xref: /freebsd/crypto/openssh/umac.c (revision 2574974648c68c738aec3ff96644d888d7913a37)
1 /* $OpenBSD: umac.c,v 1.30 2026/03/03 09:57:26 dtucker 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 https://fastcrypto.org/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 crypto 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  /* Allowable: 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 
76 #include <sys/types.h>
77 #include <endian.h>
78 #include <string.h>
79 #include <stdarg.h>
80 #include <stdint.h>
81 #include <stdlib.h>
82 #include <stddef.h>
83 
84 #include "xmalloc.h"
85 #include "umac.h"
86 #include "misc.h"
87 
88 /* ---------------------------------------------------------------------- */
89 /* --- Primitive Data Types ---                                           */
90 /* ---------------------------------------------------------------------- */
91 
92 /* The following assumptions may need change on your system */
93 typedef uint8_t	UINT8;  /* 1 byte   */
94 typedef uint16_t	UINT16; /* 2 byte   */
95 typedef uint32_t	UINT32; /* 4 byte   */
96 typedef uint64_t	UINT64; /* 8 bytes  */
97 typedef unsigned int	UWORD;  /* Register */
98 
99 /* ---------------------------------------------------------------------- */
100 /* --- Constants -------------------------------------------------------- */
101 /* ---------------------------------------------------------------------- */
102 
103 #define UMAC_KEY_LEN           16  /* UMAC takes 16 bytes of external key */
104 
105 /* Message "words" are read from memory in an endian-specific manner.     */
106 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must    */
107 /* be set true if the host computer is little-endian.                     */
108 
109 #if BYTE_ORDER == LITTLE_ENDIAN
110 #define __LITTLE_ENDIAN__ 1
111 #else
112 #define __LITTLE_ENDIAN__ 0
113 #endif
114 
115 /* ---------------------------------------------------------------------- */
116 /* ---------------------------------------------------------------------- */
117 /* ----- Architecture Specific ------------------------------------------ */
118 /* ---------------------------------------------------------------------- */
119 /* ---------------------------------------------------------------------- */
120 
121 
122 /* ---------------------------------------------------------------------- */
123 /* ---------------------------------------------------------------------- */
124 /* ----- Primitive Routines --------------------------------------------- */
125 /* ---------------------------------------------------------------------- */
126 /* ---------------------------------------------------------------------- */
127 
128 
129 /* ---------------------------------------------------------------------- */
130 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
131 /* ---------------------------------------------------------------------- */
132 
133 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
134 
135 /* ---------------------------------------------------------------------- */
136 /* --- Endian Conversion --- Forcing assembly on some platforms           */
137 /* ---------------------------------------------------------------------- */
138 
139 #if (__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 /* ----- Begin KDF & PDF Section ---------------------------------------- */
153 /* ---------------------------------------------------------------------- */
154 /* ---------------------------------------------------------------------- */
155 
156 /* UMAC uses AES with 16 byte block and key lengths */
157 #define AES_BLOCK_LEN  16
158 
159 /* OpenSSL's AES */
160 #ifdef WITH_OPENSSL
161 #include "openbsd-compat/openssl-compat.h"
162 #ifndef USE_BUILTIN_RIJNDAEL
163 # include <openssl/aes.h>
164 #endif
165 typedef AES_KEY aes_int_key[1];
166 #define aes_encryption(in,out,int_key)                  \
167   AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key)
168 #define aes_key_setup(key,int_key)                      \
169   AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key)
170 #else
171 #include "rijndael.h"
172 #define AES_ROUNDS ((UMAC_KEY_LEN / 4) + 6)
173 typedef UINT8 aes_int_key[AES_ROUNDS+1][4][4];	/* AES internal */
174 #define aes_encryption(in,out,int_key) \
175   rijndaelEncrypt((u32 *)(int_key), AES_ROUNDS, (u8 *)(in), (u8 *)(out))
176 #define aes_key_setup(key,int_key) \
177   rijndaelKeySetupEnc((u32 *)(int_key), (const unsigned char *)(key), \
178   UMAC_KEY_LEN*8)
179 #endif
180 
181 /* The user-supplied UMAC key is stretched using AES in a counter
182  * mode to supply all random bits needed by UMAC. The kdf function takes
183  * an AES internal key representation 'key' and writes a stream of
184  * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct
185  * 'ndx' causes a distinct byte stream.
186  */
kdf(void * bufp,aes_int_key key,UINT8 ndx,int nbytes)187 static void kdf(void *bufp, aes_int_key key, UINT8 ndx, int nbytes)
188 {
189     UINT8 in_buf[AES_BLOCK_LEN] = {0};
190     UINT8 out_buf[AES_BLOCK_LEN];
191     UINT8 *dst_buf = (UINT8 *)bufp;
192     int i;
193 
194     /* Setup the initial value */
195     in_buf[AES_BLOCK_LEN-9] = ndx;
196     in_buf[AES_BLOCK_LEN-1] = i = 1;
197 
198     while (nbytes >= AES_BLOCK_LEN) {
199         aes_encryption(in_buf, out_buf, key);
200         memcpy(dst_buf,out_buf,AES_BLOCK_LEN);
201         in_buf[AES_BLOCK_LEN-1] = ++i;
202         nbytes -= AES_BLOCK_LEN;
203         dst_buf += AES_BLOCK_LEN;
204     }
205     if (nbytes) {
206         aes_encryption(in_buf, out_buf, key);
207         memcpy(dst_buf,out_buf,nbytes);
208     }
209     explicit_bzero(in_buf, sizeof(in_buf));
210     explicit_bzero(out_buf, sizeof(out_buf));
211 }
212 
213 /* The final UHASH result is XOR'd with the output of a pseudorandom
214  * function. Here, we use AES to generate random output and
215  * xor the appropriate bytes depending on the last bits of nonce.
