chiark / gitweb /
symm/: New SSE2 implementations of Salsa20 and ChaCha.
[catacomb] / symm / chacha.c
1 /* -*-c-*-
2  *
3  * ChaCha stream cipher
4  *
5  * (c) 2015 Straylight/Edgeware
6  */
7
8 /*----- Licensing notice --------------------------------------------------*
9  *
10  * This file is part of Catacomb.
11  *
12  * Catacomb is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU Library General Public License as
14  * published by the Free Software Foundation; either version 2 of the
15  * License, or (at your option) any later version.
16  *
17  * Catacomb is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU Library General Public License for more details.
21  *
22  * You should have received a copy of the GNU Library General Public
23  * License along with Catacomb; if not, write to the Free
24  * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
25  * MA 02111-1307, USA.
26  */
27
28 /*----- Header files ------------------------------------------------------*/
29
30 #include "config.h"
31
32 #include <stdarg.h>
33
34 #include <mLib/bits.h>
35
36 #include "arena.h"
37 #include "chacha.h"
38 #include "chacha-core.h"
39 #include "dispatch.h"
40 #include "gcipher.h"
41 #include "grand.h"
42 #include "keysz.h"
43 #include "paranoia.h"
44
45 /*----- Global variables --------------------------------------------------*/
46
47 const octet chacha_keysz[] = { KSZ_SET, 32, 16, 10, 0 };
48
49 /*----- The ChaCha core function and utilities ----------------------------*/
50
51 /* --- @core@ --- *
52  *
53  * Arguments:   @unsigned r@ = number of rounds
54  *              @const chacha_matrix src@ = input matrix
55  *              @chacha_matrix dest@ = where to put the output
56  *
57  * Returns:     ---
58  *
59  *
60  * Use:         Apply the ChaCha/r core function to @src@, writing the
61  *              result to @dest@.  This consists of @r@ rounds followed by
62  *              the feedforward step.
63  */
64
65 CPU_DISPATCH(static, (void),
66              void, core, (unsigned r, const chacha_matrix src,
67                           chacha_matrix dest),
68              (r, src, dest),
69              pick_core, simple_core);
70
71 static void simple_core(unsigned r, const chacha_matrix src,
72                         chacha_matrix dest)
73   { CHACHA_nR(dest, src, r); CHACHA_FFWD(dest, src); }
74
75 #ifdef CPUFAM_X86
76 extern core__functype chacha_core_x86_sse2;
77 #endif
78
79 static core__functype *pick_core(void)
80 {
81 #ifdef CPUFAM_X86
82   if (cpu_feature_p(CPUFEAT_X86_SSE2)) return chacha_core_x86_sse2;
83 #endif
84   return simple_core;
85 }
86
87 /* --- @populate@ --- *
88  *
89  * Arguments:   @chacha_matrix a@ = a matrix to fill in
90  *              @const void *key@ = pointer to key material
91  *              @size_t ksz@ = size of key
92  *
93  * Returns:     ---
94  *
95  * Use:         Fills in a ChaCha matrix from the key, setting the
96  *              appropriate constants according to the key length.  The nonce
97  *              and position words are left uninitialized.
98  */
99
100 static void populate(chacha_matrix a, const void *key, size_t ksz)
101 {
102   const octet *k = key;
103
104   KSZ_ASSERT(chacha, ksz);
105
106   a[ 4] = LOAD32_L(k +  0);
107   a[ 5] = LOAD32_L(k +  4);
108   if (ksz == 10) {
109     a[ 6] = LOAD16_L(k +  8);
110     a[ 7] = 0;
111   } else {
112     a[ 6] = LOAD32_L(k +  8);
113     a[ 7] = LOAD32_L(k + 12);
114   }
115   if (ksz <= 16) {
116     a[ 8] = a[ 4];
117     a[ 9] = a[ 5];
118     a[10] = a[ 6];
119     a[11] = a[ 7];
120     a[ 0] = CHACHA_A128;
121     a[ 1] = CHACHA_B128;
122     a[ 2] = ksz == 10 ? CHACHA_C80 : CHACHA_C128;
123     a[ 3] = CHACHA_D128;
124   } else {
125     a[ 8] = LOAD32_L(k + 16);
126     a[ 9] = LOAD32_L(k + 20);
127     a[10] = LOAD32_L(k + 24);
128     a[11] = LOAD32_L(k + 28);
129     a[ 0] = CHACHA_A256;
130     a[ 1] = CHACHA_B256;
131     a[ 2] = CHACHA_C256;
132     a[ 3] = CHACHA_D256;
133   }
134 }
135
136 /*----- ChaCha implementation ---------------------------------------------*/
137
138 /* --- @chacha_init@ --- *
139  *
140  * Arguments:   @chacha_ctx *ctx@ = context to fill in
141  *              @const void *key@ = pointer to key material
142  *              @size_t ksz@ = size of key (either 32 or 16)
143  *              @const void *nonce@ = initial nonce, or null
144  *
145  * Returns:     ---
146  *
147  * Use:         Initializes a ChaCha context ready for use.
