Annotation of src/usr.bin/ssh/umac.c, Revision 1.9
1.9 ! djm 1: /* $OpenBSD: umac.c,v 1.8 2013/11/08 00:39:15 djm 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>
67: #include <sys/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:
154: /* OpenSSL's AES */
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)
1.1 pvalchev 161:
162: /* The user-supplied UMAC key is stretched using AES in a counter
163: * mode to supply all random bits needed by UMAC. The kdf function takes
164: * an AES internal key representation 'key' and writes a stream of
165: * 'nbytes' bytes to the memory pointed at by 'buffer_ptr'. Each distinct
166: * 'ndx' causes a distinct byte stream.
167: */
168: static void kdf(void *buffer_ptr, aes_int_key key, UINT8 ndx, int nbytes)
169: {
170: UINT8 in_buf[AES_BLOCK_LEN] = {0};
171: UINT8 out_buf[AES_BLOCK_LEN];
172: UINT8 *dst_buf = (UINT8 *)buffer_ptr;
173: int i;
174:
175: /* Setup the initial value */
176: in_buf[AES_BLOCK_LEN-9] = ndx;
177: in_buf[AES_BLOCK_LEN-1] = i = 1;
178:
179: while (nbytes >= AES_BLOCK_LEN) {
180: aes_encryption(in_buf, out_buf, key);
181: memcpy(dst_buf,out_buf,AES_BLOCK_LEN);
182: in_buf[AES_BLOCK_LEN-1] = ++i;
183: nbytes -= AES_BLOCK_LEN;
184: dst_buf += AES_BLOCK_LEN;
185: }
186: if (nbytes) {
187: aes_encryption(in_buf, out_buf, key);
188: memcpy(dst_buf,out_buf,nbytes);
189: }
190: }
191:
192: /* The final UHASH result is XOR'd with the output of a pseudorandom
193: * function. Here, we use AES to generate random output and
194: * xor the appropriate bytes depending on the last bits of nonce.
195: * This scheme is optimized for sequential, increasing big-endian nonces.
196: */
197:
198: typedef struct {
199: UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */
200: UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */
201: aes_int_key prf_key; /* Expanded AES key for PDF */
202: } pdf_ctx;
203:
204: static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
205: {
206: UINT8 buf[UMAC_KEY_LEN];
207:
208: kdf(buf, prf_key, 0, UMAC_KEY_LEN);
209: aes_key_setup(buf, pc->prf_key);
210:
211: /* Initialize pdf and cache */
212: memset(pc->nonce, 0, sizeof(pc->nonce));
213: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
214: }
215:
1.7 djm 216: static void pdf_gen_xor(pdf_ctx *pc, const UINT8 nonce[8], UINT8 buf[8])
1.1 pvalchev 217: {
218: /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
219: * of the AES output. If last time around we returned the ndx-1st
220: * element, then we may have the result in the cache already.
221: */
222:
223: #if (UMAC_OUTPUT_LEN == 4)
224: #define LOW_BIT_MASK 3
225: #elif (UMAC_OUTPUT_LEN == 8)
226: #define LOW_BIT_MASK 1
227: #elif (UMAC_OUTPUT_LEN > 8)
228: #define LOW_BIT_MASK 0
229: #endif
1.6 djm 230: union {
231: UINT8 tmp_nonce_lo[4];
232: UINT32 align;
233: } t;
1.1 pvalchev 234: #if LOW_BIT_MASK != 0
235: int ndx = nonce[7] & LOW_BIT_MASK;
236: #endif
1.7 djm 237: *(UINT32 *)t.tmp_nonce_lo = ((const UINT32 *)nonce)[1];
1.6 djm 238: t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
1.1 pvalchev 239:
1.6 djm 240: if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
1.7 djm 241: (((const UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
1.1 pvalchev 242: {
1.7 djm 243: ((UINT32 *)pc->nonce)[0] = ((const UINT32 *)nonce)[0];
1.6 djm 244: ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
1.1 pvalchev 245: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
246: }
247:
248: #if (UMAC_OUTPUT_LEN == 4)
249: *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
250: #elif (UMAC_OUTPUT_LEN == 8)
251: *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
252: #elif (UMAC_OUTPUT_LEN == 12)
253: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
254: ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
255: #elif (UMAC_OUTPUT_LEN == 16)
256: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
257: ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
258: #endif
259: }
260:
261: /* ---------------------------------------------------------------------- */
262: /* ---------------------------------------------------------------------- */
263: /* ----- Begin NH Hash Section ------------------------------------------ */
264: /* ---------------------------------------------------------------------- */
265: /* ---------------------------------------------------------------------- */
266:
267: /* The NH-based hash functions used in UMAC are described in the UMAC paper
268: * and specification, both of which can be found at the UMAC website.
269: * The interface to this implementation has two
270: * versions, one expects the entire message being hashed to be passed
271: * in a single buffer and returns the hash result immediately. The second
272: * allows the message to be passed in a sequence of buffers. In the
273: * muliple-buffer interface, the client calls the routine nh_update() as
274: * many times as necessary. When there is no more data to be fed to the
275: * hash, the client calls nh_final() which calculates the hash output.
