xref: /freebsd/crypto/openssh/umac.c (revision 4d3fc8b0570b29fb0d6ee9525f104d52176ff0d4)
1 /* $OpenBSD: umac.c,v 1.23 2023/03/07 01:30:52 djm 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 license 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 #ifndef UMAC_OUTPUT_LEN
56 #define UMAC_OUTPUT_LEN     8  /* Alowable: 4, 8, 12, 16                  */
57 #endif
58 
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
62 #endif
63 
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         */
69 
70 /* ---------------------------------------------------------------------- */
71 /* -- Global Includes --------------------------------------------------- */
72 /* ---------------------------------------------------------------------- */
73 
74 #include "includes.h"
75 #include <sys/types.h>
76 #include <string.h>
77 #include <stdarg.h>
78 #include <stdio.h>
79 #include <stdlib.h>
80 #include <stddef.h>
81 
82 #include "xmalloc.h"
83 #include "umac.h"
84 #include "misc.h"
85 
86 /* ---------------------------------------------------------------------- */
87 /* --- Primitive Data Types ---                                           */
88 /* ---------------------------------------------------------------------- */
89 
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 */
96 
97 /* ---------------------------------------------------------------------- */
98 /* --- Constants -------------------------------------------------------- */
99 /* ---------------------------------------------------------------------- */
100 
101 #define UMAC_KEY_LEN           16  /* UMAC takes 16 bytes of external key */
102 
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.                     */
106 
107 #if BYTE_ORDER == LITTLE_ENDIAN
108 #define __LITTLE_ENDIAN__ 1
109 #else
110 #define __LITTLE_ENDIAN__ 0
111 #endif
112 
113 /* ---------------------------------------------------------------------- */
114 /* ---------------------------------------------------------------------- */
115 /* ----- Architecture Specific ------------------------------------------ */
116 /* ---------------------------------------------------------------------- */
117 /* ---------------------------------------------------------------------- */
118 
119 
120 /* ---------------------------------------------------------------------- */
121 /* ---------------------------------------------------------------------- */
122 /* ----- Primitive Routines --------------------------------------------- */
123 /* ---------------------------------------------------------------------- */
124 /* ---------------------------------------------------------------------- */
125 
126 
127 /* ---------------------------------------------------------------------- */
128 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
129 /* ---------------------------------------------------------------------- */
130 
131 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
132 
133 /* ---------------------------------------------------------------------- */
134 /* --- Endian Conversion --- Forcing assembly on some platforms           */
135 /* ---------------------------------------------------------------------- */
136 
137 #if (__LITTLE_ENDIAN__)
138 #define LOAD_UINT32_REVERSED(p)		get_u32(p)
139 #define STORE_UINT32_REVERSED(p,v)	put_u32(p,v)
140 #else
141 #define LOAD_UINT32_REVERSED(p)		get_u32_le(p)
142 #define STORE_UINT32_REVERSED(p,v)	put_u32_le(p,v)
143 #endif
144 
145 #define LOAD_UINT32_LITTLE(p)		(get_u32_le(p))
146 #define STORE_UINT32_BIG(p,v)		put_u32(p, v)
147 
148 /* ---------------------------------------------------------------------- */
149 /* ---------------------------------------------------------------------- */
150 /* ----- Begin KDF & PDF Section ---------------------------------------- */
151 /* ---------------------------------------------------------------------- */
152 /* ---------------------------------------------------------------------- */
153 
154 /* UMAC uses AES with 16 byte block and key lengths */
155 #define AES_BLOCK_LEN  16
156 
157 /* OpenSSL's AES */
158 #ifdef WITH_OPENSSL
159 #include "openbsd-compat/openssl-compat.h"
160 #ifndef USE_BUILTIN_RIJNDAEL
161 # include <openssl/aes.h>
162 #endif
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 'bufp'. Each distinct
183  * 'ndx' causes a distinct byte stream.
184  */
kdf(void * bufp,aes_int_key key,UINT8 ndx,int nbytes)185 static void kdf(void *bufp, 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 *)bufp;
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 
pdf_gen_xor(pdf_ctx * pc,const UINT8 nonce[8],UINT8 buf[UMAC_OUTPUT_LEN])236 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8],
237     UINT8 buf[UMAC_OUTPUT_LEN])
238 {
239     /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
240      * of the AES output. If last time around we returned the ndx-1st
241      * element, then we may have the result in the cache already.
242      */
243 
244 #if (UMAC_OUTPUT_LEN == 4)
245 #define LOW_BIT_MASK 3
246 #elif (UMAC_OUTPUT_LEN == 8)
247 #define LOW_BIT_MASK 1
248 #elif (UMAC_OUTPUT_LEN > 8)
249 #define LOW_BIT_MASK 0
250 #endif
251     union {
252         UINT8 tmp_nonce_lo[4];
253         UINT32 align;
254     } t;
255 #if LOW_BIT_MASK != 0
256     int ndx = nonce[7] & LOW_BIT_MASK;
257 #endif
258     *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
259     t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
260 
261     if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
262          (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
263     {
264         ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
265         ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
266         aes_encryption(pc->nonce, pc->cache, pc->prf_key);
267     }
268 
269 #if (UMAC_OUTPUT_LEN == 4)
270     *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
271 #elif (UMAC_OUTPUT_LEN == 8)
272     *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
273 #elif (UMAC_OUTPUT_LEN == 12)
274     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
275     ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
276 #elif (UMAC_OUTPUT_LEN == 16)
277     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
278     ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
279 #endif
280 }
281 
282 /* ---------------------------------------------------------------------- */
283 /* ---------------------------------------------------------------------- */
284 /* ----- Begin NH Hash Section ------------------------------------------ */
285 /* ---------------------------------------------------------------------- */
286 /* ---------------------------------------------------------------------- */
287 
288 /* The NH-based hash functions used in UMAC are described in the UMAC paper
289  * and specification, both of which can be found at the UMAC website.
