diff options
Diffstat (limited to 'umac.c')
-rw-r--r-- | umac.c | 1272 |
1 files changed, 1272 insertions, 0 deletions
@@ -0,0 +1,1272 @@ | |||
1 | /* $OpenBSD: umac.c,v 1.1 2007/06/07 19:37:34 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 "umac.h" | ||
70 | #include <string.h> | ||
71 | #include <stdlib.h> | ||
72 | #include <stddef.h> | ||
73 | |||
74 | /* ---------------------------------------------------------------------- */ | ||
75 | /* --- Primitive Data Types --- */ | ||
76 | /* ---------------------------------------------------------------------- */ | ||
77 | |||
78 | /* The following assumptions may need change on your system */ | ||
79 | typedef u_int8_t UINT8; /* 1 byte */ | ||
80 | typedef u_int16_t UINT16; /* 2 byte */ | ||
81 | typedef u_int32_t UINT32; /* 4 byte */ | ||
82 | typedef u_int64_t UINT64; /* 8 bytes */ | ||
83 | typedef unsigned int UWORD; /* Register */ | ||
84 | |||
85 | /* ---------------------------------------------------------------------- */ | ||
86 | /* --- Constants -------------------------------------------------------- */ | ||
87 | /* ---------------------------------------------------------------------- */ | ||
88 | |||
89 | #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */ | ||
90 | |||
91 | /* Message "words" are read from memory in an endian-specific manner. */ | ||
92 | /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */ | ||
93 | /* be set true if the host computer is little-endian. */ | ||
94 | |||
95 | #if BYTE_ORDER == LITTLE_ENDIAN | ||
96 | #define __LITTLE_ENDIAN__ 1 | ||
97 | #else | ||
98 | #define __LITTLE_ENDIAN__ 0 | ||
99 | #endif | ||
100 | |||
101 | /* ---------------------------------------------------------------------- */ | ||
102 | /* ---------------------------------------------------------------------- */ | ||
103 | /* ----- Architecture Specific ------------------------------------------ */ | ||
104 | /* ---------------------------------------------------------------------- */ | ||
105 | /* ---------------------------------------------------------------------- */ | ||
106 | |||
107 | |||
108 | /* ---------------------------------------------------------------------- */ | ||
109 | /* ---------------------------------------------------------------------- */ | ||
110 | /* ----- Primitive Routines --------------------------------------------- */ | ||
111 | /* ---------------------------------------------------------------------- */ | ||
112 | /* ---------------------------------------------------------------------- */ | ||
113 | |||
114 | |||
115 | /* ---------------------------------------------------------------------- */ | ||
116 | /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */ | ||
117 | /* ---------------------------------------------------------------------- */ | ||
118 | |||
119 | #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b))) | ||
120 | |||
121 | /* ---------------------------------------------------------------------- */ | ||
122 | /* --- Endian Conversion --- Forcing assembly on some platforms */ | ||
123 | /* ---------------------------------------------------------------------- */ | ||
124 | |||
125 | #if HAVE_SWAP32 | ||
126 | #define LOAD_UINT32_REVERSED(p) (swap32(*(UINT32 *)(p))) | ||
127 | #define STORE_UINT32_REVERSED(p,v) (*(UINT32 *)(p) = swap32(v)) | ||
128 | #else /* HAVE_SWAP32 */ | ||
129 | |||
130 | static UINT32 LOAD_UINT32_REVERSED(void *ptr) | ||
131 | { | ||
132 | UINT32 temp = *(UINT32 *)ptr; | ||
133 | temp = (temp >> 24) | ((temp & 0x00FF0000) >> 8 ) | ||
134 | | ((temp & 0x0000FF00) << 8 ) | (temp << 24); | ||
135 | return (UINT32)temp; | ||
136 | } | ||
137 | |||
138 | static void STORE_UINT32_REVERSED(void *ptr, UINT32 x) | ||
139 | { | ||
140 | UINT32 i = (UINT32)x; | ||
141 | *(UINT32 *)ptr = (i >> 24) | ((i & 0x00FF0000) >> 8 ) | ||
142 | | ((i & 0x0000FF00) << 8 ) | (i << 24); | ||
143 | } | ||
144 | #endif /* HAVE_SWAP32 */ | ||
145 | |||
146 | /* The following definitions use the above reversal-primitives to do the right | ||
147 | * thing on endian specific load and stores. | ||
148 | */ | ||
149 | |||
150 | #if (__LITTLE_ENDIAN__) | ||
151 | #define LOAD_UINT32_LITTLE(ptr) (*(UINT32 *)(ptr)) | ||
152 | #define STORE_UINT32_BIG(ptr,x) STORE_UINT32_REVERSED(ptr,x) | ||
153 | #else | ||
154 | #define LOAD_UINT32_LITTLE(ptr) LOAD_UINT32_REVERSED(ptr) | ||
155 | #define STORE_UINT32_BIG(ptr,x) (*(UINT32 *)(ptr) = (UINT32)(x)) | ||
156 | #endif | ||
157 | |||
158 | /* ---------------------------------------------------------------------- */ | ||
159 | /* ---------------------------------------------------------------------- */ | ||
160 | /* ----- Begin KDF & PDF Section ---------------------------------------- */ | ||
161 | /* ---------------------------------------------------------------------- */ | ||
162 | /* ---------------------------------------------------------------------- */ | ||
163 | |||
164 | /* UMAC uses AES with 16 byte block and key lengths */ | ||
165 | #define AES_BLOCK_LEN 16 | ||
166 | |||
167 | /* OpenSSL's AES */ | ||
168 | #include "openbsd-compat/openssl-compat.h" | ||
169 | #ifndef USE_BUILTIN_RIJNDAEL | ||
170 | # include <openssl/aes.h> | ||
171 | #endif | ||
172 | typedef AES_KEY aes_int_key[1]; | ||
173 | #define aes_encryption(in,out,int_key) \ | ||
174 | AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key) | ||
175 | #define aes_key_setup(key,int_key) \ | ||
176 | AES_set_encrypt_key((u_char *)(key),UMAC_KEY_LEN*8,int_key) | ||
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 | } | ||
207 | |||
208 | /* The final UHASH result is XOR'd with the output of a pseudorandom | ||
209 | * function. Here, we use AES to generate random output and | ||
210 | * xor the appropriate bytes depending on the last bits of nonce. | ||
211 | * This scheme is optimized for sequential, increasing big-endian nonces. | ||
212 | */ | ||
213 | |||
214 | typedef struct { | ||
215 | UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */ | ||
216 | UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */ | ||
217 | aes_int_key prf_key; /* Expanded AES key for PDF */ | ||
218 | } pdf_ctx; | ||