148  */
149
150 void chacha_init(chacha_ctx *ctx, const void *key, size_t ksz,
151                   const void *nonce)
152 {
153   static const octet zerononce[CHACHA_NONCESZ];
154
155   populate(ctx->a, key, ksz);
156   chacha_setnonce(ctx, nonce ? nonce : zerononce);
157 }
158
159 /* --- @chacha_setnonce@ --- *
160  *
161  * Arguments:   @chacha_ctx *ctx@ = pointer to context
162  *              @const void *nonce@ = the nonce (@CHACHA_NONCESZ@ bytes)
163  *
164  * Returns:     ---
165  *
166  * Use:         Set a new nonce in the context @ctx@, e.g., for processing a
167  *              different message.  The stream position is reset to zero (see
168  *              @chacha_seek@ etc.).
169  */
170
171 void chacha_setnonce(chacha_ctx *ctx, const void *nonce)
172 {
173   const octet *n = nonce;
174
175   ctx->a[14] = LOAD32_L(n + 0);
176   ctx->a[15] = LOAD32_L(n + 4);
177   chacha_seek(ctx, 0);
178 }
179
180 /* --- @chacha_seek@, @chacha_seeku64@ --- *
181  *
182  * Arguments:   @chacha_ctx *ctx@ = pointer to context
183  *              @unsigned long i@, @kludge64 i@ = new position to set
184  *
185  * Returns:     ---
186  *
187  * Use:         Sets a new stream position, in units of Chacha output
188  *              blocks, which are @CHACHA_OUTSZ@ bytes each.  Byte
189  *              granularity can be achieved by calling @chachaR_encrypt@
190  *              appropriately.
191  */
192
193 void chacha_seek(chacha_ctx *ctx, unsigned long i)
194   { kludge64 ii; ASSIGN64(ii, i); chacha_seeku64(ctx, ii); }
195
196 void chacha_seeku64(chacha_ctx *ctx, kludge64 i)
197 {
198   ctx->a[12] = LO64(i); ctx->a[13] = HI64(i);
199   ctx->bufi = CHACHA_OUTSZ;
200 }
201
202 /* --- @chacha_tell@, @chacha_tellu64@ --- *
203  *
204  * Arguments:   @chacha_ctx *ctx@ = pointer to context
205  *
206  * Returns:     The current position in the output stream, in blocks,
207  *              rounding upwards.
208  */
209
210 unsigned long chacha_tell(chacha_ctx *ctx)
211   { kludge64 i = chacha_tellu64(ctx); return (GET64(unsigned long, i)); }
212
213 kludge64 chacha_tellu64(chacha_ctx *ctx)
214   { kludge64 i; SET64(i, ctx->a[9], ctx->a[8]); return (i); }
215
216 /* --- @chacha{,12,8}_encrypt@ --- *
217  *
218  * Arguments:   @chacha_ctx *ctx@ = pointer to context
219  *              @const void *src@ = source buffer (or null)
220  *              @void *dest@ = destination buffer (or null)
221  *              @size_t sz@ = size of the buffers
222  *
223  * Returns:     ---
224  *
225  * Use:         Encrypts or decrypts @sz@ bytes of data from @src@ to @dest@.
226  *              ChaCha works by XORing plaintext with a keystream, so
227  *              encryption and decryption are the same operation.  If @dest@
228  *              is null then ignore @src@ and skip @sz@ bytes of the
229  *              keystream.  If @src@ is null, then just write the keystream
230  *              to @dest@.
