xref: /freebsd/crypto/openssh/umac.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
1 /* $OpenBSD: umac.c,v 1.11 2014/07/22 07:13:42 guenther Exp $ */
2 /* -----------------------------------------------------------------------
3  *
4  * umac.c -- C Implementation UMAC Message Authentication
5  *
6  * Version 0.93b of rfc4418.txt -- 2006 July 18
7  *
8  * For a full description of UMAC message authentication see the UMAC
9  * world-wide-web page at http://www.cs.ucdavis.edu/~rogaway/umac
10  * Please report bugs and suggestions to the UMAC webpage.
11  *
12  * Copyright (c) 1999-2006 Ted Krovetz
13  *
14  * Permission to use, copy, modify, and distribute this software and
15  * its documentation for any purpose and with or without fee, is hereby
16  * granted provided that the above copyright notice appears in all copies
17  * and in supporting documentation, and that the name of the copyright
18  * holder not be used in advertising or publicity pertaining to
19  * distribution of the software without specific, written prior permission.
20  *
21  * Comments should be directed to Ted Krovetz (tdk@acm.org)
22  *
23  * ---------------------------------------------------------------------- */
24 
25  /* ////////////////////// IMPORTANT NOTES /////////////////////////////////
26   *
27   * 1) This version does not work properly on messages larger than 16MB
28   *
29   * 2) If you set the switch to use SSE2, then all data must be 16-byte
30   *    aligned
31   *
32   * 3) When calling the function umac(), it is assumed that msg is in
33   * a writable buffer of length divisible by 32 bytes. The message itself
34   * does not have to fill the entire buffer, but bytes beyond msg may be
35   * zeroed.
36   *
37   * 4) Three free AES implementations are supported by this implementation of
38   * UMAC. Paulo Barreto's version is in the public domain and can be found
39   * at http://www.esat.kuleuven.ac.be/~rijmen/rijndael/ (search for
40   * "Barreto"). The only two files needed are rijndael-alg-fst.c and
41   * rijndael-alg-fst.h. Brian Gladman's version is distributed with the GNU
42   * Public lisence at http://fp.gladman.plus.com/AES/index.htm. It
43   * includes a fast IA-32 assembly version. The OpenSSL crypo library is
44   * the third.
45   *
46   * 5) With FORCE_C_ONLY flags set to 0, incorrect results are sometimes
47   * produced under gcc with optimizations set -O3 or higher. Dunno why.
48   *
49   /////////////////////////////////////////////////////////////////////// */
50 
51 /* ---------------------------------------------------------------------- */
52 /* --- User Switches ---------------------------------------------------- */
53 /* ---------------------------------------------------------------------- */
54 
55 #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 auhthenticated encrytion         */
69 
70 /* ---------------------------------------------------------------------- */
71 /* -- Global Includes --------------------------------------------------- */
72 /* ---------------------------------------------------------------------- */
73 
74 #include "includes.h"
75 #include <sys/types.h>
76 #include <string.h>
77 #include <stdio.h>
78 #include <stdlib.h>
79 #include <stddef.h>
80 
81 #include "xmalloc.h"
82 #include "umac.h"
83 #include "misc.h"
84 
85 /* ---------------------------------------------------------------------- */
86 /* --- Primitive Data Types ---                                           */
87 /* ---------------------------------------------------------------------- */
88 
89 /* The following assumptions may need change on your system */
90 typedef u_int8_t	UINT8;  /* 1 byte   */
91 typedef u_int16_t	UINT16; /* 2 byte   */
92 typedef u_int32_t	UINT32; /* 4 byte   */
93 typedef u_int64_t	UINT64; /* 8 bytes  */
94 typedef unsigned int	UWORD;  /* Register */
95 
96 /* ---------------------------------------------------------------------- */
97 /* --- Constants -------------------------------------------------------- */
98 /* ---------------------------------------------------------------------- */
99 
100 #define UMAC_KEY_LEN           16  /* UMAC takes 16 bytes of external key */
101 
102 /* Message "words" are read from memory in an endian-specific manner.     */
103 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must    */
104 /* be set true if the host computer is little-endian.                     */
105 
106 #if BYTE_ORDER == LITTLE_ENDIAN
107 #define __LITTLE_ENDIAN__ 1
108 #else
109 #define __LITTLE_ENDIAN__ 0
110 #endif
111 
112 /* ---------------------------------------------------------------------- */
113 /* ---------------------------------------------------------------------- */
114 /* ----- Architecture Specific ------------------------------------------ */
115 /* ---------------------------------------------------------------------- */
116 /* ---------------------------------------------------------------------- */
117 
118 
119 /* ---------------------------------------------------------------------- */
120 /* ---------------------------------------------------------------------- */
121 /* ----- Primitive Routines --------------------------------------------- */
122 /* ---------------------------------------------------------------------- */
123 /* ---------------------------------------------------------------------- */
124 
125 
126 /* ---------------------------------------------------------------------- */
127 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
128 /* ---------------------------------------------------------------------- */
129 
130 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
131 
132 /* ---------------------------------------------------------------------- */
133 /* --- Endian Conversion --- Forcing assembly on some platforms           */
134 /* ---------------------------------------------------------------------- */
135 
136 #if (__LITTLE_ENDIAN__)
