1 /* $OpenBSD: umac.c,v 1.21 2021/04/03 06:58:30 djm Exp $ */
2 /* -----------------------------------------------------------------------
4 * umac.c -- C Implementation UMAC Message Authentication
6 * Version 0.93b of rfc4418.txt -- 2006 July 18
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.
12 * Copyright (c) 1999-2006 Ted Krovetz
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.
21 * Comments should be directed to Ted Krovetz (tdk@acm.org)
23 * ---------------------------------------------------------------------- */
25 /* ////////////////////// IMPORTANT NOTES /////////////////////////////////
27 * 1) This version does not work properly on messages larger than 16MB
29 * 2) If you set the switch to use SSE2, then all data must be 16-byte
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
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 license at http://fp.gladman.plus.com/AES/index.htm. It
43 * includes a fast IA-32 assembly version. The OpenSSL crypo library is
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.
49 /////////////////////////////////////////////////////////////////////// */
51 /* ---------------------------------------------------------------------- */
52 /* --- User Switches ---------------------------------------------------- */
53 /* ---------------------------------------------------------------------- */
55 #ifndef UMAC_OUTPUT_LEN
56 #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */
59 #if UMAC_OUTPUT_LEN != 4 && UMAC_OUTPUT_LEN != 8 && \
60 UMAC_OUTPUT_LEN != 12 && UMAC_OUTPUT_LEN != 16
61 # error UMAC_OUTPUT_LEN must be defined to 4, 8, 12 or 16
64 /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */
65 /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */
66 /* #define SSE2 0 Is SSE2 is available? */
67 /* #define RUN_TESTS 0 Run basic correctness/speed tests */
68 /* #define UMAC_AE_SUPPORT 0 Enable authenticated encryption */
70 /* ---------------------------------------------------------------------- */
71 /* -- Global Includes --------------------------------------------------- */
72 /* ---------------------------------------------------------------------- */
75 #include <sys/types.h>
86 /* ---------------------------------------------------------------------- */
87 /* --- Primitive Data Types --- */
88 /* ---------------------------------------------------------------------- */
90 /* The following assumptions may need change on your system */
91 typedef u_int8_t UINT8
; /* 1 byte */
92 typedef u_int16_t UINT16
; /* 2 byte */
93 typedef u_int32_t UINT32
; /* 4 byte */
94 typedef u_int64_t UINT64
; /* 8 bytes */
95 typedef unsigned int UWORD
; /* Register */
97 /* ---------------------------------------------------------------------- */
98 /* --- Constants -------------------------------------------------------- */
99 /* ---------------------------------------------------------------------- */
101 #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */
103 /* Message "words" are read from memory in an endian-specific manner. */
104 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */
105 /* be set true if the host computer is little-endian. */
107 #if BYTE_ORDER == LITTLE_ENDIAN
108 #define __LITTLE_ENDIAN__ 1
110 #define __LITTLE_ENDIAN__ 0
113 /* ---------------------------------------------------------------------- */
114 /* ---------------------------------------------------------------------- */
115 /* ----- Architecture Specific ------------------------------------------ */
116 /* ---------------------------------------------------------------------- */
117 /* ---------------------------------------------------------------------- */
120 /* ---------------------------------------------------------------------- */
121 /* ---------------------------------------------------------------------- */
122 /* ----- Primitive Routines --------------------------------------------- */
123 /* ---------------------------------------------------------------------- */
124 /* ---------------------------------------------------------------------- */
127 /* ---------------------------------------------------------------------- */
128 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
129 /* ---------------------------------------------------------------------- */
131 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
133 /* ---------------------------------------------------------------------- */
134 /* --- Endian Conversion --- Forcing assembly on some platforms */
135 /* ---------------------------------------------------------------------- */
137 #if (__LITTLE_ENDIAN__)
138 #define LOAD_UINT32_REVERSED(p) get_u32(p)
139 #define STORE_UINT32_REVERSED(p,v) put_u32(p,v)
141 #define LOAD_UINT32_REVERSED(p) get_u32_le(p)
142 #define STORE_UINT32_REVERSED(p,v) put_u32_le(p,v)
145 #define LOAD_UINT32_LITTLE(p) (get_u32_le(p))
146 #define STORE_UINT32_BIG(p,v) put_u32(p, v)
148 /* ---------------------------------------------------------------------- */
149 /* ---------------------------------------------------------------------- */
150 /* ----- Begin KDF & PDF Section ---------------------------------------- */
151 /* ---------------------------------------------------------------------- */
152 /* ---------------------------------------------------------------------- */
154 /* UMAC uses AES with 16 byte block and key lengths */
155 #define AES_BLOCK_LEN 16
159 #include "openbsd-compat/openssl-compat.h"
160 #ifndef USE_BUILTIN_RIJNDAEL
161 # include <openssl/aes.h>
163 typedef AES_KEY aes_int_key
[1];
164 #define aes_encryption(in,out,int_key) \
165 AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key)
166 #define aes_key_setup(key,int_key) \
167 AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key)
169 #include "rijndael.h"
170 #define AES_ROUNDS ((UMAC_KEY_LEN / 4) + 6)
171 typedef UINT8 aes_int_key
[AES_ROUNDS
+1][4][4]; /* AES internal */
172 #define aes_encryption(in,out,int_key) \
173 rijndaelEncrypt((u32 *)(int_key), AES_ROUNDS, (u8 *)(in), (u8 *)(out))
174 #define aes_key_setup(key,int_key) \
175 rijndaelKeySetupEnc((u32 *)(int_key), (const unsigned char *)(key), \
179 /* The user-supplied UMAC key is stretched using AES in a counter
180 * mode to supply all random bits needed by UMAC. The kdf function takes
181 * an AES internal key representation 'key' and writes a stream of
182 * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct
183 * 'ndx' causes a distinct byte stream.
