xref: /freebsd/crypto/openssh/umac.c (revision 40a8ac8f62b535d30349faf28cf47106b7041b83)
1 /* $OpenBSD: umac.c,v 1.8 2013/11/08 00:39:15 djm Exp $ */
2 /* -----------------------------------------------------------------------
3  *
4  * umac.c -- C Implementation UMAC Message Authentication
5  *
6  * Version 0.93b of rfc4418.txt -- 2006 July 18
7  *
8  * For a full description of UMAC message authentication see the UMAC
9  * world-wide-web page at http://www.cs.ucdavis.edu/~rogaway/umac
10  * Please report bugs and suggestions to the UMAC webpage.
11  *
12  * Copyright (c) 1999-2006 Ted Krovetz
13  *
14  * Permission to use, copy, modify, and distribute this software and
15  * its documentation for any purpose and with or without fee, is hereby
16  * granted provided that the above copyright notice appears in all copies
17  * and in supporting documentation, and that the name of the copyright
18  * holder not be used in advertising or publicity pertaining to
19  * distribution of the software without specific, written prior permission.
20  *
21  * Comments should be directed to Ted Krovetz (tdk@acm.org)
22  *
23  * ---------------------------------------------------------------------- */
24 
25  /* ////////////////////// IMPORTANT NOTES /////////////////////////////////
26   *
27   * 1) This version does not work properly on messages larger than 16MB
28   *
29   * 2) If you set the switch to use SSE2, then all data must be 16-byte
30   *    aligned
31   *
32   * 3) When calling the function umac(), it is assumed that msg is in
33   * a writable buffer of length divisible by 32 bytes. The message itself
34   * does not have to fill the entire buffer, but bytes beyond msg may be
35   * zeroed.
36   *
37   * 4) Three free AES implementations are supported by this implementation of
38   * UMAC. Paulo Barreto's version is in the public domain and can be found
39   * at http://www.esat.kuleuven.ac.be/~rijmen/rijndael/ (search for
40   * "Barreto"). The only two files needed are rijndael-alg-fst.c and
41   * rijndael-alg-fst.h. Brian Gladman's version is distributed with the GNU
42   * Public 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 
77 #include "xmalloc.h"
78 #include "umac.h"
79 #include <string.h>
80 #include <stdlib.h>
81 #include <stddef.h>
82 
83 /* ---------------------------------------------------------------------- */
84 /* --- Primitive Data Types ---                                           */
85 /* ---------------------------------------------------------------------- */
86 
87 /* The following assumptions may need change on your system */
88 typedef u_int8_t	UINT8;  /* 1 byte   */
89 typedef u_int16_t	UINT16; /* 2 byte   */
90 typedef u_int32_t	UINT32; /* 4 byte   */
91 typedef u_int64_t	UINT64; /* 8 bytes  */
92 typedef unsigned int	UWORD;  /* Register */
93 
94 /* ---------------------------------------------------------------------- */
95 /* --- Constants -------------------------------------------------------- */
96 /* ---------------------------------------------------------------------- */
97 
98 #define UMAC_KEY_LEN           16  /* UMAC takes 16 bytes of external key */
99 
100 /* Message "words" are read from memory in an endian-specific manner.     */
101 /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must    */
102 /* be set true if the host computer is little-endian.                     */
103 
104 #if BYTE_ORDER == LITTLE_ENDIAN
105 #define __LITTLE_ENDIAN__ 1
106 #else
107 #define __LITTLE_ENDIAN__ 0
108 #endif
109 
110 /* ---------------------------------------------------------------------- */
111 /* ---------------------------------------------------------------------- */
112 /* ----- Architecture Specific ------------------------------------------ */
113 /* ---------------------------------------------------------------------- */
114 /* ---------------------------------------------------------------------- */
115 
116 
117 /* ---------------------------------------------------------------------- */
118 /* ---------------------------------------------------------------------- */
119 /* ----- Primitive Routines --------------------------------------------- */
120 /* ---------------------------------------------------------------------- */
121 /* ---------------------------------------------------------------------- */
122 
123 
124 /* ---------------------------------------------------------------------- */
125 /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
126 /* ---------------------------------------------------------------------- */
127 
128 #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
129 
130 /* ---------------------------------------------------------------------- */
131 /* --- Endian Conversion --- Forcing assembly on some platforms           */
132 /* ---------------------------------------------------------------------- */
133 
134 #if HAVE_SWAP32
135 #define LOAD_UINT32_REVERSED(p)		(swap32(*(const UINT32 *)(p)))
136 #define STORE_UINT32_REVERSED(p,v) 	(*(UINT32 *)(p) = swap32(v))
137 #else /* HAVE_SWAP32 */
138 
139 static UINT32 LOAD_UINT32_REVERSED(const void *ptr)
140 {
141     UINT32 temp = *(const UINT32 *)ptr;
142     temp = (temp >> 24) | ((temp & 0x00FF0000) >> 8 )
143          | ((temp & 0x0000FF00) << 8 ) | (temp << 24);
144     return (UINT32)temp;
145 }
146 
147 # if (__LITTLE_ENDIAN__)
148 static void STORE_UINT32_REVERSED(void *ptr, UINT32 x)
149 {
150     UINT32 i = (UINT32)x;
151     *(UINT32 *)ptr = (i >> 24) | ((i & 0x00FF0000) >> 8 )
152                    | ((i & 0x0000FF00) << 8 ) | (i << 24);
153 }
154 # endif /* __LITTLE_ENDIAN */
155 #endif /* HAVE_SWAP32 */
156 
157 /* The following definitions use the above reversal-primitives to do the right
158  * thing on endian specific load and stores.
