1 /*
2 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
3 * Use is subject to license terms.
4 */
5
6 #pragma ident "@(#)sha1.c 1.26 07/04/10 SMI"
7
8 /*
9 * The basic framework for this code came from the reference
10 * implementation for MD5. That implementation is Copyright (C)
11 * 1991-2, RSA Data Security, Inc. Created 1991. All rights reserved.
12 *
13 * License to copy and use this software is granted provided that it
14 * is identified as the "RSA Data Security, Inc. MD5 Message-Digest
15 * Algorithm" in all material mentioning or referencing this software
16 * or this function.
17 *
18 * License is also granted to make and use derivative works provided
19 * that such works are identified as "derived from the RSA Data
20 * Security, Inc. MD5 Message-Digest Algorithm" in all material
21 * mentioning or referencing the derived work.
22 *
23 * RSA Data Security, Inc. makes no representations concerning either
24 * the merchantability of this software or the suitability of this
25 * software for any particular purpose. It is provided "as is"
26 * without express or implied warranty of any kind.
27 *
28 * These notices must be retained in any copies of any part of this
29 * documentation and/or software.
30 *
31 * NOTE: Cleaned-up and optimized, version of SHA1, based on the FIPS 180-1
32 * standard, available at http://www.itl.nist.gov/div897/pubs/fip180-1.htm
33 * Not as fast as one would like -- further optimizations are encouraged
34 * and appreciated.
35 */
36
37 #include <sys/types.h>
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/sysmacros.h>
41 #include <sys/sha1.h>
42 #include <sys/sha1_consts.h>
43
44 #ifndef _KERNEL
45 #include <strings.h>
46 #include <stdlib.h>
47 #include <errno.h>
48 #include <sys/systeminfo.h>
49 #endif /* !_KERNEL */
50
51 static void Encode(uint8_t *, const uint32_t *, size_t);
52
53 #if defined(__sparc)
54
55 #define SHA1_TRANSFORM(ctx, in) \
56 SHA1Transform((ctx)->state[0], (ctx)->state[1], (ctx)->state[2], \
57 (ctx)->state[3], (ctx)->state[4], (ctx), (in))
58
59 static void SHA1Transform(uint32_t, uint32_t, uint32_t, uint32_t, uint32_t,
60 SHA1_CTX *, const uint8_t *);
61
62 #else
63
64 #define SHA1_TRANSFORM(ctx, in) SHA1Transform((ctx), (in))
65
66 static void SHA1Transform(SHA1_CTX *, const uint8_t *);
67
68 #endif
69
70
71 static uint8_t PADDING[64] = { 0x80, /* all zeros */ };
72
73 /*
74 * F, G, and H are the basic SHA1 functions.
75 */
76 #define F(b, c, d) (((b) & (c)) | ((~b) & (d)))
77 #define G(b, c, d) ((b) ^ (c) ^ (d))
78 #define H(b, c, d) (((b) & (c)) | (((b)|(c)) & (d)))
79
80 /*
81 * ROTATE_LEFT rotates x left n bits.
82 */
83
84 #if defined(__GNUC__) && defined(_LP64)
85 static __inline__ uint64_t
86 ROTATE_LEFT(uint64_t value, uint32_t n)
87 {
88 uint32_t t32;
89
90 t32 = (uint32_t)value;
91 return ((t32 << n) | (t32 >> (32 - n)));
92 }
93
94 #else
95
96 #define ROTATE_LEFT(x, n) \
97 (((x) << (n)) | ((x) >> ((sizeof (x) * NBBY)-(n))))
98
99 #endif
100
101 #if defined(__GNUC__) && (defined(__i386) || defined(__amd64))
102
103 #define HAVE_BSWAP
104
105 extern __inline__ uint32_t bswap(uint32_t value)
106 {
107 __asm__("bswap %0" : "+r" (value));
108 return (value);
109 }
110
111 #endif
112
113 /*
114 * SHA1Init()
115 *
116 * purpose: initializes the sha1 context and begins and sha1 digest operation
117 * input: SHA1_CTX * : the context to initializes.
118 * output: void
119 */
120
121 void
122 SHA1Init(SHA1_CTX *ctx)
123 {
124 ctx->count[0] = ctx->count[1] = 0;
125
126 /*
127 * load magic initialization constants. Tell lint
128 * that these constants are unsigned by using U.