290  * The interface to this implementation has two
291  * versions, one expects the entire message being hashed to be passed
292  * in a single buffer and returns the hash result immediately. The second
293  * allows the message to be passed in a sequence of buffers. In the
294  * multiple-buffer interface, the client calls the routine nh_update() as
295  * many times as necessary. When there is no more data to be fed to the
296  * hash, the client calls nh_final() which calculates the hash output.
297  * Before beginning another hash calculation the nh_reset() routine
298  * must be called. The single-buffer routine, nh(), is equivalent to
299  * the sequence of calls nh_update() and nh_final(); however it is
300  * optimized and should be preferred whenever the multiple-buffer interface
301  * is not necessary. When using either interface, it is the client's
302  * responsibility to pass no more than L1_KEY_LEN bytes per hash result.
303  *
304  * The routine nh_init() initializes the nh_ctx data structure and
305  * must be called once, before any other PDF routine.
306  */
307 
308  /* The "nh_aux" routines do the actual NH hashing work. They
309   * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
310   * produce output for all STREAMS NH iterations in one call,
311   * allowing the parallel implementation of the streams.
312   */
313 
314 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied  */
315 #define L1_KEY_LEN         1024     /* Internal key bytes                 */
316 #define L1_KEY_SHIFT         16     /* Toeplitz key shift between streams */
317 #define L1_PAD_BOUNDARY      32     /* pad message to boundary multiple   */
318 #define ALLOC_BOUNDARY       16     /* Keep buffers aligned to this       */
319 #define HASH_BUF_BYTES       64     /* nh_aux_hb buffer multiple          */
320 
321 typedef struct {
322     UINT8  nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
323     UINT8  data   [HASH_BUF_BYTES];    /* Incoming data buffer           */
324     int next_data_empty;    /* Bookkeeping variable for data buffer.     */
325     int bytes_hashed;       /* Bytes (out of L1_KEY_LEN) incorporated.   */
326     UINT64 state[STREAMS];               /* on-line state     */
327 } nh_ctx;
328 
329 
330 #if (UMAC_OUTPUT_LEN == 4)
331 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)332 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
333 /* NH hashing primitive. Previous (partial) hash result is loaded and
334 * then stored via hp pointer. The length of the data pointed at by "dp",
335 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32).  Key
336 * is expected to be endian compensated in memory at key setup.
337 */
338 {
339     UINT64 h;
340     UWORD c = dlen / 32;
341     UINT32 *k = (UINT32 *)kp;
342     const UINT32 *d = (const UINT32 *)dp;
343     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
344     UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
345 
346     h = *((UINT64 *)hp);
347     do {
348         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
349         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
350         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
351         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
352         k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
353         k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
354         h += MUL64((k0 + d0), (k4 + d4));
355         h += MUL64((k1 + d1), (k5 + d5));
356         h += MUL64((k2 + d2), (k6 + d6));
357         h += MUL64((k3 + d3), (k7 + d7));
358 
359         d += 8;
360         k += 8;
361     } while (--c);
362   *((UINT64 *)hp) = h;
363 }
364 
365 #elif (UMAC_OUTPUT_LEN == 8)
366 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)367 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
368 /* Same as previous nh_aux, but two streams are handled in one pass,
369  * reading and writing 16 bytes of hash-state per call.
370  */
371 {
372   UINT64 h1,h2;
373   UWORD c = dlen / 32;
374   UINT32 *k = (UINT32 *)kp;
375   const UINT32 *d = (const UINT32 *)dp;
376   UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
377   UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
378         k8,k9,k10,k11;
379 
380   h1 = *((UINT64 *)hp);
381   h2 = *((UINT64 *)hp + 1);
382   k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
383   do {
384     d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
385     d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
386     d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
387     d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
388     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
389     k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
390 
391     h1 += MUL64((k0 + d0), (k4 + d4));
392     h2 += MUL64((k4 + d0), (k8 + d4));
393 
394     h1 += MUL64((k1 + d1), (k5 + d5));
395     h2 += MUL64((k5 + d1), (k9 + d5));
396 
397     h1 += MUL64((k2 + d2), (k6 + d6));
398     h2 += MUL64((k6 + d2), (k10 + d6));
399 
400     h1 += MUL64((k3 + d3), (k7 + d7));
401     h2 += MUL64((k7 + d3), (k11 + d7));
402 
403     k0 = k8; k1 = k9; k2 = k10; k3 = k11;
404 
405     d += 8;
406     k += 8;
407   } while (--c);
408   ((UINT64 *)hp)[0] = h1;
409   ((UINT64 *)hp)[1] = h2;
410 }
411 
412 #elif (UMAC_OUTPUT_LEN == 12)
413 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)414 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
415 /* Same as previous nh_aux, but two streams are handled in one pass,
416  * reading and writing 24 bytes of hash-state per call.