185 static void kdf(void *bufp
, aes_int_key key
, UINT8 ndx
, int nbytes
)
187 UINT8 in_buf
[AES_BLOCK_LEN
] = {0};
188 UINT8 out_buf
[AES_BLOCK_LEN
];
189 UINT8
*dst_buf
= (UINT8
*)bufp
;
192 /* Setup the initial value */
193 in_buf
[AES_BLOCK_LEN
-9] = ndx
;
194 in_buf
[AES_BLOCK_LEN
-1] = i
= 1;
196 while (nbytes
>= AES_BLOCK_LEN
) {
197 aes_encryption(in_buf
, out_buf
, key
);
198 memcpy(dst_buf
,out_buf
,AES_BLOCK_LEN
);
199 in_buf
[AES_BLOCK_LEN
-1] = ++i
;
200 nbytes
-= AES_BLOCK_LEN
;
201 dst_buf
+= AES_BLOCK_LEN
;
204 aes_encryption(in_buf
, out_buf
, key
);
205 memcpy(dst_buf
,out_buf
,nbytes
);
207 explicit_bzero(in_buf
, sizeof(in_buf
));
208 explicit_bzero(out_buf
, sizeof(out_buf
));
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.
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 */
223 static void pdf_init(pdf_ctx
*pc
, aes_int_key prf_key
)
225 UINT8 buf
[UMAC_KEY_LEN
];
227 kdf(buf
, prf_key
, 0, UMAC_KEY_LEN
);
228 aes_key_setup(buf
, pc
->prf_key
);
230 /* Initialize pdf and cache */
231 memset(pc
->nonce
, 0, sizeof(pc
->nonce
));
232 aes_encryption(pc
->nonce
, pc
->cache
, pc
->prf_key
);
233 explicit_bzero(buf
, sizeof(buf
));
236 static void pdf_gen_xor(pdf_ctx
*pc
, const UINT8 nonce
[8], UINT8 buf
[8])
238 /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
239 * of the AES output. If last time around we returned the ndx-1st
240 * element, then we may have the result in the cache already.
243 #if (UMAC_OUTPUT_LEN == 4)
244 #define LOW_BIT_MASK 3
245 #elif (UMAC_OUTPUT_LEN == 8)
246 #define LOW_BIT_MASK 1
247 #elif (UMAC_OUTPUT_LEN > 8)
248 #define LOW_BIT_MASK 0
251 UINT8 tmp_nonce_lo
[4];
254 #if LOW_BIT_MASK != 0
255 int ndx
= nonce
[7] & LOW_BIT_MASK
;
257 *(UINT32
*)t
.tmp_nonce_lo
= ((const UINT32
*)nonce
)[1];
258 t
.tmp_nonce_lo
[3] &= ~LOW_BIT_MASK
; /* zero last bit */
260 if ( (((UINT32
*)t
.tmp_nonce_lo
)[0] != ((UINT32
*)pc
->nonce
)[1]) ||
261 (((const UINT32
*)nonce
)[0] != ((UINT32
*)pc
->nonce
)[0]) )
263 ((UINT32
*)pc
->nonce
)[0] = ((const UINT32
*)nonce
)[0];
264 ((UINT32
*)pc
->nonce
)[1] = ((UINT32
*)t
.tmp_nonce_lo
)[0];
265 aes_encryption(pc
->nonce
, pc
->cache
, pc
->prf_key
);
268 #if (UMAC_OUTPUT_LEN == 4)
269 *((UINT32
*)buf
) ^= ((UINT32
*)pc
->cache
)[ndx
];
270 #elif (UMAC_OUTPUT_LEN == 8)
271 *((UINT64
*)buf
) ^= ((UINT64
*)pc
->cache
)[ndx
];
272 #elif (UMAC_OUTPUT_LEN == 12)
273 ((UINT64
*)buf
)[0] ^= ((UINT64
*)pc
->cache
)[0];
274 ((UINT32
*)buf
)[2] ^= ((UINT32
*)pc
->cache
)[2];
275 #elif (UMAC_OUTPUT_LEN == 16)
276 ((UINT64
*)buf
)[0] ^= ((UINT64
*)pc
->cache
)[0];
277 ((UINT64
*)buf
)[1] ^= ((UINT64
*)pc
->cache
)[1];
281 /* ---------------------------------------------------------------------- */
282 /* ---------------------------------------------------------------------- */
283 /* ----- Begin NH Hash Section ------------------------------------------ */
284 /* ---------------------------------------------------------------------- */
285 /* ---------------------------------------------------------------------- */
287 /* The NH-based hash functions used in UMAC are described in the UMAC paper
288 * and specification, both of which can be found at the UMAC website.
289 * The interface to this implementation has two
290 * versions, one expects the entire message being hashed to be passed
291 * in a single buffer and returns the hash result immediately. The second
292 * allows the message to be passed in a sequence of buffers. In the
293 * multiple-buffer interface, the client calls the routine nh_update() as
294 * many times as necessary. When there is no more data to be fed to the
295 * hash, the client calls nh_final() which calculates the hash output.
296 * Before beginning another hash calculation the nh_reset() routine
297 * must be called. The single-buffer routine, nh(), is equivalent to
298 * the sequence of calls nh_update() and nh_final(); however it is
299 * optimized and should be preferred whenever the multiple-buffer interface
300 * is not necessary. When using either interface, it is the client's
301 * responsibility to pass no more than L1_KEY_LEN bytes per hash result.
303 * The routine nh_init() initializes the nh_ctx data structure and
304 * must be called once, before any other PDF routine.
307 /* The "nh_aux" routines do the actual NH hashing work. They
308 * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
309 * produce output for all STREAMS NH iterations in one call,
310 * allowing the parallel implementation of the streams.
313 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */
314 #define L1_KEY_LEN 1024 /* Internal key bytes */
315 #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */
316 #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */
317 #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */
318 #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */
321 UINT8 nh_key
[L1_KEY_LEN
+ L1_KEY_SHIFT
* (STREAMS
- 1)]; /* NH Key */
322 UINT8 data
[HASH_BUF_BYTES
]; /* Incoming data buffer */
323 int next_data_empty
; /* Bookkeeping variable for data buffer. */
324 int bytes_hashed
; /* Bytes (out of L1_KEY_LEN) incorporated. */
325 UINT64 state
[STREAMS
]; /* on-line state */
329 #if (UMAC_OUTPUT_LEN == 4)
331 static void nh_aux(void *kp
, const void *dp
, void *hp
, UINT32 dlen
)
332 /* NH hashing primitive. Previous (partial) hash result is loaded and
333 * then stored via hp pointer. The length of the data pointed at by "dp",
334 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key
335 * is expected to be endian compensated in memory at key setup.