276: * Before beginning another hash calculation the nh_reset() routine
277: * must be called. The single-buffer routine, nh(), is equivalent to
278: * the sequence of calls nh_update() and nh_final(); however it is
279: * optimized and should be prefered whenever the multiple-buffer interface
280: * is not necessary. When using either interface, it is the client's
281: * responsability to pass no more than L1_KEY_LEN bytes per hash result.
282: *
283: * The routine nh_init() initializes the nh_ctx data structure and
284: * must be called once, before any other PDF routine.
285: */
286:
287: /* The "nh_aux" routines do the actual NH hashing work. They
288: * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
289: * produce output for all STREAMS NH iterations in one call,
290: * allowing the parallel implementation of the streams.
291: */
292:
293: #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */
294: #define L1_KEY_LEN 1024 /* Internal key bytes */
295: #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */
296: #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */
297: #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */
298: #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */
299:
300: typedef struct {
301: UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
1.4 djm 302: UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */
1.1 pvalchev 303: int next_data_empty; /* Bookeeping variable for data buffer. */
304: int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */
305: UINT64 state[STREAMS]; /* on-line state */
306: } nh_ctx;
307:
308:
309: #if (UMAC_OUTPUT_LEN == 4)
310:
1.7 djm 311: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 312: /* NH hashing primitive. Previous (partial) hash result is loaded and
313: * then stored via hp pointer. The length of the data pointed at by "dp",
314: * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key
315: * is expected to be endian compensated in memory at key setup.
316: */
317: {
318: UINT64 h;
319: UWORD c = dlen / 32;
320: UINT32 *k = (UINT32 *)kp;
1.7 djm 321: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 322: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
323: UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
324:
325: h = *((UINT64 *)hp);
326: do {
327: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
328: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
329: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
330: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
331: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
332: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
333: h += MUL64((k0 + d0), (k4 + d4));
334: h += MUL64((k1 + d1), (k5 + d5));
335: h += MUL64((k2 + d2), (k6 + d6));
336: h += MUL64((k3 + d3), (k7 + d7));
337:
338: d += 8;
339: k += 8;
340: } while (--c);
341: *((UINT64 *)hp) = h;
342: }
343:
344: #elif (UMAC_OUTPUT_LEN == 8)
345:
1.7 djm 346: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 347: /* Same as previous nh_aux, but two streams are handled in one pass,
348: * reading and writing 16 bytes of hash-state per call.
349: */
350: {
351: UINT64 h1,h2;
352: UWORD c = dlen / 32;
353: UINT32 *k = (UINT32 *)kp;
1.7 djm 354: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 355: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
356: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
357: k8,k9,k10,k11;
358:
359: h1 = *((UINT64 *)hp);
360: h2 = *((UINT64 *)hp + 1);
361: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
362: do {
363: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
364: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
365: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
366: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
367: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
368: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
369:
370: h1 += MUL64((k0 + d0), (k4 + d4));
371: h2 += MUL64((k4 + d0), (k8 + d4));
372:
373: h1 += MUL64((k1 + d1), (k5 + d5));
374: h2 += MUL64((k5 + d1), (k9 + d5));
375:
376: h1 += MUL64((k2 + d2), (k6 + d6));
377: h2 += MUL64((k6 + d2), (k10 + d6));
378:
379: h1 += MUL64((k3 + d3), (k7 + d7));
380: h2 += MUL64((k7 + d3), (k11 + d7));
381:
382: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
383:
384: d += 8;
385: k += 8;
386: } while (--c);
387: ((UINT64 *)hp)[0] = h1;
388: ((UINT64 *)hp)[1] = h2;
389: }
390:
391: #elif (UMAC_OUTPUT_LEN == 12)
392:
1.7 djm 393: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 394: /* Same as previous nh_aux, but two streams are handled in one pass,
395: * reading and writing 24 bytes of hash-state per call.