129 */
130
131 ctx->state[0] = 0x67452301U;
132 ctx->state[1] = 0xefcdab89U;
133 ctx->state[2] = 0x98badcfeU;
134 ctx->state[3] = 0x10325476U;
135 ctx->state[4] = 0xc3d2e1f0U;
136 }
137
138 #ifdef VIS_SHA1
139 #ifdef _KERNEL
140
141 #include <sys/regset.h>
142 #include <sys/vis.h>
143 #include <sys/fpu/fpusystm.h>
144
145 /* the alignment for block stores to save fp registers */
146 #define VIS_ALIGN (64)
147
148 extern int sha1_savefp(kfpu_t *, int);
149 extern void sha1_restorefp(kfpu_t *);
150
151 uint32_t vis_sha1_svfp_threshold = 128;
152
153 #endif /* _KERNEL */
154
155 /*
156 * VIS SHA-1 consts.
157 */
158 static uint64_t VIS[] = {
159 0x8000000080000000ULL,
160 0x0002000200020002ULL,
161 0x5a8279996ed9eba1ULL,
162 0x8f1bbcdcca62c1d6ULL,
163 0x012389ab456789abULL};
164
165 extern void SHA1TransformVIS(uint64_t *, uint32_t *, uint32_t *, uint64_t *);
166
167
168 /*
169 * SHA1Update()
170 *
171 * purpose: continues an sha1 digest operation, using the message block
172 * to update the context.
173 * input: SHA1_CTX * : the context to update
174 * void * : the message block
175 * size_t : the length of the message block in bytes
176 * output: void
177 */
178
179 void
180 SHA1Update(SHA1_CTX *ctx, const void *inptr, size_t input_len)
181 {
182 uint32_t i, buf_index, buf_len;
183 uint64_t X0[40], input64[8];
184 const uint8_t *input = inptr;
185 #ifdef _KERNEL
186 int usevis = 0;
187 #else
188 int usevis = 1;
189 #endif /* _KERNEL */
190
191 /* check for noop */
192 if (input_len == 0)
193 return;
194
195 /* compute number of bytes mod 64 */
196 buf_index = (ctx->count[1] >> 3) & 0x3F;
197
198 /* update number of bits */
199 if ((ctx->count[1] += (input_len << 3)) < (input_len << 3))
200 ctx->count[0]++;
201
202 ctx->count[0] += (input_len >> 29);
203
204 buf_len = 64 - buf_index;
205
206 /* transform as many times as possible */
207 i = 0;
208 if (input_len >= buf_len) {
209 #ifdef _KERNEL
210 kfpu_t *fpu;
211 if (fpu_exists) {
212 uint8_t fpua[sizeof (kfpu_t) + GSR_SIZE + VIS_ALIGN];
213 uint32_t len = (input_len + buf_index) & ~0x3f;
214 int svfp_ok;
215
216 fpu = (kfpu_t *)P2ROUNDUP((uintptr_t)fpua, 64);
217 svfp_ok = ((len >= vis_sha1_svfp_threshold) ? 1 : 0);
218 usevis = fpu_exists && sha1_savefp(fpu, svfp_ok);
219 } else {
220 usevis = 0;
221 }
222 #endif /* _KERNEL */
223
224 /*
225 * general optimization:
226 *
227 * only do initial bcopy() and SHA1Transform() if
228 * buf_index != 0. if buf_index == 0, we're just
229 * wasting our time doing the bcopy() since there
230 * wasn't any data left over from a previous call to
231 * SHA1Update().
232 */
233
234 if (buf_index) {
235 bcopy(input, &ctx->buf_un.buf8[buf_index], buf_len);
236 if (usevis) {
237 SHA1TransformVIS(X0,
238 ctx->buf_un.buf32,
239 &ctx->state[0], VIS);
240 } else {
241 SHA1_TRANSFORM(ctx, ctx->buf_un.buf8);
242 }
243 i = buf_len;
244 }
245
246 /*
247 * VIS SHA-1: uses the VIS 1.0 instructions to accelerate
248 * SHA-1 processing. This is achieved by "offloading" the
249 * computation of the message schedule (MS) to the VIS units.
250 * This allows the VIS computation of the message schedule
251 * to be performed in parallel with the standard integer
252 * processing of the remainder of the SHA-1 computation.
253 * performance by up to around 1.37X, compared to an optimized
254 * integer-only implementation.
255 *
256 * The VIS implementation of SHA1Transform has a different API
257 * to the standard integer version:
258 *
259 * void SHA1TransformVIS(
260 * uint64_t *, // Pointer to MS for ith block
261 * uint32_t *, // Pointer to ith block of message data
262 * uint32_t *, // Pointer to SHA state i.e ctx->state
263 * uint64_t *, // Pointer to various VIS constants
264 * )
265 *
266 * Note: the message data must by 4-byte aligned.
267 *
268 * Function requires VIS 1.0 support.
269 *
270 * Handling is provided to deal with arbitrary byte alingment
271 * of the input data but the performance gains are reduced
272 * for alignments other than 4-bytes.