216  * This scheme is optimized for sequential, increasing big-endian nonces.
217  */
218 
219 typedef struct {
220     UINT8 cache[AES_BLOCK_LEN];  /* Previous AES output is saved      */
221     UINT8 nonce[AES_BLOCK_LEN];  /* The AES input making above cache  */
222     aes_int_key prf_key;         /* Expanded AES key for PDF          */
223 } pdf_ctx;
224 
pdf_init(pdf_ctx * pc,aes_int_key prf_key)225 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
226 {
227     UINT8 buf[UMAC_KEY_LEN];
228 
229     kdf(buf, prf_key, 0, UMAC_KEY_LEN);
230     aes_key_setup(buf, pc->prf_key);
231 
232     /* Initialize pdf and cache */
233     memset(pc->nonce, 0, sizeof(pc->nonce));
234     aes_encryption(pc->nonce, pc->cache, pc->prf_key);
235     explicit_bzero(buf, sizeof(buf));
236 }
237 
pdf_gen_xor(pdf_ctx * pc,const UINT8 nonce[8],UINT8 buf[UMAC_OUTPUT_LEN])238 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8],
239     UINT8 buf[UMAC_OUTPUT_LEN])
240 {
241     /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
242      * of the AES output. If last time around we returned the ndx-1st
243      * element, then we may have the result in the cache already.
244      */
245 
246 #if (UMAC_OUTPUT_LEN == 4)
247 #define LOW_BIT_MASK 3
248 #elif (UMAC_OUTPUT_LEN == 8)
249 #define LOW_BIT_MASK 1
250 #elif (UMAC_OUTPUT_LEN > 8)
251 #define LOW_BIT_MASK 0
252 #endif
253     union {
254         UINT8 tmp_nonce_lo[4];
255         UINT32 align;
256     } t;
257 #if LOW_BIT_MASK != 0
258     int ndx = nonce[7] & LOW_BIT_MASK;
259 #endif
260     *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
261     t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
262 
263     if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
264          (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
265     {
266         ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
267         ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
268         aes_encryption(pc->nonce, pc->cache, pc->prf_key);
269     }
270 
271 #if (UMAC_OUTPUT_LEN == 4)
272     *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
273 #elif (UMAC_OUTPUT_LEN == 8)
274     *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
275 #elif (UMAC_OUTPUT_LEN == 12)
276     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
277     ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
278 #elif (UMAC_OUTPUT_LEN == 16)
279     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
280     ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
281 #endif
282 }
283 
284 /* ---------------------------------------------------------------------- */
285 /* ---------------------------------------------------------------------- */
286 /* ----- Begin NH Hash Section ------------------------------------------ */
287 /* ---------------------------------------------------------------------- */
288 /* ---------------------------------------------------------------------- */
289 
290 /* The NH-based hash functions used in UMAC are described in the UMAC paper
291  * and specification, both of which can be found at the UMAC website.
292  * The interface to this implementation has two
293  * versions, one expects the entire message being hashed to be passed
294  * in a single buffer and returns the hash result immediately. The second
295  * allows the message to be passed in a sequence of buffers. In the
296  * multiple-buffer interface, the client calls the routine nh_update() as
297  * many times as necessary. When there is no more data to be fed to the
298  * hash, the client calls nh_final() which calculates the hash output.
299  * Before beginning another hash calculation the nh_reset() routine
300  * must be called. The single-buffer routine, nh(), is equivalent to
301  * the sequence of calls nh_update() and nh_final(); however it is
302  * optimized and should be preferred whenever the multiple-buffer interface
303  * is not necessary. When using either interface, it is the client's
304  * responsibility to pass no more than L1_KEY_LEN bytes per hash result.
305  *
306  * The routine nh_init() initializes the nh_ctx data structure and
307  * must be called once, before any other PDF routine.
308  */
309 
310  /* The "nh_aux" routines do the actual NH hashing work. They
311   * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
312   * produce output for all STREAMS NH iterations in one call,
313   * allowing the parallel implementation of the streams.
314   */
315 
316 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied  */
317 #define L1_KEY_LEN         1024     /* Internal key bytes                 */
318 #define L1_KEY_SHIFT         16     /* Toeplitz key shift between streams */
319 #define L1_PAD_BOUNDARY      32     /* pad message to boundary multiple   */
320 #define ALLOC_BOUNDARY       16     /* Keep buffers aligned to this       */
321 #define HASH_BUF_BYTES       64     /* nh_aux_hb buffer multiple          */
322 
323 typedef struct {
324     UINT8  nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
325     UINT8  data   [HASH_BUF_BYTES];    /* Incoming data buffer           */
326     int next_data_empty;    /* Bookkeeping variable for data buffer.     */
327     int bytes_hashed;       /* Bytes (out of L1_KEY_LEN) incorporated.   */
328     UINT64 state[STREAMS];               /* on-line state     */
329 } nh_ctx;
330 
331 
332 #if (UMAC_OUTPUT_LEN == 4)
333 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)334 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
335 /* NH hashing primitive. Previous (partial) hash result is loaded and
336 * then stored via hp pointer. The length of the data pointed at by "dp",
337 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32).  Key
338 * is expected to be endian compensated in memory at key setup.
