1 /* 1
2 * VMAC: Message Authentication Code using Uni
3 *
4 * Reference: https://tools.ietf.org/html/draf
5 *
6 * Copyright (c) 2009, Intel Corporation.
7 * Copyright (c) 2018, Google Inc.
8 *
9 * This program is free software; you can redi
10 * under the terms and conditions of the GNU G
11 * version 2, as published by the Free Softwar
12 *
13 * This program is distributed in the hope it
14 * ANY WARRANTY; without even the implied warr
15 * FITNESS FOR A PARTICULAR PURPOSE. See the
16 * more details.
17 *
18 * You should have received a copy of the GNU
19 * this program; if not, write to the Free Sof
20 * Place - Suite 330, Boston, MA 02111-1307 US
21 */
22
23 /*
24 * Derived from:
25 * VMAC and VHASH Implementation by Ted K
26 * This implementation is herby placed in
27 * The authors offers no warranty. Use at
28 * Last modified: 17 APR 08, 1700 PDT
29 */
30
31 #include <linux/unaligned.h>
32 #include <linux/init.h>
33 #include <linux/types.h>
34 #include <linux/crypto.h>
35 #include <linux/module.h>
36 #include <linux/scatterlist.h>
37 #include <asm/byteorder.h>
38 #include <crypto/scatterwalk.h>
39 #include <crypto/internal/cipher.h>
40 #include <crypto/internal/hash.h>
41
42 /*
43 * User definable settings.
44 */
45 #define VMAC_TAG_LEN 64
46 #define VMAC_KEY_SIZE 128/* Must be 128, 192
47 #define VMAC_KEY_LEN (VMAC_KEY_SIZE/8)
48 #define VMAC_NHBYTES 128/* Must 2^i for any
49 #define VMAC_NONCEBYTES 16
50
51 /* per-transform (per-key) context */
52 struct vmac_tfm_ctx {
53 struct crypto_cipher *cipher;
54 u64 nhkey[(VMAC_NHBYTES/8)+2*(VMAC_TAG
55 u64 polykey[2*VMAC_TAG_LEN/64];
56 u64 l3key[2*VMAC_TAG_LEN/64];
57 };
58
59 /* per-request context */
60 struct vmac_desc_ctx {
61 union {
62 u8 partial[VMAC_NHBYTES];
63 __le64 partial_words[VMAC_NHBY
64 };
65 unsigned int partial_size; /* siz
66 bool first_block_processed;
67 u64 polytmp[2*VMAC_TAG_LEN/64]; /* run
68 union {
69 u8 bytes[VMAC_NONCEBYTES];
70 __be64 pads[VMAC_NONCEBYTES /
71 } nonce;
72 unsigned int nonce_size; /* nonce byte
73 };
74
75 /*
76 * Constants and masks
77 */
78 #define UINT64_C(x) x##ULL
79 static const u64 p64 = UINT64_C(0xffffffffff
80 static const u64 m62 = UINT64_C(0x3fffffffff
81 static const u64 m63 = UINT64_C(0x7fffffffff
82 static const u64 m64 = UINT64_C(0xffffffffff
83 static const u64 mpoly = UINT64_C(0x1fffffff1f
84
85 #define pe64_to_cpup le64_to_cpup
86
87 #ifdef __LITTLE_ENDIAN
88 #define INDEX_HIGH 1
89 #define INDEX_LOW 0
90 #else
91 #define INDEX_HIGH 0
92 #define INDEX_LOW 1
93 #endif
94
95 /*
96 * The following routines are used in this imp
97 * written via macros to simulate zero-overhea
98 *
99 * MUL64: 64x64->128-bit multiplication
100 * PMUL64: assumes top bits cleared on inputs
101 * ADD128: 128x128->128-bit addition
102 */
103
104 #define ADD128(rh, rl, ih, il)
105 do {
106 u64 _il = (il);
107 (rl) += (_il);
108 if ((rl) < (_il))
109 (rh)++;
110 (rh) += (ih);
111 } while (0)
112
113 #define MUL32(i1, i2) ((u64)(u32)(i1)*(u32)(
114
115 #define PMUL64(rh, rl, i1, i2) /* Assumes m d
116 do {
117 u64 _i1 = (i1), _i2 = (i2);
118 u64 m = MUL32(_i1, _i2>>32) +
119 rh = MUL32(_i1>>32, _i2>>32);
120 rl = MUL32(_i1, _i2);
121 ADD128(rh, rl, (m >> 32), (m <
122 } while (0)
123
124 #define MUL64(rh, rl, i1, i2)
125 do {
126 u64 _i1 = (i1), _i2 = (i2);
127 u64 m1 = MUL32(_i1, _i2>>32);
128 u64 m2 = MUL32(_i1>>32, _i2);
129 rh = MUL32(_i1>>32, _i2>>32);
130 rl = MUL32(_i1, _i2);
131 ADD128(rh, rl, (m1 >> 32), (m1
132 ADD128(rh, rl, (m2 >> 32), (m2
133 } while (0)
134
135 /*
136 * For highest performance the L1 NH and L2 po
137 * carefully implemented to take advantage of
138 * Here these two hash functions are defined m
139 * 64-bit architectures, once for 32-bit SSE2
140 * for the rest (32-bit) architectures.