159  */
160 
161 #if (__LITTLE_ENDIAN__)
162 #define LOAD_UINT32_LITTLE(ptr)     (*(const UINT32 *)(ptr))
163 #define STORE_UINT32_BIG(ptr,x)     STORE_UINT32_REVERSED(ptr,x)
164 #else
165 #define LOAD_UINT32_LITTLE(ptr)     LOAD_UINT32_REVERSED(ptr)
166 #define STORE_UINT32_BIG(ptr,x)     (*(UINT32 *)(ptr) = (UINT32)(x))
167 #endif
168 
169 /* ---------------------------------------------------------------------- */
170 /* ---------------------------------------------------------------------- */
171 /* ----- Begin KDF & PDF Section ---------------------------------------- */
172 /* ---------------------------------------------------------------------- */
173 /* ---------------------------------------------------------------------- */
174 
175 /* UMAC uses AES with 16 byte block and key lengths */
176 #define AES_BLOCK_LEN  16
177 
178 /* OpenSSL's AES */
179 #include "openbsd-compat/openssl-compat.h"
180 #ifndef USE_BUILTIN_RIJNDAEL
181 # include <openssl/aes.h>
182 #endif
183 typedef AES_KEY aes_int_key[1];
184 #define aes_encryption(in,out,int_key)                  \
185   AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key)
186 #define aes_key_setup(key,int_key)                      \
187   AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key)
188 
189 /* The user-supplied UMAC key is stretched using AES in a counter
190  * mode to supply all random bits needed by UMAC. The kdf function takes
191  * an AES internal key representation 'key' and writes a stream of
192  * 'nbytes' bytes to the memory pointed at by 'bufp'. Each distinct
193  * 'ndx' causes a distinct byte stream.
194  */
195 static void kdf(void *bufp, aes_int_key key, UINT8 ndx, int nbytes)
196 {
197     UINT8 in_buf[AES_BLOCK_LEN] = {0};
198     UINT8 out_buf[AES_BLOCK_LEN];
199     UINT8 *dst_buf = (UINT8 *)bufp;
200     int i;
201 
202     /* Setup the initial value */
203     in_buf[AES_BLOCK_LEN-9] = ndx;
204     in_buf[AES_BLOCK_LEN-1] = i = 1;
205 
206     while (nbytes >= AES_BLOCK_LEN) {
207         aes_encryption(in_buf, out_buf, key);
208         memcpy(dst_buf,out_buf,AES_BLOCK_LEN);
209         in_buf[AES_BLOCK_LEN-1] = ++i;
210         nbytes -= AES_BLOCK_LEN;
211         dst_buf += AES_BLOCK_LEN;
212     }
213     if (nbytes) {
214         aes_encryption(in_buf, out_buf, key);
215         memcpy(dst_buf,out_buf,nbytes);
216     }
217 }
218 
219 /* The final UHASH result is XOR'd with the output of a pseudorandom
220  * function. Here, we use AES to generate random output and
221  * xor the appropriate bytes depending on the last bits of nonce.
222  * This scheme is optimized for sequential, increasing big-endian nonces.
223  */
224 
225 typedef struct {
226     UINT8 cache[AES_BLOCK_LEN];  /* Previous AES output is saved      */
227     UINT8 nonce[AES_BLOCK_LEN];  /* The AES input making above cache  */
228     aes_int_key prf_key;         /* Expanded AES key for PDF          */
229 } pdf_ctx;
230 
231 static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
232 {
233     UINT8 buf[UMAC_KEY_LEN];
234 
235     kdf(buf, prf_key, 0, UMAC_KEY_LEN);
236     aes_key_setup(buf, pc->prf_key);
237 
238     /* Initialize pdf and cache */
239     memset(pc->nonce, 0, sizeof(pc->nonce));
240     aes_encryption(pc->nonce, pc->cache, pc->prf_key);
241 }
242 
243 static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8])
244 {
245     /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
246      * of the AES output. If last time around we returned the ndx-1st
247      * element, then we may have the result in the cache already.
248      */
249 
250 #if (UMAC_OUTPUT_LEN == 4)
251 #define LOW_BIT_MASK 3
252 #elif (UMAC_OUTPUT_LEN == 8)
253 #define LOW_BIT_MASK 1
254 #elif (UMAC_OUTPUT_LEN > 8)
255 #define LOW_BIT_MASK 0
256 #endif
257     union {
258         UINT8 tmp_nonce_lo[4];
259         UINT32 align;
260     } t;
261 #if LOW_BIT_MASK != 0
262     int ndx = nonce[7] & LOW_BIT_MASK;
263 #endif
264     *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
265     t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
266 
267     if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
268          (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
269     {
270         ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
271         ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
272         aes_encryption(pc->nonce, pc->cache, pc->prf_key);
273     }
274 
275 #if (UMAC_OUTPUT_LEN == 4)
276     *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
277 #elif (UMAC_OUTPUT_LEN == 8)
278     *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
279 #elif (UMAC_OUTPUT_LEN == 12)
280     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
281     ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
282 #elif (UMAC_OUTPUT_LEN == 16)
283     ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
284     ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
285 #endif
286 }
287 
288 /* ---------------------------------------------------------------------- */
289 /* ---------------------------------------------------------------------- */
290 /* ----- Begin NH Hash Section ------------------------------------------ */
291 /* ---------------------------------------------------------------------- */
292 /* ---------------------------------------------------------------------- */
293 
294 /* The NH-based hash functions used in UMAC are described in the UMAC paper
295  * and specification, both of which can be found at the UMAC website.
296  * The interface to this implementation has two
297  * versions, one expects the entire message being hashed to be passed
298  * in a single buffer and returns the hash result immediately. The second
299  * allows the message to be passed in a sequence of buffers. In the
300  * muliple-buffer interface, the client calls the routine nh_update() as
301  * many times as necessary. When there is no more data to be fed to the
302  * hash, the client calls nh_final() which calculates the hash output.
