]>
Commit | Line | Data |
---|---|---|
1 | /* $OpenBSD: umac.c,v 1.3 2008/05/12 20:52:20 pvalchev Exp $ */ | |
2 | /* ----------------------------------------------------------------------- | |
3 | * | |
4 | * umac.c -- C Implementation UMAC Message Authentication | |
5 | * | |
6 | * Version 0.93b of rfc4418.txt -- 2006 July 18 | |
7 | * | |
8 | * For a full description of UMAC message authentication see the UMAC | |
9 | * world-wide-web page at http://www.cs.ucdavis.edu/~rogaway/umac | |
10 | * Please report bugs and suggestions to the UMAC webpage. | |
11 | * | |
12 | * Copyright (c) 1999-2006 Ted Krovetz | |
13 | * | |
14 | * Permission to use, copy, modify, and distribute this software and | |
15 | * its documentation for any purpose and with or without fee, is hereby | |
16 | * granted provided that the above copyright notice appears in all copies | |
17 | * and in supporting documentation, and that the name of the copyright | |
18 | * holder not be used in advertising or publicity pertaining to | |
19 | * distribution of the software without specific, written prior permission. | |
20 | * | |
21 | * Comments should be directed to Ted Krovetz (tdk@acm.org) | |
22 | * | |
23 | * ---------------------------------------------------------------------- */ | |
24 | ||
25 | /* ////////////////////// IMPORTANT NOTES ///////////////////////////////// | |
26 | * | |
27 | * 1) This version does not work properly on messages larger than 16MB | |
28 | * | |
29 | * 2) If you set the switch to use SSE2, then all data must be 16-byte | |
30 | * aligned | |
31 | * | |
32 | * 3) When calling the function umac(), it is assumed that msg is in | |
33 | * a writable buffer of length divisible by 32 bytes. The message itself | |
34 | * does not have to fill the entire buffer, but bytes beyond msg may be | |
35 | * zeroed. | |
36 | * | |
37 | * 4) Three free AES implementations are supported by this implementation of | |
38 | * UMAC. Paulo Barreto's version is in the public domain and can be found | |
39 | * at http://www.esat.kuleuven.ac.be/~rijmen/rijndael/ (search for | |
40 | * "Barreto"). The only two files needed are rijndael-alg-fst.c and | |
41 | * rijndael-alg-fst.h. Brian Gladman's version is distributed with the GNU | |
42 | * Public lisence at http://fp.gladman.plus.com/AES/index.htm. It | |
43 | * includes a fast IA-32 assembly version. The OpenSSL crypo library is | |
44 | * the third. | |
45 | * | |
46 | * 5) With FORCE_C_ONLY flags set to 0, incorrect results are sometimes | |
47 | * produced under gcc with optimizations set -O3 or higher. Dunno why. | |
48 | * | |
49 | /////////////////////////////////////////////////////////////////////// */ | |
50 | ||
51 | /* ---------------------------------------------------------------------- */ | |
52 | /* --- User Switches ---------------------------------------------------- */ | |
53 | /* ---------------------------------------------------------------------- */ | |
54 | ||
55 | #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */ | |
56 | /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */ | |
57 | /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */ | |
58 | /* #define SSE2 0 Is SSE2 is available? */ | |
59 | /* #define RUN_TESTS 0 Run basic correctness/speed tests */ | |
60 | /* #define UMAC_AE_SUPPORT 0 Enable auhthenticated encrytion */ | |
61 | ||
62 | /* ---------------------------------------------------------------------- */ | |
63 | /* -- Global Includes --------------------------------------------------- */ | |
64 | /* ---------------------------------------------------------------------- */ | |
65 | ||
66 | #include "includes.h" | |
67 | #include <sys/types.h> | |
68 | ||
69 | #include "xmalloc.h" | |
70 | #include "umac.h" | |
71 | #include <string.h> | |
72 | #include <stdlib.h> | |
73 | #include <stddef.h> | |
74 | ||
75 | /* ---------------------------------------------------------------------- */ | |
76 | /* --- Primitive Data Types --- */ | |
77 | /* ---------------------------------------------------------------------- */ | |
78 | ||
79 | /* The following assumptions may need change on your system */ | |
80 | typedef u_int8_t UINT8; /* 1 byte */ | |
81 | typedef u_int16_t UINT16; /* 2 byte */ | |
82 | typedef u_int32_t UINT32; /* 4 byte */ | |
83 | typedef u_int64_t UINT64; /* 8 bytes */ | |
84 | typedef unsigned int UWORD; /* Register */ | |
85 | ||
86 | /* ---------------------------------------------------------------------- */ | |
87 | /* --- Constants -------------------------------------------------------- */ | |
88 | /* ---------------------------------------------------------------------- */ | |
89 | ||
90 | #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */ | |
91 | ||
92 | /* Message "words" are read from memory in an endian-specific manner. */ | |
93 | /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */ | |
94 | /* be set true if the host computer is little-endian. */ | |
95 | ||
96 | #if BYTE_ORDER == LITTLE_ENDIAN | |
97 | #define __LITTLE_ENDIAN__ 1 | |
98 | #else | |
99 | #define __LITTLE_ENDIAN__ 0 | |
100 | #endif | |
101 | ||
102 | /* ---------------------------------------------------------------------- */ | |
103 | /* ---------------------------------------------------------------------- */ | |
104 | /* ----- Architecture Specific ------------------------------------------ */ | |
105 | /* ---------------------------------------------------------------------- */ | |
106 | /* ---------------------------------------------------------------------- */ | |
107 | ||
108 | ||
109 | /* ---------------------------------------------------------------------- */ | |
110 | /* ---------------------------------------------------------------------- */ | |
111 | /* ----- Primitive Routines --------------------------------------------- */ | |
112 | /* ---------------------------------------------------------------------- */ | |
113 | /* ---------------------------------------------------------------------- */ | |
114 | ||
115 | ||
116 | /* ---------------------------------------------------------------------- */ | |
117 | /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */ | |
118 | /* ---------------------------------------------------------------------- */ | |
119 | ||
120 | #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b))) | |
121 | ||
122 | /* ---------------------------------------------------------------------- */ | |
123 | /* --- Endian Conversion --- Forcing assembly on some platforms */ | |
124 | /* ---------------------------------------------------------------------- */ | |