141 * For each, nh_16 *must* be defined (works on
142 * Optionally, nh_vmac_nhbytes can be defined
143 * VMAC_NHBYTES), and nh_16_2 and nh_vmac_nhby
144 * NH computations at once).
145 */
146
147 #ifdef CONFIG_64BIT
148
149 #define nh_16(mp, kp, nw, rh, rl)
150 do {
151 int i; u64 th, tl;
152 rh = rl = 0;
153 for (i = 0; i < nw; i += 2) {
154 MUL64(th, tl, pe64_to_
155 pe64_to_cpup((
156 ADD128(rh, rl, th, tl)
157 }
158 } while (0)
159
160 #define nh_16_2(mp, kp, nw, rh, rl, rh1, rl1)
161 do {
162 int i; u64 th, tl;
163 rh1 = rl1 = rh = rl = 0;
164 for (i = 0; i < nw; i += 2) {
165 MUL64(th, tl, pe64_to_
166 pe64_to_cpup((
167 ADD128(rh, rl, th, tl)
168 MUL64(th, tl, pe64_to_
169 pe64_to_cpup((
170 ADD128(rh1, rl1, th, t
171 }
172 } while (0)
173
174 #if (VMAC_NHBYTES >= 64) /* These versions do
175 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl)
176 do {
177 int i; u64 th, tl;
178 rh = rl = 0;
179 for (i = 0; i < nw; i += 8) {
180 MUL64(th, tl, pe64_to_
181 pe64_to_cpup((
182 ADD128(rh, rl, th, tl)
183 MUL64(th, tl, pe64_to_
184 pe64_to_cpup((
185 ADD128(rh, rl, th, tl)
186 MUL64(th, tl, pe64_to_
187 pe64_to_cpup((
188 ADD128(rh, rl, th, tl)
189 MUL64(th, tl, pe64_to_
190 pe64_to_cpup((
191 ADD128(rh, rl, th, tl)
192 }
193 } while (0)
194
195 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl,
196 do {
197 int i; u64 th, tl;
198 rh1 = rl1 = rh = rl = 0;
199 for (i = 0; i < nw; i += 8) {
200 MUL64(th, tl, pe64_to_
201 pe64_to_cpup((
202 ADD128(rh, rl, th, tl)
203 MUL64(th, tl, pe64_to_
204 pe64_to_cpup((
205 ADD128(rh1, rl1, th, t
206 MUL64(th, tl, pe64_to_
207 pe64_to_cpup((
208 ADD128(rh, rl, th, tl)
209 MUL64(th, tl, pe64_to_
210 pe64_to_cpup((
211 ADD128(rh1, rl1, th, t
212 MUL64(th, tl, pe64_to_
213 pe64_to_cpup((
214 ADD128(rh, rl, th, tl)
215 MUL64(th, tl, pe64_to_
216 pe64_to_cpup((
217 ADD128(rh1, rl1, th, t
218 MUL64(th, tl, pe64_to_
219 pe64_to_cpup((
220 ADD128(rh, rl, th, tl)
221 MUL64(th, tl, pe64_to_
222 pe64_to_cpup((
223 ADD128(rh1, rl1, th, t
224 }
225 } while (0)
226 #endif
227
228 #define poly_step(ah, al, kh, kl, mh, ml)
229 do {
230 u64 t1h, t1l, t2h, t2l, t3h, t
231 /* compute ab*cd, put bd into
232 PMUL64(t3h, t3l, al, kh);
233 PMUL64(t2h, t2l, ah, kl);
234 PMUL64(t1h, t1l, ah, 2*kh);
235 PMUL64(ah, al, al, kl);
236 /* add 2 * ac to result */
237 ADD128(ah, al, t1h, t1l);
238 /* add together ad + bc */
239 ADD128(t2h, t2l, t3h, t3l);
240 /* now (ah,al), (t2l,2*t2h) ne
241 /* first add the high register
242 ADD128(t2h, ah, z, t2l);
243 /* double t2h and add top bit
244 t2h = 2 * t2h + (ah >> 63);
245 ah &= m63;
246 /* now add the low registers *
247 ADD128(ah, al, mh, ml);
248 ADD128(ah, al, z, t2h);
