Annotation of src/usr.bin/ssh/umac.c, Revision 1.6
1.6 ! djm 1: /* $OpenBSD: umac.c,v 1.5 2013/05/17 00:13:14 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>
68:
1.2 stevesk 69: #include "xmalloc.h"
1.1 pvalchev 70: #include "umac.h"
71: #include <string.h>
72: #include <stdlib.h>
73: #include <stddef.h>
74:
75: /* ---------------------------------------------------------------------- */
76: /* --- Primitive Data Types --- */
77: /* ---------------------------------------------------------------------- */
78:
79: /* The following assumptions may need change on your system */
80: typedef u_int8_t UINT8; /* 1 byte */
81: typedef u_int16_t UINT16; /* 2 byte */
82: typedef u_int32_t UINT32; /* 4 byte */
83: typedef u_int64_t UINT64; /* 8 bytes */
84: typedef unsigned int UWORD; /* Register */
85:
86: /* ---------------------------------------------------------------------- */
87: /* --- Constants -------------------------------------------------------- */
88: /* ---------------------------------------------------------------------- */
89:
90: #define UMAC_KEY_LEN 16 /* UMAC takes 16 bytes of external key */
91:
92: /* Message "words" are read from memory in an endian-specific manner. */
93: /* For this implementation to behave correctly, __LITTLE_ENDIAN__ must */
94: /* be set true if the host computer is little-endian. */
95:
96: #if BYTE_ORDER == LITTLE_ENDIAN
97: #define __LITTLE_ENDIAN__ 1
98: #else
99: #define __LITTLE_ENDIAN__ 0
100: #endif
101:
102: /* ---------------------------------------------------------------------- */
103: /* ---------------------------------------------------------------------- */
104: /* ----- Architecture Specific ------------------------------------------ */
105: /* ---------------------------------------------------------------------- */
106: /* ---------------------------------------------------------------------- */
107:
108:
109: /* ---------------------------------------------------------------------- */
110: /* ---------------------------------------------------------------------- */
111: /* ----- Primitive Routines --------------------------------------------- */
112: /* ---------------------------------------------------------------------- */
113: /* ---------------------------------------------------------------------- */
114:
115:
116: /* ---------------------------------------------------------------------- */
117: /* --- 32-bit by 32-bit to 64-bit Multiplication ------------------------ */
118: /* ---------------------------------------------------------------------- */
119:
120: #define MUL64(a,b) ((UINT64)((UINT64)(UINT32)(a) * (UINT64)(UINT32)(b)))
121:
122: /* ---------------------------------------------------------------------- */
123: /* --- Endian Conversion --- Forcing assembly on some platforms */
124: /* ---------------------------------------------------------------------- */
125:
126: #if 0
127: static UINT32 LOAD_UINT32_REVERSED(void *ptr)
128: {
129: UINT32 temp = *(UINT32 *)ptr;
130: temp = (temp >> 24) | ((temp & 0x00FF0000) >> 8 )
131: | ((temp & 0x0000FF00) << 8 ) | (temp << 24);
132: return (UINT32)temp;
133: }
134:
135: static void STORE_UINT32_REVERSED(void *ptr, UINT32 x)
136: {
137: UINT32 i = (UINT32)x;
138: *(UINT32 *)ptr = (i >> 24) | ((i & 0x00FF0000) >> 8 )
139: | ((i & 0x0000FF00) << 8 ) | (i << 24);
140: }
141: #endif
142:
143: /* The following definitions use the above reversal-primitives to do the right
144: * thing on endian specific load and stores.
145: */
146:
147: #define LOAD_UINT32_REVERSED(p) (swap32(*(UINT32 *)(p)))
148: #define STORE_UINT32_REVERSED(p,v) (*(UINT32 *)(p) = swap32(v))
149:
150: #if (__LITTLE_ENDIAN__)
151: #define LOAD_UINT32_LITTLE(ptr) (*(UINT32 *)(ptr))
152: #define STORE_UINT32_BIG(ptr,x) STORE_UINT32_REVERSED(ptr,x)
153: #else
154: #define LOAD_UINT32_LITTLE(ptr) LOAD_UINT32_REVERSED(ptr)
155: #define STORE_UINT32_BIG(ptr,x) (*(UINT32 *)(ptr) = (UINT32)(x))
156: #endif
157:
158:
159:
160: /* ---------------------------------------------------------------------- */
161: /* ---------------------------------------------------------------------- */
162: /* ----- Begin KDF & PDF Section ---------------------------------------- */
163: /* ---------------------------------------------------------------------- */
164: /* ---------------------------------------------------------------------- */
165:
166: /* UMAC uses AES with 16 byte block and key lengths */
167: #define AES_BLOCK_LEN 16
168:
169: /* OpenSSL's AES */
170: #include <openssl/aes.h>
171: typedef AES_KEY aes_int_key[1];
172: #define aes_encryption(in,out,int_key) \
173: AES_encrypt((u_char *)(in),(u_char *)(out),(AES_KEY *)int_key)
174: #define aes_key_setup(key,int_key) \
175: AES_set_encrypt_key((u_char *)(key),UMAC_KEY_LEN*8,int_key)
176:
177: /* The user-supplied UMAC key is stretched using AES in a counter
178: * mode to supply all random bits needed by UMAC. The kdf function takes
179: * an AES internal key representation 'key' and writes a stream of
180: * 'nbytes' bytes to the memory pointed at by 'buffer_ptr'. Each distinct
181: * 'ndx' causes a distinct byte stream.
182: */
183: static void kdf(void *buffer_ptr, aes_int_key key, UINT8 ndx, int nbytes)
184: {
185: UINT8 in_buf[AES_BLOCK_LEN] = {0};
186: UINT8 out_buf[AES_BLOCK_LEN];
187: UINT8 *dst_buf = (UINT8 *)buffer_ptr;
188: int i;
189:
190: /* Setup the initial value */
191: in_buf[AES_BLOCK_LEN-9] = ndx;
192: in_buf[AES_BLOCK_LEN-1] = i = 1;
193:
194: while (nbytes >= AES_BLOCK_LEN) {
195: aes_encryption(in_buf, out_buf, key);
196: memcpy(dst_buf,out_buf,AES_BLOCK_LEN);
197: in_buf[AES_BLOCK_LEN-1] = ++i;
198: nbytes -= AES_BLOCK_LEN;
199: dst_buf += AES_BLOCK_LEN;
200: }
201: if (nbytes) {
202: aes_encryption(in_buf, out_buf, key);
203: memcpy(dst_buf,out_buf,nbytes);
204: }
205: }
206:
207: /* The final UHASH result is XOR'd with the output of a pseudorandom
208: * function. Here, we use AES to generate random output and
209: * xor the appropriate bytes depending on the last bits of nonce.