273 */
274 if (usevis) {
275 if (!IS_P2ALIGNED(&input[i], sizeof (uint32_t))) {
276 /*
277 * Main processing loop - input misaligned
278 */
279 for (; i + 63 < input_len; i += 64) {
280 bcopy(&input[i], input64, 64);
281 SHA1TransformVIS(X0, (uint32_t *)input64,
282 &ctx->state[0], VIS);
283 }
284 } else {
285 /*
286 * Main processing loop - input 8-byte aligned
287 */
288 for (; i + 63 < input_len; i += 64) {
289 SHA1TransformVIS(X0,
290 /* LINTED E_BAD_PTR_CAST_ALIGN */
291 (uint32_t *)&input[i],
292 &ctx->state[0], VIS);
293 }
294
295 }
296 #ifdef _KERNEL
297 sha1_restorefp(fpu);
298 #endif /* _KERNEL */
299 } else {
300 for (; i + 63 < input_len; i += 64) {
301 SHA1_TRANSFORM(ctx, &input[i]);
302 }
303 }
304
305 /*
306 * general optimization:
307 *
308 * if i and input_len are the same, return now instead
309 * of calling bcopy(), since the bcopy() in this case
310 * will be an expensive nop.
311 */
312
313 if (input_len == i)
314 return;
315
316 buf_index = 0;
317 }
318
319 /* buffer remaining input */
320 bcopy(&input[i], &ctx->buf_un.buf8[buf_index], input_len - i);
321 }
322
323 #else /* VIS_SHA1 */
324
325 void
326 SHA1Update(SHA1_CTX *ctx, const void *inptr, size_t input_len)
327 {
328 uint32_t i, buf_index, buf_len;
329 const uint8_t *input = inptr;
330
331 /* check for noop */
332 if (input_len == 0)
333 return;
334
335 /* compute number of bytes mod 64 */
336 buf_index = (ctx->count[1] >> 3) & 0x3F;
337
338 /* update number of bits */
339 if ((ctx->count[1] += (input_len << 3)) < (input_len << 3))
340 ctx->count[0]++;
341
342 ctx->count[0] += (input_len >> 29);
343
344 buf_len = 64 - buf_index;
345
346 /* transform as many times as possible */
347 i = 0;
348 if (input_len >= buf_len) {
349
350 /*
351 * general optimization:
352 *
353 * only do initial bcopy() and SHA1Transform() if
354 * buf_index != 0. if buf_index == 0, we're just
355 * wasting our time doing the bcopy() since there
356 * wasn't any data left over from a previous call to
357 * SHA1Update().
358 */
359
360 if (buf_index) {
361 bcopy(input, &ctx->buf_un.buf8[buf_index], buf_len);
362 SHA1_TRANSFORM(ctx, ctx->buf_un.buf8);
363 i = buf_len;
364 }
365
366 for (; i + 63 < input_len; i += 64)
367 SHA1_TRANSFORM(ctx, &input[i]);
368
369 /*
370 * general optimization:
371 *
372 * if i and input_len are the same, return now instead
373 * of calling bcopy(), since the bcopy() in this case
374 * will be an expensive nop.
375 */
376
377 if (input_len == i)
378 return;
379
380 buf_index = 0;
381 }
382
383 /* buffer remaining input */
384 bcopy(&input[i], &ctx->buf_un.buf8[buf_index], input_len - i);
385 }
386
387 #endif /* VIS_SHA1 */
388
389 /*
390 * SHA1Final()
391 *
392 * purpose: ends an sha1 digest operation, finalizing the message digest and
393 * zeroing the context.
394 * input: uchar_t * : a buffer to store the digest in
395 * : The function actually uses void* because many
396 * : callers pass things other than uchar_t here.
397 * SHA1_CTX * : the context to finalize, save, and zero
398 * output: void
399 */
400
401 void
402 SHA1Final(void *digest, SHA1_CTX *ctx)
403 {
404 uint8_t bitcount_be[sizeof (ctx->count)];
405 uint32_t index = (ctx->count[1] >> 3) & 0x3f;
406
407 /* store bit count, big endian */
408 Encode(bitcount_be, ctx->count, sizeof (bitcount_be));
409
410 /* pad out to 56 mod 64 */
411 SHA1Update(ctx, PADDING, ((index < 56) ? 56 : 120) - index);
412
413 /* append length (before padding) */
414 SHA1Update(ctx, bitcount_be, sizeof (bitcount_be));
415
416 /* store state in digest */
417 Encode(digest, ctx->state, sizeof (ctx->state));
418
419 /* zeroize sensitive information */
420 bzero(ctx, sizeof (*ctx));
421 }
422
423 typedef uint32_t sha1word;
424
425 /*
426 * sparc optimization:
427 *
428 * on the sparc, we can load big endian 32-bit data easily. note that
429 * special care must be taken to ensure the address is 32-bit aligned.