249 } while (0)
250
251 #else /* ! CONFIG_64BIT */
252
253 #ifndef nh_16
254 #define nh_16(mp, kp, nw, rh, rl)
255 do {
256 u64 t1, t2, m1, m2, t;
257 int i;
258 rh = rl = t = 0;
259 for (i = 0; i < nw; i += 2) {
260 t1 = pe64_to_cpup(mp+i
261 t2 = pe64_to_cpup(mp+i
262 m2 = MUL32(t1 >> 32, t
263 m1 = MUL32(t1, t2 >> 3
264 ADD128(rh, rl, MUL32(t
265 MUL32(t1, t2))
266 rh += (u64)(u32)(m1 >>
267 + (u32)(m2 >>
268 t += (u64)(u32)m1 + (u
269 }
270 ADD128(rh, rl, (t >> 32), (t <
271 } while (0)
272 #endif
273
274 static void poly_step_func(u64 *ahi, u64 *alo,
275 const u64 *kh, const u
276 const u64 *mh, const u
277 {
278 #define a0 (*(((u32 *)alo)+INDEX_LOW))
279 #define a1 (*(((u32 *)alo)+INDEX_HIGH))
280 #define a2 (*(((u32 *)ahi)+INDEX_LOW))
281 #define a3 (*(((u32 *)ahi)+INDEX_HIGH))
282 #define k0 (*(((u32 *)kl)+INDEX_LOW))
283 #define k1 (*(((u32 *)kl)+INDEX_HIGH))
284 #define k2 (*(((u32 *)kh)+INDEX_LOW))
285 #define k3 (*(((u32 *)kh)+INDEX_HIGH))
286
287 u64 p, q, t;
288 u32 t2;
289
290 p = MUL32(a3, k3);
291 p += p;
292 p += *(u64 *)mh;
293 p += MUL32(a0, k2);
294 p += MUL32(a1, k1);
295 p += MUL32(a2, k0);
296 t = (u32)(p);
297 p >>= 32;
298 p += MUL32(a0, k3);
299 p += MUL32(a1, k2);
300 p += MUL32(a2, k1);
301 p += MUL32(a3, k0);
302 t |= ((u64)((u32)p & 0x7fffffff)) << 3
303 p >>= 31;
304 p += (u64)(((u32 *)ml)[INDEX_LOW]);
305 p += MUL32(a0, k0);
306 q = MUL32(a1, k3);
307 q += MUL32(a2, k2);
308 q += MUL32(a3, k1);
309 q += q;
310 p += q;
311 t2 = (u32)(p);
312 p >>= 32;
313 p += (u64)(((u32 *)ml)[INDEX_HIGH]);
314 p += MUL32(a0, k1);
315 p += MUL32(a1, k0);
316 q = MUL32(a2, k3);
317 q += MUL32(a3, k2);
318 q += q;
319 p += q;
320 *(u64 *)(alo) = (p << 32) | t2;
321 p >>= 32;
322 *(u64 *)(ahi) = p + t;
323
324 #undef a0
325 #undef a1
326 #undef a2
327 #undef a3
328 #undef k0
329 #undef k1
330 #undef k2
331 #undef k3
332 }
333
334 #define poly_step(ah, al, kh, kl, mh, ml)
335 poly_step_func(&(ah), &(al), &(kh), &(
336
337 #endif /* end of specialized NH and poly defi
338
339 /* At least nh_16 is defined. Defined others a
340 #ifndef nh_16_2
341 #define nh_16_2(mp, kp, nw, rh, rl, rh2, rl2)
342 do {
343 nh_16(mp, kp, nw, rh, rl);
344 nh_16(mp, ((kp)+2), nw, rh2, r
345 } while (0)
346 #endif
347 #ifndef nh_vmac_nhbytes
348 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl)
349 nh_16(mp, kp, nw, rh, rl)
350 #endif
351 #ifndef nh_vmac_nhbytes_2
352 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl,
353 do {
354 nh_vmac_nhbytes(mp, kp, nw, rh
355 nh_vmac_nhbytes(mp, ((kp)+2),
356 } while (0)
357 #endif
358
359 static u64 l3hash(u64 p1, u64 p2, u64 k1, u64
360 {
361 u64 rh, rl, t, z = 0;
362