303  * Before beginning another hash calculation the nh_reset() routine
304  * must be called. The single-buffer routine, nh(), is equivalent to
305  * the sequence of calls nh_update() and nh_final(); however it is
306  * optimized and should be prefered whenever the multiple-buffer interface
307  * is not necessary. When using either interface, it is the client's
308  * responsability to pass no more than L1_KEY_LEN bytes per hash result.
309  *
310  * The routine nh_init() initializes the nh_ctx data structure and
311  * must be called once, before any other PDF routine.
312  */
313 
314  /* The "nh_aux" routines do the actual NH hashing work. They
315   * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
316   * produce output for all STREAMS NH iterations in one call,
317   * allowing the parallel implementation of the streams.
318   */
319 
320 #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied  */
321 #define L1_KEY_LEN         1024     /* Internal key bytes                 */
322 #define L1_KEY_SHIFT         16     /* Toeplitz key shift between streams */
323 #define L1_PAD_BOUNDARY      32     /* pad message to boundary multiple   */
324 #define ALLOC_BOUNDARY       16     /* Keep buffers aligned to this       */
325 #define HASH_BUF_BYTES       64     /* nh_aux_hb buffer multiple          */
326 
327 typedef struct {
328     UINT8  nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
329     UINT8  data   [HASH_BUF_BYTES];    /* Incoming data buffer           */
330     int next_data_empty;    /* Bookeeping variable for data buffer.       */
331     int bytes_hashed;        /* Bytes (out of L1_KEY_LEN) incorperated.   */
332     UINT64 state[STREAMS];               /* on-line state     */
333 } nh_ctx;
334 
335 
336 #if (UMAC_OUTPUT_LEN == 4)
337 
338 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
339 /* NH hashing primitive. Previous (partial) hash result is loaded and
340 * then stored via hp pointer. The length of the data pointed at by "dp",
341 * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32).  Key
342 * is expected to be endian compensated in memory at key setup.
343 */
344 {
345     UINT64 h;
346     UWORD c = dlen / 32;
347     UINT32 *k = (UINT32 *)kp;
348     const UINT32 *d = (const UINT32 *)dp;
349     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
350     UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
351 
352     h = *((UINT64 *)hp);
353     do {
354         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
355         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
356         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
357         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
358         k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
359         k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
360         h += MUL64((k0 + d0), (k4 + d4));
361         h += MUL64((k1 + d1), (k5 + d5));
362         h += MUL64((k2 + d2), (k6 + d6));
363         h += MUL64((k3 + d3), (k7 + d7));
364 
365         d += 8;
366         k += 8;
367     } while (--c);
368   *((UINT64 *)hp) = h;
369 }
370 
371 #elif (UMAC_OUTPUT_LEN == 8)
372 
373 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
374 /* Same as previous nh_aux, but two streams are handled in one pass,
375  * reading and writing 16 bytes of hash-state per call.
376  */
377 {
378   UINT64 h1,h2;
379   UWORD c = dlen / 32;
380   UINT32 *k = (UINT32 *)kp;
381   const UINT32 *d = (const UINT32 *)dp;
382   UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
383   UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
384         k8,k9,k10,k11;
385 
386   h1 = *((UINT64 *)hp);
387   h2 = *((UINT64 *)hp + 1);
388   k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
389   do {
390     d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
391     d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
392     d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
393     d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
394     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
395     k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
396 
397     h1 += MUL64((k0 + d0), (k4 + d4));
398     h2 += MUL64((k4 + d0), (k8 + d4));
399 
400     h1 += MUL64((k1 + d1), (k5 + d5));
401     h2 += MUL64((k5 + d1), (k9 + d5));
402 
403     h1 += MUL64((k2 + d2), (k6 + d6));
404     h2 += MUL64((k6 + d2), (k10 + d6));
405 
406     h1 += MUL64((k3 + d3), (k7 + d7));
407     h2 += MUL64((k7 + d3), (k11 + d7));
408 
409     k0 = k8; k1 = k9; k2 = k10; k3 = k11;
410 
411     d += 8;
412     k += 8;
413   } while (--c);
414   ((UINT64 *)hp)[0] = h1;
415   ((UINT64 *)hp)[1] = h2;
416 }
417 
418 #elif (UMAC_OUTPUT_LEN == 12)
419 
420 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
421 /* Same as previous nh_aux, but two streams are handled in one pass,
422  * reading and writing 24 bytes of hash-state per call.
423 */
424 {
425     UINT64 h1,h2,h3;
426     UWORD c = dlen / 32;
427     UINT32 *k = (UINT32 *)kp;
428     const UINT32 *d = (const UINT32 *)dp;
429     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
430     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
431         k8,k9,k10,k11,k12,k13,k14,k15;
432 
433     h1 = *((UINT64 *)hp);
434     h2 = *((UINT64 *)hp + 1);
435     h3 = *((UINT64 *)hp + 2);
436     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
437     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
438     do {
439         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
440         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
441         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
442         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
443         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
444         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
445 
446         h1 += MUL64((k0 + d0), (k4 + d4));
447         h2 += MUL64((k4 + d0), (k8 + d4));
448         h3 += MUL64((k8 + d0), (k12 + d4));
449 
450         h1 += MUL64((k1 + d1), (k5 + d5));
451         h2 += MUL64((k5 + d1), (k9 + d5));
452         h3 += MUL64((k9 + d1), (k13 + d5));
453 
454         h1 += MUL64((k2 + d2), (k6 + d6));
455         h2 += MUL64((k6 + d2), (k10 + d6));
456         h3 += MUL64((k10 + d2), (k14 + d6));
457 
458         h1 += MUL64((k3 + d3), (k7 + d7));
459         h2 += MUL64((k7 + d3), (k11 + d7));
460         h3 += MUL64((k11 + d3), (k15 + d7));
461 
462         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
463         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
464 
465         d += 8;
466         k += 8;
467     } while (--c);
468     ((UINT64 *)hp)[0] = h1;
469     ((UINT64 *)hp)[1] = h2;
470     ((UINT64 *)hp)[2] = h3;
471 }
472 
473 #elif (UMAC_OUTPUT_LEN == 16)
474 
475 static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
476 /* Same as previous nh_aux, but two streams are handled in one pass,
477  * reading and writing 24 bytes of hash-state per call.