231  */
232
233 #define CHACHA_ENCRYPT(r, ctx, src, dest, sz)                           \
234   chacha##r##_encrypt(ctx, src, dest, sz)
235 #define DEFENCRYPT(r)                                                   \
236   void CHACHA_ENCRYPT(r, chacha_ctx *ctx, const void *src,              \
237                       void *dest, size_t sz)                            \
238   {                                                                     \
239     chacha_matrix b;                                                    \
240     const octet *s = src;                                               \
241     octet *d = dest;                                                    \
242     size_t n;                                                           \
243     kludge64 pos, delta;                                                \
244                                                                         \
245     SALSA20_OUTBUF(ctx, d, s, sz);                                      \
246     if (!sz) return;                                                    \
247                                                                         \
248     if (!dest) {                                                        \
249       n = sz/CHACHA_OUTSZ;                                              \
250       pos = chacha_tellu64(ctx);                                        \
251       ASSIGN64(delta, n);                                               \
252       ADD64(pos, pos, delta);                                           \
253       chacha_seeku64(ctx, pos);                                         \
254       sz = sz%CHACHA_OUTSZ;                                             \
255     } else if (!src) {                                                  \
256       while (sz >= CHACHA_OUTSZ) {                                      \
257         core(r, ctx->a, b);                                             \
258         CHACHA_STEP(ctx->a);                                            \
259         SALSA20_GENFULL(b, d);                                          \
260         sz -= CHACHA_OUTSZ;                                             \
261       }                                                                 \
262     } else {                                                            \
263       while (sz >= CHACHA_OUTSZ) {                                      \
264         core(r, ctx->a, b);                                             \
265         CHACHA_STEP(ctx->a);                                            \
266         SALSA20_MIXFULL(b, d, s);                                       \
267         sz -= CHACHA_OUTSZ;                                             \
268       }                                                                 \
269     }                                                                   \
270                                                                         \
271     if (sz) {                                                           \
272       core(r, ctx->a, b);                                               \
273       CHACHA_STEP(ctx->a);                                              \
274       SALSA20_PREPBUF(ctx, b);                                          \
275       SALSA20_OUTBUF(ctx, d, s, sz);                                    \
276       assert(!sz);                                                      \
277     }                                                                   \
278   }
279 CHACHA_VARS(DEFENCRYPT)
280
281 /*----- HChaCha implementation --------------------------------------------*/
282
283 #define HCHACHA_RAW(r, ctx, src, dest) hchacha##r##_raw(ctx, src, dest)
284 #define HCHACHA_PRF(r, ctx, src, dest) hchacha##r##_prf(ctx, src, dest)
285
286 /* --- @hchacha{20,12,8}_prf@ --- *
287  *
288  * Arguments:   @chacha_ctx *ctx@ = pointer to context
289  *              @const void *src@ = the input (@HCHACHA_INSZ@ bytes)
290  *              @void *dest@ = the output (@HCHACHA_OUTSZ@ bytes)
291  *
292  * Returns:     ---
293  *
294  * Use:         Apply the HChacha/r pseudorandom function to @src@, writing
295  *              the result to @out@.
296  */
297
298 #define DEFHCHACHA(r)                                                   \
299   static void HCHACHA_RAW(r, chacha_matrix k,                           \
300                           const uint32 *src, uint32 *dest)              \
301   {                                                                     \
302     chacha_matrix a;                                                    \
303     int i;                                                              \
304                                                                         \
305     /* --- HChaCha, computed from full ChaCha --- *                     \
306      *                                                                  \
307      * The security proof makes use of the fact that HChaCha (i.e.,     \
308      * without the final feedforward step) can be computed from full    \
309      * ChaCha using only knowledge of the non-secret input.  I don't    \
310      * want to compromise the performance of the main function by       \
311      * making the feedforward step separate, but this operation is less \
312      * speed critical, so we do it the harder way.                      \
313      */                                                                 \
314                                                                         \
315     for (i = 0; i < 4; i++) k[12 + i] = src[i];                         \
316     core(r, k, a);                                                      \
317     for (i = 0; i < 8; i++) dest[i] = a[(i + 4)^4] - k[(i + 4)^4];      \
318   }                                                                     \
319                                                                         \
320   void HCHACHA_PRF(r, chacha_ctx *ctx, const void *src, void *dest)     \
321   {                                                                     \
322     const octet *s = src;                                               \
323     octet *d = dest;                                                    \
324     uint32 in[4], out[8];                                               \
325     int i;                                                              \
326                                                                         \
327     for (i = 0; i < 4; i++) in[i] = LOAD32_L(s + 4*i);                  \
328     HCHACHA_RAW(r, ctx->a, in, out);                                    \
329     for (i = 0; i < 8; i++) STORE32_L(d + 4*i, out[i]);                 \
330   }
331 CHACHA_VARS(DEFHCHACHA)
332
333 /*----- XChaCha implementation -------------------------------------------*/
334
335 /* --- Some convenient macros for naming functions --- *
336  *
337  * Because the crypto core is involved in XChaCha/r's per-nonce setup, we
338  * need to take an interest in the number of rounds in most of the various
339  * functions, and it will probably help if we distinguish the context
340  * structures for the various versions.
341  */
342
343 #define XCHACHA_CTX(r) xchacha##r##_ctx
344 #define XCHACHA_INIT(r, ctx, k, ksz, n) xchacha##r##_init(ctx, k, ksz, n)
345 #define XCHACHA_SETNONCE(r, ctx, n) xchacha##r##_setnonce(ctx, n)
346 #define XCHACHA_SEEK(r, ctx, i) xchacha##r##_seek(ctx, i)
347 #define XCHACHA_SEEKU64(r, ctx, i) xchacha##r##_seeku64(ctx, i)
348 #define XCHACHA_TELL(r, ctx) xchacha##r##_tell(ctx)
349 #define XCHACHA_TELLU64(r, ctx) xchacha##r##_tellu64(ctx)
350 #define XCHACHA_ENCRYPT(r, ctx, src, dest, sz)                          \
351   xchacha##r##_encrypt(ctx, src, dest, sz)
352
353 /* --- @xchacha{20,12,8}_init@ --- *
354  *
355  * Arguments:   @xchachaR_ctx *ctx@ = the context to fill in
356  *              @const void *key@ = pointer to key material
357  *              @size_t ksz@ = size of key (either 32 or 16)
358  *              @const void *nonce@ = initial nonce, or null
359  *
360  * Returns:     ---
361  *
362  * Use:         Initializes an XChaCha/r context ready for use.