417 */
418 {
419     UINT64 h1,h2,h3;
420     UWORD c = dlen / 32;
421     UINT32 *k = (UINT32 *)kp;
422     const UINT32 *d = (const UINT32 *)dp;
423     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
424     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
425         k8,k9,k10,k11,k12,k13,k14,k15;
426 
427     h1 = *((UINT64 *)hp);
428     h2 = *((UINT64 *)hp + 1);
429     h3 = *((UINT64 *)hp + 2);
430     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
431     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
432     do {
433         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
434         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
435         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
436         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
437         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
438         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
439 
440         h1 += MUL64((k0 + d0), (k4 + d4));
441         h2 += MUL64((k4 + d0), (k8 + d4));
442         h3 += MUL64((k8 + d0), (k12 + d4));
443 
444         h1 += MUL64((k1 + d1), (k5 + d5));
445         h2 += MUL64((k5 + d1), (k9 + d5));
446         h3 += MUL64((k9 + d1), (k13 + d5));
447 
448         h1 += MUL64((k2 + d2), (k6 + d6));
449         h2 += MUL64((k6 + d2), (k10 + d6));
450         h3 += MUL64((k10 + d2), (k14 + d6));
451 
452         h1 += MUL64((k3 + d3), (k7 + d7));
453         h2 += MUL64((k7 + d3), (k11 + d7));
454         h3 += MUL64((k11 + d3), (k15 + d7));
455 
456         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
457         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
458 
459         d += 8;
460         k += 8;
461     } while (--c);
462     ((UINT64 *)hp)[0] = h1;
463     ((UINT64 *)hp)[1] = h2;
464     ((UINT64 *)hp)[2] = h3;
465 }
466 
467 #elif (UMAC_OUTPUT_LEN == 16)
468 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)469 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
470 /* Same as previous nh_aux, but two streams are handled in one pass,
471  * reading and writing 24 bytes of hash-state per call.
472 */
473 {
474     UINT64 h1,h2,h3,h4;
475     UWORD c = dlen / 32;
476     UINT32 *k = (UINT32 *)kp;
477     const UINT32 *d = (const UINT32 *)dp;
478     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
479     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
480         k8,k9,k10,k11,k12,k13,k14,k15,
481         k16,k17,k18,k19;
482 
483     h1 = *((UINT64 *)hp);
484     h2 = *((UINT64 *)hp + 1);
485     h3 = *((UINT64 *)hp + 2);
486     h4 = *((UINT64 *)hp + 3);
487     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
488     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
489     do {
490         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
491         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
492         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
493         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
494         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
495         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
496         k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
497 
498         h1 += MUL64((k0 + d0), (k4 + d4));
499         h2 += MUL64((k4 + d0), (k8 + d4));
500         h3 += MUL64((k8 + d0), (k12 + d4));
501         h4 += MUL64((k12 + d0), (k16 + d4));
502 
503         h1 += MUL64((k1 + d1), (k5 + d5));
504         h2 += MUL64((k5 + d1), (k9 + d5));
505         h3 += MUL64((k9 + d1), (k13 + d5));
506         h4 += MUL64((k13 + d1), (k17 + d5));
507 
508         h1 += MUL64((k2 + d2), (k6 + d6));
509         h2 += MUL64((k6 + d2), (k10 + d6));
510         h3 += MUL64((k10 + d2), (k14 + d6));
511         h4 += MUL64((k14 + d2), (k18 + d6));
512 
513         h1 += MUL64((k3 + d3), (k7 + d7));
514         h2 += MUL64((k7 + d3), (k11 + d7));
515         h3 += MUL64((k11 + d3), (k15 + d7));
516         h4 += MUL64((k15 + d3), (k19 + d7));
517 
518         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
519         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
520         k8 = k16; k9 = k17; k10 = k18; k11 = k19;
521 
522         d += 8;
523         k += 8;
524     } while (--c);
525     ((UINT64 *)hp)[0] = h1;
526     ((UINT64 *)hp)[1] = h2;
527     ((UINT64 *)hp)[2] = h3;
528     ((UINT64 *)hp)[3] = h4;
529 }
530 
531 /* ---------------------------------------------------------------------- */
532 #endif  /* UMAC_OUTPUT_LENGTH */
533 /* ---------------------------------------------------------------------- */
534 
535 
536 /* ---------------------------------------------------------------------- */
537 
nh_transform(nh_ctx * hc,const UINT8 * buf,UINT32 nbytes)538 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
539 /* This function is a wrapper for the primitive NH hash functions. It takes
540  * as argument "hc" the current hash context and a buffer which must be a
541  * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
542  * appropriately according to how much message has been hashed already.
