Annotation of src/usr.bin/ssh/umac.c, Revision 1.12
1.12 ! markus 1: /* $OpenBSD: umac.c,v 1.11 2014/07/22 07:13:42 guenther 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: }
1.12 ! markus 200: explicit_bzero(in_buf, sizeof(in_buf));
! 201: explicit_bzero(out_buf, sizeof(out_buf));
1.1 pvalchev 202: }
203:
204: /* The final UHASH result is XOR'd with the output of a pseudorandom
205: * function. Here, we use AES to generate random output and
206: * xor the appropriate bytes depending on the last bits of nonce.
207: * This scheme is optimized for sequential, increasing big-endian nonces.
208: */
209:
210: typedef struct {
211: UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */
212: UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */
213: aes_int_key prf_key; /* Expanded AES key for PDF */
214: } pdf_ctx;
215:
216: static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
217: {
218: UINT8 buf[UMAC_KEY_LEN];
219:
220: kdf(buf, prf_key, 0, UMAC_KEY_LEN);
221: aes_key_setup(buf, pc->prf_key);
222:
223: /* Initialize pdf and cache */
224: memset(pc->nonce, 0, sizeof(pc->nonce));
225: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
1.12 ! markus 226: explicit_bzero(buf, sizeof(buf));
1.1 pvalchev 227: }
228:
1.7 djm 229: static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8])
1.1 pvalchev 230: {
231: /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
232: * of the AES output. If last time around we returned the ndx-1st
233: * element, then we may have the result in the cache already.
234: */
235:
236: #if (UMAC_OUTPUT_LEN == 4)
237: #define LOW_BIT_MASK 3
238: #elif (UMAC_OUTPUT_LEN == 8)
239: #define LOW_BIT_MASK 1
240: #elif (UMAC_OUTPUT_LEN > 8)
241: #define LOW_BIT_MASK 0
242: #endif
1.6 djm 243: union {
244: UINT8 tmp_nonce_lo[4];
245: UINT32 align;
246: } t;
1.1 pvalchev 247: #if LOW_BIT_MASK != 0
248: int ndx = nonce[7] & LOW_BIT_MASK;
249: #endif
1.7 djm 250: *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
1.6 djm 251: t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
1.1 pvalchev 252:
1.6 djm 253: if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
1.7 djm 254: (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
1.1 pvalchev 255: {
1.7 djm 256: ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
1.6 djm 257: ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
1.1 pvalchev 258: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
259: }
260:
261: #if (UMAC_OUTPUT_LEN == 4)
262: *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
263: #elif (UMAC_OUTPUT_LEN == 8)
264: *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
265: #elif (UMAC_OUTPUT_LEN == 12)
266: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
267: ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
268: #elif (UMAC_OUTPUT_LEN == 16)
269: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
270: ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
271: #endif
272: }
273:
274: /* ---------------------------------------------------------------------- */
275: /* ---------------------------------------------------------------------- */
276: /* ----- Begin NH Hash Section ------------------------------------------ */
277: /* ---------------------------------------------------------------------- */
278: /* ---------------------------------------------------------------------- */
279:
280: /* The NH-based hash functions used in UMAC are described in the UMAC paper
281: * and specification, both of which can be found at the UMAC website.
282: * The interface to this implementation has two
283: * versions, one expects the entire message being hashed to be passed
284: * in a single buffer and returns the hash result immediately. The second
285: * allows the message to be passed in a sequence of buffers. In the
286: * muliple-buffer interface, the client calls the routine nh_update() as
287: * many times as necessary. When there is no more data to be fed to the
288: * hash, the client calls nh_final() which calculates the hash output.
289: * Before beginning another hash calculation the nh_reset() routine
290: * must be called. The single-buffer routine, nh(), is equivalent to
291: * the sequence of calls nh_update() and nh_final(); however it is
292: * optimized and should be prefered whenever the multiple-buffer interface
293: * is not necessary. When using either interface, it is the client's
294: * responsability to pass no more than L1_KEY_LEN bytes per hash result.
295: *
296: * The routine nh_init() initializes the nh_ctx data structure and
297: * must be called once, before any other PDF routine.
298: */
299:
300: /* The "nh_aux" routines do the actual NH hashing work. They
301: * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
302: * produce output for all STREAMS NH iterations in one call,
303: * allowing the parallel implementation of the streams.
