xref: /openbmc/linux/crypto/vmac.c (revision e0d07278)
1 /*
2  * VMAC: Message Authentication Code using Universal Hashing
3  *
4  * Reference: https://tools.ietf.org/html/draft-krovetz-vmac-01
5  *
6  * Copyright (c) 2009, Intel Corporation.
7  * Copyright (c) 2018, Google Inc.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms and conditions of the GNU General Public License,
11  * version 2, as published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along with
19  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
20  * Place - Suite 330, Boston, MA 02111-1307 USA.
21  */
22 
23 /*
24  * Derived from:
25  *	VMAC and VHASH Implementation by Ted Krovetz (tdk@acm.org) and Wei Dai.
26  *	This implementation is herby placed in the public domain.
27  *	The authors offers no warranty. Use at your own risk.
28  *	Last modified: 17 APR 08, 1700 PDT
29  */
30 
31 #include <asm/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/hash.h>
40 
41 /*
42  * User definable settings.
43  */
44 #define VMAC_TAG_LEN	64
45 #define VMAC_KEY_SIZE	128/* Must be 128, 192 or 256			*/
46 #define VMAC_KEY_LEN	(VMAC_KEY_SIZE/8)
47 #define VMAC_NHBYTES	128/* Must 2^i for any 3 < i < 13 Standard = 128*/
48 #define VMAC_NONCEBYTES	16
49 
50 /* per-transform (per-key) context */
51 struct vmac_tfm_ctx {
52 	struct crypto_cipher *cipher;
53 	u64 nhkey[(VMAC_NHBYTES/8)+2*(VMAC_TAG_LEN/64-1)];
54 	u64 polykey[2*VMAC_TAG_LEN/64];
55 	u64 l3key[2*VMAC_TAG_LEN/64];
56 };
57 
58 /* per-request context */
59 struct vmac_desc_ctx {
60 	union {
61 		u8 partial[VMAC_NHBYTES];	/* partial block */
62 		__le64 partial_words[VMAC_NHBYTES / 8];
63 	};
64 	unsigned int partial_size;	/* size of the partial block */
65 	bool first_block_processed;
66 	u64 polytmp[2*VMAC_TAG_LEN/64];	/* running total of L2-hash */
67 	union {
68 		u8 bytes[VMAC_NONCEBYTES];
69 		__be64 pads[VMAC_NONCEBYTES / 8];
70 	} nonce;
71 	unsigned int nonce_size; /* nonce bytes filled so far */
72 };
73 
74 /*
75  * Constants and masks
76  */
77 #define UINT64_C(x) x##ULL
78 static const u64 p64   = UINT64_C(0xfffffffffffffeff);	/* 2^64 - 257 prime  */
79 static const u64 m62   = UINT64_C(0x3fffffffffffffff);	/* 62-bit mask       */
80 static const u64 m63   = UINT64_C(0x7fffffffffffffff);	/* 63-bit mask       */
81 static const u64 m64   = UINT64_C(0xffffffffffffffff);	/* 64-bit mask       */
82 static const u64 mpoly = UINT64_C(0x1fffffff1fffffff);	/* Poly key mask     */
83 
84 #define pe64_to_cpup le64_to_cpup		/* Prefer little endian */
85 
86 #ifdef __LITTLE_ENDIAN
87 #define INDEX_HIGH 1
88 #define INDEX_LOW 0
89 #else
90 #define INDEX_HIGH 0
91 #define INDEX_LOW 1
92 #endif
93 
94 /*
95  * The following routines are used in this implementation. They are
96  * written via macros to simulate zero-overhead call-by-reference.
