2 * VMAC: Message Authentication Code using Universal Hashing
4 * Reference: https://tools.ietf.org/html/draft-krovetz-vmac-01
6 * Copyright (c) 2009, Intel Corporation.
7 * Copyright (c) 2018, Google Inc.
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.
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
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.
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
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/cipher.h>
40 #include <crypto/internal/hash.h>
43 * User definable settings.
45 #define VMAC_TAG_LEN 64
46 #define VMAC_KEY_SIZE 128/* Must be 128, 192 or 256 */
47 #define VMAC_KEY_LEN (VMAC_KEY_SIZE/8)
48 #define VMAC_NHBYTES 128/* Must 2^i for any 3 < i < 13 Standard = 128*/
49 #define VMAC_NONCEBYTES 16
51 /* per-transform (per-key) context */
53 struct crypto_cipher *cipher;
54 u64 nhkey[(VMAC_NHBYTES/8)+2*(VMAC_TAG_LEN/64-1)];
55 u64 polykey[2*VMAC_TAG_LEN/64];
56 u64 l3key[2*VMAC_TAG_LEN/64];
59 /* per-request context */
60 struct vmac_desc_ctx {
62 u8 partial[VMAC_NHBYTES]; /* partial block */
63 __le64 partial_words[VMAC_NHBYTES / 8];
65 unsigned int partial_size; /* size of the partial block */
66 bool first_block_processed;
67 u64 polytmp[2*VMAC_TAG_LEN/64]; /* running total of L2-hash */
69 u8 bytes[VMAC_NONCEBYTES];
70 __be64 pads[VMAC_NONCEBYTES / 8];
72 unsigned int nonce_size; /* nonce bytes filled so far */
78 #define UINT64_C(x) x##ULL
79 static const u64 p64 = UINT64_C(0xfffffffffffffeff); /* 2^64 - 257 prime */
80 static const u64 m62 = UINT64_C(0x3fffffffffffffff); /* 62-bit mask */
81 static const u64 m63 = UINT64_C(0x7fffffffffffffff); /* 63-bit mask */
82 static const u64 m64 = UINT64_C(0xffffffffffffffff); /* 64-bit mask */
83 static const u64 mpoly = UINT64_C(0x1fffffff1fffffff); /* Poly key mask */
85 #define pe64_to_cpup le64_to_cpup /* Prefer little endian */
87 #ifdef __LITTLE_ENDIAN
96 * The following routines are used in this implementation. They are
97 * written via macros to simulate zero-overhead call-by-reference.
99 * MUL64: 64x64->128-bit multiplication
100 * PMUL64: assumes top bits cleared on inputs
101 * ADD128: 128x128->128-bit addition
104 #define ADD128(rh, rl, ih, il) \
113 #define MUL32(i1, i2) ((u64)(u32)(i1)*(u32)(i2))
115 #define PMUL64(rh, rl, i1, i2) /* Assumes m doesn't overflow */ \
117 u64 _i1 = (i1), _i2 = (i2); \
118 u64 m = MUL32(_i1, _i2>>32) + MUL32(_i1>>32, _i2); \
119 rh = MUL32(_i1>>32, _i2>>32); \
120 rl = MUL32(_i1, _i2); \
121 ADD128(rh, rl, (m >> 32), (m << 32)); \
124 #define MUL64(rh, rl, i1, i2) \
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 << 32)); \
132 ADD128(rh, rl, (m2 >> 32), (m2 << 32)); \
136 * For highest performance the L1 NH and L2 polynomial hashes should be
137 * carefully implemented to take advantage of one's target architecture.
138 * Here these two hash functions are defined multiple time; once for
139 * 64-bit architectures, once for 32-bit SSE2 architectures, and once
140 * for the rest (32-bit) architectures.
141 * For each, nh_16 *must* be defined (works on multiples of 16 bytes).
142 * Optionally, nh_vmac_nhbytes can be defined (for multiples of
143 * VMAC_NHBYTES), and nh_16_2 and nh_vmac_nhbytes_2 (versions that do two
144 * NH computations at once).