304: */
305:
306: #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */
307: #define L1_KEY_LEN 1024 /* Internal key bytes */
308: #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */
309: #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */
310: #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */
311: #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */
312:
313: typedef struct {
314: UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
1.4 djm 315: UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */
1.1 pvalchev 316: int next_data_empty; /* Bookeeping variable for data buffer. */
317: int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */
318: UINT64 state[STREAMS]; /* on-line state */
319: } nh_ctx;
320:
321:
322: #if (UMAC_OUTPUT_LEN == 4)
323:
1.7 djm 324: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 325: /* NH hashing primitive. Previous (partial) hash result is loaded and
326: * then stored via hp pointer. The length of the data pointed at by "dp",
327: * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key
328: * is expected to be endian compensated in memory at key setup.
329: */
330: {
331: UINT64 h;
332: UWORD c = dlen / 32;
333: UINT32 *k = (UINT32 *)kp;
1.7 djm 334: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 335: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
336: UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
337:
338: h = *((UINT64 *)hp);
339: do {
340: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
341: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
342: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
343: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
344: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
345: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
346: h += MUL64((k0 + d0), (k4 + d4));
347: h += MUL64((k1 + d1), (k5 + d5));
348: h += MUL64((k2 + d2), (k6 + d6));
349: h += MUL64((k3 + d3), (k7 + d7));
350:
351: d += 8;
352: k += 8;
353: } while (--c);
354: *((UINT64 *)hp) = h;
355: }
356:
357: #elif (UMAC_OUTPUT_LEN == 8)
358:
1.7 djm 359: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 360: /* Same as previous nh_aux, but two streams are handled in one pass,
361: * reading and writing 16 bytes of hash-state per call.
362: */
363: {
364: UINT64 h1,h2;
365: UWORD c = dlen / 32;
366: UINT32 *k = (UINT32 *)kp;
1.7 djm 367: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 368: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
369: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
370: k8,k9,k10,k11;
371:
372: h1 = *((UINT64 *)hp);
373: h2 = *((UINT64 *)hp + 1);
374: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
375: do {
376: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
377: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
378: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
379: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
380: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
381: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
382:
383: h1 += MUL64((k0 + d0), (k4 + d4));
384: h2 += MUL64((k4 + d0), (k8 + d4));
385:
386: h1 += MUL64((k1 + d1), (k5 + d5));
387: h2 += MUL64((k5 + d1), (k9 + d5));
388:
389: h1 += MUL64((k2 + d2), (k6 + d6));
390: h2 += MUL64((k6 + d2), (k10 + d6));
391:
392: h1 += MUL64((k3 + d3), (k7 + d7));
393: h2 += MUL64((k7 + d3), (k11 + d7));
394:
395: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
396:
397: d += 8;
398: k += 8;
399: } while (--c);
400: ((UINT64 *)hp)[0] = h1;
401: ((UINT64 *)hp)[1] = h2;
402: }
403:
404: #elif (UMAC_OUTPUT_LEN == 12)
405:
1.7 djm 406: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 407: /* Same as previous nh_aux, but two streams are handled in one pass,
408: * reading and writing 24 bytes of hash-state per call.
409: */
410: {
411: UINT64 h1,h2,h3;
412: UWORD c = dlen / 32;
413: UINT32 *k = (UINT32 *)kp;
1.7 djm 414: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 415: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
416: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
417: k8,k9,k10,k11,k12,k13,k14,k15;
418:
419: h1 = *((UINT64 *)hp);
420: h2 = *((UINT64 *)hp + 1);
421: h3 = *((UINT64 *)hp + 2);
422: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
423: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
424: do {
425: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
426: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
427: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
428: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
429: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
430: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
431:
432: h1 += MUL64((k0 + d0), (k4 + d4));
433: h2 += MUL64((k4 + d0), (k8 + d4));
434: h3 += MUL64((k8 + d0), (k12 + d4));
435:
436: h1 += MUL64((k1 + d1), (k5 + d5));
437: h2 += MUL64((k5 + d1), (k9 + d5));
438: h3 += MUL64((k9 + d1), (k13 + d5));
439:
440: h1 += MUL64((k2 + d2), (k6 + d6));
441: h2 += MUL64((k6 + d2), (k10 + d6));
442: h3 += MUL64((k10 + d2), (k14 + d6));
443:
444: h1 += MUL64((k3 + d3), (k7 + d7));
445: h2 += MUL64((k7 + d3), (k11 + d7));
446: h3 += MUL64((k11 + d3), (k15 + d7));
447:
448: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
449: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
450:
451: d += 8;
452: k += 8;
453: } while (--c);
454: ((UINT64 *)hp)[0] = h1;
455: ((UINT64 *)hp)[1] = h2;
456: ((UINT64 *)hp)[2] = h3;
457: }
458:
459: #elif (UMAC_OUTPUT_LEN == 16)
460:
1.7 djm 461: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 462: /* Same as previous nh_aux, but two streams are handled in one pass,
463: * reading and writing 24 bytes of hash-state per call.