430 * in the interest of speed, we don't check to make sure, since
431 * careful programming can guarantee this for us.
432 */
433
434 #if defined(_BIG_ENDIAN)
435
436 #define LOAD_BIG_32(addr) (*(uint32_t *)(addr))
437
438 #else /* !defined(_BIG_ENDIAN) */
439
440 #if defined(HAVE_BSWAP)
441
442 #define LOAD_BIG_32(addr) bswap(*((uint32_t *)(addr)))
443
444 #else /* !defined(HAVE_BSWAP) */
445
446 /* little endian -- will work on big endian, but slowly */
447 #define LOAD_BIG_32(addr) \
448 (((addr)[0] << 24) | ((addr)[1] << 16) | ((addr)[2] << 8) | (addr)[3])
449
450 #endif /* !defined(HAVE_BSWAP) */
451
452 #endif /* !defined(_BIG_ENDIAN) */
453
454 /*
455 * SHA1Transform()
456 */
457 #if defined(W_ARRAY)
458 #define W(n) w[n]
459 #else /* !defined(W_ARRAY) */
460 #define W(n) w_ ## n
461 #endif /* !defined(W_ARRAY) */
462
463
464 #if defined(__sparc)
465
466 /*
467 * sparc register window optimization:
468 *
469 * `a', `b', `c', `d', and `e' are passed into SHA1Transform
470 * explicitly since it increases the number of registers available to
471 * the compiler. under this scheme, these variables can be held in
472 * %i0 - %i4, which leaves more local and out registers available.
473 *
474 * purpose: sha1 transformation -- updates the digest based on `block'
475 * input: uint32_t : bytes 1 - 4 of the digest
476 * uint32_t : bytes 5 - 8 of the digest
477 * uint32_t : bytes 9 - 12 of the digest
478 * uint32_t : bytes 12 - 16 of the digest
479 * uint32_t : bytes 16 - 20 of the digest
480 * SHA1_CTX * : the context to update
481 * uint8_t [64]: the block to use to update the digest
482 * output: void
483 */
484
485 void
486 SHA1Transform(uint32_t a, uint32_t b, uint32_t c, uint32_t d, uint32_t e,
487 SHA1_CTX *ctx, const uint8_t blk[64])
488 {
489 /*
490 * sparc optimization:
491 *
492 * while it is somewhat counter-intuitive, on sparc, it is
493 * more efficient to place all the constants used in this
494 * function in an array and load the values out of the array
495 * than to manually load the constants. this is because
496 * setting a register to a 32-bit value takes two ops in most
497 * cases: a `sethi' and an `or', but loading a 32-bit value
498 * from memory only takes one `ld' (or `lduw' on v9). while
499 * this increases memory usage, the compiler can find enough
500 * other things to do while waiting to keep the pipeline does
501 * not stall. additionally, it is likely that many of these
502 * constants are cached so that later accesses do not even go
503 * out to the bus.
504 *
505 * this array is declared `static' to keep the compiler from
506 * having to bcopy() this array onto the stack frame of
507 * SHA1Transform() each time it is called -- which is
508 * unacceptably expensive.
509 *
510 * the `const' is to ensure that callers are good citizens and
511 * do not try to munge the array. since these routines are
512 * going to be called from inside multithreaded kernelland,
513 * this is a good safety check. -- `sha1_consts' will end up in
514 * .rodata.
515 *
516 * unfortunately, loading from an array in this manner hurts
517 * performance under intel. so, there is a macro,
518 * SHA1_CONST(), used in SHA1Transform(), that either expands to
519 * a reference to this array, or to the actual constant,
520 * depending on what platform this code is compiled for.
521 */
522
523 static const uint32_t sha1_consts[] = {
524 SHA1_CONST_0, SHA1_CONST_1, SHA1_CONST_2, SHA1_CONST_3,
525 };
526
527 /*
528 * general optimization:
529 *
530 * use individual integers instead of using an array. this is a
531 * win, although the amount it wins by seems to vary quite a bit.
532 */
533
534 uint32_t w_0, w_1, w_2, w_3, w_4, w_5, w_6, w_7;
535 uint32_t w_8, w_9, w_10, w_11, w_12, w_13, w_14, w_15;
536
537 /*
538 * sparc optimization:
539 *
540 * if `block' is already aligned on a 4-byte boundary, use
541 * LOAD_BIG_32() directly. otherwise, bcopy() into a
542 * buffer that *is* aligned on a 4-byte boundary and then do
543 * the LOAD_BIG_32() on that buffer. benchmarks have shown
544 * that using the bcopy() is better than loading the bytes
545 * individually and doing the endian-swap by hand.