125 | ||
126 | #if HAVE_SWAP32 | |
127 | #define LOAD_UINT32_REVERSED(p) (swap32(*(UINT32 *)(p))) | |
128 | #define STORE_UINT32_REVERSED(p,v) (*(UINT32 *)(p) = swap32(v)) | |
129 | #else /* HAVE_SWAP32 */ | |
130 | ||
131 | static UINT32 LOAD_UINT32_REVERSED(void *ptr) | |
132 | { | |
133 | UINT32 temp = *(UINT32 *)ptr; | |
134 | temp = (temp >> 24) | ((temp & 0x00FF0000) >> 8 ) | |
135 | | ((temp & 0x0000FF00) << 8 ) | (temp << 24); | |
136 | return (UINT32)temp; | |
137 | } | |
138 | ||
139 | # if (__LITTLE_ENDIAN__) | |
140 | static void STORE_UINT32_REVERSED(void *ptr, UINT32 x) | |
141 | { | |
142 | UINT32 i = (UINT32)x; | |
143 | *(UINT32 *)ptr = (i >> 24) | ((i & 0x00FF0000) >> 8 ) | |
144 | | ((i & 0x0000FF00) << 8 ) | (i << 24); | |
145 | } | |
146 | # endif /* __LITTLE_ENDIAN */ | |
147 | #endif /* HAVE_SWAP32 */ | |
148 | ||
149 | /* The following definitions use the above reversal-primitives to do the right | |
150 | * thing on endian specific load and stores. | |
151 | */ | |
152 | ||
153 | #if (__LITTLE_ENDIAN__) | |
154 | #define LOAD_UINT32_LITTLE(ptr) (*(UINT32 *)(ptr)) | |
155 | #define STORE_UINT32_BIG(ptr,x) STORE_UINT32_REVERSED(ptr,x) | |
156 | #else | |
157 | #define LOAD_UINT32_LITTLE(ptr) LOAD_UINT32_REVERSED(ptr) | |
158 | #define STORE_UINT32_BIG(ptr,x) (*(UINT32 *)(ptr) = (UINT32)(x)) | |
159 | #endif | |
160 | ||
161 | /* ---------------------------------------------------------------------- */ | |
162 | /* ---------------------------------------------------------------------- */ | |
163 | /* ----- Begin KDF & PDF Section ---------------------------------------- */ | |
164 | /* ---------------------------------------------------------------------- */ | |
165 | /* ---------------------------------------------------------------------- */ | |
166 | ||
167 | /* UMAC uses AES with 16 byte block and key lengths */ | |
168 | #define AES_BLOCK_LEN 16 | |
169 | ||
170 | /* OpenSSL's AES */ | |
171 | #include "openbsd-compat/openssl-compat.h" | |
172 | #ifndef USE_BUILTIN_RIJNDAEL | |
173 | # include <openssl/aes.h> | |
174 | #endif | |
175 | typedef AES_KEY aes_int_key[1]; | |
176 | #define aes_encryption(in,out,int_key) \ | |
177 | AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key) | |
178 | #define aes_key_setup(key,int_key) \ | |
179 | AES_set_encrypt_key((u_char *)(key),UMAC_KEY_LEN*8,int_key) | |
180 | ||
181 | /* The user-supplied UMAC key is stretched using AES in a counter | |
182 | * mode to supply all random bits needed by UMAC. The kdf function takes | |
183 | * an AES internal key representation 'key' and writes a stream of | |
184 | * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct | |
185 | * 'ndx' causes a distinct byte stream. | |
186 | */ | |
187 | static void kdf(void *bufp, aes_int_key key, UINT8 ndx, int nbytes) | |
188 | { | |
189 | UINT8 in_buf[AES_BLOCK_LEN] = {0}; | |
190 | UINT8 out_buf[AES_BLOCK_LEN]; | |
191 | UINT8 *dst_buf = (UINT8 *)bufp; | |
192 | int i; | |
193 | ||
194 | /* Setup the initial value */ | |
195 | in_buf[AES_BLOCK_LEN-9] = ndx; | |
196 | in_buf[AES_BLOCK_LEN-1] = i = 1; | |
197 | ||
198 | while (nbytes >= AES_BLOCK_LEN) { | |
199 | aes_encryption(in_buf, out_buf, key); | |
200 | memcpy(dst_buf,out_buf,AES_BLOCK_LEN); | |
201 | in_buf[AES_BLOCK_LEN-1] = ++i; | |
202 | nbytes -= AES_BLOCK_LEN; | |
203 | dst_buf += AES_BLOCK_LEN; | |
204 | } | |
205 | if (nbytes) { | |
206 | aes_encryption(in_buf, out_buf, key); | |
207 | memcpy(dst_buf,out_buf,nbytes); | |
208 | } | |
209 | } | |
210 | ||
211 | /* The final UHASH result is XOR'd with the output of a pseudorandom | |
212 | * function. Here, we use AES to generate random output and | |
213 | * xor the appropriate bytes depending on the last bits of nonce. | |
214 | * This scheme is optimized for sequential, increasing big-endian nonces. | |
215 | */ | |
216 | ||
217 | typedef struct { | |
218 | UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */ | |
219 | UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */ | |
220 | aes_int_key prf_key; /* Expanded AES key for PDF */ | |
221 | } pdf_ctx; | |
222 | ||
223 | static void pdf_init(pdf_ctx *pc, aes_int_key prf_key) | |
224 | { | |
225 | UINT8 buf[UMAC_KEY_LEN]; | |
226 | ||
227 | kdf(buf, prf_key, 0, UMAC_KEY_LEN); | |
228 | aes_key_setup(buf, pc->prf_key); | |
229 | ||
230 | /* Initialize pdf and cache */ | |
231 | memset(pc->nonce, 0, sizeof(pc->nonce)); | |
232 | aes_encryption(pc->nonce, pc->cache, pc->prf_key); | |
233 | } | |
234 | ||
235 | static void pdf_gen_xor(pdf_ctx *pc, UINT8 nonce[8], UINT8 buf[8]) | |
236 | { | |
237 | /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes | |
238 | * of the AES output. If last time around we returned the ndx-1st | |
239 | * element, then we may have the result in the cache already. | |
240 | */ | |
241 | ||
242 | #if (UMAC_OUTPUT_LEN == 4) | |
243 | #define LOW_BIT_MASK 3 | |
244 | #elif (UMAC_OUTPUT_LEN == 8) | |
245 | #define LOW_BIT_MASK 1 | |
246 | #elif (UMAC_OUTPUT_LEN > 8) | |
247 | #define LOW_BIT_MASK 0 | |
248 | #endif | |
249 | ||
250 | UINT8 tmp_nonce_lo[4]; | |
251 | #if LOW_BIT_MASK != 0 | |
252 | int ndx = nonce[7] & LOW_BIT_MASK; | |
253 | #endif | |
254 | *(UINT32 *)tmp_nonce_lo = ((UINT32 *)nonce)[1]; | |
255 | tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */ | |
256 | ||
257 | if ( (((UINT32 *)tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) || | |
258 | (((UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) ) | |
259 | { | |
260 | ((UINT32 *)pc->nonce)[0] = ((UINT32 *)nonce)[0]; | |
261 | ((UINT32 *)pc->nonce)[1] = ((UINT32 *)tmp_nonce_lo)[0]; | |
262 | aes_encryption(pc->nonce, pc->cache, pc->prf_key); | |
263 | } | |
264 | ||
265 | #if (UMAC_OUTPUT_LEN == 4) | |
266 | *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx]; | |
267 | #elif (UMAC_OUTPUT_LEN == 8) | |
268 | *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx]; | |
269 | #elif (UMAC_OUTPUT_LEN == 12) | |
270 | ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; | |
271 | ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2]; | |
272 | #elif (UMAC_OUTPUT_LEN == 16) | |
273 | ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; | |
274 | ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1]; | |
275 | #endif | |
276 | } | |
277 | ||
278 | /* ---------------------------------------------------------------------- */ | |
279 | /* ---------------------------------------------------------------------- */ | |
280 | /* ----- Begin NH Hash Section ------------------------------------------ */ | |
281 | /* ---------------------------------------------------------------------- */ | |