396: */
397: {
398: UINT64 h1,h2,h3;
399: UWORD c = dlen / 32;
400: UINT32 *k = (UINT32 *)kp;
1.7 djm 401: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 402: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
403: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
404: k8,k9,k10,k11,k12,k13,k14,k15;
405:
406: h1 = *((UINT64 *)hp);
407: h2 = *((UINT64 *)hp + 1);
408: h3 = *((UINT64 *)hp + 2);
409: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
410: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
411: do {
412: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
413: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
414: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
415: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
416: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
417: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
418:
419: h1 += MUL64((k0 + d0), (k4 + d4));
420: h2 += MUL64((k4 + d0), (k8 + d4));
421: h3 += MUL64((k8 + d0), (k12 + d4));
422:
423: h1 += MUL64((k1 + d1), (k5 + d5));
424: h2 += MUL64((k5 + d1), (k9 + d5));
425: h3 += MUL64((k9 + d1), (k13 + d5));
426:
427: h1 += MUL64((k2 + d2), (k6 + d6));
428: h2 += MUL64((k6 + d2), (k10 + d6));
429: h3 += MUL64((k10 + d2), (k14 + d6));
430:
431: h1 += MUL64((k3 + d3), (k7 + d7));
432: h2 += MUL64((k7 + d3), (k11 + d7));
433: h3 += MUL64((k11 + d3), (k15 + d7));
434:
435: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
436: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
437:
438: d += 8;
439: k += 8;
440: } while (--c);
441: ((UINT64 *)hp)[0] = h1;
442: ((UINT64 *)hp)[1] = h2;
443: ((UINT64 *)hp)[2] = h3;
444: }
445:
446: #elif (UMAC_OUTPUT_LEN == 16)
447:
1.7 djm 448: static void nh_aux(void *kp, const void *dp, void *hp, UINT32 dlen)
1.1 pvalchev 449: /* Same as previous nh_aux, but two streams are handled in one pass,
450: * reading and writing 24 bytes of hash-state per call.
451: */
452: {
453: UINT64 h1,h2,h3,h4;
454: UWORD c = dlen / 32;
455: UINT32 *k = (UINT32 *)kp;
1.7 djm 456: const UINT32 *d = (const UINT32 *)dp;
1.1 pvalchev 457: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
458: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
459: k8,k9,k10,k11,k12,k13,k14,k15,
460: k16,k17,k18,k19;
461:
462: h1 = *((UINT64 *)hp);
463: h2 = *((UINT64 *)hp + 1);
464: h3 = *((UINT64 *)hp + 2);
465: h4 = *((UINT64 *)hp + 3);
466: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
467: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
468: do {
469: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
470: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
471: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
472: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
473: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
474: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
475: k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
476:
477: h1 += MUL64((k0 + d0), (k4 + d4));
478: h2 += MUL64((k4 + d0), (k8 + d4));
479: h3 += MUL64((k8 + d0), (k12 + d4));
480: h4 += MUL64((k12 + d0), (k16 + d4));
481:
482: h1 += MUL64((k1 + d1), (k5 + d5));
483: h2 += MUL64((k5 + d1), (k9 + d5));
484: h3 += MUL64((k9 + d1), (k13 + d5));
485: h4 += MUL64((k13 + d1), (k17 + d5));
486:
487: h1 += MUL64((k2 + d2), (k6 + d6));
488: h2 += MUL64((k6 + d2), (k10 + d6));
489: h3 += MUL64((k10 + d2), (k14 + d6));
490: h4 += MUL64((k14 + d2), (k18 + d6));
491:
492: h1 += MUL64((k3 + d3), (k7 + d7));
493: h2 += MUL64((k7 + d3), (k11 + d7));
494: h3 += MUL64((k11 + d3), (k15 + d7));
495: h4 += MUL64((k15 + d3), (k19 + d7));
496:
497: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
498: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
499: k8 = k16; k9 = k17; k10 = k18; k11 = k19;
500:
501: d += 8;
502: k += 8;
503: } while (--c);
504: ((UINT64 *)hp)[0] = h1;
505: ((UINT64 *)hp)[1] = h2;
506: ((UINT64 *)hp)[2] = h3;
507: ((UINT64 *)hp)[3] = h4;
508: }
509:
510: /* ---------------------------------------------------------------------- */
511: #endif /* UMAC_OUTPUT_LENGTH */
512: /* ---------------------------------------------------------------------- */
513:
514:
515: /* ---------------------------------------------------------------------- */
516:
1.7 djm 517: static void nh_transform(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
1.1 pvalchev 518: /* This function is a wrapper for the primitive NH hash functions. It takes
519: * as argument "hc" the current hash context and a buffer which must be a
520: * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
521: * appropriately according to how much message has been hashed already.