340 UINT32
*k
= (UINT32
*)kp
;
341 const UINT32
*d
= (const UINT32
*)dp
;
342 UINT32 d0
,d1
,d2
,d3
,d4
,d5
,d6
,d7
;
343 UINT32 k0
,k1
,k2
,k3
,k4
,k5
,k6
,k7
;
347 d0
= LOAD_UINT32_LITTLE(d
+0); d1
= LOAD_UINT32_LITTLE(d
+1);
348 d2
= LOAD_UINT32_LITTLE(d
+2); d3
= LOAD_UINT32_LITTLE(d
+3);
349 d4
= LOAD_UINT32_LITTLE(d
+4); d5
= LOAD_UINT32_LITTLE(d
+5);
350 d6
= LOAD_UINT32_LITTLE(d
+6); d7
= LOAD_UINT32_LITTLE(d
+7);
351 k0
= *(k
+0); k1
= *(k
+1); k2
= *(k
+2); k3
= *(k
+3);
352 k4
= *(k
+4); k5
= *(k
+5); k6
= *(k
+6); k7
= *(k
+7);
353 h
+= MUL64((k0
+ d0
), (k4
+ d4
));
354 h
+= MUL64((k1
+ d1
), (k5
+ d5
));
355 h
+= MUL64((k2
+ d2
), (k6
+ d6
));
356 h
+= MUL64((k3
+ d3
), (k7
+ d7
));
364 #elif (UMAC_OUTPUT_LEN == 8)
366 static void nh_aux(void *kp
, const void *dp
, void *hp
, UINT32 dlen
)
367 /* Same as previous nh_aux, but two streams are handled in one pass,
368 * reading and writing 16 bytes of hash-state per call.
373 UINT32
*k
= (UINT32
*)kp
;
374 const UINT32
*d
= (const UINT32
*)dp
;
375 UINT32 d0
,d1
,d2
,d3
,d4
,d5
,d6
,d7
;
376 UINT32 k0
,k1
,k2
,k3
,k4
,k5
,k6
,k7
,
379 h1
= *((UINT64
*)hp
);
380 h2
= *((UINT64
*)hp
+ 1);
381 k0
= *(k
+0); k1
= *(k
+1); k2
= *(k
+2); k3
= *(k
+3);
383 d0
= LOAD_UINT32_LITTLE(d
+0); d1
= LOAD_UINT32_LITTLE(d
+1);
384 d2
= LOAD_UINT32_LITTLE(d
+2); d3
= LOAD_UINT32_LITTLE(d
+3);
385 d4
= LOAD_UINT32_LITTLE(d
+4); d5
= LOAD_UINT32_LITTLE(d
+5);
386 d6
= LOAD_UINT32_LITTLE(d
+6); d7
= LOAD_UINT32_LITTLE(d
+7);
387 k4
= *(k
+4); k5
= *(k
+5); k6
= *(k
+6); k7
= *(k
+7);
388 k8
= *(k
+8); k9
= *(k
+9); k10
= *(k
+10); k11
= *(k
+11);
390 h1
+= MUL64((k0
+ d0
), (k4
+ d4
));
391 h2
+= MUL64((k4
+ d0
), (k8
+ d4
));
393 h1
+= MUL64((k1
+ d1
), (k5
+ d5
));
394 h2
+= MUL64((k5
+ d1
), (k9
+ d5
));
396 h1
+= MUL64((k2
+ d2
), (k6
+ d6
));
397 h2
+= MUL64((k6
+ d2
), (k10
+ d6
));
399 h1
+= MUL64((k3
+ d3
), (k7
+ d7
));
400 h2
+= MUL64((k7
+ d3
), (k11
+ d7
));
402 k0
= k8
; k1
= k9
; k2
= k10
; k3
= k11
;
407 ((UINT64
*)hp
)[0] = h1
;
408 ((UINT64
*)hp
)[1] = h2
;
411 #elif (UMAC_OUTPUT_LEN == 12)
413 static void nh_aux(void *kp
, const void *dp
, void *hp
, UINT32 dlen
)
414 /* Same as previous nh_aux, but two streams are handled in one pass,
415 * reading and writing 24 bytes of hash-state per call.
420 UINT32
*k
= (UINT32
*)kp
;
421 const UINT32
*d
= (const UINT32
*)dp
;
422 UINT32 d0
,d1
,d2
,d3
,d4
,d5
,d6
,d7
;
423 UINT32 k0
,k1
,k2
,k3
,k4
,k5
,k6
,k7
,
424 k8
,k9
,k10
,k11
,k12
,k13
,k14
,k15
;
426 h1
= *((UINT64
*)hp
);
427 h2
= *((UINT64
*)hp
+ 1);
428 h3
= *((UINT64
*)hp
+ 2);
429 k0
= *(k
+0); k1
= *(k
+1); k2
= *(k
+2); k3
= *(k
+3);
430 k4
= *(k
+4); k5
= *(k
+5); k6
= *(k
+6); k7
= *(k
+7);
432 d0
= LOAD_UINT32_LITTLE(d
+0); d1
= LOAD_UINT32_LITTLE(d
+1);
433 d2
= LOAD_UINT32_LITTLE(d
+2); d3
= LOAD_UINT32_LITTLE(d
+3);
434 d4
= LOAD_UINT32_LITTLE(d
+4); d5
= LOAD_UINT32_LITTLE(d
+5);
435 d6
= LOAD_UINT32_LITTLE(d
+6); d7
= LOAD_UINT32_LITTLE(d
+7);
436 k8
= *(k
+8); k9
= *(k
+9); k10
= *(k
+10); k11
= *(k
+11);
437 k12
= *(k
+12); k13
= *(k
+13); k14
= *(k
+14); k15
= *(k
+15);
439 h1
+= MUL64((k0
+ d0
), (k4
+ d4
));
440 h2
+= MUL64((k4
+ d0
), (k8
+ d4
));
441 h3
+= MUL64((k8
+ d0
), (k12
+ d4
));
443 h1
+= MUL64((k1
+ d1
), (k5
+ d5
));
444 h2
+= MUL64((k5
+ d1
), (k9
+ d5
));
445 h3
+= MUL64((k9
+ d1
), (k13
+ d5
));
447 h1
+= MUL64((k2
+ d2
), (k6
+ d6
));
448 h2
+= MUL64((k6
+ d2
), (k10
+ d6
));
449 h3
+= MUL64((k10
+ d2
), (k14
+ d6
));
451 h1
+= MUL64((k3
+ d3
), (k7
+ d7
));
452 h2
+= MUL64((k7
+ d3
), (k11
+ d7
));
453 h3
+= MUL64((k11
+ d3
), (k15
+ d7
));
455 k0
= k8
; k1
= k9
; k2
= k10
; k3
= k11
;
456 k4
= k12
; k5
= k13
; k6
= k14
; k7
= k15
;
461 ((UINT64
*)hp
)[0] = h1
;
462 ((UINT64
*)hp
)[1] = h2
;
463 ((UINT64
*)hp
)[2] = h3
;
466 #elif (UMAC_OUTPUT_LEN == 16)
468 static void nh_aux(void *kp
, const void *dp
, void *hp
, UINT32 dlen
)
469 /* Same as previous nh_aux, but two streams are handled in one pass,
470 * reading and writing 24 bytes of hash-state per call.