219 | |||
220 | static void pdf_init(pdf_ctx *pc, aes_int_key prf_key) | ||
221 | { | ||
222 | UINT8 buf[UMAC_KEY_LEN]; | ||
223 | |||
224 | kdf(buf, prf_key, 0, UMAC_KEY_LEN); | ||
225 | aes_key_setup(buf, pc->prf_key); | ||
226 | |||
227 | /* Initialize pdf and cache */ | ||
228 | memset(pc->nonce, 0, sizeof(pc->nonce)); | ||
229 | aes_encryption(pc->nonce, pc->cache, pc->prf_key); | ||
230 | } | ||
231 | |||
232 | static void pdf_gen_xor(pdf_ctx *pc, UINT8 nonce[8], UINT8 buf[8]) | ||
233 | { | ||
234 | /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes | ||
235 | * of the AES output. If last time around we returned the ndx-1st | ||
236 | * element, then we may have the result in the cache already. | ||
237 | */ | ||
238 | |||
239 | #if (UMAC_OUTPUT_LEN == 4) | ||
240 | #define LOW_BIT_MASK 3 | ||
241 | #elif (UMAC_OUTPUT_LEN == 8) | ||
242 | #define LOW_BIT_MASK 1 | ||
243 | #elif (UMAC_OUTPUT_LEN > 8) | ||
244 | #define LOW_BIT_MASK 0 | ||
245 | #endif | ||
246 | |||
247 | UINT8 tmp_nonce_lo[4]; | ||
248 | #if LOW_BIT_MASK != 0 | ||
249 | int ndx = nonce[7] & LOW_BIT_MASK; | ||
250 | #endif | ||
251 | *(UINT32 *)tmp_nonce_lo = ((UINT32 *)nonce)[1]; | ||
252 | tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */ | ||
253 | |||
254 | if ( (((UINT32 *)tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) || | ||
255 | (((UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) ) | ||
256 | { | ||
257 | ((UINT32 *)pc->nonce)[0] = ((UINT32 *)nonce)[0]; | ||
258 | ((UINT32 *)pc->nonce)[1] = ((UINT32 *)tmp_nonce_lo)[0]; | ||
259 | aes_encryption(pc->nonce, pc->cache, pc->prf_key); | ||
260 | } | ||
261 | |||
262 | #if (UMAC_OUTPUT_LEN == 4) | ||
263 | *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx]; | ||
264 | #elif (UMAC_OUTPUT_LEN == 8) | ||
265 | *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx]; | ||
266 | #elif (UMAC_OUTPUT_LEN == 12) | ||
267 | ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; | ||
268 | ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2]; | ||
269 | #elif (UMAC_OUTPUT_LEN == 16) | ||
270 | ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0]; | ||
271 | ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1]; | ||
272 | #endif | ||
273 | } | ||
274 | |||
275 | /* ---------------------------------------------------------------------- */ | ||
276 | /* ---------------------------------------------------------------------- */ | ||
277 | /* ----- Begin NH Hash Section ------------------------------------------ */ | ||
278 | /* ---------------------------------------------------------------------- */ | ||
279 | /* ---------------------------------------------------------------------- */ | ||
280 | |||
281 | /* The NH-based hash functions used in UMAC are described in the UMAC paper | ||
282 | * and specification, both of which can be found at the UMAC website. | ||
283 | * The interface to this implementation has two | ||
284 | * versions, one expects the entire message being hashed to be passed | ||
285 | * in a single buffer and returns the hash result immediately. The second | ||
286 | * allows the message to be passed in a sequence of buffers. In the | ||
287 | * muliple-buffer interface, the client calls the routine nh_update() as | ||
288 | * many times as necessary. When there is no more data to be fed to the | ||
289 | * hash, the client calls nh_final() which calculates the hash output. | ||
290 | * Before beginning another hash calculation the nh_reset() routine | ||
291 | * must be called. The single-buffer routine, nh(), is equivalent to | ||
292 | * the sequence of calls nh_update() and nh_final(); however it is | ||
293 | * optimized and should be prefered whenever the multiple-buffer interface | ||
294 | * is not necessary. When using either interface, it is the client's | ||
295 | * responsability to pass no more than L1_KEY_LEN bytes per hash result. | ||
296 | * | ||
297 | * The routine nh_init() initializes the nh_ctx data structure and | ||
298 | * must be called once, before any other PDF routine. | ||
299 | */ | ||
300 | |||
301 | /* The "nh_aux" routines do the actual NH hashing work. They | ||
302 | * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines | ||
303 | * produce output for all STREAMS NH iterations in one call, | ||
304 | * allowing the parallel implementation of the streams. | ||
305 | */ | ||
306 | |||
307 | #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */ | ||
308 | #define L1_KEY_LEN 1024 /* Internal key bytes */ | ||
309 | #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */ | ||
310 | #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */ | ||
311 | #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */ | ||
312 | #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */ | ||
313 | |||
314 | typedef struct { | ||
315 | UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */ | ||
316 | UINT8 data [HASH_BUF_BYTES]; /* Incomming data buffer */ | ||
317 | int next_data_empty; /* Bookeeping variable for data buffer. */ | ||
318 | int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */ | ||
319 | UINT64 state[STREAMS]; /* on-line state */ | ||
320 | } nh_ctx; | ||
321 | |||
322 | |||
323 | #if (UMAC_OUTPUT_LEN == 4) | ||
324 | |||
325 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | ||
326 | /* NH hashing primitive. Previous (partial) hash result is loaded and | ||
327 | * then stored via hp pointer. The length of the data pointed at by "dp", | ||
328 | * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key | ||
329 | * is expected to be endian compensated in memory at key setup. | ||
330 | */ | ||
331 | { | ||
332 | UINT64 h; | ||
333 | UWORD c = dlen / 32; | ||
334 | UINT32 *k = (UINT32 *)kp; | ||
335 | UINT32 *d = (UINT32 *)dp; | ||
336 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | ||
337 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7; | ||
338 | |||
339 | h = *((UINT64 *)hp); | ||
340 | do { | ||
341 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | ||
342 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | ||
343 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | ||