210: * This scheme is optimized for sequential, increasing big-endian nonces.
211: */
212:
213: typedef struct {
214: UINT8 cache[AES_BLOCK_LEN]; /* Previous AES output is saved */
215: UINT8 nonce[AES_BLOCK_LEN]; /* The AES input making above cache */
216: aes_int_key prf_key; /* Expanded AES key for PDF */
217: } pdf_ctx;
218:
219: static void pdf_init(pdf_ctx *pc, aes_int_key prf_key)
220: {
221: UINT8 buf[UMAC_KEY_LEN];
222:
223: kdf(buf, prf_key, 0, UMAC_KEY_LEN);
224: aes_key_setup(buf, pc->prf_key);
225:
226: /* Initialize pdf and cache */
227: memset(pc->nonce, 0, sizeof(pc->nonce));
228: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
229: }
230:
231: static void pdf_gen_xor(pdf_ctx *pc, UINT8 nonce[8], UINT8 buf[8])
232: {
233: /* 'ndx' indicates that we'll be using the 0th or 1st eight bytes
234: * of the AES output. If last time around we returned the ndx-1st
235: * element, then we may have the result in the cache already.
236: */
237:
238: #if (UMAC_OUTPUT_LEN == 4)
239: #define LOW_BIT_MASK 3
240: #elif (UMAC_OUTPUT_LEN == 8)
241: #define LOW_BIT_MASK 1
242: #elif (UMAC_OUTPUT_LEN > 8)
243: #define LOW_BIT_MASK 0
244: #endif
1.6 ! djm 245: union {
! 246: UINT8 tmp_nonce_lo[4];
! 247: UINT32 align;
! 248: } t;
1.1 pvalchev 249: #if LOW_BIT_MASK != 0
250: int ndx = nonce[7] & LOW_BIT_MASK;
251: #endif
1.6 ! djm 252: *(UINT32 *)t.tmp_nonce_lo = ((UINT32 *)nonce)[1];
! 253: t.tmp_nonce_lo[3] &= ~LOW_BIT_MASK; /* zero last bit */
1.1 pvalchev 254:
1.6 ! djm 255: if ( (((UINT32 *)t.tmp_nonce_lo)[0] != ((UINT32 *)pc->nonce)[1]) ||
1.1 pvalchev 256: (((UINT32 *)nonce)[0] != ((UINT32 *)pc->nonce)[0]) )
257: {
258: ((UINT32 *)pc->nonce)[0] = ((UINT32 *)nonce)[0];
1.6 ! djm 259: ((UINT32 *)pc->nonce)[1] = ((UINT32 *)t.tmp_nonce_lo)[0];
1.1 pvalchev 260: aes_encryption(pc->nonce, pc->cache, pc->prf_key);
261: }
262:
263: #if (UMAC_OUTPUT_LEN == 4)
264: *((UINT32 *)buf) ^= ((UINT32 *)pc->cache)[ndx];
265: #elif (UMAC_OUTPUT_LEN == 8)
266: *((UINT64 *)buf) ^= ((UINT64 *)pc->cache)[ndx];
267: #elif (UMAC_OUTPUT_LEN == 12)
268: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
269: ((UINT32 *)buf)[2] ^= ((UINT32 *)pc->cache)[2];
270: #elif (UMAC_OUTPUT_LEN == 16)
271: ((UINT64 *)buf)[0] ^= ((UINT64 *)pc->cache)[0];
272: ((UINT64 *)buf)[1] ^= ((UINT64 *)pc->cache)[1];
273: #endif
274: }
275:
276: /* ---------------------------------------------------------------------- */
277: /* ---------------------------------------------------------------------- */
278: /* ----- Begin NH Hash Section ------------------------------------------ */
279: /* ---------------------------------------------------------------------- */
280: /* ---------------------------------------------------------------------- */
281:
282: /* The NH-based hash functions used in UMAC are described in the UMAC paper
283: * and specification, both of which can be found at the UMAC website.
284: * The interface to this implementation has two
285: * versions, one expects the entire message being hashed to be passed
286: * in a single buffer and returns the hash result immediately. The second
287: * allows the message to be passed in a sequence of buffers. In the
288: * muliple-buffer interface, the client calls the routine nh_update() as
289: * many times as necessary. When there is no more data to be fed to the
290: * hash, the client calls nh_final() which calculates the hash output.
291: * Before beginning another hash calculation the nh_reset() routine
292: * must be called. The single-buffer routine, nh(), is equivalent to
293: * the sequence of calls nh_update() and nh_final(); however it is
294: * optimized and should be prefered whenever the multiple-buffer interface
295: * is not necessary. When using either interface, it is the client's
296: * responsability to pass no more than L1_KEY_LEN bytes per hash result.
297: *
298: * The routine nh_init() initializes the nh_ctx data structure and
299: * must be called once, before any other PDF routine.
300: */
301:
302: /* The "nh_aux" routines do the actual NH hashing work. They
303: * expect buffers to be multiples of L1_PAD_BOUNDARY. These routines
304: * produce output for all STREAMS NH iterations in one call,
305: * allowing the parallel implementation of the streams.
306: */
307:
308: #define STREAMS (UMAC_OUTPUT_LEN / 4) /* Number of times hash is applied */
309: #define L1_KEY_LEN 1024 /* Internal key bytes */
310: #define L1_KEY_SHIFT 16 /* Toeplitz key shift between streams */
311: #define L1_PAD_BOUNDARY 32 /* pad message to boundary multiple */
312: #define ALLOC_BOUNDARY 16 /* Keep buffers aligned to this */
313: #define HASH_BUF_BYTES 64 /* nh_aux_hb buffer multiple */
314:
315: typedef struct {
316: UINT8 nh_key [L1_KEY_LEN + L1_KEY_SHIFT * (STREAMS - 1)]; /* NH Key */
1.4 djm 317: UINT8 data [HASH_BUF_BYTES]; /* Incoming data buffer */
1.1 pvalchev 318: int next_data_empty; /* Bookeeping variable for data buffer. */
319: int bytes_hashed; /* Bytes (out of L1_KEY_LEN) incorperated. */
320: UINT64 state[STREAMS]; /* on-line state */
321: } nh_ctx;
322:
323:
324: #if (UMAC_OUTPUT_LEN == 4)
325:
326: static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen)
327: /* NH hashing primitive. Previous (partial) hash result is loaded and
328: * then stored via hp pointer. The length of the data pointed at by "dp",
329: * "dlen", is guaranteed to be divisible by L1_PAD_BOUNDARY (32). Key
330: * is expected to be endian compensated in memory at key setup.