478 */
479 {
480     UINT64 h1,h2,h3,h4;
481     UWORD c = dlen / 32;
482     UINT32 *k = (UINT32 *)kp;
483     const UINT32 *d = (const UINT32 *)dp;
484     UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
485     UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
486         k8,k9,k10,k11,k12,k13,k14,k15,
487         k16,k17,k18,k19;
488 
489     h1 = *((UINT64 *)hp);
490     h2 = *((UINT64 *)hp + 1);
491     h3 = *((UINT64 *)hp + 2);
492     h4 = *((UINT64 *)hp + 3);
493     k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
494     k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
495     do {
496         d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
497         d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
498         d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
499         d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
500         k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
501         k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
502         k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
503 
504         h1 += MUL64((k0 + d0), (k4 + d4));
505         h2 += MUL64((k4 + d0), (k8 + d4));
506         h3 += MUL64((k8 + d0), (k12 + d4));
507         h4 += MUL64((k12 + d0), (k16 + d4));
508 
509         h1 += MUL64((k1 + d1), (k5 + d5));
510         h2 += MUL64((k5 + d1), (k9 + d5));
511         h3 += MUL64((k9 + d1), (k13 + d5));
512         h4 += MUL64((k13 + d1), (k17 + d5));
513 
514         h1 += MUL64((k2 + d2), (k6 + d6));
515         h2 += MUL64((k6 + d2), (k10 + d6));
516         h3 += MUL64((k10 + d2), (k14 + d6));
517         h4 += MUL64((k14 + d2), (k18 + d6));
518 
519         h1 += MUL64((k3 + d3), (k7 + d7));
520         h2 += MUL64((k7 + d3), (k11 + d7));
521         h3 += MUL64((k11 + d3), (k15 + d7));
522         h4 += MUL64((k15 + d3), (k19 + d7));
523 
524         k0 = k8; k1 = k9; k2 = k10; k3 = k11;
525         k4 = k12; k5 = k13; k6 = k14; k7 = k15;
526         k8 = k16; k9 = k17; k10 = k18; k11 = k19;
527 
528         d += 8;
529         k += 8;
530     } while (--c);
531     ((UINT64 *)hp)[0] = h1;
532     ((UINT64 *)hp)[1] = h2;
533     ((UINT64 *)hp)[2] = h3;
534     ((UINT64 *)hp)[3] = h4;
535 }
536 
537 /* ---------------------------------------------------------------------- */
538 #endif  /* UMAC_OUTPUT_LENGTH */
539 /* ---------------------------------------------------------------------- */
540 
541 
542 /* ---------------------------------------------------------------------- */
543 
544 static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
545 /* This function is a wrapper for the primitive NH hash functions. It takes
546  * as argument "hc" the current hash context and a buffer which must be a
547  * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
548  * appropriately according to how much message has been hashed already.
549  */
550 {
551     UINT8 *key;
552 
553     key = hc->nh_key + hc->bytes_hashed;
554     nh_aux(key, buf, hc->state, nbytes);
555 }
556 
557 /* ---------------------------------------------------------------------- */
558 
559 #if (__LITTLE_ENDIAN__)
560 static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
561 /* We endian convert the keys on little-endian computers to               */
562 /* compensate for the lack of big-endian memory reads during hashing.     */
563 {
564     UWORD iters = num_bytes / bpw;
565     if (bpw == 4) {
566         UINT32 *p = (UINT32 *)buf;
567         do {
568             *p = LOAD_UINT32_REVERSED(p);
569             p++;
570         } while (--iters);
571     } else if (bpw == 8) {
572         UINT32 *p = (UINT32 *)buf;
573         UINT32 t;
574         do {
575             t = LOAD_UINT32_REVERSED(p+1);
576             p[1] = LOAD_UINT32_REVERSED(p);
577             p[0] = t;
578             p += 2;
579         } while (--iters);
580     }
581 }
582 #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
583 #else
584 #define endian_convert_if_le(x,y,z) do{}while(0)  /* Do nothing */
585 #endif
586 
587 /* ---------------------------------------------------------------------- */
588 
589 static void nh_reset(nh_ctx *hc)
590 /* Reset nh_ctx to ready for hashing of new data */
591 {
592     hc->bytes_hashed = 0;
593     hc->next_data_empty = 0;
594     hc->state[0] = 0;
595 #if (UMAC_OUTPUT_LEN >= 8)
596     hc->state[1] = 0;
597 #endif
598 #if (UMAC_OUTPUT_LEN >= 12)
599     hc->state[2] = 0;
600 #endif
601 #if (UMAC_OUTPUT_LEN == 16)
602     hc->state[3] = 0;
603 #endif
604 
605 }
606 
607 /* ---------------------------------------------------------------------- */
608 
609 static void nh_init(nh_ctx *hc, aes_int_key prf_key)
610 /* Generate nh_key, endian convert and reset to be ready for hashing.   */
611 {
612     kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
613     endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
614     nh_reset(hc);
615 }
616 
617 /* ---------------------------------------------------------------------- */
618 
619 static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
620 /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an    */
621 /* even multiple of HASH_BUF_BYTES.                                       */
622 {
623     UINT32 i,j;
624 
625     j = hc->next_data_empty;
626     if ((j + nbytes) >= HASH_BUF_BYTES) {
