += MUL64((k3 + d3), (k7 + d7)); 376 377 d += 8; 378 k += 8; 379 } while (--c); 380 *((UINT64 *)hp) = h; 381 } 382 383 #elif (UMAC_OUTPUT_LEN == 8) 384 385 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 386 /* Same as previous nh_aux, but two streams are handled in one pass, 387 * reading and writing 16 bytes of hash-state per call. 388 */ 389 { 390 UINT64 h1,h2; 391 UWORD c = dlen / 32; 392 UINT32 *k = (UINT32 *)kp; 393 const UINT32 *d = (const UINT32 *)dp; 394 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 395 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 396 k8,k9,k10,k11; 397 398 h1 = *((UINT64 *)hp); 399 h2 = *((UINT64 *)hp + 1); 400 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 401 do { 402 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 403 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 404 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 405 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 406 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 407 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 408 409 h1 += MUL64((k0 + d0), (k4 + d4)); 410 h2 += MUL64((k4 + d0), (k8 + d4)); 411 412 h1 += MUL64((k1 + d1), (k5 + d5)); 413 h2 += MUL64((k5 + d1), (k9 + d5)); 414 415 h1 += MUL64((k2 + d2), (k6 + d6)); 416 h2 += MUL64((k6 + d2), (k10 + d6)); 417 418 h1 += MUL64((k3 + d3), (k7 + d7)); 419 h2 += MUL64((k7 + d3), (k11 + d7)); 420 421 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 422 423 d += 8; 424 k += 8; 425 } while (--c); 426 ((UINT64 *)hp)[0] = h1; 427 ((UINT64 *)hp)[1] = h2; 428 } 429 430 #elif (UMAC_OUTPUT_LEN == 12) 431 432 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 433 /* Same as previous nh_aux, but two streams are handled in one pass, 434 * reading and writing 24 bytes of hash-state per call. 435 */ 436 { 437 UINT64 h1,h2,h3; 438 UWORD c = dlen / 32; 439 UINT32 *k = (UINT32 *)kp; 440 const UINT32 *d = (const UINT32 *)dp; 441 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 442 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 443 k8,k9,k10,k11,k12,k13,k14,k15; 444 445 h1 = *((UINT64 *)hp); 446 h2 = *((UINT64 *)hp + 1); 447 h3 = *((UINT64 *)hp + 2); 448 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 449 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 450 do { 451 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 452 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 453 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 454 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 455 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 456 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 457 458 h1 += MUL64((k0 + d0), (k4 + d4)); 459 h2 += MUL64((k4 + d0), (k8 + d4)); 460 h3 += MUL64((k8 + d0), (k12 + d4)); 461 462 h1 += MUL64((k1 + d1), (k5 + d5)); 463 h2 += MUL64((k5 + d1), (k9 + d5)); 464 h3 += MUL64((k9 + d1), (k13 + d5)); 465 466 h1 += MUL64((k2 + d2), (k6 + d6)); 467 h2 += MUL64((k6 + d2), (k10 + d6)); 468 h3 += MUL64((k10 + d2), (k14 + d6)); 469 470 h1 += MUL64((k3 + d3), (k7 + d7)); 471 h2 += MUL64((k7 + d3), (k11 + d7)); 472 h3 += MUL64((k11 + d3), (k15 + d7)); 473 474 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 475 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 476 477 d += 8; 478 k += 8; 479 } while (--c); 480 ((UINT64 *)hp)[0] = h1; 481 ((UINT64 *)hp)[1] = h2; 482 ((UINT64 *)hp)[2] = h3; 483 } 484 485 #elif (UMAC_OUTPUT_LEN == 16) 486 487 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen) 488 /* Same as previous nh_aux, but two streams are handled in one pass, 489 * reading and writing 24 bytes of hash-state per call. 490 */ 491 { 492 UINT64 h1,h2,h3,h4; 493 UWORD c = dlen / 32; 494 UINT32 *k = (UINT32 *)kp; 495 const UINT32 *d = (const UINT32 *)dp; 496 UINT32 d0,d1,d2,d3,d4,d5,d6,d7; 497 UINT32 k0,k1,k2,k3,k4,k5,k6,k7, 