282 | /* ---------------------------------------------------------------------- */ | |
283 | ||
284 | /* The NH-based hash functions used in UMAC are described in the UMAC paper | |
285 | * and specification, both of which can be found at the UMAC website. | |
286 | * The interface to this implementation has two | |
287 | * versions, one expects the entire message being hashed to be passed | |
288 | * in a single buffer and returns the hash result immediately. The second | |
289 | * allows the message to be passed in a sequence of buffers. In the | |
290 | * muliple-buffer interface, the client calls the routine nh_update() as | |
291 | * many times as necessary. When there is no more data to be fed to the | |
292 | * hash, the client calls nh_final() which calculates the hash output. | |
293 | * Before beginning another hash calculation the nh_reset() routine | |
294 | * must be called. The single-buffer routine, nh(), is equivalent to | |
295 | * the sequence of calls nh_update() and nh_final(); however it is | |
296 | * optimized and should be prefered whenever the multiple-buffer interface | |
297 | * is not necessary. When using either interface, it is the client's | |
298 | * responsability to pass no more than L1_KEY_LEN bytes per hash result. | |
299 | * | |
300 | * The routine nh_init() initializes the nh_ctx data structure and | |
301 | * must be called once, before any other PDF routine. | |
302 | */ | |
303 | ||
304 | /* The "nh_aux" routines do the actual NH hashing work. They | |
305 | * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines | |
306 | * produce output for all STREAMS NH iterations in one call, | |
307 | * allowing the parallel implementation of the streams. | |
308 | */ | |
309 | ||
310 | #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */ | |
311 | #define L1_KEY_LEN 1024 /* Internal key bytes */ | |
312 | #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */ | |
313 | #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */ | |
314 | #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */ | |
315 | #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */ | |
316 | ||
317 | typedef struct { | |
318 | UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */ | |
319 | UINT8 data [HASH_BUF_BYTES]; /* Incomming data buffer */ | |
320 | int next_data_empty; /* Bookeeping variable for data buffer. */ | |
321 | int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */ | |
322 | UINT64 state[STREAMS]; /* on-line state */ | |
323 | } nh_ctx; | |
324 | ||
325 | ||
326 | #if (UMAC_OUTPUT_LEN == 4) | |
327 | ||
328 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | |
329 | /* NH hashing primitive. Previous (partial) hash result is loaded and | |
330 | * then stored via hp pointer. The length of the data pointed at by "dp", | |
331 | * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key | |
332 | * is expected to be endian compensated in memory at key setup. | |
333 | */ | |
334 | { | |
335 | UINT64 h; | |
336 | UWORD c = dlen / 32; | |
337 | UINT32 *k = (UINT32 *)kp; | |
338 | UINT32 *d = (UINT32 *)dp; | |
339 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | |
340 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7; | |
341 | ||
342 | h = *((UINT64 *)hp); | |
343 | do { | |
344 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | |
345 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | |
346 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | |
347 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | |
348 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | |
349 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | |
350 | h += MUL64((k0 + d0), (k4 + d4)); | |
351 | h += MUL64((k1 + d1), (k5 + d5)); | |
352 | h += MUL64((k2 + d2), (k6 + d6)); | |
353 | h += MUL64((k3 + d3), (k7 + d7)); | |
354 | ||
355 | d += 8; | |
356 | k += 8; | |
357 | } while (--c); | |
358 | *((UINT64 *)hp) = h; | |
359 | } | |
360 | ||
361 | #elif (UMAC_OUTPUT_LEN == 8) | |
362 | ||
363 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | |
364 | /* Same as previous nh_aux, but two streams are handled in one pass, | |
365 | * reading and writing 16 bytes of hash-state per call. | |
366 | */ | |
367 | { | |
368 | UINT64 h1,h2; | |
369 | UWORD c = dlen / 32; | |
370 | UINT32 *k = (UINT32 *)kp; | |
371 | UINT32 *d = (UINT32 *)dp; | |
372 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | |
373 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7, | |
374 | k8,k9,k10,k11; | |
375 | ||
376 | h1 = *((UINT64 *)hp); | |
377 | h2 = *((UINT64 *)hp + 1); | |
378 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | |
379 | do { | |
380 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | |
381 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | |
382 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | |
383 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | |
384 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | |
385 | k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); | |
386 | ||
387 | h1 += MUL64((k0 + d0), (k4 + d4)); | |
388 | h2 += MUL64((k4 + d0), (k8 + d4)); | |
389 | ||
390 | h1 += MUL64((k1 + d1), (k5 + d5)); | |
391 | h2 += MUL64((k5 + d1), (k9 + d5)); | |
392 | ||
393 | h1 += MUL64((k2 + d2), (k6 + d6)); | |
394 | h2 += MUL64((k6 + d2), (k10 + d6)); | |
395 | ||
396 | h1 += MUL64((k3 + d3), (k7 + d7)); | |
397 | h2 += MUL64((k7 + d3), (k11 + d7)); | |
398 | ||
399 | k0 = k8; k1 = k9; k2 = k10; k3 = k11; | |
400 | ||
401 | d += 8; | |
402 | k += 8; | |
403 | } while (--c); | |
404 | ((UINT64 *)hp)[0] = h1; | |
405 | ((UINT64 *)hp)[1] = h2; | |
406 | } | |
407 | ||
408 | #elif (UMAC_OUTPUT_LEN == 12) | |
409 | ||
410 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | |
411 | /* Same as previous nh_aux, but two streams are handled in one pass, | |
412 | * reading and writing 24 bytes of hash-state per call. | |
413 | */ | |
414 | { | |
415 | UINT64 h1,h2,h3; | |
416 | UWORD c = dlen / 32; | |
417 | UINT32 *k = (UINT32 *)kp; | |
418 | UINT32 *d = (UINT32 *)dp; | |
419 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | |
420 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7, | |
421 | k8,k9,k10,k11,k12,k13,k14,k15; | |
422 | ||
423 | h1 = *((UINT64 *)hp); | |
424 | h2 = *((UINT64 *)hp + 1); | |
425 | h3 = *((UINT64 *)hp + 2); | |
426 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | |
427 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | |
428 | do { | |
429 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | |
430 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | |
431 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | |
432 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | |
433 | k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); | |
434 | k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); | |
435 | ||
436 | h1 += MUL64((k0 + d0), (k4 + d4)); | |
437 | h2 += MUL64((k4 + d0), (k8 + d4)); | |
438 | h3 += MUL64((k8 + d0), (k12 + d4)); | |
439 | ||
440 | h1 += MUL64((k1 + d1), (k5 + d5)); | |
441 | h2 += MUL64((k5 + d1), (k9 + d5)); | |
442 | h3 += MUL64((k9 + d1), (k13 + d5)); | |
443 | ||
444 | h1 += MUL64((k2 + d2), (k6 + d6)); | |
445 | h2 += MUL64((k6 + d2), (k10 + d6)); | |
446 | h3 += MUL64((k10 + d2), (k14 + d6)); | |
447 | ||
448 | h1 += MUL64((k3 + d3), (k7 + d7)); | |
449 | h2 += MUL64((k7 + d3), (k11 + d7)); | |
450 | h3 += MUL64((k11 + d3), (k15 + d7)); | |
451 | ||
452 | k0 = k8; k1 = k9; k2 = k10; k3 = k11; | |
453 | k4 = k12; k5 = k13; k6 = k14; k7 = k15; | |
454 | ||
455 | d += 8; | |
456 | k += 8; | |
457 | } while (--c); | |
458 | ((UINT64 *)hp)[0] = h1; | |
459 | ((UINT64 *)hp)[1] = h2; | |
460 | ((UINT64 *)hp)[2] = h3; | |
461 | } | |
462 | ||
463 | #elif (UMAC_OUTPUT_LEN == 16) | |
464 | ||
465 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | |
466 | /* Same as previous nh_aux, but two streams are handled in one pass, | |
467 | * reading and writing 24 bytes of hash-state per call. | |
468 | */ | |
469 | { | |
470 | UINT64 h1,h2,h3,h4; | |
471 | UWORD c = dlen / 32; | |
472 | UINT32 *k = (UINT32 *)kp; | |
473 | UINT32 *d = (UINT32 *)dp; | |
474 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | |
475 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7, | |
476 | k8,k9,k10,k11,k12,k13,k14,k15, | |
477 | k16,k17,k18,k19; | |
478 | ||
479 | h1 = *((UINT64 *)hp); | |
480 | h2 = *((UINT64 *)hp + 1); | |
481 | h3 = *((UINT64 *)hp + 2); | |
482 | h4 = *((UINT64 *)hp + 3); | |
483 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | |
484 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | |
485 | do { | |
486 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | |
487 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | |
488 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | |
489 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | |
490 | k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); | |
491 | k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); | |
492 | k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19); | |
493 | ||
494 | h1 += MUL64((k0 + d0), (k4 + d4)); | |
495 | h2 += MUL64((k4 + d0), (k8 + d4)); | |
496 | h3 += MUL64((k8 + d0), (k12 + d4)); | |
497 | h4 += MUL64((k12 + d0), (k16 + d4)); | |
498 | ||
499 | h1 += MUL64((k1 + d1), (k5 + d5)); | |
500 | h2 += MUL64((k5 + d1), (k9 + d5)); | |
501 | h3 += MUL64((k9 + d1), (k13 + d5)); | |
502 | h4 += MUL64((k13 + d1), (k17 + d5)); | |
503 | ||
504 | h1 += MUL64((k2 + d2), (k6 + d6)); | |
505 | h2 += MUL64((k6 + d2), (k10 + d6)); | |
506 | h3 += MUL64((k10 + d2), (k14 + d6)); | |
507 | h4 += MUL64((k14 + d2), (k18 + d6)); | |
508 | ||
509 | h1 += MUL64((k3 + d3), (k7 + d7)); | |
510 | h2 += MUL64((k7 + d3), (k11 + d7)); | |
511 | h3 += MUL64((k11 + d3), (k15 + d7)); | |
512 | h4 += MUL64((k15 + d3), (k19 + d7)); | |
513 | ||
514 | k0 = k8; k1 = k9; k2 = k10; k3 = k11; | |
515 | k4 = k12; k5 = k13; k6 = k14; k7 = k15; | |
516 | k8 = k16; k9 = k17; k10 = k18; k11 = k19; | |
517 | ||
518 | d += 8; | |
519 | k += 8; | |
520 | } while (--c); | |
521 | ((UINT64 *)hp)[0] = h1; | |
522 | ((UINT64 *)hp)[1] = h2; | |
523 | ((UINT64 *)hp)[2] = h3; | |
524 | ((UINT64 *)hp)[3] = h4; | |
525 | } | |
526 | ||
527 | /* ---------------------------------------------------------------------- */ | |
528 | #endif /* UMAC_OUTPUT_LENGTH */ | |
529 | /* ---------------------------------------------------------------------- */ | |
530 | ||
531 | ||
532 | /* ---------------------------------------------------------------------- */ | |
533 | ||
534 | static void nh_transform(nh_ctx *hc, UINT8 *buf, UINT32 nbytes) | |
535 | /* This function is a wrapper for the primitive NH hash functions. It takes | |
536 | * as argument "hc" the current hash context and a buffer which must be a | |
537 | * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset | |
538 | * appropriately according to how much message has been hashed already. | |
539 | */ | |
540 | { | |
541 | UINT8 *key; | |
542 | ||
543 | key = hc->nh_key + hc->bytes_hashed; | |
544 | nh_aux(key, buf, hc->state, nbytes); | |
545 | } | |
546 | ||
547 | /* ---------------------------------------------------------------------- */ | |
548 | ||
549 | #if (__LITTLE_ENDIAN__) | |
550 | static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes) | |
551 | /* We endian convert the keys on little-endian computers to */ | |
552 | /* compensate for the lack of big-endian memory reads during hashing. */ | |
553 | { | |
554 | UWORD iters = num_bytes / bpw; | |
555 | if (bpw == 4) { | |
556 | UINT32 *p = (UINT32 *)buf; | |
557 | do { | |
558 | *p = LOAD_UINT32_REVERSED(p); | |
559 | p++; | |
560 | } while (--iters); | |
561 | } else if (bpw == 8) { | |
562 | UINT32 *p = (UINT32 *)buf; | |
563 | UINT32 t; | |
564 | do { | |
565 | t = LOAD_UINT32_REVERSED(p+1); | |
566 | p[1] = LOAD_UINT32_REVERSED(p); | |
567 | p[0] = t; | |
568 | p += 2; | |
569 | } while (--iters); | |
570 | } | |
571 | } | |
572 | #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) | |
573 | #else | |
574 | #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */ | |
575 | #endif | |
576 | ||
577 | /* ---------------------------------------------------------------------- */ | |
578 | ||
579 | static void nh_reset(nh_ctx *hc) | |
580 | /* Reset nh_ctx to ready for hashing of new data */ | |
581 | { | |
582 | hc->bytes_hashed = 0; | |
583 | hc->next_data_empty = 0; | |
584 | hc->state[0] = 0; | |
585 | #if (UMAC_OUTPUT_LEN >= 8) | |
586 | hc->state[1] = 0; | |
587 | #endif | |
588 | #if (UMAC_OUTPUT_LEN >= 12) | |
589 | hc->state[2] = 0; | |
590 | #endif | |
591 | #if (UMAC_OUTPUT_LEN == 16) | |
592 | hc->state[3] = 0; | |
593 | #endif | |
594 | ||
595 | } | |
596 | ||
597 | /* ---------------------------------------------------------------------- */ | |
598 | ||
599 | static void nh_init(nh_ctx *hc, aes_int_key prf_key) | |
600 | /* Generate nh_key, endian convert and reset to be ready for hashing. */ | |
601 | { | |
602 | kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key)); | |
603 | endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key)); | |
604 | nh_reset(hc); | |
605 | } | |
606 | ||