543  */
544 {
545     UINT8 *key;
546 
547     key = hc->nh_key + hc->bytes_hashed;
548     nh_aux(key, buf, hc->state, nbytes);
549 }
550 
551 /* ---------------------------------------------------------------------- */
552 
553 #if (__LITTLE_ENDIAN__)
endian_convert(void * buf,UWORD bpw,UINT32 num_bytes)554 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
555 /* We endian convert the keys on little-endian computers to               */
556 /* compensate for the lack of big-endian memory reads during hashing.     */
557 {
558     UWORD iters = num_bytes / bpw;
559     if (bpw == 4) {
560         UINT32 *p = (UINT32 *)buf;
561         do {
562             *p = LOAD_UINT32_REVERSED(p);
563             p++;
564         } while (--iters);
565     } else if (bpw == 8) {
566         UINT32 *p = (UINT32 *)buf;
567         UINT32 t;
568         do {
569             t = LOAD_UINT32_REVERSED(p+1);
570             p[1] = LOAD_UINT32_REVERSED(p);
571             p[0] = t;
572             p += 2;
573         } while (--iters);
574     }
575 }
576 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
577 #else
578 #define endian_convert_if_le(x,y,z) do{}while(0)  /* Do nothing */
579 #endif
580 
581 /* ---------------------------------------------------------------------- */
582 
nh_reset(nh_ctx * hc)583 static void nh_reset(nh_ctx *hc)
584 /* Reset nh_ctx to ready for hashing of new data */
585 {
586     hc->bytes_hashed = 0;
587     hc->next_data_empty = 0;
588     hc->state[0] = 0;
589 #if (UMAC_OUTPUT_LEN >= 8)
590     hc->state[1] = 0;
591 #endif
592 #if (UMAC_OUTPUT_LEN >= 12)
593     hc->state[2] = 0;
594 #endif
595 #if (UMAC_OUTPUT_LEN == 16)
596     hc->state[3] = 0;
597 #endif
598 
599 }
600 
601 /* ---------------------------------------------------------------------- */
602 
nh_init(nh_ctx * hc,aes_int_key prf_key)603 static void nh_init(nh_ctx *hc, aes_int_key prf_key)
604 /* Generate nh_key, endian convert and reset to be ready for hashing.   */
605 {
606     kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
607     endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
608     nh_reset(hc);
609 }
610 
611 /* ---------------------------------------------------------------------- */
612 
nh_update(nh_ctx * hc,const UINT8 * buf,UINT32 nbytes)613 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
614 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an    */
615 /* even multiple of HASH_BUF_BYTES.                                       */
616 {
617     UINT32 i,j;
618 
619     j = hc->next_data_empty;
620     if ((j + nbytes) >= HASH_BUF_BYTES) {
621         if (j) {
622             i = HASH_BUF_BYTES - j;
623             memcpy(hc->data+j, buf, i);
624             nh_transform(hc,hc->data,HASH_BUF_BYTES);
625             nbytes -= i;
626             buf += i;
627             hc->bytes_hashed += HASH_BUF_BYTES;
628         }
629         if (nbytes >= HASH_BUF_BYTES) {
630             i = nbytes & ~(HASH_BUF_BYTES - 1);
631             nh_transform(hc, buf, i);
632             nbytes -= i;
633             buf += i;
634             hc->bytes_hashed += i;
635         }
636         j = 0;
637     }
638     memcpy(hc->data + j, buf, nbytes);
639     hc->next_data_empty = j + nbytes;
640 }
641 
642 /* ---------------------------------------------------------------------- */
643 
zero_pad(UINT8 * p,int nbytes)644 static void zero_pad(UINT8 *p, int nbytes)
645 {
646 /* Write "nbytes" of zeroes, beginning at "p" */
647     if (nbytes >= (int)sizeof(UWORD)) {
648         while ((ptrdiff_t)p % sizeof(UWORD)) {
649             *p = 0;
650             nbytes--;
651             p++;
652         }
653         while (nbytes >= (int)sizeof(UWORD)) {
654             *(UWORD *)p = 0;
655             nbytes -= sizeof(UWORD);
656             p += sizeof(UWORD);
657         }
658     }
659     while (nbytes) {
660         *p = 0;
661         nbytes--;
662         p++;
663     }
664 }
665 
666 /* ---------------------------------------------------------------------- */
667 
nh_final(nh_ctx * hc,UINT8 * result)668 static void nh_final(nh_ctx *hc, UINT8 *result)
669 /* After passing some number of data buffers to nh_update() for integration
670  * into an NH context, nh_final is called to produce a hash result. If any
671  * bytes are in the buffer hc->data, incorporate them into the
672  * NH context. Finally, add into the NH accumulation "state" the total number
673  * of bits hashed. The resulting numbers are written to the buffer "result".
674  * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
675  */
676 {
677     int nh_len, nbits;
678 
679     if (hc->next_data_empty != 0) {
680         nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
681                                                 ~(L1_PAD_BOUNDARY - 1));
682         zero_pad(hc->data + hc->next_data_empty,
683                                           nh_len - hc->next_data_empty);
684         nh_transform(hc, hc->data, nh_len);
685         hc->bytes_hashed += hc->next_data_empty;
686     } else if (hc->bytes_hashed == 0) {
687 	nh_len = L1_PAD_BOUNDARY;
688         zero_pad(hc->data, L1_PAD_BOUNDARY);
689         nh_transform(hc, hc->data, nh_len);
690     }
691 
692     nbits = (hc->bytes_hashed << 3);
693     ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
694 #if (UMAC_OUTPUT_LEN >= 8)
695     ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
696 #endif
697 #if (UMAC_OUTPUT_LEN >= 12)
698     ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
699 #endif
700 #if (UMAC_OUTPUT_LEN == 16)
701     ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
702 #endif
703     nh_reset(hc);
704 }
705 
706 /* ---------------------------------------------------------------------- */
707 
nh(nh_ctx * hc,const UINT8 * buf,UINT32 padded_len,UINT32 unpadded_len,UINT8 * result)708 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
709                UINT32 unpadded_len, UINT8 *result)
710 /* All-in-one nh_update() and nh_final() equivalent.