498 k8,k9,k10,k11,k12,k13,k14,k15, 499 k16,k17,k18,k19; 500 501 h1 = *((UINT64 *)hp); 502 h2 = *((UINT64 *)hp + 1); 503 h3 = *((UINT64 *)hp + 2); 504 h4 = *((UINT64 *)hp + 3); 505 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); 506 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); 507 do { 508 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); 509 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); 510 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); 511 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); 512 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); 513 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); 514 k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19); 515 516 h1 += MUL64((k0 + d0), (k4 + d4)); 517 h2 += MUL64((k4 + d0), (k8 + d4)); 518 h3 += MUL64((k8 + d0), (k12 + d4)); 519 h4 += MUL64((k12 + d0), (k16 + d4)); 520 521 h1 += MUL64((k1 + d1), (k5 + d5)); 522 h2 += MUL64((k5 + d1), (k9 + d5)); 523 h3 += MUL64((k9 + d1), (k13 + d5)); 524 h4 += MUL64((k13 + d1), (k17 + d5)); 525 526 h1 += MUL64((k2 + d2), (k6 + d6)); 527 h2 += MUL64((k6 + d2), (k10 + d6)); 528 h3 += MUL64((k10 + d2), (k14 + d6)); 529 h4 += MUL64((k14 + d2), (k18 + d6)); 530 531 h1 += MUL64((k3 + d3), (k7 + d7)); 532 h2 += MUL64((k7 + d3), (k11 + d7)); 533 h3 += MUL64((k11 + d3), (k15 + d7)); 534 h4 += MUL64((k15 + d3), (k19 + d7)); 535 536 k0 = k8; k1 = k9; k2 = k10; k3 = k11; 537 k4 = k12; k5 = k13; k6 = k14; k7 = k15; 538 k8 = k16; k9 = k17; k10 = k18; k11 = k19; 539 540 d += 8; 541 k += 8; 542 } while (--c); 543 ((UINT64 *)hp)[0] = h1; 544 ((UINT64 *)hp)[1] = h2; 545 ((UINT64 *)hp)[2] = h3; 546 ((UINT64 *)hp)[3] = h4; 547 } 548 549 /* ---------------------------------------------------------------------- */ 550 #endif /* UMAC_OUTPUT_LENGTH */ 551 /* ---------------------------------------------------------------------- */ 552 553 554 /* ---------------------------------------------------------------------- */ 555 556 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes) 557 /* This function is a wrapper for the primitive NH hash functions. It takes 558 * as argument "hc" the current hash context and a buffer which must be a 559 * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset 560 * appropriately according to how much message has been hashed already. 561 */ 562 { 563 UINT8 *key; 564 565 key = hc->nh_key + hc->bytes_hashed; 566 nh_aux(key, buf, hc->state, nbytes); 567 } 568 569 /* ---------------------------------------------------------------------- */ 570 571 #if (__LITTLE_ENDIAN__) 572 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes) 573 /* We endian convert the keys on little-endian computers to */ 574 /* compensate for the lack of big-endian memory reads during hashing. */ 575 { 576 UWORD iters = num_bytes / bpw; 577 if (bpw == 4) { 578 UINT32 *p = (UINT32 *)buf; 579 do { 580 *p = LOAD_UINT32_REVERSED(p); 581 p++; 582 } while (--iters); 583 } else if (bpw == 8) { 584 UINT64 *p = (UINT64 *)buf; 585 UINT64 th; 586 UINT64 t; 587 do { 588 t = LOAD_UINT32_REVERSED((UINT32 *)p+1); 589 th = LOAD_UINT32_REVERSED((UINT32 *)p); 590 *p++ = t | (th << 32); 591 } while (--iters); 592 } 593 } 594 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) 595 #else 596 #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */ 597 #endif 598 599 /* ---------------------------------------------------------------------- */ 600 601 static void nh_reset(nh_ctx *hc) 602 /* Reset nh_ctx to ready for hashing of new data */ 603 { 604 hc->bytes_hashed = 0; 605 hc->next_data_empty = 0; 606 hc->state[0] = 0; 607 #if (UMAC_OUTPUT_LEN >= 8) 608 hc->state[1] = 0; 609 #endif 610 #if (UMAC_OUTPUT_LEN >= 12) 611 hc->state[2] = 0; 612 #endif 613 #if (UMAC_OUTPUT_LEN == 16) 614 hc->state[3] = 0; 615 #endif 616 617 } 618 619 /* ---------------------------------------------------------------------- */ 620 621 static void nh_init(nh_ctx *hc, aes_int_key prf_key) 622 /* Generate nh_key, endian convert and reset to be ready for hashing. */ 623 { 624 kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key)); 625 endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key)); 626 nh_reset(hc); 627 } 628 629 /* ---------------------------------------------------------------------- */ 630 631 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes) 632 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */ 633 /* even multiple of HASH_BUF_BYTES. */ 634 { 635 UINT32 i,j; 636 637 j = hc->next_data_empty; 638 if ((j + nbytes) >= HASH_BUF_BYTES) { 639 if (j) { 640 i = HASH_BUF_BYTES - j; 641 memcpy(hc->data+j, buf, i); 642 nh_transform(hc,hc->data,HASH_BUF_BYTES); 643 nbytes -= i; 644 buf += i; 645 hc->bytes_hashed += HASH_BUF_BYTES; 646 } 647 if (nbytes >= HASH_BUF_BYTES) { 648 i = nbytes & ~(HASH_BUF_BYTES - 1); 649 nh_transform(hc, buf, i); 650 nbytes -= i; 651 buf += i; 652 hc->bytes_hashed += i; 653 } 654 j = 0; 655 } 656 memcpy(hc->data + j, buf, nbytes); 657 hc->next_data_empty = j + nbytes; 658 } 659 660 /* ---------------------------------------------------------------------- */ 661 662 static void zero_pad(UINT8 *p, int nbytes) 663 { 664 /* Write "nbytes" of zeroes, beginning at "p" */ 665 if (nbytes >= (int)sizeof(UWORD)) { 666 while ((ptrdiff_t)p % sizeof(UWORD)) { 667 *p = 0; 668 nbytes--; 669 p++; 670 } 671 while (nbytes >= (int)sizeof(UWORD)) { 672 *(UWORD *)p = 0; 673 nbytes -= sizeof(UWORD); 674 p += sizeof(UWORD); 675 } 676 } 677 while (nbytes) { 678 *p = 0; 679 nbytes--; 680 p++; 681 } 682 } 683 684 /* ---------------------------------------------------------------------- */ 685 686 static void nh_final(nh_ctx *hc, UINT8 *result) 687 /* After passing some number of data buffers to nh_update() for integration 688 * into an NH context, nh_final is called to produce a hash result. If any 689 * bytes are in the buffer hc->data, incorporate them into the 690 * NH context. Finally, add into the NH accumulation "state" the total number 691 * of bits hashed. The resulting numbers are written to the buffer "result". 692 * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated. 693 */ 694 { 695 int nh_len, nbits; 696 697 if (hc->next_data_empty != 0) { 698 nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) & 699 ~(L1_PAD_BOUNDARY - 1)); 700 zero_pad(hc->data + hc->next_data_empty, 701 nh_len - hc->next_data_empty); 702 nh_transform(hc, hc->data, nh_len); 703 hc->bytes_hashed += hc->next_data_empty; 704 } else if (hc->bytes_hashed == 0) { 705 nh_len = L1_PAD_BOUNDARY; 706 zero_pad(hc->data, L1_PAD_BOUNDARY); 707 nh_transform(hc, hc->data, nh_len); 708 } 709 710 nbits = (hc->bytes_hashed << 3); 711 ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits; 712 #if (UMAC_OUTPUT_LEN >= 8) 713 ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits; 714 #endif 715 #if (UMAC_OUTPUT_LEN >= 12) 716 ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits; 717 #endif 718 #if (UMAC_OUTPUT_LEN == 16) 719 ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits; 720 #endif 721 nh_reset(hc); 722 } 723 724 /* ---------------------------------------------------------------------- */ 725 726 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len, 727 UINT32 unpadded_len, UINT8 *result) 728 /* All-in-one nh_update() and nh_final() equivalent. 