464: */
465: {
466: UINT64 h1,h2,h3,h4;
467: UWORD c = dlen / 32;
468: UINT32 *k = (UINT32 *)kp;
1.7 djm 469: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 470: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
471: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
472: k8,k9,k10,k11,k12,k13,k14,k15,
473: k16,k17,k18,k19;
474:
475: h1 = *((UINT64 *)hp);
476: h2 = *((UINT64 *)hp + 1);
477: h3 = *((UINT64 *)hp + 2);
478: h4 = *((UINT64 *)hp + 3);
479: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
480: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
481: do {
482: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
483: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
484: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
485: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
486: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
487: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
488: k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
489:
490: h1 += MUL64((k0 + d0), (k4 + d4));
491: h2 += MUL64((k4 + d0), (k8 + d4));
492: h3 += MUL64((k8 + d0), (k12 + d4));
493: h4 += MUL64((k12 + d0), (k16 + d4));
494:
495: h1 += MUL64((k1 + d1), (k5 + d5));
496: h2 += MUL64((k5 + d1), (k9 + d5));
497: h3 += MUL64((k9 + d1), (k13 + d5));
498: h4 += MUL64((k13 + d1), (k17 + d5));
499:
500: h1 += MUL64((k2 + d2), (k6 + d6));
501: h2 += MUL64((k6 + d2), (k10 + d6));
502: h3 += MUL64((k10 + d2), (k14 + d6));
503: h4 += MUL64((k14 + d2), (k18 + d6));
504:
505: h1 += MUL64((k3 + d3), (k7 + d7));
506: h2 += MUL64((k7 + d3), (k11 + d7));
507: h3 += MUL64((k11 + d3), (k15 + d7));
508: h4 += MUL64((k15 + d3), (k19 + d7));
509:
510: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
511: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
512: k8 = k16; k9 = k17; k10 = k18; k11 = k19;
513:
514: d += 8;
515: k += 8;
516: } while (--c);
517: ((UINT64 *)hp)[0] = h1;
518: ((UINT64 *)hp)[1] = h2;
519: ((UINT64 *)hp)[2] = h3;
520: ((UINT64 *)hp)[3] = h4;
521: }
522:
523: /* ---------------------------------------------------------------------- */
524: #endif /* UMAC_OUTPUT_LENGTH */
525: /* ---------------------------------------------------------------------- */
526:
527:
528: /* ---------------------------------------------------------------------- */
529:
1.7 djm 530: static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
1.1 pvalchev 531: /* This function is a wrapper for the primitive NH hash functions. It takes
532: * as argument "hc" the current hash context and a buffer which must be a
533: * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
534: * appropriately according to how much message has been hashed already.
535: */
536: {
537: UINT8 *key;
538:
539: key = hc->nh_key + hc->bytes_hashed;
540: nh_aux(key, buf, hc->state, nbytes);
541: }
542:
543: /* ---------------------------------------------------------------------- */
544:
1.12 ! markus 545: #if (__LITTLE_ENDIAN__)
1.1 pvalchev 546: static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
547: /* We endian convert the keys on little-endian computers to */
548: /* compensate for the lack of big-endian memory reads during hashing. */
549: {
550: UWORD iters = num_bytes / bpw;
551: if (bpw == 4) {
552: UINT32 *p = (UINT32 *)buf;
553: do {
554: *p = LOAD_UINT32_REVERSED(p);
555: p++;
556: } while (--iters);
557: } else if (bpw == 8) {
558: UINT32 *p = (UINT32 *)buf;
559: UINT32 t;
560: do {
561: t = LOAD_UINT32_REVERSED(p+1);
562: p[1] = LOAD_UINT32_REVERSED(p);
563: p[0] = t;
564: p += 2;
565: } while (--iters);
566: }
567: }
568: #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
569: #else
570: #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */
571: #endif
572:
573: /* ---------------------------------------------------------------------- */
574:
575: static void nh_reset(nh_ctx *hc)
576: /* Reset nh_ctx to ready for hashing of new data */
577: {
578: hc->bytes_hashed = 0;
579: hc->next_data_empty = 0;
580: hc->state[0] = 0;
581: #if (UMAC_OUTPUT_LEN >= 8)
582: hc->state[1] = 0;
583: #endif
584: #if (UMAC_OUTPUT_LEN >= 12)
585: hc->state[2] = 0;
586: #endif
587: #if (UMAC_OUTPUT_LEN == 16)
588: hc->state[3] = 0;
589: #endif
590:
591: }
592:
593: /* ---------------------------------------------------------------------- */
594:
595: static void nh_init(nh_ctx *hc, aes_int_key prf_key)
596: /* Generate nh_key, endian convert and reset to be ready for hashing. */
597: {
598: kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