363 /* fully reduce (p1,p2)+(len,0) mod p1
364 t = p1 >> 63;
365 p1 &= m63;
366 ADD128(p1, p2, len, t);
367 /* At this point, (p1,p2) is at most 2
368 t = (p1 > m63) + ((p1 == m63) && (p2 =
369 ADD128(p1, p2, z, t);
370 p1 &= m63;
371
372 /* compute (p1,p2)/(2^64-2^32) and (p1
373 t = p1 + (p2 >> 32);
374 t += (t >> 32);
375 t += (u32)t > 0xfffffffeu;
376 p1 += (t >> 32);
377 p2 += (p1 << 32);
378
379 /* compute (p1+k1)%p64 and (p2+k2)%p64
380 p1 += k1;
381 p1 += (0 - (p1 < k1)) & 257;
382 p2 += k2;
383 p2 += (0 - (p2 < k2)) & 257;
384
385 /* compute (p1+k1)*(p2+k2)%p64 */
386 MUL64(rh, rl, p1, p2);
387 t = rh >> 56;
388 ADD128(t, rl, z, rh);
389 rh <<= 8;
390 ADD128(t, rl, z, rh);
391 t += t << 8;
392 rl += t;
393 rl += (0 - (rl < t)) & 257;
394 rl += (0 - (rl > p64-1)) & 257;
395 return rl;
396 }
397
398 /* L1 and L2-hash one or more VMAC_NHBYTES-byt
399 static void vhash_blocks(const struct vmac_tfm
400 struct vmac_desc_ctx
401 const __le64 *mptr, u
402 {
403 const u64 *kptr = tctx->nhkey;
404 const u64 pkh = tctx->polykey[0];
405 const u64 pkl = tctx->polykey[1];
406 u64 ch = dctx->polytmp[0];
407 u64 cl = dctx->polytmp[1];
408 u64 rh, rl;
409
410 if (!dctx->first_block_processed) {
411 dctx->first_block_processed =
412 nh_vmac_nhbytes(mptr, kptr, VM
413 rh &= m62;
414 ADD128(ch, cl, rh, rl);
415 mptr += (VMAC_NHBYTES/sizeof(u
416 blocks--;
417 }
418
419 while (blocks--) {
420 nh_vmac_nhbytes(mptr, kptr, VM
421 rh &= m62;
422 poly_step(ch, cl, pkh, pkl, rh
423 mptr += (VMAC_NHBYTES/sizeof(u
424 }
425
426 dctx->polytmp[0] = ch;
427 dctx->polytmp[1] = cl;
428 }
429
430 static int vmac_setkey(struct crypto_shash *tf
431 const u8 *key, unsigned
432 {
433 struct vmac_tfm_ctx *tctx = crypto_sha
434 __be64 out[2];
435 u8 in[16] = { 0 };
436 unsigned int i;
437 int err;
438
439 if (keylen != VMAC_KEY_LEN)
440 return -EINVAL;
441
442 err = crypto_cipher_setkey(tctx->ciphe
443 if (err)
444 return err;
445
446 /* Fill nh key */
447 in[0] = 0x80;
448 for (i = 0; i < ARRAY_SIZE(tctx->nhkey
449 crypto_cipher_encrypt_one(tctx
450 tctx->nhkey[i] = be64_to_cpu(o
451 tctx->nhkey[i+1] = be64_to_cpu
452 in[15]++;
453 }
454
455 /* Fill poly key */
456 in[0] = 0xC0;
457 in[15] = 0;
458 for (i = 0; i < ARRAY_SIZE(tctx->polyk
459 crypto_cipher_encrypt_one(tctx
460 tctx->polykey[i] = be64_to_cpu
461 tctx->polykey[i+1] = be64_to_c
462 in[15]++;
463 }
464
465 /* Fill ip key */
466 in[0] = 0xE0;
467 in[15] = 0;
468 for (i = 0; i < ARRAY_SIZE(tctx->l3key
469 do {
470 crypto_cipher_encrypt_
471 tctx->l3key[i] = be64_
472 tctx->l3key[i+1] = be6
473 in[15]++;
474 } while (tctx->l3key[i] >= p64
475 }
476
477 return 0;
478 }
479