363  *
364  *              There is a different function for each number of rounds,
365  *              unlike for plain ChaCha.
366  */
367
368 #define DEFXINIT(r)                                                     \
369   void XCHACHA_INIT(r, XCHACHA_CTX(r) *ctx,                             \
370                     const void *key, size_t ksz, const void *nonce)     \
371   {                                                                     \
372     static const octet zerononce[XCHACHA_NONCESZ];                      \
373                                                                         \
374     populate(ctx->k, key, ksz);                                         \
375     ctx->s.a[ 0] = CHACHA_A256;                                         \
376     ctx->s.a[ 1] = CHACHA_B256;                                         \
377     ctx->s.a[ 2] = CHACHA_C256;                                         \
378     ctx->s.a[ 3] = CHACHA_D256;                                         \
379     XCHACHA_SETNONCE(r, ctx, nonce ? nonce : zerononce);                \
380   }
381 CHACHA_VARS(DEFXINIT)
382
383 /* --- @xchacha{20,12,8}_setnonce@ --- *
384  *
385  * Arguments:   @xchachaR_ctx *ctx@ = pointer to context
386  *              @const void *nonce@ = the nonce (@XCHACHA_NONCESZ@ bytes)
387  *
388  * Returns:     ---
389  *
390  * Use:         Set a new nonce in the context @ctx@, e.g., for processing a
391  *              different message.  The stream position is reset to zero (see
392  *              @chacha_seek@ etc.).
393  *
394  *              There is a different function for each number of rounds,
395  *              unlike for plain ChaCha.
396  */
397
398 #define DEFXNONCE(r)                                                    \
399   void XCHACHA_SETNONCE(r, XCHACHA_CTX(r) *ctx, const void *nonce)      \
400   {                                                                     \
401     const octet *n = nonce;                                             \
402     uint32 in[4];                                                       \
403     int i;                                                              \
404                                                                         \
405     for (i = 0; i < 4; i++) in[i] = LOAD32_L(n + 4*i);                  \
406     HCHACHA_RAW(r, ctx->k, in, ctx->s.a + 4);                           \
407     chacha_setnonce(&ctx->s, n + 16);                                   \
408   }
409 CHACHA_VARS(DEFXNONCE)
410
411 /* --- @xchacha{20,12,8}_seek@, @xchacha{20,12,8}_seeku64@ --- *
412  *
413  * Arguments:   @xchachaR_ctx *ctx@ = pointer to context
414  *              @unsigned long i@, @kludge64 i@ = new position to set
415  *
416  * Returns:     ---
417  *
418  * Use:         Sets a new stream position, in units of ChaCha output
419  *              blocks, which are @XCHACHA_OUTSZ@ bytes each.  Byte
420  *              granularity can be achieved by calling @xchachaR_encrypt@
421  *              appropriately.
422  *
423  *              There is a different function for each number of rounds,
424  *              unlike for plain ChaCha, because the context structures are
425  *              different.
426  */
427
428 /* --- @xchacha{20,12,8}_tell@, @xchacha{20,12,8}_tellu64@ --- *
429  *
430  * Arguments:   @chacha_ctx *ctx@ = pointer to context
431  *
432  * Returns:     The current position in the output stream, in blocks,
433  *              rounding upwards.
434  *
435  *              There is a different function for each number of rounds,
436  *              unlike for plain ChaCha, because the context structures are
437  *              different.
438  */
439
440 /* --- @xchacha{,12,8}_encrypt@ --- *
441  *
442  * Arguments:   @xchachaR_ctx *ctx@ = pointer to context
443  *              @const void *src@ = source buffer (or null)
444  *              @void *dest@ = destination buffer (or null)
445  *              @size_t sz@ = size of the buffers
446  *
447  * Returns:     ---
448  *
449  * Use:         Encrypts or decrypts @sz@ bytes of data from @src@ to @dest@.
450  *              XChaCha works by XORing plaintext with a keystream, so
451  *              encryption and decryption are the same operation.  If @dest@
452  *              is null then ignore @src@ and skip @sz@ bytes of the
453  *              keystream.  If @src@ is null, then just write the keystream
454  *              to @dest@.