137 #define LOAD_UINT32_REVERSED(p)		get_u32(p)
138 #define STORE_UINT32_REVERSED(p,v)	put_u32(p,v)
139 #else
140 #define LOAD_UINT32_REVERSED(p)		get_u32_le(p)
141 #define STORE_UINT32_REVERSED(p,v)	put_u32_le(p,v)
142 #endif
143 
144 #define LOAD_UINT32_LITTLE(p)		(get_u32_le(p))
145 #define STORE_UINT32_BIG(p,v)		put_u32(p, v)
146 
147 /* ---------------------------------------------------------------------- */
148 /* ---------------------------------------------------------------------- */
149 /* ----- Begin KDF & PDF Section ---------------------------------------- */
150 /* ---------------------------------------------------------------------- */
151 /* ---------------------------------------------------------------------- */
152 
153 /* UMAC uses AES with 16 byte block and key lengths */
154 #define AES_BLOCK_LEN  16
155 
156 /* OpenSSL's AES */
157 #ifdef WITH_OPENSSL
158 #include "openbsd-compat/openssl-compat.h"
159 #ifndef USE_BUILTIN_RIJNDAEL
160 # include <openssl/aes.h>
161 #endif
162 typedef AES_KEY aes_int_key[1];
163 #define aes_encryption(in,out,int_key)                  \
164   AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key)
165 #define aes_key_setup(key,int_key)                      \
166   AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key)
167 #else
168 #include "rijndael.h"
169 #define AES_ROUNDS ((UMAC_KEY_LEN / 4) + 6)
170 typedef UINT8 aes_int_key[AES_ROUNDS+1][4][4];	/* AES internal */
171 #define aes_encryption(in,out,int_key) \
172   rijndaelEncrypt((u32 *)(int_key), AES_ROUNDS, (u8 *)(in), (u8 *)(out))
173 #define aes_key_setup(key,int_key) \
174   rijndaelKeySetupEnc((u32 *)(int_key), (const unsigned char *)(key), \
175   UMAC_KEY_LEN*8)
176 #endif
177 
178 /* The user-supplied UMAC key is stretched using AES in a counter
179  * mode to supply all random bits needed by UMAC. The kdf function takes
180  * an AES internal key representation 'key' and writes a stream of
181  * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct
182  * 'ndx' causes a distinct byte stream.
183  */
184 static void kdf(void *bufp, aes_int_key key, UINT8 ndx, int nbytes)
185 {
186     UINT8 in_buf[AES_BLOCK_LEN] = {0};
187     UINT8 out_buf[AES_BLOCK_LEN];
188     UINT8 *dst_buf = (UINT8 *)bufp;
189     int i;
190 
191     /* Setup the initial value */
192     in_buf[AES_BLOCK_LEN-9] = ndx;
193     in_buf[AES_BLOCK_LEN-1] = i = 1;
194 
195     while (nbytes >= AES_BLOCK_LEN) {
196         aes_encryption(in_buf, out_buf, key);
197         memcpy(dst_buf,out_buf,AES_BLOCK_LEN);
198         in_buf[AES_BLOCK_LEN-1] = ++i;
199         nbytes -= AES_BLOCK_LEN;
200         dst_buf += AES_BLOCK_LEN;
201     }
202     if (nbytes) {
203         aes_encryption(in_buf, out_buf, key);
204         memcpy(dst_buf,out_buf,nbytes);
205     }
206 }
207 
208 /* The final UHASH result is XOR'd with the output of a pseudorandom
209  * function. Here, we use AES to generate random output and
210  * xor the appropriate bytes depending on the last bits of nonce.
211  * This scheme is optimized for sequential, increasing big-endian nonces.
212  */
213 
214 typedef struct {
215     UINT8 cache[AES_BLOCK_LEN];  /* Previous AES output is saved      */
216     UINT8 nonce[AES_BLOCK_LEN];  /* The AES input making above cache  */
217     aes_int_key prf_key;         /* Expanded AES key for PDF          */
218 } pdf_ctx;
219 
220 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
221 {
222     UINT8 buf[UMAC_KEY_LEN];
223 
224     kdf(buf, prf_key, 0, UMAC_KEY_LEN);
225     aes_key_setup(buf, pc->prf_key);
226 
227     /* Initialize pdf and cache */
228     memset(pc->nonce, 0, sizeof(pc->nonce));
229     aes_encryption(pc->nonce, pc->cache, pc->prf_key);
230 }
231 
232 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8])
233 {
234     /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
235      * of the AES output. If last time around we returned the ndx-1st
236      * element, then we may have the result in the cache already.
237      */
238 
239 #if (UMAC_OUTPUT_LEN == 4)
240 #define LOW_BIT_MASK 3
241 #elif (UMAC_OUTPUT_LEN == 8)
242 #define LOW_BIT_MASK 1
243 #elif (UMAC_OUTPUT_LEN > 8)
244 #define LOW_BIT_MASK 0
245 #endif
246     union {
247         UINT8 tmp_nonce_lo[4];
248         UINT32 align;
249     } t;
250 #if LOW_BIT_MASK != 0
251     int ndx = nonce[7] & LOW_BIT_MASK;
252 #endif
253     *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
254     t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
255 
256     if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
257          (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
258     {
259         ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
260         ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
261         aes_encryption(pc->nonce, pc->cache, pc->prf_key);
262     }
263 
264 #if (UMAC_OUTPUT_LEN == 4)
265     *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
266 #elif (UMAC_OUTPUT_LEN == 8)
267     *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
268 #elif (UMAC_OUTPUT_LEN == 12)
269     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
270     ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
271 #elif (UMAC_OUTPUT_LEN == 16)
272     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
273     ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
274 #endif
275 }
276 
277 /* ---------------------------------------------------------------------- */
278 /* ---------------------------------------------------------------------- */
279 /* ----- Begin NH Hash Section ------------------------------------------ */
280 /* ---------------------------------------------------------------------- */
281 /* ---------------------------------------------------------------------- */
282 
283 /* The NH-based hash functions used in UMAC are described in the UMAC paper
284  * and specification, both of which can be found at the UMAC website.