729 * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is 730 * well aligned 731 */ 732 { 733 UINT32 nbits; 734 735 /* Initialize the hash state */ 736 nbits = (unpadded_len << 3); 737 738 ((UINT64 *)result)[0] = nbits; 739 #if (UMAC_OUTPUT_LEN >= 8) 740 ((UINT64 *)result)[1] = nbits; 741 #endif 742 #if (UMAC_OUTPUT_LEN >= 12) 743 ((UINT64 *)result)[2] = nbits; 744 #endif 745 #if (UMAC_OUTPUT_LEN == 16) 746 ((UINT64 *)result)[3] = nbits; 747 #endif 748 749 nh_aux(hc->nh_key, buf, result, padded_len); 750 } 751 752 /* ---------------------------------------------------------------------- */ 753 /* ---------------------------------------------------------------------- */ 754 /* ----- Begin UHASH Section -------------------------------------------- */ 755 /* ---------------------------------------------------------------------- */ 756 /* ---------------------------------------------------------------------- */ 757 758 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first 759 * hashed by NH. The NH output is then hashed by a polynomial-hash layer 760 * unless the initial data to be hashed is short. After the polynomial- 761 * layer, an inner-product hash is used to produce the final UHASH output. 762 * 763 * UHASH provides two interfaces, one all-at-once and another where data 764 * buffers are presented sequentially. In the sequential interface, the 765 * UHASH client calls the routine uhash_update() as many times as necessary. 766 * When there is no more data to be fed to UHASH, the client calls 767 * uhash_final() which 768 * calculates the UHASH output. Before beginning another UHASH calculation 769 * the uhash_reset() routine must be called. The all-at-once UHASH routine, 770 * uhash(), is equivalent to the sequence of calls uhash_update() and 771 * uhash_final(); however it is optimized and should be 772 * used whenever the sequential interface is not necessary. 773 * 774 * The routine uhash_init() initializes the uhash_ctx data structure and 775 * must be called once, before any other UHASH routine. 776 */ 777 778 /* ---------------------------------------------------------------------- */ 779 /* ----- Constants and uhash_ctx ---------------------------------------- */ 780 /* ---------------------------------------------------------------------- */ 781 782 /* ---------------------------------------------------------------------- */ 783 /* ----- Poly hash and Inner-Product hash Constants --------------------- */ 784 /* ---------------------------------------------------------------------- */ 785 786 /* Primes and masks */ 787 #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */ 788 #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */ 789 #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */ 790 791 792 /* ---------------------------------------------------------------------- */ 793 794 typedef struct uhash_ctx { 795 nh_ctx hash; /* Hash context for L1 NH hash */ 796 UINT64 poly_key_8[STREAMS]; /* p64 poly keys */ 797 UINT64 poly_accum[STREAMS]; /* poly hash result */ 798 UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */ 799 UINT32 ip_trans[STREAMS]; /* Inner-product translation */ 800 UINT32 msg_len; /* Total length of data passed */ 801 /* to uhash */ 802 } uhash_ctx; 803 typedef struct uhash_ctx *uhash_ctx_t; 804 805 /* ---------------------------------------------------------------------- */ 806 807 808 /* The polynomial hashes use Horner's rule to evaluate a polynomial one 809 * word at a time. As described in the specification, poly32 and poly64 810 * require keys from special domains. The following implementations exploit 811 * the special domains to avoid overflow. The results are not guaranteed to 812 * be within Z_p32 and Z_p64, but the Inner-Product hash implementation 813 * patches any errant values. 