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/hash.h>
42 * User definable settings.
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*/
49 /* per-transform (per-key) context */
51 struct crypto_cipher *cipher;
52 u64 nhkey[(VMAC_NHBYTES/8)+2*(VMAC_TAG_LEN/64-1)];
53 u64 polykey[2*VMAC_TAG_LEN/64];
54 u64 l3key[2*VMAC_TAG_LEN/64];
57 /* per-request context */
58 struct vmac_desc_ctx {
60 u8 partial[VMAC_NHBYTES]; /* partial block */
61 __le64 partial_words[VMAC_NHBYTES / 8];
63 unsigned int partial_size; /* size of the partial block */
64 bool first_block_processed;
65 u64 polytmp[2*VMAC_TAG_LEN/64]; /* running total of L2-hash */
71 #define UINT64_C(x) x##ULL
72 static const u64 p64 = UINT64_C(0xfffffffffffffeff); /* 2^64 - 257 prime */
73 static const u64 m62 = UINT64_C(0x3fffffffffffffff); /* 62-bit mask */
74 static const u64 m63 = UINT64_C(0x7fffffffffffffff); /* 63-bit mask */
75 static const u64 m64 = UINT64_C(0xffffffffffffffff); /* 64-bit mask */
76 static const u64 mpoly = UINT64_C(0x1fffffff1fffffff); /* Poly key mask */
78 #define pe64_to_cpup le64_to_cpup /* Prefer little endian */
80 #ifdef __LITTLE_ENDIAN
89 * The following routines are used in this implementation. They are
90 * written via macros to simulate zero-overhead call-by-reference.
92 * MUL64: 64x64->128-bit multiplication
93 * PMUL64: assumes top bits cleared on inputs
94 * ADD128: 128x128->128-bit addition
97 #define ADD128(rh, rl, ih, il) \
106 #define MUL32(i1, i2) ((u64)(u32)(i1)*(u32)(i2))
108 #define PMUL64(rh, rl, i1, i2) /* Assumes m doesn't overflow */ \
110 u64 _i1 = (i1), _i2 = (i2); \
111 u64 m = MUL32(_i1, _i2>>32) + MUL32(_i1>>32, _i2); \
112 rh = MUL32(_i1>>32, _i2>>32); \
113 rl = MUL32(_i1, _i2); \
114 ADD128(rh, rl, (m >> 32), (m << 32)); \
117 #define MUL64(rh, rl, i1, i2) \
119 u64 _i1 = (i1), _i2 = (i2); \
120 u64 m1 = MUL32(_i1, _i2>>32); \
121 u64 m2 = MUL32(_i1>>32, _i2); \
122 rh = MUL32(_i1>>32, _i2>>32); \
123 rl = MUL32(_i1, _i2); \
124 ADD128(rh, rl, (m1 >> 32), (m1 << 32)); \
125 ADD128(rh, rl, (m2 >> 32), (m2 << 32)); \
129 * For highest performance the L1 NH and L2 polynomial hashes should be
130 * carefully implemented to take advantage of one's target architecture.
131 * Here these two hash functions are defined multiple time; once for
132 * 64-bit architectures, once for 32-bit SSE2 architectures, and once
133 * for the rest (32-bit) architectures.
134 * For each, nh_16 *must* be defined (works on multiples of 16 bytes).
135 * Optionally, nh_vmac_nhbytes can be defined (for multiples of
136 * VMAC_NHBYTES), and nh_16_2 and nh_vmac_nhbytes_2 (versions that do two
137 * NH computations at once).