285  * The interface to this implementation has two
286  * versions, one expects the entire message being hashed to be passed
287  * in a single buffer and returns the hash result immediately. The second
288  * allows the message to be passed in a sequence of buffers. In the
289  * muliple-buffer interface, the client calls the routine nh_update() as
290  * many times as necessary. When there is no more data to be fed to the
291  * hash, the client calls nh_final() which calculates the hash output.
292  * Before beginning another hash calculation the nh_reset() routine
293  * must be called. The single-buffer routine, nh(), is equivalent to
294  * the sequence of calls nh_update() and nh_final(); however it is
295  * optimized and should be prefered whenever the multiple-buffer interface
296  * is not necessary. When using either interface, it is the client's
297  * responsability to pass no more than L1_KEY_LEN bytes per hash result.
298  *
299  * The routine nh_init() initializes the nh_ctx data structure and
300  * must be called once, before any other PDF routine.
301  */
302 
303  /* The "nh_aux" routines do the actual NH hashing work. They
304   * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
305   * produce output for all STREAMS NH iterations in one call,
306   * allowing the parallel implementation of the streams.
307   */
308 
309 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied  */
310 #define L1_KEY_LEN         1024     /* Internal key bytes                 */
311 #define L1_KEY_SHIFT         16     /* Toeplitz key shift between streams */
312 #define L1_PAD_BOUNDARY      32     /* pad message to boundary multiple   */
313 #define ALLOC_BOUNDARY       16     /* Keep buffers aligned to this       */
314 #define HASH_BUF_BYTES       64     /* nh_aux_hb buffer multiple          */
315 
316 typedef struct {
317     UINT8  nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
318     UINT8  data   [HASH_BUF_BYTES];    /* Incoming data buffer           */
319     int next_data_empty;    /* Bookeeping variable for data buffer.       */
320     int bytes_hashed;        /* Bytes (out of L1_KEY_LEN) incorperated.   */
321     UINT64 state[STREAMS];               /* on-line state     */
322 } nh_ctx;
323 
324 
325 #if (UMAC_OUTPUT_LEN == 4)
326 
327 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
328 /* NH hashing primitive. Previous (partial) hash result is loaded and
329 * then stored via hp pointer. The length of the data pointed at by "dp",
330 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32).  Key
331 * is expected to be endian compensated in memory at key setup.
332 */
333 {
334     UINT64 h;
335     UWORD c = dlen / 32;
336     UINT32 *k = (UINT32 *)kp;
337     const UINT32 *d = (const UINT32 *)dp;
338     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
339     UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
340 
341     h = *((UINT64 *)hp);
342     do {
343         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
344         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
345         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
346         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
347         k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
348         k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
349         h += MUL64((k0 + d0), (k4 + d4));
350         h += MUL64((k1 + d1), (k5 + d5));
351         h += MUL64((k2 + d2), (k6 + d6));
352         h += MUL64((k3 + d3), (k7 + d7));
353 
354         d += 8;
355         k += 8;
356     } while (--c);
357   *((UINT64 *)hp) = h;
358 }
359 
360 #elif (UMAC_OUTPUT_LEN == 8)
361 
362 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
363 /* Same as previous nh_aux, but two streams are handled in one pass,
364  * reading and writing 16 bytes of hash-state per call.
365  */
366 {
367   UINT64 h1,h2;
368   UWORD c = dlen / 32;
369   UINT32 *k = (UINT32 *)kp;
370   const UINT32 *d = (const UINT32 *)dp;
371   UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
372   UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
373         k8,k9,k10,k11;
374 
375   h1 = *((UINT64 *)hp);
376   h2 = *((UINT64 *)hp + 1);
377   k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
378   do {
379     d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
380     d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
381     d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
382     d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
383     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
384     k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
385 
386     h1 += MUL64((k0 + d0), (k4 + d4));
387     h2 += MUL64((k4 + d0), (k8 + d4));
388 
389     h1 += MUL64((k1 + d1), (k5 + d5));
390     h2 += MUL64((k5 + d1), (k9 + d5));
391 
392     h1 += MUL64((k2 + d2), (k6 + d6));
393     h2 += MUL64((k6 + d2), (k10 + d6));
394 
395     h1 += MUL64((k3 + d3), (k7 + d7));
396     h2 += MUL64((k7 + d3), (k11 + d7));
397 
398     k0 = k8; k1 = k9; k2 = k10; k3 = k11;
399 
400     d += 8;
401     k += 8;
402   } while (--c);
403   ((UINT64 *)hp)[0] = h1;
404   ((UINT64 *)hp)[1] = h2;
405 }
406 
407 #elif (UMAC_OUTPUT_LEN == 12)
408 
409 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
410 /* Same as previous nh_aux, but two streams are handled in one pass,
411  * reading and writing 24 bytes of hash-state per call.
412 */
413 {
414     UINT64 h1,h2,h3;
415     UWORD c = dlen / 32;
416     UINT32 *k = (UINT32 *)kp;
417     const UINT32 *d = (const UINT32 *)dp;
418     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
419     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
420         k8,k9,k10,k11,k12,k13,k14,k15;
421 
422     h1 = *((UINT64 *)hp);
423     h2 = *((UINT64 *)hp + 1);
424     h3 = *((UINT64 *)hp + 2);
425     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
426     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
427     do {
428         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
429         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
430         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
431         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
432         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
433         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
434 
435         h1 += MUL64((k0 + d0), (k4 + d4));
436         h2 += MUL64((k4 + d0), (k8 + d4));
437         h3 += MUL64((k8 + d0), (k12 + d4));
438 
439         h1 += MUL64((k1 + d1), (k5 + d5));
440         h2 += MUL64((k5 + d1), (k9 + d5));
441         h3 += MUL64((k9 + d1), (k13 + d5));
442 
443         h1 += MUL64((k2 + d2), (k6 + d6));
444         h2 += MUL64((k6 + d2), (k10 + d6));
445         h3 += MUL64((k10 + d2), (k14 + d6));
446 
447         h1 += MUL64((k3 + d3), (k7 + d7));
448         h2 += MUL64((k7 + d3), (k11 + d7));
449         h3 += MUL64((k11 + d3), (k15 + d7));
450 
451         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
452         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
453 
454         d += 8;
455         k += 8;
456     } while (--c);
457     ((UINT64 *)hp)[0] = h1;
458     ((UINT64 *)hp)[1] = h2;
459     ((UINT64 *)hp)[2] = h3;
460 }
461 
462 #elif (UMAC_OUTPUT_LEN == 16)
463 
464 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
465 /* Same as previous nh_aux, but two streams are handled in one pass,
466  * reading and writing 24 bytes of hash-state per call.