711  * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
712  * well aligned
713  */
714 {
715     UINT32 nbits;
716 
717     /* Initialize the hash state */
718     nbits = (unpadded_len << 3);
719 
720     ((UINT64 *)result)[0] = nbits;
721 #if (UMAC_OUTPUT_LEN >= 8)
722     ((UINT64 *)result)[1] = nbits;
723 #endif
724 #if (UMAC_OUTPUT_LEN >= 12)
725     ((UINT64 *)result)[2] = nbits;
726 #endif
727 #if (UMAC_OUTPUT_LEN == 16)
728     ((UINT64 *)result)[3] = nbits;
729 #endif
730 
731     nh_aux(hc->nh_key, buf, result, padded_len);
732 }
733 
734 /* ---------------------------------------------------------------------- */
735 /* ---------------------------------------------------------------------- */
736 /* ----- Begin UHASH Section -------------------------------------------- */
737 /* ---------------------------------------------------------------------- */
738 /* ---------------------------------------------------------------------- */
739 
740 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
741  * hashed by NH. The NH output is then hashed by a polynomial-hash layer
742  * unless the initial data to be hashed is short. After the polynomial-
743  * layer, an inner-product hash is used to produce the final UHASH output.
744  *
745  * UHASH provides two interfaces, one all-at-once and another where data
746  * buffers are presented sequentially. In the sequential interface, the
747  * UHASH client calls the routine uhash_update() as many times as necessary.
748  * When there is no more data to be fed to UHASH, the client calls
749  * uhash_final() which
750  * calculates the UHASH output. Before beginning another UHASH calculation
751  * the uhash_reset() routine must be called. The all-at-once UHASH routine,
752  * uhash(), is equivalent to the sequence of calls uhash_update() and
753  * uhash_final(); however it is optimized and should be
754  * used whenever the sequential interface is not necessary.
755  *
756  * The routine uhash_init() initializes the uhash_ctx data structure and
757  * must be called once, before any other UHASH routine.
758  */
759 
760 /* ---------------------------------------------------------------------- */
761 /* ----- Constants and uhash_ctx ---------------------------------------- */
762 /* ---------------------------------------------------------------------- */
763 
764 /* ---------------------------------------------------------------------- */
765 /* ----- Poly hash and Inner-Product hash Constants --------------------- */
766 /* ---------------------------------------------------------------------- */
767 
768 /* Primes and masks */
769 #define p36    ((UINT64)0x0000000FFFFFFFFBull)              /* 2^36 -  5 */
770 #define p64    ((UINT64)0xFFFFFFFFFFFFFFC5ull)              /* 2^64 - 59 */
771 #define m36    ((UINT64)0x0000000FFFFFFFFFull)  /* The low 36 of 64 bits */
772 
773 
774 /* ---------------------------------------------------------------------- */
775 
776 typedef struct uhash_ctx {
777     nh_ctx hash;                          /* Hash context for L1 NH hash  */
778     UINT64 poly_key_8[STREAMS];           /* p64 poly keys                */
779     UINT64 poly_accum[STREAMS];           /* poly hash result             */
780     UINT64 ip_keys[STREAMS*4];            /* Inner-product keys           */
781     UINT32 ip_trans[STREAMS];             /* Inner-product translation    */
782     UINT32 msg_len;                       /* Total length of data passed  */
783                                           /* to uhash */
784 } uhash_ctx;
785 typedef struct uhash_ctx *uhash_ctx_t;
786 
787 /* ---------------------------------------------------------------------- */
788 
789 
790 /* The polynomial hashes use Horner's rule to evaluate a polynomial one
791  * word at a time. As described in the specification, poly32 and poly64
792  * require keys from special domains. The following implementations exploit
793  * the special domains to avoid overflow. The results are not guaranteed to
794  * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
795  * patches any errant values.
796  */
797 
poly64(UINT64 cur,UINT64 key,UINT64 data)798 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
799 {
800     UINT32 key_hi = (UINT32)(key >> 32),
801            key_lo = (UINT32)key,
802            cur_hi = (UINT32)(cur >> 32),
803            cur_lo = (UINT32)cur,
804            x_lo,
805            x_hi;
806     UINT64 X,T,res;
807 
808     X =  MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
809     x_lo = (UINT32)X;
810     x_hi = (UINT32)(X >> 32);
811 
812     res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
813 
814     T = ((UINT64)x_lo << 32);
815     res += T;
816     if (res < T)
817         res += 59;
818 
819     res += data;
820     if (res < data)
821         res += 59;
822 
823     return res;
824 }
825 
826 
827 /* Although UMAC is specified to use a ramped polynomial hash scheme, this
828  * implementation does not handle all ramp levels. Because we don't handle
829  * the ramp up to p128 modulus in this implementation, we are limited to
830  * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
831  * bytes input to UMAC per tag, ie. 16MB).