331: */
332: {
333: UINT64 h;
334: UWORD c = dlen / 32;
335: UINT32 *k = (UINT32 *)kp;
336: UINT32 *d = (UINT32 *)dp;
337: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
338: UINT32 k0,k1,k2,k3,k4,k5,k6,k7;
339:
340: h = *((UINT64 *)hp);
341: do {
342: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
343: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
344: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
345: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
346: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
347: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
348: h += MUL64((k0 + d0), (k4 + d4));
349: h += MUL64((k1 + d1), (k5 + d5));
350: h += MUL64((k2 + d2), (k6 + d6));
351: h += MUL64((k3 + d3), (k7 + d7));
352:
353: d += 8;
354: k += 8;
355: } while (--c);
356: *((UINT64 *)hp) = h;
357: }
358:
359: #elif (UMAC_OUTPUT_LEN == 8)
360:
361: static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen)
362: /* Same as previous nh_aux, but two streams are handled in one pass,
363: * reading and writing 16 bytes of hash-state per call.
364: */
365: {
366: UINT64 h1,h2;
367: UWORD c = dlen / 32;
368: UINT32 *k = (UINT32 *)kp;
369: UINT32 *d = (UINT32 *)dp;
370: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
371: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
372: k8,k9,k10,k11;
373:
374: h1 = *((UINT64 *)hp);
375: h2 = *((UINT64 *)hp + 1);
376: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
377: do {
378: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
379: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
380: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
381: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
382: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
383: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
384:
385: h1 += MUL64((k0 + d0), (k4 + d4));
386: h2 += MUL64((k4 + d0), (k8 + d4));
387:
388: h1 += MUL64((k1 + d1), (k5 + d5));
389: h2 += MUL64((k5 + d1), (k9 + d5));
390:
391: h1 += MUL64((k2 + d2), (k6 + d6));
392: h2 += MUL64((k6 + d2), (k10 + d6));
393:
394: h1 += MUL64((k3 + d3), (k7 + d7));
395: h2 += MUL64((k7 + d3), (k11 + d7));
396:
397: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
398:
399: d += 8;
400: k += 8;
401: } while (--c);
402: ((UINT64 *)hp)[0] = h1;
403: ((UINT64 *)hp)[1] = h2;
404: }
405:
406: #elif (UMAC_OUTPUT_LEN == 12)
407:
408: static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen)
409: /* Same as previous nh_aux, but two streams are handled in one pass,
410: * reading and writing 24 bytes of hash-state per call.
411: */
412: {
413: UINT64 h1,h2,h3;
414: UWORD c = dlen / 32;
415: UINT32 *k = (UINT32 *)kp;
416: UINT32 *d = (UINT32 *)dp;
417: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
418: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
419: k8,k9,k10,k11,k12,k13,k14,k15;
420:
421: h1 = *((UINT64 *)hp);
422: h2 = *((UINT64 *)hp + 1);
423: h3 = *((UINT64 *)hp + 2);
424: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
425: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
426: do {
427: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
428: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
429: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
430: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
431: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
432: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
433:
434: h1 += MUL64((k0 + d0), (k4 + d4));
435: h2 += MUL64((k4 + d0), (k8 + d4));
436: h3 += MUL64((k8 + d0), (k12 + d4));
437:
438: h1 += MUL64((k1 + d1), (k5 + d5));
439: h2 += MUL64((k5 + d1), (k9 + d5));
440: h3 += MUL64((k9 + d1), (k13 + d5));
441:
442: h1 += MUL64((k2 + d2), (k6 + d6));
443: h2 += MUL64((k6 + d2), (k10 + d6));
444: h3 += MUL64((k10 + d2), (k14 + d6));
445:
446: h1 += MUL64((k3 + d3), (k7 + d7));
447: h2 += MUL64((k7 + d3), (k11 + d7));
448: h3 += MUL64((k11 + d3), (k15 + d7));
449:
450: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
451: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
452:
453: d += 8;
454: k += 8;
455: } while (--c);
456: ((UINT64 *)hp)[0] = h1;
457: ((UINT64 *)hp)[1] = h2;
458: ((UINT64 *)hp)[2] = h3;
459: }
460:
461: #elif (UMAC_OUTPUT_LEN == 16)
462:
463: static void nh_aux(void *kp, void *dp, void *hp, UINT32 dlen)
464: /* Same as previous nh_aux, but two streams are handled in one pass,
465: * reading and writing 24 bytes of hash-state per call.