339 */
340 {
341     UINT64 h;
342     UWORD c = dlen / 32;
343     UINT32 *k = (UINT32 *)kp;
344     const UINT32 *d = (const UINT32 *)dp;
345     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
346     UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
347 
348     h = *((UINT64 *)hp);
349     do {
350         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
351         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
352         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
353         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
354         k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
355         k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
356         h += MUL64((k0 + d0), (k4 + d4));
357         h += MUL64((k1 + d1), (k5 + d5));
358         h += MUL64((k2 + d2), (k6 + d6));
359         h += MUL64((k3 + d3), (k7 + d7));
360 
361         d += 8;
362         k += 8;
363     } while (--c);
364   *((UINT64 *)hp) = h;
365 }
366 
367 #elif (UMAC_OUTPUT_LEN == 8)
368 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)369 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
370 /* Same as previous nh_aux, but two streams are handled in one pass,
371  * reading and writing 16 bytes of hash-state per call.
372  */
373 {
374   UINT64 h1,h2;
375   UWORD c = dlen / 32;
376   UINT32 *k = (UINT32 *)kp;
377   const UINT32 *d = (const UINT32 *)dp;
378   UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
379   UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
380         k8,k9,k10,k11;
381 
382   h1 = *((UINT64 *)hp);
383   h2 = *((UINT64 *)hp + 1);
384   k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
385   do {
386     d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
387     d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
388     d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
389     d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
390     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
391     k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
392 
393     h1 += MUL64((k0 + d0), (k4 + d4));
394     h2 += MUL64((k4 + d0), (k8 + d4));
395 
396     h1 += MUL64((k1 + d1), (k5 + d5));
397     h2 += MUL64((k5 + d1), (k9 + d5));
398 
399     h1 += MUL64((k2 + d2), (k6 + d6));
400     h2 += MUL64((k6 + d2), (k10 + d6));
401 
402     h1 += MUL64((k3 + d3), (k7 + d7));
403     h2 += MUL64((k7 + d3), (k11 + d7));
404 
405     k0 = k8; k1 = k9; k2 = k10; k3 = k11;
406 
407     d += 8;
408     k += 8;
409   } while (--c);
410   ((UINT64 *)hp)[0] = h1;
411   ((UINT64 *)hp)[1] = h2;
412 }
413 
414 #elif (UMAC_OUTPUT_LEN == 12)
415 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)416 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
417 /* Same as previous nh_aux, but two streams are handled in one pass,
418  * reading and writing 24 bytes of hash-state per call.
419 */
420 {
421     UINT64 h1,h2,h3;
422     UWORD c = dlen / 32;
423     UINT32 *k = (UINT32 *)kp;
424     const UINT32 *d = (const UINT32 *)dp;
425     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
426     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
427         k8,k9,k10,k11,k12,k13,k14,k15;
428 
429     h1 = *((UINT64 *)hp);
430     h2 = *((UINT64 *)hp + 1);
431     h3 = *((UINT64 *)hp + 2);
432     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
433     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
434     do {
435         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
436         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
437         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
438         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
439         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
440         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
441 
442         h1 += MUL64((k0 + d0), (k4 + d4));
443         h2 += MUL64((k4 + d0), (k8 + d4));
444         h3 += MUL64((k8 + d0), (k12 + d4));
445 
446         h1 += MUL64((k1 + d1), (k5 + d5));
447         h2 += MUL64((k5 + d1), (k9 + d5));
448         h3 += MUL64((k9 + d1), (k13 + d5));
449 
450         h1 += MUL64((k2 + d2), (k6 + d6));
451         h2 += MUL64((k6 + d2), (k10 + d6));
452         h3 += MUL64((k10 + d2), (k14 + d6));
453 
454         h1 += MUL64((k3 + d3), (k7 + d7));
455         h2 += MUL64((k7 + d3), (k11 + d7));
456         h3 += MUL64((k11 + d3), (k15 + d7));
457 
458         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
459         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
460 
461         d += 8;
462         k += 8;
463     } while (--c);
464     ((UINT64 *)hp)[0] = h1;
465     ((UINT64 *)hp)[1] = h2;
466     ((UINT64 *)hp)[2] = h3;
467 }
468 
469 #elif (UMAC_OUTPUT_LEN == 16)
470 
nh_aux(void * kp,const void * dp,void * hp,UINT32 dlen)471 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
472 /* Same as previous nh_aux, but two streams are handled in one pass,
473  * reading and writing 24 bytes of hash-state per call.
474 */
475 {
476     UINT64 h1,h2,h3,h4;
477     UWORD c = dlen / 32;
478     UINT32 *k = (UINT32 *)kp;
479     const UINT32 *d = (const UINT32 *)dp;
480     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
481     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
482         k8,k9,k10,k11,k12,k13,k14,k15,
483         k16,k17,k18,k19;
484 
485     h1 = *((UINT64 *)hp);
486     h2 = *((UINT64 *)hp + 1);
487     h3 = *((UINT64 *)hp + 2);
488     h4 = *((UINT64 *)hp + 3);
489     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
490     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
491     do {
492         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
493         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
494         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
495         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
496         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
497         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
498         k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
499 
500         h1 += MUL64((k0 + d0), (k4 + d4));
501         h2 += MUL64((k4 + d0), (k8 + d4));
502         h3 += MUL64((k8 + d0), (k12 + d4));
503         h4 += MUL64((k12 + d0), (k16 + d4));
504 
505         h1 += MUL64((k1 + d1), (k5 + d5));
506         h2 += MUL64((k5 + d1), (k9 + d5));
507         h3 += MUL64((k9 + d1), (k13 + d5));
508         h4 += MUL64((k13 + d1), (k17 + d5));
509 
510         h1 += MUL64((k2 + d2), (k6 + d6));
511         h2 += MUL64((k6 + d2), (k10 + d6));
512         h3 += MUL64((k10 + d2), (k14 + d6));
513         h4 += MUL64((k14 + d2), (k18 + d6));
514 
515         h1 += MUL64((k3 + d3), (k7 + d7));
516         h2 += MUL64((k7 + d3), (k11 + d7));
517         h3 += MUL64((k11 + d3), (k15 + d7));
518         h4 += MUL64((k15 + d3), (k19 + d7));
519 
520         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
521         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
522         k8 = k16; k9 = k17; k10 = k18; k11 = k19;
523 
524         d += 8;
525         k += 8;
526     } while (--c);
527     ((UINT64 *)hp)[0] = h1;
528     ((UINT64 *)hp)[1] = h2;
529     ((UINT64 *)hp)[2] = h3;
530     ((UINT64 *)hp)[3] = h4;
531 }
532 
533 /* ---------------------------------------------------------------------- */
534 #endif  /* UMAC_OUTPUT_LENGTH */
535 /* ---------------------------------------------------------------------- */
536 
537 
538 /* ---------------------------------------------------------------------- */
539 
nh_transform(nh_ctx * hc,const UINT8 * buf,UINT32 nbytes)540 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
541 /* This function is a wrapper for the primitive NH hash functions. It takes
542  * as argument "hc" the current hash context and a buffer which must be a
543  * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
544  * appropriately according to how much message has been hashed already.