814 */ 815 816 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data) 817 { 818 UINT32 key_hi = (UINT32)(key >> 32), 819 key_lo = (UINT32)key, 820 cur_hi = (UINT32)(cur >> 32), 821 cur_lo = (UINT32)cur, 822 x_lo, 823 x_hi; 824 UINT64 X,T,res; 825 826 X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo); 827 x_lo = (UINT32)X; 828 x_hi = (UINT32)(X >> 32); 829 830 res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo); 831 832 T = ((UINT64)x_lo << 32); 833 res += T; 834 if (res < T) 835 res += 59; 836 837 res += data; 838 if (res < data) 839 res += 59; 840 841 return res; 842 } 843 844 845 /* Although UMAC is specified to use a ramped polynomial hash scheme, this 846 * implementation does not handle all ramp levels. Because we don't handle 847 * the ramp up to p128 modulus in this implementation, we are limited to 848 * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24 849 * bytes input to UMAC per tag, ie. 16MB). 850 */ 851 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[]) 852 { 853 int i; 854 UINT64 *data=(UINT64*)data_in; 855 856 for (i = 0; i < STREAMS; i++) { 857 if ((UINT32)(data[i] >> 32) == 0xfffffffful) { 858 hc->poly_accum[i] = poly64(hc->poly_accum[i], 859 hc->poly_key_8[i], p64 - 1); 860 hc->poly_accum[i] = poly64(hc->poly_accum[i], 861 hc->poly_key_8[i], (data[i] - 59)); 862 } else { 863 hc->poly_accum[i] = poly64(hc->poly_accum[i], 864 hc->poly_key_8[i], data[i]); 865 } 866 } 867 } 868 869 870 /* ---------------------------------------------------------------------- */ 871 872 873 /* The final step in UHASH is an inner-product hash. The poly hash 874 * produces a result not neccesarily WORD_LEN bytes long. The inner- 875 * product hash breaks the polyhash output into 16-bit chunks and 876 * multiplies each with a 36 bit key. 877 */ 878 879 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data) 880 { 881 t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48); 882 t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32); 883 t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16); 884 t = t + ipkp[3] * (UINT64)(UINT16)(data); 885 886 return t; 887 } 888 889 static UINT32 ip_reduce_p36(UINT64 t) 890 { 891 /* Divisionless modular reduction */ 892 UINT64 ret; 893 894 ret = (t & m36) + 5 * (t >> 36); 895 if (ret >= p36) 896 ret -= p36; 897 898 /* return least significant 32 bits */ 899 return (UINT32)(ret); 900 } 901 902 903 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then 904 * the polyhash stage is skipped and ip_short is applied directly to the 905 * NH output. 906 */ 907 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res) 908 { 909 UINT64 t; 910 UINT64 *nhp = (UINT64 *)nh_res; 911 912 t = ip_aux(0,ahc->ip_keys, nhp[0]); 913 STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]); 914 #if (UMAC_OUTPUT_LEN >= 8) 915 t = ip_aux(0,ahc->ip_keys+4, nhp[1]); 916 STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]); 917 #endif 918 #if (UMAC_OUTPUT_LEN >= 12) 919 t = ip_aux(0,ahc->ip_keys+8, nhp[2]); 920 STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]); 921 #endif 922 #if (UMAC_OUTPUT_LEN == 16) 923 t = ip_aux(0,ahc->ip_keys+12, nhp[3]); 924 STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]); 925 #endif 926 } 927 928 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then 929 * the polyhash stage is not skipped and ip_long is applied to the 930 * polyhash output. 931 */ 932 static void ip_long(uhash_ctx_t ahc, u_char *res) 933 { 934 int i; 935 UINT64 t; 936 937 for (i = 0; i < STREAMS; i++) { 938 /* fix polyhash output not in Z_p64 */ 939 if (ahc->poly_accum[i] >= p64) 940 ahc->poly_accum[i] -= p64; 941 t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]); 942 STORE_UINT32_BIG((UINT32 *)res+i, 943 ip_reduce_p36(t) ^ ahc->ip_trans[i]); 944 } 945 } 946 947 948 /* ---------------------------------------------------------------------- */ 949 950 /* ---------------------------------------------------------------------- */ 951 952 /* Reset uhash context for next hash session */ 953 static int uhash_reset(uhash_ctx_t pc) 954 { 955 nh_reset(&pc->hash); 956 pc->msg_len = 0; 957 pc->poly_accum[0] = 1; 958 #if (UMAC_OUTPUT_LEN >= 8) 959 pc->poly_accum[1] = 1; 960 #endif 961 #if (UMAC_OUTPUT_LEN >= 12) 962 pc->poly_accum[2] = 1; 963 #endif 964 #if (UMAC_OUTPUT_LEN == 16) 965 pc->poly_accum[3] = 1; 966 #endif 967 return 1; 968 } 969 970 /* ---------------------------------------------------------------------- */ 971 972 /* Given a pointer to the internal key needed by kdf() and a uhash context, 973 * initialize the NH context and generate keys needed for poly and inner- 974 * product hashing. All keys are endian adjusted in memory so that native 975 * loads cause correct keys to be in registers during calculation. 