546 *
547 * even though it's quite tempting to assign to do:
548 *
549 * blk = bcopy(ctx->buf_un.buf32, blk, sizeof (ctx->buf_un.buf32));
550 *
551 * and only have one set of LOAD_BIG_32()'s, the compiler
552 * *does not* like that, so please resist the urge.
553 */
554
555 if ((uintptr_t)blk & 0x3) { /* not 4-byte aligned? */
556 bcopy(blk, ctx->buf_un.buf32, sizeof (ctx->buf_un.buf32));
557 w_15 = LOAD_BIG_32(ctx->buf_un.buf32 + 15);
558 w_14 = LOAD_BIG_32(ctx->buf_un.buf32 + 14);
559 w_13 = LOAD_BIG_32(ctx->buf_un.buf32 + 13);
560 w_12 = LOAD_BIG_32(ctx->buf_un.buf32 + 12);
561 w_11 = LOAD_BIG_32(ctx->buf_un.buf32 + 11);
562 w_10 = LOAD_BIG_32(ctx->buf_un.buf32 + 10);
563 w_9 = LOAD_BIG_32(ctx->buf_un.buf32 + 9);
564 w_8 = LOAD_BIG_32(ctx->buf_un.buf32 + 8);
565 w_7 = LOAD_BIG_32(ctx->buf_un.buf32 + 7);
566 w_6 = LOAD_BIG_32(ctx->buf_un.buf32 + 6);
567 w_5 = LOAD_BIG_32(ctx->buf_un.buf32 + 5);
568 w_4 = LOAD_BIG_32(ctx->buf_un.buf32 + 4);
569 w_3 = LOAD_BIG_32(ctx->buf_un.buf32 + 3);
570 w_2 = LOAD_BIG_32(ctx->buf_un.buf32 + 2);
571 w_1 = LOAD_BIG_32(ctx->buf_un.buf32 + 1);
572 w_0 = LOAD_BIG_32(ctx->buf_un.buf32 + 0);
573 } else {
574 /*LINTED*/
575 w_15 = LOAD_BIG_32(blk + 60);
576 /*LINTED*/
577 w_14 = LOAD_BIG_32(blk + 56);
578 /*LINTED*/
579 w_13 = LOAD_BIG_32(blk + 52);
580 /*LINTED*/
581 w_12 = LOAD_BIG_32(blk + 48);
582 /*LINTED*/
583 w_11 = LOAD_BIG_32(blk + 44);
584 /*LINTED*/
585 w_10 = LOAD_BIG_32(blk + 40);
586 /*LINTED*/
587 w_9 = LOAD_BIG_32(blk + 36);
588 /*LINTED*/
589 w_8 = LOAD_BIG_32(blk + 32);
590 /*LINTED*/
591 w_7 = LOAD_BIG_32(blk + 28);
592 /*LINTED*/
593 w_6 = LOAD_BIG_32(blk + 24);
594 /*LINTED*/
595 w_5 = LOAD_BIG_32(blk + 20);
596 /*LINTED*/
597 w_4 = LOAD_BIG_32(blk + 16);
598 /*LINTED*/
599 w_3 = LOAD_BIG_32(blk + 12);
600 /*LINTED*/
601 w_2 = LOAD_BIG_32(blk + 8);
602 /*LINTED*/
603 w_1 = LOAD_BIG_32(blk + 4);
604 /*LINTED*/
605 w_0 = LOAD_BIG_32(blk + 0);
606 }
607 #else /* !defined(__sparc) */
608
609 void
610 SHA1Transform(SHA1_CTX *ctx, const uint8_t blk[64])
611 {
612 sha1word a = ctx->state[0];
613 sha1word b = ctx->state[1];
614 sha1word c = ctx->state[2];
615 sha1word d = ctx->state[3];
616 sha1word e = ctx->state[4];
617
618 #if defined(W_ARRAY)
619 sha1word w[16];
620 #else /* !defined(W_ARRAY) */
621 sha1word w_0, w_1, w_2, w_3, w_4, w_5, w_6, w_7;
622 sha1word w_8, w_9, w_10, w_11, w_12, w_13, w_14, w_15;
623 #endif /* !defined(W_ARRAY) */
624
625 W(0) = LOAD_BIG_32(blk + 0);
626 W(1) = LOAD_BIG_32(blk + 4);
627 W(2) = LOAD_BIG_32(blk + 8);
628 W(3) = LOAD_BIG_32(blk + 12);
629 W(4) = LOAD_BIG_32(blk + 16);
630 W(5) = LOAD_BIG_32(blk + 20);
631 W(6) = LOAD_BIG_32(blk + 24);
632 W(7) = LOAD_BIG_32(blk + 28);
633 W(8) = LOAD_BIG_32(blk + 32);
634 W(9) = LOAD_BIG_32(blk + 36);
635 W(10) = LOAD_BIG_32(blk + 40);
636 W(11) = LOAD_BIG_32(blk + 44);
637 W(12) = LOAD_BIG_32(blk + 48);
638 W(13) = LOAD_BIG_32(blk + 52);
639 W(14) = LOAD_BIG_32(blk + 56);
640 W(15) = LOAD_BIG_32(blk + 60);
641
642 #endif /* !defined(__sparc) */
643
644 /*
645 * general optimization:
646 *
647 * even though this approach is described in the standard as
648 * being slower algorithmically, it is 30-40% faster than the
649 * "faster" version under SPARC, because this version has more
650 * of the constraints specified at compile-time and uses fewer
651 * variables (and therefore has better register utilization)
652 * than its "speedier" brother. (i've tried both, trust me)
653 *
654 * for either method given in the spec, there is an "assignment"
655 * phase where the following takes place:
656 *
657 * tmp = (main_computation);
658 * e = d; d = c; c = rotate_left(b, 30); b = a; a = tmp;
659 *
660 * we can make the algorithm go faster by not doing this work,
661 * but just pretending that `d' is now `e', etc. this works
662 * really well and obviates the need for a temporary variable.