344 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | ||
345 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | ||
346 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | ||
347 | h += MUL64((k0 + d0), (k4 + d4)); | ||
348 | h += MUL64((k1 + d1), (k5 + d5)); | ||
349 | h += MUL64((k2 + d2), (k6 + d6)); | ||
350 | h += MUL64((k3 + d3), (k7 + d7)); | ||
351 | |||
352 | d += 8; | ||
353 | k += 8; | ||
354 | } while (--c); | ||
355 | *((UINT64 *)hp) = h; | ||
356 | } | ||
357 | |||
358 | #elif (UMAC_OUTPUT_LEN == 8) | ||
359 | |||
360 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | ||
361 | /* Same as previous nh_aux, but two streams are handled in one pass, | ||
362 | * reading and writing 16 bytes of hash-state per call. | ||
363 | */ | ||
364 | { | ||
365 | UINT64 h1,h2; | ||
366 | UWORD c = dlen / 32; | ||
367 | UINT32 *k = (UINT32 *)kp; | ||
368 | UINT32 *d = (UINT32 *)dp; | ||
369 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | ||
370 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7, | ||
371 | k8,k9,k10,k11; | ||
372 | |||
373 | h1 = *((UINT64 *)hp); | ||
374 | h2 = *((UINT64 *)hp + 1); | ||
375 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | ||
376 | do { | ||
377 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | ||
378 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | ||
379 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | ||
380 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | ||
381 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | ||
382 | k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); | ||
383 | |||
384 | h1 += MUL64((k0 + d0), (k4 + d4)); | ||
385 | h2 += MUL64((k4 + d0), (k8 + d4)); | ||
386 | |||
387 | h1 += MUL64((k1 + d1), (k5 + d5)); | ||
388 | h2 += MUL64((k5 + d1), (k9 + d5)); | ||
389 | |||
390 | h1 += MUL64((k2 + d2), (k6 + d6)); | ||
391 | h2 += MUL64((k6 + d2), (k10 + d6)); | ||
392 | |||
393 | h1 += MUL64((k3 + d3), (k7 + d7)); | ||
394 | h2 += MUL64((k7 + d3), (k11 + d7)); | ||
395 | |||
396 | k0 = k8; k1 = k9; k2 = k10; k3 = k11; | ||
397 | |||
398 | d += 8; | ||
399 | k += 8; | ||
400 | } while (--c); | ||
401 | ((UINT64 *)hp)[0] = h1; | ||
402 | ((UINT64 *)hp)[1] = h2; | ||
403 | } | ||
404 | |||
405 | #elif (UMAC_OUTPUT_LEN == 12) | ||
406 | |||
407 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | ||
408 | /* Same as previous nh_aux, but two streams are handled in one pass, | ||
409 | * reading and writing 24 bytes of hash-state per call. | ||
410 | */ | ||
411 | { | ||
412 | UINT64 h1,h2,h3; | ||
413 | UWORD c = dlen / 32; | ||
414 | UINT32 *k = (UINT32 *)kp; | ||
415 | UINT32 *d = (UINT32 *)dp; | ||
416 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | ||
417 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7, | ||
418 | k8,k9,k10,k11,k12,k13,k14,k15; | ||
419 | |||
420 | h1 = *((UINT64 *)hp); | ||
421 | h2 = *((UINT64 *)hp + 1); | ||
422 | h3 = *((UINT64 *)hp + 2); | ||
423 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | ||
424 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | ||
425 | do { | ||
426 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | ||
427 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | ||
428 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | ||
429 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | ||
430 | k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); | ||
431 | k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); | ||
432 | |||
433 | h1 += MUL64((k0 + d0), (k4 + d4)); | ||
434 | h2 += MUL64((k4 + d0), (k8 + d4)); | ||
435 | h3 += MUL64((k8 + d0), (k12 + d4)); | ||
436 | |||
437 | h1 += MUL64((k1 + d1), (k5 + d5)); | ||
438 | h2 += MUL64((k5 + d1), (k9 + d5)); | ||
439 | h3 += MUL64((k9 + d1), (k13 + d5)); | ||
440 | |||
441 | h1 += MUL64((k2 + d2), (k6 + d6)); | ||
442 | h2 += MUL64((k6 + d2), (k10 + d6)); | ||
443 | h3 += MUL64((k10 + d2), (k14 + d6)); | ||
444 | |||
445 | h1 += MUL64((k3 + d3), (k7 + d7)); | ||
446 | h2 += MUL64((k7 + d3), (k11 + d7)); | ||
447 | h3 += MUL64((k11 + d3), (k15 + d7)); | ||
448 | |||
449 | k0 = k8; k1 = k9; k2 = k10; k3 = k11; | ||
450 | k4 = k12; k5 = k13; k6 = k14; k7 = k15; | ||
451 | |||
452 | d += 8; | ||
453 | k += 8; | ||
454 | } while (--c); | ||
455 | ((UINT64 *)hp)[0] = h1; | ||
456 | ((UINT64 *)hp)[1] = h2; | ||
457 | ((UINT64 *)hp)[2] = h3; | ||
458 | } | ||
459 | |||
460 | #elif (UMAC_OUTPUT_LEN == 16) | ||
461 | |||
462 | static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen) | ||
463 | /* Same as previous nh_aux, but two streams are handled in one pass, | ||
464 | * reading and writing 24 bytes of hash-state per call. | ||
465 | */ | ||
466 | { | ||
467 | UINT64 h1,h2,h3,h4; | ||
468 | UWORD c = dlen / 32; | ||
469 | UINT32 *k = (UINT32 *)kp; | ||
470 | UINT32 *d = (UINT32 *)dp; | ||
471 | UINT32 d0,d1,d2,d3,d4,d5,d6,d7; | ||
472 | UINT32 k0,k1,k2,k3,k4,k5,k6,k7, | ||
473 | k8,k9,k10,k11,k12,k13,k14,k15, | ||
474 | k16,k17,k18,k19; | ||
475 | |||
476 | h1 = *((UINT64 *)hp); | ||
477 | h2 = *((UINT64 *)hp + 1); | ||
478 | h3 = *((UINT64 *)hp + 2); | ||
479 | h4 = *((UINT64 *)hp + 3); | ||
480 | k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3); | ||
481 | k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7); | ||
482 | do { | ||
483 | d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1); | ||
484 | d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3); | ||
485 | d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5); | ||
486 | d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7); | ||
487 | k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11); | ||
488 | k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15); | ||
489 | k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19); | ||
490 | |||
491 | h1 += MUL64((k0 + d0), (k4 + d4)); | ||
492 | h2 += MUL64((k4 + d0), (k8 + d4)); | ||
493 | h3 += MUL64((k8 + d0), (k12 + d4)); | ||
494 | h4 += MUL64((k12 + d0), (k16 + d4)); | ||
495 | |||
496 | h1 += MUL64((k1 + d1), (k5 + d5)); | ||
497 | h2 += MUL64((k5 + d1), (k9 + d5)); | ||
498 | h3 += MUL64((k9 + d1), (k13 + d5)); | ||
499 | h4 += MUL64((k13 + d1), (k17 + d5)); | ||
500 | |||
501 | h1 += MUL64((k2 + d2), (k6 + d6)); | ||