627         if (j) {
628             i = HASH_BUF_BYTES - j;
629             memcpy(hc->data+j, buf, i);
630             nh_transform(hc,hc->data,HASH_BUF_BYTES);
631             nbytes -= i;
632             buf += i;
633             hc->bytes_hashed += HASH_BUF_BYTES;
634         }
635         if (nbytes >= HASH_BUF_BYTES) {
636             i = nbytes & ~(HASH_BUF_BYTES - 1);
637             nh_transform(hc, buf, i);
638             nbytes -= i;
639             buf += i;
640             hc->bytes_hashed += i;
641         }
642         j = 0;
643     }
644     memcpy(hc->data + j, buf, nbytes);
645     hc->next_data_empty = j + nbytes;
646 }
647 
648 /* ---------------------------------------------------------------------- */
649 
650 static void zero_pad(UINT8 *p, int nbytes)
651 {
652 /* Write "nbytes" of zeroes, beginning at "p" */
653     if (nbytes >= (int)sizeof(UWORD)) {
654         while ((ptrdiff_t)p % sizeof(UWORD)) {
655             *p = 0;
656             nbytes--;
657             p++;
658         }
659         while (nbytes >= (int)sizeof(UWORD)) {
660             *(UWORD *)p = 0;
661             nbytes -= sizeof(UWORD);
662             p += sizeof(UWORD);
663         }
664     }
665     while (nbytes) {
666         *p = 0;
667         nbytes--;
668         p++;
669     }
670 }
671 
672 /* ---------------------------------------------------------------------- */
673 
674 static void nh_final(nh_ctx *hc, UINT8 *result)
675 /* After passing some number of data buffers to nh_update() for integration
676  * into an NH context, nh_final is called to produce a hash result. If any
677  * bytes are in the buffer hc->data, incorporate them into the
678  * NH context. Finally, add into the NH accumulation "state" the total number
679  * of bits hashed. The resulting numbers are written to the buffer "result".
680  * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
681  */
682 {
683     int nh_len, nbits;
684 
685     if (hc->next_data_empty != 0) {
686         nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
687                                                 ~(L1_PAD_BOUNDARY - 1));
688         zero_pad(hc->data + hc->next_data_empty,
689                                           nh_len - hc->next_data_empty);
690         nh_transform(hc, hc->data, nh_len);
691         hc->bytes_hashed += hc->next_data_empty;
692     } else if (hc->bytes_hashed == 0) {
693     	nh_len = L1_PAD_BOUNDARY;
694         zero_pad(hc->data, L1_PAD_BOUNDARY);
695         nh_transform(hc, hc->data, nh_len);
696     }
697 
698     nbits = (hc->bytes_hashed << 3);
699     ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
700 #if (UMAC_OUTPUT_LEN >= 8)
701     ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
702 #endif
703 #if (UMAC_OUTPUT_LEN >= 12)
704     ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
705 #endif
706 #if (UMAC_OUTPUT_LEN == 16)
707     ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
708 #endif
709     nh_reset(hc);
710 }
711 
712 /* ---------------------------------------------------------------------- */
713 
714 static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
715                UINT32 unpadded_len, UINT8 *result)
716 /* All-in-one nh_update() and nh_final() equivalent.
717  * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
718  * well aligned
719  */
720 {
721     UINT32 nbits;
722 
723     /* Initialize the hash state */
724     nbits = (unpadded_len << 3);
725 
726     ((UINT64 *)result)[0] = nbits;
727 #if (UMAC_OUTPUT_LEN >= 8)
728     ((UINT64 *)result)[1] = nbits;
729 #endif
730 #if (UMAC_OUTPUT_LEN >= 12)
731     ((UINT64 *)result)[2] = nbits;
732 #endif
733 #if (UMAC_OUTPUT_LEN == 16)
734     ((UINT64 *)result)[3] = nbits;
735 #endif
736 
737     nh_aux(hc->nh_key, buf, result, padded_len);
738 }
739 
740 /* ---------------------------------------------------------------------- */
741 /* ---------------------------------------------------------------------- */
742 /* ----- Begin UHASH Section -------------------------------------------- */
743 /* ---------------------------------------------------------------------- */
744 /* ---------------------------------------------------------------------- */
745 
746 /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
747  * hashed by NH. The NH output is then hashed by a polynomial-hash layer
748  * unless the initial data to be hashed is short. After the polynomial-
749  * layer, an inner-product hash is used to produce the final UHASH output.
750  *
751  * UHASH provides two interfaces, one all-at-once and another where data
752  * buffers are presented sequentially. In the sequential interface, the
753  * UHASH client calls the routine uhash_update() as many times as necessary.
754  * When there is no more data to be fed to UHASH, the client calls
755  * uhash_final() which
756  * calculates the UHASH output. Before beginning another UHASH calculation
757  * the uhash_reset() routine must be called. The all-at-once UHASH routine,
758  * uhash(), is equivalent to the sequence of calls uhash_update() and
759  * uhash_final(); however it is optimized and should be
760  * used whenever the sequential interface is not necessary.
761  *
762  * The routine uhash_init() initializes the uhash_ctx data structure and
763  * must be called once, before any other UHASH routine.