663 * however, we still explictly perform the rotate action,
664 * since it is cheaper on SPARC to do it once than to have to
665 * do it over and over again.
666 */
667
668 /* round 1 */
669 e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(0) + SHA1_CONST(0); /* 0 */
670 b = ROTATE_LEFT(b, 30);
671
672 d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(1) + SHA1_CONST(0); /* 1 */
673 a = ROTATE_LEFT(a, 30);
674
675 c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(2) + SHA1_CONST(0); /* 2 */
676 e = ROTATE_LEFT(e, 30);
677
678 b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(3) + SHA1_CONST(0); /* 3 */
679 d = ROTATE_LEFT(d, 30);
680
681 a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(4) + SHA1_CONST(0); /* 4 */
682 c = ROTATE_LEFT(c, 30);
683
684 e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(5) + SHA1_CONST(0); /* 5 */
685 b = ROTATE_LEFT(b, 30);
686
687 d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(6) + SHA1_CONST(0); /* 6 */
688 a = ROTATE_LEFT(a, 30);
689
690 c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(7) + SHA1_CONST(0); /* 7 */
691 e = ROTATE_LEFT(e, 30);
692
693 b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(8) + SHA1_CONST(0); /* 8 */
694 d = ROTATE_LEFT(d, 30);
695
696 a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(9) + SHA1_CONST(0); /* 9 */
697 c = ROTATE_LEFT(c, 30);
698
699 e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(10) + SHA1_CONST(0); /* 10 */
700 b = ROTATE_LEFT(b, 30);
701
702 d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(11) + SHA1_CONST(0); /* 11 */
703 a = ROTATE_LEFT(a, 30);
704
705 c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(12) + SHA1_CONST(0); /* 12 */
706 e = ROTATE_LEFT(e, 30);
707
708 b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(13) + SHA1_CONST(0); /* 13 */
709 d = ROTATE_LEFT(d, 30);
710
711 a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(14) + SHA1_CONST(0); /* 14 */
712 c = ROTATE_LEFT(c, 30);
713
714 e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(15) + SHA1_CONST(0); /* 15 */
715 b = ROTATE_LEFT(b, 30);
716
717 W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1); /* 16 */
718 d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(0) + SHA1_CONST(0);
719 a = ROTATE_LEFT(a, 30);
720
721 W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1); /* 17 */
722 c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(1) + SHA1_CONST(0);
723 e = ROTATE_LEFT(e, 30);
724
725 W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1); /* 18 */
726 b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(2) + SHA1_CONST(0);
727 d = ROTATE_LEFT(d, 30);
728
729 W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1); /* 19 */
730 a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(3) + SHA1_CONST(0);
731 c = ROTATE_LEFT(c, 30);
732
733 /* round 2 */
734 W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1); /* 20 */
735 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(4) + SHA1_CONST(1);
736 b = ROTATE_LEFT(b, 30);
737
738 W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1); /* 21 */
739 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(5) + SHA1_CONST(1);
740 a = ROTATE_LEFT(a, 30);
741
742 W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1); /* 22 */
743 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(6) + SHA1_CONST(1);
744 e = ROTATE_LEFT(e, 30);
745
746 W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1); /* 23 */
747 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(7) + SHA1_CONST(1);
748 d = ROTATE_LEFT(d, 30);
749
750 W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1); /* 24 */
751 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(8) + SHA1_CONST(1);
752 c = ROTATE_LEFT(c, 30);
753
754 W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1); /* 25 */
755 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(9) + SHA1_CONST(1);
756 b = ROTATE_LEFT(b, 30);
757
758 W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1); /* 26 */
759 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(10) + SHA1_CONST(1);
760 a = ROTATE_LEFT(a, 30);
761
762 W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1); /* 27 */
763 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(11) + SHA1_CONST(1);
764 e = ROTATE_LEFT(e, 30);
765
766 W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1); /* 28 */
767 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(12) + SHA1_CONST(1);
768 d = ROTATE_LEFT(d, 30);
769
770 W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1); /* 29 */
771 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(13) + SHA1_CONST(1);
772 c = ROTATE_LEFT(c, 30);
773
774 W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1); /* 30 */
775 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(14) + SHA1_CONST(1);
776 b = ROTATE_LEFT(b, 30);
777
778 W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1); /* 31 */
779 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(15) + SHA1_CONST(1);
780 a = ROTATE_LEFT(a, 30);
781