142 #define nh_16(mp, kp, nw, rh, rl) \
146 for (i = 0; i < nw; i += 2) { \
147 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
148 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
149 ADD128(rh, rl, th, tl); \
153 #define nh_16_2(mp, kp, nw, rh, rl, rh1, rl1) \
156 rh1 = rl1 = rh = rl = 0; \
157 for (i = 0; i < nw; i += 2) { \
158 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
159 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
160 ADD128(rh, rl, th, tl); \
161 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
162 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
163 ADD128(rh1, rl1, th, tl); \
167 #if (VMAC_NHBYTES >= 64) /* These versions do 64-bytes of message at a time */
168 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
172 for (i = 0; i < nw; i += 8) { \
173 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
174 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
175 ADD128(rh, rl, th, tl); \
176 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
177 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
178 ADD128(rh, rl, th, tl); \
179 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
180 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
181 ADD128(rh, rl, th, tl); \
182 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
183 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
184 ADD128(rh, rl, th, tl); \
188 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh1, rl1) \
191 rh1 = rl1 = rh = rl = 0; \
192 for (i = 0; i < nw; i += 8) { \
193 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
194 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
195 ADD128(rh, rl, th, tl); \
196 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
197 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
198 ADD128(rh1, rl1, th, tl); \
199 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
200 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
201 ADD128(rh, rl, th, tl); \
202 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+4], \
203 pe64_to_cpup((mp)+i+3)+(kp)[i+5]); \
204 ADD128(rh1, rl1, th, tl); \
205 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
206 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
207 ADD128(rh, rl, th, tl); \
208 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+6], \
209 pe64_to_cpup((mp)+i+5)+(kp)[i+7]); \
210 ADD128(rh1, rl1, th, tl); \
211 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
212 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
213 ADD128(rh, rl, th, tl); \
214 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+8], \
215 pe64_to_cpup((mp)+i+7)+(kp)[i+9]); \
216 ADD128(rh1, rl1, th, tl); \
221 #define poly_step(ah, al, kh, kl, mh, ml) \
223 u64 t1h, t1l, t2h, t2l, t3h, t3l, z = 0; \
224 /* compute ab*cd, put bd into result registers */ \
225 PMUL64(t3h, t3l, al, kh); \
226 PMUL64(t2h, t2l, ah, kl); \
227 PMUL64(t1h, t1l, ah, 2*kh); \
228 PMUL64(ah, al, al, kl); \
229 /* add 2 * ac to result */ \
230 ADD128(ah, al, t1h, t1l); \
231 /* add together ad + bc */ \
232 ADD128(t2h, t2l, t3h, t3l); \
233 /* now (ah,al), (t2l,2*t2h) need summing */ \
234 /* first add the high registers, carrying into t2h */ \
235 ADD128(t2h, ah, z, t2l); \
236 /* double t2h and add top bit of ah */ \
237 t2h = 2 * t2h + (ah >> 63); \
239 /* now add the low registers */ \
240 ADD128(ah, al, mh, ml); \
241 ADD128(ah, al, z, t2h); \
244 #else /* ! CONFIG_64BIT */
247 #define nh_16(mp, kp, nw, rh, rl) \
249 u64 t1, t2, m1, m2, t; \
252 for (i = 0; i < nw; i += 2) { \
253 t1 = pe64_to_cpup(mp+i) + kp[i]; \
254 t2 = pe64_to_cpup(mp+i+1) + kp[i+1]; \
255 m2 = MUL32(t1 >> 32, t2); \
256 m1 = MUL32(t1, t2 >> 32); \
257 ADD128(rh, rl, MUL32(t1 >> 32, t2 >> 32), \
259 rh += (u64)(u32)(m1 >> 32) \
261 t += (u64)(u32)m1 + (u32)m2; \
263 ADD128(rh, rl, (t >> 32), (t << 32)); \
267 static void poly_step_func(u64 *ahi, u64 *alo,
268 const u64 *kh, const u64 *kl,
269 const u64 *mh, const u64 *ml)
271 #define a0 (*(((u32 *)alo)+INDEX_LOW))
272 #define a1 (*(((u32 *)alo)+INDEX_HIGH))
273 #define a2 (*(((u32 *)ahi)+INDEX_LOW))
