1 /* $NetBSD: umac.c,v 1.22 2023/07/26 17:58:16 christos Exp $ */
2 /* $OpenBSD: umac.c,v 1.23 2023/03/07 01:30:52 djm Exp $ */
3 /* -----------------------------------------------------------------------
4 *
5 * umac.c -- C Implementation UMAC Message Authentication
6 *
7 * Version 0.93b of rfc4418.txt -- 2006 July 18
8 *
9 * For a full description of UMAC message authentication see the UMAC
10 * world-wide-web page at http://www.cs.ucdavis.edu/~rogaway/umac
11 * Please report bugs and suggestions to the UMAC webpage.
12 *
13 * Copyright (c) 1999-2006 Ted Krovetz
14 *
15 * Permission to use, copy, modify, and distribute this software and
16 * its documentation for any purpose and with or without fee, is hereby
17 * granted provided that the above copyright notice appears in all copies
18 * and in supporting documentation, and that the name of the copyright
19 * holder not be used in advertising or publicity pertaining to
20 * distribution of the software without specific, written prior permission.
21 *
22 * Comments should be directed to Ted Krovetz (tdk@acm.org)
23 *
24 * ---------------------------------------------------------------------- */
25
26 /* ////////////////////// IMPORTANT NOTES /////////////////////////////////
27 *
28 * 1) This version does not work properly on messages larger than 16MB
29 *
30 * 2) If you set the switch to use SSE2, then all data must be 16-byte
31 * aligned
32 *
33 * 3) When calling the function umac(), it is assumed that msg is in
34 * a writable buffer of length divisible by 32 bytes. The message itself
35 * does not have to fill the entire buffer, but bytes beyond msg may be
36 * zeroed.
37 *
38 * 4) Three free AES implementations are supported by this implementation of
39 * UMAC. Paulo Barreto's version is in the public domain and can be found
40 * at http://www.esat.kuleuven.ac.be/~rijmen/rijndael/ (search for
41 * "Barreto"). The only two files needed are rijndael-alg-fst.c and
42 * rijndael-alg-fst.h. Brian Gladman's version is distributed with the GNU
43 * Public license at http://fp.gladman.plus.com/AES/index.htm. It
44 * includes a fast IA-32 assembly version. The OpenSSL crypo library is
45 * the third.
46 *
47 * 5) With FORCE_C_ONLY flags set to 0, incorrect results are sometimes
48 * produced under gcc with optimizations set -O3 or higher. Dunno why.
49 *
50 /////////////////////////////////////////////////////////////////////// */
51
52 /* ---------------------------------------------------------------------- */
53 /* --- User Switches ---------------------------------------------------- */
54 /* ---------------------------------------------------------------------- */
55
56 #ifndef UMAC_OUTPUT_LEN
57 #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */
58 #endif
59 /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */
60 /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */
61 /* #define SSE2 0 Is SSE2 is available? */
62 /* #define RUN_TESTS 0 Run basic correctness/speed tests */
63 /* #define UMAC_AE_SUPPORT 0 Enable authenticated encryption */
64
65 /* ---------------------------------------------------------------------- */
66 /* -- Global Includes --------------------------------------------------- */
67 /* ---------------------------------------------------------------------- */
68
69 #include "includes.h"
70 __RCSID("$NetBSD: umac.c,v 1.22 2023/07/26 17:58:16 christos Exp $");
71 #include <sys/types.h>
72 #include <sys/endian.h>
73 #include <string.h>
74 #include <stdarg.h>
75 #include <stdio.h>
76 #include <stdlib.h>
77 #include <stddef.h>
78 #include <time.h>
79
80 #include "xmalloc.h"
81 #include "umac.h"
82 #include "misc.h"
83
84 /* ---------------------------------------------------------------------- */
85 /* --- Primitive Data Types --- */
86 /* ---------------------------------------------------------------------- */
87
88 /* The following assumptions may need change on your system */
89 typedef u_int8_t UINT8; /* 1 byte */
90 typedef u_int16_t UINT16; /* 2 byte */
91 typedef u_int32_t UINT32; /* 4 byte */
92 typedef u_int64_t UINT64; /* 8 bytes */
93 typedef unsigned int UWORD; /* Register */
94
95 /* ---------------------------------------------------------------------- */
96 /* --- Constants -------------------------------------------------------- */
97 /* ---------------------------------------------------------------------- */
98
99 #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */
100
101 /* Message "words" are read from memory in an endian-specific manner. */
102 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */
103 /* be set true if the host computer is little-endian. */
104
105 #if BYTE_ORDER == LITTLE_ENDIAN
106 #define __LITTLE_ENDIAN__ 1
107 #else
108 #define __LITTLE_ENDIAN__ 0
109 #endif
110
111 /* ---------------------------------------------------------------------- */
112 /* ---------------------------------------------------------------------- */
113 /* ----- Architecture Specific ------------------------------------------ */
114 /* ---------------------------------------------------------------------- */
115 /* ---------------------------------------------------------------------- */
116
117
118 /* ---------------------------------------------------------------------- */
119 /* ---------------------------------------------------------------------- */
120 /* ----- Primitive Routines --------------------------------------------- */
121 /* ---------------------------------------------------------------------- */
122 /* ---------------------------------------------------------------------- */
123
124
125 /* ---------------------------------------------------------------------- */
126 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
127 /* ---------------------------------------------------------------------- */
128
129 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
130
131 /* ---------------------------------------------------------------------- */
132 /* --- Endian Conversion --- Forcing assembly on some platforms */
133 /* ---------------------------------------------------------------------- */
134
135 /* The following definitions use the above reversal-primitives to do the right
136 * thing on endian specific load and stores.