475 UINT32
*k
= (UINT32
*)kp
;
476 const UINT32
*d
= (const UINT32
*)dp
;
477 UINT32 d0
,d1
,d2
,d3
,d4
,d5
,d6
,d7
;
478 UINT32 k0
,k1
,k2
,k3
,k4
,k5
,k6
,k7
,
479 k8
,k9
,k10
,k11
,k12
,k13
,k14
,k15
,
482 h1
= *((UINT64
*)hp
);
483 h2
= *((UINT64
*)hp
+ 1);
484 h3
= *((UINT64
*)hp
+ 2);
485 h4
= *((UINT64
*)hp
+ 3);
486 k0
= *(k
+0); k1
= *(k
+1); k2
= *(k
+2); k3
= *(k
+3);
487 k4
= *(k
+4); k5
= *(k
+5); k6
= *(k
+6); k7
= *(k
+7);
489 d0
= LOAD_UINT32_LITTLE(d
+0); d1
= LOAD_UINT32_LITTLE(d
+1);
490 d2
= LOAD_UINT32_LITTLE(d
+2); d3
= LOAD_UINT32_LITTLE(d
+3);
491 d4
= LOAD_UINT32_LITTLE(d
+4); d5
= LOAD_UINT32_LITTLE(d
+5);
492 d6
= LOAD_UINT32_LITTLE(d
+6); d7
= LOAD_UINT32_LITTLE(d
+7);
493 k8
= *(k
+8); k9
= *(k
+9); k10
= *(k
+10); k11
= *(k
+11);
494 k12
= *(k
+12); k13
= *(k
+13); k14
= *(k
+14); k15
= *(k
+15);
495 k16
= *(k
+16); k17
= *(k
+17); k18
= *(k
+18); k19
= *(k
+19);
497 h1
+= MUL64((k0
+ d0
), (k4
+ d4
));
498 h2
+= MUL64((k4
+ d0
), (k8
+ d4
));
499 h3
+= MUL64((k8
+ d0
), (k12
+ d4
));
500 h4
+= MUL64((k12
+ d0
), (k16
+ d4
));
502 h1
+= MUL64((k1
+ d1
), (k5
+ d5
));
503 h2
+= MUL64((k5
+ d1
), (k9
+ d5
));
504 h3
+= MUL64((k9
+ d1
), (k13
+ d5
));
505 h4
+= MUL64((k13
+ d1
), (k17
+ d5
));
507 h1
+= MUL64((k2
+ d2
), (k6
+ d6
));
508 h2
+= MUL64((k6
+ d2
), (k10
+ d6
));
509 h3
+= MUL64((k10
+ d2
), (k14
+ d6
));
510 h4
+= MUL64((k14
+ d2
), (k18
+ d6
));
512 h1
+= MUL64((k3
+ d3
), (k7
+ d7
));
513 h2
+= MUL64((k7
+ d3
), (k11
+ d7
));
514 h3
+= MUL64((k11
+ d3
), (k15
+ d7
));
515 h4
+= MUL64((k15
+ d3
), (k19
+ d7
));
517 k0
= k8
; k1
= k9
; k2
= k10
; k3
= k11
;
518 k4
= k12
; k5
= k13
; k6
= k14
; k7
= k15
;
519 k8
= k16
; k9
= k17
; k10
= k18
; k11
= k19
;
524 ((UINT64
*)hp
)[0] = h1
;
525 ((UINT64
*)hp
)[1] = h2
;
526 ((UINT64
*)hp
)[2] = h3
;
527 ((UINT64
*)hp
)[3] = h4
;
530 /* ---------------------------------------------------------------------- */
531 #endif /* UMAC_OUTPUT_LENGTH */
532 /* ---------------------------------------------------------------------- */
535 /* ---------------------------------------------------------------------- */
537 static void nh_transform(nh_ctx
*hc
, const UINT8
*buf
, UINT32 nbytes
)
538 /* This function is a wrapper for the primitive NH hash functions. It takes
539 * as argument "hc" the current hash context and a buffer which must be a
540 * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
541 * appropriately according to how much message has been hashed already.