274 #define a3 (*(((u32 *)ahi)+INDEX_HIGH))
275 #define k0 (*(((u32 *)kl)+INDEX_LOW))
276 #define k1 (*(((u32 *)kl)+INDEX_HIGH))
277 #define k2 (*(((u32 *)kh)+INDEX_LOW))
278 #define k3 (*(((u32 *)kh)+INDEX_HIGH))
295 t |= ((u64)((u32)p & 0x7fffffff)) << 32;
297 p += (u64)(((u32 *)ml)[INDEX_LOW]);
306 p += (u64)(((u32 *)ml)[INDEX_HIGH]);
313 *(u64 *)(alo) = (p << 32) | t2;
315 *(u64 *)(ahi) = p + t;
327 #define poly_step(ah, al, kh, kl, mh, ml) \
328 poly_step_func(&(ah), &(al), &(kh), &(kl), &(mh), &(ml))
330 #endif /* end of specialized NH and poly definitions */
332 /* At least nh_16 is defined. Defined others as needed here */
334 #define nh_16_2(mp, kp, nw, rh, rl, rh2, rl2) \
336 nh_16(mp, kp, nw, rh, rl); \
337 nh_16(mp, ((kp)+2), nw, rh2, rl2); \
340 #ifndef nh_vmac_nhbytes
341 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
342 nh_16(mp, kp, nw, rh, rl)
344 #ifndef nh_vmac_nhbytes_2
345 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh2, rl2) \
347 nh_vmac_nhbytes(mp, kp, nw, rh, rl); \
348 nh_vmac_nhbytes(mp, ((kp)+2), nw, rh2, rl2); \
352 static u64 l3hash(u64 p1, u64 p2, u64 k1, u64 k2, u64 len)
354 u64 rh, rl, t, z = 0;
356 /* fully reduce (p1,p2)+(len,0) mod p127 */
359 ADD128(p1, p2, len, t);
360 /* At this point, (p1,p2) is at most 2^127+(len<<64) */
361 t = (p1 > m63) + ((p1 == m63) && (p2 == m64));
362 ADD128(p1, p2, z, t);
365 /* compute (p1,p2)/(2^64-2^32) and (p1,p2)%(2^64-2^32) */
368 t += (u32)t > 0xfffffffeu;
372 /* compute (p1+k1)%p64 and (p2+k2)%p64 */
374 p1 += (0 - (p1 < k1)) & 257;
376 p2 += (0 - (p2 < k2)) & 257;
378 /* compute (p1+k1)*(p2+k2)%p64 */
379 MUL64(rh, rl, p1, p2);
381 ADD128(t, rl, z, rh);
383 ADD128(t, rl, z, rh);
386 rl += (0 - (rl < t)) & 257;
387 rl += (0 - (rl > p64-1)) & 257;
391 /* L1 and L2-hash one or more VMAC_NHBYTES-byte blocks */
392 static void vhash_blocks(const struct vmac_tfm_ctx *tctx,
393 struct vmac_desc_ctx *dctx,
394 const __le64 *mptr, unsigned int blocks)
396 const u64 *kptr = tctx->nhkey;
397 const u64 pkh = tctx->polykey[0];
398 const u64 pkl = tctx->polykey[1];
399 u64 ch = dctx->polytmp[0];
400 u64 cl = dctx->polytmp[1];
403 if (!dctx->first_block_processed) {
404 dctx->first_block_processed = true;
405 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
407 ADD128(ch, cl, rh, rl);
408 mptr += (VMAC_NHBYTES/sizeof(u64));
413 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
415 poly_step(ch, cl, pkh, pkl, rh, rl);
416 mptr += (VMAC_NHBYTES/sizeof(u64));
419 dctx->polytmp[0] = ch;
420 dctx->polytmp[1] = cl;
423 static int vmac_setkey(struct crypto_shash *tfm,
424 const u8 *key, unsigned int keylen)
426 struct vmac_tfm_ctx *tctx = crypto_shash_ctx(tfm);
432 if (keylen != VMAC_KEY_LEN) {
433 crypto_shash_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN);
437 err = crypto_cipher_setkey(tctx->cipher, key, keylen);
443 for (i = 0; i < ARRAY_SIZE(tctx->nhkey); i += 2) {
444 crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
445 tctx->nhkey[i] = be64_to_cpu(out[0]);
446 tctx->nhkey[i+1] = be64_to_cpu(out[1]);
453 for (i = 0; i < ARRAY_SIZE(tctx->polykey); i += 2) {
454 crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
455 tctx->polykey[i] = be64_to_cpu(out[0]) & mpoly;
456 tctx->polykey[i+1] = be64_to_cpu(out[1]) & mpoly;
463 for (i = 0; i < ARRAY_SIZE(tctx->l3key); i += 2) {
465 crypto_cipher_encrypt_one(tctx->cipher, (u8 *)out, in);
466 tctx->l3key[i] = be64_to_cpu(out[0]);
467 tctx->l3key[i+1] = be64_to_cpu(out[1]);
469 } while (tctx->l3key[i] >= p64 || tctx->l3key[i+1] >= p64);
475 static int vmac_init(struct shash_desc *desc)
477 const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
478 struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
480 dctx->partial_size = 0;
481 dctx->first_block_processed = false;
482 memcpy(dctx->polytmp, tctx->polykey, sizeof(dctx->polytmp));
486 static int vmac_update(struct shash_desc *desc, const u8 *p, unsigned int len)
488 const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
489 struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