137 */
138
139 #if BYTE_ORDER == LITTLE_ENDIAN
140 #define LOAD_UINT32_REVERSED(p) get_u32(p)
141 #define STORE_UINT32_REVERSED(p,v) put_u32(p,v)
142 #else
143 #define LOAD_UINT32_REVERSED(p) get_u32_le(p)
144 #define STORE_UINT32_REVERSED(p,v) put_u32_le(p,v)
145 #endif
146
147 #define LOAD_UINT32_LITTLE(p) (get_u32_le(p))
148 #define STORE_UINT32_BIG(p,v) put_u32(p, v)
149
150
151
152 /* ---------------------------------------------------------------------- */
153 /* ---------------------------------------------------------------------- */
154 /* ----- Begin KDF & PDF Section ---------------------------------------- */
155 /* ---------------------------------------------------------------------- */
156 /* ---------------------------------------------------------------------- */
157
158 /* UMAC uses AES with 16 byte block and key lengths */
159 #define AES_BLOCK_LEN 16
160
161 #ifdef WITH_OPENSSL
162 #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)
168 #else
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), \
176 UMAC_KEY_LEN*8)
177 #endif
178
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 'buffer_ptr'. Each distinct
183 * 'ndx' causes a distinct byte stream.
184 */
kdf(void * buffer_ptr,aes_int_key key,UINT8 ndx,int nbytes)185 static void kdf(void *buffer_ptr, aes_int_key key, UINT8 ndx, int nbytes)
186 {
187 UINT8 in_buf[AES_BLOCK_LEN] = {0};
188 UINT8 out_buf[AES_BLOCK_LEN];
189 UINT8 *dst_buf = (UINT8 *)buffer_ptr;
190 int i;
191
192 /* Setup the initial value */
193 in_buf[AES_BLOCK_LEN-9] = ndx;
194 in_buf[AES_BLOCK_LEN-1] = i = 1;
195
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;
202 }
203 if (nbytes) {
204 aes_encryption(in_buf, out_buf, key);
205 memcpy(dst_buf,out_buf,nbytes);
206 }
207 explicit_bzero(in_buf, sizeof(in_buf));
208 explicit_bzero(out_buf, sizeof(out_buf));
209 }
210
211 /* The final UHASH result is XOR'd with the output of a pseudorandom
212 * function. Here, we use AES to generate random output and
213 * xor the appropriate bytes depending on the last bits of nonce.
214 * This scheme is optimized for sequential, increasing big-endian nonces.
215 */
216
217 typedef struct {
218 UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */
219 UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */
220 aes_int_key prf_key; /* Expanded AES key for PDF */
221 } pdf_ctx;
222
pdf_init(pdf_ctx * pc,aes_int_key prf_key)223 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
224 {
225 UINT8 buf[UMAC_KEY_LEN];
226
227 kdf(buf, prf_key, 0, UMAC_KEY_LEN);
228 aes_key_setup(buf, pc->prf_key);
229
230 /* Initialize pdf and cache */
231 memset(pc->nonce, 0, sizeof(pc->nonce));
232 aes_encryption(pc->nonce, pc->cache, pc->prf_key);
233 explicit_bzero(buf, sizeof(buf));
234 }
235
236 static inline void
xor64(uint8_t * dp,int di,uint8_t * sp,int si)237 xor64(uint8_t *dp, int di, uint8_t *sp, int si)
238 {
239 uint64_t dst, src;
240 memcpy(&dst, dp + sizeof(dst) * di, sizeof(dst));
241 memcpy(&src, sp + sizeof(src) * si, sizeof(src));
242 dst ^= src;
243 memcpy(dp + sizeof(dst) * di, &dst, sizeof(dst));
244 }
245
246 __unused static inline void
xor32(uint8_t * dp,int di,uint8_t * sp,int si)247 xor32(uint8_t *dp, int di, uint8_t *sp, int si)
248 {
249 uint32_t dst, src;
250 memcpy(&dst, dp + sizeof(dst) * di, sizeof(dst));
251 memcpy(&src, sp + sizeof(src) * si, sizeof(src));
252 dst ^= src;
253 memcpy(dp + sizeof(dst) * di, &dst, sizeof(dst));
254 }
255
pdf_gen_xor(pdf_ctx * pc,const UINT8 nonce[8],UINT8 buf[UMAC_OUTPUT_LEN])256 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8],
257 UINT8 buf[UMAC_OUTPUT_LEN])
258 {
259 /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
260 * of the AES output. If last time around we returned the ndx-1st
261 * element, then we may have the result in the cache already.
262 */
263
264 #if (UMAC_OUTPUT_LEN == 4)
265 #define LOW_BIT_MASK 3
266 #elif (UMAC_OUTPUT_LEN == 8)
267 #define LOW_BIT_MASK 1
268 #elif (UMAC_OUTPUT_LEN > 8)
269 #define LOW_BIT_MASK 0
270 #endif
271 union {
272 UINT8 tmp_nonce_lo[4];
273 UINT32 align;
274 } t;
275 #if LOW_BIT_MASK != 0
276 int ndx = nonce[7] & LOW_BIT_MASK;
277 #endif
278 memcpy(t.tmp_nonce_lo, nonce + 4, sizeof(t.tmp_nonce_lo));
279 t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
280
281 if (memcmp(t.tmp_nonce_lo, pc->nonce + 1, sizeof(t.tmp_nonce_lo)) != 0 ||
282 memcmp(nonce, pc->nonce, sizeof(t.tmp_nonce_lo)) != 0)
283 {
284 memcpy(pc->nonce, nonce, sizeof(t.tmp_nonce_lo));
285 memcpy(pc->nonce + 4, t.tmp_nonce_lo, sizeof(t.tmp_nonce_lo));
286 aes_encryption(pc->nonce, pc->cache, pc->prf_key);
287 }
288
289 #if (UMAC_OUTPUT_LEN == 4)
290 xor32(buf, 0, pc->cache, ndx);
291 #elif (UMAC_OUTPUT_LEN == 8)
292 xor64(buf, 0, pc->cache, ndx);
293 #elif (UMAC_OUTPUT_LEN == 12)
294 xor64(buf, 0, pc->cache, 0);
295 xor32(buf, 2, pc->cache, 2);
296 #elif (UMAC_OUTPUT_LEN == 16)
297 xor64(buf, 0, pc->cache, 0);
298 xor64(buf, 1, pc->cache, 1);
299 #endif
300 }
301
302 /* ---------------------------------------------------------------------- */
303 /* ---------------------------------------------------------------------- */
304 /* ----- Begin NH Hash Section ------------------------------------------ */
305 /* ---------------------------------------------------------------------- */
306 /* ---------------------------------------------------------------------- */
307
308 /* The NH-based hash functions used in UMAC are described in the UMAC paper
309 * and specification, both of which can be found at the UMAC website.