466: */
467: {
468: UINT64 h1,h2,h3,h4;
469: UWORD c = dlen / 32;
470: UINT32 *k = (UINT32 *)kp;
471: UINT32 *d = (UINT32 *)dp;
472: UINT32 d0,d1,d2,d3,d4,d5,d6,d7;
473: UINT32 k0,k1,k2,k3,k4,k5,k6,k7,
474: k8,k9,k10,k11,k12,k13,k14,k15,
475: k16,k17,k18,k19;
476:
477: h1 = *((UINT64 *)hp);
478: h2 = *((UINT64 *)hp + 1);
479: h3 = *((UINT64 *)hp + 2);
480: h4 = *((UINT64 *)hp + 3);
481: k0 = *(k+0); k1 = *(k+1); k2 = *(k+2); k3 = *(k+3);
482: k4 = *(k+4); k5 = *(k+5); k6 = *(k+6); k7 = *(k+7);
483: do {
484: d0 = LOAD_UINT32_LITTLE(d+0); d1 = LOAD_UINT32_LITTLE(d+1);
485: d2 = LOAD_UINT32_LITTLE(d+2); d3 = LOAD_UINT32_LITTLE(d+3);
486: d4 = LOAD_UINT32_LITTLE(d+4); d5 = LOAD_UINT32_LITTLE(d+5);
487: d6 = LOAD_UINT32_LITTLE(d+6); d7 = LOAD_UINT32_LITTLE(d+7);
488: k8 = *(k+8); k9 = *(k+9); k10 = *(k+10); k11 = *(k+11);
489: k12 = *(k+12); k13 = *(k+13); k14 = *(k+14); k15 = *(k+15);
490: k16 = *(k+16); k17 = *(k+17); k18 = *(k+18); k19 = *(k+19);
491:
492: h1 += MUL64((k0 + d0), (k4 + d4));
493: h2 += MUL64((k4 + d0), (k8 + d4));
494: h3 += MUL64((k8 + d0), (k12 + d4));
495: h4 += MUL64((k12 + d0), (k16 + d4));
496:
497: h1 += MUL64((k1 + d1), (k5 + d5));
498: h2 += MUL64((k5 + d1), (k9 + d5));
499: h3 += MUL64((k9 + d1), (k13 + d5));
500: h4 += MUL64((k13 + d1), (k17 + d5));
501:
502: h1 += MUL64((k2 + d2), (k6 + d6));
503: h2 += MUL64((k6 + d2), (k10 + d6));
504: h3 += MUL64((k10 + d2), (k14 + d6));
505: h4 += MUL64((k14 + d2), (k18 + d6));
506:
507: h1 += MUL64((k3 + d3), (k7 + d7));
508: h2 += MUL64((k7 + d3), (k11 + d7));
509: h3 += MUL64((k11 + d3), (k15 + d7));
510: h4 += MUL64((k15 + d3), (k19 + d7));
511:
512: k0 = k8; k1 = k9; k2 = k10; k3 = k11;
513: k4 = k12; k5 = k13; k6 = k14; k7 = k15;
514: k8 = k16; k9 = k17; k10 = k18; k11 = k19;
515:
516: d += 8;
517: k += 8;
518: } while (--c);
519: ((UINT64 *)hp)[0] = h1;
520: ((UINT64 *)hp)[1] = h2;
521: ((UINT64 *)hp)[2] = h3;
522: ((UINT64 *)hp)[3] = h4;
523: }
524:
525: /* ---------------------------------------------------------------------- */
526: #endif /* UMAC_OUTPUT_LENGTH */
527: /* ---------------------------------------------------------------------- */
528:
529:
530: /* ---------------------------------------------------------------------- */
531:
532: static void nh_transform(nh_ctx *hc, UINT8 *buf, UINT32 nbytes)
533: /* This function is a wrapper for the primitive NH hash functions. It takes
534: * as argument "hc" the current hash context and a buffer which must be a
535: * multiple of L1_PAD_BOUNDARY. The key passed to nh_aux is offset
536: * appropriately according to how much message has been hashed already.
537: */
538: {
539: UINT8 *key;
540:
541: key = hc->nh_key + hc->bytes_hashed;
542: nh_aux(key, buf, hc->state, nbytes);
543: }
544:
545: /* ---------------------------------------------------------------------- */
546:
547: static void endian_convert(void *buf, UWORD bpw, UINT32 num_bytes)
548: /* We endian convert the keys on little-endian computers to */
549: /* compensate for the lack of big-endian memory reads during hashing. */
550: {
551: UWORD iters = num_bytes / bpw;
552: if (bpw == 4) {
553: UINT32 *p = (UINT32 *)buf;
554: do {
555: *p = LOAD_UINT32_REVERSED(p);
556: p++;
557: } while (--iters);
558: } else if (bpw == 8) {
559: UINT32 *p = (UINT32 *)buf;
560: UINT32 t;
561: do {
562: t = LOAD_UINT32_REVERSED(p+1);
563: p[1] = LOAD_UINT32_REVERSED(p);
564: p[0] = t;
565: p += 2;
566: } while (--iters);
567: }
568: }
569: #if (__LITTLE_ENDIAN__)
570: #define endian_convert_if_le(x,y,z) endian_convert((x),(y),(z))
571: #else
572: #define endian_convert_if_le(x,y,z) do{}while(0) /* Do nothing */
573: #endif
574:
575: /* ---------------------------------------------------------------------- */
576:
577: static void nh_reset(nh_ctx *hc)
578: /* Reset nh_ctx to ready for hashing of new data */
579: {
580: hc->bytes_hashed = 0;
581: hc->next_data_empty = 0;
582: hc->state[0] = 0;
583: #if (UMAC_OUTPUT_LEN >= 8)
584: hc->state[1] = 0;
585: #endif
586: #if (UMAC_OUTPUT_LEN >= 12)
587: hc->state[2] = 0;
588: #endif
589: #if (UMAC_OUTPUT_LEN == 16)
590: hc->state[3] = 0;
591: #endif
592:
593: }
594:
595: /* ---------------------------------------------------------------------- */
596:
597: static void nh_init(nh_ctx *hc, aes_int_key prf_key)
598: /* Generate nh_key, endian convert and reset to be ready for hashing. */
599: {
600: kdf(hc->nh_key, prf_key, 1, sizeof(hc->nh_key));
601: endian_convert_if_le(hc->nh_key, 4, sizeof(hc->nh_key));
602: nh_reset(hc);
603: }
604:
605: /* ---------------------------------------------------------------------- */
606:
607: static void nh_update(nh_ctx *hc, UINT8 *buf, UINT32 nbytes)
608: /* Incorporate nbytes of data into a nh_ctx, buffer whatever is not an */
609: /* even multiple of HASH_BUF_BYTES. */
610: {
611: UINT32 i,j;
612:
613: j = hc->next_data_empty;
614: if ((j + nbytes) >= HASH_BUF_BYTES) {
615: if (j) {
616: i = HASH_BUF_BYTES - j;
617: memcpy(hc->data+j, buf, i);
618: nh_transform(hc,hc->data,HASH_BUF_BYTES);
619: nbytes -= i;
620: buf += i;
621: hc->bytes_hashed += HASH_BUF_BYTES;
622: }
623: if (nbytes >= HASH_BUF_BYTES) {
624: i = nbytes & ~(HASH_BUF_BYTES - 1);
625: nh_transform(hc, buf, i);
626: nbytes -= i;
627: buf += i;
628: hc->bytes_hashed += i;
629: }
630: j = 0;
631: }
632: memcpy(hc->data + j, buf, nbytes);
633: hc->next_data_empty = j + nbytes;
634: }
635:
636: /* ---------------------------------------------------------------------- */
637:
638: static void zero_pad(UINT8 *p, int nbytes)
639: {
640: /* Write "nbytes" of zeroes, beginning at "p" */
641: if (nbytes >= (int)sizeof(UWORD)) {
642: while ((ptrdiff_t)p % sizeof(UWORD)) {
643: *p = 0;
644: nbytes--;
645: p++;
646: }
647: while (nbytes >= (int)sizeof(UWORD)) {
648: *(UWORD *)p = 0;
649: nbytes -= sizeof(UWORD);
650: p += sizeof(UWORD);
651: }
652: }
653: while (nbytes) {
654: *p = 0;
655: nbytes--;
656: p++;
657: }
658: }
659:
660: /* ---------------------------------------------------------------------- */
661:
662: static void nh_final(nh_ctx *hc, UINT8 *result)
663: /* After passing some number of data buffers to nh_update() for integration
664: * into an NH context, nh_final is called to produce a hash result. If any
665: * bytes are in the buffer hc->data, incorporate them into the
666: * NH context. Finally, add into the NH accumulation "state" the total number
667: * of bits hashed. The resulting numbers are written to the buffer "result".