149 #define nh_16(mp, kp, nw, rh, rl) \
153 for (i = 0; i < nw; i += 2) { \
154 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
155 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
156 ADD128(rh, rl, th, tl); \
160 #define nh_16_2(mp, kp, nw, rh, rl, rh1, rl1) \
163 rh1 = rl1 = rh = rl = 0; \
164 for (i = 0; i < nw; i += 2) { \
165 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
166 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
167 ADD128(rh, rl, th, tl); \
168 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
169 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
170 ADD128(rh1, rl1, th, tl); \
174 #if (VMAC_NHBYTES >= 64) /* These versions do 64-bytes of message at a time */
175 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
179 for (i = 0; i < nw; i += 8) { \
180 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
181 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
182 ADD128(rh, rl, th, tl); \
183 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
184 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
185 ADD128(rh, rl, th, tl); \
186 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
187 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
188 ADD128(rh, rl, th, tl); \
189 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
190 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
191 ADD128(rh, rl, th, tl); \
195 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh1, rl1) \
198 rh1 = rl1 = rh = rl = 0; \
199 for (i = 0; i < nw; i += 8) { \
200 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
201 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
202 ADD128(rh, rl, th, tl); \
203 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
204 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
205 ADD128(rh1, rl1, th, tl); \
206 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
207 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
208 ADD128(rh, rl, th, tl); \
209 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+4], \
210 pe64_to_cpup((mp)+i+3)+(kp)[i+5]); \
211 ADD128(rh1, rl1, th, tl); \
212 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
213 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
214 ADD128(rh, rl, th, tl); \
215 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+6], \
216 pe64_to_cpup((mp)+i+5)+(kp)[i+7]); \
217 ADD128(rh1, rl1, th, tl); \
218 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
219 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
220 ADD128(rh, rl, th, tl); \
221 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+8], \
222 pe64_to_cpup((mp)+i+7)+(kp)[i+9]); \
223 ADD128(rh1, rl1, th, tl); \
228 #define poly_step(ah, al, kh, kl, mh, ml) \
230 u64 t1h, t1l, t2h, t2l, t3h, t3l, z = 0; \
231 /* compute ab*cd, put bd into result registers */ \
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) need summing */ \
241 /* first add the high registers, carrying into t2h */ \
242 ADD128(t2h, ah, z, t2l); \
243 /* double t2h and add top bit of ah */ \
244 t2h = 2 * t2h + (ah >> 63); \
246 /* now add the low registers */ \
247 ADD128(ah, al, mh, ml); \
248 ADD128(ah, al, z, t2h); \
251 #else /* ! CONFIG_64BIT */
254 #define nh_16(mp, kp, nw, rh, rl) \
256 u64 t1, t2, m1, m2, t; \
259 for (i = 0; i < nw; i += 2) { \
260 t1 = pe64_to_cpup(mp+i) + kp[i]; \
261 t2 = pe64_to_cpup(mp+i+1) + kp[i+1]; \
262 m2 = MUL32(t1 >> 32, t2); \
263 m1 = MUL32(t1, t2 >> 32); \
264 ADD128(rh, rl, MUL32(t1 >> 32, t2 >> 32), \
266 rh += (u64)(u32)(m1 >> 32) \
268 t += (u64)(u32)m1 + (u32)m2; \
270 ADD128(rh, rl, (t >> 32), (t << 32)); \
274 static void poly_step_func(u64 *ahi, u64 *alo,
275 const u64 *kh, const u64 *kl,
276 const u64 *mh, const u64 *ml)
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))
302 t |= ((u64)((u32)p & 0x7fffffff)) << 32;
304 p += (u64)(((u32 *)ml)[INDEX_LOW]);
313 p += (u64)(((u32 *)ml)[INDEX_HIGH]);
320 *(u64 *)(alo) = (p << 32) | t2;
322 *(u64 *)(ahi) = p + t;
334 #define poly_step(ah, al, kh, kl, mh, ml) \
335 poly_step_func(&(ah), &(al), &(kh), &(kl), &(mh), &(ml))
337 #endif /* end of specialized NH and poly definitions */
339 /* At least nh_16 is defined. Defined others as needed here */
341 #define nh_16_2(mp, kp, nw, rh, rl, rh2, rl2) \
343 nh_16(mp, kp, nw, rh, rl); \
344 nh_16(mp, ((kp)+2), nw, rh2, rl2); \
347 #ifndef nh_vmac_nhbytes
348 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
349 nh_16(mp, kp, nw, rh, rl)
351 #ifndef nh_vmac_nhbytes_2
352 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh2, rl2) \
354 nh_vmac_nhbytes(mp, kp, nw, rh, rl); \
355 nh_vmac_nhbytes(mp, ((kp)+2), nw, rh2, rl2); \
359 static u64 l3hash(u64 p1, u64 p2, u64 k1, u64 k2, u64 len)
361 u64 rh, rl, t, z = 0;