502 | h2 += MUL64((k6 + d2), (k10 + d6)); | ||
503 | h3 += MUL64((k10 + d2), (k14 + d6)); | ||
504 | h4 += MUL64((k14 + d2), (k18 + d6)); | ||
505 | |||
506 | h1 += MUL64((k3 + d3), (k7 + d7)); | ||
507 | h2 += MUL64((k7 + d3), (k11 + d7)); | ||
508 | h3 += MUL64((k11 + d3), (k15 + d7)); | ||
509 | h4 += MUL64((k15 + d3), (k19 + d7)); | ||
510 | |||
511 | k0 = k8; k1 = k9; k2 = k10; k3 = k11; | ||
512 | k4 = k12; k5 = k13; k6 = k14; k7 = k15; | ||
513 | k8 = k16; k9 = k17; k10 = k18; k11 = k19; | ||
514 | |||
515 | d += 8; | ||
516 | k += 8; | ||
517 | } while (--c); | ||
518 | ((UINT64 *)hp)[0] = h1; | ||
519 | ((UINT64 *)hp)[1] = h2; | ||
520 | ((UINT64 *)hp)[2] = h3; | ||
521 | ((UINT64 *)hp)[3] = h4; | ||
522 | } | ||
523 | |||
524 | /* ---------------------------------------------------------------------- */ | ||
525 | #endif /* UMAC_OUTPUT_LENGTH */ | ||
526 | /* ---------------------------------------------------------------------- */ | ||
527 | |||
528 | |||
529 | /* ---------------------------------------------------------------------- */ | ||
530 | |||
531 | static void nh_transform(nh_ctx *hc, UINT8 *buf, UINT32 nbytes) | ||
532 | /* This function is a wrapper for the primitive NH hash functions. It takes | ||
533 | * as argument "hc" the current hash context and a buffer which must be a | ||
534 | * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset | ||
535 | * appropriately according to how much message has been hashed already. | ||
536 | */ | ||
537 | { | ||
538 | UINT8 *key; | ||
539 | |||
540 | key = hc->nh_key + hc->bytes_hashed; | ||
541 | nh_aux(key, buf, hc->state, nbytes); | ||
542 | } | ||
543 | |||
544 | /* ---------------------------------------------------------------------- */ | ||
545 | |||
546 | static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes) | ||
547 | /* We endian convert the keys on little-endian computers to */ | ||
548 | /* compensate for the lack of big-endian memory reads during hashing. */ | ||
549 | { | ||
550 | UWORD iters = num_bytes / bpw; | ||
551 | if (bpw == 4) { | ||
552 | UINT32 *p = (UINT32 *)buf; | ||
553 | do { | ||
554 | *p = LOAD_UINT32_REVERSED(p); | ||
555 | p++; | ||
556 | } while (--iters); | ||
557 | } else if (bpw == 8) { | ||
558 | UINT32 *p = (UINT32 *)buf; | ||
559 | UINT32 t; | ||
560 | do { | ||
561 | t = LOAD_UINT32_REVERSED(p+1); | ||
562 | p[1] = LOAD_UINT32_REVERSED(p); | ||
563 | p[0] = t; | ||
564 | p += 2; | ||
565 | } while (--iters); | ||
566 | } | ||
567 | } | ||
568 | #if (__LITTLE_ENDIAN__) | ||
569 | #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z)) | ||
570 | #else | ||
571 | #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */ | ||
572 | #endif | ||
573 | |||
574 | /* ---------------------------------------------------------------------- */ | ||
575 | |||
576 | static void nh_reset(nh_ctx *hc) | ||
577 | /* Reset nh_ctx to ready for hashing of new data */ | ||
578 | { | ||
579 | hc->bytes_hashed = 0; | ||
580 | hc->next_data_empty = 0; | ||
581 | hc->state[0] = 0; | ||
582 | #if (UMAC_OUTPUT_LEN >= 8) | ||
583 | hc->state[1] = 0; | ||
584 | #endif | ||
585 | #if (UMAC_OUTPUT_LEN >= 12) | ||
586 | hc->state[2] = 0; | ||
587 | #endif | ||
588 | #if (UMAC_OUTPUT_LEN == 16) | ||
589 | hc->state[3] = 0; | ||
590 | #endif | ||
591 | |||
592 | } | ||
593 | |||
594 | /* ---------------------------------------------------------------------- */ | ||
595 | |||
596 | static void nh_init(nh_ctx *hc, aes_int_key prf_key) | ||
597 | /* Generate nh_key, endian convert and reset to be ready for hashing. */ | ||
598 | { | ||
599 | kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key)); | ||
600 | endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key)); | ||
601 | nh_reset(hc); | ||
602 | } | ||
603 | |||
604 | /* ---------------------------------------------------------------------- */ | ||
605 | |||
606 | static void nh_update(nh_ctx *hc, UINT8 *buf, UINT32 nbytes) | ||
607 | /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */ | ||
608 | /* even multiple of HASH_BUF_BYTES. */ | ||
609 | { | ||
610 | UINT32 i,j; | ||
611 | |||
612 | j = hc->next_data_empty; | ||
613 | if ((j + nbytes) >= HASH_BUF_BYTES) { | ||
614 | if (j) { | ||
615 | i = HASH_BUF_BYTES - j; | ||
616 | memcpy(hc->data+j, buf, i); | ||
617 | nh_transform(hc,hc->data,HASH_BUF_BYTES); | ||
618 | nbytes -= i; | ||
619 | buf += i; | ||
620 | hc->bytes_hashed += HASH_BUF_BYTES; | ||
621 | } | ||
622 | if (nbytes >= HASH_BUF_BYTES) { | ||
623 | i = nbytes & ~(HASH_BUF_BYTES - 1); | ||
624 | nh_transform(hc, buf, i); | ||
625 | nbytes -= i; | ||
626 | buf += i; | ||
627 | hc->bytes_hashed += i; | ||
628 | } | ||
629 | j = 0; | ||
630 | } | ||
631 | memcpy(hc->data + j, buf, nbytes); | ||
632 | hc->next_data_empty = j + nbytes; | ||
633 | } | ||
634 | |||
635 | /* ---------------------------------------------------------------------- */ | ||
636 | |||
637 | static void zero_pad(UINT8 *p, int nbytes) | ||
638 | { | ||
639 | /* Write "nbytes" of zeroes, beginning at "p" */ | ||
640 | if (nbytes >= (int)sizeof(UWORD)) { | ||
641 | while ((ptrdiff_t)p % sizeof(UWORD)) { | ||
642 | *p = 0; | ||
643 | nbytes--; | ||
644 | p++; | ||
645 | } | ||
646 | while (nbytes >= (int)sizeof(UWORD)) { | ||
647 | *(UWORD *)p = 0; | ||
648 | nbytes -= sizeof(UWORD); | ||
649 | p += sizeof(UWORD); | ||
650 | } | ||
651 | } | ||
652 | while (nbytes) { | ||
653 | *p = 0; | ||
654 | nbytes--; | ||
655 | p++; | ||
656 | } | ||
657 | } | ||
658 | |||
659 | /* ---------------------------------------------------------------------- */ | ||
660 | |||
661 | static void nh_final(nh_ctx *hc, UINT8 *result) | ||
662 | /* After passing some number of data buffers to nh_update() for integration | ||
663 | * into an NH context, nh_final is called to produce a hash result. If any | ||
664 | * bytes are in the buffer hc->data, incorporate them into the | ||
665 | * NH context. Finally, add into the NH accumulation "state" the total number | ||
666 | * of bits hashed. The resulting numbers are written to the buffer "result". | ||
667 | * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated. | ||