546 key
= hc
->nh_key
+ hc
->bytes_hashed
;
547 nh_aux(key
, buf
, hc
->state
, nbytes
);
550 /* ---------------------------------------------------------------------- */
552 #if (__LITTLE_ENDIAN__)
553 static void endian_convert(void *buf
, UWORD bpw
, UINT32 num_bytes
)
554 /* We endian convert the keys on little-endian computers to */
555 /* compensate for the lack of big-endian memory reads during hashing. */
557 UWORD iters
= num_bytes
/ bpw
;
559 UINT32
*p
= (UINT32
*)buf
;
561 *p
= LOAD_UINT32_REVERSED(p
);
564 } else if (bpw
== 8) {
565 UINT32
*p
= (UINT32
*)buf
;
568 t
= LOAD_UINT32_REVERSED(p
+1);
569 p
[1] = LOAD_UINT32_REVERSED(p
);
575 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
577 #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */
580 /* ---------------------------------------------------------------------- */
582 static void nh_reset(nh_ctx
*hc
)
583 /* Reset nh_ctx to ready for hashing of new data */
585 hc
->bytes_hashed
= 0;
586 hc
->next_data_empty
= 0;
588 #if (UMAC_OUTPUT_LEN >= 8)
591 #if (UMAC_OUTPUT_LEN >= 12)
594 #if (UMAC_OUTPUT_LEN == 16)
600 /* ---------------------------------------------------------------------- */
602 static void nh_init(nh_ctx
*hc
, aes_int_key prf_key
)
603 /* Generate nh_key, endian convert and reset to be ready for hashing. */
605 kdf(hc
->nh_key
, prf_key
, 1, sizeof(hc
->nh_key
));
606 endian_convert_if_le(hc
->nh_key
, 4, sizeof(hc
->nh_key
));
610 /* ---------------------------------------------------------------------- */
612 static void nh_update(nh_ctx
*hc
, const UINT8
*buf
, UINT32 nbytes
)
613 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */
614 /* even multiple of HASH_BUF_BYTES. */
618 j
= hc
->next_data_empty
;
619 if ((j
+ nbytes
) >= HASH_BUF_BYTES
) {
621 i
= HASH_BUF_BYTES
- j
;
622 memcpy(hc
->data
+j
, buf
, i
);
623 nh_transform(hc
,hc
->data
,HASH_BUF_BYTES
);
626 hc
->bytes_hashed
+= HASH_BUF_BYTES
;
628 if (nbytes
>= HASH_BUF_BYTES
) {
629 i
= nbytes
& ~(HASH_BUF_BYTES
- 1);
630 nh_transform(hc
, buf
, i
);
633 hc
->bytes_hashed
+= i
;
637 memcpy(hc
->data
+ j
, buf
, nbytes
);
638 hc
->next_data_empty
= j
+ nbytes
;
641 /* ---------------------------------------------------------------------- */
643 static void zero_pad(UINT8
*p
, int nbytes
)
645 /* Write "nbytes" of zeroes, beginning at "p" */
646 if (nbytes
>= (int)sizeof(UWORD
)) {
647 while ((ptrdiff_t)p
% sizeof(UWORD
)) {
652 while (nbytes
>= (int)sizeof(UWORD
)) {
654 nbytes
-= sizeof(UWORD
);
665 /* ---------------------------------------------------------------------- */
667 static void nh_final(nh_ctx
*hc
, UINT8
*result
)
668 /* After passing some number of data buffers to nh_update() for integration
669 * into an NH context, nh_final is called to produce a hash result. If any
670 * bytes are in the buffer hc->data, incorporate them into the
671 * NH context. Finally, add into the NH accumulation "state" the total number
672 * of bits hashed. The resulting numbers are written to the buffer "result".
673 * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
678 if (hc
->next_data_empty
!= 0) {
679 nh_len
= ((hc
->next_data_empty
+ (L1_PAD_BOUNDARY
- 1)) &
680 ~(L1_PAD_BOUNDARY
- 1));
681 zero_pad(hc
->data
+ hc
->next_data_empty
,
682 nh_len
- hc
->next_data_empty
);
683 nh_transform(hc
, hc
->data
, nh_len
);
684 hc
->bytes_hashed
+= hc
->next_data_empty
;
685 } else if (hc
->bytes_hashed
== 0) {
686 nh_len
= L1_PAD_BOUNDARY
;
687 zero_pad(hc
->data
, L1_PAD_BOUNDARY
);
688 nh_transform(hc
, hc
->data
, nh_len
);
691 nbits
= (hc
->bytes_hashed
<< 3);
692 ((UINT64
*)result
)[0] = ((UINT64
*)hc
->state
)[0] + nbits
;
693 #if (UMAC_OUTPUT_LEN >= 8)
694 ((UINT64
*)result
)[1] = ((UINT64
*)hc
->state
)[1] + nbits
;
696 #if (UMAC_OUTPUT_LEN >= 12)
697 ((UINT64
*)result
)[2] = ((UINT64
*)hc
->state
)[2] + nbits
;
699 #if (UMAC_OUTPUT_LEN == 16)
700 ((UINT64
*)result
)[3] = ((UINT64
*)hc
->state
)[3] + nbits
;
705 /* ---------------------------------------------------------------------- */
707 static void nh(nh_ctx
*hc
, const UINT8
*buf
, UINT32 padded_len
,
708 UINT32 unpadded_len
, UINT8
*result
)
709 /* All-in-one nh_update() and nh_final() equivalent.
710 * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
716 /* Initialize the hash state */
717 nbits
= (unpadded_len
<< 3);
719 ((UINT64
*)result
)[0] = nbits
;
720 #if (UMAC_OUTPUT_LEN >= 8)
721 ((UINT64
*)result
)[1] = nbits
;
723 #if (UMAC_OUTPUT_LEN >= 12)
724 ((UINT64
*)result
)[2] = nbits
;
726 #if (UMAC_OUTPUT_LEN == 16)
727 ((UINT64
*)result
)[3] = nbits
;
730 nh_aux(hc
->nh_key
, buf
, result
, padded_len
);
733 /* ---------------------------------------------------------------------- */
734 /* ---------------------------------------------------------------------- */
735 /* ----- Begin UHASH Section -------------------------------------------- */
736 /* ---------------------------------------------------------------------- */
737 /* ---------------------------------------------------------------------- */
739 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
740 * hashed by NH. The NH output is then hashed by a polynomial-hash layer
741 * unless the initial data to be hashed is short. After the polynomial-
742 * layer, an inner-product hash is used to produce the final UHASH output.
744 * UHASH provides two interfaces, one all-at-once and another where data
745 * buffers are presented sequentially. In the sequential interface, the
746 * UHASH client calls the routine uhash_update() as many times as necessary.
747 * When there is no more data to be fed to UHASH, the client calls
748 * uhash_final() which
749 * calculates the UHASH output. Before beginning another UHASH calculation
750 * the uhash_reset() routine must be called. The all-at-once UHASH routine,
751 * uhash(), is equivalent to the sequence of calls uhash_update() and
752 * uhash_final(); however it is optimized and should be
753 * used whenever the sequential interface is not necessary.
755 * The routine uhash_init() initializes the uhash_ctx data structure and
756 * must be called once, before any other UHASH routine.