97  *
98  * MUL64: 64x64->128-bit multiplication
99  * PMUL64: assumes top bits cleared on inputs
100  * ADD128: 128x128->128-bit addition
101  */
102 
103 #define ADD128(rh, rl, ih, il)						\
104 	do {								\
105 		u64 _il = (il);						\
106 		(rl) += (_il);						\
107 		if ((rl) < (_il))					\
108 			(rh)++;						\
109 		(rh) += (ih);						\
110 	} while (0)
111 
112 #define MUL32(i1, i2)	((u64)(u32)(i1)*(u32)(i2))
113 
114 #define PMUL64(rh, rl, i1, i2)	/* Assumes m doesn't overflow */	\
115 	do {								\
116 		u64 _i1 = (i1), _i2 = (i2);				\
117 		u64 m = MUL32(_i1, _i2>>32) + MUL32(_i1>>32, _i2);	\
118 		rh = MUL32(_i1>>32, _i2>>32);				\
119 		rl = MUL32(_i1, _i2);					\
120 		ADD128(rh, rl, (m >> 32), (m << 32));			\
121 	} while (0)
122 
123 #define MUL64(rh, rl, i1, i2)						\
124 	do {								\
125 		u64 _i1 = (i1), _i2 = (i2);				\
126 		u64 m1 = MUL32(_i1, _i2>>32);				\
127 		u64 m2 = MUL32(_i1>>32, _i2);				\
128 		rh = MUL32(_i1>>32, _i2>>32);				\
129 		rl = MUL32(_i1, _i2);					\
130 		ADD128(rh, rl, (m1 >> 32), (m1 << 32));			\
131 		ADD128(rh, rl, (m2 >> 32), (m2 << 32));			\
132 	} while (0)
133 
134 /*
135  * For highest performance the L1 NH and L2 polynomial hashes should be
136  * carefully implemented to take advantage of one's target architecture.
137  * Here these two hash functions are defined multiple time; once for
138  * 64-bit architectures, once for 32-bit SSE2 architectures, and once
139  * for the rest (32-bit) architectures.
140  * For each, nh_16 *must* be defined (works on multiples of 16 bytes).
141  * Optionally, nh_vmac_nhbytes can be defined (for multiples of
142  * VMAC_NHBYTES), and nh_16_2 and nh_vmac_nhbytes_2 (versions that do two
143  * NH computations at once).
144  */
145 
146 #ifdef CONFIG_64BIT
147 
148 #define nh_16(mp, kp, nw, rh, rl)					\
149 	do {								\
150 		int i; u64 th, tl;					\
151 		rh = rl = 0;						\
152 		for (i = 0; i < nw; i += 2) {				\
153 			MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i],	\
154 				pe64_to_cpup((mp)+i+1)+(kp)[i+1]);	\
155 			ADD128(rh, rl, th, tl);				\
156 		}							\
157 	} while (0)
158 
159 #define nh_16_2(mp, kp, nw, rh, rl, rh1, rl1)				\
160 	do {								\
161 		int i; u64 th, tl;					\
162 		rh1 = rl1 = rh = rl = 0;				\
163 		for (i = 0; i < nw; i += 2) {				\
164 			MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i],	\
165 				pe64_to_cpup((mp)+i+1)+(kp)[i+1]);	\
166 			ADD128(rh, rl, th, tl);				\
167 			MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2],	\
168 				pe64_to_cpup((mp)+i+1)+(kp)[i+3]);	\
169 			ADD128(rh1, rl1, th, tl);			\
170 		}							\
171 	} while (0)
172 
173 #if (VMAC_NHBYTES >= 64) /* These versions do 64-bytes of message at a time */
174 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl)				\
175 	do {								\
176 		int i; u64 th, tl;					\
177 		rh = rl = 0;						\
178 		for (i = 0; i < nw; i += 8) {				\
179 			MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i],	\
180 				pe64_to_cpup((mp)+i+1)+(kp)[i+1]);	\
181 			ADD128(rh, rl, th, tl);				\