668: * If nh_update was never called, L1_PAD_BOUNDARY zeroes are incorporated.
669: */
670: {
671: int nh_len, nbits;
672:
673: if (hc->next_data_empty != 0) {
674: nh_len = ((hc->next_data_empty + (L1_PAD_BOUNDARY - 1)) &
675: ~(L1_PAD_BOUNDARY - 1));
676: zero_pad(hc->data + hc->next_data_empty,
677: nh_len - hc->next_data_empty);
678: nh_transform(hc, hc->data, nh_len);
679: hc->bytes_hashed += hc->next_data_empty;
680: } else if (hc->bytes_hashed == 0) {
681: nh_len = L1_PAD_BOUNDARY;
682: zero_pad(hc->data, L1_PAD_BOUNDARY);
683: nh_transform(hc, hc->data, nh_len);
684: }
685:
686: nbits = (hc->bytes_hashed << 3);
687: ((UINT64 *)result)[0] = ((UINT64 *)hc->state)[0] + nbits;
688: #if (UMAC_OUTPUT_LEN >= 8)
689: ((UINT64 *)result)[1] = ((UINT64 *)hc->state)[1] + nbits;
690: #endif
691: #if (UMAC_OUTPUT_LEN >= 12)
692: ((UINT64 *)result)[2] = ((UINT64 *)hc->state)[2] + nbits;
693: #endif
694: #if (UMAC_OUTPUT_LEN == 16)
695: ((UINT64 *)result)[3] = ((UINT64 *)hc->state)[3] + nbits;
696: #endif
697: nh_reset(hc);
698: }
699:
700: /* ---------------------------------------------------------------------- */
701:
702: static void nh(nh_ctx *hc, UINT8 *buf, UINT32 padded_len,
703: UINT32 unpadded_len, UINT8 *result)
704: /* All-in-one nh_update() and nh_final() equivalent.
705: * Assumes that padded_len is divisible by L1_PAD_BOUNDARY and result is
706: * well aligned
707: */
708: {
709: UINT32 nbits;
710:
711: /* Initialize the hash state */
712: nbits = (unpadded_len << 3);
713:
714: ((UINT64 *)result)[0] = nbits;
715: #if (UMAC_OUTPUT_LEN >= 8)
716: ((UINT64 *)result)[1] = nbits;
717: #endif
718: #if (UMAC_OUTPUT_LEN >= 12)
719: ((UINT64 *)result)[2] = nbits;
720: #endif
721: #if (UMAC_OUTPUT_LEN == 16)
722: ((UINT64 *)result)[3] = nbits;
723: #endif
724:
725: nh_aux(hc->nh_key, buf, result, padded_len);
726: }
727:
728: /* ---------------------------------------------------------------------- */
729: /* ---------------------------------------------------------------------- */
730: /* ----- Begin UHASH Section -------------------------------------------- */
731: /* ---------------------------------------------------------------------- */
732: /* ---------------------------------------------------------------------- */
733:
734: /* UHASH is a multi-layered algorithm. Data presented to UHASH is first
735: * hashed by NH. The NH output is then hashed by a polynomial-hash layer
736: * unless the initial data to be hashed is short. After the polynomial-
737: * layer, an inner-product hash is used to produce the final UHASH output.
738: *
739: * UHASH provides two interfaces, one all-at-once and another where data
740: * buffers are presented sequentially. In the sequential interface, the
741: * UHASH client calls the routine uhash_update() as many times as necessary.
742: * When there is no more data to be fed to UHASH, the client calls
743: * uhash_final() which
744: * calculates the UHASH output. Before beginning another UHASH calculation
745: * the uhash_reset() routine must be called. The all-at-once UHASH routine,
746: * uhash(), is equivalent to the sequence of calls uhash_update() and
747: * uhash_final(); however it is optimized and should be
748: * used whenever the sequential interface is not necessary.
749: *
750: * The routine uhash_init() initializes the uhash_ctx data structure and
751: * must be called once, before any other UHASH routine.
752: */
753:
754: /* ---------------------------------------------------------------------- */
755: /* ----- Constants and uhash_ctx ---------------------------------------- */
756: /* ---------------------------------------------------------------------- */
757:
758: /* ---------------------------------------------------------------------- */
759: /* ----- Poly hash and Inner-Product hash Constants --------------------- */
760: /* ---------------------------------------------------------------------- */
761:
762: /* Primes and masks */
763: #define p36 ((UINT64)0x0000000FFFFFFFFBull) /* 2^36 - 5 */
764: #define p64 ((UINT64)0xFFFFFFFFFFFFFFC5ull) /* 2^64 - 59 */
765: #define m36 ((UINT64)0x0000000FFFFFFFFFull) /* The low 36 of 64 bits */
766:
767:
768: /* ---------------------------------------------------------------------- */
769:
770: typedef struct uhash_ctx {
771: nh_ctx hash; /* Hash context for L1 NH hash */
772: UINT64 poly_key_8[STREAMS]; /* p64 poly keys */
773: UINT64 poly_accum[STREAMS]; /* poly hash result */
774: UINT64 ip_keys[STREAMS*4]; /* Inner-product keys */
775: UINT32 ip_trans[STREAMS]; /* Inner-product translation */
776: UINT32 msg_len; /* Total length of data passed */
777: /* to uhash */
778: } uhash_ctx;
779: typedef struct uhash_ctx *uhash_ctx_t;
780:
781: /* ---------------------------------------------------------------------- */
782:
783:
784: /* The polynomial hashes use Horner's rule to evaluate a polynomial one
785: * word at a time. As described in the specification, poly32 and poly64
786: * require keys from special domains. The following implementations exploit
787: * the special domains to avoid overflow. The results are not guaranteed to
788: * be within Z_p32 and Z_p64, but the Inner-Product hash implementation
789: * patches any errant values.