759 /* ---------------------------------------------------------------------- */
760 /* ----- Constants and uhash_ctx ---------------------------------------- */
761 /* ---------------------------------------------------------------------- */
763 /* ---------------------------------------------------------------------- */
764 /* ----- Poly hash and Inner-Product hash Constants --------------------- */
765 /* ---------------------------------------------------------------------- */
767 /* Primes and masks */
768 #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */
769 #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */
770 #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */
773 /* ---------------------------------------------------------------------- */
775 typedef struct uhash_ctx
{
776 nh_ctx hash
; /* Hash context for L1 NH hash */
777 UINT64 poly_key_8
[STREAMS
]; /* p64 poly keys */
778 UINT64 poly_accum
[STREAMS
]; /* poly hash result */
779 UINT64 ip_keys
[STREAMS
*4]; /* Inner-product keys */
780 UINT32 ip_trans
[STREAMS
]; /* Inner-product translation */
781 UINT32 msg_len
; /* Total length of data passed */
784 typedef struct uhash_ctx
*uhash_ctx_t
;
786 /* ---------------------------------------------------------------------- */
789 /* The polynomial hashes use Horner's rule to evaluate a polynomial one
790 * word at a time. As described in the specification, poly32 and poly64
791 * require keys from special domains. The following implementations exploit
792 * the special domains to avoid overflow. The results are not guaranteed to
793 * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
794 * patches any errant values.
797 static UINT64
poly64(UINT64 cur
, UINT64 key
, UINT64 data
)
799 UINT32 key_hi
= (UINT32
)(key
>> 32),
800 key_lo
= (UINT32
)key
,
801 cur_hi
= (UINT32
)(cur
>> 32),
802 cur_lo
= (UINT32
)cur
,
807 X
= MUL64(key_hi
, cur_lo
) + MUL64(cur_hi
, key_lo
);
809 x_hi
= (UINT32
)(X
>> 32);
811 res
= (MUL64(key_hi
, cur_hi
) + x_hi
) * 59 + MUL64(key_lo
, cur_lo
);
813 T
= ((UINT64
)x_lo
<< 32);
826 /* Although UMAC is specified to use a ramped polynomial hash scheme, this
827 * implementation does not handle all ramp levels. Because we don't handle
828 * the ramp up to p128 modulus in this implementation, we are limited to
829 * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
830 * bytes input to UMAC per tag, ie. 16MB).
832 static void poly_hash(uhash_ctx_t hc
, UINT32 data_in
[])
835 UINT64
*data
=(UINT64
*)data_in
;
837 for (i
= 0; i
< STREAMS
; i
++) {
838 if ((UINT32
)(data
[i
] >> 32) == 0xfffffffful
) {
839 hc
->poly_accum
[i
] = poly64(hc
->poly_accum
[i
],
840 hc
->poly_key_8
[i
], p64
- 1);
841 hc
->poly_accum
[i
] = poly64(hc
->poly_accum
[i
],
842 hc
->poly_key_8
[i
], (data
[i
] - 59));
844 hc
->poly_accum
[i
] = poly64(hc
->poly_accum
[i
],
845 hc
->poly_key_8
[i
], data
[i
]);
851 /* ---------------------------------------------------------------------- */
854 /* The final step in UHASH is an inner-product hash. The poly hash
855 * produces a result not necessarily WORD_LEN bytes long. The inner-
856 * product hash breaks the polyhash output into 16-bit chunks and
857 * multiplies each with a 36 bit key.
860 static UINT64
ip_aux(UINT64 t
, UINT64
*ipkp
, UINT64 data
)
862 t
= t
+ ipkp
[0] * (UINT64
)(UINT16
)(data
>> 48);
863 t
= t
+ ipkp
[1] * (UINT64
)(UINT16
)(data
>> 32);
864 t
= t
+ ipkp
[2] * (UINT64
)(UINT16
)(data
>> 16);
865 t
= t
+ ipkp
[3] * (UINT64
)(UINT16
)(data
);
870 static UINT32
ip_reduce_p36(UINT64 t
)
872 /* Divisionless modular reduction */
875 ret
= (t
& m36
) + 5 * (t
>> 36);
879 /* return least significant 32 bits */
880 return (UINT32
)(ret
);
884 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
885 * the polyhash stage is skipped and ip_short is applied directly to the
888 static void ip_short(uhash_ctx_t ahc
, UINT8
*nh_res
, u_char
*res
)
891 UINT64
*nhp
= (UINT64
*)nh_res
;
893 t
= ip_aux(0,ahc
->ip_keys
, nhp
[0]);
894 STORE_UINT32_BIG((UINT32
*)res
+0, ip_reduce_p36(t
) ^ ahc
->ip_trans
[0]);
895 #if (UMAC_OUTPUT_LEN >= 8)
896 t
= ip_aux(0,ahc
->ip_keys
+4, nhp
[1]);
897 STORE_UINT32_BIG((UINT32
*)res
+1, ip_reduce_p36(t
) ^ ahc
->ip_trans
[1]);
899 #if (UMAC_OUTPUT_LEN >= 12)
900 t
= ip_aux(0,ahc
->ip_keys
+8, nhp
[2]);
901 STORE_UINT32_BIG((UINT32
*)res
+2, ip_reduce_p36(t
) ^ ahc
->ip_trans
[2]);
903 #if (UMAC_OUTPUT_LEN == 16)
904 t
= ip_aux(0,ahc
->ip_keys
+12, nhp
[3]);
905 STORE_UINT32_BIG((UINT32
*)res
+3, ip_reduce_p36(t
) ^ ahc
->ip_trans
[3]);
909 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
910 * the polyhash stage is not skipped and ip_long is applied to the
913 static void ip_long(uhash_ctx_t ahc
, u_char
*res
)
918 for (i
= 0; i
< STREAMS
; i
++) {
919 /* fix polyhash output not in Z_p64 */
920 if (ahc
->poly_accum
[i
] >= p64
)
921 ahc
->poly_accum
[i
] -= p64
;
922 t
= ip_aux(0,ahc
->ip_keys
+(i
*4), ahc
->poly_accum
[i
]);
923 STORE_UINT32_BIG((UINT32
*)res
+i
,
924 ip_reduce_p36(t
) ^ ahc
->ip_trans
[i
]);
929 /* ---------------------------------------------------------------------- */
931 /* ---------------------------------------------------------------------- */
933 /* Reset uhash context for next hash session */
934 static int uhash_reset(uhash_ctx_t pc
)
938 pc
->poly_accum
[0] = 1;
939 #if (UMAC_OUTPUT_LEN >= 8)
940 pc
->poly_accum
[1] = 1;
942 #if (UMAC_OUTPUT_LEN >= 12)
943 pc
->poly_accum
[2] = 1;
945 #if (UMAC_OUTPUT_LEN == 16)
946 pc
->poly_accum
[3] = 1;
951 /* ---------------------------------------------------------------------- */
953 /* Given a pointer to the internal key needed by kdf() and a uhash context,
954 * initialize the NH context and generate keys needed for poly and inner-
955 * product hashing. All keys are endian adjusted in memory so that native
956 * loads cause correct keys to be in registers during calculation.