310 * The interface to this implementation has two
311 * versions, one expects the entire message being hashed to be passed
312 * in a single buffer and returns the hash result immediately. The second
313 * allows the message to be passed in a sequence of buffers. In the
314 * multiple-buffer interface, the client calls the routine nh_update() as
315 * many times as necessary. When there is no more data to be fed to the
316 * hash, the client calls nh_final() which calculates the hash output.
317 * Before beginning another hash calculation the nh_reset() routine
318 * must be called. The single-buffer routine, nh(), is equivalent to
319 * the sequence of calls nh_update() and nh_final(); however it is
320 * optimized and should be preferred whenever the multiple-buffer interface
321 * is not necessary. When using either interface, it is the client's
322 * responsibility to pass no more than L1_KEY_LEN bytes per hash result.
323 *
324 * The routine nh_init() initializes the nh_ctx data structure and
325 * must be called once, before any other PDF routine.
326 */
327
328 /* The "nh_aux" routines do the actual NH hashing work. They
329 * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
330 * produce output for all STREAMS NH iterations in one call,
331 * allowing the parallel implementation of the streams.
332 */
333
334 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */
335 #define L1_KEY_LEN 1024 /* Internal key bytes */
336 #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */
337 #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */
338 #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */
339 #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */
340
341 typedef struct {
342 UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
343 UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */
344 int next_data_empty; /* Bookkeeping variable for data buffer. */
345 int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorporated. */
346 UINT64 state[STREAMS]; /* on-line state */
347 } nh_ctx;
348
349
350 #if (UMAC_OUTPUT_LEN == 4)
351
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)352 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
353 /* NH hashing primitive. Previous (partial) hash result is loaded and
354 * then stored via hp pointer. The length of the data pointed at by "dp",
355 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key
356 * is expected to be endian compensated in memory at key setup.
357 */
358 {
359 UINT64 h;
360 UWORD c = dlen / 32;
361 UINT32 *k = (UINT32 *)kp;
362 const UINT32 *d = (const UINT32 *)dp;
363 UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
364 UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
365
366 h = *((UINT64 *)hp);
367 do {
368 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
369 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
370 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
371 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
372 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
373 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
374 h += MUL64((k0 + d0), (k4 + d4));
375 h += MUL64((k1 + d1), (k5 + d5));
376 h += MUL64((k2 + d2), (k6 + d6));
377 h += MUL64((k3 + d3), (k7 + d7));
378
379 d += 8;
380 k += 8;
381 } while (--c);
382 *((UINT64 *)hp) = h;
383 }
384
385 #elif (UMAC_OUTPUT_LEN == 8)
386
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)387 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
388 /* Same as previous nh_aux, but two streams are handled in one pass,
389 * reading and writing 16 bytes of hash-state per call.
390 */
391 {
392 UINT64 h1,h2;
393 UWORD c = dlen / 32;
394 UINT32 *k = (UINT32 *)kp;
395 const UINT32 *d = (const UINT32 *)dp;
396 UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
397 UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
398 k8,k9,k10,k11;
399
400 h1 = *((UINT64 *)hp);
401 h2 = *((UINT64 *)hp + 1);
402 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
403 do {
404 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
405 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
406 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
407 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
408 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
409 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
410
411 h1 += MUL64((k0 + d0), (k4 + d4));
412 h2 += MUL64((k4 + d0), (k8 + d4));
413
414 h1 += MUL64((k1 + d1), (k5 + d5));
415 h2 += MUL64((k5 + d1), (k9 + d5));
416
417 h1 += MUL64((k2 + d2), (k6 + d6));
418 h2 += MUL64((k6 + d2), (k10 + d6));
419
420 h1 += MUL64((k3 + d3), (k7 + d7));
421 h2 += MUL64((k7 + d3), (k11 + d7));
422
423 k0 = k8; k1 = k9; k2 = k10; k3 = k11;
424
425 d += 8;
426 k += 8;
427 } while (--c);
428 ((UINT64 *)hp)[0] = h1;
429 ((UINT64 *)hp)[1] = h2;
430 }
431
432 #elif (UMAC_OUTPUT_LEN == 12)
433
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)434 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
435 /* Same as previous nh_aux, but two streams are handled in one pass,
436 * reading and writing 24 bytes of hash-state per call.
437 */
438 {
439 UINT64 h1,h2,h3;
440 UWORD c = dlen / 32;
441 UINT32 *k = (UINT32 *)kp;
442 const UINT32 *d = (const UINT32 *)dp;
443 UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
444 UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
445 k8,k9,k10,k11,k12,k13,k14,k15;
446
447 h1 = *((UINT64 *)hp);
448 h2 = *((UINT64 *)hp + 1);
449 h3 = *((UINT64 *)hp + 2);
450 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
451 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
452 do {
453 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
454 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
455 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
456 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
457 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
458 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
459
460 h1 += MUL64((k0 + d0), (k4 + d4));
461 h2 += MUL64((k4 + d0), (k8 + d4));
462 h3 += MUL64((k8 + d0), (k12 + d4));
463
464 h1 += MUL64((k1 + d1), (k5 + d5));
465 h2 += MUL64((k5 + d1), (k9 + d5));
466 h3 += MUL64((k9 + d1), (k13 + d5));
467
468 h1 += MUL64((k2 + d2), (k6 + d6));
469 h2 += MUL64((k6 + d2), (k10 + d6));
470 h3 += MUL64((k10 + d2), (k14 + d6));
471
472 h1 += MUL64((k3 + d3), (k7 + d7));
473 h2 += MUL64((k7 + d3), (k11 + d7));
474 h3 += MUL64((k11 + d3), (k15 + d7));
475
476 k0 = k8; k1 = k9; k2 = k10; k3 = k11;
477 k4 = k12; k5 = k13; k6 = k14; k7 = k15;
478
479 d += 8;
480 k += 8;
481 } while (--c);
482 ((UINT64 *)hp)[0] = h1;
483 ((UINT64 *)hp)[1] = h2;
484 ((UINT64 *)hp)[2] = h3;
485 }
486
487 #elif (UMAC_OUTPUT_LEN == 16)
488
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)489 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
490 /* Same as previous nh_aux, but two streams are handled in one pass,
491 * reading and writing 24 bytes of hash-state per call.