607 | /* ---------------------------------------------------------------------- */ | |
608 | ||
609 | static void nh_update(nh_ctx *hc, UINT8 *buf, UINT32 nbytes) | |
610 | /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */ | |
611 | /* even multiple of HASH_BUF_BYTES. */ | |
612 | { | |
613 | UINT32 i,j; | |
614 | ||
615 | j = hc->next_data_empty; | |
616 | if ((j + nbytes) >= HASH_BUF_BYTES) { | |
617 | if (j) { | |
618 | i = HASH_BUF_BYTES - j; | |
619 | memcpy(hc->data+j, buf, i); | |
620 | nh_transform(hc,hc->data,HASH_BUF_BYTES); | |
621 | nbytes -= i; | |
622 | buf += i; | |
623 | hc->bytes_hashed += HASH_BUF_BYTES; | |
624 | } | |
625 | if (nbytes >= HASH_BUF_BYTES) { | |
626 | i = nbytes & ~(HASH_BUF_BYTES - 1); | |
627 | nh_transform(hc, buf, i); | |
628 | nbytes -= i; | |
629 | buf += i; | |
630 | hc->bytes_hashed += i; | |
631 | } | |
632 | j = 0; | |
633 | } | |
634 | memcpy(hc->data + j, buf, nbytes); | |
635 | hc->next_data_empty = j + nbytes; | |
636 | } | |
637 | ||
638 | /* ---------------------------------------------------------------------- */ | |
639 | ||
640 | static void zero_pad(UINT8 *p, int nbytes) | |
641 | { | |
642 | /* Write "nbytes" of zeroes, beginning at "p" */ | |
643 | if (nbytes >= (int)sizeof(UWORD)) { | |
644 | while ((ptrdiff_t)p % sizeof(UWORD)) { | |
645 | *p = 0; | |
646 | nbytes--; | |
647 | p++; | |
648 | } | |
649 | while (nbytes >= (int)sizeof(UWORD)) { | |
650 | *(UWORD *)p = 0; | |
651 | nbytes -= sizeof(UWORD); | |
652 | p += sizeof(UWORD); | |
653 | } | |
654 | } | |
655 | while (nbytes) { | |
656 | *p = 0; | |
657 | nbytes--; | |
658 | p++; | |
659 | } | |
660 | } | |
661 | ||
662 | /* ---------------------------------------------------------------------- */ | |
663 | ||
664 | static void nh_final(nh_ctx *hc, UINT8 *result) | |
665 | /* After passing some number of data buffers to nh_update() for integration | |
666 | * into an NH context, nh_final is called to produce a hash result. If any | |
667 | * bytes are in the buffer hc->data, incorporate them into the | |
668 | * NH context. Finally, add into the NH accumulation "state" the total number | |
669 | * of bits hashed. The resulting numbers are written to the buffer "result". | |
670 | * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated. | |
671 | */ | |
672 | { | |
673 | int nh_len, nbits; | |
674 | ||
675 | if (hc->next_data_empty != 0) { | |
676 | nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) & | |
677 | ~(L1_PAD_BOUNDARY - 1)); | |
678 | zero_pad(hc->data + hc->next_data_empty, | |
679 | nh_len - hc->next_data_empty); | |
680 | nh_transform(hc, hc->data, nh_len); | |
681 | hc->bytes_hashed += hc->next_data_empty; | |
682 | } else if (hc->bytes_hashed == 0) { | |
683 | nh_len = L1_PAD_BOUNDARY; | |
684 | zero_pad(hc->data, L1_PAD_BOUNDARY); | |
685 | nh_transform(hc, hc->data, nh_len); | |
686 | } | |
687 | ||
688 | nbits = (hc->bytes_hashed << 3); | |
689 | ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits; | |
690 | #if (UMAC_OUTPUT_LEN >= 8) | |
691 | ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits; | |
692 | #endif | |
693 | #if (UMAC_OUTPUT_LEN >= 12) | |
694 | ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits; | |
695 | #endif | |
696 | #if (UMAC_OUTPUT_LEN == 16) | |
697 | ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits; | |
698 | #endif | |
699 | nh_reset(hc); | |
700 | } | |
701 | ||
702 | /* ---------------------------------------------------------------------- */ | |
703 | ||
704 | static void nh(nh_ctx *hc, UINT8 *buf, UINT32 padded_len, | |
705 | UINT32 unpadded_len, UINT8 *result) | |
706 | /* All-in-one nh_update() and nh_final() equivalent. | |
707 | * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is | |
708 | * well aligned | |
709 | */ | |
710 | { | |
711 | UINT32 nbits; | |
712 | ||
713 | /* Initialize the hash state */ | |
714 | nbits = (unpadded_len << 3); | |
715 | ||
716 | ((UINT64 *)result)[0] = nbits; | |
717 | #if (UMAC_OUTPUT_LEN >= 8) | |
718 | ((UINT64 *)result)[1] = nbits; | |
719 | #endif | |
720 | #if (UMAC_OUTPUT_LEN >= 12) | |
721 | ((UINT64 *)result)[2] = nbits; | |
722 | #endif | |
723 | #if (UMAC_OUTPUT_LEN == 16) | |
724 | ((UINT64 *)result)[3] = nbits; | |
725 | #endif | |
726 | ||
727 | nh_aux(hc->nh_key, buf, result, padded_len); | |
728 | } | |
729 | ||
730 | /* ---------------------------------------------------------------------- */ | |
731 | /* ---------------------------------------------------------------------- */ | |
732 | /* ----- Begin UHASH Section -------------------------------------------- */ | |
733 | /* ---------------------------------------------------------------------- */ | |
734 | /* ---------------------------------------------------------------------- */ | |
735 | ||
736 | /* UHASH is a multi-layered algorithm. Data presented to UHASH is first | |
737 | * hashed by NH. The NH output is then hashed by a polynomial-hash layer | |
738 | * unless the initial data to be hashed is short. After the polynomial- | |
739 | * layer, an inner-product hash is used to produce the final UHASH output. | |
740 | * | |
741 | * UHASH provides two interfaces, one all-at-once and another where data | |
742 | * buffers are presented sequentially. In the sequential interface, the | |
743 | * UHASH client calls the routine uhash_update() as many times as necessary. | |
744 | * When there is no more data to be fed to UHASH, the client calls | |
745 | * uhash_final() which | |
746 | * calculates the UHASH output. Before beginning another UHASH calculation | |
747 | * the uhash_reset() routine must be called. The all-at-once UHASH routine, | |
748 | * uhash(), is equivalent to the sequence of calls uhash_update() and | |
749 | * uhash_final(); however it is optimized and should be | |
750 | * used whenever the sequential interface is not necessary. | |
751 | * | |
752 | * The routine uhash_init() initializes the uhash_ctx data structure and | |
753 | * must be called once, before any other UHASH routine. | |
754 | */ | |
755 | ||
756 | /* ---------------------------------------------------------------------- */ | |
757 | /* ----- Constants and uhash_ctx ---------------------------------------- */ | |
758 | /* ---------------------------------------------------------------------- */ | |
759 | ||
760 | /* ---------------------------------------------------------------------- */ | |
761 | /* ----- Poly hash and Inner-Product hash Constants --------------------- */ | |
762 | /* ---------------------------------------------------------------------- */ | |
763 | ||
764 | /* Primes and masks */ | |
