Annotation of src/usr.bin/ssh/umac.c, Revision 1.11
1.11 ! guenther 1: /* $OpenBSD: umac.c,v 1.10 2014/04/30 19:07:48 naddy Exp $ */
1.1 pvalchev 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: #define UMAC_OUTPUT_LEN 8 /* Alowable: 4, 8, 12, 16 */
56: /* #define FORCE_C_ONLY 1 ANSI C and 64-bit integers req'd */
57: /* #define AES_IMPLEMENTAION 1 1 = OpenSSL, 2 = Barreto, 3 = Gladman */
58: /* #define SSE2 0 Is SSE2 is available? */
59: /* #define RUN_TESTS 0 Run basic correctness/speed tests */
60: /* #define UMAC_AE_SUPPORT 0 Enable auhthenticated encrytion */
61:
62: /* ---------------------------------------------------------------------- */
63: /* -- Global Includes --------------------------------------------------- */
64: /* ---------------------------------------------------------------------- */
65:
66: #include <sys/types.h>
1.11 ! guenther 67: #include <endian.h>
1.9 djm 68: #include <string.h>
69: #include <stdio.h>
70: #include <stdlib.h>
71: #include <stddef.h>
1.1 pvalchev 72:
1.2 stevesk 73: #include "xmalloc.h"
1.1 pvalchev 74: #include "umac.h"
1.9 djm 75: #include "misc.h"
1.1 pvalchev 76:
77: /* ---------------------------------------------------------------------- */
78: /* --- Primitive Data Types --- */
79: /* ---------------------------------------------------------------------- */
80:
81: /* The following assumptions may need change on your system */
82: typedef u_int8_t UINT8; /* 1 byte */
83: typedef u_int16_t UINT16; /* 2 byte */
84: typedef u_int32_t UINT32; /* 4 byte */
85: typedef u_int64_t UINT64; /* 8 bytes */
86: typedef unsigned int UWORD; /* Register */
87:
88: /* ---------------------------------------------------------------------- */
89: /* --- Constants -------------------------------------------------------- */
90: /* ---------------------------------------------------------------------- */
91:
92: #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */
93:
94: /* Message "words" are read from memory in an endian-specific manner. */
95: /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */
96: /* be set true if the host computer is little-endian. */
97:
98: #if BYTE_ORDER == LITTLE_ENDIAN
99: #define __LITTLE_ENDIAN__ 1
100: #else
101: #define __LITTLE_ENDIAN__ 0
102: #endif
103:
104: /* ---------------------------------------------------------------------- */
105: /* ---------------------------------------------------------------------- */
106: /* ----- Architecture Specific ------------------------------------------ */
107: /* ---------------------------------------------------------------------- */
108: /* ---------------------------------------------------------------------- */
109:
110:
111: /* ---------------------------------------------------------------------- */
112: /* ---------------------------------------------------------------------- */
113: /* ----- Primitive Routines --------------------------------------------- */
114: /* ---------------------------------------------------------------------- */
115: /* ---------------------------------------------------------------------- */
116:
117:
118: /* ---------------------------------------------------------------------- */
119: /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
120: /* ---------------------------------------------------------------------- */
121:
122: #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
123:
124: /* ---------------------------------------------------------------------- */
125: /* --- Endian Conversion --- Forcing assembly on some platforms */
126: /* ---------------------------------------------------------------------- */
127:
128: /* The following definitions use the above reversal-primitives to do the right
129: * thing on endian specific load and stores.
130: */
131:
1.9 djm 132: #if BYTE_ORDER == LITTLE_ENDIAN
133: #define LOAD_UINT32_REVERSED(p) get_u32(p)
134: #define STORE_UINT32_REVERSED(p,v) put_u32(p,v)
1.1 pvalchev 135: #else
1.9 djm 136: #define LOAD_UINT32_REVERSED(p) get_u32_le(p)
137: #define STORE_UINT32_REVERSED(p,v) put_u32_le(p,v)
1.1 pvalchev 138: #endif
1.9 djm 139:
140: #define LOAD_UINT32_LITTLE(p) (get_u32_le(p))
141: #define STORE_UINT32_BIG(p,v) put_u32(p, v)
1.1 pvalchev 142:
143:
144:
145: /* ---------------------------------------------------------------------- */
146: /* ---------------------------------------------------------------------- */
147: /* ----- Begin KDF & PDF Section ---------------------------------------- */
148: /* ---------------------------------------------------------------------- */
149: /* ---------------------------------------------------------------------- */
150:
151: /* UMAC uses AES with 16 byte block and key lengths */
152: #define AES_BLOCK_LEN 16
153:
1.10 naddy 154: #ifdef WITH_OPENSSL
1.1 pvalchev 155: #include <openssl/aes.h>
156: typedef AES_KEY aes_int_key[1];
157: #define aes_encryption(in,out,int_key) \
158: AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key)
159: #define aes_key_setup(key,int_key) \
1.7 djm 160: AES_set_encrypt_key((const u_char *)(key),UMAC_KEY_LEN*8,int_key)
1.10 naddy 161: #else
162: #include "rijndael.h"
163: #define AES_ROUNDS ((UMAC_KEY_LEN / 4) + 6)
164: typedef UINT8 aes_int_key[AES_ROUNDS+1][4][4]; /* AES internal */
165: #define aes_encryption(in,out,int_key) \
166: rijndaelEncrypt((u32 *)(int_key), AES_ROUNDS, (u8 *)(in), (u8 *)(out))
167: #define aes_key_setup(key,int_key) \
168: rijndaelKeySetupEnc((u32 *)(int_key), (const unsigned char *)(key), \
169: UMAC_KEY_LEN*8)
170: #endif
1.1 pvalchev 171:
172: /* The user-supplied UMAC key is stretched using AES in a counter
173: * mode to supply all random bits needed by UMAC. The kdf function takes
174: * an AES internal key representation 'key' and writes a stream of
175: * 'nbytes' bytes to the memory pointed at by 'buffer_ptr'. Each distinct
176: * 'ndx' causes a distinct byte stream.