545  */
546 {
547     UINT8 *key;
548 
549     key = hc->nh_key + hc->bytes_hashed;
550     nh_aux(key, buf, hc->state, nbytes);
551 }
552 
553 /* ---------------------------------------------------------------------- */
554 
555 #if (__LITTLE_ENDIAN__)
endian_convert(void * buf,UWORD bpw,UINT32 num_bytes)556 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
557 /* We endian convert the keys on little-endian computers to               */
558 /* compensate for the lack of big-endian memory reads during hashing.     */
559 {
560     UWORD iters = num_bytes / bpw;
561     if (bpw == 4) {
562         UINT32 *p = (UINT32 *)buf;
563         do {
564             *p = LOAD_UINT32_REVERSED(p);
565             p++;
566         } while (--iters);
567     } else if (bpw == 8) {
568         UINT32 *p = (UINT32 *)buf;
569         UINT32 t;
570         do {
571             t = LOAD_UINT32_REVERSED(p+1);
572             p[1] = LOAD_UINT32_REVERSED(p);
573             p[0] = t;
574             p += 2;
575         } while (--iters);
576     }
577 }
578 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
579 #else
580 #define endian_convert_if_le(x,y,z) do{}while(0)  /* Do nothing */
581 #endif
582 
583 /* ---------------------------------------------------------------------- */
584 
nh_reset(nh_ctx * hc)585 static void nh_reset(nh_ctx *hc)
586 /* Reset nh_ctx to ready for hashing of new data */
587 {
588     hc->bytes_hashed = 0;
589     hc->next_data_empty = 0;
590     hc->state[0] = 0;
591 #if (UMAC_OUTPUT_LEN >= 8)
592     hc->state[1] = 0;
593 #endif
594 #if (UMAC_OUTPUT_LEN >= 12)
595     hc->state[2] = 0;
596 #endif
597 #if (UMAC_OUTPUT_LEN == 16)
598     hc->state[3] = 0;
599 #endif
600 
601 }
602 
603 /* ---------------------------------------------------------------------- */
604 
nh_init(nh_ctx * hc,aes_int_key prf_key)605 static void nh_init(nh_ctx *hc, aes_int_key prf_key)
606 /* Generate nh_key, endian convert and reset to be ready for hashing.   */
607 {
608     kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
609     endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
610     nh_reset(hc);
611 }
612 
613 /* ---------------------------------------------------------------------- */
614 
nh_update(nh_ctx * hc,const UINT8 * buf,UINT32 nbytes)615 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
616 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an    */
617 /* even multiple of HASH_BUF_BYTES.                                       */
618 {
619     UINT32 i,j;
620 
621     j = hc->next_data_empty;
622     if ((j + nbytes) >= HASH_BUF_BYTES) {
623         if (j) {
624             i = HASH_BUF_BYTES - j;
625             memcpy(hc->data+j, buf, i);
626             nh_transform(hc,hc->data,HASH_BUF_BYTES);
627             nbytes -= i;
628             buf += i;
629             hc->bytes_hashed += HASH_BUF_BYTES;
630         }
631         if (nbytes >= HASH_BUF_BYTES) {
632             i = nbytes & ~(HASH_BUF_BYTES - 1);
633             nh_transform(hc, buf, i);
634             nbytes -= i;
635             buf += i;
636             hc->bytes_hashed += i;
637         }
638         j = 0;
639     }
640     memcpy(hc->data + j, buf, nbytes);
641     hc->next_data_empty = j + nbytes;
642 }
643 
644 /* ---------------------------------------------------------------------- */
645 
zero_pad(UINT8 * p,int nbytes)646 static void zero_pad(UINT8 *p, int nbytes)
647 {
648 /* Write "nbytes" of zeroes, beginning at "p" */
649     if (nbytes >= (int)sizeof(UWORD)) {
650         while ((ptrdiff_t)p % sizeof(UWORD)) {
651             *p = 0;
652             nbytes--;
653             p++;
654         }
655         while (nbytes >= (int)sizeof(UWORD)) {
656             *(UWORD *)p = 0;
657             nbytes -= sizeof(UWORD);
658             p += sizeof(UWORD);
659         }
660     }
661     while (nbytes) {
662         *p = 0;
663         nbytes--;
664         p++;
665     }
666 }
667 
668 /* ---------------------------------------------------------------------- */
669 
nh_final(nh_ctx * hc,UINT8 * result)670 static void nh_final(nh_ctx *hc, UINT8 *result)
671 /* After passing some number of data buffers to nh_update() for integration
672  * into an NH context, nh_final is called to produce a hash result. If any
673  * bytes are in the buffer hc->data, incorporate them into the
674  * NH context. Finally, add into the NH accumulation "state" the total number
675  * of bits hashed. The resulting numbers are written to the buffer "result".