790: */
791:
792: static UINT64 poly64(UINT64 cur, UINT64 key, UINT64 data)
793: {
794: UINT32 key_hi = (UINT32)(key >> 32),
795: key_lo = (UINT32)key,
796: cur_hi = (UINT32)(cur >> 32),
797: cur_lo = (UINT32)cur,
798: x_lo,
799: x_hi;
800: UINT64 X,T,res;
801:
802: X = MUL64(key_hi, cur_lo) + MUL64(cur_hi, key_lo);
803: x_lo = (UINT32)X;
804: x_hi = (UINT32)(X >> 32);
805:
806: res = (MUL64(key_hi, cur_hi) + x_hi) * 59 + MUL64(key_lo, cur_lo);
807:
808: T = ((UINT64)x_lo << 32);
809: res += T;
810: if (res < T)
811: res += 59;
812:
813: res += data;
814: if (res < data)
815: res += 59;
816:
817: return res;
818: }
819:
820:
821: /* Although UMAC is specified to use a ramped polynomial hash scheme, this
822: * implementation does not handle all ramp levels. Because we don't handle
823: * the ramp up to p128 modulus in this implementation, we are limited to
824: * 2^14 poly_hash() invocations per stream (for a total capacity of 2^24
825: * bytes input to UMAC per tag, ie. 16MB).
826: */
827: static void poly_hash(uhash_ctx_t hc, UINT32 data_in[])
828: {
829: int i;
830: UINT64 *data=(UINT64*)data_in;
831:
832: for (i = 0; i < STREAMS; i++) {
833: if ((UINT32)(data[i] >> 32) == 0xfffffffful) {
834: hc->poly_accum[i] = poly64(hc->poly_accum[i],
835: hc->poly_key_8[i], p64 - 1);
836: hc->poly_accum[i] = poly64(hc->poly_accum[i],
837: hc->poly_key_8[i], (data[i] - 59));
838: } else {
839: hc->poly_accum[i] = poly64(hc->poly_accum[i],
840: hc->poly_key_8[i], data[i]);
841: }
842: }
843: }
844:
845:
846: /* ---------------------------------------------------------------------- */
847:
848:
849: /* The final step in UHASH is an inner-product hash. The poly hash
850: * produces a result not neccesarily WORD_LEN bytes long. The inner-
851: * product hash breaks the polyhash output into 16-bit chunks and
852: * multiplies each with a 36 bit key.
853: */
854:
855: static UINT64 ip_aux(UINT64 t, UINT64 *ipkp, UINT64 data)
856: {
857: t = t + ipkp[0] * (UINT64)(UINT16)(data >> 48);
858: t = t + ipkp[1] * (UINT64)(UINT16)(data >> 32);
859: t = t + ipkp[2] * (UINT64)(UINT16)(data >> 16);
860: t = t + ipkp[3] * (UINT64)(UINT16)(data);
861:
862: return t;
863: }
864:
865: static UINT32 ip_reduce_p36(UINT64 t)
866: {
867: /* Divisionless modular reduction */
868: UINT64 ret;
869:
870: ret = (t & m36) + 5 * (t >> 36);
871: if (ret >= p36)
872: ret -= p36;
873:
874: /* return least significant 32 bits */
875: return (UINT32)(ret);
876: }
877:
878:
879: /* If the data being hashed by UHASH is no longer than L1_KEY_LEN, then
880: * the polyhash stage is skipped and ip_short is applied directly to the
881: * NH output.
882: */
883: static void ip_short(uhash_ctx_t ahc, UINT8 *nh_res, u_char *res)
884: {
885: UINT64 t;
886: UINT64 *nhp = (UINT64 *)nh_res;
887:
888: t = ip_aux(0,ahc->ip_keys, nhp[0]);
889: STORE_UINT32_BIG((UINT32 *)res+0, ip_reduce_p36(t) ^ ahc->ip_trans[0]);
890: #if (UMAC_OUTPUT_LEN >= 8)
891: t = ip_aux(0,ahc->ip_keys+4, nhp[1]);
892: STORE_UINT32_BIG((UINT32 *)res+1, ip_reduce_p36(t) ^ ahc->ip_trans[1]);
893: #endif
894: #if (UMAC_OUTPUT_LEN >= 12)
895: t = ip_aux(0,ahc->ip_keys+8, nhp[2]);
896: STORE_UINT32_BIG((UINT32 *)res+2, ip_reduce_p36(t) ^ ahc->ip_trans[2]);
897: #endif
898: #if (UMAC_OUTPUT_LEN == 16)
899: t = ip_aux(0,ahc->ip_keys+12, nhp[3]);
900: STORE_UINT32_BIG((UINT32 *)res+3, ip_reduce_p36(t) ^ ahc->ip_trans[3]);
901: #endif
902: }
903:
904: /* If the data being hashed by UHASH is longer than L1_KEY_LEN, then
905: * the polyhash stage is not skipped and ip_long is applied to the
906: * polyhash output.
907: */
908: static void ip_long(uhash_ctx_t ahc, u_char *res)
909: {
910: int i;
911: UINT64 t;
912:
913: for (i = 0; i < STREAMS; i++) {
914: /* fix polyhash output not in Z_p64 */
915: if (ahc->poly_accum[i] >= p64)
916: ahc->poly_accum[i] -= p64;
917: t = ip_aux(0,ahc->ip_keys+(i*4), ahc->poly_accum[i]);
918: STORE_UINT32_BIG((UINT32 *)res+i,
919: ip_reduce_p36(t) ^ ahc->ip_trans[i]);
920: }
921: }
922:
923:
924: /* ---------------------------------------------------------------------- */
925:
926: /* ---------------------------------------------------------------------- */
927:
928: /* Reset uhash context for next hash session */
929: static int uhash_reset(uhash_ctx_t pc)
930: {
931: nh_reset(&pc->hash);
932: pc->msg_len = 0;
933: pc->poly_accum[0] = 1;
934: #if (UMAC_OUTPUT_LEN >= 8)
935: pc->poly_accum[1] = 1;
936: #endif
937: #if (UMAC_OUTPUT_LEN >= 12)
938: pc->poly_accum[2] = 1;
939: #endif
940: #if (UMAC_OUTPUT_LEN == 16)
941: pc->poly_accum[3] = 1;
942: #endif
943: return 1;
944: }
945:
946: /* ---------------------------------------------------------------------- */
947:
948: /* Given a pointer to the internal key needed by kdf() and a uhash context,
949: * initialize the NH context and generate keys needed for poly and inner-
950: * product hashing. All keys are endian adjusted in memory so that native
951: * loads cause correct keys to be in registers during calculation.