599: endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
600: nh_reset(hc);
601: }
602:
603: /* ---------------------------------------------------------------------- */
604:
1.7 djm 605: static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
1.1 pvalchev 606: /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */
607: /* even multiple of HASH_BUF_BYTES. */
608: {
609: UINT32 i,j;
610:
611: j = hc->next_data_empty;
612: if ((j + nbytes) >= HASH_BUF_BYTES) {
613: if (j) {
614: i = HASH_BUF_BYTES - j;
615: memcpy(hc->data+j, buf, i);
616: nh_transform(hc,hc->data,HASH_BUF_BYTES);
617: nbytes -= i;
618: buf += i;
619: hc->bytes_hashed += HASH_BUF_BYTES;
620: }
621: if (nbytes >= HASH_BUF_BYTES) {
622: i = nbytes & ~(HASH_BUF_BYTES - 1);
623: nh_transform(hc, buf, i);
624: nbytes -= i;
625: buf += i;
626: hc->bytes_hashed += i;
627: }
628: j = 0;
629: }
630: memcpy(hc->data + j, buf, nbytes);
631: hc->next_data_empty = j + nbytes;
632: }
633:
634: /* ---------------------------------------------------------------------- */
635:
636: static void zero_pad(UINT8 *p, int nbytes)
637: {
638: /* Write "nbytes" of zeroes, beginning at "p" */
639: if (nbytes >= (int)sizeof(UWORD)) {
640: while ((ptrdiff_t)p % sizeof(UWORD)) {
641: *p = 0;
642: nbytes--;
643: p++;
644: }
645: while (nbytes >= (int)sizeof(UWORD)) {
646: *(UWORD *)p = 0;
647: nbytes -= sizeof(UWORD);
648: p += sizeof(UWORD);
649: }
650: }
651: while (nbytes) {
652: *p = 0;
653: nbytes--;
654: p++;
655: }
656: }
657:
658: /* ---------------------------------------------------------------------- */
659:
660: static void nh_final(nh_ctx *hc, UINT8 *result)
661: /* After passing some number of data buffers to nh_update() for integration
662: * into an NH context, nh_final is called to produce a hash result. If any
663: * bytes are in the buffer hc->data, incorporate them into the
664: * NH context. Finally, add into the NH accumulation "state" the total number
665: * of bits hashed. The resulting numbers are written to the buffer "result".
666: * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
667: */
668: {
669: int nh_len, nbits;
670:
671: if (hc->next_data_empty != 0) {
672: nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
673: ~(L1_PAD_BOUNDARY - 1));
674: zero_pad(hc->data + hc->next_data_empty,
675: nh_len - hc->next_data_empty);
676: nh_transform(hc, hc->data, nh_len);
677: hc->bytes_hashed += hc->next_data_empty;
678: } else if (hc->bytes_hashed == 0) {
679: nh_len = L1_PAD_BOUNDARY;
680: zero_pad(hc->data, L1_PAD_BOUNDARY);
681: nh_transform(hc, hc->data, nh_len);
682: }
683:
684: nbits = (hc->bytes_hashed << 3);
685: ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
686: #if (UMAC_OUTPUT_LEN >= 8)
687: ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
688: #endif
689: #if (UMAC_OUTPUT_LEN >= 12)
690: ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
691: #endif
692: #if (UMAC_OUTPUT_LEN == 16)
693: ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
694: #endif
695: nh_reset(hc);
696: }
697:
698: /* ---------------------------------------------------------------------- */
699:
1.7 djm 700: static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
1.1 pvalchev 701: UINT32 unpadded_len, UINT8 *result)
702: /* All-in-one nh_update() and nh_final() equivalent.
703: * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
704: * well aligned
705: */
706: {
707: UINT32 nbits;
708:
709: /* Initialize the hash state */
710: nbits = (unpadded_len << 3);
711:
712: ((UINT64 *)result)[0] = nbits;
713: #if (UMAC_OUTPUT_LEN >= 8)
714: ((UINT64 *)result)[1] = nbits;
715: #endif
716: #if (UMAC_OUTPUT_LEN >= 12)
717: ((UINT64 *)result)[2] = nbits;
718: #endif
719: #if (UMAC_OUTPUT_LEN == 16)
720: ((UINT64 *)result)[3] = nbits;
721: #endif
722:
723: nh_aux(hc->nh_key, buf, result, padded_len);
724: }
725:
726: /* ---------------------------------------------------------------------- */
727: /* ---------------------------------------------------------------------- */
728: /* ----- Begin UHASH Section -------------------------------------------- */
729: /* ---------------------------------------------------------------------- */
730: /* ---------------------------------------------------------------------- */
731:
732: /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
733: * hashed by NH. The NH output is then hashed by a polynomial-hash layer
734: * unless the initial data to be hashed is short. After the polynomial-
735: * layer, an inner-product hash is used to produce the final UHASH output.