832  */
poly_hash(uhash_ctx_t hc,UINT32 data_in[])833 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
834 {
835     int i;
836     UINT64 *data=(UINT64*)data_in;
837 
838     for (i = 0; i < STREAMS; i++) {
839         if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
840             hc->poly_accum[i] = poly64(hc->poly_accum[i],
841                                        hc->poly_key_8[i], p64 - 1);
842             hc->poly_accum[i] = poly64(hc->poly_accum[i],
843                                        hc->poly_key_8[i], (data[i] - 59));
844         } else {
845             hc->poly_accum[i] = poly64(hc->poly_accum[i],
846                                        hc->poly_key_8[i], data[i]);
847         }
848     }
849 }
850 
851 
852 /* ---------------------------------------------------------------------- */
853 
854 
855 /* The final step in UHASH is an inner-product hash. The poly hash
856  * produces a result not necessarily WORD_LEN bytes long. The inner-
857  * product hash breaks the polyhash output into 16-bit chunks and
858  * multiplies each with a 36 bit key.
859  */
860 
ip_aux(UINT64 t,UINT64 * ipkp,UINT64 data)861 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
862 {
863     t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
864     t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
865     t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
866     t = t + ipkp[3] * (UINT64)(UINT16)(data);
867 
868     return t;
869 }
870 
ip_reduce_p36(UINT64 t)871 static UINT32 ip_reduce_p36(UINT64 t)
872 {
873 /* Divisionless modular reduction */
874     UINT64 ret;
875 
876     ret = (t & m36) + 5 * (t >> 36);
877     if (ret >= p36)
878         ret -= p36;
879 
880     /* return least significant 32 bits */
881     return (UINT32)(ret);
882 }
883 
884 
885 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
886  * the polyhash stage is skipped and ip_short is applied directly to the
887  * NH output.
888  */
ip_short(uhash_ctx_t ahc,UINT8 * nh_res,u_char * res)889 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
890 {
891     UINT64 t;
892     UINT64 *nhp = (UINT64 *)nh_res;
893 
894     t  = ip_aux(0,ahc->ip_keys, nhp[0]);
895     STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
896 #if (UMAC_OUTPUT_LEN >= 8)
897     t  = ip_aux(0,ahc->ip_keys+4, nhp[1]);
898     STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
899 #endif
900 #if (UMAC_OUTPUT_LEN >= 12)
901     t  = ip_aux(0,ahc->ip_keys+8, nhp[2]);
902     STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
903 #endif
904 #if (UMAC_OUTPUT_LEN == 16)
905     t  = ip_aux(0,ahc->ip_keys+12, nhp[3]);
906     STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
907 #endif
908 }
909 
910 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
911  * the polyhash stage is not skipped and ip_long is applied to the
912  * polyhash output.
913  */
ip_long(uhash_ctx_t ahc,u_char * res)914 static void ip_long(uhash_ctx_t ahc, u_char *res)
915 {
916     int i;
917     UINT64 t;
918 
919     for (i = 0; i < STREAMS; i++) {
920         /* fix polyhash output not in Z_p64 */
921         if (ahc->poly_accum[i] >= p64)
922             ahc->poly_accum[i] -= p64;
923         t  = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
924         STORE_UINT32_BIG((UINT32 *)res+i,
925                          ip_reduce_p36(t) ^ ahc->ip_trans[i]);
926     }
927 }
928 
929 
930 /* ---------------------------------------------------------------------- */
931 
932 /* ---------------------------------------------------------------------- */
933 
934 /* Reset uhash context for next hash session */
uhash_reset(uhash_ctx_t pc)935 static int uhash_reset(uhash_ctx_t pc)
936 {
937     nh_reset(&pc->hash);
938     pc->msg_len = 0;
939     pc->poly_accum[0] = 1;
940 #if (UMAC_OUTPUT_LEN >= 8)
941     pc->poly_accum[1] = 1;
942 #endif
943 #if (UMAC_OUTPUT_LEN >= 12)
944     pc->poly_accum[2] = 1;
945 #endif
946 #if (UMAC_OUTPUT_LEN == 16)
947     pc->poly_accum[3] = 1;
948 #endif
949     return 1;
950 }
951 
952 /* ---------------------------------------------------------------------- */
953 
954 /* Given a pointer to the internal key needed by kdf() and a uhash context,
955  * initialize the NH context and generate keys needed for poly and inner-
956  * product hashing. All keys are endian adjusted in memory so that native
957  * loads cause correct keys to be in registers during calculation.
958  */
uhash_init(uhash_ctx_t ahc,aes_int_key prf_key)959 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
960 {
961     int i;
962     UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
963 
964     /* Zero the entire uhash context */
965     memset(ahc, 0, sizeof(uhash_ctx));
966 
967     /* Initialize the L1 hash */
968     nh_init(&ahc->hash, prf_key);
969 
970     /* Setup L2 hash variables */
971     kdf(buf, prf_key, 2, sizeof(buf));    /* Fill buffer with index 1 key */
972     for (i = 0; i < STREAMS; i++) {
973         /* Fill keys from the buffer, skipping bytes in the buffer not
974          * used by this implementation. Endian reverse the keys if on a
975          * little-endian computer.