480 static int vmac_init(struct shash_desc *desc)
481 {
482 const struct vmac_tfm_ctx *tctx = cryp
483 struct vmac_desc_ctx *dctx = shash_des
484
485 dctx->partial_size = 0;
486 dctx->first_block_processed = false;
487 memcpy(dctx->polytmp, tctx->polykey, s
488 dctx->nonce_size = 0;
489 return 0;
490 }
491
492 static int vmac_update(struct shash_desc *desc
493 {
494 const struct vmac_tfm_ctx *tctx = cryp
495 struct vmac_desc_ctx *dctx = shash_des
496 unsigned int n;
497
498 /* Nonce is passed as first VMAC_NONCE
499 if (dctx->nonce_size < VMAC_NONCEBYTES
500 n = min(len, VMAC_NONCEBYTES -
501 memcpy(&dctx->nonce.bytes[dctx
502 dctx->nonce_size += n;
503 p += n;
504 len -= n;
505 }
506
507 if (dctx->partial_size) {
508 n = min(len, VMAC_NHBYTES - dc
509 memcpy(&dctx->partial[dctx->pa
510 dctx->partial_size += n;
511 p += n;
512 len -= n;
513 if (dctx->partial_size == VMAC
514 vhash_blocks(tctx, dct
515 dctx->partial_size = 0
516 }
517 }
518
519 if (len >= VMAC_NHBYTES) {
520 n = round_down(len, VMAC_NHBYT
521 /* TODO: 'p' may be misaligned
522 vhash_blocks(tctx, dctx, (cons
523 p += n;
524 len -= n;
525 }
526
527 if (len) {
528 memcpy(dctx->partial, p, len);
529 dctx->partial_size = len;
530 }
531
532 return 0;
533 }
534
535 static u64 vhash_final(const struct vmac_tfm_c
536 struct vmac_desc_ctx *d
537 {
538 unsigned int partial = dctx->partial_s
539 u64 ch = dctx->polytmp[0];
540 u64 cl = dctx->polytmp[1];
541
542 /* L1 and L2-hash the final block if n
543 if (partial) {
544 /* Zero-pad to next 128-bit bo
545 unsigned int n = round_up(part
546 u64 rh, rl;
547
548 memset(&dctx->partial[partial]
549 nh_16(dctx->partial_words, tct
550 rh &= m62;
551 if (dctx->first_block_processe
552 poly_step(ch, cl, tctx
553 rh, rl);
554 else
555 ADD128(ch, cl, rh, rl)
556 }
557
558 /* L3-hash the 128-bit output of L2-ha
559 return l3hash(ch, cl, tctx->l3key[0],
560 }
561
562 static int vmac_final(struct shash_desc *desc,
563 {
564 const struct vmac_tfm_ctx *tctx = cryp
565 struct vmac_desc_ctx *dctx = shash_des
566 int index;
567 u64 hash, pad;
568
569 if (dctx->nonce_size != VMAC_NONCEBYTE
570 return -EINVAL;
571
572 /*
573 * The VMAC specification requires a n
574 * the block cipher's block length, so
575 * nonce. We define the unused bit to
576 * it be 0, so the needed prepending o
577 */
578 if (dctx->nonce.bytes[0] & 0x80)
579 return -EINVAL;
580
581 /* Finish calculating the VHASH of the
582 hash = vhash_final(tctx, dctx);
583
584 /* Generate pseudorandom pad by encryp
585 BUILD_BUG_ON(VMAC_NONCEBYTES != 2 * (V
586 index = dctx->nonce.bytes[VMAC_NONCEBY
587 dctx->nonce.bytes[VMAC_NONCEBYTES - 1]
588 crypto_cipher_encrypt_one(tctx->cipher
589 dctx->nonce.