492 */
493 {
494 UINT64 h1,h2,h3,h4;
495 UWORD c = dlen / 32;
496 UINT32 *k = (UINT32 *)kp;
497 const UINT32 *d = (const UINT32 *)dp;
498 UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
499 UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
500 k8,k9,k10,k11,k12,k13,k14,k15,
501 k16,k17,k18,k19;
502
503 h1 = *((UINT64 *)hp);
504 h2 = *((UINT64 *)hp + 1);
505 h3 = *((UINT64 *)hp + 2);
506 h4 = *((UINT64 *)hp + 3);
507 k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
508 k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
509 do {
510 d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
511 d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
512 d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
513 d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
514 k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
515 k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
516 k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
517
518 h1 += MUL64((k0 + d0), (k4 + d4));
519 h2 += MUL64((k4 + d0), (k8 + d4));
520 h3 += MUL64((k8 + d0), (k12 + d4));
521 h4 += MUL64((k12 + d0), (k16 + d4));
522
523 h1 += MUL64((k1 + d1), (k5 + d5));
524 h2 += MUL64((k5 + d1), (k9 + d5));
525 h3 += MUL64((k9 + d1), (k13 + d5));
526 h4 += MUL64((k13 + d1), (k17 + d5));
527
528 h1 += MUL64((k2 + d2), (k6 + d6));
529 h2 += MUL64((k6 + d2), (k10 + d6));
530 h3 += MUL64((k10 + d2), (k14 + d6));
531 h4 += MUL64((k14 + d2), (k18 + d6));
532
533 h1 += MUL64((k3 + d3), (k7 + d7));
534 h2 += MUL64((k7 + d3), (k11 + d7));
535 h3 += MUL64((k11 + d3), (k15 + d7));
536 h4 += MUL64((k15 + d3), (k19 + d7));
537
538 k0 = k8; k1 = k9; k2 = k10; k3 = k11;
539 k4 = k12; k5 = k13; k6 = k14; k7 = k15;
540 k8 = k16; k9 = k17; k10 = k18; k11 = k19;
541
542 d += 8;
543 k += 8;
544 } while (--c);
545 ((UINT64 *)hp)[0] = h1;
546 ((UINT64 *)hp)[1] = h2;
547 ((UINT64 *)hp)[2] = h3;
548 ((UINT64 *)hp)[3] = h4;
549 }
550
551 /* ---------------------------------------------------------------------- */
552 #endif /* UMAC_OUTPUT_LENGTH */
553 /* ---------------------------------------------------------------------- */
554
555
556 /* ---------------------------------------------------------------------- */
557
nh_transform(nh_ctx * hc,const UINT8 * buf,UINT32 nbytes)558 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
559 /* This function is a wrapper for the primitive NH hash functions. It takes
560 * as argument "hc" the current hash context and a buffer which must be a
561 * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
562 * appropriately according to how much message has been hashed already.
563 */
564 {
565 UINT8 *key;
566
567 key = hc->nh_key + hc->bytes_hashed;
568 nh_aux(key, buf, hc->state, nbytes);
569 }
570
571 /* ---------------------------------------------------------------------- */
572
573 #if (__LITTLE_ENDIAN__)
endian_convert(void * buf,UWORD bpw,UINT32 num_bytes)574 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
575 /* We endian convert the keys on little-endian computers to */
576 /* compensate for the lack of big-endian memory reads during hashing. */
577 {
578 UWORD iters = num_bytes / bpw;
579 if (bpw == 4) {
580 UINT32 *p = (UINT32 *)buf;
581 do {
582 *p = LOAD_UINT32_REVERSED(p);
583 p++;
584 } while (--iters);
585 } else if (bpw == 8) {
586 UINT64 *p = (UINT64 *)buf;
587 UINT64 th;
588 UINT64 t;
589 do {
590 t = LOAD_UINT32_REVERSED((UINT32 *)p+1);
591 th = LOAD_UINT32_REVERSED((UINT32 *)p);
592 *p++ = t | (th << 32);
593 } while (--iters);
594 }
595 }
596 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
597 #else
598 #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */
599 #endif
600
601 /* ---------------------------------------------------------------------- */
602
nh_reset(nh_ctx * hc)603 static void nh_reset(nh_ctx *hc)
604 /* Reset nh_ctx to ready for hashing of new data */
605 {
606 hc->bytes_hashed = 0;
607 hc->next_data_empty = 0;
608 hc->state[0] = 0;
609 #if (UMAC_OUTPUT_LEN >= 8)
610 hc->state[1] = 0;
611 #endif
612 #if (UMAC_OUTPUT_LEN >= 12)
613 hc->state[2] = 0;
614 #endif
615 #if (UMAC_OUTPUT_LEN == 16)
616 hc->state[3] = 0;
617 #endif
618
619 }
620
621 /* ---------------------------------------------------------------------- */
622
nh_init(nh_ctx * hc,aes_int_key prf_key)623 static void nh_init(nh_ctx *hc, aes_int_key prf_key)
624 /* Generate nh_key, endian convert and reset to be ready for hashing. */
625 {
626 kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
627 endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
628 nh_reset(hc);
629 }
630
631 /* ---------------------------------------------------------------------- */
632
nh_update(nh_ctx * hc,const UINT8 * buf,UINT32 nbytes)633 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
634 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */
635 /* even multiple of HASH_BUF_BYTES. */
636 {
637 UINT32 i,j;
638
639 j = hc->next_data_empty;