765 | #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */ | |
766 | #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */ | |
767 | #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */ | |
768 | ||
769 | ||
770 | /* ---------------------------------------------------------------------- */ | |
771 | ||
772 | typedef struct uhash_ctx { | |
773 | nh_ctx hash; /* Hash context for L1 NH hash */ | |
774 | UINT64 poly_key_8[STREAMS]; /* p64 poly keys */ | |
775 | UINT64 poly_accum[STREAMS]; /* poly hash result */ | |
776 | UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */ | |
777 | UINT32 ip_trans[STREAMS]; /* Inner-product translation */ | |
778 | UINT32 msg_len; /* Total length of data passed */ | |
779 | /* to uhash */ | |
780 | } uhash_ctx; | |
781 | typedef struct uhash_ctx *uhash_ctx_t; | |
782 | ||
783 | /* ---------------------------------------------------------------------- */ | |
784 | ||
785 | ||
786 | /* The polynomial hashes use Horner's rule to evaluate a polynomial one | |
787 | * word at a time. As described in the specification, poly32 and poly64 | |
788 | * require keys from special domains. The following implementations exploit | |
789 | * the special domains to avoid overflow. The results are not guaranteed to | |
790 | * be within Z_p32 and Z_p64, but the Inner-Product hash implementation | |
791 | * patches any errant values. | |
792 | */ | |
793 | ||
794 | static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data) | |
795 | { | |
796 | UINT32 key_hi = (UINT32)(key >> 32), | |
797 | key_lo = (UINT32)key, | |
798 | cur_hi = (UINT32)(cur >> 32), | |
799 | cur_lo = (UINT32)cur, | |
800 | x_lo, | |
801 | x_hi; | |
802 | UINT64 X,T,res; | |
803 | ||
804 | X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo); | |
805 | x_lo = (UINT32)X; | |
806 | x_hi = (UINT32)(X >> 32); | |
807 | ||
808 | res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo); | |
809 | ||
810 | T = ((UINT64)x_lo << 32); | |
811 | res += T; | |
812 | if (res < T) | |
813 | res += 59; | |
814 | ||
815 | res += data; | |
816 | if (res < data) | |
817 | res += 59; | |
818 | ||
819 | return res; | |
820 | } | |
821 | ||
822 | ||
823 | /* Although UMAC is specified to use a ramped polynomial hash scheme, this | |
824 | * implementation does not handle all ramp levels. Because we don't handle | |
825 | * the ramp up to p128 modulus in this implementation, we are limited to | |
826 | * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24 | |
827 | * bytes input to UMAC per tag, ie. 16MB). | |
828 | */ | |
829 | static void poly_hash(uhash_ctx_t hc, UINT32 data_in[]) | |
830 | { | |
831 | int i; | |
832 | UINT64 *data=(UINT64*)data_in; | |
833 | ||
834 | for (i = 0; i < STREAMS; i++) { | |
835 | if ((UINT32)(data[i] >> 32) == 0xfffffffful) { | |
836 | hc->poly_accum[i] = poly64(hc->poly_accum[i], | |
837 | hc->poly_key_8[i], p64 - 1); | |
838 | hc->poly_accum[i] = poly64(hc->poly_accum[i], | |
839 | hc->poly_key_8[i], (data[i] - 59)); | |
840 | } else { | |
841 | hc->poly_accum[i] = poly64(hc->poly_accum[i], | |
842 | hc->poly_key_8[i], data[i]); | |
843 | } | |
844 | } | |
845 | } | |
846 | ||
847 | ||
848 | /* ---------------------------------------------------------------------- */ | |
849 | ||
850 | ||
851 | /* The final step in UHASH is an inner-product hash. The poly hash | |
852 | * produces a result not neccesarily WORD_LEN bytes long. The inner- | |
853 | * product hash breaks the polyhash output into 16-bit chunks and | |
854 | * multiplies each with a 36 bit key. | |
855 | */ | |
856 | ||
857 | static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data) | |
858 | { | |
859 | t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48); | |
860 | t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32); | |
861 | t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16); | |
862 | t = t + ipkp[3] * (UINT64)(UINT16)(data); | |
863 | ||
864 | return t; | |
865 | } | |
866 | ||
867 | static UINT32 ip_reduce_p36(UINT64 t) | |
868 | { | |
869 | /* Divisionless modular reduction */ | |
870 | UINT64 ret; | |
871 | ||
872 | ret = (t & m36) + 5 * (t >> 36); | |
873 | if (ret >= p36) | |
874 | ret -= p36; | |
875 | ||
876 | /* return least significant 32 bits */ | |
877 | return (UINT32)(ret); | |
878 | } | |
879 | ||
880 | ||
881 | /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then | |
882 | * the polyhash stage is skipped and ip_short is applied directly to the | |
883 | * NH output. | |
884 | */ | |
885 | static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res) | |
886 | { | |
887 | UINT64 t; | |
888 | UINT64 *nhp = (UINT64 *)nh_res; | |
889 | ||
890 | t = ip_aux(0,ahc->ip_keys, nhp[0]); | |
891 | STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]); | |
892 | #if (UMAC_OUTPUT_LEN >= 8) | |
893 | t = ip_aux(0,ahc->ip_keys+4, nhp[1]); | |
894 | STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]); | |
895 | #endif | |
896 | #if (UMAC_OUTPUT_LEN >= 12) | |
897 | t = ip_aux(0,ahc->ip_keys+8, nhp[2]); | |
898 | STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]); | |
899 | #endif | |
900 | #if (UMAC_OUTPUT_LEN == 16) | |
901 | t = ip_aux(0,ahc->ip_keys+12, nhp[3]); | |
902 | STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]); | |
903 | #endif | |
904 | } | |
905 | ||
906 | /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then | |
907 | * the polyhash stage is not skipped and ip_long is applied to the | |
908 | * polyhash output. | |
909 | */ | |
910 | static void ip_long(uhash_ctx_t ahc, u_char *res) | |
911 | { | |
912 | int i; | |
913 | UINT64 t; | |
914 | ||
915 | for (i = 0; i < STREAMS; i++) { | |
916 | /* fix polyhash output not in Z_p64 */ | |
917 | if (ahc->poly_accum[i] >= p64) | |
918 | ahc->poly_accum[i] -= p64; | |
919 | t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]); | |
920 | STORE_UINT32_BIG((UINT32 *)res+i, | |
921 | ip_reduce_p36(t) ^ ahc->ip_trans[i]); | |
922 | } | |
923 | } | |
924 | ||
925 | ||
926 | /* ---------------------------------------------------------------------- */ | |
927 | ||
928 | /* ---------------------------------------------------------------------- */ | |
929 | ||
930 | /* Reset uhash context for next hash session */ | |
931 | static int uhash_reset(uhash_ctx_t pc) | |
932 | { | |
933 | nh_reset(&pc->hash); | |
934 | pc->msg_len = 0; | |
935 | pc->poly_accum[0] = 1; | |
936 | #if (UMAC_OUTPUT_LEN >= 8) | |
937 | pc->poly_accum[1] = 1; | |
938 | #endif | |
939 | #if (UMAC_OUTPUT_LEN >= 12) | |
940 | pc->poly_accum[2] = 1; | |
941 | #endif | |
942 | #if (UMAC_OUTPUT_LEN == 16) | |