976          */
977         memcpy(ahc->poly_key_8+i, buf+24*i, 8);
978         endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
979         /* Mask the 64-bit keys to their special domain */
980         ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
981         ahc->poly_accum[i] = 1;  /* Our polyhash prepends a non-zero word */
982     }
983 
984     /* Setup L3-1 hash variables */
985     kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
986     for (i = 0; i < STREAMS; i++)
987           memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
988                                                  4*sizeof(UINT64));
989     endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
990                                                   sizeof(ahc->ip_keys));
991     for (i = 0; i < STREAMS*4; i++)
992         ahc->ip_keys[i] %= p36;  /* Bring into Z_p36 */
993 
994     /* Setup L3-2 hash variables    */
995     /* Fill buffer with index 4 key */
996     kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
997     endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
998                          STREAMS * sizeof(UINT32));
999     explicit_bzero(buf, sizeof(buf));
1000 }
1001 
1002 /* ---------------------------------------------------------------------- */
1003 
1004 #if 0
1005 static uhash_ctx_t uhash_alloc(u_char key[])
1006 {
1007 /* Allocate memory and force to a 16-byte boundary. */
1008     uhash_ctx_t ctx;
1009     u_char bytes_to_add;
1010     aes_int_key prf_key;
1011 
1012     ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1013     if (ctx) {
1014         if (ALLOC_BOUNDARY) {
1015             bytes_to_add = ALLOC_BOUNDARY -
1016                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1017             ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1018             *((u_char *)ctx - 1) = bytes_to_add;
1019         }
1020         aes_key_setup(key,prf_key);
1021         uhash_init(ctx, prf_key);
1022     }
1023     return (ctx);
1024 }
1025 #endif
1026 
1027 /* ---------------------------------------------------------------------- */
1028 
1029 #if 0
1030 static int uhash_free(uhash_ctx_t ctx)
1031 {
1032 /* Free memory allocated by uhash_alloc */
1033     u_char bytes_to_sub;
1034 
1035     if (ctx) {
1036         if (ALLOC_BOUNDARY) {
1037             bytes_to_sub = *((u_char *)ctx - 1);
1038             ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1039         }
1040         free(ctx);
1041     }
1042     return (1);
1043 }
1044 #endif
1045 /* ---------------------------------------------------------------------- */
1046 
uhash_update(uhash_ctx_t ctx,const u_char * input,long len)1047 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1048 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1049  * hash each one with NH, calling the polyhash on each NH output.
1050  */
1051 {
1052     UWORD bytes_hashed, bytes_remaining;
1053     UINT64 result_buf[STREAMS];
1054     UINT8 *nh_result = (UINT8 *)&result_buf;
1055 
1056     if (ctx->msg_len + len <= L1_KEY_LEN) {
1057         nh_update(&ctx->hash, (const UINT8 *)input, len);
1058         ctx->msg_len += len;
1059     } else {
1060 
1061          bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1062          if (ctx->msg_len == L1_KEY_LEN)
1063              bytes_hashed = L1_KEY_LEN;
1064 
1065          if (bytes_hashed + len >= L1_KEY_LEN) {
1066 
1067              /* If some bytes have been passed to the hash function      */
1068              /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1069              /* bytes to complete the current nh_block.                  */
1070              if (bytes_hashed) {
1071                  bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1072                  nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1073                  nh_final(&ctx->hash, nh_result);
1074                  ctx->msg_len += bytes_remaining;
1075                  poly_hash(ctx,(UINT32 *)nh_result);
1076                  len -= bytes_remaining;
1077                  input += bytes_remaining;
1078              }
1079 
1080              /* Hash directly from input stream if enough bytes */
1081              while (len >= L1_KEY_LEN) {
1082                  nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1083                                    L1_KEY_LEN, nh_result);
1084                  ctx->msg_len += L1_KEY_LEN;
1085                  len -= L1_KEY_LEN;
1086                  input += L1_KEY_LEN;
1087                  poly_hash(ctx,(UINT32 *)nh_result);
1088              }
1089          }
1090 
1091          /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1092          if (len) {
1093              nh_update(&ctx->hash, (const UINT8 *)input, len);
1094              ctx->msg_len += len;
1095          }
1096      }
1097 
1098     return (1);
1099 }
1100 
1101 /* ---------------------------------------------------------------------- */
1102 
uhash_final(uhash_ctx_t ctx,u_char * res)1103 static int uhash_final(uhash_ctx_t ctx, u_char *res)
1104 /* Incorporate any pending data, pad, and generate tag */
1105 {
1106     UINT64 result_buf[STREAMS];
1107     UINT8 *nh_result = (UINT8 *)&result_buf;
1108 
1109     if (ctx->msg_len > L1_KEY_LEN) {
1110         if (ctx->msg_len % L1_KEY_LEN) {
1111             nh_final(&ctx->hash, nh_result);
1112             poly_hash(ctx,(UINT32 *)nh_result);
1113         }
1114         ip_long(ctx, res);
1115     } else {
1116         nh_final(&ctx->hash, nh_result);
1117         ip_short(ctx,nh_result, res);
1118     }
1119     uhash_reset(ctx);
1120     return (1);
1121 }
1122 
1123 /* ---------------------------------------------------------------------- */
1124 
1125 #if 0
1126 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1127 /* assumes that msg is in a writable buffer of length divisible by */
1128 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed.           */
1129 {
1130     UINT8 nh_result[STREAMS*sizeof(UINT64)];
1131     UINT32 nh_len;
1132     int extra_zeroes_needed;
1133 
1134     /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1135      * the polyhash.