640 if ((j + nbytes) >= HASH_BUF_BYTES) {
641 if (j) {
642 i = HASH_BUF_BYTES - j;
643 memcpy(hc->data+j, buf, i);
644 nh_transform(hc,hc->data,HASH_BUF_BYTES);
645 nbytes -= i;
646 buf += i;
647 hc->bytes_hashed += HASH_BUF_BYTES;
648 }
649 if (nbytes >= HASH_BUF_BYTES) {
650 i = nbytes & ~(HASH_BUF_BYTES - 1);
651 nh_transform(hc, buf, i);
652 nbytes -= i;
653 buf += i;
654 hc->bytes_hashed += i;
655 }
656 j = 0;
657 }
658 memcpy(hc->data + j, buf, nbytes);
659 hc->next_data_empty = j + nbytes;
660 }
661
662 /* ---------------------------------------------------------------------- */
663
zero_pad(UINT8 * p,int nbytes)664 static void zero_pad(UINT8 *p, int nbytes)
665 {
666 /* Write "nbytes" of zeroes, beginning at "p" */
667 if (nbytes >= (int)sizeof(UWORD)) {
668 while ((ptrdiff_t)p % sizeof(UWORD)) {
669 *p = 0;
670 nbytes--;
671 p++;
672 }
673 while (nbytes >= (int)sizeof(UWORD)) {
674 *(UWORD *)p = 0;
675 nbytes -= sizeof(UWORD);
676 p += sizeof(UWORD);
677 }
678 }
679 while (nbytes) {
680 *p = 0;
681 nbytes--;
682 p++;
683 }
684 }
685
686 /* ---------------------------------------------------------------------- */
687
nh_final(nh_ctx * hc,UINT8 * result)688 static void nh_final(nh_ctx *hc, UINT8 *result)
689 /* After passing some number of data buffers to nh_update() for integration
690 * into an NH context, nh_final is called to produce a hash result. If any
691 * bytes are in the buffer hc->data, incorporate them into the
692 * NH context. Finally, add into the NH accumulation "state" the total number
693 * of bits hashed. The resulting numbers are written to the buffer "result".
694 * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
695 */
696 {
697 int nh_len, nbits;
698
699 if (hc->next_data_empty != 0) {
700 nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
701 ~(L1_PAD_BOUNDARY - 1));
702 zero_pad(hc->data + hc->next_data_empty,
703 nh_len - hc->next_data_empty);
704 nh_transform(hc, hc->data, nh_len);
705 hc->bytes_hashed += hc->next_data_empty;
706 } else if (hc->bytes_hashed == 0) {
707 nh_len = L1_PAD_BOUNDARY;
708 zero_pad(hc->data, L1_PAD_BOUNDARY);
709 nh_transform(hc, hc->data, nh_len);
710 }
711
712 nbits = (hc->bytes_hashed << 3);
713 ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
714 #if (UMAC_OUTPUT_LEN >= 8)
715 ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
716 #endif
717 #if (UMAC_OUTPUT_LEN >= 12)
718 ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
719 #endif
720 #if (UMAC_OUTPUT_LEN == 16)
721 ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
722 #endif
723 nh_reset(hc);
724 }
725
726 /* ---------------------------------------------------------------------- */
727
nh(nh_ctx * hc,const UINT8 * buf,UINT32 padded_len,UINT32 unpadded_len,UINT8 * result)728 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
729 UINT32 unpadded_len, UINT8 *result)
730 /* All-in-one nh_update() and nh_final() equivalent.
731 * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
732 * well aligned
733 */
734 {
735 UINT32 nbits;
736
737 /* Initialize the hash state */
738 nbits = (unpadded_len << 3);
739
740 ((UINT64 *)result)[0] = nbits;
741 #if (UMAC_OUTPUT_LEN >= 8)
742 ((UINT64 *)result)[1] = nbits;
743 #endif
744 #if (UMAC_OUTPUT_LEN >= 12)
745 ((UINT64 *)result)[2] = nbits;
746 #endif
747 #if (UMAC_OUTPUT_LEN == 16)
748 ((UINT64 *)result)[3] = nbits;
749 #endif
750
751 nh_aux(hc->nh_key, buf, result, padded_len);
752 }
753
754 /* ---------------------------------------------------------------------- */
755 /* ---------------------------------------------------------------------- */
756 /* ----- Begin UHASH Section -------------------------------------------- */
757 /* ---------------------------------------------------------------------- */
758 /* ---------------------------------------------------------------------- */
759
760 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
761 * hashed by NH. The NH output is then hashed by a polynomial-hash layer
762 * unless the initial data to be hashed is short. After the polynomial-
763 * layer, an inner-product hash is used to produce the final UHASH output.
764 *
765 * UHASH provides two interfaces, one all-at-once and another where data
766 * buffers are presented sequentially. In the sequential interface, the
767 * UHASH client calls the routine uhash_update() as many times as necessary.
768 * When there is no more data to be fed to UHASH, the client calls
769 * uhash_final() which
770 * calculates the UHASH output. Before beginning another UHASH calculation
771 * the uhash_reset() routine must be called. The all-at-once UHASH routine,
772 * uhash(), is equivalent to the sequence of calls uhash_update() and
773 * uhash_final(); however it is optimized and should be
774 * used whenever the sequential interface is not necessary.
775 *
776 * The routine uhash_init() initializes the uhash_ctx data structure and
777 * must be called once, before any other UHASH routine.