492 if (dctx->partial_size) {
493 n = min(len, VMAC_NHBYTES - dctx->partial_size);
494 memcpy(&dctx->partial[dctx->partial_size], p, n);
495 dctx->partial_size += n;
498 if (dctx->partial_size == VMAC_NHBYTES) {
499 vhash_blocks(tctx, dctx, dctx->partial_words, 1);
500 dctx->partial_size = 0;
504 if (len >= VMAC_NHBYTES) {
505 n = round_down(len, VMAC_NHBYTES);
506 /* TODO: 'p' may be misaligned here */
507 vhash_blocks(tctx, dctx, (const __le64 *)p, n / VMAC_NHBYTES);
513 memcpy(dctx->partial, p, len);
514 dctx->partial_size = len;
520 static u64 vhash_final(const struct vmac_tfm_ctx *tctx,
521 struct vmac_desc_ctx *dctx)
523 unsigned int partial = dctx->partial_size;
524 u64 ch = dctx->polytmp[0];
525 u64 cl = dctx->polytmp[1];
527 /* L1 and L2-hash the final block if needed */
529 /* Zero-pad to next 128-bit boundary */
530 unsigned int n = round_up(partial, 16);
533 memset(&dctx->partial[partial], 0, n - partial);
534 nh_16(dctx->partial_words, tctx->nhkey, n / 8, rh, rl);
536 if (dctx->first_block_processed)
537 poly_step(ch, cl, tctx->polykey[0], tctx->polykey[1],
540 ADD128(ch, cl, rh, rl);
543 /* L3-hash the 128-bit output of L2-hash */
544 return l3hash(ch, cl, tctx->l3key[0], tctx->l3key[1], partial * 8);
547 static int vmac_final(struct shash_desc *desc, u8 *out)
549 const struct vmac_tfm_ctx *tctx = crypto_shash_ctx(desc->tfm);
550 struct vmac_desc_ctx *dctx = shash_desc_ctx(desc);
551 static const u8 nonce[16] = {}; /* TODO: this is insecure */
559 /* Finish calculating the VHASH of the message */
560 hash = vhash_final(tctx, dctx);
562 /* Generate pseudorandom pad by encrypting the nonce */
563 memcpy(&block, nonce, 16);
564 index = block.bytes[15] & 1;
565 block.bytes[15] &= ~1;
566 crypto_cipher_encrypt_one(tctx->cipher, block.bytes, block.bytes);
567 pad = be64_to_cpu(block.pads[index]);
569 /* The VMAC is the sum of VHASH and the pseudorandom pad */
570 put_unaligned_le64(hash + pad, out);
574 static int vmac_init_tfm(struct crypto_tfm *tfm)
576 struct crypto_instance *inst = crypto_tfm_alg_instance(tfm);
577 struct crypto_spawn *spawn = crypto_instance_ctx(inst);
578 struct vmac_tfm_ctx *tctx = crypto_tfm_ctx(tfm);
579 struct crypto_cipher *cipher;
581 cipher = crypto_spawn_cipher(spawn);
583 return PTR_ERR(cipher);
585 tctx->cipher = cipher;
589 static void vmac_exit_tfm(struct crypto_tfm *tfm)
591 struct vmac_tfm_ctx *tctx = crypto_tfm_ctx(tfm);
593 crypto_free_cipher(tctx->cipher);
596 static int vmac_create(struct crypto_template *tmpl, struct rtattr **tb)
598 struct shash_instance *inst;
599 struct crypto_alg *alg;
602 err = crypto_check_attr_type(tb, CRYPTO_ALG_TYPE_SHASH);
606 alg = crypto_get_attr_alg(tb, CRYPTO_ALG_TYPE_CIPHER,
607 CRYPTO_ALG_TYPE_MASK);
612 if (alg->cra_blocksize != 16)
615 inst = shash_alloc_instance("vmac", alg);
620 err = crypto_init_spawn(shash_instance_ctx(inst), alg,
621 shash_crypto_instance(inst),
622 CRYPTO_ALG_TYPE_MASK);
626 inst->alg.base.cra_priority = alg->cra_priority;
627 inst->alg.base.cra_blocksize = alg->cra_blocksize;
628 inst->alg.base.cra_alignmask = alg->cra_alignmask;
630 inst->alg.base.cra_ctxsize = sizeof(struct vmac_tfm_ctx);
631 inst->alg.base.cra_init = vmac_init_tfm;
632 inst->alg.base.cra_exit = vmac_exit_tfm;
634 inst->alg.descsize = sizeof(struct vmac_desc_ctx);
635 inst->alg.digestsize = VMAC_TAG_LEN / 8;
636 inst->alg.init = vmac_init;
637 inst->alg.update = vmac_update;
638 inst->alg.final = vmac_final;
639 inst->alg.setkey = vmac_setkey;
641 err = shash_register_instance(tmpl, inst);
644 shash_free_instance(shash_crypto_instance(inst));
652 static struct crypto_template vmac_tmpl = {
654 .create = vmac_create,
655 .free = shash_free_instance,
656 .module = THIS_MODULE,
659 static int __init vmac_module_init(void)
661 return crypto_register_template(&vmac_tmpl);
664 static void __exit vmac_module_exit(void)
666 crypto_unregister_template(&vmac_tmpl);
669 module_init(vmac_module_init);
670 module_exit(vmac_module_exit);
672 MODULE_LICENSE("GPL");
673 MODULE_DESCRIPTION("VMAC hash algorithm");
674 MODULE_ALIAS_CRYPTO("vmac");