764  */
765 
766 /* ---------------------------------------------------------------------- */
767 /* ----- Constants and uhash_ctx ---------------------------------------- */
768 /* ---------------------------------------------------------------------- */
769 
770 /* ---------------------------------------------------------------------- */
771 /* ----- Poly hash and Inner-Product hash Constants --------------------- */
772 /* ---------------------------------------------------------------------- */
773 
774 /* Primes and masks */
775 #define p36    ((UINT64)0x0000000FFFFFFFFBull)              /* 2^36 -  5 */
776 #define p64    ((UINT64)0xFFFFFFFFFFFFFFC5ull)              /* 2^64 - 59 */
777 #define m36    ((UINT64)0x0000000FFFFFFFFFull)  /* The low 36 of 64 bits */
778 
779 
780 /* ---------------------------------------------------------------------- */
781 
782 typedef struct uhash_ctx {
783     nh_ctx hash;                          /* Hash context for L1 NH hash  */
784     UINT64 poly_key_8[STREAMS];           /* p64 poly keys                */
785     UINT64 poly_accum[STREAMS];           /* poly hash result             */
786     UINT64 ip_keys[STREAMS*4];            /* Inner-product keys           */
787     UINT32 ip_trans[STREAMS];             /* Inner-product translation    */
788     UINT32 msg_len;                       /* Total length of data passed  */
789                                           /* to uhash */
790 } uhash_ctx;
791 typedef struct uhash_ctx *uhash_ctx_t;
792 
793 /* ---------------------------------------------------------------------- */
794 
795 
796 /* The polynomial hashes use Horner's rule to evaluate a polynomial one
797  * word at a time. As described in the specification, poly32 and poly64
798  * require keys from special domains. The following implementations exploit
799  * the special domains to avoid overflow. The results are not guaranteed to
800  * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
801  * patches any errant values.
802  */
803 
804 static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
805 {
806     UINT32 key_hi = (UINT32)(key >> 32),
807            key_lo = (UINT32)key,
808            cur_hi = (UINT32)(cur >> 32),
809            cur_lo = (UINT32)cur,
810            x_lo,
811            x_hi;
812     UINT64 X,T,res;
813 
814     X =  MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
815     x_lo = (UINT32)X;
816     x_hi = (UINT32)(X >> 32);
817 
818     res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
819 
820     T = ((UINT64)x_lo << 32);
821     res += T;
822     if (res < T)
823         res += 59;
824 
825     res += data;
826     if (res < data)
827         res += 59;
828 
829     return res;
830 }
831 
832 
833 /* Although UMAC is specified to use a ramped polynomial hash scheme, this
834  * implementation does not handle all ramp levels. Because we don't handle
835  * the ramp up to p128 modulus in this implementation, we are limited to
836  * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
837  * bytes input to UMAC per tag, ie. 16MB).
838  */
839 static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
840 {
841     int i;
842     UINT64 *data=(UINT64*)data_in;
843 
844     for (i = 0; i < STREAMS; i++) {
845         if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
846             hc->poly_accum[i] = poly64(hc->poly_accum[i],
847                                        hc->poly_key_8[i], p64 - 1);
848             hc->poly_accum[i] = poly64(hc->poly_accum[i],
849                                        hc->poly_key_8[i], (data[i] - 59));
850         } else {
851             hc->poly_accum[i] = poly64(hc->poly_accum[i],
852                                        hc->poly_key_8[i], data[i]);
853         }
854     }
855 }
856 
857 
858 /* ---------------------------------------------------------------------- */
859 
860 
861 /* The final step in UHASH is an inner-product hash. The poly hash
862  * produces a result not neccesarily WORD_LEN bytes long. The inner-
863  * product hash breaks the polyhash output into 16-bit chunks and
864  * multiplies each with a 36 bit key.
865  */
866 
867 static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
868 {
869     t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
870     t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
871     t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
872     t = t + ipkp[3] * (UINT64)(UINT16)(data);
873 
874     return t;
875 }
876 
877 static UINT32 ip_reduce_p36(UINT64 t)
878 {
879 /* Divisionless modular reduction */
880     UINT64 ret;
881 
882     ret = (t & m36) + 5 * (t >> 36);
883     if (ret >= p36)
884         ret -= p36;
885 
886     /* return least significant 32 bits */
887     return (UINT32)(ret);
888 }
889 
890 
891 /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
892  * the polyhash stage is skipped and ip_short is applied directly to the
893  * NH output.
894  */
895 static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
896 {
897     UINT64 t;
898     UINT64 *nhp = (UINT64 *)nh_res;
899 
900     t  = ip_aux(0,ahc->ip_keys, nhp[0]);
901     STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
902 #if (UMAC_OUTPUT_LEN >= 8)
903     t  = ip_aux(0,ahc->ip_keys+4, nhp[1]);
904     STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
905 #endif
906 #if (UMAC_OUTPUT_LEN >= 12)
907     t  = ip_aux(0,ahc->ip_keys+8, nhp[2]);
908     STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
909 #endif
910 #if (UMAC_OUTPUT_LEN == 16)
911     t  = ip_aux(0,ahc->ip_keys+12, nhp[3]);
912     STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
913 #endif
914 }
915 
916 /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
917  * the polyhash stage is not skipped and ip_long is applied to the
918  * polyhash output.
919  */
920 static void ip_long(uhash_ctx_t ahc, u_char *res)
921 {
922     int i;
923     UINT64 t;
924 
925     for (i = 0; i < STREAMS; i++) {
926         /* fix polyhash output not in Z_p64 */
927         if (ahc->poly_accum[i] >= p64)
928             ahc->poly_accum[i] -= p64;
929         t  = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
930         STORE_UINT32_BIG((UINT32 *)res+i,
931                          ip_reduce_p36(t) ^ ahc->ip_trans[i]);
932     }
933 }
934 
935 
936 /* ---------------------------------------------------------------------- */
937 
938 /* ---------------------------------------------------------------------- */
939 
940 /* Reset uhash context for next hash session */
941 static int uhash_reset(uhash_ctx_t pc)
942 {
943     nh_reset(&pc->hash);
944     pc->msg_len = 0;
945     pc->poly_accum[0] = 1;
946 #if (UMAC_OUTPUT_LEN >= 8)
947     pc->poly_accum[1] = 1;
948 #endif
949 #if (UMAC_OUTPUT_LEN >= 12)
950     pc->poly_accum[2] = 1;
951 #endif
952 #if (UMAC_OUTPUT_LEN == 16)
953     pc->poly_accum[3] = 1;
954 #endif
955     return 1;
956 }
957 
958 /* ---------------------------------------------------------------------- */
959 
960 /* Given a pointer to the internal key needed by kdf() and a uhash context,
961  * initialize the NH context and generate keys needed for poly and inner-
962  * product hashing. All keys are endian adjusted in memory so that native
963  * loads cause correct keys to be in registers during calculation.