467 */
468 {
469     UINT64 h1,h2,h3,h4;
470     UWORD c = dlen / 32;
471     UINT32 *k = (UINT32 *)kp;
472     const UINT32 *d = (const UINT32 *)dp;
473     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
474     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
475         k8,k9,k10,k11,k12,k13,k14,k15,
476         k16,k17,k18,k19;
477 
478     h1 = *((UINT64 *)hp);
479     h2 = *((UINT64 *)hp + 1);
480     h3 = *((UINT64 *)hp + 2);
481     h4 = *((UINT64 *)hp + 3);
482     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
483     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
484     do {
485         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
486         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
487         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
488         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
489         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
490         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
491         k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
492 
493         h1 += MUL64((k0 + d0), (k4 + d4));
494         h2 += MUL64((k4 + d0), (k8 + d4));
495         h3 += MUL64((k8 + d0), (k12 + d4));
496         h4 += MUL64((k12 + d0), (k16 + d4));
497 
498         h1 += MUL64((k1 + d1), (k5 + d5));
499         h2 += MUL64((k5 + d1), (k9 + d5));
500         h3 += MUL64((k9 + d1), (k13 + d5));
501         h4 += MUL64((k13 + d1), (k17 + d5));
502 
503         h1 += MUL64((k2 + d2), (k6 + d6));
504         h2 += MUL64((k6 + d2), (k10 + d6));
505         h3 += MUL64((k10 + d2), (k14 + d6));
506         h4 += MUL64((k14 + d2), (k18 + d6));
507 
508         h1 += MUL64((k3 + d3), (k7 + d7));
509         h2 += MUL64((k7 + d3), (k11 + d7));
510         h3 += MUL64((k11 + d3), (k15 + d7));
511         h4 += MUL64((k15 + d3), (k19 + d7));
512 
513         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
514         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
515         k8 = k16; k9 = k17; k10 = k18; k11 = k19;
516 
517         d += 8;
518         k += 8;
519     } while (--c);
520     ((UINT64 *)hp)[0] = h1;
521     ((UINT64 *)hp)[1] = h2;
522     ((UINT64 *)hp)[2] = h3;
523     ((UINT64 *)hp)[3] = h4;
524 }
525 
526 /* ---------------------------------------------------------------------- */
527 #endif  /* UMAC_OUTPUT_LENGTH */
528 /* ---------------------------------------------------------------------- */
529 
530 
531 /* ---------------------------------------------------------------------- */
532 
533 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
534 /* This function is a wrapper for the primitive NH hash functions. It takes
535  * as argument "hc" the current hash context and a buffer which must be a
536  * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
537  * appropriately according to how much message has been hashed already.
538  */
539 {
540     UINT8 *key;
541 
542     key = hc->nh_key + hc->bytes_hashed;
543     nh_aux(key, buf, hc->state, nbytes);
544 }
545 
546 /* ---------------------------------------------------------------------- */
547 
548 #if (__LITTLE_ENDIAN__)
549 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
550 /* We endian convert the keys on little-endian computers to               */
551 /* compensate for the lack of big-endian memory reads during hashing.     */
552 {
553     UWORD iters = num_bytes / bpw;
554     if (bpw == 4) {
555         UINT32 *p = (UINT32 *)buf;
556         do {
557             *p = LOAD_UINT32_REVERSED(p);
558             p++;
559         } while (--iters);
560     } else if (bpw == 8) {
561         UINT32 *p = (UINT32 *)buf;
562         UINT32 t;
563         do {
564             t = LOAD_UINT32_REVERSED(p+1);
565             p[1] = LOAD_UINT32_REVERSED(p);
566             p[0] = t;
567             p += 2;
568         } while (--iters);
569     }
570 }
571 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
572 #else
573 #define endian_convert_if_le(x,y,z) do{}while(0)  /* Do nothing */
574 #endif
575 
576 /* ---------------------------------------------------------------------- */
577 
578 static void nh_reset(nh_ctx *hc)
579 /* Reset nh_ctx to ready for hashing of new data */
580 {
581     hc->bytes_hashed = 0;
582     hc->next_data_empty = 0;
583     hc->state[0] = 0;
584 #if (UMAC_OUTPUT_LEN >= 8)
585     hc->state[1] = 0;
586 #endif
587 #if (UMAC_OUTPUT_LEN >= 12)
588     hc->state[2] = 0;
589 #endif
590 #if (UMAC_OUTPUT_LEN == 16)
591     hc->state[3] = 0;
592 #endif
593 
594 }
595 
596 /* ---------------------------------------------------------------------- */
597 
598 static void nh_init(nh_ctx *hc, aes_int_key prf_key)
599 /* Generate nh_key, endian convert and reset to be ready for hashing.   */
600 {
601     kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
602     endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
603     nh_reset(hc);
604 }
605 
606 /* ---------------------------------------------------------------------- */
607 
608 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
609 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an    */
610 /* even multiple of HASH_BUF_BYTES.                                       */
611 {
612     UINT32 i,j;
613 
614     j = hc->next_data_empty;
615     if ((j + nbytes) >= HASH_BUF_BYTES) {