736: *
737: * UHASH provides two interfaces, one all-at-once and another where data
738: * buffers are presented sequentially. In the sequential interface, the
739: * UHASH client calls the routine uhash_update() as many times as necessary.
740: * When there is no more data to be fed to UHASH, the client calls
741: * uhash_final() which
742: * calculates the UHASH output. Before beginning another UHASH calculation
743: * the uhash_reset() routine must be called. The all-at-once UHASH routine,
744: * uhash(), is equivalent to the sequence of calls uhash_update() and
745: * uhash_final(); however it is optimized and should be
746: * used whenever the sequential interface is not necessary.
747: *
748: * The routine uhash_init() initializes the uhash_ctx data structure and
749: * must be called once, before any other UHASH routine.
750: */
751:
752: /* ---------------------------------------------------------------------- */
753: /* ----- Constants and uhash_ctx ---------------------------------------- */
754: /* ---------------------------------------------------------------------- */
755:
756: /* ---------------------------------------------------------------------- */
757: /* ----- Poly hash and Inner-Product hash Constants --------------------- */
758: /* ---------------------------------------------------------------------- */
759:
760: /* Primes and masks */
761: #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */
762: #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */
763: #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */
764:
765:
766: /* ---------------------------------------------------------------------- */
767:
768: typedef struct uhash_ctx {
769: nh_ctx hash; /* Hash context for L1 NH hash */
770: UINT64 poly_key_8[STREAMS]; /* p64 poly keys */
771: UINT64 poly_accum[STREAMS]; /* poly hash result */
772: UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */
773: UINT32 ip_trans[STREAMS]; /* Inner-product translation */
774: UINT32 msg_len; /* Total length of data passed */
775: /* to uhash */
776: } uhash_ctx;
777: typedef struct uhash_ctx *uhash_ctx_t;
778:
779: /* ---------------------------------------------------------------------- */
780:
781:
782: /* The polynomial hashes use Horner's rule to evaluate a polynomial one
783: * word at a time. As described in the specification, poly32 and poly64
784: * require keys from special domains. The following implementations exploit
785: * the special domains to avoid overflow. The results are not guaranteed to
786: * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
787: * patches any errant values.
788: */
789:
790: static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
791: {
792: UINT32 key_hi = (UINT32)(key >> 32),
793: key_lo = (UINT32)key,
794: cur_hi = (UINT32)(cur >> 32),
795: cur_lo = (UINT32)cur,
796: x_lo,
797: x_hi;
798: UINT64 X,T,res;
799:
800: X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
801: x_lo = (UINT32)X;
802: x_hi = (UINT32)(X >> 32);
803:
804: res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
805:
806: T = ((UINT64)x_lo << 32);
807: res += T;
808: if (res < T)
809: res += 59;
810:
811: res += data;
812: if (res < data)
813: res += 59;
814:
815: return res;
816: }
817:
818:
819: /* Although UMAC is specified to use a ramped polynomial hash scheme, this
820: * implementation does not handle all ramp levels. Because we don't handle
821: * the ramp up to p128 modulus in this implementation, we are limited to
822: * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
823: * bytes input to UMAC per tag, ie. 16MB).
824: */
825: static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
826: {
827: int i;
828: UINT64 *data=(UINT64*)data_in;
829:
830: for (i = 0; i < STREAMS; i++) {
831: if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
832: hc->poly_accum[i] = poly64(hc->poly_accum[i],
833: hc->poly_key_8[i], p64 - 1);
834: hc->poly_accum[i] = poly64(hc->poly_accum[i],
835: hc->poly_key_8[i], (data[i] - 59));
836: } else {
837: hc->poly_accum[i] = poly64(hc->poly_accum[i],
838: hc->poly_key_8[i], data[i]);
839: }
840: }
841: }
842:
843:
844: /* ---------------------------------------------------------------------- */
845:
846:
847: /* The final step in UHASH is an inner-product hash. The poly hash
848: * produces a result not neccesarily WORD_LEN bytes long. The inner-
849: * product hash breaks the polyhash output into 16-bit chunks and
850: * multiplies each with a 36 bit key.
851: */
852:
853: static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
854: {
855: t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
856: t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
857: t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
858: t = t + ipkp[3] * (UINT64)(UINT16)(data);
859:
860: return t;
861: }
862:
863: static UINT32 ip_reduce_p36(UINT64 t)
864: {
865: /* Divisionless modular reduction */
866: UINT64 ret;
867:
868: ret = (t & m36) + 5 * (t >> 36);
869: if (ret >= p36)
870: ret -= p36;
871:
872: /* return least significant 32 bits */
873: return (UINT32)(ret);
874: }
875:
876:
877: /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
878: * the polyhash stage is skipped and ip_short is applied directly to the
879: * NH output.