778 */
779
780 /* ---------------------------------------------------------------------- */
781 /* ----- Constants and uhash_ctx ---------------------------------------- */
782 /* ---------------------------------------------------------------------- */
783
784 /* ---------------------------------------------------------------------- */
785 /* ----- Poly hash and Inner-Product hash Constants --------------------- */
786 /* ---------------------------------------------------------------------- */
787
788 /* Primes and masks */
789 #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */
790 #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */
791 #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */
792
793
794 /* ---------------------------------------------------------------------- */
795
796 typedef struct uhash_ctx {
797 nh_ctx hash; /* Hash context for L1 NH hash */
798 UINT64 poly_key_8[STREAMS]; /* p64 poly keys */
799 UINT64 poly_accum[STREAMS]; /* poly hash result */
800 UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */
801 UINT32 ip_trans[STREAMS]; /* Inner-product translation */
802 UINT32 msg_len; /* Total length of data passed */
803 /* to uhash */
804 } uhash_ctx;
805 typedef struct uhash_ctx *uhash_ctx_t;
806
807 /* ---------------------------------------------------------------------- */
808
809
810 /* The polynomial hashes use Horner's rule to evaluate a polynomial one
811 * word at a time. As described in the specification, poly32 and poly64
812 * require keys from special domains. The following implementations exploit
813 * the special domains to avoid overflow. The results are not guaranteed to
814 * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
815 * patches any errant values.
816 */
817
poly64(UINT64 cur,UINT64 key,UINT64 data)818 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
819 {
820 UINT32 key_hi = (UINT32)(key >> 32),
821 key_lo = (UINT32)key,
822 cur_hi = (UINT32)(cur >> 32),
823 cur_lo = (UINT32)cur,
824 x_lo,
825 x_hi;
826 UINT64 X,T,res;
827
828 X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
829 x_lo = (UINT32)X;
830 x_hi = (UINT32)(X >> 32);
831
832 res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
833
834 T = ((UINT64)x_lo << 32);
835 res += T;
836 if (res < T)
837 res += 59;
838
839 res += data;
840 if (res < data)
841 res += 59;
842
843 return res;
844 }
845
846
847 /* Although UMAC is specified to use a ramped polynomial hash scheme, this
848 * implementation does not handle all ramp levels. Because we don't handle
849 * the ramp up to p128 modulus in this implementation, we are limited to
850 * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
851 * bytes input to UMAC per tag, ie. 16MB).
852 */
poly_hash(uhash_ctx_t hc,UINT32 data_in[])853 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
854 {
855 int i;
856 UINT64 *data=(UINT64*)data_in;
857
858 for (i = 0; i < STREAMS; i++) {
859 if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
860 hc->poly_accum[i] = poly64(hc->poly_accum[i],
861 hc->poly_key_8[i], p64 - 1);
862 hc->poly_accum[i] = poly64(hc->poly_accum[i],
863 hc->poly_key_8[i], (data[i] - 59));
864 } else {
865 hc->poly_accum[i] = poly64(hc->poly_accum[i],
866 hc->poly_key_8[i], data[i]);
867 }
868 }
869 }
870
871
872 /* ---------------------------------------------------------------------- */
873
874
875 /* The final step in UHASH is an inner-product hash. The poly hash
876 * produces a result not necessarily WORD_LEN bytes long. The inner-
877 * product hash breaks the polyhash output into 16-bit chunks and
878 * multiplies each with a 36 bit key.
879 */
880
ip_aux(UINT64 t,UINT64 * ipkp,UINT64 data)881 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
882 {
883 t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
884 t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
885 t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
886 t = t + ipkp[3] * (UINT64)(UINT16)(data);
887
888 return t;
889 }
890
ip_reduce_p36(UINT64 t)891 static UINT32 ip_reduce_p36(UINT64 t)
892 {
893 /* Divisionless modular reduction */
894 UINT64 ret;
895
896 ret = (t & m36) + 5 * (t >> 36);
897 if (ret >= p36)
898 ret -= p36;
899
900 /* return least significant 32 bits */
901 return (UINT32)(ret);
902 }
903
904
905 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
906 * the polyhash stage is skipped and ip_short is applied directly to the
907 * NH output.
908 */
ip_short(uhash_ctx_t ahc,UINT8 * nh_res,u_char * res)909 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
910 {
911 UINT64 t;
912 UINT64 *nhp = (UINT64 *)nh_res;
913
914 t = ip_aux(0,ahc->ip_keys, nhp[0]);
915 STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
916 #if (UMAC_OUTPUT_LEN >= 8)
917 t = ip_aux(0,ahc->ip_keys+4, nhp[1]);
918 STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
919 #endif
920 #if (UMAC_OUTPUT_LEN >= 12)
921 t = ip_aux(0,ahc->ip_keys+8, nhp[2]);
922 STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
923 #endif
924 #if (UMAC_OUTPUT_LEN == 16)
925 t = ip_aux(0,ahc->ip_keys+12, nhp[3]);
926 STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
927 #endif
928 }
929
930 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
931 * the polyhash stage is not skipped and ip_long is applied to the
932 * polyhash output.
933 */
ip_long(uhash_ctx_t ahc,u_char * res)934 static void ip_long(uhash_ctx_t ahc, u_char *res)
935 {
936 int i;
937 UINT64 t;
938
939 for (i = 0; i < STREAMS; i++) {
940 /* fix polyhash output not in Z_p64 */
941 if (ahc->poly_accum[i] >= p64)
942 ahc->poly_accum[i] -= p64;
943 t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
944 STORE_UINT32_BIG((UINT32 *)res+i,
945 ip_reduce_p36(t) ^ ahc->ip_trans[i]);
946 }
947 }
948
949
950 /* ---------------------------------------------------------------------- */
951
952 /* ---------------------------------------------------------------------- */
953
954 /* Reset uhash context for next hash session */
uhash_reset(uhash_ctx_t pc)955 static int uhash_reset(uhash_ctx_t pc)
956 {
957 nh_reset(&pc->hash);
958 pc->msg_len = 0;
959 pc->poly_accum[0] = 1;
960 #if (UMAC_OUTPUT_LEN >= 8)
961 pc->poly_accum[1] = 1;
962 #endif
963 #if (UMAC_OUTPUT_LEN >= 12)
964 pc->poly_accum[2] = 1;
965 #endif
966 #if (UMAC_OUTPUT_LEN == 16)
967 pc->poly_accum[3] = 1;
968 #endif
969 return 1;
970 }
971
972 /* ---------------------------------------------------------------------- */
973
974 /* Given a pointer to the internal key needed by kdf() and a uhash context,
975 * initialize the NH context and generate keys needed for poly and inner-
976 * product hashing. All keys are endian adjusted in memory so that native
977 * loads cause correct keys to be in registers during calculation.