455  */
456
457 #define DEFXPASSTHRU(r)                                                 \
458   void XCHACHA_SEEK(r, XCHACHA_CTX(r) *ctx, unsigned long i)            \
459     { chacha_seek(&ctx->s, i); }                                        \
460   void XCHACHA_SEEKU64(r, XCHACHA_CTX(r) *ctx, kludge64 i)              \
461     { chacha_seeku64(&ctx->s, i); }                                     \
462   unsigned long XCHACHA_TELL(r, XCHACHA_CTX(r) *ctx)                    \
463     { return chacha_tell(&ctx->s); }                                    \
464   kludge64 XCHACHA_TELLU64(r, XCHACHA_CTX(r) *ctx)                      \
465     { return chacha_tellu64(&ctx->s); }                                 \
466   void XCHACHA_ENCRYPT(r, XCHACHA_CTX(r) *ctx,                          \
467                         const void *src, void *dest, size_t sz)         \
468     { CHACHA_ENCRYPT(r, &ctx->s, src, dest, sz); }
469 CHACHA_VARS(DEFXPASSTHRU)
470
471 /*----- Generic cipher interface ------------------------------------------*/
472
473 typedef struct gctx { gcipher c; chacha_ctx ctx; } gctx;
474
475 static void gsetiv(gcipher *c, const void *iv)
476   { gctx *g = (gctx *)c; chacha_setnonce(&g->ctx, iv); }
477
478 static void gdestroy(gcipher *c)
479   { gctx *g = (gctx *)c; BURN(*g); S_DESTROY(g); }
480
481 #define DEFGCIPHER(r)                                                   \
482                                                                         \
483   static const gcipher_ops gops_##r;                                    \
484                                                                         \
485   static gcipher *ginit_##r(const void *k, size_t sz)                   \
486   {                                                                     \
487     gctx *g = S_CREATE(gctx);                                           \
488     g->c.ops = &gops_##r;                                               \
489     chacha_init(&g->ctx, k, sz, 0);                                     \
490     return (&g->c);                                                     \
491   }                                                                     \
492                                                                         \
493   static void gencrypt_##r(gcipher *c, const void *s,                   \
494                            void *t, size_t sz)                          \
495     { gctx *g = (gctx *)c; CHACHA_ENCRYPT(r, &g->ctx, s, t, sz); }      \
496                                                                         \
497   static const gcipher_ops gops_##r = {                                 \
498     &chacha##r,                                                         \
499     gencrypt_##r, gencrypt_##r, gdestroy, gsetiv, 0                     \
500   };                                                                    \
501                                                                         \
502   const gccipher chacha##r = {                                          \
503     "chacha" #r, chacha_keysz,                                          \
504     CHACHA_NONCESZ, ginit_##r                                           \
505   };
506
507 CHACHA_VARS(DEFGCIPHER)
508
509 #define DEFGXCIPHER(r)                                                  \
510                                                                         \
511   typedef struct { gcipher c; XCHACHA_CTX(r) ctx; } gxctx_##r;          \
512                                                                         \
513   static void gxsetiv_##r(gcipher *c, const void *iv)                   \
514     { gxctx_##r *g = (gxctx_##r *)c; XCHACHA_SETNONCE(r, &g->ctx, iv); } \
515                                                                         \
516   static void gxdestroy_##r(gcipher *c)                                 \
517     { gxctx_##r *g = (gxctx_##r *)c; BURN(*g); S_DESTROY(g); }          \
518                                                                         \
519   static const gcipher_ops gxops_##r;                                   \
520                                                                         \
521   static gcipher *gxinit_##r(const void *k, size_t sz)                  \
522   {                                                                     \
523     gxctx_##r *g = S_CREATE(gxctx_##r);                                 \
524     g->c.ops = &gxops_##r;                                              \
525     XCHACHA_INIT(r, &g->ctx, k, sz, 0);                                 \
526     return (&g->c);                                                     \
527   }                                                                     \
528                                                                         \
529   static void gxencrypt_##r(gcipher *c, const void *s,                  \
530                             void *t, size_t sz)                         \
531   {                                                                     \
532     gxctx_##r *g = (gxctx_##r *)c;                                      \
533     XCHACHA_ENCRYPT(r, &g->ctx, s, t, sz);                              \
534   }                                                                     \
535                                                                         \
536   static const gcipher_ops gxops_##r = {                                \
537     &xchacha##r,                                                        \
538     gxencrypt_##r, gxencrypt_##r, gxdestroy_##r, gxsetiv_##r, 0         \
539   };                                                                    \
540                                                                         \
541   const gccipher xchacha##r = {                                         \
542     "xchacha" #r, chacha_keysz,                                         \
543     CHACHA_NONCESZ, gxinit_##r                                          \
544   };
545
546 CHACHA_VARS(DEFGXCIPHER)
547
548 /*----- Generic random number generator interface -------------------------*/
549
550 typedef struct grops {
551   size_t noncesz;
552   void (*seek)(void *, kludge64);
553   kludge64 (*tell)(void *);
554   void (*setnonce)(void *, const void *);
555   void (*generate)(void *, void *, size_t);
556 } grops;
557
558 typedef struct grbasectx {
559   grand r;
560   const grops *ops;
561 } grbasectx;
562
563 static int grmisc(grand *r, unsigned op, ...)