880: */
881: static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
882: {
883: UINT64 t;
884: UINT64 *nhp = (UINT64 *)nh_res;
885:
886: t = ip_aux(0,ahc->ip_keys, nhp[0]);
887: STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
888: #if (UMAC_OUTPUT_LEN >= 8)
889: t = ip_aux(0,ahc->ip_keys+4, nhp[1]);
890: STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
891: #endif
892: #if (UMAC_OUTPUT_LEN >= 12)
893: t = ip_aux(0,ahc->ip_keys+8, nhp[2]);
894: STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
895: #endif
896: #if (UMAC_OUTPUT_LEN == 16)
897: t = ip_aux(0,ahc->ip_keys+12, nhp[3]);
898: STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
899: #endif
900: }
901:
902: /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
903: * the polyhash stage is not skipped and ip_long is applied to the
904: * polyhash output.
905: */
906: static void ip_long(uhash_ctx_t ahc, u_char *res)
907: {
908: int i;
909: UINT64 t;
910:
911: for (i = 0; i < STREAMS; i++) {
912: /* fix polyhash output not in Z_p64 */
913: if (ahc->poly_accum[i] >= p64)
914: ahc->poly_accum[i] -= p64;
915: t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
916: STORE_UINT32_BIG((UINT32 *)res+i,
917: ip_reduce_p36(t) ^ ahc->ip_trans[i]);
918: }
919: }
920:
921:
922: /* ---------------------------------------------------------------------- */
923:
924: /* ---------------------------------------------------------------------- */
925:
926: /* Reset uhash context for next hash session */
927: static int uhash_reset(uhash_ctx_t pc)
928: {
929: nh_reset(&pc->hash);
930: pc->msg_len = 0;
931: pc->poly_accum[0] = 1;
932: #if (UMAC_OUTPUT_LEN >= 8)
933: pc->poly_accum[1] = 1;
934: #endif
935: #if (UMAC_OUTPUT_LEN >= 12)
936: pc->poly_accum[2] = 1;
937: #endif
938: #if (UMAC_OUTPUT_LEN == 16)
939: pc->poly_accum[3] = 1;
940: #endif
941: return 1;
942: }
943:
944: /* ---------------------------------------------------------------------- */
945:
946: /* Given a pointer to the internal key needed by kdf() and a uhash context,
947: * initialize the NH context and generate keys needed for poly and inner-
948: * product hashing. All keys are endian adjusted in memory so that native
949: * loads cause correct keys to be in registers during calculation.
950: */
951: static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
952: {
953: int i;
954: UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
955:
956: /* Zero the entire uhash context */
957: memset(ahc, 0, sizeof(uhash_ctx));
958:
959: /* Initialize the L1 hash */
960: nh_init(&ahc->hash, prf_key);
961:
962: /* Setup L2 hash variables */
963: kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */
964: for (i = 0; i < STREAMS; i++) {
965: /* Fill keys from the buffer, skipping bytes in the buffer not
966: * used by this implementation. Endian reverse the keys if on a
967: * little-endian computer.
968: */
969: memcpy(ahc->poly_key_8+i, buf+24*i, 8);
970: endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
971: /* Mask the 64-bit keys to their special domain */
972: ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
973: ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */
974: }
975:
976: /* Setup L3-1 hash variables */
977: kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
978: for (i = 0; i < STREAMS; i++)
979: memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
980: 4*sizeof(UINT64));
981: endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
982: sizeof(ahc->ip_keys));
983: for (i = 0; i < STREAMS*4; i++)
984: ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */
985:
986: /* Setup L3-2 hash variables */
987: /* Fill buffer with index 4 key */
988: kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
989: endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
990: STREAMS * sizeof(UINT32));
1.12 ! markus 991: explicit_bzero(buf, sizeof(buf));
1.1 pvalchev 992: }
993:
994: /* ---------------------------------------------------------------------- */
995:
996: #if 0
997: static uhash_ctx_t uhash_alloc(u_char key[])
998: {
999: /* Allocate memory and force to a 16-byte boundary. */
1000: uhash_ctx_t ctx;
1001: u_char bytes_to_add;
1002: aes_int_key prf_key;
1003:
1004: ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1005: if (ctx) {
1006: if (ALLOC_BOUNDARY) {
1007: bytes_to_add = ALLOC_BOUNDARY -
1008: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1009: ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1010: *((u_char *)ctx - 1) = bytes_to_add;
1011: }
1012: aes_key_setup(key,prf_key);
1013: uhash_init(ctx, prf_key);
1014: }
1015: return (ctx);
1016: }
1017: #endif
1018:
1019: /* ---------------------------------------------------------------------- */
1020:
1021: #if 0
1022: static int uhash_free(uhash_ctx_t ctx)
1023: {
1024: /* Free memory allocated by uhash_alloc */
1025: u_char bytes_to_sub;
1026:
1027: if (ctx) {
1028: if (ALLOC_BOUNDARY) {
1029: bytes_to_sub = *((u_char *)ctx - 1);
1030: ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1031: }
1032: free(ctx);
1033: }
1034: return (1);
1035: }
1036: #endif
1037: /* ---------------------------------------------------------------------- */
1038:
1.7 djm 1039: static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1.1 pvalchev 1040: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1041: * hash each one with NH, calling the polyhash on each NH output.