676  * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
677  */
678 {
679     int nh_len, nbits;
680 
681     if (hc->next_data_empty != 0) {
682         nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
683                                                 ~(L1_PAD_BOUNDARY - 1));
684         zero_pad(hc->data + hc->next_data_empty,
685                                           nh_len - hc->next_data_empty);
686         nh_transform(hc, hc->data, nh_len);
687         hc->bytes_hashed += hc->next_data_empty;
688     } else if (hc->bytes_hashed == 0) {
689 	nh_len = L1_PAD_BOUNDARY;
690         zero_pad(hc->data, L1_PAD_BOUNDARY);
691         nh_transform(hc, hc->data, nh_len);
692     }
693 
694     nbits = (hc->bytes_hashed << 3);
695     ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
696 #if (UMAC_OUTPUT_LEN >= 8)
697     ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
698 #endif
699 #if (UMAC_OUTPUT_LEN >= 12)
700     ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
701 #endif
702 #if (UMAC_OUTPUT_LEN == 16)
703     ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
704 #endif
705     nh_reset(hc);
706 }
707 
708 /* ---------------------------------------------------------------------- */
709 
nh(nh_ctx * hc,const UINT8 * buf,UINT32 padded_len,UINT32 unpadded_len,UINT8 * result)710 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
711                UINT32 unpadded_len, UINT8 *result)
712 /* All-in-one nh_update() and nh_final() equivalent.
713  * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
714  * well aligned
715  */
716 {
717     UINT32 nbits;
718 
719     /* Initialize the hash state */
720     nbits = (unpadded_len << 3);
721 
722     ((UINT64 *)result)[0] = nbits;
723 #if (UMAC_OUTPUT_LEN >= 8)
724     ((UINT64 *)result)[1] = nbits;
725 #endif
726 #if (UMAC_OUTPUT_LEN >= 12)
727     ((UINT64 *)result)[2] = nbits;
728 #endif
729 #if (UMAC_OUTPUT_LEN == 16)
730     ((UINT64 *)result)[3] = nbits;
731 #endif
732 
733     nh_aux(hc->nh_key, buf, result, padded_len);
734 }
735 
736 /* ---------------------------------------------------------------------- */
737 /* ---------------------------------------------------------------------- */
738 /* ----- Begin UHASH Section -------------------------------------------- */
739 /* ---------------------------------------------------------------------- */
740 /* ---------------------------------------------------------------------- */
741 
742 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
743  * hashed by NH. The NH output is then hashed by a polynomial-hash layer
744  * unless the initial data to be hashed is short. After the polynomial-
745  * layer, an inner-product hash is used to produce the final UHASH output.
746  *
747  * UHASH provides two interfaces, one all-at-once and another where data
748  * buffers are presented sequentially. In the sequential interface, the
749  * UHASH client calls the routine uhash_update() as many times as necessary.
750  * When there is no more data to be fed to UHASH, the client calls
751  * uhash_final() which
752  * calculates the UHASH output. Before beginning another UHASH calculation
753  * the uhash_reset() routine must be called. The all-at-once UHASH routine,
754  * uhash(), is equivalent to the sequence of calls uhash_update() and
755  * uhash_final(); however it is optimized and should be
756  * used whenever the sequential interface is not necessary.
757  *
758  * The routine uhash_init() initializes the uhash_ctx data structure and
759  * must be called once, before any other UHASH routine.
760  */
761 
762 /* ---------------------------------------------------------------------- */
763 /* ----- Constants and uhash_ctx ---------------------------------------- */
764 /* ---------------------------------------------------------------------- */
765 
766 /* ---------------------------------------------------------------------- */
767 /* ----- Poly hash and Inner-Product hash Constants --------------------- */
768 /* ---------------------------------------------------------------------- */
769 
770 /* Primes and masks */
771 #define p36    ((UINT64)0x0000000FFFFFFFFBull)              /* 2^36 -  5 */
772 #define p64    ((UINT64)0xFFFFFFFFFFFFFFC5ull)              /* 2^64 - 59 */
773 #define m36    ((UINT64)0x0000000FFFFFFFFFull)  /* The low 36 of 64 bits */
774 
775 
776 /* ---------------------------------------------------------------------- */
777 
778 typedef struct uhash_ctx {
779     nh_ctx hash;                          /* Hash context for L1 NH hash  */
780     UINT64 poly_key_8[STREAMS];           /* p64 poly keys                */
781     UINT64 poly_accum[STREAMS];           /* poly hash result             */
782     UINT64 ip_keys[STREAMS*4];            /* Inner-product keys           */
783     UINT32 ip_trans[STREAMS];             /* Inner-product translation    */
784     UINT32 msg_len;                       /* Total length of data passed  */
785                                           /* to uhash */
786 } uhash_ctx;
787 typedef struct uhash_ctx *uhash_ctx_t;
788 
789 /* ---------------------------------------------------------------------- */
790 
791 
792 /* The polynomial hashes use Horner's rule to evaluate a polynomial one
793  * word at a time. As described in the specification, poly32 and poly64
794  * require keys from special domains. The following implementations exploit
795  * the special domains to avoid overflow. The results are not guaranteed to
796  * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
797  * patches any errant values.