177: */
178: static void kdf(void *buffer_ptr, aes_int_key key, UINT8 ndx, int nbytes)
179: {
180: UINT8 in_buf[AES_BLOCK_LEN] = {0};
181: UINT8 out_buf[AES_BLOCK_LEN];
182: UINT8 *dst_buf = (UINT8 *)buffer_ptr;
183: int i;
184:
185: /* Setup the initial value */
186: in_buf[AES_BLOCK_LEN-9] = ndx;
187: in_buf[AES_BLOCK_LEN-1] = i = 1;
188:
189: while (nbytes >= AES_BLOCK_LEN) {
190: aes_encryption(in_buf, out_buf, key);
191: memcpy(dst_buf,out_buf,AES_BLOCK_LEN);
192: in_buf[AES_BLOCK_LEN-1] = ++i;
193: nbytes -= AES_BLOCK_LEN;
194: dst_buf += AES_BLOCK_LEN;
195: }
196: if (nbytes) {
197: aes_encryption(in_buf, out_buf, key);
198: memcpy(dst_buf,out_buf,nbytes);
199: }
200: }
201:
202: /* The final UHASH result is XOR'd with the output of a pseudorandom
203: * function. Here, we use AES to generate random output and
204: * xor the appropriate bytes depending on the last bits of nonce.
205: * This scheme is optimized for sequential, increasing big-endian nonces.
206: */
207:
208: typedef struct {
209: UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */
210: UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */
211: aes_int_key prf_key; /* Expanded AES key for PDF */
212: } pdf_ctx;
213:
214: static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
215: {
216: UINT8 buf[UMAC_KEY_LEN];
217:
218: kdf(buf, prf_key, 0, UMAC_KEY_LEN);
219: aes_key_setup(buf, pc->prf_key);
220:
221: /* Initialize pdf and cache */
222: memset(pc->nonce, 0, sizeof(pc->nonce));
223: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
224: }
225:
1.7 djm 226: static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8])
1.1 pvalchev 227: {
228: /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
229: * of the AES output. If last time around we returned the ndx-1st
230: * element, then we may have the result in the cache already.
231: */
232:
233: #if (UMAC_OUTPUT_LEN == 4)
234: #define LOW_BIT_MASK 3
235: #elif (UMAC_OUTPUT_LEN == 8)
236: #define LOW_BIT_MASK 1
237: #elif (UMAC_OUTPUT_LEN > 8)
238: #define LOW_BIT_MASK 0
239: #endif
1.6 djm 240: union {
241: UINT8 tmp_nonce_lo[4];
242: UINT32 align;
243: } t;
1.1 pvalchev 244: #if LOW_BIT_MASK != 0
245: int ndx = nonce[7] & LOW_BIT_MASK;
246: #endif
1.7 djm 247: *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
1.6 djm 248: t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
1.1 pvalchev 249:
1.6 djm 250: if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
1.7 djm 251: (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
1.1 pvalchev 252: {
1.7 djm 253: ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
1.6 djm 254: ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
1.1 pvalchev 255: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
256: }
257:
258: #if (UMAC_OUTPUT_LEN == 4)
259: *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
260: #elif (UMAC_OUTPUT_LEN == 8)
261: *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
262: #elif (UMAC_OUTPUT_LEN == 12)
263: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
264: ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
265: #elif (UMAC_OUTPUT_LEN == 16)
266: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
267: ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
268: #endif
269: }
270:
271: /* ---------------------------------------------------------------------- */
272: /* ---------------------------------------------------------------------- */
273: /* ----- Begin NH Hash Section ------------------------------------------ */
274: /* ---------------------------------------------------------------------- */
275: /* ---------------------------------------------------------------------- */
276:
277: /* The NH-based hash functions used in UMAC are described in the UMAC paper
278: * and specification, both of which can be found at the UMAC website.
279: * The interface to this implementation has two
280: * versions, one expects the entire message being hashed to be passed
281: * in a single buffer and returns the hash result immediately. The second
282: * allows the message to be passed in a sequence of buffers. In the
283: * muliple-buffer interface, the client calls the routine nh_update() as
284: * many times as necessary. When there is no more data to be fed to the
285: * hash, the client calls nh_final() which calculates the hash output.
286: * Before beginning another hash calculation the nh_reset() routine
287: * must be called. The single-buffer routine, nh(), is equivalent to
288: * the sequence of calls nh_update() and nh_final(); however it is
289: * optimized and should be prefered whenever the multiple-buffer interface
290: * is not necessary. When using either interface, it is the client's
291: * responsability to pass no more than L1_KEY_LEN bytes per hash result.
292: *
293: * The routine nh_init() initializes the nh_ctx data structure and
294: * must be called once, before any other PDF routine.
295: */
296:
297: /* The "nh_aux" routines do the actual NH hashing work. They
298: * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
299: * produce output for all STREAMS NH iterations in one call,
300: * allowing the parallel implementation of the streams.