616         if (j) {
617             i = HASH_BUF_BYTES - j;
618             memcpy(hc->data+j, buf, i);
619             nh_transform(hc,hc->data,HASH_BUF_BYTES);
620             nbytes -= i;
621             buf += i;
622             hc->bytes_hashed += HASH_BUF_BYTES;
623         }
624         if (nbytes >= HASH_BUF_BYTES) {
625             i = nbytes & ~(HASH_BUF_BYTES - 1);
626             nh_transform(hc, buf, i);
627             nbytes -= i;
628             buf += i;
629             hc->bytes_hashed += i;
630         }
631         j = 0;
632     }
633     memcpy(hc->data + j, buf, nbytes);
634     hc->next_data_empty = j + nbytes;
635 }
636 
637 /* ---------------------------------------------------------------------- */
638 
639 static void zero_pad(UINT8 *p, int nbytes)
640 {
641 /* Write "nbytes" of zeroes, beginning at "p" */
642     if (nbytes >= (int)sizeof(UWORD)) {
643         while ((ptrdiff_t)p % sizeof(UWORD)) {
644             *p = 0;
645             nbytes--;
646             p++;
647         }
648         while (nbytes >= (int)sizeof(UWORD)) {
649             *(UWORD *)p = 0;
650             nbytes -= sizeof(UWORD);
651             p += sizeof(UWORD);
652         }
653     }
654     while (nbytes) {
655         *p = 0;
656         nbytes--;
657         p++;
658     }
659 }
660 
661 /* ---------------------------------------------------------------------- */
662 
663 static void nh_final(nh_ctx *hc, UINT8 *result)
664 /* After passing some number of data buffers to nh_update() for integration
665  * into an NH context, nh_final is called to produce a hash result. If any
666  * bytes are in the buffer hc->data, incorporate them into the
667  * NH context. Finally, add into the NH accumulation "state" the total number
668  * of bits hashed. The resulting numbers are written to the buffer "result".
669  * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
670  */
671 {
672     int nh_len, nbits;
673 
674     if (hc->next_data_empty != 0) {
675         nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
676                                                 ~(L1_PAD_BOUNDARY - 1));
677         zero_pad(hc->data + hc->next_data_empty,
678                                           nh_len - hc->next_data_empty);
679         nh_transform(hc, hc->data, nh_len);
680         hc->bytes_hashed += hc->next_data_empty;
681     } else if (hc->bytes_hashed == 0) {
682     	nh_len = L1_PAD_BOUNDARY;
683         zero_pad(hc->data, L1_PAD_BOUNDARY);
684         nh_transform(hc, hc->data, nh_len);
685     }
686 
687     nbits = (hc->bytes_hashed << 3);
688     ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
689 #if (UMAC_OUTPUT_LEN >= 8)
690     ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
691 #endif
692 #if (UMAC_OUTPUT_LEN >= 12)
693     ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
694 #endif
695 #if (UMAC_OUTPUT_LEN == 16)
696     ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
697 #endif
698     nh_reset(hc);
699 }
700 
701 /* ---------------------------------------------------------------------- */
702 
703 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
704                UINT32 unpadded_len, UINT8 *result)
705 /* All-in-one nh_update() and nh_final() equivalent.
706  * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
707  * well aligned
708  */
709 {
710     UINT32 nbits;
711 
712     /* Initialize the hash state */
713     nbits = (unpadded_len << 3);
714 
715     ((UINT64 *)result)[0] = nbits;
716 #if (UMAC_OUTPUT_LEN >= 8)
717     ((UINT64 *)result)[1] = nbits;
718 #endif
719 #if (UMAC_OUTPUT_LEN >= 12)
720     ((UINT64 *)result)[2] = nbits;
721 #endif
722 #if (UMAC_OUTPUT_LEN == 16)
723     ((UINT64 *)result)[3] = nbits;
724 #endif
725 
726     nh_aux(hc->nh_key, buf, result, padded_len);
727 }
728 
729 /* ---------------------------------------------------------------------- */
730 /* ---------------------------------------------------------------------- */
731 /* ----- Begin UHASH Section -------------------------------------------- */
732 /* ---------------------------------------------------------------------- */
733 /* ---------------------------------------------------------------------- */
734 
735 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
736  * hashed by NH. The NH output is then hashed by a polynomial-hash layer
737  * unless the initial data to be hashed is short. After the polynomial-
738  * layer, an inner-product hash is used to produce the final UHASH output.
739  *
740  * UHASH provides two interfaces, one all-at-once and another where data
741  * buffers are presented sequentially. In the sequential interface, the
742  * UHASH client calls the routine uhash_update() as many times as necessary.
743  * When there is no more data to be fed to UHASH, the client calls
744  * uhash_final() which
745  * calculates the UHASH output. Before beginning another UHASH calculation
746  * the uhash_reset() routine must be called. The all-at-once UHASH routine,
747  * uhash(), is equivalent to the sequence of calls uhash_update() and
748  * uhash_final(); however it is optimized and should be
749  * used whenever the sequential interface is not necessary.
750  *
751  * The routine uhash_init() initializes the uhash_ctx data structure and
752  * must be called once, before any other UHASH routine.