522: */
523: {
524: UINT8 *key;
525:
526: key = hc->nh_key + hc->bytes_hashed;
527: nh_aux(key, buf, hc->state, nbytes);
528: }
529:
530: /* ---------------------------------------------------------------------- */
531:
532: static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
533: /* We endian convert the keys on little-endian computers to */
534: /* compensate for the lack of big-endian memory reads during hashing. */
535: {
536: UWORD iters = num_bytes / bpw;
537: if (bpw == 4) {
538: UINT32 *p = (UINT32 *)buf;
539: do {
540: *p = LOAD_UINT32_REVERSED(p);
541: p++;
542: } while (--iters);
543: } else if (bpw == 8) {
544: UINT32 *p = (UINT32 *)buf;
545: UINT32 t;
546: do {
547: t = LOAD_UINT32_REVERSED(p+1);
548: p[1] = LOAD_UINT32_REVERSED(p);
549: p[0] = t;
550: p += 2;
551: } while (--iters);
552: }
553: }
554: #if (__LITTLE_ENDIAN__)
555: #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
556: #else
557: #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */
558: #endif
559:
560: /* ---------------------------------------------------------------------- */
561:
562: static void nh_reset(nh_ctx *hc)
563: /* Reset nh_ctx to ready for hashing of new data */
564: {
565: hc->bytes_hashed = 0;
566: hc->next_data_empty = 0;
567: hc->state[0] = 0;
568: #if (UMAC_OUTPUT_LEN >= 8)
569: hc->state[1] = 0;
570: #endif
571: #if (UMAC_OUTPUT_LEN >= 12)
572: hc->state[2] = 0;
573: #endif
574: #if (UMAC_OUTPUT_LEN == 16)
575: hc->state[3] = 0;
576: #endif
577:
578: }
579:
580: /* ---------------------------------------------------------------------- */
581:
582: static void nh_init(nh_ctx *hc, aes_int_key prf_key)
583: /* Generate nh_key, endian convert and reset to be ready for hashing. */
584: {
585: kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
586: endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
587: nh_reset(hc);
588: }
589:
590: /* ---------------------------------------------------------------------- */
591:
1.7 djm 592: static void nh_update(nh_ctx *hc, const UINT8 *buf, UINT32 nbytes)
1.1 pvalchev 593: /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */
594: /* even multiple of HASH_BUF_BYTES. */
595: {
596: UINT32 i,j;
597:
598: j = hc->next_data_empty;
599: if ((j + nbytes) >= HASH_BUF_BYTES) {
600: if (j) {
601: i = HASH_BUF_BYTES - j;
602: memcpy(hc->data+j, buf, i);
603: nh_transform(hc,hc->data,HASH_BUF_BYTES);
604: nbytes -= i;
605: buf += i;
606: hc->bytes_hashed += HASH_BUF_BYTES;
607: }
608: if (nbytes >= HASH_BUF_BYTES) {
609: i = nbytes & ~(HASH_BUF_BYTES - 1);
610: nh_transform(hc, buf, i);
611: nbytes -= i;
612: buf += i;
613: hc->bytes_hashed += i;
614: }
615: j = 0;
616: }
617: memcpy(hc->data + j, buf, nbytes);
618: hc->next_data_empty = j + nbytes;
619: }
620:
621: /* ---------------------------------------------------------------------- */
622:
623: static void zero_pad(UINT8 *p, int nbytes)
624: {
625: /* Write "nbytes" of zeroes, beginning at "p" */
626: if (nbytes >= (int)sizeof(UWORD)) {
627: while ((ptrdiff_t)p % sizeof(UWORD)) {
628: *p = 0;
629: nbytes--;
630: p++;
631: }
632: while (nbytes >= (int)sizeof(UWORD)) {
633: *(UWORD *)p = 0;
634: nbytes -= sizeof(UWORD);
635: p += sizeof(UWORD);
636: }
637: }
638: while (nbytes) {
639: *p = 0;
640: nbytes--;
641: p++;
642: }
643: }
644:
645: /* ---------------------------------------------------------------------- */
646:
647: static void nh_final(nh_ctx *hc, UINT8 *result)
648: /* After passing some number of data buffers to nh_update() for integration
649: * into an NH context, nh_final is called to produce a hash result. If any
650: * bytes are in the buffer hc->data, incorporate them into the
651: * NH context. Finally, add into the NH accumulation "state" the total number
652: * of bits hashed. The resulting numbers are written to the buffer "result".
653: * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
654: */
655: {
656: int nh_len, nbits;
657:
658: if (hc->next_data_empty != 0) {
659: nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
660: ~(L1_PAD_BOUNDARY - 1));
661: zero_pad(hc->data + hc->next_data_empty,
662: nh_len - hc->next_data_empty);
663: nh_transform(hc, hc->data, nh_len);
664: hc->bytes_hashed += hc->next_data_empty;
665: } else if (hc->bytes_hashed == 0) {
666: nh_len = L1_PAD_BOUNDARY;
667: zero_pad(hc->data, L1_PAD_BOUNDARY);
668: nh_transform(hc, hc->data, nh_len);
669: }
670:
671: nbits = (hc->bytes_hashed << 3);
672: ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
673: #if (UMAC_OUTPUT_LEN >= 8)
674: ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
675: #endif
676: #if (UMAC_OUTPUT_LEN >= 12)
677: ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
678: #endif
679: #if (UMAC_OUTPUT_LEN == 16)
680: ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
681: #endif
682: nh_reset(hc);
683: }
684:
685: /* ---------------------------------------------------------------------- */
686:
1.7 djm 687: static void nh(nh_ctx *hc, const UINT8 *buf, UINT32 padded_len,
1.1 pvalchev 688: UINT32 unpadded_len, UINT8 *result)
689: /* All-in-one nh_update() and nh_final() equivalent.