301: */
302:
303: #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */
304: #define L1_KEY_LEN 1024 /* Internal key bytes */
305: #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */
306: #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */
307: #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */
308: #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */
309:
310: typedef struct {
311: UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
1.4 djm 312: UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */
1.1 pvalchev 313: int next_data_empty; /* Bookeeping variable for data buffer. */
314: int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */
315: UINT64 state[STREAMS]; /* on-line state */
316: } nh_ctx;
317:
318:
319: #if (UMAC_OUTPUT_LEN == 4)
320:
1.7 djm 321: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 322: /* NH hashing primitive. Previous (partial) hash result is loaded and
323: * then stored via hp pointer. The length of the data pointed at by "dp",
324: * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key
325: * is expected to be endian compensated in memory at key setup.
326: */
327: {
328: UINT64 h;
329: UWORD c = dlen / 32;
330: UINT32 *k = (UINT32 *)kp;
1.7 djm 331: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 332: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
333: UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
334:
335: h = *((UINT64 *)hp);
336: do {
337: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
338: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
339: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
340: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
341: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
342: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
343: h += MUL64((k0 + d0), (k4 + d4));
344: h += MUL64((k1 + d1), (k5 + d5));
345: h += MUL64((k2 + d2), (k6 + d6));
346: h += MUL64((k3 + d3), (k7 + d7));
347:
348: d += 8;
349: k += 8;
350: } while (--c);
351: *((UINT64 *)hp) = h;
352: }
353:
354: #elif (UMAC_OUTPUT_LEN == 8)
355:
1.7 djm 356: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 357: /* Same as previous nh_aux, but two streams are handled in one pass,
358: * reading and writing 16 bytes of hash-state per call.
359: */
360: {
361: UINT64 h1,h2;
362: UWORD c = dlen / 32;
363: UINT32 *k = (UINT32 *)kp;
1.7 djm 364: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 365: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
366: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
367: k8,k9,k10,k11;
368:
369: h1 = *((UINT64 *)hp);
370: h2 = *((UINT64 *)hp + 1);
371: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
372: do {
373: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
374: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
375: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
376: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
377: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
378: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
379:
380: h1 += MUL64((k0 + d0), (k4 + d4));
381: h2 += MUL64((k4 + d0), (k8 + d4));
382:
383: h1 += MUL64((k1 + d1), (k5 + d5));
384: h2 += MUL64((k5 + d1), (k9 + d5));
385:
386: h1 += MUL64((k2 + d2), (k6 + d6));
387: h2 += MUL64((k6 + d2), (k10 + d6));
388:
389: h1 += MUL64((k3 + d3), (k7 + d7));
390: h2 += MUL64((k7 + d3), (k11 + d7));
391:
392: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
393:
394: d += 8;
395: k += 8;
396: } while (--c);
397: ((UINT64 *)hp)[0] = h1;
398: ((UINT64 *)hp)[1] = h2;
399: }
400:
401: #elif (UMAC_OUTPUT_LEN == 12)
402:
1.7 djm 403: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 404: /* Same as previous nh_aux, but two streams are handled in one pass,
405: * reading and writing 24 bytes of hash-state per call.
406: */
407: {
408: UINT64 h1,h2,h3;
409: UWORD c = dlen / 32;
410: UINT32 *k = (UINT32 *)kp;
1.7 djm 411: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 412: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
413: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
414: k8,k9,k10,k11,k12,k13,k14,k15;
415:
416: h1 = *((UINT64 *)hp);
417: h2 = *((UINT64 *)hp + 1);
418: h3 = *((UINT64 *)hp + 2);
419: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
420: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
421: do {
422: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
423: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
424: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
425: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
426: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
427: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
428:
429: h1 += MUL64((k0 + d0), (k4 + d4));
430: h2 += MUL64((k4 + d0), (k8 + d4));
431: h3 += MUL64((k8 + d0), (k12 + d4));
432:
433: h1 += MUL64((k1 + d1), (k5 + d5));
434: h2 += MUL64((k5 + d1), (k9 + d5));
435: h3 += MUL64((k9 + d1), (k13 + d5));
436:
437: h1 += MUL64((k2 + d2), (k6 + d6));
438: h2 += MUL64((k6 + d2), (k10 + d6));
439: h3 += MUL64((k10 + d2), (k14 + d6));
440:
441: h1 += MUL64((k3 + d3), (k7 + d7));
442: h2 += MUL64((k7 + d3), (k11 + d7));
443: h3 += MUL64((k11 + d3), (k15 + d7));
444:
445: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
446: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
447:
448: d += 8;
449: k += 8;
450: } while (--c);
451: ((UINT64 *)hp)[0] = h1;
452: ((UINT64 *)hp)[1] = h2;
453: ((UINT64 *)hp)[2] = h3;
454: }
455:
456: #elif (UMAC_OUTPUT_LEN == 16)
457:
1.7 djm 458: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 459: /* Same as previous nh_aux, but two streams are handled in one pass,
460: * reading and writing 24 bytes of hash-state per call.