798  */
799 
poly64(UINT64 cur,UINT64 key,UINT64 data)800 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
801 {
802     UINT32 key_hi = (UINT32)(key >> 32),
803            key_lo = (UINT32)key,
804            cur_hi = (UINT32)(cur >> 32),
805            cur_lo = (UINT32)cur,
806            x_lo,
807            x_hi;
808     UINT64 X,T,res;
809 
810     X =  MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
811     x_lo = (UINT32)X;
812     x_hi = (UINT32)(X >> 32);
813 
814     res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
815 
816     T = ((UINT64)x_lo << 32);
817     res += T;
818     if (res < T)
819         res += 59;
820 
821     res += data;
822     if (res < data)
823         res += 59;
824 
825     return res;
826 }
827 
828 
829 /* Although UMAC is specified to use a ramped polynomial hash scheme, this
830  * implementation does not handle all ramp levels. Because we don't handle
831  * the ramp up to p128 modulus in this implementation, we are limited to
832  * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
833  * bytes input to UMAC per tag, ie. 16MB).
834  */
poly_hash(uhash_ctx_t hc,UINT32 data_in[])835 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
836 {
837     int i;
838     UINT64 *data=(UINT64*)data_in;
839 
840     for (i = 0; i < STREAMS; i++) {
841         if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
842             hc->poly_accum[i] = poly64(hc->poly_accum[i],
843                                        hc->poly_key_8[i], p64 - 1);
844             hc->poly_accum[i] = poly64(hc->poly_accum[i],
845                                        hc->poly_key_8[i], (data[i] - 59));
846         } else {
847             hc->poly_accum[i] = poly64(hc->poly_accum[i],
848                                        hc->poly_key_8[i], data[i]);
849         }
850     }
851 }
852 
853 
854 /* ---------------------------------------------------------------------- */
855 
856 
857 /* The final step in UHASH is an inner-product hash. The poly hash
858  * produces a result not necessarily WORD_LEN bytes long. The inner-
859  * product hash breaks the polyhash output into 16-bit chunks and
860  * multiplies each with a 36 bit key.
861  */
862 
ip_aux(UINT64 t,UINT64 * ipkp,UINT64 data)863 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
864 {
865     t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
866     t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
867     t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
868     t = t + ipkp[3] * (UINT64)(UINT16)(data);
869 
870     return t;
871 }
872 
ip_reduce_p36(UINT64 t)873 static UINT32 ip_reduce_p36(UINT64 t)
874 {
875 /* Divisionless modular reduction */
876     UINT64 ret;
877 
878     ret = (t & m36) + 5 * (t >> 36);
879     if (ret >= p36)
880         ret -= p36;
881 
882     /* return least significant 32 bits */
883     return (UINT32)(ret);
884 }
885 
886 
887 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
888  * the polyhash stage is skipped and ip_short is applied directly to the
889  * NH output.
890  */
ip_short(uhash_ctx_t ahc,UINT8 * nh_res,u_char * res)891 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
892 {
893     UINT64 t;
894     UINT64 *nhp = (UINT64 *)nh_res;
895 
896     t  = ip_aux(0,ahc->ip_keys, nhp[0]);
897     STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
898 #if (UMAC_OUTPUT_LEN >= 8)
899     t  = ip_aux(0,ahc->ip_keys+4, nhp[1]);
900     STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
901 #endif
902 #if (UMAC_OUTPUT_LEN >= 12)
903     t  = ip_aux(0,ahc->ip_keys+8, nhp[2]);
904     STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
905 #endif
906 #if (UMAC_OUTPUT_LEN == 16)
907     t  = ip_aux(0,ahc->ip_keys+12, nhp[3]);
908     STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
909 #endif
910 }
911 
912 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
913  * the polyhash stage is not skipped and ip_long is applied to the
914  * polyhash output.
915  */
ip_long(uhash_ctx_t ahc,u_char * res)916 static void ip_long(uhash_ctx_t ahc, u_char *res)
917 {
918     int i;
919     UINT64 t;
920 
921     for (i = 0; i < STREAMS; i++) {
922         /* fix polyhash output not in Z_p64 */
923         if (ahc->poly_accum[i] >= p64)
924             ahc->poly_accum[i] -= p64;
925         t  = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
926         STORE_UINT32_BIG((UINT32 *)res+i,
927                          ip_reduce_p36(t) ^ ahc->ip_trans[i]);
928     }
929 }
930 
931 
932 /* ---------------------------------------------------------------------- */
933 
934 /* ---------------------------------------------------------------------- */
935 
936 /* Reset uhash context for next hash session */
uhash_reset(uhash_ctx_t pc)937 static int uhash_reset(uhash_ctx_t pc)
938 {
939     nh_reset(&pc->hash);
940     pc->msg_len = 0;
941     pc->poly_accum[0] = 1;
942 #if (UMAC_OUTPUT_LEN >= 8)
943     pc->poly_accum[1] = 1;
944 #endif
945 #if (UMAC_OUTPUT_LEN >= 12)
946     pc->poly_accum[2] = 1;
947 #endif
948 #if (UMAC_OUTPUT_LEN == 16)
949     pc->poly_accum[3] = 1;
950 #endif
951     return 1;
952 }
953 
954 /* ---------------------------------------------------------------------- */
955 
956 /* Given a pointer to the internal key needed by kdf() and a uhash context,
957  * initialize the NH context and generate keys needed for poly and inner-
958  * product hashing. All keys are endian adjusted in memory so that native
959  * loads cause correct keys to be in registers during calculation.