690: * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
691: * well aligned
692: */
693: {
694: UINT32 nbits;
695:
696: /* Initialize the hash state */
697: nbits = (unpadded_len << 3);
698:
699: ((UINT64 *)result)[0] = nbits;
700: #if (UMAC_OUTPUT_LEN >= 8)
701: ((UINT64 *)result)[1] = nbits;
702: #endif
703: #if (UMAC_OUTPUT_LEN >= 12)
704: ((UINT64 *)result)[2] = nbits;
705: #endif
706: #if (UMAC_OUTPUT_LEN == 16)
707: ((UINT64 *)result)[3] = nbits;
708: #endif
709:
710: nh_aux(hc->nh_key, buf, result, padded_len);
711: }
712:
713: /* ---------------------------------------------------------------------- */
714: /* ---------------------------------------------------------------------- */
715: /* ----- Begin UHASH Section -------------------------------------------- */
716: /* ---------------------------------------------------------------------- */
717: /* ---------------------------------------------------------------------- */
718:
719: /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
720: * hashed by NH. The NH output is then hashed by a polynomial-hash layer
721: * unless the initial data to be hashed is short. After the polynomial-
722: * layer, an inner-product hash is used to produce the final UHASH output.
723: *
724: * UHASH provides two interfaces, one all-at-once and another where data
725: * buffers are presented sequentially. In the sequential interface, the
726: * UHASH client calls the routine uhash_update() as many times as necessary.
727: * When there is no more data to be fed to UHASH, the client calls
728: * uhash_final() which
729: * calculates the UHASH output. Before beginning another UHASH calculation
730: * the uhash_reset() routine must be called. The all-at-once UHASH routine,
731: * uhash(), is equivalent to the sequence of calls uhash_update() and
732: * uhash_final(); however it is optimized and should be
733: * used whenever the sequential interface is not necessary.
734: *
735: * The routine uhash_init() initializes the uhash_ctx data structure and
736: * must be called once, before any other UHASH routine.
737: */
738:
739: /* ---------------------------------------------------------------------- */
740: /* ----- Constants and uhash_ctx ---------------------------------------- */
741: /* ---------------------------------------------------------------------- */
742:
743: /* ---------------------------------------------------------------------- */
744: /* ----- Poly hash and Inner-Product hash Constants --------------------- */
745: /* ---------------------------------------------------------------------- */
746:
747: /* Primes and masks */
748: #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */
749: #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */
750: #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */
751:
752:
753: /* ---------------------------------------------------------------------- */
754:
755: typedef struct uhash_ctx {
756: nh_ctx hash; /* Hash context for L1 NH hash */
757: UINT64 poly_key_8[STREAMS]; /* p64 poly keys */
758: UINT64 poly_accum[STREAMS]; /* poly hash result */
759: UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */
760: UINT32 ip_trans[STREAMS]; /* Inner-product translation */
761: UINT32 msg_len; /* Total length of data passed */
762: /* to uhash */
763: } uhash_ctx;
764: typedef struct uhash_ctx *uhash_ctx_t;
765:
766: /* ---------------------------------------------------------------------- */
767:
768:
769: /* The polynomial hashes use Horner's rule to evaluate a polynomial one
770: * word at a time. As described in the specification, poly32 and poly64
771: * require keys from special domains. The following implementations exploit
772: * the special domains to avoid overflow. The results are not guaranteed to
773: * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
774: * patches any errant values.
775: */
776:
777: static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
778: {
779: UINT32 key_hi = (UINT32)(key >> 32),
780: key_lo = (UINT32)key,
781: cur_hi = (UINT32)(cur >> 32),
782: cur_lo = (UINT32)cur,
783: x_lo,
784: x_hi;
785: UINT64 X,T,res;
786:
787: X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
788: x_lo = (UINT32)X;
789: x_hi = (UINT32)(X >> 32);
790:
791: res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
792:
793: T = ((UINT64)x_lo << 32);
794: res += T;
795: if (res < T)
796: res += 59;
797:
798: res += data;
799: if (res < data)
800: res += 59;
801:
802: return res;
803: }
804:
805:
806: /* Although UMAC is specified to use a ramped polynomial hash scheme, this
807: * implementation does not handle all ramp levels. Because we don't handle
808: * the ramp up to p128 modulus in this implementation, we are limited to
809: * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
810: * bytes input to UMAC per tag, ie. 16MB).
811: */
812: static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
813: {
814: int i;
815: UINT64 *data=(UINT64*)data_in;
816:
817: for (i = 0; i < STREAMS; i++) {
818: if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
819: hc->poly_accum[i] = poly64(hc->poly_accum[i],
820: hc->poly_key_8[i], p64 - 1);
821: hc->poly_accum[i] = poly64(hc->poly_accum[i],
822: hc->poly_key_8[i], (data[i] - 59));
823: } else {
824: hc->poly_accum[i] = poly64(hc->poly_accum[i],
825: hc->poly_key_8[i], data[i]);
826: }
827: }
828: }
829:
830:
831: /* ---------------------------------------------------------------------- */
832:
833:
834: /* The final step in UHASH is an inner-product hash. The poly hash
835: * produces a result not neccesarily WORD_LEN bytes long. The inner-
836: * product hash breaks the polyhash output into 16-bit chunks and
837: * multiplies each with a 36 bit key.