564 {
565   octet buf[XCHACHA_NONCESZ];
566   grbasectx *g = (grbasectx *)r;
567   grand *rr;
568   const octet *p;
569   size_t sz;
570   uint32 i;
571   unsigned long ul;
572   kludge64 pos;
573   va_list ap;
574   int rc = 0;
575
576   va_start(ap, op);
577
578   switch (op) {
579     case GRAND_CHECK:
580       switch (va_arg(ap, unsigned)) {
581         case GRAND_CHECK:
582         case GRAND_SEEDINT:
583         case GRAND_SEEDUINT32:
584         case GRAND_SEEDBLOCK:
585         case GRAND_SEEDRAND:
586         case CHACHA_SEEK:
587         case CHACHA_SEEKU64:
588         case CHACHA_TELL:
589         case CHACHA_TELLU64:
590           rc = 1;
591           break;
592         default:
593           rc = 0;
594           break;
595       }
596       break;
597
598     case GRAND_SEEDINT:
599       i = va_arg(ap, unsigned); STORE32_L(buf, i);
600       memset(buf + 4, 0, g->ops->noncesz - 4);
601       g->ops->setnonce(g, buf);
602       break;
603     case GRAND_SEEDUINT32:
604       i = va_arg(ap, uint32); STORE32_L(buf, i);
605       memset(buf + 4, 0, g->ops->noncesz - 4);
606       g->ops->setnonce(g, buf);
607       break;
608     case GRAND_SEEDBLOCK:
609       p = va_arg(ap, const void *);
610       sz = va_arg(ap, size_t);
611       if (sz < g->ops->noncesz) {
612         memcpy(buf, p, sz);
613         memset(buf + sz, 0, g->ops->noncesz - sz);
614         p = buf;
615       }
616       g->ops->setnonce(g, p);
617       break;
618     case GRAND_SEEDRAND:
619       rr = va_arg(ap, grand *);
620       rr->ops->fill(rr, buf, g->ops->noncesz);
621       g->ops->setnonce(g, buf);
622       break;
623     case CHACHA_SEEK:
624       ul = va_arg(ap, unsigned long); ASSIGN64(pos, ul);
625       g->ops->seek(g, pos);
626       break;
627     case CHACHA_SEEKU64:
628       pos = va_arg(ap, kludge64);
629       g->ops->seek(g, pos);
630       break;
631     case CHACHA_TELL:
632       pos = g->ops->tell(g);
633       *va_arg(ap, unsigned long *) = GET64(unsigned long, pos);
634       break;
635     case CHACHA_TELLU64:
636       *va_arg(ap, kludge64 *) = g->ops->tell(g);
637       break;
638     default:
639       GRAND_BADOP;
640       break;
641   }
642
643   return (rc);
644 }
645
646 static octet grbyte(grand *r)
647 {
648   grbasectx *g = (grbasectx *)r;
649   octet o;
650   g->ops->generate(g, &o, 1);
651   return (o);
652 }
653
654 static uint32 grword(grand *r)
655 {
656   grbasectx *g = (grbasectx *)r;
657   octet b[4];
658   g->ops->generate(g, b, sizeof(b));
659   return (LOAD32_L(b));
660 }
661
662 static void grfill(grand *r, void *p, size_t sz)
663 {
664   grbasectx *g = (grbasectx *)r;
665   g->ops->generate(r, p, sz);
666 }
667
668 typedef struct grctx {
669   grbasectx r;
670   chacha_ctx ctx;
671 } grctx;
672
673 static void gr_seek(void *r, kludge64 pos)
674   { grctx *g = r; chacha_seeku64(&g->ctx, pos); }
675
676 static kludge64 gr_tell(void *r)
677   { grctx *g = r; return (chacha_tellu64(&g->ctx)); }
678
679 static void gr_setnonce(void *r, const void *n)
680   { grctx *g = r; chacha_setnonce(&g->ctx, n); }
681
682 static void grdestroy(grand *r)
683   { grctx *g = (grctx *)r; BURN(*g); S_DESTROY(g); }
684
685 #define DEFGRAND(rr)                                                    \
686                                                                         \
687   static void gr_generate_##rr(void *r, void *b, size_t sz)             \
688     { grctx *g = r; CHACHA_ENCRYPT(rr, &g->ctx, 0, b, sz); }            \
689                                                                         \
690   static const grops grops_##rr =                                       \
691     { CHACHA_NONCESZ, gr_seek, gr_tell,                                 \
692       gr_setnonce, gr_generate_##rr };                                  \
693                                                                         \
694   static const grand_ops grops_rand_##rr = {                            \
695     "chacha" #rr, GRAND_CRYPTO, 0,                                      \
696     grmisc, grdestroy, grword,                                          \
697     grbyte, grword, grand_range, grfill                                 \
698   };                                                                    \
699                                                                         \
700   grand *chacha##rr##_rand(const void *k, size_t ksz, const void *n)    \
701   {                                                                     \
702     grctx *g = S_CREATE(g);                                             \
703     g->r.r.ops = &grops_rand_##rr;                                      \
704     g->r.ops = &grops_##rr;                                             \
705     chacha_init(&g->ctx, k, ksz, n);                                    \
706     return (&g->r.r);                                                   \
707   }
708 CHACHA_VARS(DEFGRAND)
709
710 #define DEFXGRAND(rr)                                                   \
711                                                                         \
712   typedef struct grxctx_##rr {                                          \
713     grbasectx r;                                                        \
714     XCHACHA_CTX(rr) ctx;                                                \
715   } grxctx_##rr;                                                        \
716                                                                         \
717   static void grx_seek_##rr(void *r, kludge64 pos)                      \
718     { grxctx_##rr *g = r; XCHACHA_SEEKU64(rr, &g->ctx, pos); }          \