461: */
462: {
463: UINT64 h1,h2,h3,h4;
464: UWORD c = dlen / 32;
465: UINT32 *k = (UINT32 *)kp;
1.7 djm 466: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 467: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
468: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
469: k8,k9,k10,k11,k12,k13,k14,k15,
470: k16,k17,k18,k19;
471:
472: h1 = *((UINT64 *)hp);
473: h2 = *((UINT64 *)hp + 1);
474: h3 = *((UINT64 *)hp + 2);
475: h4 = *((UINT64 *)hp + 3);
476: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
477: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
478: do {
479: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
480: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
481: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
482: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
483: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
484: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
485: k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
486:
487: h1 += MUL64((k0 + d0), (k4 + d4));
488: h2 += MUL64((k4 + d0), (k8 + d4));
489: h3 += MUL64((k8 + d0), (k12 + d4));
490: h4 += MUL64((k12 + d0), (k16 + d4));
491:
492: h1 += MUL64((k1 + d1), (k5 + d5));
493: h2 += MUL64((k5 + d1), (k9 + d5));
494: h3 += MUL64((k9 + d1), (k13 + d5));
495: h4 += MUL64((k13 + d1), (k17 + d5));
496:
497: h1 += MUL64((k2 + d2), (k6 + d6));
498: h2 += MUL64((k6 + d2), (k10 + d6));
499: h3 += MUL64((k10 + d2), (k14 + d6));
500: h4 += MUL64((k14 + d2), (k18 + d6));
501:
502: h1 += MUL64((k3 + d3), (k7 + d7));
503: h2 += MUL64((k7 + d3), (k11 + d7));
504: h3 += MUL64((k11 + d3), (k15 + d7));
505: h4 += MUL64((k15 + d3), (k19 + d7));
506:
507: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
508: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
509: k8 = k16; k9 = k17; k10 = k18; k11 = k19;
510:
511: d += 8;
512: k += 8;
513: } while (--c);
514: ((UINT64 *)hp)[0] = h1;
515: ((UINT64 *)hp)[1] = h2;
516: ((UINT64 *)hp)[2] = h3;
517: ((UINT64 *)hp)[3] = h4;
518: }
519:
520: /* ---------------------------------------------------------------------- */
521: #endif /* UMAC_OUTPUT_LENGTH */
522: /* ---------------------------------------------------------------------- */
523:
524:
525: /* ---------------------------------------------------------------------- */
526:
1.7 djm 527: static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
1.1 pvalchev 528: /* This function is a wrapper for the primitive NH hash functions. It takes
529: * as argument "hc" the current hash context and a buffer which must be a
530: * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
531: * appropriately according to how much message has been hashed already.
532: */
533: {
534: UINT8 *key;
535:
536: key = hc->nh_key + hc->bytes_hashed;
537: nh_aux(key, buf, hc->state, nbytes);
538: }
539:
540: /* ---------------------------------------------------------------------- */
541:
542: static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
543: /* We endian convert the keys on little-endian computers to */
544: /* compensate for the lack of big-endian memory reads during hashing. */
545: {
546: UWORD iters = num_bytes / bpw;
547: if (bpw == 4) {
548: UINT32 *p = (UINT32 *)buf;
549: do {
550: *p = LOAD_UINT32_REVERSED(p);
551: p++;
552: } while (--iters);
553: } else if (bpw == 8) {
554: UINT32 *p = (UINT32 *)buf;
555: UINT32 t;
556: do {
557: t = LOAD_UINT32_REVERSED(p+1);
558: p[1] = LOAD_UINT32_REVERSED(p);
559: p[0] = t;
560: p += 2;
561: } while (--iters);
562: }
563: }
564: #if (__LITTLE_ENDIAN__)
565: #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
566: #else
567: #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */
568: #endif
569:
570: /* ---------------------------------------------------------------------- */
571:
572: static void nh_reset(nh_ctx *hc)
573: /* Reset nh_ctx to ready for hashing of new data */
574: {
575: hc->bytes_hashed = 0;
576: hc->next_data_empty = 0;
577: hc->state[0] = 0;
578: #if (UMAC_OUTPUT_LEN >= 8)
579: hc->state[1] = 0;
580: #endif
581: #if (UMAC_OUTPUT_LEN >= 12)
582: hc->state[2] = 0;
583: #endif
584: #if (UMAC_OUTPUT_LEN == 16)
585: hc->state[3] = 0;
586: #endif
587:
588: }
589:
590: /* ---------------------------------------------------------------------- */
591:
592: static void nh_init(nh_ctx *hc, aes_int_key prf_key)
593: /* Generate nh_key, endian convert and reset to be ready for hashing. */
594: {
595: kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
596: endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
597: nh_reset(hc);
598: }
599:
600: /* ---------------------------------------------------------------------- */
601:
1.7 djm 602: static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
1.1 pvalchev 603: /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */
604: /* even multiple of HASH_BUF_BYTES. */
605: {
606: UINT32 i,j;
607:
608: j = hc->next_data_empty;
609: if ((j + nbytes) >= HASH_BUF_BYTES) {
610: if (j) {
611: i = HASH_BUF_BYTES - j;
612: memcpy(hc->data+j, buf, i);
613: nh_transform(hc,hc->data,HASH_BUF_BYTES);
614: nbytes -= i;
615: buf += i;
616: hc->bytes_hashed += HASH_BUF_BYTES;
617: }
618: if (nbytes >= HASH_BUF_BYTES) {
619: i = nbytes & ~(HASH_BUF_BYTES - 1);
620: nh_transform(hc, buf, i);
621: nbytes -= i;
622: buf += i;
623: hc->bytes_hashed += i;
624: }
625: j = 0;
626: }
627: memcpy(hc->data + j, buf, nbytes);
628: hc->next_data_empty = j + nbytes;
629: }
630:
631: /* ---------------------------------------------------------------------- */
632:
633: static void zero_pad(UINT8 *p, int nbytes)
634: {
635: /* Write "nbytes" of zeroes, beginning at "p" */
636: if (nbytes >= (int)sizeof(UWORD)) {
637: while ((ptrdiff_t)p % sizeof(UWORD)) {
638: *p = 0;
639: nbytes--;
640: p++;
641: }
642: while (nbytes >= (int)sizeof(UWORD)) {
643: *(UWORD *)p = 0;
644: nbytes -= sizeof(UWORD);
645: p += sizeof(UWORD);
646: }
647: }
648: while (nbytes) {
649: *p = 0;
650: nbytes--;
651: p++;
652: }
653: }
654:
655: /* ---------------------------------------------------------------------- */
656:
657: static void nh_final(nh_ctx *hc, UINT8 *result)
658: /* After passing some number of data buffers to nh_update() for integration
659: * into an NH context, nh_final is called to produce a hash result. If any
660: * bytes are in the buffer hc->data, incorporate them into the
661: * NH context. Finally, add into the NH accumulation "state" the total number
662: * of bits hashed. The resulting numbers are written to the buffer "result".