838: */
839:
840: static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
841: {
842: t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
843: t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
844: t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
845: t = t + ipkp[3] * (UINT64)(UINT16)(data);
846:
847: return t;
848: }
849:
850: static UINT32 ip_reduce_p36(UINT64 t)
851: {
852: /* Divisionless modular reduction */
853: UINT64 ret;
854:
855: ret = (t & m36) + 5 * (t >> 36);
856: if (ret >= p36)
857: ret -= p36;
858:
859: /* return least significant 32 bits */
860: return (UINT32)(ret);
861: }
862:
863:
864: /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
865: * the polyhash stage is skipped and ip_short is applied directly to the
866: * NH output.
867: */
868: static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
869: {
870: UINT64 t;
871: UINT64 *nhp = (UINT64 *)nh_res;
872:
873: t = ip_aux(0,ahc->ip_keys, nhp[0]);
874: STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
875: #if (UMAC_OUTPUT_LEN >= 8)
876: t = ip_aux(0,ahc->ip_keys+4, nhp[1]);
877: STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
878: #endif
879: #if (UMAC_OUTPUT_LEN >= 12)
880: t = ip_aux(0,ahc->ip_keys+8, nhp[2]);
881: STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
882: #endif
883: #if (UMAC_OUTPUT_LEN == 16)
884: t = ip_aux(0,ahc->ip_keys+12, nhp[3]);
885: STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
886: #endif
887: }
888:
889: /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
890: * the polyhash stage is not skipped and ip_long is applied to the
891: * polyhash output.
892: */
893: static void ip_long(uhash_ctx_t ahc, u_char *res)
894: {
895: int i;
896: UINT64 t;
897:
898: for (i = 0; i < STREAMS; i++) {
899: /* fix polyhash output not in Z_p64 */
900: if (ahc->poly_accum[i] >= p64)
901: ahc->poly_accum[i] -= p64;
902: t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
903: STORE_UINT32_BIG((UINT32 *)res+i,
904: ip_reduce_p36(t) ^ ahc->ip_trans[i]);
905: }
906: }
907:
908:
909: /* ---------------------------------------------------------------------- */
910:
911: /* ---------------------------------------------------------------------- */
912:
913: /* Reset uhash context for next hash session */
914: static int uhash_reset(uhash_ctx_t pc)
915: {
916: nh_reset(&pc->hash);
917: pc->msg_len = 0;
918: pc->poly_accum[0] = 1;
919: #if (UMAC_OUTPUT_LEN >= 8)
920: pc->poly_accum[1] = 1;
921: #endif
922: #if (UMAC_OUTPUT_LEN >= 12)
923: pc->poly_accum[2] = 1;
924: #endif
925: #if (UMAC_OUTPUT_LEN == 16)
926: pc->poly_accum[3] = 1;
927: #endif
928: return 1;
929: }
930:
931: /* ---------------------------------------------------------------------- */
932:
933: /* Given a pointer to the internal key needed by kdf() and a uhash context,
934: * initialize the NH context and generate keys needed for poly and inner-
935: * product hashing. All keys are endian adjusted in memory so that native
936: * loads cause correct keys to be in registers during calculation.
937: */
938: static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
939: {
940: int i;
941: UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
942:
943: /* Zero the entire uhash context */
944: memset(ahc, 0, sizeof(uhash_ctx));
945:
946: /* Initialize the L1 hash */
947: nh_init(&ahc->hash, prf_key);
948:
949: /* Setup L2 hash variables */
950: kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */
951: for (i = 0; i < STREAMS; i++) {
952: /* Fill keys from the buffer, skipping bytes in the buffer not
953: * used by this implementation. Endian reverse the keys if on a
954: * little-endian computer.
955: */
956: memcpy(ahc->poly_key_8+i, buf+24*i, 8);
957: endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
958: /* Mask the 64-bit keys to their special domain */
959: ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
960: ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */
961: }
962:
963: /* Setup L3-1 hash variables */
964: kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
965: for (i = 0; i < STREAMS; i++)
966: memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
967: 4*sizeof(UINT64));
968: endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
969: sizeof(ahc->ip_keys));
970: for (i = 0; i < STREAMS*4; i++)
971: ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */
972:
973: /* Setup L3-2 hash variables */
974: /* Fill buffer with index 4 key */
975: kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
976: endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
977: STREAMS * sizeof(UINT32));
978: }
979:
980: /* ---------------------------------------------------------------------- */
981:
982: #if 0
983: static uhash_ctx_t uhash_alloc(u_char key[])
984: {
985: /* Allocate memory and force to a 16-byte boundary. */
986: uhash_ctx_t ctx;
987: u_char bytes_to_add;
988: aes_int_key prf_key;
989:
990: ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
991: if (ctx) {
992: if (ALLOC_BOUNDARY) {
993: bytes_to_add = ALLOC_BOUNDARY -
994: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
995: ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
996: *((u_char *)ctx - 1) = bytes_to_add;
997: }
998: aes_key_setup(key,prf_key);
999: uhash_init(ctx, prf_key);
1000: }
1001: return (ctx);
1002: }
1003: #endif
1004:
1005: /* ---------------------------------------------------------------------- */
1006:
1007: #if 0
1008: static int uhash_free(uhash_ctx_t ctx)
1009: {
1010: /* Free memory allocated by uhash_alloc */
1011: u_char bytes_to_sub;
1012:
1013: if (ctx) {
1014: if (ALLOC_BOUNDARY) {
1015: bytes_to_sub = *((u_char *)ctx - 1);
1016: ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1017: }
1018: free(ctx);
1019: }
1020: return (1);
1021: }
1022: #endif
1023: /* ---------------------------------------------------------------------- */
1024:
1.7 djm 1025: static int uhash_update(uhash_ctx_t ctx, const u_char *input, long len)
1.1 pvalchev 1026: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1027: * hash each one with NH, calling the polyhash on each NH output.