958 static void uhash_init(uhash_ctx_t ahc
, aes_int_key prf_key
)
961 UINT8 buf
[(8*STREAMS
+4)*sizeof(UINT64
)];
963 /* Zero the entire uhash context */
964 memset(ahc
, 0, sizeof(uhash_ctx
));
966 /* Initialize the L1 hash */
967 nh_init(&ahc
->hash
, prf_key
);
969 /* Setup L2 hash variables */
970 kdf(buf
, prf_key
, 2, sizeof(buf
)); /* Fill buffer with index 1 key */
971 for (i
= 0; i
< STREAMS
; i
++) {
972 /* Fill keys from the buffer, skipping bytes in the buffer not
973 * used by this implementation. Endian reverse the keys if on a
974 * little-endian computer.
976 memcpy(ahc
->poly_key_8
+i
, buf
+24*i
, 8);
977 endian_convert_if_le(ahc
->poly_key_8
+i
, 8, 8);
978 /* Mask the 64-bit keys to their special domain */
979 ahc
->poly_key_8
[i
] &= ((UINT64
)0x01ffffffu
<< 32) + 0x01ffffffu
;
980 ahc
->poly_accum
[i
] = 1; /* Our polyhash prepends a non-zero word */
983 /* Setup L3-1 hash variables */
984 kdf(buf
, prf_key
, 3, sizeof(buf
)); /* Fill buffer with index 2 key */
985 for (i
= 0; i
< STREAMS
; i
++)
986 memcpy(ahc
->ip_keys
+4*i
, buf
+(8*i
+4)*sizeof(UINT64
),
988 endian_convert_if_le(ahc
->ip_keys
, sizeof(UINT64
),
989 sizeof(ahc
->ip_keys
));
990 for (i
= 0; i
< STREAMS
*4; i
++)
991 ahc
->ip_keys
[i
] %= p36
; /* Bring into Z_p36 */
993 /* Setup L3-2 hash variables */
994 /* Fill buffer with index 4 key */
995 kdf(ahc
->ip_trans
, prf_key
, 4, STREAMS
* sizeof(UINT32
));
996 endian_convert_if_le(ahc
->ip_trans
, sizeof(UINT32
),
997 STREAMS
* sizeof(UINT32
));
998 explicit_bzero(buf
, sizeof(buf
));
1001 /* ---------------------------------------------------------------------- */
1004 static uhash_ctx_t
uhash_alloc(u_char key
[])
1006 /* Allocate memory and force to a 16-byte boundary. */
1008 u_char bytes_to_add
;
1009 aes_int_key prf_key
;
1011 ctx
= (uhash_ctx_t
)malloc(sizeof(uhash_ctx
)+ALLOC_BOUNDARY
);
1013 if (ALLOC_BOUNDARY
) {
1014 bytes_to_add
= ALLOC_BOUNDARY
-
1015 ((ptrdiff_t)ctx
& (ALLOC_BOUNDARY
-1));
1016 ctx
= (uhash_ctx_t
)((u_char
*)ctx
+ bytes_to_add
);
1017 *((u_char
*)ctx
- 1) = bytes_to_add
;
1019 aes_key_setup(key
,prf_key
);
1020 uhash_init(ctx
, prf_key
);
1026 /* ---------------------------------------------------------------------- */
1029 static int uhash_free(uhash_ctx_t ctx
)
1031 /* Free memory allocated by uhash_alloc */
1032 u_char bytes_to_sub
;
1035 if (ALLOC_BOUNDARY
) {
1036 bytes_to_sub
= *((u_char
*)ctx
- 1);
1037 ctx
= (uhash_ctx_t
)((u_char
*)ctx
- bytes_to_sub
);
1044 /* ---------------------------------------------------------------------- */
1046 static int uhash_update(uhash_ctx_t ctx
, const u_char
*input
, long len
)
1047 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1048 * hash each one with NH, calling the polyhash on each NH output.
1051 UWORD bytes_hashed
, bytes_remaining
;
1052 UINT64 result_buf
[STREAMS
];
1053 UINT8
*nh_result
= (UINT8
*)&result_buf
;
1055 if (ctx
->msg_len
+ len
<= L1_KEY_LEN
) {
1056 nh_update(&ctx
->hash
, (const UINT8
*)input
, len
);
1057 ctx
->msg_len
+= len
;
1060 bytes_hashed
= ctx
->msg_len
% L1_KEY_LEN
;
1061 if (ctx
->msg_len
== L1_KEY_LEN
)
1062 bytes_hashed
= L1_KEY_LEN
;
1064 if (bytes_hashed
+ len
>= L1_KEY_LEN
) {
1066 /* If some bytes have been passed to the hash function */
1067 /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1068 /* bytes to complete the current nh_block. */
1070 bytes_remaining
= (L1_KEY_LEN
- bytes_hashed
);
1071 nh_update(&ctx
->hash
, (const UINT8
*)input
, bytes_remaining
);
1072 nh_final(&ctx
->hash
, nh_result
);
1073 ctx
->msg_len
+= bytes_remaining
;
1074 poly_hash(ctx
,(UINT32
*)nh_result
);
1075 len
-= bytes_remaining
;
1076 input
+= bytes_remaining
;
1079 /* Hash directly from input stream if enough bytes */
1080 while (len
>= L1_KEY_LEN
) {
1081 nh(&ctx
->hash
, (const UINT8
*)input
, L1_KEY_LEN
,
1082 L1_KEY_LEN
, nh_result
);
1083 ctx
->msg_len
+= L1_KEY_LEN
;
1085 input
+= L1_KEY_LEN
;
1086 poly_hash(ctx
,(UINT32
*)nh_result
);
1090 /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1092 nh_update(&ctx
->hash
, (const UINT8
*)input
, len
);
1093 ctx
->msg_len
+= len
;
1100 /* ---------------------------------------------------------------------- */
1102 static int uhash_final(uhash_ctx_t ctx
, u_char
*res
)
1103 /* Incorporate any pending data, pad, and generate tag */
1105 UINT64 result_buf
[STREAMS
];
1106 UINT8
*nh_result
= (UINT8
*)&result_buf
;
1108 if (ctx
->msg_len
> L1_KEY_LEN
) {
1109 if (ctx
->msg_len
% L1_KEY_LEN
) {
1110 nh_final(&ctx
->hash
, nh_result
);
1111 poly_hash(ctx
,(UINT32
*)nh_result
);
1115 nh_final(&ctx
->hash
, nh_result
);
1116 ip_short(ctx
,nh_result
, res
);
1122 /* ---------------------------------------------------------------------- */
1125 static int uhash(uhash_ctx_t ahc
, u_char
*msg
, long len
, u_char
*res
)
1126 /* assumes that msg is in a writable buffer of length divisible by */
1127 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */
1129 UINT8 nh_result
[STREAMS
*sizeof(UINT64
)];
1131 int extra_zeroes_needed
;
1133 /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1136 if (len
<= L1_KEY_LEN
) {
1137 if (len
== 0) /* If zero length messages will not */
1138 nh_len
= L1_PAD_BOUNDARY
; /* be seen, comment out this case */
1140 nh_len
= ((len
+ (L1_PAD_BOUNDARY
- 1)) & ~(L1_PAD_BOUNDARY
- 1));
1141 extra_zeroes_needed
= nh_len
- len
;
1142 zero_pad((UINT8
*)msg
+ len
, extra_zeroes_needed
);
1143 nh(&ahc
->hash
, (UINT8
*)msg
, nh_len
, len
, nh_result
);
1144 ip_short(ahc
,nh_result
, res
);
1146 /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1147 * output to poly_hash().