976 */ 977 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key) 978 { 979 int i; 980 UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)]; 981 982 /* Zero the entire uhash context */ 983 memset(ahc, 0, sizeof(uhash_ctx)); 984 985 /* Initialize the L1 hash */ 986 nh_init(&ahc->hash, prf_key); 987 988 /* Setup L2 hash variables */ 989 kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */ 990 for (i = 0; i < STREAMS; i++) { 991 /* Fill keys from the buffer, skipping bytes in the buffer not 992 * used by this implementation. Endian reverse the keys if on a 993 * little-endian computer. 994 */ 995 memcpy(ahc->poly_key_8+i, buf+24*i, 8); 996 endian_convert_if_le(ahc->poly_key_8+i, 8, 8); 997 /* Mask the 64-bit keys to their special domain */ 998 ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu; 999 ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */ 1000 } 1001 1002 /* Setup L3-1 hash variables */ 1003 kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */ 1004 for (i = 0; i < STREAMS; i++) 1005 memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64), 1006 4*sizeof(UINT64)); 1007 endian_convert_if_le(ahc->ip_keys, sizeof(UINT64), 1008 sizeof(ahc->ip_keys)); 1009 for (i = 0; i < STREAMS*4; i++) 1010 ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */ 1011 1012 /* Setup L3-2 hash variables */ 1013 /* Fill buffer with index 4 key */ 1014 kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32)); 1015 endian_convert_if_le(ahc->ip_trans, sizeof(UINT32), 1016 STREAMS * sizeof(UINT32)); 1017 explicit_bzero(buf, sizeof(buf)); 1018 } 1019 1020 /* ---------------------------------------------------------------------- */ 1021 1022 #if 0 1023 static uhash_ctx_t uhash_alloc(u_char key[]) 1024 { 1025 /* Allocate memory and force to a 16-byte boundary. */ 1026 uhash_ctx_t ctx; 1027 u_char bytes_to_add; 1028 aes_int_key prf_key; 1029 1030 ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY); 1031 if (ctx) { 1032 if (ALLOC_BOUNDARY) { 1033 bytes_to_add = ALLOC_BOUNDARY - 1034 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1)); 1035 ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add); 1036 *((u_char *)ctx - 1) = bytes_to_add; 1037 } 1038 aes_key_setup(key,prf_key); 1039 uhash_init(ctx, prf_key); 1040 } 1041 return (ctx); 1042 } 1043 #endif 1044 1045 /* ---------------------------------------------------------------------- */ 1046 1047 #if 0 1048 static int uhash_free(uhash_ctx_t ctx) 1049 { 1050 /* Free memory allocated by uhash_alloc */ 1051 u_char bytes_to_sub; 1052 1053 if (ctx) { 1054 if (ALLOC_BOUNDARY) { 1055 bytes_to_sub = *((u_char *)ctx - 1); 1056 ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub); 1057 } 1058 free(ctx); 1059 } 1060 return (1); 1061 } 1062 #endif 1063 /* ---------------------------------------------------------------------- */ 1064 1065 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len) 1066 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and 1067 * hash each one with NH, calling the polyhash on each NH output. 