964  */
965 static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
966 {
967     int i;
968     UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
969 
970     /* Zero the entire uhash context */
971     memset(ahc, 0, sizeof(uhash_ctx));
972 
973     /* Initialize the L1 hash */
974     nh_init(&ahc->hash, prf_key);
975 
976     /* Setup L2 hash variables */
977     kdf(buf, prf_key, 2, sizeof(buf));    /* Fill buffer with index 1 key */
978     for (i = 0; i < STREAMS; i++) {
979         /* Fill keys from the buffer, skipping bytes in the buffer not
980          * used by this implementation. Endian reverse the keys if on a
981          * little-endian computer.
982          */
983         memcpy(ahc->poly_key_8+i, buf+24*i, 8);
984         endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
985         /* Mask the 64-bit keys to their special domain */
986         ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
987         ahc->poly_accum[i] = 1;  /* Our polyhash prepends a non-zero word */
988     }
989 
990     /* Setup L3-1 hash variables */
991     kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
992     for (i = 0; i < STREAMS; i++)
993           memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
994                                                  4*sizeof(UINT64));
995     endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
996                                                   sizeof(ahc->ip_keys));
997     for (i = 0; i < STREAMS*4; i++)
998         ahc->ip_keys[i] %= p36;  /* Bring into Z_p36 */
999 
1000     /* Setup L3-2 hash variables    */
1001     /* Fill buffer with index 4 key */
1002     kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
1003     endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
1004                          STREAMS * sizeof(UINT32));
1005 }
1006 
1007 /* ---------------------------------------------------------------------- */
1008 
1009 #if 0
1010 static uhash_ctx_t uhash_alloc(u_char key[])
1011 {
1012 /* Allocate memory and force to a 16-byte boundary. */
1013     uhash_ctx_t ctx;
1014     u_char bytes_to_add;
1015     aes_int_key prf_key;
1016 
1017     ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1018     if (ctx) {
1019         if (ALLOC_BOUNDARY) {
1020             bytes_to_add = ALLOC_BOUNDARY -
1021                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1022             ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1023             *((u_char *)ctx - 1) = bytes_to_add;
1024         }
1025         aes_key_setup(key,prf_key);
1026         uhash_init(ctx, prf_key);
1027     }
1028     return (ctx);
1029 }
1030 #endif
1031 
1032 /* ---------------------------------------------------------------------- */
1033 
1034 #if 0
1035 static int uhash_free(uhash_ctx_t ctx)
1036 {
1037 /* Free memory allocated by uhash_alloc */
1038     u_char bytes_to_sub;
1039 
1040     if (ctx) {
1041         if (ALLOC_BOUNDARY) {
1042             bytes_to_sub = *((u_char *)ctx - 1);
1043             ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1044         }
1045         free(ctx);
1046     }
1047     return (1);
1048 }
1049 #endif
1050 /* ---------------------------------------------------------------------- */
1051 
1052 static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1053 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1054  * hash each one with NH, calling the polyhash on each NH output.
1055  */
1056 {
1057     UWORD bytes_hashed, bytes_remaining;
1058     UINT64 result_buf[STREAMS];
1059     UINT8 *nh_result = (UINT8 *)&result_buf;
1060 
1061     if (ctx->msg_len + len <= L1_KEY_LEN) {
1062         nh_update(&ctx->hash, (const UINT8 *)input, len);
1063         ctx->msg_len += len;
1064     } else {
1065 
1066          bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1067          if (ctx->msg_len == L1_KEY_LEN)
1068              bytes_hashed = L1_KEY_LEN;
1069 
1070          if (bytes_hashed + len >= L1_KEY_LEN) {
1071 
1072              /* If some bytes have been passed to the hash function      */
1073              /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1074              /* bytes to complete the current nh_block.                  */
1075              if (bytes_hashed) {
1076                  bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1077                  nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1078                  nh_final(&ctx->hash, nh_result);
1079                  ctx->msg_len += bytes_remaining;
1080                  poly_hash(ctx,(UINT32 *)nh_result);
1081                  len -= bytes_remaining;
1082                  input += bytes_remaining;
1083              }
1084 
1085              /* Hash directly from input stream if enough bytes */
1086              while (len >= L1_KEY_LEN) {
1087                  nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1088                                    L1_KEY_LEN, nh_result);
1089                  ctx->msg_len += L1_KEY_LEN;
1090                  len -= L1_KEY_LEN;
1091                  input += L1_KEY_LEN;
1092                  poly_hash(ctx,(UINT32 *)nh_result);
1093              }
1094          }
1095 
1096          /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1097          if (len) {
1098              nh_update(&ctx->hash, (const UINT8 *)input, len);
1099              ctx->msg_len += len;
1100          }
1101      }
1102 
1103     return (1);
1104 }
1105 
1106 /* ---------------------------------------------------------------------- */
1107 
1108 static int uhash_final(uhash_ctx_t ctx, u_char *res)
1109 /* Incorporate any pending data, pad, and generate tag */
1110 {
1111     UINT64 result_buf[STREAMS];
1112     UINT8 *nh_result = (UINT8 *)&result_buf;
1113 
1114     if (ctx->msg_len > L1_KEY_LEN) {
1115         if (ctx->msg_len % L1_KEY_LEN) {
1116             nh_final(&ctx->hash, nh_result);
1117             poly_hash(ctx,(UINT32 *)nh_result);
1118         }
1119         ip_long(ctx, res);
1120     } else {
1121         nh_final(&ctx->hash, nh_result);
1122         ip_short(ctx,nh_result, res);
1123     }
1124     uhash_reset(ctx);
1125     return (1);
1126 }
1127 
1128 /* ---------------------------------------------------------------------- */
1129 
1130 #if 0
1131 static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1132 /* assumes that msg is in a writable buffer of length divisible by */
1133 /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed.           */
1134 {
1135     UINT8 nh_result[STREAMS*sizeof(UINT64)];
1136     UINT32 nh_len;
1137     int extra_zeroes_needed;
1138 
1139     /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1140      * the polyhash.