668 | */ | ||
669 | { | ||
670 | int nh_len, nbits; | ||
671 | |||
672 | if (hc->next_data_empty != 0) { | ||
673 | nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) & | ||
674 | ~(L1_PAD_BOUNDARY - 1)); | ||
675 | zero_pad(hc->data + hc->next_data_empty, | ||
676 | nh_len - hc->next_data_empty); | ||
677 | nh_transform(hc, hc->data, nh_len); | ||
678 | hc->bytes_hashed += hc->next_data_empty; | ||
679 | } else if (hc->bytes_hashed == 0) { | ||
680 | nh_len = L1_PAD_BOUNDARY; | ||
681 | zero_pad(hc->data, L1_PAD_BOUNDARY); | ||
682 | nh_transform(hc, hc->data, nh_len); | ||
683 | } | ||
684 | |||
685 | nbits = (hc->bytes_hashed << 3); | ||
686 | ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits; | ||
687 | #if (UMAC_OUTPUT_LEN >= 8) | ||
688 | ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits; | ||
689 | #endif | ||
690 | #if (UMAC_OUTPUT_LEN >= 12) | ||
691 | ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits; | ||
692 | #endif | ||
693 | #if (UMAC_OUTPUT_LEN == 16) | ||
694 | ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits; | ||
695 | #endif | ||
696 | nh_reset(hc); | ||
697 | } | ||
698 | |||
699 | /* ---------------------------------------------------------------------- */ | ||
700 | |||
701 | static void nh(nh_ctx *hc, UINT8 *buf, UINT32 padded_len, | ||
702 | UINT32 unpadded_len, UINT8 *result) | ||
703 | /* All-in-one nh_update() and nh_final() equivalent. | ||
704 | * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is | ||
705 | * well aligned | ||
706 | */ | ||
707 | { | ||
708 | UINT32 nbits; | ||
709 | |||
710 | /* Initialize the hash state */ | ||
711 | nbits = (unpadded_len << 3); | ||
712 | |||
713 | ((UINT64 *)result)[0] = nbits; | ||
714 | #if (UMAC_OUTPUT_LEN >= 8) | ||
715 | ((UINT64 *)result)[1] = nbits; | ||
716 | #endif | ||
717 | #if (UMAC_OUTPUT_LEN >= 12) | ||
718 | ((UINT64 *)result)[2] = nbits; | ||
719 | #endif | ||
720 | #if (UMAC_OUTPUT_LEN == 16) | ||
721 | ((UINT64 *)result)[3] = nbits; | ||
722 | #endif | ||
723 | |||
724 | nh_aux(hc->nh_key, buf, result, padded_len); | ||
725 | } | ||
726 | |||
727 | /* ---------------------------------------------------------------------- */ | ||
728 | /* ---------------------------------------------------------------------- */ | ||
729 | /* ----- Begin UHASH Section -------------------------------------------- */ | ||
730 | /* ---------------------------------------------------------------------- */ | ||
731 | /* ---------------------------------------------------------------------- */ | ||
732 | |||
733 | /* UHASH is a multi-layered algorithm. Data presented to UHASH is first | ||
734 | * hashed by NH. The NH output is then hashed by a polynomial-hash layer | ||
735 | * unless the initial data to be hashed is short. After the polynomial- | ||
736 | * layer, an inner-product hash is used to produce the final UHASH output. | ||
737 | * | ||
738 | * UHASH provides two interfaces, one all-at-once and another where data | ||
739 | * buffers are presented sequentially. In the sequential interface, the | ||
740 | * UHASH client calls the routine uhash_update() as many times as necessary. | ||
741 | * When there is no more data to be fed to UHASH, the client calls | ||
742 | * uhash_final() which | ||
743 | * calculates the UHASH output. Before beginning another UHASH calculation | ||
744 | * the uhash_reset() routine must be called. The all-at-once UHASH routine, | ||
745 | * uhash(), is equivalent to the sequence of calls uhash_update() and | ||
746 | * uhash_final(); however it is optimized and should be | ||
747 | * used whenever the sequential interface is not necessary. | ||
748 | * | ||
749 | * The routine uhash_init() initializes the uhash_ctx data structure and | ||
750 | * must be called once, before any other UHASH routine. | ||
751 | */ | ||
752 | |||
753 | /* ---------------------------------------------------------------------- */ | ||
754 | /* ----- Constants and uhash_ctx ---------------------------------------- */ | ||
755 | /* ---------------------------------------------------------------------- */ | ||
756 | |||
757 | /* ---------------------------------------------------------------------- */ | ||
758 | /* ----- Poly hash and Inner-Product hash Constants --------------------- */ | ||
759 | /* ---------------------------------------------------------------------- */ | ||
760 | |||
761 | /* Primes and masks */ | ||
762 | #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */ | ||
763 | #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */ | ||
764 | #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */ | ||
765 | |||
766 | |||
767 | /* ---------------------------------------------------------------------- */ | ||
768 | |||
769 | typedef struct uhash_ctx { | ||
770 | nh_ctx hash; /* Hash context for L1 NH hash */ | ||
771 | UINT64 poly_key_8[STREAMS]; /* p64 poly keys */ | ||
772 | UINT64 poly_accum[STREAMS]; /* poly hash result */ | ||
773 | UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */ | ||
774 | UINT32 ip_trans[STREAMS]; /* Inner-product translation */ | ||
775 | UINT32 msg_len; /* Total length of data passed */ | ||
776 | /* to uhash */ | ||
777 | } uhash_ctx; | ||
778 | typedef struct uhash_ctx *uhash_ctx_t; | ||
779 | |||
780 | /* ---------------------------------------------------------------------- */ | ||
781 | |||
782 | |||
783 | /* The polynomial hashes use Horner's rule to evaluate a polynomial one | ||
784 | * word at a time. As described in the specification, poly32 and poly64 | ||
785 | * require keys from special domains. The following implementations exploit | ||
786 | * the special domains to avoid overflow. The results are not guaranteed to | ||
787 | * be within Z_p32 and Z_p64, but the Inner-Product hash implementation | ||
788 | * patches any errant values. | ||
789 | */ | ||
790 | |||
791 | static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data) | ||
792 | { | ||
793 | UINT32 key_hi = (UINT32)(key >> 32), | ||
794 | key_lo = (UINT32)key, | ||
795 | cur_hi = (UINT32)(cur >> 32), | ||
796 | cur_lo = (UINT32)cur, | ||
797 | x_lo, | ||
798 | x_hi; | ||
799 | UINT64 X,T,res; | ||
800 | |||