182 			MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2],	\
183 				pe64_to_cpup((mp)+i+3)+(kp)[i+3]);	\
184 			ADD128(rh, rl, th, tl);				\
185 			MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4],	\
186 				pe64_to_cpup((mp)+i+5)+(kp)[i+5]);	\
187 			ADD128(rh, rl, th, tl);				\
188 			MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6],	\
189 				pe64_to_cpup((mp)+i+7)+(kp)[i+7]);	\
190 			ADD128(rh, rl, th, tl);				\
191 		}							\
192 	} while (0)
193 
194 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh1, rl1)			\
195 	do {								\
196 		int i; u64 th, tl;					\
197 		rh1 = rl1 = rh = rl = 0;				\
198 		for (i = 0; i < nw; i += 8) {				\
199 			MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i],	\
200 				pe64_to_cpup((mp)+i+1)+(kp)[i+1]);	\
201 			ADD128(rh, rl, th, tl);				\
202 			MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2],	\
203 				pe64_to_cpup((mp)+i+1)+(kp)[i+3]);	\
204 			ADD128(rh1, rl1, th, tl);			\
205 			MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2],	\
206 				pe64_to_cpup((mp)+i+3)+(kp)[i+3]);	\
207 			ADD128(rh, rl, th, tl);				\
208 			MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+4],	\
209 				pe64_to_cpup((mp)+i+3)+(kp)[i+5]);	\
210 			ADD128(rh1, rl1, th, tl);			\
211 			MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4],	\
212 				pe64_to_cpup((mp)+i+5)+(kp)[i+5]);	\
213 			ADD128(rh, rl, th, tl);				\
214 			MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+6],	\
215 				pe64_to_cpup((mp)+i+5)+(kp)[i+7]);	\
216 			ADD128(rh1, rl1, th, tl);			\
217 			MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6],	\
218 				pe64_to_cpup((mp)+i+7)+(kp)[i+7]);	\
219 			ADD128(rh, rl, th, tl);				\
220 			MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+8],	\
221 				pe64_to_cpup((mp)+i+7)+(kp)[i+9]);	\
222 			ADD128(rh1, rl1, th, tl);			\
223 		}							\
224 	} while (0)
225 #endif
226 
227 #define poly_step(ah, al, kh, kl, mh, ml)				\
228 	do {								\
229 		u64 t1h, t1l, t2h, t2l, t3h, t3l, z = 0;		\
230 		/* compute ab*cd, put bd into result registers */	\
231 		PMUL64(t3h, t3l, al, kh);				\
232 		PMUL64(t2h, t2l, ah, kl);				\
233 		PMUL64(t1h, t1l, ah, 2*kh);				\
234 		PMUL64(ah, al, al, kl);					\
235 		/* add 2 * ac to result */				\
236 		ADD128(ah, al, t1h, t1l);				\
237 		/* add together ad + bc */				\
238 		ADD128(t2h, t2l, t3h, t3l);				\
239 		/* now (ah,al), (t2l,2*t2h) need summing */		\
240 		/* first add the high registers, carrying into t2h */	\
241 		ADD128(t2h, ah, z, t2l);				\
242 		/* double t2h and add top bit of ah */			\
243 		t2h = 2 * t2h + (ah >> 63);				\
244 		ah &= m63;						\
245 		/* now add the low registers */				\
246 		ADD128(ah, al, mh, ml);					\
247 		ADD128(ah, al, z, t2h);					\
248 	} while (0)
249 
250 #else /* ! CONFIG_64BIT */
251 
252 #ifndef nh_16
253 #define nh_16(mp, kp, nw, rh, rl)					\
254 	do {								\
255 		u64 t1, t2, m1, m2, t;					\
256 		int i;							\
257 		rh = rl = t = 0;					\
258 		for (i = 0; i < nw; i += 2)  {				\