1141      */
1142     if (len <= L1_KEY_LEN) {
1143     	if (len == 0)                  /* If zero length messages will not */
1144     		nh_len = L1_PAD_BOUNDARY;  /* be seen, comment out this case   */
1145     	else
1146         	nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1147         extra_zeroes_needed = nh_len - len;
1148         zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1149         nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1150         ip_short(ahc,nh_result, res);
1151     } else {
1152         /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1153          * output to poly_hash().
1154          */
1155         do {
1156             nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1157             poly_hash(ahc,(UINT32 *)nh_result);
1158             len -= L1_KEY_LEN;
1159             msg += L1_KEY_LEN;
1160         } while (len >= L1_KEY_LEN);
1161         if (len) {
1162             nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1163             extra_zeroes_needed = nh_len - len;
1164             zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1165             nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1166             poly_hash(ahc,(UINT32 *)nh_result);
1167         }
1168 
1169         ip_long(ahc, res);
1170     }
1171 
1172     uhash_reset(ahc);
1173     return 1;
1174 }
1175 #endif
1176 
1177 /* ---------------------------------------------------------------------- */
1178 /* ---------------------------------------------------------------------- */
1179 /* ----- Begin UMAC Section --------------------------------------------- */
1180 /* ---------------------------------------------------------------------- */
1181 /* ---------------------------------------------------------------------- */
1182 
1183 /* The UMAC interface has two interfaces, an all-at-once interface where
1184  * the entire message to be authenticated is passed to UMAC in one buffer,
1185  * and a sequential interface where the message is presented a little at a
1186  * time. The all-at-once is more optimaized than the sequential version and
1187  * should be preferred when the sequential interface is not required.
1188  */
1189 struct umac_ctx {
1190     uhash_ctx hash;          /* Hash function for message compression    */
1191     pdf_ctx pdf;             /* PDF for hashed output                    */
1192     void *free_ptr;          /* Address to free this struct via          */
1193 } umac_ctx;
1194 
1195 /* ---------------------------------------------------------------------- */
1196 
1197 #if 0
1198 int umac_reset(struct umac_ctx *ctx)
1199 /* Reset the hash function to begin a new authentication.        */
1200 {
1201     uhash_reset(&ctx->hash);
1202     return (1);
1203 }
1204 #endif
1205 
1206 /* ---------------------------------------------------------------------- */
1207 
1208 int umac_delete(struct umac_ctx *ctx)
1209 /* Deallocate the ctx structure */
1210 {
1211     if (ctx) {
1212         if (ALLOC_BOUNDARY)
1213             ctx = (struct umac_ctx *)ctx->free_ptr;
1214         free(ctx);
1215     }
1216     return (1);
1217 }
1218 
1219 /* ---------------------------------------------------------------------- */
1220 
1221 struct umac_ctx *umac_new(const u_char key[])
1222 /* Dynamically allocate a umac_ctx struct, initialize variables,
1223  * generate subkeys from key. Align to 16-byte boundary.
1224  */
1225 {
1226     struct umac_ctx *ctx, *octx;
1227     size_t bytes_to_add;
1228     aes_int_key prf_key;
1229 
1230     octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1231     if (ctx) {
1232         if (ALLOC_BOUNDARY) {
1233             bytes_to_add = ALLOC_BOUNDARY -
1234                               ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1235             ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1236         }
1237         ctx->free_ptr = octx;
1238         aes_key_setup(key, prf_key);
1239         pdf_init(&ctx->pdf, prf_key);
1240         uhash_init(&ctx->hash, prf_key);
1241     }
1242 
1243     return (ctx);
1244 }
1245 
1246 /* ---------------------------------------------------------------------- */
1247 
1248 int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1249 /* Incorporate any pending data, pad, and generate tag */
1250 {
1251     uhash_final(&ctx->hash, (u_char *)tag);
1252     pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1253 
1254     return (1);
1255 }
1256 
1257 /* ---------------------------------------------------------------------- */
1258 
1259 int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1260 /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and   */
1261 /* hash each one, calling the PDF on the hashed output whenever the hash- */
1262 /* output buffer is full.                                                 */
1263 {
1264     uhash_update(&ctx->hash, input, len);
1265     return (1);
1266 }
1267 
1268 /* ---------------------------------------------------------------------- */
1269 
1270 #if 0
1271 int umac(struct umac_ctx *ctx, u_char *input,
1272          long len, u_char tag[],
1273          u_char nonce[8])
1274 /* All-in-one version simply calls umac_update() and umac_final().        */
1275 {
1276     uhash(&ctx->hash, input, len, (u_char *)tag);
1277     pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1278 
1279     return (1);
1280 }
1281 #endif
1282 
1283 /* ---------------------------------------------------------------------- */
1284 /* ---------------------------------------------------------------------- */
1285 /* ----- End UMAC Section ----------------------------------------------- */
1286 /* ---------------------------------------------------------------------- */
1287 /* ---------------------------------------------------------------------- */
1288