663: * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
664: */
665: {
666: int nh_len, nbits;
667:
668: if (hc->next_data_empty != 0) {
669: nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
670: ~(L1_PAD_BOUNDARY - 1));
671: zero_pad(hc->data + hc->next_data_empty,
672: nh_len - hc->next_data_empty);
673: nh_transform(hc, hc->data, nh_len);
674: hc->bytes_hashed += hc->next_data_empty;
675: } else if (hc->bytes_hashed == 0) {
676: nh_len = L1_PAD_BOUNDARY;
677: zero_pad(hc->data, L1_PAD_BOUNDARY);
678: nh_transform(hc, hc->data, nh_len);
679: }
680:
681: nbits = (hc->bytes_hashed << 3);
682: ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
683: #if (UMAC_OUTPUT_LEN >= 8)
684: ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
685: #endif
686: #if (UMAC_OUTPUT_LEN >= 12)
687: ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
688: #endif
689: #if (UMAC_OUTPUT_LEN == 16)
690: ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
691: #endif
692: nh_reset(hc);
693: }
694:
695: /* ---------------------------------------------------------------------- */
696:
1.7 djm 697: static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
1.1 pvalchev 698: UINT32 unpadded_len, UINT8 *result)
699: /* All-in-one nh_update() and nh_final() equivalent.
700: * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
701: * well aligned
702: */
703: {
704: UINT32 nbits;
705:
706: /* Initialize the hash state */
707: nbits = (unpadded_len << 3);
708:
709: ((UINT64 *)result)[0] = nbits;
710: #if (UMAC_OUTPUT_LEN >= 8)
711: ((UINT64 *)result)[1] = nbits;
712: #endif
713: #if (UMAC_OUTPUT_LEN >= 12)
714: ((UINT64 *)result)[2] = nbits;
715: #endif
716: #if (UMAC_OUTPUT_LEN == 16)
717: ((UINT64 *)result)[3] = nbits;
718: #endif
719:
720: nh_aux(hc->nh_key, buf, result, padded_len);
721: }
722:
723: /* ---------------------------------------------------------------------- */
724: /* ---------------------------------------------------------------------- */
725: /* ----- Begin UHASH Section -------------------------------------------- */
726: /* ---------------------------------------------------------------------- */
727: /* ---------------------------------------------------------------------- */
728:
729: /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
730: * hashed by NH. The NH output is then hashed by a polynomial-hash layer
731: * unless the initial data to be hashed is short. After the polynomial-
732: * layer, an inner-product hash is used to produce the final UHASH output.
733: *
734: * UHASH provides two interfaces, one all-at-once and another where data
735: * buffers are presented sequentially. In the sequential interface, the
736: * UHASH client calls the routine uhash_update() as many times as necessary.
737: * When there is no more data to be fed to UHASH, the client calls
738: * uhash_final() which
739: * calculates the UHASH output. Before beginning another UHASH calculation
740: * the uhash_reset() routine must be called. The all-at-once UHASH routine,
741: * uhash(), is equivalent to the sequence of calls uhash_update() and
742: * uhash_final(); however it is optimized and should be
743: * used whenever the sequential interface is not necessary.
744: *
745: * The routine uhash_init() initializes the uhash_ctx data structure and
746: * must be called once, before any other UHASH routine.
747: */
748:
749: /* ---------------------------------------------------------------------- */
750: /* ----- Constants and uhash_ctx ---------------------------------------- */
751: /* ---------------------------------------------------------------------- */
752:
753: /* ---------------------------------------------------------------------- */
754: /* ----- Poly hash and Inner-Product hash Constants --------------------- */
755: /* ---------------------------------------------------------------------- */
756:
757: /* Primes and masks */
758: #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */
759: #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */
760: #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */
761:
762:
763: /* ---------------------------------------------------------------------- */
764:
765: typedef struct uhash_ctx {
766: nh_ctx hash; /* Hash context for L1 NH hash */
767: UINT64 poly_key_8[STREAMS]; /* p64 poly keys */
768: UINT64 poly_accum[STREAMS]; /* poly hash result */
769: UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */
770: UINT32 ip_trans[STREAMS]; /* Inner-product translation */
771: UINT32 msg_len; /* Total length of data passed */
772: /* to uhash */
773: } uhash_ctx;
774: typedef struct uhash_ctx *uhash_ctx_t;
775:
776: /* ---------------------------------------------------------------------- */
777:
778:
779: /* The polynomial hashes use Horner's rule to evaluate a polynomial one
780: * word at a time. As described in the specification, poly32 and poly64
781: * require keys from special domains. The following implementations exploit
782: * the special domains to avoid overflow. The results are not guaranteed to
783: * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
784: * patches any errant values.