363 /* fully reduce (p1,p2)+(len,0) mod p127 */
366 ADD128(p1, p2, len, t);
367 /* At this point, (p1,p2) is at most 2^127+(len<<64) */
368 t = (p1 > m63) + ((p1 == m63) && (p2 == m64));
369 ADD128(p1, p2, z, t);
372 /* compute (p1,p2)/(2^64-2^32) and (p1,p2)%(2^64-2^32) */
375 t += (u32)t > 0xfffffffeu;
379 /* compute (p1+k1)%p64 and (p2+k2)%p64 */
381 p1 += (0 - (p1 < k1)) & 257;
383 p2 += (0 - (p2 < k2)) & 257;
385 /* compute (p1+k1)*(p2+k2)%p64 */
386 MUL64(rh, rl, p1, p2);
388 ADD128(t, rl, z, rh);
390 ADD128(t, rl, z, rh);
393 rl += (0 - (rl < t)) & 257;
394 rl += (0 - (rl > p64-1)) & 257;
398 /* L1 and L2-hash one or more VMAC_NHBYTES-byte blocks */
399 static void vhash_blocks(const struct vmac_tfm_ctx *tctx,
400 struct vmac_desc_ctx *dctx,
401 const __le64 *mptr, unsigned int blocks)
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];
410 if (!dctx->first_block_processed) {
411 dctx->first_block_processed = true;
412 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
414 ADD128(ch, cl, rh, rl);
415 mptr += (VMAC_NHBYTES/sizeof(u64));
420 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
422 poly_step(ch, cl, pkh, pkl, rh, rl);
423 mptr += (VMAC_NHBYTES/sizeof(u64));
426 dctx->polytmp[0] = ch;
427 dctx->polytmp[1] = cl;
430 static int vmac_setkey(struct crypto_shash *tfm,
431 const u8 *key, unsigned int keylen)
433 struct vmac_tfm_ctx *tctx = crypto_shash_ctx(tfm);
439 if (keylen != VMAC_KEY_LEN)
442 err = crypto_cipher_setkey(tctx->cipher, key, keylen);
448 for (i = 0; i < ARRAY_SIZE(tctx->nhkey); i += 2) {
449 crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
450 tctx->nhkey[i] = be64_to_cpu(out[0]);
451 tctx->nhkey[i+1] = be64_to_cpu(out[1]);
458 for (i = 0; i < ARRAY_SIZE(tctx->polykey); i += 2) {
459 crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
460 tctx->polykey[i] = be64_to_cpu(out[0]) & mpoly;
461 tctx->polykey[i+1] = be64_to_cpu(out[1]) & mpoly;
468 for (i = 0; i < ARRAY_SIZE(tctx->l3key); i += 2) {
470 crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
471 tctx->l3key[i] = be64_to_cpu(out[0]);
472 tctx->l3key[i+1] = be64_to_cpu(out[1]);
474 } while (tctx->l3key[i] >= p64 || tctx->l3key[i+1] >= p64);
480 static int vmac_init(struct shash_desc *desc)
482 const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
483 struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
485 dctx->partial_size = 0;
486 dctx->first_block_processed = false;
487 memcpy(dctx->polytmp, tctx->polykey, sizeof(dctx->polytmp));
488 dctx->nonce_size = 0;
492 static int vmac_update(struct shash_desc *desc, const u8 *p, unsigned int len)
494 const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
495 struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
498 /* Nonce is passed as first VMAC_NONCEBYTES bytes of data */
499 if (dctx->nonce_size < VMAC_NONCEBYTES) {
500 n = min(len, VMAC_NONCEBYTES - dctx->nonce_size);
501 memcpy(&dctx->nonce.bytes[dctx->nonce_size], p, n);
502 dctx->nonce_size += n;
507 if (dctx->partial_size) {
508 n = min(len, VMAC_NHBYTES - dctx->partial_size);
509 memcpy(&dctx->partial[dctx->partial_size], p, n);
510 dctx->partial_size += n;
513 if (dctx->partial_size == VMAC_NHBYTES) {
514 vhash_blocks(tctx, dctx, dctx->partial_words, 1);
515 dctx->partial_size = 0;
519 if (len >= VMAC_NHBYTES) {
520 n = round_down(len, VMAC_NHBYTES);
521 /* TODO: 'p' may be misaligned here */
522 vhash_blocks(tctx, dctx, (const __le64 *)p, n / VMAC_NHBYTES);
528 memcpy(dctx->partial, p, len);
529 dctx->partial_size = len;
535 static u64 vhash_final(const struct vmac_tfm_ctx *tctx,
536 struct vmac_desc_ctx *dctx)
538 unsigned int partial = dctx->partial_size;
539 u64 ch = dctx->polytmp[0];
540 u64 cl = dctx->polytmp[1];
542 /* L1 and L2-hash the final block if needed */
544 /* Zero-pad to next 128-bit boundary */
545 unsigned int n = round_up(partial, 16);
548 memset(&dctx->partial[partial], 0, n - partial);
549 nh_16(dctx->partial_words, tctx->nhkey, n / 8, rh, rl);
551 if (dctx->first_block_processed)
552 poly_step(ch, cl, tctx->polykey[0], tctx->polykey[1],
555 ADD128(ch, cl, rh, rl);
558 /* L3-hash the 128-bit output of L2-hash */
559 return l3hash(ch, cl, tctx->l3key[0], tctx->l3key[1], partial * 8);
562 static int vmac_final(struct shash_desc *desc, u8 *out)
564 const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
565 struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
569 if (dctx->nonce_size != VMAC_NONCEBYTES)
573 * The VMAC specification requires a nonce at least 1 bit shorter than
574 * the block cipher's block length, so we actually only accept a 127-bit
575 * nonce. We define the unused bit to be the first one and require that
576 * it be 0, so the needed prepending of a 0 bit is implicit.