1068 */ 1069 { 1070 UWORD bytes_hashed, bytes_remaining; 1071 UINT64 result_buf[STREAMS]; 1072 UINT8 *nh_result = (UINT8 *)&result_buf; 1073 1074 if (ctx->msg_len + len <= L1_KEY_LEN) { 1075 nh_update(&ctx->hash, (const UINT8 *)input, len); 1076 ctx->msg_len += len; 1077 } else { 1078 1079 bytes_hashed = ctx->msg_len % L1_KEY_LEN; 1080 if (ctx->msg_len == L1_KEY_LEN) 1081 bytes_hashed = L1_KEY_LEN; 1082 1083 if (bytes_hashed + len >= L1_KEY_LEN) { 1084 1085 /* If some bytes have been passed to the hash function */ 1086 /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */ 1087 /* bytes to complete the current nh_block. */ 1088 if (bytes_hashed) { 1089 bytes_remaining = (L1_KEY_LEN - bytes_hashed); 1090 nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining); 1091 nh_final(&ctx->hash, nh_result); 1092 ctx->msg_len += bytes_remaining; 1093 poly_hash(ctx,(UINT32 *)nh_result); 1094 len -= bytes_remaining; 1095 input += bytes_remaining; 1096 } 1097 1098 /* Hash directly from input stream if enough bytes */ 1099 while (len >= L1_KEY_LEN) { 1100 nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN, 1101 L1_KEY_LEN, nh_result); 1102 ctx->msg_len += L1_KEY_LEN; 1103 len -= L1_KEY_LEN; 1104 input += L1_KEY_LEN; 1105 poly_hash(ctx,(UINT32 *)nh_result); 1106 } 1107 } 1108 1109 /* pass remaining < L1_KEY_LEN bytes of input data to NH */ 1110 if (len) { 1111 nh_update(&ctx->hash, (const UINT8 *)input, len); 1112 ctx->msg_len += len; 1113 } 1114 } 1115 1116 return (1); 1117 } 1118 1119 /* ---------------------------------------------------------------------- */ 1120 1121 static int uhash_final(uhash_ctx_t ctx, u_char *res) 1122 /* Incorporate any pending data, pad, and generate tag */ 1123 { 1124 UINT64 result_buf[STREAMS]; 1125 UINT8 *nh_result = (UINT8 *)&result_buf; 1126 1127 if (ctx->msg_len > L1_KEY_LEN) { 1128 if (ctx->msg_len % L1_KEY_LEN) { 1129 nh_final(&ctx->hash, nh_result); 1130 poly_hash(ctx,(UINT32 *)nh_result); 1131 } 1132 ip_long(ctx, res); 1133 } else { 1134 nh_final(&ctx->hash, nh_result); 1135 ip_short(ctx,nh_result, res); 1136 } 1137 uhash_reset(ctx); 1138 return (1); 1139 } 1140 1141 /* ---------------------------------------------------------------------- */ 1142 1143 #if 0 1144 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res) 1145 /* assumes that msg is in a writable buffer of length divisible by */ 1146 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */ 1147 { 1148 UINT8 nh_result[STREAMS*sizeof(UINT64)]; 1149 UINT32 nh_len; 1150 int extra_zeroes_needed; 1151 1152 /* If the message to be hashed is no longer than L1_HASH_LEN, we skip 1153 * the polyhash. 1154 */ 1155 if (len <= L1_KEY_LEN) { 1156 if (len == 0) /* If zero length messages will not */ 1157 nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */ 1158 else 1159 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1160 extra_zeroes_needed = nh_len - len; 1161 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1162 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1163 ip_short(ahc,nh_result, res); 1164 } else { 1165 /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH 1166 * output to poly_hash(). 1167 */ 1168 do { 1169 nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result); 1170 poly_hash(ahc,(UINT32 *)nh_result); 1171 len -= L1_KEY_LEN; 1172 msg += L1_KEY_LEN; 1173 } while (len >= L1_KEY_LEN); 1174 if (len) { 1175 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); 1176 extra_zeroes_needed = nh_len - len; 1177 zero_pad((UINT8 *)msg + len, extra_zeroes_needed); 1178 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); 1179 poly_hash(ahc,(UINT32 *)nh_result); 1180 } 1181 1182 ip_long(ahc, res); 1183 } 1184 1185 uhash_reset(ahc); 1186 return 1; 1187 } 1188 #endif 1189 1190 /* ---------------------------------------------------------------------- */ 1191 /* ---------------------------------------------------------------------- */ 1192 /* ----- Begin UMAC Section --------------------------------------------- */ 1193 /* ---------------------------------------------------------------------- */ 1194 /* ---------------------------------------------------------------------- */ 1195 1196 /* The UMAC interface has two interfaces, an all-at-once interface where 1197 * the entire message to be authenticated is passed to UMAC in one buffer, 1198 * and a sequential interface where the message is presented a little at a 1199 * time. The all-at-once is more optimaized than the sequential version and 1200 * should be preferred when the sequential interface is not required. 