952: */
953: static void uhash_init(uhash_ctx_t ahc, aes_int_key prf_key)
954: {
955: int i;
956: UINT8 buf[(8*STREAMS+4)*sizeof(UINT64)];
957:
958: /* Zero the entire uhash context */
959: memset(ahc, 0, sizeof(uhash_ctx));
960:
961: /* Initialize the L1 hash */
962: nh_init(&ahc->hash, prf_key);
963:
964: /* Setup L2 hash variables */
965: kdf(buf, prf_key, 2, sizeof(buf)); /* Fill buffer with index 1 key */
966: for (i = 0; i < STREAMS; i++) {
967: /* Fill keys from the buffer, skipping bytes in the buffer not
968: * used by this implementation. Endian reverse the keys if on a
969: * little-endian computer.
970: */
971: memcpy(ahc->poly_key_8+i, buf+24*i, 8);
972: endian_convert_if_le(ahc->poly_key_8+i, 8, 8);
973: /* Mask the 64-bit keys to their special domain */
974: ahc->poly_key_8[i] &= ((UINT64)0x01ffffffu << 32) + 0x01ffffffu;
975: ahc->poly_accum[i] = 1; /* Our polyhash prepends a non-zero word */
976: }
977:
978: /* Setup L3-1 hash variables */
979: kdf(buf, prf_key, 3, sizeof(buf)); /* Fill buffer with index 2 key */
980: for (i = 0; i < STREAMS; i++)
981: memcpy(ahc->ip_keys+4*i, buf+(8*i+4)*sizeof(UINT64),
982: 4*sizeof(UINT64));
983: endian_convert_if_le(ahc->ip_keys, sizeof(UINT64),
984: sizeof(ahc->ip_keys));
985: for (i = 0; i < STREAMS*4; i++)
986: ahc->ip_keys[i] %= p36; /* Bring into Z_p36 */
987:
988: /* Setup L3-2 hash variables */
989: /* Fill buffer with index 4 key */
990: kdf(ahc->ip_trans, prf_key, 4, STREAMS * sizeof(UINT32));
991: endian_convert_if_le(ahc->ip_trans, sizeof(UINT32),
992: STREAMS * sizeof(UINT32));
993: }
994:
995: /* ---------------------------------------------------------------------- */
996:
997: #if 0
998: static uhash_ctx_t uhash_alloc(u_char key[])
999: {
1000: /* Allocate memory and force to a 16-byte boundary. */
1001: uhash_ctx_t ctx;
1002: u_char bytes_to_add;
1003: aes_int_key prf_key;
1004:
1005: ctx = (uhash_ctx_t)malloc(sizeof(uhash_ctx)+ALLOC_BOUNDARY);
1006: if (ctx) {
1007: if (ALLOC_BOUNDARY) {
1008: bytes_to_add = ALLOC_BOUNDARY -
1009: ((ptrdiff_t)ctx & (ALLOC_BOUNDARY -1));
1010: ctx = (uhash_ctx_t)((u_char *)ctx + bytes_to_add);
1011: *((u_char *)ctx - 1) = bytes_to_add;
1012: }
1013: aes_key_setup(key,prf_key);
1014: uhash_init(ctx, prf_key);
1015: }
1016: return (ctx);
1017: }
1018: #endif
1019:
1020: /* ---------------------------------------------------------------------- */
1021:
1022: #if 0
1023: static int uhash_free(uhash_ctx_t ctx)
1024: {
1025: /* Free memory allocated by uhash_alloc */
1026: u_char bytes_to_sub;
1027:
1028: if (ctx) {
1029: if (ALLOC_BOUNDARY) {
1030: bytes_to_sub = *((u_char *)ctx - 1);
1031: ctx = (uhash_ctx_t)((u_char *)ctx - bytes_to_sub);
1032: }
1033: free(ctx);
1034: }
1035: return (1);
1036: }
1037: #endif
1038: /* ---------------------------------------------------------------------- */
1039:
1040: static int uhash_update(uhash_ctx_t ctx, u_char *input, long len)
1041: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and
1042: * hash each one with NH, calling the polyhash on each NH output.
1043: */
1044: {
1045: UWORD bytes_hashed, bytes_remaining;
1.3 pvalchev 1046: UINT64 result_buf[STREAMS];
1047: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1048:
1049: if (ctx->msg_len + len <= L1_KEY_LEN) {
1050: nh_update(&ctx->hash, (UINT8 *)input, len);
1051: ctx->msg_len += len;
1052: } else {
1053:
1054: bytes_hashed = ctx->msg_len % L1_KEY_LEN;
1055: if (ctx->msg_len == L1_KEY_LEN)
1056: bytes_hashed = L1_KEY_LEN;
1057:
1058: if (bytes_hashed + len >= L1_KEY_LEN) {
1059:
1060: /* If some bytes have been passed to the hash function */
1061: /* then we want to pass at most (L1_KEY_LEN - bytes_hashed) */
1062: /* bytes to complete the current nh_block. */
1063: if (bytes_hashed) {
1064: bytes_remaining = (L1_KEY_LEN - bytes_hashed);
1065: nh_update(&ctx->hash, (UINT8 *)input, bytes_remaining);
1066: nh_final(&ctx->hash, nh_result);
1067: ctx->msg_len += bytes_remaining;
1068: poly_hash(ctx,(UINT32 *)nh_result);
1069: len -= bytes_remaining;
1070: input += bytes_remaining;
1071: }
1072:
1073: /* Hash directly from input stream if enough bytes */
1074: while (len >= L1_KEY_LEN) {
1075: nh(&ctx->hash, (UINT8 *)input, L1_KEY_LEN,
1076: L1_KEY_LEN, nh_result);
1077: ctx->msg_len += L1_KEY_LEN;
1078: len -= L1_KEY_LEN;
1079: input += L1_KEY_LEN;
1080: poly_hash(ctx,(UINT32 *)nh_result);
1081: }
1082: }
1083:
1084: /* pass remaining < L1_KEY_LEN bytes of input data to NH */
1085: if (len) {
1086: nh_update(&ctx->hash, (UINT8 *)input, len);
1087: ctx->msg_len += len;
1088: }
1089: }
1090:
1091: return (1);
1092: }
1093:
1094: /* ---------------------------------------------------------------------- */
1095:
1096: static int uhash_final(uhash_ctx_t ctx, u_char *res)
1097: /* Incorporate any pending data, pad, and generate tag */
1098: {
1.3 pvalchev 1099: UINT64 result_buf[STREAMS];
1100: UINT8 *nh_result = (UINT8 *)&result_buf;
1.1 pvalchev 1101:
1102: if (ctx->msg_len > L1_KEY_LEN) {
1103: if (ctx->msg_len % L1_KEY_LEN) {
1104: nh_final(&ctx->hash, nh_result);
1105: poly_hash(ctx,(UINT32 *)nh_result);
1106: }
1107: ip_long(ctx, res);
1108: } else {
1109: nh_final(&ctx->hash, nh_result);
1110: ip_short(ctx,nh_result, res);
1111: }
1112: uhash_reset(ctx);
1113: return (1);
1114: }
1115:
1116: /* ---------------------------------------------------------------------- */
1117:
1118: #if 0
1119: static int uhash(uhash_ctx_t ahc, u_char *msg, long len, u_char *res)
1120: /* assumes that msg is in a writable buffer of length divisible by */
1121: /* L1_PAD_BOUNDARY. Bytes beyond msg[len] may be zeroed. */
1122: {
1123: UINT8 nh_result[STREAMS*sizeof(UINT64)];
1124: UINT32 nh_len;
1125: int extra_zeroes_needed;
1126:
1127: /* If the message to be hashed is no longer than L1_HASH_LEN, we skip
1128: * the polyhash.