719                                                                         \
720   static kludge64 grx_tell_##rr(void *r)                                \
721     { grxctx_##rr *g = r; return (XCHACHA_TELLU64(rr, &g->ctx)); }      \
722                                                                         \
723   static void grx_setnonce_##rr(void *r, const void *n)                 \
724     { grxctx_##rr *g = r; XCHACHA_SETNONCE(rr, &g->ctx, n); }           \
725                                                                         \
726   static void grxdestroy_##rr(grand *r)                                 \
727     { grxctx_##rr *g = (grxctx_##rr *)r; BURN(*g); S_DESTROY(g); }      \
728                                                                         \
729   static void grx_generate_##rr(void *r, void *b, size_t sz)            \
730     { grxctx_##rr *g = r; XCHACHA_ENCRYPT(rr, &g->ctx, 0, b, sz); }     \
731                                                                         \
732   static const grops grxops_##rr =                                      \
733     { XCHACHA_NONCESZ, grx_seek_##rr, grx_tell_##rr,                    \
734       grx_setnonce_##rr, grx_generate_##rr };                           \
735                                                                         \
736   static const grand_ops grxops_rand_##rr = {                           \
737     "xchacha" #rr, GRAND_CRYPTO, 0,                                     \
738     grmisc, grxdestroy_##rr, grword,                                    \
739     grbyte, grword, grand_range, grfill                                 \
740   };                                                                    \
741                                                                         \
742   grand *xchacha##rr##_rand(const void *k, size_t ksz, const void *n)   \
743   {                                                                     \
744     grxctx_##rr *g = S_CREATE(g);                                       \
745     g->r.r.ops = &grxops_rand_##rr;                                     \
746     g->r.ops = &grxops_##rr;                                            \
747     XCHACHA_INIT(rr, &g->ctx, k, ksz, n);                               \
748     return (&g->r.r);                                                   \
749   }
750 CHACHA_VARS(DEFXGRAND)
751
752 /*----- Test rig ----------------------------------------------------------*/
753
754 #ifdef TEST_RIG
755
756 #include <stdio.h>
757 #include <string.h>
758
759 #include <mLib/quis.h>
760 #include <mLib/testrig.h>
761
762 #define DEFVCORE(r)                                                     \
763   static int v_core_##r(dstr *v)                                        \
764   {                                                                     \
765     chacha_matrix a, b;                                                 \
766     dstr d = DSTR_INIT;                                                 \
767     int i, n;                                                           \
768     int ok = 1;                                                         \
769                                                                         \
770     DENSURE(&d, CHACHA_OUTSZ); d.len = CHACHA_OUTSZ;                    \
771     n = *(int *)v[0].buf;                                               \
772     for (i = 0; i < CHACHA_OUTSZ/4; i++)                                \
773       a[i] = LOAD32_L(v[1].buf + 4*i);                                  \
774     for (i = 0; i < n; i++) {                                           \
775       core(r, a, b);                                                    \
776       memcpy(a, b, sizeof(a));                                          \
777     }                                                                   \
778     for (i = 0; i < CHACHA_OUTSZ/4; i++) STORE32_L(d.buf + 4*i, a[i]);  \
779                                                                         \
780     if (d.len != v[2].len || memcmp(d.buf, v[2].buf, v[2].len) != 0) {  \
781       ok = 0;                                                           \
782       printf("\nfail core:"                                             \
783              "\n\titerations = %d"                                      \
784              "\n\tin       = ", n);                                     \
785       type_hex.dump(&v[1], stdout);                                     \
786       printf("\n\texpected   = ");                                      \
787       type_hex.dump(&v[2], stdout);                                     \
788       printf("\n\tcalculated = ");                                      \
789       type_hex.dump(&d, stdout);                                        \
790       putchar('\n');                                                    \
791     }                                                                   \
792                                                                         \
793     dstr_destroy(&d);                                                   \
794     return (ok);                                                        \
795   }
796 CHACHA_VARS(DEFVCORE)
797
798 #define CHACHA_CTX(r) chacha_ctx
799 #define CHACHA_INIT(r, ctx, k, ksz, n) chacha_init(ctx, k, ksz, n)
800 #define CHACHA_SEEKU64(r, ctx, i) chacha_seeku64(ctx, i)
801 #define XCHACHA_SEEKU64(r, ctx, i) xchacha##r##_seeku64(ctx, i)
802
803 #define DEFxVENC(base, BASE, r)                                         \
804   static int v_encrypt_##base##_##r(dstr *v)                            \