943 | pc->poly_accum[3] = 1; | |
944 | #endif | |
945 | return 1; | |
946 | } | |
947 | ||
948 | /* ---------------------------------------------------------------------- */ | |
949 | ||
950 | /* Given a pointer to the internal key needed by kdf() and a uhash context, | |
951 | * initialize the NH context and generate keys needed for poly and inner- | |
952 | * product hashing. All keys are endian adjusted in memory so that native | |
953 | * loads cause correct keys to be in registers during calculation. | |
954 | */ | |
955 | static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key) | |
956 | { | |
957 | int i; | |
958 | UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)]; | |
959 | ||
960 | /* Zero the entire uhash context */ | |
961 | memset(ahc, 0, sizeof(uhash_ctx)); | |
962 | ||
963 | /* Initialize the L1 hash */ | |
964 | nh_init(&ahc->hash, prf_key); | |
965 | ||
966 | /* Setup L2 hash variables */ | |
967 | kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */ | |
968 | for (i = 0; i < STREAMS; i++) { | |
969 | /* Fill keys from the buffer, skipping bytes in the buffer not | |
970 | * used by this implementation. Endian reverse the keys if on a | |
971 | * little-endian computer. | |
972 | */ | |
973 | memcpy(ahc->poly_key_8+i, buf+24*i, 8); | |
974 | endian_convert_if_le(ahc->poly_key_8+i, 8, 8); | |
975 | /* Mask the 64-bit keys to their special domain */ | |
976 | ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu; | |
977 | ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */ | |
978 | } | |
979 | ||
980 | /* Setup L3-1 hash variables */ | |
981 | kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */ | |
982 | for (i = 0; i < STREAMS; i++) | |
983 | memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64), | |
984 | 4*sizeof(UINT64)); | |
985 | endian_convert_if_le(ahc->ip_keys, sizeof(UINT64), | |
986 | sizeof(ahc->ip_keys)); | |
987 | for (i = 0; i < STREAMS*4; i++) | |
988 | ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */ | |
989 | ||
990 | /* Setup L3-2 hash variables */ | |
991 | /* Fill buffer with index 4 key */ | |
992 | kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32)); | |
993 | endian_convert_if_le(ahc->ip_trans, sizeof(UINT32), | |
994 | STREAMS * sizeof(UINT32)); | |
995 | } | |
996 | ||
997 | /* ---------------------------------------------------------------------- */ | |
998 | ||
999 | #if 0 | |
1000 | static uhash_ctx_t uhash_alloc(u_char key[]) | |
1001 | { | |
1002 | /* Allocate memory and force to a 16-byte boundary. */ | |
1003 | uhash_ctx_t ctx; | |
1004 | u_char bytes_to_add; | |
1005 | aes_int_key prf_key; | |
1006 | ||
1007 | ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY); | |
1008 | if (ctx) { | |
1009 | if (ALLOC_BOUNDARY) { | |
1010 | bytes_to_add = ALLOC_BOUNDARY - | |
1011 | ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1)); | |
1012 | ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add); | |
1013 | *((u_char *)ctx - 1) = bytes_to_add; | |
1014 | } | |
1015 | aes_key_setup(key,prf_key); | |
1016 | uhash_init(ctx, prf_key); | |
1017 | } | |
1018 | return (ctx); | |
1019 | } | |
1020 | #endif | |
1021 | ||
1022 | /* ---------------------------------------------------------------------- */ | |
1023 | ||
1024 | #if 0 | |
1025 | static int uhash_free(uhash_ctx_t ctx) | |
1026 | { | |
1027 | /* Free memory allocated by uhash_alloc */ | |
1028 | u_char bytes_to_sub; | |
1029 | ||
1030 | if (ctx) { | |
1031 | if (ALLOC_BOUNDARY) { | |
1032 | bytes_to_sub = *((u_char *)ctx - 1); | |
1033 | ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub); | |
1034 | } | |
1035 | free(ctx); | |
1036 | } | |
1037 | return (1); | |
1038 | } | |
1039 | #endif | |
1040 | /* ---------------------------------------------------------------------- */ | |
1041 | ||
1042 | static int uhash_update(uhash_ctx_t ctx, u_char *input, long len) | |
1043 | /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and | |
1044 | * hash each one with NH, calling the polyhash on each NH output. | |
1045 | */ | |
1046 | { | |
1047 | UWORD bytes_hashed, bytes_remaining; | |
1048 | UINT64 result_buf[STREAMS]; | |
1049 | UINT8 *nh_result = (UINT8 *)&result_buf; | |
1050 | ||
1051 | if (ctx->msg_len + len <= L1_KEY_LEN) { | |
1052 | nh_update(&ctx->hash, (UINT8 *)input, len); | |
1053 | ctx->msg_len += len; | |
1054 | } else { | |
1055 | ||
1056 | bytes_hashed = ctx->msg_len % L1_KEY_LEN; | |
1057 | if (ctx->msg_len == L1_KEY_LEN) | |
1058 | bytes_hashed = L1_KEY_LEN; | |
1059 | ||
1060 | if (bytes_hashed + len >= L1_KEY_LEN) { | |
1061 | ||
1062 | /* If some bytes have been passed to the hash function */ | |
1063 | /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */ | |
1064 | /* bytes to complete the current nh_block. */ | |
1065 | if (bytes_hashed) { | |
1066 | bytes_remaining = (L1_KEY_LEN - bytes_hashed); | |
1067 | nh_update(&ctx->hash, (UINT8 *)input, bytes_remaining); | |
1068 | nh_final(&ctx->hash, nh_result); | |
1069 | ctx->msg_len += bytes_remaining; | |
1070 | poly_hash(ctx,(UINT32 *)nh_result); | |
1071 | len -= bytes_remaining; | |
1072 | input += bytes_remaining; | |
1073 | } | |
1074 | ||
1075 | /* Hash directly from input stream if enough bytes */ | |
1076 | while (len >= L1_KEY_LEN) { | |
1077 | nh(&ctx->hash, (UINT8 *)input, L1_KEY_LEN, | |
1078 | L1_KEY_LEN, nh_result); | |
1079 | ctx->msg_len += L1_KEY_LEN; | |
1080 | len -= L1_KEY_LEN; | |
1081 | input += L1_KEY_LEN; | |
1082 | poly_hash(ctx,(UINT32 *)nh_result); | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | /* pass remaining < L1_KEY_LEN bytes of input data to NH */ | |
1087 | if (len) { | |
1088 | nh_update(&ctx->hash, (UINT8 *)input, len); | |
1089 | ctx->msg_len += len; | |
1090 | } | |
1091 | } | |
1092 | ||
1093 | return (1); | |
1094 | } | |
1095 | ||
1096 | /* ---------------------------------------------------------------------- */ | |
1097 | ||
1098 | static int uhash_final(uhash_ctx_t ctx, u_char *res) | |
1099 | /* Incorporate any pending data, pad, and generate tag */ | |
1100 | { | |
1101 | UINT64 result_buf[STREAMS]; | |
1102 | UINT8 *nh_result = (UINT8 *)&result_buf; | |
1103 | ||
1104 | if (ctx->msg_len > L1_KEY_LEN) { | |
1105 | if (ctx->msg_len % L1_KEY_LEN) { | |
1106 | nh_final(&ctx->hash, nh_result); | |
1107 | poly_hash(ctx,(UINT32 *)nh_result); | |
1108 | } | |
1109 | ip_long(ctx, res); | |
1110 | } else { | |
1111 | nh_final(&ctx->hash, nh_result); | |
1112 | ip_short(ctx,nh_result, res); | |
1113 | } | |
1114 | uhash_reset(ctx); | |
1115 | return (1); | |
1116 | } | |
1117 | ||