753  */
754 
755 /* ---------------------------------------------------------------------- */
756 /* ----- Constants and uhash_ctx ---------------------------------------- */
757 /* ---------------------------------------------------------------------- */
758 
759 /* ---------------------------------------------------------------------- */
760 /* ----- Poly hash and Inner-Product hash Constants --------------------- */
761 /* ---------------------------------------------------------------------- */
762 
763 /* Primes and masks */
764 #define p36    ((UINT64)0x0000000FFFFFFFFBull)              /* 2^36 -  5 */
765 #define p64    ((UINT64)0xFFFFFFFFFFFFFFC5ull)              /* 2^64 - 59 */
766 #define m36    ((UINT64)0x0000000FFFFFFFFFull)  /* The low 36 of 64 bits */
767 
768 
769 /* ---------------------------------------------------------------------- */
770 
771 typedef struct uhash_ctx {
772     nh_ctx hash;                          /* Hash context for L1 NH hash  */
773     UINT64 poly_key_8[STREAMS];           /* p64 poly keys                */
774     UINT64 poly_accum[STREAMS];           /* poly hash result             */
775     UINT64 ip_keys[STREAMS*4];            /* Inner-product keys           */
776     UINT32 ip_trans[STREAMS];             /* Inner-product translation    */
777     UINT32 msg_len;                       /* Total length of data passed  */
778                                           /* to uhash */
779 } uhash_ctx;
780 typedef struct uhash_ctx *uhash_ctx_t;
781 
782 /* ---------------------------------------------------------------------- */
783 
784 
785 /* The polynomial hashes use Horner's rule to evaluate a polynomial one
786  * word at a time. As described in the specification, poly32 and poly64
787  * require keys from special domains. The following implementations exploit
788  * the special domains to avoid overflow. The results are not guaranteed to
789  * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
790  * patches any errant values.
791  */
792 
793 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
794 {
795     UINT32 key_hi = (UINT32)(key >> 32),
796            key_lo = (UINT32)key,
797            cur_hi = (UINT32)(cur >> 32),
798            cur_lo = (UINT32)cur,
799            x_lo,
800            x_hi;
801     UINT64 X,T,res;
802 
803     X =  MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
804     x_lo = (UINT32)X;
805     x_hi = (UINT32)(X >> 32);
806 
807     res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
808 
809     T = ((UINT64)x_lo << 32);
810     res += T;
811     if (res < T)
812         res += 59;
813 
814     res += data;
815     if (res < data)
816         res += 59;
817 
818     return res;
819 }
820 
821 
822 /* Although UMAC is specified to use a ramped polynomial hash scheme, this
823  * implementation does not handle all ramp levels. Because we don't handle
824  * the ramp up to p128 modulus in this implementation, we are limited to
825  * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
826  * bytes input to UMAC per tag, ie. 16MB).
827  */
828 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
829 {
830     int i;
831     UINT64 *data=(UINT64*)data_in;
832 
833     for (i = 0; i < STREAMS; i++) {
834         if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
835             hc->poly_accum[i] = poly64(hc->poly_accum[i],
836                                        hc->poly_key_8[i], p64 - 1);
837             hc->poly_accum[i] = poly64(hc->poly_accum[i],
838                                        hc->poly_key_8[i], (data[i] - 59));
839         } else {
840             hc->poly_accum[i] = poly64(hc->poly_accum[i],
841                                        hc->poly_key_8[i], data[i]);
842         }
843     }
844 }
845 
846 
847 /* ---------------------------------------------------------------------- */
848 
849 
850 /* The final step in UHASH is an inner-product hash. The poly hash
851  * produces a result not neccesarily WORD_LEN bytes long. The inner-
852  * product hash breaks the polyhash output into 16-bit chunks and
853  * multiplies each with a 36 bit key.
854  */
855 
856 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
857 {
858     t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
859     t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
860     t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
861     t = t + ipkp[3] * (UINT64)(UINT16)(data);
862 
863     return t;
864 }
865 
866 static UINT32 ip_reduce_p36(UINT64 t)
867 {
868 /* Divisionless modular reduction */
869     UINT64 ret;
870 
871     ret = (t & m36) + 5 * (t >> 36);
872     if (ret >= p36)
873         ret -= p36;
874 
875     /* return least significant 32 bits */
876     return (UINT32)(ret);
877 }
878 
879 
880 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
881  * the polyhash stage is skipped and ip_short is applied directly to the
882  * NH output.
883  */
884 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
885 {
886     UINT64 t;
887     UINT64 *nhp = (UINT64 *)nh_res;
888 
889     t  = ip_aux(0,ahc->ip_keys, nhp[0]);
890     STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
891 #if (UMAC_OUTPUT_LEN >= 8)
892     t  = ip_aux(0,ahc->ip_keys+4, nhp[1]);
893     STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
894 #endif
895 #if (UMAC_OUTPUT_LEN >= 12)
896     t  = ip_aux(0,ahc->ip_keys+8, nhp[2]);
897     STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
898 #endif
899 #if (UMAC_OUTPUT_LEN == 16)
900     t  = ip_aux(0,ahc->ip_keys+12, nhp[3]);
901     STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
902 #endif
903 }
904 
905 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
906  * the polyhash stage is not skipped and ip_long is applied to the
907  * polyhash output.
908  */
909 static void ip_long(uhash_ctx_t ahc, u_char *res)
910 {
911     int i;
912     UINT64 t;
913 
914     for (i = 0; i < STREAMS; i++) {
915         /* fix polyhash output not in Z_p64 */
916         if (ahc->poly_accum[i] >= p64)
917             ahc->poly_accum[i] -= p64;
918         t  = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
919         STORE_UINT32_BIG((UINT32 *)res+i,
920                          ip_reduce_p36(t) ^ ahc->ip_trans[i]);
921     }
922 }
923 
924 
925 /* ---------------------------------------------------------------------- */
926 
927 /* ---------------------------------------------------------------------- */
928 
929 /* Reset uhash context for next hash session */
930 static int uhash_reset(uhash_ctx_t pc)
931 {
932     nh_reset(&pc->hash);
933     pc->msg_len = 0;
934     pc->poly_accum[0] = 1;
935 #if (UMAC_OUTPUT_LEN >= 8)
936     pc->poly_accum[1] = 1;
937 #endif
938 #if (UMAC_OUTPUT_LEN >= 12)
939     pc->poly_accum[2] = 1;
940 #endif
941 #if (UMAC_OUTPUT_LEN == 16)
942     pc->poly_accum[3] = 1;
943 #endif
944     return 1;
945 }
946 
947 /* ---------------------------------------------------------------------- */
948 
949 /* Given a pointer to the internal key needed by kdf() and a uhash context,
950  * initialize the NH context and generate keys needed for poly and inner-
951  * product hashing. All keys are endian adjusted in memory so that native
952  * loads cause correct keys to be in registers during calculation.