590 pad = be64_to_cpu(dctx->nonce.pads[ind
591
592 /* The VMAC is the sum of VHASH and th
593 put_unaligned_be64(hash + pad, out);
594 return 0;
595 }
596
597 static int vmac_init_tfm(struct crypto_tfm *tf
598 {
599 struct crypto_instance *inst = crypto_
600 struct crypto_cipher_spawn *spawn = cr
601 struct vmac_tfm_ctx *tctx = crypto_tfm
602 struct crypto_cipher *cipher;
603
604 cipher = crypto_spawn_cipher(spawn);
605 if (IS_ERR(cipher))
606 return PTR_ERR(cipher);
607
608 tctx->cipher = cipher;
609 return 0;
610 }
611
612 static void vmac_exit_tfm(struct crypto_tfm *t
613 {
614 struct vmac_tfm_ctx *tctx = crypto_tfm
615
616 crypto_free_cipher(tctx->cipher);
617 }
618
619 static int vmac_create(struct crypto_template
620 {
621 struct shash_instance *inst;
622 struct crypto_cipher_spawn *spawn;
623 struct crypto_alg *alg;
624 u32 mask;
625 int err;
626
627 err = crypto_check_attr_type(tb, CRYPT
628 if (err)
629 return err;
630
631 inst = kzalloc(sizeof(*inst) + sizeof(
632 if (!inst)
633 return -ENOMEM;
634 spawn = shash_instance_ctx(inst);
635
636 err = crypto_grab_cipher(spawn, shash_
637 crypto_attr_a
638 if (err)
639 goto err_free_inst;
640 alg = crypto_spawn_cipher_alg(spawn);
641
642 err = -EINVAL;
643 if (alg->cra_blocksize != VMAC_NONCEBY
644 goto err_free_inst;
645
646 err = crypto_inst_setname(shash_crypto
647 if (err)
648 goto err_free_inst;
649
650 inst->alg.base.cra_priority = alg->cra
651 inst->alg.base.cra_blocksize = alg->cr
652
653 inst->alg.base.cra_ctxsize = sizeof(st
654 inst->alg.base.cra_init = vmac_init_tf
655 inst->alg.base.cra_exit = vmac_exit_tf
656
657 inst->alg.descsize = sizeof(struct vma
658 inst->alg.digestsize = VMAC_TAG_LEN /
659 inst->alg.init = vmac_init;
660 inst->alg.update = vmac_update;
661 inst->alg.final = vmac_final;
662 inst->alg.setkey = vmac_setkey;
663
664 inst->free = shash_free_singlespawn_in
665
666 err = shash_register_instance(tmpl, in
667 if (err) {
668 err_free_inst:
669 shash_free_singlespawn_instanc
670 }
671 return err;
672 }
673
674 static struct crypto_template vmac64_tmpl = {
675 .name = "vmac64",
676 .create = vmac_create,
677 .module = THIS_MODULE,
678 };
679
680 static int __init vmac_module_init(void)
681 {
682 return crypto_register_template(&vmac6
683 }
684
685 static void __exit vmac_module_exit(void)
686 {
687 crypto_unregister_template(&vmac64_tmp
688 }
689
690 subsys_initcall(vmac_module_init);
691 module_exit(vmac_module_exit);
692
693 MODULE_LICENSE("GPL");
694 MODULE_DESCRIPTION("VMAC hash algorithm");
695 MODULE_ALIAS_CRYPTO("vmac64");
696 MODULE_IMPORT_NS(CRYPTO_INTERNAL);
697
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.