1118 | /* ---------------------------------------------------------------------- */ | |
1119 | ||
1120 | #if 0 | |
1121 | static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res) | |
1122 | /* assumes that msg is in a writable buffer of length divisible by */ | |
1123 | /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */ | |
1124 | { | |
1125 | UINT8 nh_result[STREAMS*sizeof(UINT64)]; | |
1126 | UINT32 nh_len; | |
1127 | int extra_zeroes_needed; | |
1128 | ||
1129 | /* If the message to be hashed is no longer than L1_HASH_LEN, we skip | |
1130 | * the polyhash. | |
1131 | */ | |
1132 | if (len <= L1_KEY_LEN) { | |
1133 | if (len == 0) /* If zero length messages will not */ | |
1134 | nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */ | |
1135 | else | |
1136 | nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); | |
1137 | extra_zeroes_needed = nh_len - len; | |
1138 | zero_pad((UINT8 *)msg + len, extra_zeroes_needed); | |
1139 | nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); | |
1140 | ip_short(ahc,nh_result, res); | |
1141 | } else { | |
1142 | /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH | |
1143 | * output to poly_hash(). | |
1144 | */ | |
1145 | do { | |
1146 | nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result); | |
1147 | poly_hash(ahc,(UINT32 *)nh_result); | |
1148 | len -= L1_KEY_LEN; | |
1149 | msg += L1_KEY_LEN; | |
1150 | } while (len >= L1_KEY_LEN); | |
1151 | if (len) { | |
1152 | nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); | |
1153 | extra_zeroes_needed = nh_len - len; | |
1154 | zero_pad((UINT8 *)msg + len, extra_zeroes_needed); | |
1155 | nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); | |
1156 | poly_hash(ahc,(UINT32 *)nh_result); | |
1157 | } | |
1158 | ||
1159 | ip_long(ahc, res); | |
1160 | } | |
1161 | ||
1162 | uhash_reset(ahc); | |
1163 | return 1; | |
1164 | } | |
1165 | #endif | |
1166 | ||
1167 | /* ---------------------------------------------------------------------- */ | |
1168 | /* ---------------------------------------------------------------------- */ | |
1169 | /* ----- Begin UMAC Section --------------------------------------------- */ | |
1170 | /* ---------------------------------------------------------------------- */ | |
1171 | /* ---------------------------------------------------------------------- */ | |
1172 | ||
1173 | /* The UMAC interface has two interfaces, an all-at-once interface where | |
1174 | * the entire message to be authenticated is passed to UMAC in one buffer, | |
1175 | * and a sequential interface where the message is presented a little at a | |
1176 | * time. The all-at-once is more optimaized than the sequential version and | |
1177 | * should be preferred when the sequential interface is not required. | |
1178 | */ | |
1179 | struct umac_ctx { | |
1180 | uhash_ctx hash; /* Hash function for message compression */ | |
1181 | pdf_ctx pdf; /* PDF for hashed output */ | |
1182 | void *free_ptr; /* Address to free this struct via */ | |
1183 | } umac_ctx; | |
1184 | ||
1185 | /* ---------------------------------------------------------------------- */ | |
1186 | ||
1187 | #if 0 | |
1188 | int umac_reset(struct umac_ctx *ctx) | |
1189 | /* Reset the hash function to begin a new authentication. */ | |
1190 | { | |
1191 | uhash_reset(&ctx->hash); | |
1192 | return (1); | |
1193 | } | |
1194 | #endif | |
1195 | ||
1196 | /* ---------------------------------------------------------------------- */ | |
1197 | ||
1198 | int umac_delete(struct umac_ctx *ctx) | |
1199 | /* Deallocate the ctx structure */ | |
1200 | { | |
1201 | if (ctx) { | |
1202 | if (ALLOC_BOUNDARY) | |
1203 | ctx = (struct umac_ctx *)ctx->free_ptr; | |
1204 | xfree(ctx); | |
1205 | } | |
1206 | return (1); | |
1207 | } | |
1208 | ||
1209 | /* ---------------------------------------------------------------------- */ | |
1210 | ||
1211 | struct umac_ctx *umac_new(u_char key[]) | |
1212 | /* Dynamically allocate a umac_ctx struct, initialize variables, | |
1213 | * generate subkeys from key. Align to 16-byte boundary. | |
1214 | */ | |
1215 | { | |
1216 | struct umac_ctx *ctx, *octx; | |
1217 | size_t bytes_to_add; | |
1218 | aes_int_key prf_key; | |
1219 | ||
1220 | octx = ctx = xmalloc(sizeof(*ctx) + ALLOC_BOUNDARY); | |
1221 | if (ctx) { | |
1222 | if (ALLOC_BOUNDARY) { | |
1223 | bytes_to_add = ALLOC_BOUNDARY - | |
1224 | ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1)); | |
1225 | ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add); | |
1226 | } | |
1227 | ctx->free_ptr = octx; | |
1228 | aes_key_setup(key,prf_key); | |
1229 | pdf_init(&ctx->pdf, prf_key); | |
1230 | uhash_init(&ctx->hash, prf_key); | |
1231 | } | |
1232 | ||
1233 | return (ctx); | |
1234 | } | |
1235 | ||
1236 | /* ---------------------------------------------------------------------- */ | |
1237 | ||
1238 | int umac_final(struct umac_ctx *ctx, u_char tag[], u_char nonce[8]) | |
1239 | /* Incorporate any pending data, pad, and generate tag */ | |
1240 | { | |
1241 | uhash_final(&ctx->hash, (u_char *)tag); | |
1242 | pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); | |
1243 | ||
1244 | return (1); | |
1245 | } | |
1246 | ||
1247 | /* ---------------------------------------------------------------------- */ | |
1248 | ||
1249 | int umac_update(struct umac_ctx *ctx, u_char *input, long len) | |
1250 | /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */ | |
1251 | /* hash each one, calling the PDF on the hashed output whenever the hash- */ | |
1252 | /* output buffer is full. */ | |
1253 | { | |
1254 | uhash_update(&ctx->hash, input, len); | |
1255 | return (1); | |
1256 | } | |
1257 | ||
1258 | /* ---------------------------------------------------------------------- */ | |
1259 | ||
1260 | #if 0 | |
1261 | int umac(struct umac_ctx *ctx, u_char *input, | |
1262 | long len, u_char tag[], | |
1263 | u_char nonce[8]) | |
1264 | /* All-in-one version simply calls umac_update() and umac_final(). */ | |
1265 | { | |
1266 | uhash(&ctx->hash, input, len, (u_char *)tag); | |
1267 | pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); | |
1268 | ||
1269 | return (1); | |
1270 | } | |
1271 | #endif | |
1272 | ||
1273 | /* ---------------------------------------------------------------------- */ | |
1274 | /* ---------------------------------------------------------------------- */ | |
1275 | /* ----- End UMAC Section ----------------------------------------------- */ | |
1276 | /* ---------------------------------------------------------------------- */ | |
1277 | /* ---------------------------------------------------------------------- */ |