805   {                                                                     \
806     BASE##_CTX(r) ctx;                                                  \
807     dstr d = DSTR_INIT;                                                 \
808     kludge64 pos;                                                       \
809     const octet *p, *p0;                                                \
810     octet *q;                                                           \
811     size_t sz, sz0, step;                                               \
812     unsigned long skip;                                                 \
813     int ok = 1;                                                         \
814                                                                         \
815     if (v[4].len) { p0 = (const octet *)v[4].buf; sz0 = v[4].len; }     \
816     else { p0 = 0; sz0 = v[5].len; }                                    \
817     DENSURE(&d, sz0); d.len = sz0;                                      \
818     skip = *(unsigned long *)v[3].buf;                                  \
819                                                                         \
820     step = 0;                                                           \
821     while (step < sz0 + skip) {                                         \
822       step = step ? 3*step + 4 : 1;                                     \
823       if (step > sz0 + skip) step = sz0 + skip;                         \
824       BASE##_INIT(r, &ctx, v[0].buf, v[0].len, v[1].buf);               \
825       if (v[2].len) {                                                   \
826         LOAD64_(pos, v[2].buf);                                         \
827         BASE##_SEEKU64(r, &ctx, pos);                                   \
828       }                                                                 \
829                                                                         \
830       for (sz = skip; sz >= step; sz -= step)                           \
831         BASE##_ENCRYPT(r, &ctx, 0, 0, step);                            \
832       if (sz) BASE##_ENCRYPT(r, &ctx, 0, 0, sz);                        \
833       for (p = p0, q = (octet *)d.buf, sz = sz0;                        \
834            sz >= step;                                                  \
835            sz -= step, q += step) {                                     \
836         BASE##_ENCRYPT(r, &ctx, p, q, step);                            \
837         if (p) p += step;                                               \
838       }                                                                 \
839       if (sz) BASE##_ENCRYPT(r, &ctx, p, q, sz);                        \
840                                                                         \
841       if (d.len != v[5].len || memcmp(d.buf, v[5].buf, v[5].len) != 0) { \
842         ok = 0;                                                         \
843         printf("\nfail encrypt:"                                        \
844                "\n\tstep           = %lu"                               \
845                "\n\tkey    = ", (unsigned long)step);                   \
846         type_hex.dump(&v[0], stdout);                                   \
847         printf("\n\tnonce          = ");                                \
848         type_hex.dump(&v[1], stdout);                                   \
849         printf("\n\tposition   = ");                                    \
850         type_hex.dump(&v[2], stdout);                                   \
851         printf("\n\tskip           = %lu", skip);                       \
852         printf("\n\tmessage    = ");                                    \
853         type_hex.dump(&v[4], stdout);                                   \
854         printf("\n\texpected   = ");                                    \
855         type_hex.dump(&v[5], stdout);                                   \
856         printf("\n\tcalculated = ");                                    \
857         type_hex.dump(&d, stdout);                                      \
858         putchar('\n');                                                  \
859       }                                                                 \
860     }                                                                   \
861                                                                         \
862     dstr_destroy(&d);                                                   \
863     return (ok);                                                        \
864   }
865 #define DEFVENC(r) DEFxVENC(chacha, CHACHA, r)
866 #define DEFXVENC(r) DEFxVENC(xchacha, XCHACHA, r)
867 CHACHA_VARS(DEFVENC)
868 CHACHA_VARS(DEFXVENC)
869
870 static test_chunk defs[] = {
871 #define DEFxTAB(base, r)                                                \
872   { #base #r, v_encrypt_##base##_##r,                                   \
873     { &type_hex, &type_hex, &type_hex, &type_ulong,                     \
874       &type_hex, &type_hex, 0 } },
875 #define DEFTAB(r)                                                       \
876   { "chacha" #r "-core", v_core_##r,                                    \
877     { &type_int, &type_hex, &type_hex, 0 } },                           \
878   DEFxTAB(chacha, r)
879 #define DEFXTAB(r) DEFxTAB(xchacha, r)
880 CHACHA_VARS(DEFTAB)
881 CHACHA_VARS(DEFXTAB)
882   { 0, 0, { 0 } }
883 };
884
885 int main(int argc, char *argv[])
886 {
887   test_run(argc, argv, defs, SRCDIR"/t/chacha");
888   return (0);
889 }
890
891 #endif
892
893 /*----- That's all, folks -------------------------------------------------*/