1201 */ 1202 struct umac_ctx { 1203 uhash_ctx hash; /* Hash function for message compression */ 1204 pdf_ctx pdf; /* PDF for hashed output */ 1205 void *free_ptr; /* Address to free this struct via */ 1206 } umac_ctx; 1207 1208 /* ---------------------------------------------------------------------- */ 1209 1210 #if 0 1211 int umac_reset(struct umac_ctx *ctx) 1212 /* Reset the hash function to begin a new authentication. */ 1213 { 1214 uhash_reset(&ctx->hash); 1215 return (1); 1216 } 1217 #endif 1218 1219 /* ---------------------------------------------------------------------- */ 1220 1221 int umac_delete(struct umac_ctx *ctx) 1222 /* Deallocate the ctx structure */ 1223 { 1224 if (ctx) { 1225 if (ALLOC_BOUNDARY) 1226 ctx = (struct umac_ctx *)ctx->free_ptr; 1227 explicit_bzero(ctx, sizeof(*ctx) + ALLOC_BOUNDARY); 1228 free(ctx); 1229 } 1230 return (1); 1231 } 1232 1233 /* ---------------------------------------------------------------------- */ 1234 1235 struct umac_ctx *umac_new(const u_char key[]) 1236 /* Dynamically allocate a umac_ctx struct, initialize variables, 1237 * generate subkeys from key. Align to 16-byte boundary. 1238 */ 1239 { 1240 struct umac_ctx *ctx, *octx; 1241 size_t bytes_to_add; 1242 aes_int_key prf_key; 1243 1244 octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY); 1245 if (ctx) { 1246 if (ALLOC_BOUNDARY) { 1247 bytes_to_add = ALLOC_BOUNDARY - 1248 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1)); 1249 ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add); 1250 } 1251 ctx->free_ptr = octx; 1252 aes_key_setup(key, prf_key); 1253 pdf_init(&ctx->pdf, prf_key); 1254 uhash_init(&ctx->hash, prf_key); 1255 explicit_bzero(prf_key, sizeof(prf_key)); 1256 } 1257 1258 return (ctx); 1259 } 1260 1261 /* ---------------------------------------------------------------------- */ 1262 1263 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8]) 1264 /* Incorporate any pending data, pad, and generate tag */ 1265 { 1266 uhash_final(&ctx->hash, (u_char *)tag); 1267 pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag); 1268 1269 return (1); 1270 } 1271 1272 /* ---------------------------------------------------------------------- */ 1273 1274 int umac_update(struct umac_ctx *ctx, const u_char *input, long len) 1275 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */ 1276 /* hash each one, calling the PDF on the hashed output whenever the hash- */ 1277 /* output buffer is full. */ 1278 { 1279 uhash_update(&ctx->hash, input, len); 1280 return (1); 1281 } 1282 1283 /* ---------------------------------------------------------------------- */ 1284 1285 #if 0 1286 int umac(struct umac_ctx *ctx, u_char *input, 1287 long len, u_char tag[], 1288 u_char nonce[8]) 1289 /* All-in-one version simply calls umac_update() and umac_final(). */ 1290 { 1291 uhash(&ctx->hash, input, len, (u_char *)tag); 1292 pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); 1293 1294 return (1); 1295 } 1296 #endif 1297 1298 /* ---------------------------------------------------------------------- */ 1299 /* ---------------------------------------------------------------------- */ 1300 /* ----- End UMAC Section ----------------------------------------------- */ 1301 /* ---------------------------------------------------------------------- */ 1302 /* ---------------------------------------------------------------------- */ 1303