953  */
954 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
955 {
956     int i;
957     UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
958 
959     /* Zero the entire uhash context */
960     memset(ahc, 0, sizeof(uhash_ctx));
961 
962     /* Initialize the L1 hash */
963     nh_init(&ahc->hash, prf_key);
964 
965     /* Setup L2 hash variables */
966     kdf(buf, prf_key, 2, sizeof(buf));    /* Fill buffer with index 1 key */
967     for (i = 0; i < STREAMS; i++) {
968         /* Fill keys from the buffer, skipping bytes in the buffer not
969          * used by this implementation. Endian reverse the keys if on a
970          * little-endian computer.
971          */
972         memcpy(ahc->poly_key_8+i, buf+24*i, 8);
973         endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
974         /* Mask the 64-bit keys to their special domain */
975         ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
976         ahc->poly_accum[i] = 1;  /* Our polyhash prepends a non-zero word */
977     }
978 
979     /* Setup L3-1 hash variables */
980     kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
981     for (i = 0; i < STREAMS; i++)
982           memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
983                                                  4*sizeof(UINT64));
984     endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
985                                                   sizeof(ahc->ip_keys));
986     for (i = 0; i < STREAMS*4; i++)
987         ahc->ip_keys[i] %= p36;  /* Bring into Z_p36 */
988 
989     /* Setup L3-2 hash variables    */
990     /* Fill buffer with index 4 key */
991     kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
992     endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
993                          STREAMS * sizeof(UINT32));
994 }
995 
996 /* ---------------------------------------------------------------------- */
997 
998 #if 0
999 static uhash_ctx_t uhash_alloc(u_char key[])
1000 {
1001 /* Allocate memory and force to a 16-byte boundary. */
1002     uhash_ctx_t ctx;
1003     u_char bytes_to_add;
1004     aes_int_key prf_key;
1005 
1006     ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1007     if (ctx) {
1008         if (ALLOC_BOUNDARY) {
1009             bytes_to_add = ALLOC_BOUNDARY -
1010                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1011             ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1012             *((u_char *)ctx - 1) = bytes_to_add;
1013         }
1014         aes_key_setup(key,prf_key);
1015         uhash_init(ctx, prf_key);
1016     }
1017     return (ctx);
1018 }
1019 #endif
1020 
1021 /* ---------------------------------------------------------------------- */
1022 
1023 #if 0
1024 static int uhash_free(uhash_ctx_t ctx)
1025 {
1026 /* Free memory allocated by uhash_alloc */
1027     u_char bytes_to_sub;
1028 
1029     if (ctx) {
1030         if (ALLOC_BOUNDARY) {
1031             bytes_to_sub = *((u_char *)ctx - 1);
1032             ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1033         }
1034         free(ctx);
1035     }
1036     return (1);
1037 }
1038 #endif
1039 /* ---------------------------------------------------------------------- */
1040 
1041 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1042 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1043  * hash each one with NH, calling the polyhash on each NH output.
1044  */
1045 {
1046     UWORD bytes_hashed, bytes_remaining;
1047     UINT64 result_buf[STREAMS];
1048     UINT8 *nh_result = (UINT8 *)&result_buf;
1049 
1050     if (ctx->msg_len + len <= L1_KEY_LEN) {
1051         nh_update(&ctx->hash, (const UINT8 *)input, len);
1052         ctx->msg_len += len;
1053     } else {
1054 
1055          bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1056          if (ctx->msg_len == L1_KEY_LEN)
1057              bytes_hashed = L1_KEY_LEN;
1058 
1059          if (bytes_hashed + len >= L1_KEY_LEN) {
1060 
1061              /* If some bytes have been passed to the hash function      */
1062              /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1063              /* bytes to complete the current nh_block.                  */
1064              if (bytes_hashed) {
1065                  bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1066                  nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1067                  nh_final(&ctx->hash, nh_result);
1068                  ctx->msg_len += bytes_remaining;
1069                  poly_hash(ctx,(UINT32 *)nh_result);
1070                  len -= bytes_remaining;
1071                  input += bytes_remaining;
1072              }
1073 
1074              /* Hash directly from input stream if enough bytes */
1075              while (len >= L1_KEY_LEN) {
1076                  nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1077                                    L1_KEY_LEN, nh_result);
1078                  ctx->msg_len += L1_KEY_LEN;
1079                  len -= L1_KEY_LEN;
1080                  input += L1_KEY_LEN;
1081                  poly_hash(ctx,(UINT32 *)nh_result);
1082              }
1083          }
1084 
1085          /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1086          if (len) {
1087              nh_update(&ctx->hash, (const UINT8 *)input, len);
1088              ctx->msg_len += len;
1089          }
1090      }
1091 
1092     return (1);
1093 }
1094 
1095 /* ---------------------------------------------------------------------- */
1096 
1097 static int uhash_final(uhash_ctx_t ctx, u_char *res)
1098 /* Incorporate any pending data, pad, and generate tag */
1099 {
1100     UINT64 result_buf[STREAMS];
1101     UINT8 *nh_result = (UINT8 *)&result_buf;
1102 
1103     if (ctx->msg_len > L1_KEY_LEN) {
1104         if (ctx->msg_len % L1_KEY_LEN) {
1105             nh_final(&ctx->hash, nh_result);
1106             poly_hash(ctx,(UINT32 *)nh_result);
1107         }
1108         ip_long(ctx, res);
1109     } else {
1110         nh_final(&ctx->hash, nh_result);
1111         ip_short(ctx,nh_result, res);
1112     }
1113     uhash_reset(ctx);
1114     return (1);
1115 }
1116 
1117 /* ---------------------------------------------------------------------- */
1118 
1119 #if 0
1120 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1121 /* assumes that msg is in a writable buffer of length divisible by */
1122 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed.           */
1123 {
1124     UINT8 nh_result[STREAMS*sizeof(UINT64)];
1125     UINT32 nh_len;
1126     int extra_zeroes_needed;
1127 
1128     /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1129      * the polyhash.