1150 nh(&ahc
->hash
, (UINT8
*)msg
, L1_KEY_LEN
, L1_KEY_LEN
, nh_result
);
1151 poly_hash(ahc
,(UINT32
*)nh_result
);
1154 } while (len
>= L1_KEY_LEN
);
1156 nh_len
= ((len
+ (L1_PAD_BOUNDARY
- 1)) & ~(L1_PAD_BOUNDARY
- 1));
1157 extra_zeroes_needed
= nh_len
- len
;
1158 zero_pad((UINT8
*)msg
+ len
, extra_zeroes_needed
);
1159 nh(&ahc
->hash
, (UINT8
*)msg
, nh_len
, len
, nh_result
);
1160 poly_hash(ahc
,(UINT32
*)nh_result
);
1171 /* ---------------------------------------------------------------------- */
1172 /* ---------------------------------------------------------------------- */
1173 /* ----- Begin UMAC Section --------------------------------------------- */
1174 /* ---------------------------------------------------------------------- */
1175 /* ---------------------------------------------------------------------- */
1177 /* The UMAC interface has two interfaces, an all-at-once interface where
1178 * the entire message to be authenticated is passed to UMAC in one buffer,
1179 * and a sequential interface where the message is presented a little at a
1180 * time. The all-at-once is more optimaized than the sequential version and
1181 * should be preferred when the sequential interface is not required.
1184 uhash_ctx hash
; /* Hash function for message compression */
1185 pdf_ctx pdf
; /* PDF for hashed output */
1186 void *free_ptr
; /* Address to free this struct via */
1189 /* ---------------------------------------------------------------------- */
1192 int umac_reset(struct umac_ctx
*ctx
)
1193 /* Reset the hash function to begin a new authentication. */
1195 uhash_reset(&ctx
->hash
);
1200 /* ---------------------------------------------------------------------- */
1202 int umac_delete(struct umac_ctx
*ctx
)
1203 /* Deallocate the ctx structure */
1207 ctx
= (struct umac_ctx
*)ctx
->free_ptr
;
1208 freezero(ctx
, sizeof(*ctx
) + ALLOC_BOUNDARY
);
1213 /* ---------------------------------------------------------------------- */
1215 struct umac_ctx
*umac_new(const u_char key
[])
1216 /* Dynamically allocate a umac_ctx struct, initialize variables,
1217 * generate subkeys from key. Align to 16-byte boundary.
1220 struct umac_ctx
*ctx
, *octx
;
1221 size_t bytes_to_add
;
1222 aes_int_key prf_key
;
1224 octx
= ctx
= xcalloc(1, sizeof(*ctx
) + ALLOC_BOUNDARY
);
1226 if (ALLOC_BOUNDARY
) {
1227 bytes_to_add
= ALLOC_BOUNDARY
-
1228 ((ptrdiff_t)ctx
& (ALLOC_BOUNDARY
- 1));
1229 ctx
= (struct umac_ctx
*)((u_char
*)ctx
+ bytes_to_add
);
1231 ctx
->free_ptr
= octx
;
1232 aes_key_setup(key
, prf_key
);
1233 pdf_init(&ctx
->pdf
, prf_key
);
1234 uhash_init(&ctx
->hash
, prf_key
);
1235 explicit_bzero(prf_key
, sizeof(prf_key
));
1241 /* ---------------------------------------------------------------------- */
1243 int umac_final(struct umac_ctx
*ctx
, u_char tag
[], const u_char nonce
[8])
1244 /* Incorporate any pending data, pad, and generate tag */
1246 uhash_final(&ctx
->hash
, (u_char
*)tag
);
1247 pdf_gen_xor(&ctx
->pdf
, (const UINT8
*)nonce
, (UINT8
*)tag
);
1252 /* ---------------------------------------------------------------------- */
1254 int umac_update(struct umac_ctx
*ctx
, const u_char
*input
, long len
)
1255 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */
1256 /* hash each one, calling the PDF on the hashed output whenever the hash- */
1257 /* output buffer is full. */
1259 uhash_update(&ctx
->hash
, input
, len
);
1263 /* ---------------------------------------------------------------------- */
1266 int umac(struct umac_ctx
*ctx
, u_char
*input
,
1267 long len
, u_char tag
[],
1269 /* All-in-one version simply calls umac_update() and umac_final(). */
1271 uhash(&ctx
->hash
, input
, len
, (u_char
*)tag
);
1272 pdf_gen_xor(&ctx
->pdf
, (UINT8
*)nonce
, (UINT8
*)tag
);
1278 /* ---------------------------------------------------------------------- */
1279 /* ---------------------------------------------------------------------- */
1280 /* ----- End UMAC Section ----------------------------------------------- */
1281 /* ---------------------------------------------------------------------- */
1282 /* ---------------------------------------------------------------------- */