259 			t1 = pe64_to_cpup(mp+i) + kp[i];		\
260 			t2 = pe64_to_cpup(mp+i+1) + kp[i+1];		\
261 			m2 = MUL32(t1 >> 32, t2);			\
262 			m1 = MUL32(t1, t2 >> 32);			\
263 			ADD128(rh, rl, MUL32(t1 >> 32, t2 >> 32),	\
264 				MUL32(t1, t2));				\
265 			rh += (u64)(u32)(m1 >> 32)			\
266 				+ (u32)(m2 >> 32);			\
267 			t += (u64)(u32)m1 + (u32)m2;			\
268 		}							\
269 		ADD128(rh, rl, (t >> 32), (t << 32));			\
270 	} while (0)
271 #endif
272 
273 static void poly_step_func(u64 *ahi, u64 *alo,
274 			const u64 *kh, const u64 *kl,
275 			const u64 *mh, const u64 *ml)
276 {
277 #define a0 (*(((u32 *)alo)+INDEX_LOW))
278 #define a1 (*(((u32 *)alo)+INDEX_HIGH))
279 #define a2 (*(((u32 *)ahi)+INDEX_LOW))
280 #define a3 (*(((u32 *)ahi)+INDEX_HIGH))
281 #define k0 (*(((u32 *)kl)+INDEX_LOW))
282 #define k1 (*(((u32 *)kl)+INDEX_HIGH))
283 #define k2 (*(((u32 *)kh)+INDEX_LOW))
284 #define k3 (*(((u32 *)kh)+INDEX_HIGH))
285 
286 	u64 p, q, t;
287 	u32 t2;
288 
289 	p = MUL32(a3, k3);
290 	p += p;
291 	p += *(u64 *)mh;
292 	p += MUL32(a0, k2);
293 	p += MUL32(a1, k1);
294 	p += MUL32(a2, k0);
295 	t = (u32)(p);
296 	p >>= 32;
297 	p += MUL32(a0, k3);
298 	p += MUL32(a1, k2);
299 	p += MUL32(a2, k1);
300 	p += MUL32(a3, k0);
301 	t |= ((u64)((u32)p & 0x7fffffff)) << 32;
302 	p >>= 31;
303 	p += (u64)(((u32 *)ml)[INDEX_LOW]);
304 	p += MUL32(a0, k0);
305 	q =  MUL32(a1, k3);
306 	q += MUL32(a2, k2);
307 	q += MUL32(a3, k1);
308 	q += q;
309 	p += q;
310 	t2 = (u32)(p);
311 	p >>= 32;
312 	p += (u64)(((u32 *)ml)[INDEX_HIGH]);
313 	p += MUL32(a0, k1);
314 	p += MUL32(a1, k0);
315 	q =  MUL32(a2, k3);
316 	q += MUL32(a3, k2);
317 	q += q;
318 	p += q;
319 	*(u64 *)(alo) = (p << 32) | t2;
320 	p >>= 32;
321 	*(u64 *)(ahi) = p + t;
322 
323 #undef a0
324 #undef a1
325 #undef a2
326 #undef a3
327 #undef k0
328 #undef k1
329 #undef k2
330 #undef k3
331 }
332 
333 #define poly_step(ah, al, kh, kl, mh, ml)				\
334 	poly_step_func(&(ah), &(al), &(kh), &(kl), &(mh), &(ml))
335 
336 #endif  /* end of specialized NH and poly definitions */
337 
338 /* At least nh_16 is defined. Defined others as needed here */
339 #ifndef nh_16_2
340 #define nh_16_2(mp, kp, nw, rh, rl, rh2, rl2)				\
341 	do { 								\
342 		nh_16(mp, kp, nw, rh, rl);				\
343 		nh_16(mp, ((kp)+2), nw, rh2, rl2);			\
344 	} while (0)
345 #endif
346 #ifndef nh_vmac_nhbytes
347 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl)				\
348 	nh_16(mp, kp, nw, rh, rl)
349 #endif
350 #ifndef nh_vmac_nhbytes_2
351 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh2, rl2)			\
352 	do {								\
353 		nh_vmac_nhbytes(mp, kp, nw, rh, rl);			\
354 		nh_vmac_nhbytes(mp, ((kp)+2), nw, rh2, rl2);		\
355 	} while (0)
356 #endif
357 
358 static u64 l3hash(u64 p1, u64 p2, u64 k1, u64 k2, u64 len)
359 {
360 	u64 rh, rl, t, z = 0;
361 
362 	/* fully reduce (p1,p2)+(len,0) mod p127 */
363 	t = p1 >> 63;
364 	p1 &= m63;
365 	ADD128(p1, p2, len, t);
366 	/* At this point, (p1,p2) is at most 2^127+(len<<64) */
367 	t = (p1 > m63) + ((p1 == m63) && (p2 == m64));