1042: */
1043: {
1044: UWORD bytes_hashed, bytes_remaining;
1.3 pvalchev 1045: UINT64 result_buf[STREAMS];
1046: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1047:
1048: if (ctx->msg_len + len <= L1_KEY_LEN) {
1.7 djm 1049: nh_update(&ctx->hash, (const UINT8 *)input, len);
1.1 pvalchev 1050: ctx->msg_len += len;
1051: } else {
1052:
1053: bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1054: if (ctx->msg_len == L1_KEY_LEN)
1055: bytes_hashed = L1_KEY_LEN;
1056:
1057: if (bytes_hashed + len >= L1_KEY_LEN) {
1058:
1059: /* If some bytes have been passed to the hash function */
1060: /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1061: /* bytes to complete the current nh_block. */
1062: if (bytes_hashed) {
1063: bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1.7 djm 1064: nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1.1 pvalchev 1065: nh_final(&ctx->hash, nh_result);
1066: ctx->msg_len += bytes_remaining;
1067: poly_hash(ctx,(UINT32 *)nh_result);
1068: len -= bytes_remaining;
1069: input += bytes_remaining;
1070: }
1071:
1072: /* Hash directly from input stream if enough bytes */
1073: while (len >= L1_KEY_LEN) {
1.7 djm 1074: nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1.1 pvalchev 1075: L1_KEY_LEN, nh_result);
1076: ctx->msg_len += L1_KEY_LEN;
1077: len -= L1_KEY_LEN;
1078: input += L1_KEY_LEN;
1079: poly_hash(ctx,(UINT32 *)nh_result);
1080: }
1081: }
1082:
1083: /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1084: if (len) {
1.7 djm 1085: nh_update(&ctx->hash, (const UINT8 *)input, len);
1.1 pvalchev 1086: ctx->msg_len += len;
1087: }
1088: }
1089:
1090: return (1);
1091: }
1092:
1093: /* ---------------------------------------------------------------------- */
1094:
1095: static int uhash_final(uhash_ctx_t ctx, u_char *res)
1096: /* Incorporate any pending data, pad, and generate tag */
1097: {
1.3 pvalchev 1098: UINT64 result_buf[STREAMS];
1099: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1100:
1101: if (ctx->msg_len > L1_KEY_LEN) {
1102: if (ctx->msg_len % L1_KEY_LEN) {
1103: nh_final(&ctx->hash, nh_result);
1104: poly_hash(ctx,(UINT32 *)nh_result);
1105: }
1106: ip_long(ctx, res);
1107: } else {
1108: nh_final(&ctx->hash, nh_result);
1109: ip_short(ctx,nh_result, res);
1110: }
1111: uhash_reset(ctx);
1112: return (1);
1113: }
1114:
1115: /* ---------------------------------------------------------------------- */
1116:
1117: #if 0
1118: static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1119: /* assumes that msg is in a writable buffer of length divisible by */
1120: /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */
1121: {
1122: UINT8 nh_result[STREAMS*sizeof(UINT64)];
1123: UINT32 nh_len;
1124: int extra_zeroes_needed;
1125:
1126: /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1127: * the polyhash.
1128: */
1129: if (len <= L1_KEY_LEN) {
1130: if (len == 0) /* If zero length messages will not */
1131: nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */
1132: else
1133: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1134: extra_zeroes_needed = nh_len - len;
1135: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1136: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1137: ip_short(ahc,nh_result, res);
1138: } else {
1139: /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1140: * output to poly_hash().