978 */
uhash_init(uhash_ctx_t ahc,aes_int_key prf_key)979 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
980 {
981 int i;
982 UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
983
984 /* Zero the entire uhash context */
985 memset(ahc, 0, sizeof(uhash_ctx));
986
987 /* Initialize the L1 hash */
988 nh_init(&ahc->hash, prf_key);
989
990 /* Setup L2 hash variables */
991 kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */
992 for (i = 0; i < STREAMS; i++) {
993 /* Fill keys from the buffer, skipping bytes in the buffer not
994 * used by this implementation. Endian reverse the keys if on a
995 * little-endian computer.
996 */
997 memcpy(ahc->poly_key_8+i, buf+24*i, 8);
998 endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
999 /* Mask the 64-bit keys to their special domain */
1000 ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
1001 ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */
1002 }
1003
1004 /* Setup L3-1 hash variables */
1005 kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
1006 for (i = 0; i < STREAMS; i++)
1007 memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
1008 4*sizeof(UINT64));
1009 endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
1010 sizeof(ahc->ip_keys));
1011 for (i = 0; i < STREAMS*4; i++)
1012 ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */
1013
1014 /* Setup L3-2 hash variables */
1015 /* Fill buffer with index 4 key */
1016 kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
1017 endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
1018 STREAMS * sizeof(UINT32));
1019 explicit_bzero(buf, sizeof(buf));
1020 }
1021
1022 /* ---------------------------------------------------------------------- */
1023
1024 #if 0
1025 static uhash_ctx_t uhash_alloc(u_char key[])
1026 {
1027 /* Allocate memory and force to a 16-byte boundary. */
1028 uhash_ctx_t ctx;
1029 u_char bytes_to_add;
1030 aes_int_key prf_key;
1031
1032 ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1033 if (ctx) {
1034 if (ALLOC_BOUNDARY) {
1035 bytes_to_add = ALLOC_BOUNDARY -
1036 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1037 ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1038 *((u_char *)ctx - 1) = bytes_to_add;
1039 }
1040 aes_key_setup(key,prf_key);
1041 uhash_init(ctx, prf_key);
1042 }
1043 return (ctx);
1044 }
1045 #endif
1046
1047 /* ---------------------------------------------------------------------- */
1048
1049 #if 0
1050 static int uhash_free(uhash_ctx_t ctx)
1051 {
1052 /* Free memory allocated by uhash_alloc */
1053 u_char bytes_to_sub;
1054
1055 if (ctx) {
1056 if (ALLOC_BOUNDARY) {
1057 bytes_to_sub = *((u_char *)ctx - 1);
1058 ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1059 }
1060 free(ctx);
1061 }
1062 return (1);
1063 }
1064 #endif
1065 /* ---------------------------------------------------------------------- */
1066
uhash_update(uhash_ctx_t ctx,const u_char * input,long len)1067 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1068 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1069 * hash each one with NH, calling the polyhash on each NH output.
1070 */
1071 {
1072 UWORD bytes_hashed, bytes_remaining;
1073 UINT64 result_buf[STREAMS];
1074 UINT8 *nh_result = (UINT8 *)&result_buf;
1075
1076 if (ctx->msg_len + len <= L1_KEY_LEN) {
1077 nh_update(&ctx->hash, (const UINT8 *)input, len);
1078 ctx->msg_len += len;
1079 } else {
1080
1081 bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1082 if (ctx->msg_len == L1_KEY_LEN)
1083 bytes_hashed = L1_KEY_LEN;
1084
1085 if (bytes_hashed + len >= L1_KEY_LEN) {
1086
1087 /* If some bytes have been passed to the hash function */
1088 /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1089 /* bytes to complete the current nh_block. */
1090 if (bytes_hashed) {
1091 bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1092 nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1093 nh_final(&ctx->hash, nh_result);
1094 ctx->msg_len += bytes_remaining;
1095 poly_hash(ctx,(UINT32 *)nh_result);
1096 len -= bytes_remaining;
1097 input += bytes_remaining;
1098 }
1099
1100 /* Hash directly from input stream if enough bytes */
1101 while (len >= L1_KEY_LEN) {
1102 nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1103 L1_KEY_LEN, nh_result);
1104 ctx->msg_len += L1_KEY_LEN;
1105 len -= L1_KEY_LEN;
1106 input += L1_KEY_LEN;
1107 poly_hash(ctx,(UINT32 *)nh_result);
1108 }
1109 }
1110
1111 /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1112 if (len) {
1113 nh_update(&ctx->hash, (const UINT8 *)input, len);
1114 ctx->msg_len += len;
1115 }
1116 }
1117
1118 return (1);
1119 }
1120
1121 /* ---------------------------------------------------------------------- */
1122
uhash_final(uhash_ctx_t ctx,u_char * res)1123 static int uhash_final(uhash_ctx_t ctx, u_char *res)
1124 /* Incorporate any pending data, pad, and generate tag */
1125 {
1126 UINT64 result_buf[STREAMS];
1127 UINT8 *nh_result = (UINT8 *)&result_buf;
1128
1129 if (ctx->msg_len > L1_KEY_LEN) {
1130 if (ctx->msg_len % L1_KEY_LEN) {
1131 nh_final(&ctx->hash, nh_result);
1132 poly_hash(ctx,(UINT32 *)nh_result);
1133 }
1134 ip_long(ctx, res);
1135 } else {
1136 nh_final(&ctx->hash, nh_result);
1137 ip_short(ctx,nh_result, res);
1138 }
1139 uhash_reset(ctx);
1140 return (1);
1141 }
1142
1143 /* ---------------------------------------------------------------------- */
1144
1145 #if 0
1146 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1147 /* assumes that msg is in a writable buffer of length divisible by */
1148 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */
1149 {
1150 UINT8 nh_result[STREAMS*sizeof(UINT64)];
1151 UINT32 nh_len;
1152 int extra_zeroes_needed;
1153
1154 /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1155 * the polyhash.