785: */
786:
787: static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
788: {
789: UINT32 key_hi = (UINT32)(key >> 32),
790: key_lo = (UINT32)key,
791: cur_hi = (UINT32)(cur >> 32),
792: cur_lo = (UINT32)cur,
793: x_lo,
794: x_hi;
795: UINT64 X,T,res;
796:
797: X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
798: x_lo = (UINT32)X;
799: x_hi = (UINT32)(X >> 32);
800:
801: res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
802:
803: T = ((UINT64)x_lo << 32);
804: res += T;
805: if (res < T)
806: res += 59;
807:
808: res += data;
809: if (res < data)
810: res += 59;
811:
812: return res;
813: }
814:
815:
816: /* Although UMAC is specified to use a ramped polynomial hash scheme, this
817: * implementation does not handle all ramp levels. Because we don't handle
818: * the ramp up to p128 modulus in this implementation, we are limited to
819: * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
820: * bytes input to UMAC per tag, ie. 16MB).
821: */
822: static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
823: {
824: int i;
825: UINT64 *data=(UINT64*)data_in;
826:
827: for (i = 0; i < STREAMS; i++) {
828: if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
829: hc->poly_accum[i] = poly64(hc->poly_accum[i],
830: hc->poly_key_8[i], p64 - 1);
831: hc->poly_accum[i] = poly64(hc->poly_accum[i],
832: hc->poly_key_8[i], (data[i] - 59));
833: } else {
834: hc->poly_accum[i] = poly64(hc->poly_accum[i],
835: hc->poly_key_8[i], data[i]);
836: }
837: }
838: }
839:
840:
841: /* ---------------------------------------------------------------------- */
842:
843:
844: /* The final step in UHASH is an inner-product hash. The poly hash
845: * produces a result not neccesarily WORD_LEN bytes long. The inner-
846: * product hash breaks the polyhash output into 16-bit chunks and
847: * multiplies each with a 36 bit key.
848: */
849:
850: static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
851: {
852: t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
853: t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
854: t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
855: t = t + ipkp[3] * (UINT64)(UINT16)(data);
856:
857: return t;
858: }
859:
860: static UINT32 ip_reduce_p36(UINT64 t)
861: {
862: /* Divisionless modular reduction */
863: UINT64 ret;
864:
865: ret = (t & m36) + 5 * (t >> 36);
866: if (ret >= p36)
867: ret -= p36;
868:
869: /* return least significant 32 bits */
870: return (UINT32)(ret);
871: }
872:
873:
874: /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
875: * the polyhash stage is skipped and ip_short is applied directly to the
876: * NH output.
877: */
878: static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
879: {
880: UINT64 t;
881: UINT64 *nhp = (UINT64 *)nh_res;
882:
883: t = ip_aux(0,ahc->ip_keys, nhp[0]);
884: STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
885: #if (UMAC_OUTPUT_LEN >= 8)
886: t = ip_aux(0,ahc->ip_keys+4, nhp[1]);
887: STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
888: #endif
889: #if (UMAC_OUTPUT_LEN >= 12)
890: t = ip_aux(0,ahc->ip_keys+8, nhp[2]);
891: STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
892: #endif
893: #if (UMAC_OUTPUT_LEN == 16)
894: t = ip_aux(0,ahc->ip_keys+12, nhp[3]);
895: STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
896: #endif
897: }
898:
899: /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
900: * the polyhash stage is not skipped and ip_long is applied to the
901: * polyhash output.
902: */
903: static void ip_long(uhash_ctx_t ahc, u_char *res)
904: {
905: int i;
906: UINT64 t;
907:
908: for (i = 0; i < STREAMS; i++) {
909: /* fix polyhash output not in Z_p64 */
910: if (ahc->poly_accum[i] >= p64)
911: ahc->poly_accum[i] -= p64;
912: t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
913: STORE_UINT32_BIG((UINT32 *)res+i,
914: ip_reduce_p36(t) ^ ahc->ip_trans[i]);
915: }
916: }
917:
918:
919: /* ---------------------------------------------------------------------- */
920:
921: /* ---------------------------------------------------------------------- */
922:
923: /* Reset uhash context for next hash session */
924: static int uhash_reset(uhash_ctx_t pc)
925: {
926: nh_reset(&pc->hash);
927: pc->msg_len = 0;
928: pc->poly_accum[0] = 1;
929: #if (UMAC_OUTPUT_LEN >= 8)
930: pc->poly_accum[1] = 1;
931: #endif
932: #if (UMAC_OUTPUT_LEN >= 12)
933: pc->poly_accum[2] = 1;
934: #endif
935: #if (UMAC_OUTPUT_LEN == 16)
936: pc->poly_accum[3] = 1;
937: #endif
938: return 1;
939: }
940:
941: /* ---------------------------------------------------------------------- */
942:
943: /* Given a pointer to the internal key needed by kdf() and a uhash context,
944: * initialize the NH context and generate keys needed for poly and inner-
945: * product hashing. All keys are endian adjusted in memory so that native
946: * loads cause correct keys to be in registers during calculation.