801 | X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo); | ||
802 | x_lo = (UINT32)X; | ||
803 | x_hi = (UINT32)(X >> 32); | ||
804 | |||
805 | res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo); | ||
806 | |||
807 | T = ((UINT64)x_lo << 32); | ||
808 | res += T; | ||
809 | if (res < T) | ||
810 | res += 59; | ||
811 | |||
812 | res += data; | ||
813 | if (res < data) | ||
814 | res += 59; | ||
815 | |||
816 | return res; | ||
817 | } | ||
818 | |||
819 | |||
820 | /* Although UMAC is specified to use a ramped polynomial hash scheme, this | ||
821 | * implementation does not handle all ramp levels. Because we don't handle | ||
822 | * the ramp up to p128 modulus in this implementation, we are limited to | ||
823 | * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24 | ||
824 | * bytes input to UMAC per tag, ie. 16MB). | ||
825 | */ | ||
826 | static void poly_hash(uhash_ctx_t hc, UINT32 data_in[]) | ||
827 | { | ||
828 | int i; | ||
829 | UINT64 *data=(UINT64*)data_in; | ||
830 | |||
831 | for (i = 0; i < STREAMS; i++) { | ||
832 | if ((UINT32)(data[i] >> 32) == 0xfffffffful) { | ||
833 | hc->poly_accum[i] = poly64(hc->poly_accum[i], | ||
834 | hc->poly_key_8[i], p64 - 1); | ||
835 | hc->poly_accum[i] = poly64(hc->poly_accum[i], | ||
836 | hc->poly_key_8[i], (data[i] - 59)); | ||
837 | } else { | ||
838 | hc->poly_accum[i] = poly64(hc->poly_accum[i], | ||
839 | hc->poly_key_8[i], data[i]); | ||
840 | } | ||
841 | } | ||
842 | } | ||
843 | |||
844 | |||
845 | /* ---------------------------------------------------------------------- */ | ||
846 | |||
847 | |||
848 | /* The final step in UHASH is an inner-product hash. The poly hash | ||
849 | * produces a result not neccesarily WORD_LEN bytes long. The inner- | ||
850 | * product hash breaks the polyhash output into 16-bit chunks and | ||
851 | * multiplies each with a 36 bit key. | ||
852 | */ | ||
853 | |||
854 | static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data) | ||
855 | { | ||
856 | t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48); | ||
857 | t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32); | ||
858 | t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16); | ||
859 | t = t + ipkp[3] * (UINT64)(UINT16)(data); | ||
860 | |||
861 | return t; | ||
862 | } | ||
863 | |||
864 | static UINT32 ip_reduce_p36(UINT64 t) | ||
865 | { | ||
866 | /* Divisionless modular reduction */ | ||
867 | UINT64 ret; | ||
868 | |||
869 | ret = (t & m36) + 5 * (t >> 36); | ||
870 | if (ret >= p36) | ||
871 | ret -= p36; | ||
872 | |||
873 | /* return least significant 32 bits */ | ||
874 | return (UINT32)(ret); | ||
875 | } | ||
876 | |||
877 | |||
878 | /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then | ||
879 | * the polyhash stage is skipped and ip_short is applied directly to the | ||
880 | * NH output. | ||
881 | */ | ||
882 | static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res) | ||
883 | { | ||
884 | UINT64 t; | ||
885 | UINT64 *nhp = (UINT64 *)nh_res; | ||
886 | |||
887 | t = ip_aux(0,ahc->ip_keys, nhp[0]); | ||
888 | STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]); | ||
889 | #if (UMAC_OUTPUT_LEN >= 8) | ||
890 | t = ip_aux(0,ahc->ip_keys+4, nhp[1]); | ||
891 | STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]); | ||
892 | #endif | ||
893 | #if (UMAC_OUTPUT_LEN >= 12) | ||
894 | t = ip_aux(0,ahc->ip_keys+8, nhp[2]); | ||
895 | STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]); | ||
896 | #endif | ||
897 | #if (UMAC_OUTPUT_LEN == 16) | ||
898 | t = ip_aux(0,ahc->ip_keys+12, nhp[3]); | ||
899 | STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]); | ||
900 | #endif | ||
901 | } | ||
902 | |||
903 | /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then | ||
904 | * the polyhash stage is not skipped and ip_long is applied to the | ||
905 | * polyhash output. | ||
906 | */ | ||
907 | static void ip_long(uhash_ctx_t ahc, u_char *res) | ||
908 | { | ||
909 | int i; | ||
910 | UINT64 t; | ||
911 | |||
912 | for (i = 0; i < STREAMS; i++) { | ||
913 | /* fix polyhash output not in Z_p64 */ | ||
914 | if (ahc->poly_accum[i] >= p64) | ||
915 | ahc->poly_accum[i] -= p64; | ||
916 | t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]); | ||
917 | STORE_UINT32_BIG((UINT32 *)res+i, | ||
918 | ip_reduce_p36(t) ^ ahc->ip_trans[i]); | ||
919 | } | ||
920 | } | ||
921 | |||
922 | |||
923 | /* ---------------------------------------------------------------------- */ | ||
924 | |||
925 | /* ---------------------------------------------------------------------- */ | ||
926 | |||
927 | /* Reset uhash context for next hash session */ | ||
928 | static int uhash_reset(uhash_ctx_t pc) | ||
929 | { | ||
930 | nh_reset(&pc->hash); | ||
931 | pc->msg_len = 0; | ||
932 | pc->poly_accum[0] = 1; | ||
933 | #if (UMAC_OUTPUT_LEN >= 8) | ||
934 | pc->poly_accum[1] = 1; | ||
935 | #endif | ||
936 | #if (UMAC_OUTPUT_LEN >= 12) | ||
937 | pc->poly_accum[2] = 1; | ||
938 | #endif | ||
939 | #if (UMAC_OUTPUT_LEN == 16) | ||
940 | pc->poly_accum[3] = 1; | ||
941 | #endif | ||
942 | return 1; | ||
943 | } | ||
944 | |||
945 | /* ---------------------------------------------------------------------- */ | ||
946 | |||
947 | /* Given a pointer to the internal key needed by kdf() and a uhash context, | ||
948 | * initialize the NH context and generate keys needed for poly and inner- | ||
949 | * product hashing. All keys are endian adjusted in memory so that native | ||
950 | * loads cause correct keys to be in registers during calculation. | ||
951 | */ | ||
952 | static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key) | ||
953 | { | ||
954 | int i; | ||
955 | UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)]; | ||
956 | |||
957 | /* Zero the entire uhash context */ | ||
958 | memset(ahc, 0, sizeof(uhash_ctx)); | ||
959 | |||
960 | /* Initialize the L1 hash */ | ||
961 | nh_init(&ahc->hash, prf_key); | ||
962 | |||
963 | /* Setup L2 hash variables */ | ||
964 | kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */ | ||
965 | for (i = 0; i < STREAMS; i++) { | ||
966 | /* Fill keys from the buffer, skipping bytes in the buffer not | ||