1141: */
1142: do {
1143: nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1144: poly_hash(ahc,(UINT32 *)nh_result);
1145: len -= L1_KEY_LEN;
1146: msg += L1_KEY_LEN;
1147: } while (len >= L1_KEY_LEN);
1148: if (len) {
1149: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1150: extra_zeroes_needed = nh_len - len;
1151: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1152: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1153: poly_hash(ahc,(UINT32 *)nh_result);
1154: }
1155:
1156: ip_long(ahc, res);
1157: }
1158:
1159: uhash_reset(ahc);
1160: return 1;
1161: }
1162: #endif
1163:
1164: /* ---------------------------------------------------------------------- */
1165: /* ---------------------------------------------------------------------- */
1166: /* ----- Begin UMAC Section --------------------------------------------- */
1167: /* ---------------------------------------------------------------------- */
1168: /* ---------------------------------------------------------------------- */
1169:
1170: /* The UMAC interface has two interfaces, an all-at-once interface where
1171: * the entire message to be authenticated is passed to UMAC in one buffer,
1172: * and a sequential interface where the message is presented a little at a
1173: * time. The all-at-once is more optimaized than the sequential version and
1174: * should be preferred when the sequential interface is not required.
1175: */
1176: struct umac_ctx {
1177: uhash_ctx hash; /* Hash function for message compression */
1178: pdf_ctx pdf; /* PDF for hashed output */
1179: void *free_ptr; /* Address to free this struct via */
1180: } umac_ctx;
1181:
1182: /* ---------------------------------------------------------------------- */
1183:
1184: #if 0
1185: int umac_reset(struct umac_ctx *ctx)
1186: /* Reset the hash function to begin a new authentication. */
1187: {
1188: uhash_reset(&ctx->hash);
1189: return (1);
1190: }
1191: #endif
1192:
1193: /* ---------------------------------------------------------------------- */
1194:
1195: int umac_delete(struct umac_ctx *ctx)
1196: /* Deallocate the ctx structure */
1197: {
1198: if (ctx) {
1199: if (ALLOC_BOUNDARY)
1200: ctx = (struct umac_ctx *)ctx->free_ptr;
1.12 ! markus 1201: explicit_bzero(ctx, sizeof(*ctx) + ALLOC_BOUNDARY);
1.5 djm 1202: free(ctx);
1.1 pvalchev 1203: }
1204: return (1);
1205: }
1206:
1207: /* ---------------------------------------------------------------------- */
1208:
1.7 djm 1209: struct umac_ctx *umac_new(const u_char key[])
1.1 pvalchev 1210: /* Dynamically allocate a umac_ctx struct, initialize variables,
1211: * generate subkeys from key. Align to 16-byte boundary.
1212: */
1213: {
1214: struct umac_ctx *ctx, *octx;
1215: size_t bytes_to_add;
1216: aes_int_key prf_key;
1217:
1.8 djm 1218: octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1.1 pvalchev 1219: if (ctx) {
1220: if (ALLOC_BOUNDARY) {
1221: bytes_to_add = ALLOC_BOUNDARY -
1222: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1223: ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1224: }
1225: ctx->free_ptr = octx;
1.7 djm 1226: aes_key_setup(key, prf_key);
1.1 pvalchev 1227: pdf_init(&ctx->pdf, prf_key);
1228: uhash_init(&ctx->hash, prf_key);
1.12 ! markus 1229: explicit_bzero(prf_key, sizeof(prf_key));
1.1 pvalchev 1230: }
1231:
1232: return (ctx);
1233: }
1234:
1235: /* ---------------------------------------------------------------------- */
1236:
1.7 djm 1237: int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1.1 pvalchev 1238: /* Incorporate any pending data, pad, and generate tag */
1239: {
1240: uhash_final(&ctx->hash, (u_char *)tag);
1.7 djm 1241: pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1.1 pvalchev 1242:
1243: return (1);
1244: }
1245:
1246: /* ---------------------------------------------------------------------- */
1247:
1.7 djm 1248: int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1.1 pvalchev 1249: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */
1250: /* hash each one, calling the PDF on the hashed output whenever the hash- */
1251: /* output buffer is full. */
1252: {
1253: uhash_update(&ctx->hash, input, len);
1254: return (1);
1255: }
1256:
1257: /* ---------------------------------------------------------------------- */
1258:
1259: #if 0
1260: int umac(struct umac_ctx *ctx, u_char *input,
1261: long len, u_char tag[],
1262: u_char nonce[8])
1263: /* All-in-one version simply calls umac_update() and umac_final(). */
1264: {
1265: uhash(&ctx->hash, input, len, (u_char *)tag);
1266: pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1267:
1268: return (1);
1269: }
1270: #endif
1271:
1272: /* ---------------------------------------------------------------------- */
1273: /* ---------------------------------------------------------------------- */
1274: /* ----- End UMAC Section ----------------------------------------------- */
1275: /* ---------------------------------------------------------------------- */
1276: /* ---------------------------------------------------------------------- */