782 W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1); /* 32 */
783 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(0) + SHA1_CONST(1);
784 e = ROTATE_LEFT(e, 30);
785
786 W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1); /* 33 */
787 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(1) + SHA1_CONST(1);
788 d = ROTATE_LEFT(d, 30);
789
790 W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1); /* 34 */
791 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(2) + SHA1_CONST(1);
792 c = ROTATE_LEFT(c, 30);
793
794 W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1); /* 35 */
795 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(3) + SHA1_CONST(1);
796 b = ROTATE_LEFT(b, 30);
797
798 W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1); /* 36 */
799 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(4) + SHA1_CONST(1);
800 a = ROTATE_LEFT(a, 30);
801
802 W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1); /* 37 */
803 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(5) + SHA1_CONST(1);
804 e = ROTATE_LEFT(e, 30);
805
806 W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1); /* 38 */
807 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(6) + SHA1_CONST(1);
808 d = ROTATE_LEFT(d, 30);
809
810 W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1); /* 39 */
811 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(7) + SHA1_CONST(1);
812 c = ROTATE_LEFT(c, 30);
813
814 /* round 3 */
815 W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1); /* 40 */
816 e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(8) + SHA1_CONST(2);
817 b = ROTATE_LEFT(b, 30);
818
819 W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1); /* 41 */
820 d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(9) + SHA1_CONST(2);
821 a = ROTATE_LEFT(a, 30);
822
823 W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1); /* 42 */
824 c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(10) + SHA1_CONST(2);
825 e = ROTATE_LEFT(e, 30);
826
827 W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1); /* 43 */
828 b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(11) + SHA1_CONST(2);
829 d = ROTATE_LEFT(d, 30);
830
831 W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1); /* 44 */
832 a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(12) + SHA1_CONST(2);
833 c = ROTATE_LEFT(c, 30);
834
835 W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1); /* 45 */
836 e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(13) + SHA1_CONST(2);
837 b = ROTATE_LEFT(b, 30);
838
839 W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1); /* 46 */
840 d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(14) + SHA1_CONST(2);
841 a = ROTATE_LEFT(a, 30);
842
843 W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1); /* 47 */
844 c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(15) + SHA1_CONST(2);
845 e = ROTATE_LEFT(e, 30);
846
847 W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1); /* 48 */
848 b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(0) + SHA1_CONST(2);
849 d = ROTATE_LEFT(d, 30);
850
851 W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1); /* 49 */
852 a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(1) + SHA1_CONST(2);
853 c = ROTATE_LEFT(c, 30);
854
855 W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1); /* 50 */
856 e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(2) + SHA1_CONST(2);
857 b = ROTATE_LEFT(b, 30);
858
859 W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1); /* 51 */
860 d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(3) + SHA1_CONST(2);
861 a = ROTATE_LEFT(a, 30);
862
863 W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1); /* 52 */
864 c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(4) + SHA1_CONST(2);
865 e = ROTATE_LEFT(e, 30);
866
867 W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1); /* 53 */
868 b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(5) + SHA1_CONST(2);
869 d = ROTATE_LEFT(d, 30);
870
871 W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1); /* 54 */
872 a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(6) + SHA1_CONST(2);
873 c = ROTATE_LEFT(c, 30);
874
875 W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1); /* 55 */
876 e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(7) + SHA1_CONST(2);
877 b = ROTATE_LEFT(b, 30);
878
879 W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1); /* 56 */
880 d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(8) + SHA1_CONST(2);
881 a = ROTATE_LEFT(a, 30);
882
883 W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1); /* 57 */
884 c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(9) + SHA1_CONST(2);
885 e = ROTATE_LEFT(e, 30);
886
887 W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1); /* 58 */
888 b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(10) + SHA1_CONST(2);
889 d = ROTATE_LEFT(d, 30);
890
891 W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1); /* 59 */
892 a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(11) + SHA1_CONST(2);