1130      */
1131     if (len <= L1_KEY_LEN) {
1132     	if (len == 0)                  /* If zero length messages will not */
1133     		nh_len = L1_PAD_BOUNDARY;  /* be seen, comment out this case   */
1134     	else
1135         	nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1136         extra_zeroes_needed = nh_len - len;
1137         zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1138         nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1139         ip_short(ahc,nh_result, res);
1140     } else {
1141         /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1142          * output to poly_hash().
1143          */
1144         do {
1145             nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1146             poly_hash(ahc,(UINT32 *)nh_result);
1147             len -= L1_KEY_LEN;
1148             msg += L1_KEY_LEN;
1149         } while (len >= L1_KEY_LEN);
1150         if (len) {
1151             nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1152             extra_zeroes_needed = nh_len - len;
1153             zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1154             nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1155             poly_hash(ahc,(UINT32 *)nh_result);
1156         }
1157 
1158         ip_long(ahc, res);
1159     }
1160 
1161     uhash_reset(ahc);
1162     return 1;
1163 }
1164 #endif
1165 
1166 /* ---------------------------------------------------------------------- */
1167 /* ---------------------------------------------------------------------- */
1168 /* ----- Begin UMAC Section --------------------------------------------- */
1169 /* ---------------------------------------------------------------------- */
1170 /* ---------------------------------------------------------------------- */
1171 
1172 /* The UMAC interface has two interfaces, an all-at-once interface where
1173  * the entire message to be authenticated is passed to UMAC in one buffer,
1174  * and a sequential interface where the message is presented a little at a
1175  * time. The all-at-once is more optimaized than the sequential version and
1176  * should be preferred when the sequential interface is not required.
1177  */
1178 struct umac_ctx {
1179     uhash_ctx hash;          /* Hash function for message compression    */
1180     pdf_ctx pdf;             /* PDF for hashed output                    */
1181     void *free_ptr;          /* Address to free this struct via          */
1182 } umac_ctx;
1183 
1184 /* ---------------------------------------------------------------------- */
1185 
1186 #if 0
1187 int umac_reset(struct umac_ctx *ctx)
1188 /* Reset the hash function to begin a new authentication.        */
1189 {
1190     uhash_reset(&ctx->hash);
1191     return (1);
1192 }
1193 #endif
1194 
1195 /* ---------------------------------------------------------------------- */
1196 
1197 int umac_delete(struct umac_ctx *ctx)
1198 /* Deallocate the ctx structure */
1199 {
1200     if (ctx) {
1201         if (ALLOC_BOUNDARY)
1202             ctx = (struct umac_ctx *)ctx->free_ptr;
1203         free(ctx);
1204     }
1205     return (1);
1206 }
1207 
1208 /* ---------------------------------------------------------------------- */
1209 
1210 struct umac_ctx *umac_new(const u_char key[])
1211 /* Dynamically allocate a umac_ctx struct, initialize variables,
1212  * generate subkeys from key. Align to 16-byte boundary.
1213  */
1214 {
1215     struct umac_ctx *ctx, *octx;
1216     size_t bytes_to_add;
1217     aes_int_key prf_key;
1218 
1219     octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1220     if (ctx) {
1221         if (ALLOC_BOUNDARY) {
1222             bytes_to_add = ALLOC_BOUNDARY -
1223                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1224             ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1225         }
1226         ctx->free_ptr = octx;
1227         aes_key_setup(key, prf_key);
1228         pdf_init(&ctx->pdf, prf_key);
1229         uhash_init(&ctx->hash, prf_key);
1230     }
1231 
1232     return (ctx);
1233 }
1234 
1235 /* ---------------------------------------------------------------------- */
1236 
1237 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1238 /* Incorporate any pending data, pad, and generate tag */
1239 {
1240     uhash_final(&ctx->hash, (u_char *)tag);
1241     pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1242 
1243     return (1);
1244 }
1245 
1246 /* ---------------------------------------------------------------------- */
1247 
1248 int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1249 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and   */
1250 /* hash each one, calling the PDF on the hashed output whenever the hash- */
1251 /* output buffer is full.                                                 */
1252 {
1253     uhash_update(&ctx->hash, input, len);
1254     return (1);
1255 }
1256 
1257 /* ---------------------------------------------------------------------- */
1258 
1259 #if 0
1260 int umac(struct umac_ctx *ctx, u_char *input,
1261          long len, u_char tag[],
1262          u_char nonce[8])
1263 /* All-in-one version simply calls umac_update() and umac_final().        */
1264 {
1265     uhash(&ctx->hash, input, len, (u_char *)tag);
1266     pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1267 
1268     return (1);
1269 }
1270 #endif
1271 
1272 /* ---------------------------------------------------------------------- */
1273 /* ---------------------------------------------------------------------- */
1274 /* ----- End UMAC Section ----------------------------------------------- */
1275 /* ---------------------------------------------------------------------- */
1276 /* ---------------------------------------------------------------------- */
1277