368 	ADD128(p1, p2, z, t);
369 	p1 &= m63;
370 
371 	/* compute (p1,p2)/(2^64-2^32) and (p1,p2)%(2^64-2^32) */
372 	t = p1 + (p2 >> 32);
373 	t += (t >> 32);
374 	t += (u32)t > 0xfffffffeu;
375 	p1 += (t >> 32);
376 	p2 += (p1 << 32);
377 
378 	/* compute (p1+k1)%p64 and (p2+k2)%p64 */
379 	p1 += k1;
380 	p1 += (0 - (p1 < k1)) & 257;
381 	p2 += k2;
382 	p2 += (0 - (p2 < k2)) & 257;
383 
384 	/* compute (p1+k1)*(p2+k2)%p64 */
385 	MUL64(rh, rl, p1, p2);
386 	t = rh >> 56;
387 	ADD128(t, rl, z, rh);
388 	rh <<= 8;
389 	ADD128(t, rl, z, rh);
390 	t += t << 8;
391 	rl += t;
392 	rl += (0 - (rl < t)) & 257;
393 	rl += (0 - (rl > p64-1)) & 257;
394 	return rl;
395 }
396 
397 /* L1 and L2-hash one or more VMAC_NHBYTES-byte blocks */
398 static void vhash_blocks(const struct vmac_tfm_ctx *tctx,
399 			 struct vmac_desc_ctx *dctx,
400 			 const __le64 *mptr, unsigned int blocks)
401 {
402 	const u64 *kptr = tctx->nhkey;
403 	const u64 pkh = tctx->polykey[0];
404 	const u64 pkl = tctx->polykey[1];
405 	u64 ch = dctx->polytmp[0];
406 	u64 cl = dctx->polytmp[1];
407 	u64 rh, rl;
408 
409 	if (!dctx->first_block_processed) {
410 		dctx->first_block_processed = true;
411 		nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
412 		rh &= m62;
413 		ADD128(ch, cl, rh, rl);
414 		mptr += (VMAC_NHBYTES/sizeof(u64));
415 		blocks--;
416 	}
417 
418 	while (blocks--) {
419 		nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
420 		rh &= m62;
421 		poly_step(ch, cl, pkh, pkl, rh, rl);
422 		mptr += (VMAC_NHBYTES/sizeof(u64));
423 	}
424 
425 	dctx->polytmp[0] = ch;
426 	dctx->polytmp[1] = cl;
427 }
428 
429 static int vmac_setkey(struct crypto_shash *tfm,
430 		       const u8 *key, unsigned int keylen)
431 {
432 	struct vmac_tfm_ctx *tctx = crypto_shash_ctx(tfm);
433 	__be64 out[2];
434 	u8 in[16] = { 0 };
435 	unsigned int i;
436 	int err;
437 
438 	if (keylen != VMAC_KEY_LEN)
439 		return -EINVAL;
440 
441 	err = crypto_cipher_setkey(tctx->cipher, key, keylen);
442 	if (err)
443 		return err;
444 
445 	/* Fill nh key */
446 	in[0] = 0x80;
447 	for (i = 0; i < ARRAY_SIZE(tctx->nhkey); i += 2) {
448 		crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
449 		tctx->nhkey[i] = be64_to_cpu(out[0]);
450 		tctx->nhkey[i+1] = be64_to_cpu(out[1]);
451 		in[15]++;
452 	}
453 
454 	/* Fill poly key */
455 	in[0] = 0xC0;
456 	in[15] = 0;
457 	for (i = 0; i < ARRAY_SIZE(tctx->polykey); i += 2) {
458 		crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
459 		tctx->polykey[i] = be64_to_cpu(out[0]) & mpoly;
460 		tctx->polykey[i+1] = be64_to_cpu(out[1]) & mpoly;
461 		in[15]++;
462 	}
463 
464 	/* Fill ip key */
465 	in[0] = 0xE0;
466 	in[15] = 0;
467 	for (i = 0; i < ARRAY_SIZE(tctx->l3key); i += 2) {
468 		do {
469 			crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
470 			tctx->l3key[i] = be64_to_cpu(out[0]);
471 			tctx->l3key[i+1] = be64_to_cpu(out[1]);
472 			in[15]++;
473 		} while (tctx->l3key[i] >= p64 || tctx->l3key[i+1] >= p64);
474 	}
475 
476 	return 0;
477 }
478 
479 static int vmac_init(struct shash_desc *desc)