1028: */
1029: {
1030: UWORD bytes_hashed, bytes_remaining;
1.3 pvalchev 1031: UINT64 result_buf[STREAMS];
1032: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1033:
1034: if (ctx->msg_len + len <= L1_KEY_LEN) {
1.7 djm 1035: nh_update(&ctx->hash, (const UINT8 *)input, len);
1.1 pvalchev 1036: ctx->msg_len += len;
1037: } else {
1038:
1039: bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1040: if (ctx->msg_len == L1_KEY_LEN)
1041: bytes_hashed = L1_KEY_LEN;
1042:
1043: if (bytes_hashed + len >= L1_KEY_LEN) {
1044:
1045: /* If some bytes have been passed to the hash function */
1046: /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1047: /* bytes to complete the current nh_block. */
1048: if (bytes_hashed) {
1049: bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1.7 djm 1050: nh_update(&ctx->hash, (const UINT8 *)input, bytes_remaining);
1.1 pvalchev 1051: nh_final(&ctx->hash, nh_result);
1052: ctx->msg_len += bytes_remaining;
1053: poly_hash(ctx,(UINT32 *)nh_result);
1054: len -= bytes_remaining;
1055: input += bytes_remaining;
1056: }
1057:
1058: /* Hash directly from input stream if enough bytes */
1059: while (len >= L1_KEY_LEN) {
1.7 djm 1060: nh(&ctx->hash, (const UINT8 *)input, L1_KEY_LEN,
1.1 pvalchev 1061: L1_KEY_LEN, nh_result);
1062: ctx->msg_len += L1_KEY_LEN;
1063: len -= L1_KEY_LEN;
1064: input += L1_KEY_LEN;
1065: poly_hash(ctx,(UINT32 *)nh_result);
1066: }
1067: }
1068:
1069: /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1070: if (len) {
1.7 djm 1071: nh_update(&ctx->hash, (const UINT8 *)input, len);
1.1 pvalchev 1072: ctx->msg_len += len;
1073: }
1074: }
1075:
1076: return (1);
1077: }
1078:
1079: /* ---------------------------------------------------------------------- */
1080:
1081: static int uhash_final(uhash_ctx_t ctx, u_char *res)
1082: /* Incorporate any pending data, pad, and generate tag */
1083: {
1.3 pvalchev 1084: UINT64 result_buf[STREAMS];
1085: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1086:
1087: if (ctx->msg_len > L1_KEY_LEN) {
1088: if (ctx->msg_len % L1_KEY_LEN) {
1089: nh_final(&ctx->hash, nh_result);
1090: poly_hash(ctx,(UINT32 *)nh_result);
1091: }
1092: ip_long(ctx, res);
1093: } else {
1094: nh_final(&ctx->hash, nh_result);
1095: ip_short(ctx,nh_result, res);
1096: }
1097: uhash_reset(ctx);
1098: return (1);
1099: }
1100:
1101: /* ---------------------------------------------------------------------- */
1102:
1103: #if 0
1104: static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1105: /* assumes that msg is in a writable buffer of length divisible by */
1106: /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */
1107: {
1108: UINT8 nh_result[STREAMS*sizeof(UINT64)];
1109: UINT32 nh_len;
1110: int extra_zeroes_needed;
1111:
1112: /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1113: * the polyhash.
1114: */
1115: if (len <= L1_KEY_LEN) {
1116: if (len == 0) /* If zero length messages will not */
1117: nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */
1118: else
1119: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1120: extra_zeroes_needed = nh_len - len;
1121: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1122: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1123: ip_short(ahc,nh_result, res);
1124: } else {
1125: /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1126: * output to poly_hash().