578 if (dctx->nonce.bytes[0] & 0x80)
581 /* Finish calculating the VHASH of the message */
582 hash = vhash_final(tctx, dctx);
584 /* Generate pseudorandom pad by encrypting the nonce */
585 BUILD_BUG_ON(VMAC_NONCEBYTES != 2 * (VMAC_TAG_LEN / 8));
586 index = dctx->nonce.bytes[VMAC_NONCEBYTES - 1] & 1;
587 dctx->nonce.bytes[VMAC_NONCEBYTES - 1] &= ~1;
588 crypto_cipher_encrypt_one(tctx->cipher, dctx->nonce.bytes,
590 pad = be64_to_cpu(dctx->nonce.pads[index]);
592 /* The VMAC is the sum of VHASH and the pseudorandom pad */
593 put_unaligned_be64(hash + pad, out);
597 static int vmac_init_tfm(struct crypto_tfm *tfm)
599 struct crypto_instance *inst = crypto_tfm_alg_instance(tfm);
600 struct crypto_cipher_spawn *spawn = crypto_instance_ctx(inst);
601 struct vmac_tfm_ctx *tctx = crypto_tfm_ctx(tfm);
602 struct crypto_cipher *cipher;
604 cipher = crypto_spawn_cipher(spawn);
606 return PTR_ERR(cipher);
608 tctx->cipher = cipher;
612 static void vmac_exit_tfm(struct crypto_tfm *tfm)
614 struct vmac_tfm_ctx *tctx = crypto_tfm_ctx(tfm);
616 crypto_free_cipher(tctx->cipher);
619 static int vmac_create(struct crypto_template *tmpl, struct rtattr **tb)
621 struct shash_instance *inst;
622 struct crypto_cipher_spawn *spawn;
623 struct crypto_alg *alg;
627 err = crypto_check_attr_type(tb, CRYPTO_ALG_TYPE_SHASH, &mask);
631 inst = kzalloc(sizeof(*inst) + sizeof(*spawn), GFP_KERNEL);
634 spawn = shash_instance_ctx(inst);
636 err = crypto_grab_cipher(spawn, shash_crypto_instance(inst),
637 crypto_attr_alg_name(tb[1]), 0, mask);
640 alg = crypto_spawn_cipher_alg(spawn);
643 if (alg->cra_blocksize != VMAC_NONCEBYTES)
646 err = crypto_inst_setname(shash_crypto_instance(inst), tmpl->name, alg);
650 inst->alg.base.cra_priority = alg->cra_priority;
651 inst->alg.base.cra_blocksize = alg->cra_blocksize;
652 inst->alg.base.cra_alignmask = alg->cra_alignmask;
654 inst->alg.base.cra_ctxsize = sizeof(struct vmac_tfm_ctx);
655 inst->alg.base.cra_init = vmac_init_tfm;
656 inst->alg.base.cra_exit = vmac_exit_tfm;
658 inst->alg.descsize = sizeof(struct vmac_desc_ctx);
659 inst->alg.digestsize = VMAC_TAG_LEN / 8;
660 inst->alg.init = vmac_init;
661 inst->alg.update = vmac_update;
662 inst->alg.final = vmac_final;
663 inst->alg.setkey = vmac_setkey;
665 inst->free = shash_free_singlespawn_instance;
667 err = shash_register_instance(tmpl, inst);
670 shash_free_singlespawn_instance(inst);
675 static struct crypto_template vmac64_tmpl = {
677 .create = vmac_create,
678 .module = THIS_MODULE,
681 static int __init vmac_module_init(void)
683 return crypto_register_template(&vmac64_tmpl);
686 static void __exit vmac_module_exit(void)
688 crypto_unregister_template(&vmac64_tmpl);
691 subsys_initcall(vmac_module_init);
692 module_exit(vmac_module_exit);
694 MODULE_LICENSE("GPL");
695 MODULE_DESCRIPTION("VMAC hash algorithm");
696 MODULE_ALIAS_CRYPTO("vmac64");
697 MODULE_IMPORT_NS(CRYPTO_INTERNAL);