1129: */
1130: if (len <= L1_KEY_LEN) {
1131: if (len == 0) /* If zero length messages will not */
1132: nh_len = L1_PAD_BOUNDARY; /* be seen, comment out this case */
1133: else
1134: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1135: extra_zeroes_needed = nh_len - len;
1136: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1137: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1138: ip_short(ahc,nh_result, res);
1139: } else {
1140: /* Otherwise, we hash each L1_KEY_LEN chunk with NH, passing the NH
1141: * output to poly_hash().
1142: */
1143: do {
1144: nh(&ahc->hash, (UINT8 *)msg, L1_KEY_LEN, L1_KEY_LEN, nh_result);
1145: poly_hash(ahc,(UINT32 *)nh_result);
1146: len -= L1_KEY_LEN;
1147: msg += L1_KEY_LEN;
1148: } while (len >= L1_KEY_LEN);
1149: if (len) {
1150: nh_len = ((len + (L1_PAD_BOUNDARY - 1)) & ~(L1_PAD_BOUNDARY - 1));
1151: extra_zeroes_needed = nh_len - len;
1152: zero_pad((UINT8 *)msg + len, extra_zeroes_needed);
1153: nh(&ahc->hash, (UINT8 *)msg, nh_len, len, nh_result);
1154: poly_hash(ahc,(UINT32 *)nh_result);
1155: }
1156:
1157: ip_long(ahc, res);
1158: }
1159:
1160: uhash_reset(ahc);
1161: return 1;
1162: }
1163: #endif
1164:
1165: /* ---------------------------------------------------------------------- */
1166: /* ---------------------------------------------------------------------- */
1167: /* ----- Begin UMAC Section --------------------------------------------- */
1168: /* ---------------------------------------------------------------------- */
1169: /* ---------------------------------------------------------------------- */
1170:
1171: /* The UMAC interface has two interfaces, an all-at-once interface where
1172: * the entire message to be authenticated is passed to UMAC in one buffer,
1173: * and a sequential interface where the message is presented a little at a
1174: * time. The all-at-once is more optimaized than the sequential version and
1175: * should be preferred when the sequential interface is not required.
1176: */
1177: struct umac_ctx {
1178: uhash_ctx hash; /* Hash function for message compression */
1179: pdf_ctx pdf; /* PDF for hashed output */
1180: void *free_ptr; /* Address to free this struct via */
1181: } umac_ctx;
1182:
1183: /* ---------------------------------------------------------------------- */
1184:
1185: #if 0
1186: int umac_reset(struct umac_ctx *ctx)
1187: /* Reset the hash function to begin a new authentication. */
1188: {
1189: uhash_reset(&ctx->hash);
1190: return (1);
1191: }
1192: #endif
1193:
1194: /* ---------------------------------------------------------------------- */
1195:
1196: int umac_delete(struct umac_ctx *ctx)
1197: /* Deallocate the ctx structure */
1198: {
1199: if (ctx) {
1200: if (ALLOC_BOUNDARY)
1201: ctx = (struct umac_ctx *)ctx->free_ptr;
1.5 djm 1202: free(ctx);
1.1 pvalchev 1203: }
1204: return (1);
1205: }
1206:
1207: /* ---------------------------------------------------------------------- */
1208:
1209: struct umac_ctx *umac_new(u_char key[])
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:
1.2 stevesk 1218: octx = ctx = xmalloc(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;
1226: aes_key_setup(key,prf_key);
1227: pdf_init(&ctx->pdf, prf_key);
1228: uhash_init(&ctx->hash, prf_key);
1229: }
1230:
1231: return (ctx);
1232: }
1233:
1234: /* ---------------------------------------------------------------------- */
1235:
1236: int umac_final(struct umac_ctx *ctx, u_char tag[], u_char nonce[8])
1237: /* Incorporate any pending data, pad, and generate tag */
1238: {
1239: uhash_final(&ctx->hash, (u_char *)tag);
1240: pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1241:
1242: return (1);
1243: }
1244:
1245: /* ---------------------------------------------------------------------- */
1246:
1247: int umac_update(struct umac_ctx *ctx, u_char *input, long len)
1248: /* Given len bytes of data, we parse it into L1_KEY_LEN chunks and */
1249: /* hash each one, calling the PDF on the hashed output whenever the hash- */
1250: /* output buffer is full. */
1251: {
1252: uhash_update(&ctx->hash, input, len);
1253: return (1);
1254: }
1255:
1256: /* ---------------------------------------------------------------------- */
1257:
1258: #if 0
1259: int umac(struct umac_ctx *ctx, u_char *input,
1260: long len, u_char tag[],
1261: u_char nonce[8])
1262: /* All-in-one version simply calls umac_update() and umac_final(). */
1263: {
1264: uhash(&ctx->hash, input, len, (u_char *)tag);
1265: pdf_gen_xor(&ctx->pdf, (UINT8 *)nonce, (UINT8 *)tag);
1266:
1267: return (1);
1268: }
1269: #endif
1270:
1271: /* ---------------------------------------------------------------------- */
1272: /* ---------------------------------------------------------------------- */
1273: /* ----- End UMAC Section ----------------------------------------------- */
1274: /* ---------------------------------------------------------------------- */
1275: /* ---------------------------------------------------------------------- */