947: */
948: static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
949: {
950: int i;
951: UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
952:
953: /* Zero the entire uhash context */
954: memset(ahc, 0, sizeof(uhash_ctx));
955:
956: /* Initialize the L1 hash */
957: nh_init(&ahc->hash, prf_key);
958:
959: /* Setup L2 hash variables */
960: kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */
961: for (i = 0; i < STREAMS; i++) {
962: /* Fill keys from the buffer, skipping bytes in the buffer not
963: * used by this implementation. Endian reverse the keys if on a
964: * little-endian computer.
965: */
966: memcpy(ahc->poly_key_8+i, buf+24*i, 8);
967: endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
968: /* Mask the 64-bit keys to their special domain */
969: ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
970: ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */
971: }
972:
973: /* Setup L3-1 hash variables */
974: kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
975: for (i = 0; i < STREAMS; i++)
976: memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
977: 4*sizeof(UINT64));
978: endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
979: sizeof(ahc->ip_keys));
980: for (i = 0; i < STREAMS*4; i++)
981: ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */
982:
983: /* Setup L3-2 hash variables */
984: /* Fill buffer with index 4 key */
985: kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
986: endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
987: STREAMS * sizeof(UINT32));
988: }
989:
990: /* ---------------------------------------------------------------------- */
991:
992: #if 0
993: static uhash_ctx_t uhash_alloc(u_char key[])
994: {
995: /* Allocate memory and force to a 16-byte boundary. */
996: uhash_ctx_t ctx;
997: u_char bytes_to_add;
998: aes_int_key prf_key;
999:
1000: ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1001: if (ctx) {
1002: if (ALLOC_BOUNDARY) {
1003: bytes_to_add = ALLOC_BOUNDARY -
1004: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1005: ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1006: *((u_char *)ctx - 1) = bytes_to_add;
1007: }
1008: aes_key_setup(key,prf_key);
1009: uhash_init(ctx, prf_key);
1010: }
1011: return (ctx);
1012: }
1013: #endif
1014:
1015: /* ---------------------------------------------------------------------- */
1016:
1017: #if 0
1018: static int uhash_free(uhash_ctx_t ctx)
1019: {
1020: /* Free memory allocated by uhash_alloc */
1021: u_char bytes_to_sub;
1022:
1023: if (ctx) {
1024: if (ALLOC_BOUNDARY) {
1025: bytes_to_sub = *((u_char *)ctx - 1);
1026: ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1027: }
1028: free(ctx);
1029: }
1030: return (1);
1031: }
1032: #endif
1033: /* ---------------------------------------------------------------------- */
1034:
1.7 djm 1035: static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1.1 pvalchev 1036: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1037: * hash each one with NH, calling the polyhash on each NH output.
1038: */
1039: {
1040: UWORD bytes_hashed, bytes_remaining;
1.3 pvalchev 1041: UINT64 result_buf[STREAMS];
1042: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1043:
1044: if (ctx->msg_len + len <= L1_KEY_LEN) {
1.7 djm 1045: nh_update(&ctx->hash, (const UINT8 *)input, len);
1.1 pvalchev 1046: ctx->msg_len += len;
1047: } else {
1048:
1049: bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1050: if (ctx->msg_len == L1_KEY_LEN)
1051: bytes_hashed = L1_KEY_LEN;
1052:
1053: if (bytes_hashed + len >= L1_KEY_LEN) {
1054:
1055: /* If some bytes have been passed to the hash function */
1056: /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1057: /* bytes to complete the current nh_block. */
1058: if (bytes_hashed) {
1059: bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1.7 djm 1060: nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1.1 pvalchev 1061: nh_final(&ctx->hash, nh_result);
1062: ctx->msg_len += bytes_remaining;
1063: poly_hash(ctx,(UINT32 *)nh_result);
1064: len -= bytes_remaining;
1065: input += bytes_remaining;
1066: }
1067:
1068: /* Hash directly from input stream if enough bytes */
1069: while (len >= L1_KEY_LEN) {
1.7 djm 1070: nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1.1 pvalchev 1071: L1_KEY_LEN, nh_result);
1072: ctx->msg_len += L1_KEY_LEN;
1073: len -= L1_KEY_LEN;
1074: input += L1_KEY_LEN;
1075: poly_hash(ctx,(UINT32 *)nh_result);
1076: }
1077: }
1078:
1079: /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1080: if (len) {
1.7 djm 1081: nh_update(&ctx->hash, (const UINT8 *)input, len);
1.1 pvalchev 1082: ctx->msg_len += len;
1083: }
1084: }
1085:
1086: return (1);
1087: }
1088:
1089: /* ---------------------------------------------------------------------- */
1090:
1091: static int uhash_final(uhash_ctx_t ctx, u_char *res)
1092: /* Incorporate any pending data, pad, and generate tag */
1093: {
1.3 pvalchev 1094: UINT64 result_buf[STREAMS];
1095: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1096:
1097: if (ctx->msg_len > L1_KEY_LEN) {
1098: if (ctx->msg_len % L1_KEY_LEN) {
1099: nh_final(&ctx->hash, nh_result);
1100: poly_hash(ctx,(UINT32 *)nh_result);
1101: }
1102: ip_long(ctx, res);
1103: } else {
1104: nh_final(&ctx->hash, nh_result);
1105: ip_short(ctx,nh_result, res);
1106: }
1107: uhash_reset(ctx);
1108: return (1);
1109: }
1110:
1111: /* ---------------------------------------------------------------------- */
1112:
1113: #if 0
1114: static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1115: /* assumes that msg is in a writable buffer of length divisible by */
1116: /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */
1117: {
1118: UINT8 nh_result[STREAMS*sizeof(UINT64)];
1119: UINT32 nh_len;
1120: int extra_zeroes_needed;
1121:
1122: /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1123: * the polyhash.