1136      */
1137     if (len <= L1_KEY_LEN) {
1138 	if (len == 0)                  /* If zero length messages will not */
1139 		nh_len = L1_PAD_BOUNDARY;  /* be seen, comment out this case   */
1140 	else
1141 		nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1142         extra_zeroes_needed = nh_len - len;
1143         zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1144         nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1145         ip_short(ahc,nh_result, res);
1146     } else {
1147         /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1148          * output to poly_hash().
1149          */
1150         do {
1151             nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1152             poly_hash(ahc,(UINT32 *)nh_result);
1153             len -= L1_KEY_LEN;
1154             msg += L1_KEY_LEN;
1155         } while (len >= L1_KEY_LEN);
1156         if (len) {
1157             nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1158             extra_zeroes_needed = nh_len - len;
1159             zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1160             nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1161             poly_hash(ahc,(UINT32 *)nh_result);
1162         }
1163 
1164         ip_long(ahc, res);
1165     }
1166 
1167     uhash_reset(ahc);
1168     return 1;
1169 }
1170 #endif
1171 
1172 /* ---------------------------------------------------------------------- */
1173 /* ---------------------------------------------------------------------- */
1174 /* ----- Begin UMAC Section --------------------------------------------- */
1175 /* ---------------------------------------------------------------------- */
1176 /* ---------------------------------------------------------------------- */
1177 
1178 /* The UMAC interface has two interfaces, an all-at-once interface where
1179  * the entire message to be authenticated is passed to UMAC in one buffer,
1180  * and a sequential interface where the message is presented a little at a
1181  * time. The all-at-once is more optimized than the sequential version and
1182  * should be preferred when the sequential interface is not required.
1183  */
1184 struct umac_ctx {
1185     uhash_ctx hash;          /* Hash function for message compression    */
1186     pdf_ctx pdf;             /* PDF for hashed output                    */
1187     void *free_ptr;          /* Address to free this struct via          */
1188 } umac_ctx;
1189 
1190 /* ---------------------------------------------------------------------- */
1191 
1192 #if 0
1193 int umac_reset(struct umac_ctx *ctx)
1194 /* Reset the hash function to begin a new authentication.        */
1195 {
1196     uhash_reset(&ctx->hash);
1197     return (1);
1198 }
1199 #endif
1200 
1201 /* ---------------------------------------------------------------------- */
1202 
umac_delete(struct umac_ctx * ctx)1203 int umac_delete(struct umac_ctx *ctx)
1204 /* Deallocate the ctx structure */
1205 {
1206     if (ctx) {
1207         if (ALLOC_BOUNDARY)
1208             ctx = (struct umac_ctx *)ctx->free_ptr;
1209         freezero(ctx, sizeof(*ctx) + ALLOC_BOUNDARY);
1210     }
1211     return (1);
1212 }
1213 
1214 /* ---------------------------------------------------------------------- */
1215 
umac_new(const u_char key[])1216 struct umac_ctx *umac_new(const u_char key[])
1217 /* Dynamically allocate a umac_ctx struct, initialize variables,
1218  * generate subkeys from key. Align to 16-byte boundary.
1219  */
1220 {
1221     struct umac_ctx *ctx, *octx;
1222     size_t bytes_to_add;
1223     aes_int_key prf_key;
1224 
1225     octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1226     if (ctx) {
1227         if (ALLOC_BOUNDARY) {
1228             bytes_to_add = ALLOC_BOUNDARY -
1229                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1230             ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1231         }
1232         ctx->free_ptr = octx;
1233         aes_key_setup(key, prf_key);
1234         pdf_init(&ctx->pdf, prf_key);
1235         uhash_init(&ctx->hash, prf_key);
1236         explicit_bzero(prf_key, sizeof(prf_key));
1237     }
1238 
1239     return (ctx);
1240 }
1241 
1242 /* ---------------------------------------------------------------------- */
1243 
umac_final(struct umac_ctx * ctx,u_char tag[],const u_char nonce[8])1244 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1245 /* Incorporate any pending data, pad, and generate tag */
1246 {
1247     uhash_final(&ctx->hash, (u_char *)tag);
1248     pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1249 
1250     return (1);
1251 }
1252 
1253 /* ---------------------------------------------------------------------- */
1254 
umac_update(struct umac_ctx * ctx,const u_char * input,long len)1255 int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1256 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and   */
1257 /* hash each one, calling the PDF on the hashed output whenever the hash- */
1258 /* output buffer is full.                                                 */
1259 {
1260     uhash_update(&ctx->hash, input, len);
1261     return (1);
1262 }
1263 
1264 /* ---------------------------------------------------------------------- */
1265 
1266 #if 0
1267 int umac(struct umac_ctx *ctx, u_char *input,
1268          long len, u_char tag[],
1269          u_char nonce[8])
1270 /* All-in-one version simply calls umac_update() and umac_final().        */
1271 {
1272     uhash(&ctx->hash, input, len, (u_char *)tag);
1273     pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1274 
1275     return (1);
1276 }
1277 #endif
1278 
1279 /* ---------------------------------------------------------------------- */
1280 /* ---------------------------------------------------------------------- */
1281 /* ----- End UMAC Section ----------------------------------------------- */
1282 /* ---------------------------------------------------------------------- */
1283 /* ---------------------------------------------------------------------- */
1284