1127: */
1128: do {
1129: nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1130: poly_hash(ahc,(UINT32 *)nh_result);
1131: len -= L1_KEY_LEN;
1132: msg += L1_KEY_LEN;
1133: } while (len >= L1_KEY_LEN);
1134: if (len) {
1135: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1136: extra_zeroes_needed = nh_len - len;
1137: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1138: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1139: poly_hash(ahc,(UINT32 *)nh_result);
1140: }
1141:
1142: ip_long(ahc, res);
1143: }
1144:
1145: uhash_reset(ahc);
1146: return 1;
1147: }
1148: #endif
1149:
1150: /* ---------------------------------------------------------------------- */
1151: /* ---------------------------------------------------------------------- */
1152: /* ----- Begin UMAC Section --------------------------------------------- */
1153: /* ---------------------------------------------------------------------- */
1154: /* ---------------------------------------------------------------------- */
1155:
1156: /* The UMAC interface has two interfaces, an all-at-once interface where
1157: * the entire message to be authenticated is passed to UMAC in one buffer,
1158: * and a sequential interface where the message is presented a little at a
1159: * time. The all-at-once is more optimaized than the sequential version and
1160: * should be preferred when the sequential interface is not required.
1161: */
1162: struct umac_ctx {
1163: uhash_ctx hash; /* Hash function for message compression */
1164: pdf_ctx pdf; /* PDF for hashed output */
1165: void *free_ptr; /* Address to free this struct via */
1166: } umac_ctx;
1167:
1168: /* ---------------------------------------------------------------------- */
1169:
1170: #if 0
1171: int umac_reset(struct umac_ctx *ctx)
1172: /* Reset the hash function to begin a new authentication. */
1173: {
1174: uhash_reset(&ctx->hash);
1175: return (1);
1176: }
1177: #endif
1178:
1179: /* ---------------------------------------------------------------------- */
1180:
1181: int umac_delete(struct umac_ctx *ctx)
1182: /* Deallocate the ctx structure */
1183: {
1184: if (ctx) {
1185: if (ALLOC_BOUNDARY)
1186: ctx = (struct umac_ctx *)ctx->free_ptr;
1.5 djm 1187: free(ctx);
1.1 pvalchev 1188: }
1189: return (1);
1190: }
1191:
1192: /* ---------------------------------------------------------------------- */
1193:
1.7 djm 1194: struct umac_ctx *umac_new(const u_char key[])
1.1 pvalchev 1195: /* Dynamically allocate a umac_ctx struct, initialize variables,
1196: * generate subkeys from key. Align to 16-byte boundary.
1197: */
1198: {
1199: struct umac_ctx *ctx, *octx;
1200: size_t bytes_to_add;
1201: aes_int_key prf_key;
1202:
1.8 djm 1203: octx = ctx = xcalloc(1, sizeof(*ctx) + ALLOC_BOUNDARY);
1.1 pvalchev 1204: if (ctx) {
1205: if (ALLOC_BOUNDARY) {
1206: bytes_to_add = ALLOC_BOUNDARY -
1207: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY - 1));
1208: ctx = (struct umac_ctx *)((u_char *)ctx + bytes_to_add);
1209: }
1210: ctx->free_ptr = octx;
1.7 djm 1211: aes_key_setup(key, prf_key);
1.1 pvalchev 1212: pdf_init(&ctx->pdf, prf_key);
1213: uhash_init(&ctx->hash, prf_key);
1214: }
1215:
1216: return (ctx);
1217: }
1218:
1219: /* ---------------------------------------------------------------------- */
1220:
1.7 djm 1221: int umac_final(struct umac_ctx *ctx, u_char tag[], const u_char nonce[8])
1.1 pvalchev 1222: /* Incorporate any pending data, pad, and generate tag */
1223: {
1224: uhash_final(&ctx->hash, (u_char *)tag);
1.7 djm 1225: pdf_gen_xor(&ctx->pdf, (const UINT8 *)nonce, (UINT8 *)tag);
1.1 pvalchev 1226:
1227: return (1);
1228: }
1229:
1230: /* ---------------------------------------------------------------------- */
1231:
1.7 djm 1232: int umac_update(struct umac_ctx *ctx, const u_char *input, long len)
1.1 pvalchev 1233: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */
1234: /* hash each one, calling the PDF on the hashed output whenever the hash- */
1235: /* output buffer is full. */
1236: {
1237: uhash_update(&ctx->hash, input, len);
1238: return (1);
1239: }
1240:
1241: /* ---------------------------------------------------------------------- */
1242:
1243: #if 0
1244: int umac(struct umac_ctx *ctx, u_char *input,
1245: long len, u_char tag[],
1246: u_char nonce[8])
1247: /* All-in-one version simply calls umac_update() and umac_final(). */
1248: {
1249: uhash(&ctx->hash, input, len, (u_char *)tag);
1250: pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1251:
1252: return (1);
1253: }
1254: #endif
1255:
1256: /* ---------------------------------------------------------------------- */
1257: /* ---------------------------------------------------------------------- */
1258: /* ----- End UMAC Section ----------------------------------------------- */
1259: /* ---------------------------------------------------------------------- */
1260: /* ---------------------------------------------------------------------- */
![[BACK]](/icons/back.gif){kind=icon}
Return to umac.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [local] / src / usr.bin / ssh