1124: */
1125: if (len <= L1_KEY_LEN) {
1126: if (len == 0) /* If zero length messages will not */
1127: nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */
1128: else
1129: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1130: extra_zeroes_needed = nh_len - len;
1131: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1132: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1133: ip_short(ahc,nh_result, res);
1134: } else {
1135: /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1136: * output to poly_hash().
1137: */
1138: do {
1139: nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1140: poly_hash(ahc,(UINT32 *)nh_result);
1141: len -= L1_KEY_LEN;
1142: msg += L1_KEY_LEN;
1143: } while (len >= L1_KEY_LEN);
1144: if (len) {
1145: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1146: extra_zeroes_needed = nh_len - len;
1147: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1148: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1149: poly_hash(ahc,(UINT32 *)nh_result);
1150: }
1151:
1152: ip_long(ahc, res);
1153: }
1154:
1155: uhash_reset(ahc);
1156: return 1;
1157: }
1158: #endif
1159:
1160: /* ---------------------------------------------------------------------- */
1161: /* ---------------------------------------------------------------------- */
1162: /* ----- Begin UMAC Section --------------------------------------------- */
1163: /* ---------------------------------------------------------------------- */
1164: /* ---------------------------------------------------------------------- */
1165:
1166: /* The UMAC interface has two interfaces, an all-at-once interface where
1167: * the entire message to be authenticated is passed to UMAC in one buffer,
1168: * and a sequential interface where the message is presented a little at a
1169: * time. The all-at-once is more optimaized than the sequential version and
1170: * should be preferred when the sequential interface is not required.
1171: */
1172: struct umac_ctx {
1173: uhash_ctx hash; /* Hash function for message compression */
1174: pdf_ctx pdf; /* PDF for hashed output */
1175: void *free_ptr; /* Address to free this struct via */
1176: } umac_ctx;
1177:
1178: /* ---------------------------------------------------------------------- */
1179:
1180: #if 0
1181: int umac_reset(struct umac_ctx *ctx)
1182: /* Reset the hash function to begin a new authentication. */
1183: {
1184: uhash_reset(&ctx->hash);
1185: return (1);
1186: }
1187: #endif
1188:
1189: /* ---------------------------------------------------------------------- */
1190:
1191: int umac_delete(struct umac_ctx *ctx)
1192: /* Deallocate the ctx structure */
1193: {
1194: if (ctx) {
1195: if (ALLOC_BOUNDARY)
1196: ctx = (struct umac_ctx *)ctx->free_ptr;
1.5 djm 1197: free(ctx);
1.1 pvalchev 1198: }
1199: return (1);
1200: }
1201:
1202: /* ---------------------------------------------------------------------- */
1203:
1.7 djm 1204: struct umac_ctx *umac_new(const u_char key[])
1.1 pvalchev 1205: /* Dynamically allocate a umac_ctx struct, initialize variables,
1206: * generate subkeys from key. Align to 16-byte boundary.
1207: */
1208: {
1209: struct umac_ctx *ctx, *octx;
1210: size_t bytes_to_add;
1211: aes_int_key prf_key;
1212:
1.8 djm 1213: octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1.1 pvalchev 1214: if (ctx) {
1215: if (ALLOC_BOUNDARY) {
1216: bytes_to_add = ALLOC_BOUNDARY -
1217: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1218: ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1219: }
1220: ctx->free_ptr = octx;
1.7 djm 1221: aes_key_setup(key, prf_key);
1.1 pvalchev 1222: pdf_init(&ctx->pdf, prf_key);
1223: uhash_init(&ctx->hash, prf_key);
1224: }
1225:
1226: return (ctx);
1227: }
1228:
1229: /* ---------------------------------------------------------------------- */
1230:
1.7 djm 1231: int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1.1 pvalchev 1232: /* Incorporate any pending data, pad, and generate tag */
1233: {
1234: uhash_final(&ctx->hash, (u_char *)tag);
1.7 djm 1235: pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1.1 pvalchev 1236:
1237: return (1);
1238: }
1239:
1240: /* ---------------------------------------------------------------------- */
1241:
1.7 djm 1242: int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1.1 pvalchev 1243: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */
1244: /* hash each one, calling the PDF on the hashed output whenever the hash- */
1245: /* output buffer is full. */
1246: {
1247: uhash_update(&ctx->hash, input, len);
1248: return (1);
1249: }
1250:
1251: /* ---------------------------------------------------------------------- */
1252:
1253: #if 0
1254: int umac(struct umac_ctx *ctx, u_char *input,
1255: long len, u_char tag[],
1256: u_char nonce[8])
1257: /* All-in-one version simply calls umac_update() and umac_final(). */
1258: {
1259: uhash(&ctx->hash, input, len, (u_char *)tag);
1260: pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1261:
1262: return (1);
1263: }
1264: #endif
1265:
1266: /* ---------------------------------------------------------------------- */
1267: /* ---------------------------------------------------------------------- */
1268: /* ----- End UMAC Section ----------------------------------------------- */
1269: /* ---------------------------------------------------------------------- */
1270: /* ---------------------------------------------------------------------- */