960  */
uhash_init(uhash_ctx_t ahc,aes_int_key prf_key)961 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
962 {
963     int i;
964     UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
965 
966     /* Zero the entire uhash context */
967     memset(ahc, 0, sizeof(uhash_ctx));
968 
969     /* Initialize the L1 hash */
970     nh_init(&ahc->hash, prf_key);
971 
972     /* Setup L2 hash variables */
973     kdf(buf, prf_key, 2, sizeof(buf));    /* Fill buffer with index 1 key */
974     for (i = 0; i < STREAMS; i++) {
975         /* Fill keys from the buffer, skipping bytes in the buffer not
976          * used by this implementation. Endian reverse the keys if on a
977          * little-endian computer.
978          */
979         memcpy(ahc->poly_key_8+i, buf+24*i, 8);
980         endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
981         /* Mask the 64-bit keys to their special domain */
982         ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
983         ahc->poly_accum[i] = 1;  /* Our polyhash prepends a non-zero word */
984     }
985 
986     /* Setup L3-1 hash variables */
987     kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
988     for (i = 0; i < STREAMS; i++)
989           memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
990                                                  4*sizeof(UINT64));
991     endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
992                                                   sizeof(ahc->ip_keys));
993     for (i = 0; i < STREAMS*4; i++)
994         ahc->ip_keys[i] %= p36;  /* Bring into Z_p36 */
995 
996     /* Setup L3-2 hash variables    */
997     /* Fill buffer with index 4 key */
998     kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
999     endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
1000                          STREAMS * sizeof(UINT32));
1001     explicit_bzero(buf, sizeof(buf));
1002 }
1003 
1004 /* ---------------------------------------------------------------------- */
1005 
1006 #if 0
1007 static uhash_ctx_t uhash_alloc(u_char key[])
1008 {
1009 /* Allocate memory and force to a 16-byte boundary. */
1010     uhash_ctx_t ctx;
1011     u_char bytes_to_add;
1012     aes_int_key prf_key;
1013 
1014     ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1015     if (ctx) {
1016         if (ALLOC_BOUNDARY) {
1017             bytes_to_add = ALLOC_BOUNDARY -
1018                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1019             ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1020             *((u_char *)ctx - 1) = bytes_to_add;
1021         }
1022         aes_key_setup(key,prf_key);
1023         uhash_init(ctx, prf_key);
1024     }
1025     return (ctx);
1026 }
1027 #endif
1028 
1029 /* ---------------------------------------------------------------------- */
1030 
1031 #if 0
1032 static int uhash_free(uhash_ctx_t ctx)
1033 {
1034 /* Free memory allocated by uhash_alloc */
1035     u_char bytes_to_sub;
1036 
1037     if (ctx) {
1038         if (ALLOC_BOUNDARY) {
1039             bytes_to_sub = *((u_char *)ctx - 1);
1040             ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1041         }
1042         free(ctx);
1043     }
1044     return (1);
1045 }
1046 #endif
1047 /* ---------------------------------------------------------------------- */
1048 
uhash_update(uhash_ctx_t ctx,const u_char * input,long len)1049 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1050 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1051  * hash each one with NH, calling the polyhash on each NH output.
1052  */
1053 {
1054     UWORD bytes_hashed, bytes_remaining;
1055     UINT64 result_buf[STREAMS];
1056     UINT8 *nh_result = (UINT8 *)&result_buf;
1057 
1058     if (ctx->msg_len + len <= L1_KEY_LEN) {
1059         nh_update(&ctx->hash, (const UINT8 *)input, len);
1060         ctx->msg_len += len;
1061     } else {
1062 
1063          bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1064          if (ctx->msg_len == L1_KEY_LEN)
1065              bytes_hashed = L1_KEY_LEN;
1066 
1067          if (bytes_hashed + len >= L1_KEY_LEN) {
1068 
1069              /* If some bytes have been passed to the hash function      */
1070              /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1071              /* bytes to complete the current nh_block.                  */
1072              if (bytes_hashed) {
1073                  bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1074                  nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1075                  nh_final(&ctx->hash, nh_result);
1076                  ctx->msg_len += bytes_remaining;
1077                  poly_hash(ctx,(UINT32 *)nh_result);
1078                  len -= bytes_remaining;
1079                  input += bytes_remaining;
1080              }
1081 
1082              /* Hash directly from input stream if enough bytes */
1083              while (len >= L1_KEY_LEN) {
1084                  nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1085                                    L1_KEY_LEN, nh_result);
1086                  ctx->msg_len += L1_KEY_LEN;
1087                  len -= L1_KEY_LEN;
1088                  input += L1_KEY_LEN;
1089                  poly_hash(ctx,(UINT32 *)nh_result);
1090              }
1091          }
1092 
1093          /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1094          if (len > 0 && (unsigned long)len <= UINT32_MAX) {
1095              nh_update(&ctx->hash, (const UINT8 *)input, len);
1096              ctx->msg_len += len;
1097          }
1098      }
1099 
1100     return (1);
1101 }
1102 
1103 /* ---------------------------------------------------------------------- */
1104 
uhash_final(uhash_ctx_t ctx,u_char * res)1105 static int uhash_final(uhash_ctx_t ctx, u_char *res)
1106 /* Incorporate any pending data, pad, and generate tag */
1107 {
1108     UINT64 result_buf[STREAMS];
1109     UINT8 *nh_result = (UINT8 *)&result_buf;
1110 
1111     if (ctx->msg_len > L1_KEY_LEN) {
1112         if (ctx->msg_len % L1_KEY_LEN) {
1113             nh_final(&ctx->hash, nh_result);
1114             poly_hash(ctx,(UINT32 *)nh_result);
1115         }
1116         ip_long(ctx, res);
1117     } else {
1118         nh_final(&ctx->hash, nh_result);
1119         ip_short(ctx,nh_result, res);
1120     }
1121     uhash_reset(ctx);
1122     return (1);
1123 }
1124 
1125 /* ---------------------------------------------------------------------- */
1126 
1127 #if 0
1128 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1129 /* assumes that msg is in a writable buffer of length divisible by */
1130 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed.           */
1131 {
1132     UINT8 nh_result[STREAMS*sizeof(UINT64)];
1133     UINT32 nh_len;
1134     int extra_zeroes_needed;
1135 
1136     /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1137      * the polyhash.