480 {
481 	const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
482 	struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
483 
484 	dctx->partial_size = 0;
485 	dctx->first_block_processed = false;
486 	memcpy(dctx->polytmp, tctx->polykey, sizeof(dctx->polytmp));
487 	dctx->nonce_size = 0;
488 	return 0;
489 }
490 
491 static int vmac_update(struct shash_desc *desc, const u8 *p, unsigned int len)
492 {
493 	const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
494 	struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
495 	unsigned int n;
496 
497 	/* Nonce is passed as first VMAC_NONCEBYTES bytes of data */
498 	if (dctx->nonce_size < VMAC_NONCEBYTES) {
499 		n = min(len, VMAC_NONCEBYTES - dctx->nonce_size);
500 		memcpy(&dctx->nonce.bytes[dctx->nonce_size], p, n);
501 		dctx->nonce_size += n;
502 		p += n;
503 		len -= n;
504 	}
505 
506 	if (dctx->partial_size) {
507 		n = min(len, VMAC_NHBYTES - dctx->partial_size);
508 		memcpy(&dctx->partial[dctx->partial_size], p, n);
509 		dctx->partial_size += n;
510 		p += n;
511 		len -= n;
512 		if (dctx->partial_size == VMAC_NHBYTES) {
513 			vhash_blocks(tctx, dctx, dctx->partial_words, 1);
514 			dctx->partial_size = 0;
515 		}
516 	}
517 
518 	if (len >= VMAC_NHBYTES) {
519 		n = round_down(len, VMAC_NHBYTES);
520 		/* TODO: 'p' may be misaligned here */
521 		vhash_blocks(tctx, dctx, (const __le64 *)p, n / VMAC_NHBYTES);
522 		p += n;
523 		len -= n;
524 	}
525 
526 	if (len) {
527 		memcpy(dctx->partial, p, len);
528 		dctx->partial_size = len;
529 	}
530 
531 	return 0;
532 }
533 
534 static u64 vhash_final(const struct vmac_tfm_ctx *tctx,
535 		       struct vmac_desc_ctx *dctx)
536 {
537 	unsigned int partial = dctx->partial_size;
538 	u64 ch = dctx->polytmp[0];
539 	u64 cl = dctx->polytmp[1];
540 
541 	/* L1 and L2-hash the final block if needed */
542 	if (partial) {
543 		/* Zero-pad to next 128-bit boundary */
544 		unsigned int n = round_up(partial, 16);
545 		u64 rh, rl;
546 
547 		memset(&dctx->partial[partial], 0, n - partial);
548 		nh_16(dctx->partial_words, tctx->nhkey, n / 8, rh, rl);
549 		rh &= m62;
550 		if (dctx->first_block_processed)
551 			poly_step(ch, cl, tctx->polykey[0], tctx->polykey[1],
552 				  rh, rl);
553 		else
554 			ADD128(ch, cl, rh, rl);
555 	}
556 
557 	/* L3-hash the 128-bit output of L2-hash */
558 	return l3hash(ch, cl, tctx->l3key[0], tctx->l3key[1], partial * 8);
559 }
560 
561 static int vmac_final(struct shash_desc *desc, u8 *out)
562 {
563 	const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
564 	struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
565 	int index;
566 	u64 hash, pad;
567 
568 	if (dctx->nonce_size != VMAC_NONCEBYTES)
569 		return -EINVAL;
570 
571 	/*
572 	 * The VMAC specification requires a nonce at least 1 bit shorter than
573 	 * the block cipher's block length, so we actually only accept a 127-bit
574 	 * nonce.  We define the unused bit to be the first one and require that
575 	 * it be 0, so the needed prepending of a 0 bit is implicit.