967 | * used by this implementation. Endian reverse the keys if on a | ||
968 | * little-endian computer. | ||
969 | */ | ||
970 | memcpy(ahc->poly_key_8+i, buf+24*i, 8); | ||
971 | endian_convert_if_le(ahc->poly_key_8+i, 8, 8); | ||
972 | /* Mask the 64-bit keys to their special domain */ | ||
973 | ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu; | ||
974 | ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */ | ||
975 | } | ||
976 | |||
977 | /* Setup L3-1 hash variables */ | ||
978 | kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */ | ||
979 | for (i = 0; i < STREAMS; i++) | ||
980 | memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64), | ||
981 | 4*sizeof(UINT64)); | ||
982 | endian_convert_if_le(ahc->ip_keys, sizeof(UINT64), | ||
983 | sizeof(ahc->ip_keys)); | ||
984 | for (i = 0; i < STREAMS*4; i++) | ||
985 | ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */ | ||
986 | |||
987 | /* Setup L3-2 hash variables */ | ||
988 | /* Fill buffer with index 4 key */ | ||
989 | kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32)); | ||
990 | endian_convert_if_le(ahc->ip_trans, sizeof(UINT32), | ||
991 | STREAMS * sizeof(UINT32)); | ||
992 | } | ||
993 | |||
994 | /* ---------------------------------------------------------------------- */ | ||
995 | |||
996 | #if 0 | ||
997 | static uhash_ctx_t uhash_alloc(u_char key[]) | ||
998 | { | ||
999 | /* Allocate memory and force to a 16-byte boundary. */ | ||
1000 | uhash_ctx_t ctx; | ||
1001 | u_char bytes_to_add; | ||
1002 | aes_int_key prf_key; | ||
1003 | |||
1004 | ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY); | ||
1005 | if (ctx) { | ||
1006 | if (ALLOC_BOUNDARY) { | ||
1007 | bytes_to_add = ALLOC_BOUNDARY - | ||
1008 | ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1)); | ||
1009 | ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add); | ||
1010 | *((u_char *)ctx - 1) = bytes_to_add; | ||
1011 | } | ||
1012 | aes_key_setup(key,prf_key); | ||
1013 | uhash_init(ctx, prf_key); | ||
1014 | } | ||
1015 | return (ctx); | ||
1016 | } | ||
1017 | #endif | ||
1018 | |||
1019 | /* ---------------------------------------------------------------------- */ | ||
1020 | |||
1021 | #if 0 | ||
1022 | static int uhash_free(uhash_ctx_t ctx) | ||
1023 | { | ||
1024 | /* Free memory allocated by uhash_alloc */ | ||
1025 | u_char bytes_to_sub; | ||
1026 | |||
1027 | if (ctx) { | ||
1028 | if (ALLOC_BOUNDARY) { | ||
1029 | bytes_to_sub = *((u_char *)ctx - 1); | ||
1030 | ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub); | ||
1031 | } | ||
1032 | free(ctx); | ||
1033 | } | ||
1034 | return (1); | ||
1035 | } | ||
1036 | #endif | ||
1037 | /* ---------------------------------------------------------------------- */ | ||
1038 | |||
1039 | static int uhash_update(uhash_ctx_t ctx, u_char *input, long len) | ||
1040 | /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and | ||
1041 | * hash each one with NH, calling the polyhash on each NH output. | ||
1042 | */ | ||
1043 | { | ||
1044 | UWORD bytes_hashed, bytes_remaining; | ||
1045 | UINT8 nh_result[STREAMS*sizeof(UINT64)]; | ||
1046 | |||
1047 | if (ctx->msg_len + len <= L1_KEY_LEN) { | ||
1048 | nh_update(&ctx->hash, (UINT8 *)input, len); | ||
1049 | ctx->msg_len += len; | ||
1050 | } else { | ||
1051 | |||
1052 | bytes_hashed = ctx->msg_len % L1_KEY_LEN; | ||
1053 | if (ctx->msg_len == L1_KEY_LEN) | ||
1054 | bytes_hashed = L1_KEY_LEN; | ||
1055 | |||
1056 | if (bytes_hashed + len >= L1_KEY_LEN) { | ||
1057 | |||
1058 | /* If some bytes have been passed to the hash function */ | ||
1059 | /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */ | ||
1060 | /* bytes to complete the current nh_block. */ | ||
1061 | if (bytes_hashed) { | ||
1062 | bytes_remaining = (L1_KEY_LEN - bytes_hashed); | ||
1063 | nh_update(&ctx->hash, (UINT8 *)input, bytes_remaining); | ||
1064 | nh_final(&ctx->hash, nh_result); | ||
1065 | ctx->msg_len += bytes_remaining; | ||
1066 | poly_hash(ctx,(UINT32 *)nh_result); | ||
1067 | len -= bytes_remaining; | ||
1068 | input += bytes_remaining; | ||
1069 | } | ||
1070 | |||
1071 | /* Hash directly from input stream if enough bytes */ | ||
1072 | while (len >= L1_KEY_LEN) { | ||
1073 | nh(&ctx->hash, (UINT8 *)input, L1_KEY_LEN, | ||
1074 | L1_KEY_LEN, nh_result); | ||
1075 | ctx->msg_len += L1_KEY_LEN; | ||
1076 | len -= L1_KEY_LEN; | ||
1077 | input += L1_KEY_LEN; | ||
1078 | poly_hash(ctx,(UINT32 *)nh_result); | ||
1079 | } | ||
1080 | } | ||
1081 | |||
1082 | /* pass remaining < L1_KEY_LEN bytes of input data to NH */ | ||
1083 | if (len) { | ||
1084 | nh_update(&ctx->hash, (UINT8 *)input, len); | ||
1085 | ctx->msg_len += len; | ||
1086 | } | ||
1087 | } | ||
1088 | |||
1089 | return (1); | ||
1090 | } | ||
1091 | |||
1092 | /* ---------------------------------------------------------------------- */ | ||
1093 | |||
1094 | static int uhash_final(uhash_ctx_t ctx, u_char *res) | ||
1095 | /* Incorporate any pending data, pad, and generate tag */ | ||
1096 | { | ||
1097 | UINT8 nh_result[STREAMS*sizeof(UINT64)]; | ||
1098 | |||
1099 | if (ctx->msg_len > L1_KEY_LEN) { | ||
1100 | if (ctx->msg_len % L1_KEY_LEN) { | ||
1101 | nh_final(&ctx->hash, nh_result); | ||
1102 | poly_hash(ctx,(UINT32 *)nh_result); | ||
1103 | } | ||
1104 | ip_long(ctx, res); | ||
1105 | } else { | ||
1106 | nh_final(&ctx->hash, nh_result); | ||
1107 | ip_short(ctx,nh_result, res); | ||
1108 | } | ||
1109 | uhash_reset(ctx); | ||
1110 | return (1); | ||
1111 | } | ||
1112 | |||
1113 | /* ---------------------------------------------------------------------- */ | ||
1114 | |||
1115 | #if 0 | ||
1116 | static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res) | ||
1117 | /* assumes that msg is in a writable buffer of length divisible by */ | ||
1118 | /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */ | ||
1119 | { | ||
1120 | UINT8 nh_result[STREAMS*sizeof(UINT64)]; | ||
1121 | UINT32 nh_len; | ||
1122 | int extra_zeroes_needed; | ||
1123 | |||
1124 | /* If the message to be hashed is no longer than L1_HASH_LEN, we skip | ||
1125 | * the polyhash. | ||
1126 | */ | ||
1127 | if (len <= L1_KEY_LEN) { | ||