1138      */
1139     if (len <= L1_KEY_LEN) {
1140 	if (len == 0)                  /* If zero length messages will not */
1141 		nh_len = L1_PAD_BOUNDARY;  /* be seen, comment out this case   */
1142 	else
1143 		nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1144         extra_zeroes_needed = nh_len - len;
1145         zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1146         nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1147         ip_short(ahc,nh_result, res);
1148     } else {
1149         /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1150          * output to poly_hash().
1151          */
1152         do {
1153             nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1154             poly_hash(ahc,(UINT32 *)nh_result);
1155             len -= L1_KEY_LEN;
1156             msg += L1_KEY_LEN;
1157         } while (len >= L1_KEY_LEN);
1158         if (len) {
1159             nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1160             extra_zeroes_needed = nh_len - len;
1161             zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1162             nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1163             poly_hash(ahc,(UINT32 *)nh_result);
1164         }
1165 
1166         ip_long(ahc, res);
1167     }
1168 
1169     uhash_reset(ahc);
1170     return 1;
1171 }
1172 #endif
1173 
1174 /* ---------------------------------------------------------------------- */
1175 /* ---------------------------------------------------------------------- */
1176 /* ----- Begin UMAC Section --------------------------------------------- */
1177 /* ---------------------------------------------------------------------- */
1178 /* ---------------------------------------------------------------------- */
1179 
1180 /* The UMAC interface has two interfaces, an all-at-once interface where
1181  * the entire message to be authenticated is passed to UMAC in one buffer,
1182  * and a sequential interface where the message is presented a little at a
1183  * time. The all-at-once is more optimized than the sequential version and
1184  * should be preferred when the sequential interface is not required.
1185  */
1186 struct umac_ctx {
1187     uhash_ctx hash;          /* Hash function for message compression    */
1188     pdf_ctx pdf;             /* PDF for hashed output                    */
1189     void *free_ptr;          /* Address to free this struct via          */
1190 } umac_ctx;
1191 
1192 /* ---------------------------------------------------------------------- */
1193 
1194 #if 0
1195 int umac_reset(struct umac_ctx *ctx)
1196 /* Reset the hash function to begin a new authentication.        */
1197 {
1198     uhash_reset(&ctx->hash);
1199     return (1);
1200 }
1201 #endif
1202 
1203 /* ---------------------------------------------------------------------- */
1204 
umac_delete(struct umac_ctx * ctx)1205 int umac_delete(struct umac_ctx *ctx)
1206 /* Deallocate the ctx structure */
1207 {
1208     if (ctx) {
1209         if (ALLOC_BOUNDARY)
1210             ctx = (struct umac_ctx *)ctx->free_ptr;
1211         freezero(ctx, sizeof(*ctx) + ALLOC_BOUNDARY);
1212     }
1213     return (1);
1214 }
1215 
1216 /* ---------------------------------------------------------------------- */
1217 
umac_new(const u_char key[])1218 struct umac_ctx *umac_new(const u_char key[])
1219 /* Dynamically allocate a umac_ctx struct, initialize variables,
1220  * generate subkeys from key. Align to 16-byte boundary.
1221  */
1222 {
1223     struct umac_ctx *ctx, *octx;
1224     size_t bytes_to_add;
1225     aes_int_key prf_key;
1226 
1227     octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1228     if (ctx) {
1229         if (ALLOC_BOUNDARY) {
1230             bytes_to_add = ALLOC_BOUNDARY -
1231                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1232             ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1233         }
1234         ctx->free_ptr = octx;
1235         aes_key_setup(key, prf_key);
1236         pdf_init(&ctx->pdf, prf_key);
1237         uhash_init(&ctx->hash, prf_key);
1238         explicit_bzero(prf_key, sizeof(prf_key));
1239     }
1240 
1241     return (ctx);
1242 }
1243 
1244 /* ---------------------------------------------------------------------- */
1245 
umac_final(struct umac_ctx * ctx,u_char tag[],const u_char nonce[8])1246 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1247 /* Incorporate any pending data, pad, and generate tag */
1248 {
1249     uhash_final(&ctx->hash, (u_char *)tag);
1250     pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1251 
1252     return (1);
1253 }
1254 
1255 /* ---------------------------------------------------------------------- */
1256 
umac_update(struct umac_ctx * ctx,const u_char * input,long len)1257 int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1258 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and   */
1259 /* hash each one, calling the PDF on the hashed output whenever the hash- */
1260 /* output buffer is full.                                                 */
1261 {
1262     uhash_update(&ctx->hash, input, len);
1263     return (1);
1264 }
1265 
1266 /* ---------------------------------------------------------------------- */
1267 
1268 #if 0
1269 int umac(struct umac_ctx *ctx, u_char *input,
1270          long len, u_char tag[],
1271          u_char nonce[8])
1272 /* All-in-one version simply calls umac_update() and umac_final().        */
1273 {
1274     uhash(&ctx->hash, input, len, (u_char *)tag);
1275     pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1276 
1277     return (1);
1278 }
1279 #endif
1280 
1281 /* ---------------------------------------------------------------------- */
1282 /* ---------------------------------------------------------------------- */
1283 /* ----- End UMAC Section ----------------------------------------------- */
1284 /* ---------------------------------------------------------------------- */
1285 /* ---------------------------------------------------------------------- */
1286