576 	 */
577 	if (dctx->nonce.bytes[0] & 0x80)
578 		return -EINVAL;
579 
580 	/* Finish calculating the VHASH of the message */
581 	hash = vhash_final(tctx, dctx);
582 
583 	/* Generate pseudorandom pad by encrypting the nonce */
584 	BUILD_BUG_ON(VMAC_NONCEBYTES != 2 * (VMAC_TAG_LEN / 8));
585 	index = dctx->nonce.bytes[VMAC_NONCEBYTES - 1] & 1;
586 	dctx->nonce.bytes[VMAC_NONCEBYTES - 1] &= ~1;
587 	crypto_cipher_encrypt_one(tctx->cipher, dctx->nonce.bytes,
588 				  dctx->nonce.bytes);
589 	pad = be64_to_cpu(dctx->nonce.pads[index]);
590 
591 	/* The VMAC is the sum of VHASH and the pseudorandom pad */
592 	put_unaligned_be64(hash + pad, out);
593 	return 0;
594 }
595 
596 static int vmac_init_tfm(struct crypto_tfm *tfm)
597 {
598 	struct crypto_instance *inst = crypto_tfm_alg_instance(tfm);
599 	struct crypto_cipher_spawn *spawn = crypto_instance_ctx(inst);
600 	struct vmac_tfm_ctx *tctx = crypto_tfm_ctx(tfm);
601 	struct crypto_cipher *cipher;
602 
603 	cipher = crypto_spawn_cipher(spawn);
604 	if (IS_ERR(cipher))
605 		return PTR_ERR(cipher);
606 
607 	tctx->cipher = cipher;
608 	return 0;
609 }
610 
611 static void vmac_exit_tfm(struct crypto_tfm *tfm)
612 {
613 	struct vmac_tfm_ctx *tctx = crypto_tfm_ctx(tfm);
614 
615 	crypto_free_cipher(tctx->cipher);
616 }
617 
618 static int vmac_create(struct crypto_template *tmpl, struct rtattr **tb)
619 {
620 	struct shash_instance *inst;
621 	struct crypto_cipher_spawn *spawn;
622 	struct crypto_alg *alg;
623 	u32 mask;
624 	int err;
625 
626 	err = crypto_check_attr_type(tb, CRYPTO_ALG_TYPE_SHASH, &mask);
627 	if (err)
628 		return err;
629 
630 	inst = kzalloc(sizeof(*inst) + sizeof(*spawn), GFP_KERNEL);
631 	if (!inst)
632 		return -ENOMEM;
633 	spawn = shash_instance_ctx(inst);
634 
635 	err = crypto_grab_cipher(spawn, shash_crypto_instance(inst),
636 				 crypto_attr_alg_name(tb[1]), 0, mask);
637 	if (err)
638 		goto err_free_inst;
639 	alg = crypto_spawn_cipher_alg(spawn);
640 
641 	err = -EINVAL;
642 	if (alg->cra_blocksize != VMAC_NONCEBYTES)
643 		goto err_free_inst;
644 
645 	err = crypto_inst_setname(shash_crypto_instance(inst), tmpl->name, alg);
646 	if (err)
647 		goto err_free_inst;
648 
649 	inst->alg.base.cra_priority = alg->cra_priority;
650 	inst->alg.base.cra_blocksize = alg->cra_blocksize;
651 	inst->alg.base.cra_alignmask = alg->cra_alignmask;
652 
653 	inst->alg.base.cra_ctxsize = sizeof(struct vmac_tfm_ctx);
654 	inst->alg.base.cra_init = vmac_init_tfm;
655 	inst->alg.base.cra_exit = vmac_exit_tfm;
656 
657 	inst->alg.descsize = sizeof(struct vmac_desc_ctx);
658 	inst->alg.digestsize = VMAC_TAG_LEN / 8;
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_instance;
665 
666 	err = shash_register_instance(tmpl, inst);
667 	if (err) {
668 err_free_inst:
669 		shash_free_singlespawn_instance(inst);
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(&vmac64_tmpl);
683 }
684 
685 static void __exit vmac_module_exit(void)
686 {
687 	crypto_unregister_template(&vmac64_tmpl);
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