893 c = ROTATE_LEFT(c, 30);
894
895 /* round 4 */
896 W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1); /* 60 */
897 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(12) + SHA1_CONST(3);
898 b = ROTATE_LEFT(b, 30);
899
900 W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1); /* 61 */
901 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(13) + SHA1_CONST(3);
902 a = ROTATE_LEFT(a, 30);
903
904 W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1); /* 62 */
905 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(14) + SHA1_CONST(3);
906 e = ROTATE_LEFT(e, 30);
907
908 W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1); /* 63 */
909 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(15) + SHA1_CONST(3);
910 d = ROTATE_LEFT(d, 30);
911
912 W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1); /* 64 */
913 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(0) + SHA1_CONST(3);
914 c = ROTATE_LEFT(c, 30);
915
916 W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1); /* 65 */
917 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(1) + SHA1_CONST(3);
918 b = ROTATE_LEFT(b, 30);
919
920 W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1); /* 66 */
921 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(2) + SHA1_CONST(3);
922 a = ROTATE_LEFT(a, 30);
923
924 W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1); /* 67 */
925 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(3) + SHA1_CONST(3);
926 e = ROTATE_LEFT(e, 30);
927
928 W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1); /* 68 */
929 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(4) + SHA1_CONST(3);
930 d = ROTATE_LEFT(d, 30);
931
932 W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1); /* 69 */
933 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(5) + SHA1_CONST(3);
934 c = ROTATE_LEFT(c, 30);
935
936 W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1); /* 70 */
937 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(6) + SHA1_CONST(3);
938 b = ROTATE_LEFT(b, 30);
939
940 W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1); /* 71 */
941 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(7) + SHA1_CONST(3);
942 a = ROTATE_LEFT(a, 30);
943
944 W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1); /* 72 */
945 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(8) + SHA1_CONST(3);
946 e = ROTATE_LEFT(e, 30);
947
948 W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1); /* 73 */
949 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(9) + SHA1_CONST(3);
950 d = ROTATE_LEFT(d, 30);
951
952 W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1); /* 74 */
953 a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(10) + SHA1_CONST(3);
954 c = ROTATE_LEFT(c, 30);
955
956 W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1); /* 75 */
957 e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(11) + SHA1_CONST(3);
958 b = ROTATE_LEFT(b, 30);
959
960 W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1); /* 76 */
961 d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(12) + SHA1_CONST(3);
962 a = ROTATE_LEFT(a, 30);
963
964 W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1); /* 77 */
965 c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(13) + SHA1_CONST(3);
966 e = ROTATE_LEFT(e, 30);
967
968 W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1); /* 78 */
969 b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(14) + SHA1_CONST(3);
970 d = ROTATE_LEFT(d, 30);
971
972 W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1); /* 79 */
973
974 ctx->state[0] += ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(15) +
975 SHA1_CONST(3);
976 ctx->state[1] += b;
977 ctx->state[2] += ROTATE_LEFT(c, 30);
978 ctx->state[3] += d;
979 ctx->state[4] += e;
980
981 /* zeroize sensitive information */
982 W(0) = W(1) = W(2) = W(3) = W(4) = W(5) = W(6) = W(7) = W(8) = 0;
983 W(9) = W(10) = W(11) = W(12) = W(13) = W(14) = W(15) = 0;
984 }
985
986 /*
987 * Encode()
988 *
989 * purpose: to convert a list of numbers from little endian to big endian
990 * input: uint8_t * : place to store the converted big endian numbers
991 * uint32_t * : place to get numbers to convert from
992 * size_t : the length of the input in bytes
993 * output: void
994 */
995
996 static void
997 Encode(uint8_t *_RESTRICT_KYWD output, const uint32_t *_RESTRICT_KYWD input,
998 size_t len)
999 {
1000 size_t i, j;
1001
1002 #if defined(__sparc)
1003 if (IS_P2ALIGNED(output, sizeof (uint32_t))) {
1004 for (i = 0, j = 0; j < len; i++, j += 4) {
1005 /* LINTED: pointer alignment */
1006 *((uint32_t *)(output + j)) = input[i];
1007 }
1008 } else {
1009 #endif /* little endian -- will work on big endian, but slowly */
1010 for (i = 0, j = 0; j < len; i++, j += 4) {
1011 output[j] = (input[i] >> 24) & 0xff;
1012 output[j + 1] = (input[i] >> 16) & 0xff;
1013 output[j + 2] = (input[i] >> 8) & 0xff;
1014 output[j + 3] = input[i] & 0xff;
1015 }
1016 #if defined(__sparc)
1017 }
1018 #endif
1019 }