1156 */
1157 if (len <= L1_KEY_LEN) {
1158 if (len == 0) /* If zero length messages will not */
1159 nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */
1160 else
1161 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1162 extra_zeroes_needed = nh_len - len;
1163 zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1164 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1165 ip_short(ahc,nh_result, res);
1166 } else {
1167 /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1168 * output to poly_hash().
1169 */
1170 do {
1171 nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1172 poly_hash(ahc,(UINT32 *)nh_result);
1173 len -= L1_KEY_LEN;
1174 msg += L1_KEY_LEN;
1175 } while (len >= L1_KEY_LEN);
1176 if (len) {
1177 nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1178 extra_zeroes_needed = nh_len - len;
1179 zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1180 nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1181 poly_hash(ahc,(UINT32 *)nh_result);
1182 }
1183
1184 ip_long(ahc, res);
1185 }
1186
1187 uhash_reset(ahc);
1188 return 1;
1189 }
1190 #endif
1191
1192 /* ---------------------------------------------------------------------- */
1193 /* ---------------------------------------------------------------------- */
1194 /* ----- Begin UMAC Section --------------------------------------------- */
1195 /* ---------------------------------------------------------------------- */
1196 /* ---------------------------------------------------------------------- */
1197
1198 /* The UMAC interface has two interfaces, an all-at-once interface where
1199 * the entire message to be authenticated is passed to UMAC in one buffer,
1200 * and a sequential interface where the message is presented a little at a
1201 * time. The all-at-once is more optimized than the sequential version and
1202 * should be preferred when the sequential interface is not required.
1203 */
1204 struct umac_ctx {
1205 uhash_ctx hash; /* Hash function for message compression */
1206 pdf_ctx pdf; /* PDF for hashed output */
1207 void *free_ptr; /* Address to free this struct via */
1208 } umac_ctx;
1209
1210 /* ---------------------------------------------------------------------- */
1211
1212 #if 0
1213 int umac_reset(struct umac_ctx *ctx)
1214 /* Reset the hash function to begin a new authentication. */
1215 {
1216 uhash_reset(&ctx->hash);
1217 return (1);
1218 }
1219 #endif
1220
1221 /* ---------------------------------------------------------------------- */
1222
umac_delete(struct umac_ctx * ctx)1223 int umac_delete(struct umac_ctx *ctx)
1224 /* Deallocate the ctx structure */
1225 {
1226 if (ctx) {
1227 if (ALLOC_BOUNDARY)
1228 ctx = (struct umac_ctx *)ctx->free_ptr;
1229 freezero(ctx, sizeof(*ctx) + ALLOC_BOUNDARY);
1230 }
1231 return (1);
1232 }
1233
1234 /* ---------------------------------------------------------------------- */
1235
umac_new(const u_char key[])1236 struct umac_ctx *umac_new(const u_char key[])
1237 /* Dynamically allocate a umac_ctx struct, initialize variables,
1238 * generate subkeys from key. Align to 16-byte boundary.
1239 */
1240 {
1241 struct umac_ctx *ctx, *octx;
1242 size_t bytes_to_add;
1243 aes_int_key prf_key;
1244
1245 octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1246 if (ctx) {
1247 if (ALLOC_BOUNDARY) {
1248 bytes_to_add = ALLOC_BOUNDARY -
1249 ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1250 ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1251 }
1252 ctx->free_ptr = octx;
1253 aes_key_setup(key, prf_key);
1254 pdf_init(&ctx->pdf, prf_key);
1255 uhash_init(&ctx->hash, prf_key);
1256 explicit_bzero(prf_key, sizeof(prf_key));
1257 }
1258
1259 return (ctx);
1260 }
1261
1262 /* ---------------------------------------------------------------------- */
1263
umac_final(struct umac_ctx * ctx,u_char tag[],const u_char nonce[8])1264 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1265 /* Incorporate any pending data, pad, and generate tag */
1266 {
1267 uhash_final(&ctx->hash, (u_char *)tag);
1268 pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1269
1270 return (1);
1271 }
1272
1273 /* ---------------------------------------------------------------------- */
1274
umac_update(struct umac_ctx * ctx,const u_char * input,long len)1275 int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1276 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */
1277 /* hash each one, calling the PDF on the hashed output whenever the hash- */
1278 /* output buffer is full. */
1279 {
1280 uhash_update(&ctx->hash, input, len);
1281 return (1);
1282 }
1283
1284 /* ---------------------------------------------------------------------- */
1285
1286 #if 0
1287 int umac(struct umac_ctx *ctx, u_char *input,
1288 long len, u_char tag[],
1289 u_char nonce[8])
1290 /* All-in-one version simply calls umac_update() and umac_final(). */
1291 {
1292 uhash(&ctx->hash, input, len, (u_char *)tag);
1293 pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1294
1295 return (1);
1296 }
1297 #endif
1298
1299 /* ---------------------------------------------------------------------- */
1300 /* ---------------------------------------------------------------------- */
1301 /* ----- End UMAC Section ----------------------------------------------- */
1302 /* ---------------------------------------------------------------------- */
1303 /* ---------------------------------------------------------------------- */
1304