1128 | if (len == 0) /* If zero length messages will not */ | ||
1129 | nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */ | ||
1130 | else | ||
1131 | nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); | ||
1132 | extra_zeroes_needed = nh_len - len; | ||
1133 | zero_pad((UINT8 *)msg + len, extra_zeroes_needed); | ||
1134 | nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); | ||
1135 | ip_short(ahc,nh_result, res); | ||
1136 | } else { | ||
1137 | /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH | ||
1138 | * output to poly_hash(). | ||
1139 | */ | ||
1140 | do { | ||
1141 | nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result); | ||
1142 | poly_hash(ahc,(UINT32 *)nh_result); | ||
1143 | len -= L1_KEY_LEN; | ||
1144 | msg += L1_KEY_LEN; | ||
1145 | } while (len >= L1_KEY_LEN); | ||
1146 | if (len) { | ||
1147 | nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1)); | ||
1148 | extra_zeroes_needed = nh_len - len; | ||
1149 | zero_pad((UINT8 *)msg + len, extra_zeroes_needed); | ||
1150 | nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result); | ||
1151 | poly_hash(ahc,(UINT32 *)nh_result); | ||
1152 | } | ||
1153 | |||
1154 | ip_long(ahc, res); | ||
1155 | } | ||
1156 | |||
1157 | uhash_reset(ahc); | ||
1158 | return 1; | ||
1159 | } | ||
1160 | #endif | ||
1161 | |||
1162 | /* ---------------------------------------------------------------------- */ | ||
1163 | /* ---------------------------------------------------------------------- */ | ||
1164 | /* ----- Begin UMAC Section --------------------------------------------- */ | ||
1165 | /* ---------------------------------------------------------------------- */ | ||
1166 | /* ---------------------------------------------------------------------- */ | ||
1167 | |||
1168 | /* The UMAC interface has two interfaces, an all-at-once interface where | ||
1169 | * the entire message to be authenticated is passed to UMAC in one buffer, | ||
1170 | * and a sequential interface where the message is presented a little at a | ||
1171 | * time. The all-at-once is more optimaized than the sequential version and | ||
1172 | * should be preferred when the sequential interface is not required. | ||
1173 | */ | ||
1174 | struct umac_ctx { | ||
1175 | uhash_ctx hash; /* Hash function for message compression */ | ||
1176 | pdf_ctx pdf; /* PDF for hashed output */ | ||
1177 | void *free_ptr; /* Address to free this struct via */ | ||
1178 | } umac_ctx; | ||
1179 | |||
1180 | /* ---------------------------------------------------------------------- */ | ||
1181 | |||
1182 | #if 0 | ||
1183 | int umac_reset(struct umac_ctx *ctx) | ||
1184 | /* Reset the hash function to begin a new authentication. */ | ||
1185 | { | ||
1186 | uhash_reset(&ctx->hash); | ||
1187 | return (1); | ||
1188 | } | ||
1189 | #endif | ||
1190 | |||
1191 | /* ---------------------------------------------------------------------- */ | ||
1192 | |||
1193 | int umac_delete(struct umac_ctx *ctx) | ||
1194 | /* Deallocate the ctx structure */ | ||
1195 | { | ||
1196 | if (ctx) { | ||
1197 | if (ALLOC_BOUNDARY) | ||
1198 | ctx = (struct umac_ctx *)ctx->free_ptr; | ||
1199 | free(ctx); | ||
1200 | } | ||
1201 | return (1); | ||
1202 | } | ||
1203 | |||
1204 | /* ---------------------------------------------------------------------- */ | ||
1205 | |||
1206 | struct umac_ctx *umac_new(u_char key[]) | ||
1207 | /* Dynamically allocate a umac_ctx struct, initialize variables, | ||
1208 | * generate subkeys from key. Align to 16-byte boundary. | ||
1209 | */ | ||
1210 | { | ||
1211 | struct umac_ctx *ctx, *octx; | ||
1212 | size_t bytes_to_add; | ||
1213 | aes_int_key prf_key; | ||
1214 | |||
1215 | octx = ctx = malloc(sizeof(*ctx) + ALLOC_BOUNDARY); | ||
1216 | if (ctx) { | ||
1217 | if (ALLOC_BOUNDARY) { | ||
1218 | bytes_to_add = ALLOC_BOUNDARY - | ||
1219 | ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1)); | ||
1220 | ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add); | ||
1221 | } | ||
1222 | ctx->free_ptr = octx; | ||
1223 | aes_key_setup(key,prf_key); | ||
1224 | pdf_init(&ctx->pdf, prf_key); | ||
1225 | uhash_init(&ctx->hash, prf_key); | ||
1226 | } | ||
1227 | |||
1228 | return (ctx); | ||
1229 | } | ||
1230 | |||
1231 | /* ---------------------------------------------------------------------- */ | ||
1232 | |||
1233 | int umac_final(struct umac_ctx *ctx, u_char tag[], u_char nonce[8]) | ||
1234 | /* Incorporate any pending data, pad, and generate tag */ | ||
1235 | { | ||
1236 | uhash_final(&ctx->hash, (u_char *)tag); | ||
1237 | pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); | ||
1238 | |||
1239 | return (1); | ||
1240 | } | ||
1241 | |||
1242 | /* ---------------------------------------------------------------------- */ | ||
1243 | |||
1244 | int umac_update(struct umac_ctx *ctx, u_char *input, long len) | ||
1245 | /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */ | ||
1246 | /* hash each one, calling the PDF on the hashed output whenever the hash- */ | ||
1247 | /* output buffer is full. */ | ||
1248 | { | ||
1249 | uhash_update(&ctx->hash, input, len); | ||
1250 | return (1); | ||
1251 | } | ||
1252 | |||
1253 | /* ---------------------------------------------------------------------- */ | ||
1254 | |||
1255 | #if 0 | ||
1256 | int umac(struct umac_ctx *ctx, u_char *input, | ||
1257 | long len, u_char tag[], | ||
1258 | u_char nonce[8]) | ||
1259 | /* All-in-one version simply calls umac_update() and umac_final(). */ | ||
1260 | { | ||
1261 | uhash(&ctx->hash, input, len, (u_char *)tag); | ||
1262 | pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag); | ||
1263 | |||
1264 | return (1); | ||
1265 | } | ||
1266 | #endif | ||
1267 | |||
1268 | /* ---------------------------------------------------------------------- */ | ||
1269 | /* ---------------------------------------------------------------------- */ | ||
1270 | /* ----- End UMAC Section ----------------------------------------------- */ | ||
1271 | /* ---------------------------------------------------------------------- */ | ||
1272 | /* ---------------------------------------------------------------------- */ | ||