2 * Modified to interface to the Linux kernel
3 * Copyright (c) 2009, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
16 * Place - Suite 330, Boston, MA 02111-1307 USA.
19 /* --------------------------------------------------------------------------
20 * VMAC and VHASH Implementation by Ted Krovetz (tdk@acm.org) and Wei Dai.
21 * This implementation is herby placed in the public domain.
22 * The authors offers no warranty. Use at your own risk.
23 * Please send bug reports to the authors.
24 * Last modified: 17 APR 08, 1700 PDT
25 * ----------------------------------------------------------------------- */
27 #include <linux/init.h>
28 #include <linux/types.h>
29 #include <linux/crypto.h>
30 #include <linux/scatterlist.h>
31 #include <asm/byteorder.h>
32 #include <crypto/scatterwalk.h>
33 #include <crypto/vmac.h>
34 #include <crypto/internal/hash.h>
39 #define UINT64_C(x) x##ULL
40 const u64 p64 = UINT64_C(0xfffffffffffffeff); /* 2^64 - 257 prime */
41 const u64 m62 = UINT64_C(0x3fffffffffffffff); /* 62-bit mask */
42 const u64 m63 = UINT64_C(0x7fffffffffffffff); /* 63-bit mask */
43 const u64 m64 = UINT64_C(0xffffffffffffffff); /* 64-bit mask */
44 const u64 mpoly = UINT64_C(0x1fffffff1fffffff); /* Poly key mask */
46 #define pe64_to_cpup le64_to_cpup /* Prefer little endian */
48 #ifdef __LITTLE_ENDIAN
57 * The following routines are used in this implementation. They are
58 * written via macros to simulate zero-overhead call-by-reference.
60 * MUL64: 64x64->128-bit multiplication
61 * PMUL64: assumes top bits cleared on inputs
62 * ADD128: 128x128->128-bit addition
65 #define ADD128(rh, rl, ih, il) \
74 #define MUL32(i1, i2) ((u64)(u32)(i1)*(u32)(i2))
76 #define PMUL64(rh, rl, i1, i2) /* Assumes m doesn't overflow */ \
78 u64 _i1 = (i1), _i2 = (i2); \
79 u64 m = MUL32(_i1, _i2>>32) + MUL32(_i1>>32, _i2); \
80 rh = MUL32(_i1>>32, _i2>>32); \
81 rl = MUL32(_i1, _i2); \
82 ADD128(rh, rl, (m >> 32), (m << 32)); \
85 #define MUL64(rh, rl, i1, i2) \
87 u64 _i1 = (i1), _i2 = (i2); \
88 u64 m1 = MUL32(_i1, _i2>>32); \
89 u64 m2 = MUL32(_i1>>32, _i2); \
90 rh = MUL32(_i1>>32, _i2>>32); \
91 rl = MUL32(_i1, _i2); \
92 ADD128(rh, rl, (m1 >> 32), (m1 << 32)); \
93 ADD128(rh, rl, (m2 >> 32), (m2 << 32)); \
97 * For highest performance the L1 NH and L2 polynomial hashes should be
98 * carefully implemented to take advantage of one's target architechture.
99 * Here these two hash functions are defined multiple time; once for
100 * 64-bit architectures, once for 32-bit SSE2 architectures, and once
101 * for the rest (32-bit) architectures.
102 * For each, nh_16 *must* be defined (works on multiples of 16 bytes).
103 * Optionally, nh_vmac_nhbytes can be defined (for multiples of
104 * VMAC_NHBYTES), and nh_16_2 and nh_vmac_nhbytes_2 (versions that do two
105 * NH computations at once).
110 #define nh_16(mp, kp, nw, rh, rl) \
114 for (i = 0; i < nw; i += 2) { \
115 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
116 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
117 ADD128(rh, rl, th, tl); \
121 #define nh_16_2(mp, kp, nw, rh, rl, rh1, rl1) \
124 rh1 = rl1 = rh = rl = 0; \
125 for (i = 0; i < nw; i += 2) { \
126 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
127 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
128 ADD128(rh, rl, th, tl); \
129 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
130 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
131 ADD128(rh1, rl1, th, tl); \
135 #if (VMAC_NHBYTES >= 64) /* These versions do 64-bytes of message at a time */
136 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
140 for (i = 0; i < nw; i += 8) { \
141 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
142 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
143 ADD128(rh, rl, th, tl); \
144 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
145 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
146 ADD128(rh, rl, th, tl); \
147 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
148 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
149 ADD128(rh, rl, th, tl); \
150 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
151 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
152 ADD128(rh, rl, th, tl); \
156 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh1, rl1) \
159 rh1 = rl1 = rh = rl = 0; \
160 for (i = 0; i < nw; i += 8) { \
161 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
162 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
163 ADD128(rh, rl, th, tl); \
164 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
165 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
166 ADD128(rh1, rl1, th, tl); \
167 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
168 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
169 ADD128(rh, rl, th, tl); \
170 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+4], \
171 pe64_to_cpup((mp)+i+3)+(kp)[i+5]); \
172 ADD128(rh1, rl1, th, tl); \
173 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
174 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
175 ADD128(rh, rl, th, tl); \
176 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+6], \
177 pe64_to_cpup((mp)+i+5)+(kp)[i+7]); \
178 ADD128(rh1, rl1, th, tl); \
179 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
180 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
181 ADD128(rh, rl, th, tl); \
182 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+8], \
183 pe64_to_cpup((mp)+i+7)+(kp)[i+9]); \
184 ADD128(rh1, rl1, th, tl); \
189 #define poly_step(ah, al, kh, kl, mh, ml) \
191 u64 t1h, t1l, t2h, t2l, t3h, t3l, z = 0; \
192 /* compute ab*cd, put bd into result registers */ \
193 PMUL64(t3h, t3l, al, kh); \
194 PMUL64(t2h, t2l, ah, kl); \
195 PMUL64(t1h, t1l, ah, 2*kh); \
196 PMUL64(ah, al, al, kl); \
197 /* add 2 * ac to result */ \
198 ADD128(ah, al, t1h, t1l); \
199 /* add together ad + bc */ \
200 ADD128(t2h, t2l, t3h, t3l); \
201 /* now (ah,al), (t2l,2*t2h) need summing */ \
202 /* first add the high registers, carrying into t2h */ \
203 ADD128(t2h, ah, z, t2l); \
204 /* double t2h and add top bit of ah */ \
205 t2h = 2 * t2h + (ah >> 63); \
207 /* now add the low registers */ \
208 ADD128(ah, al, mh, ml); \
209 ADD128(ah, al, z, t2h); \
212 #else /* ! CONFIG_64BIT */
215 #define nh_16(mp, kp, nw, rh, rl) \
217 u64 t1, t2, m1, m2, t; \
220 for (i = 0; i < nw; i += 2) { \
221 t1 = pe64_to_cpup(mp+i) + kp[i]; \
222 t2 = pe64_to_cpup(mp+i+1) + kp[i+1]; \
223 m2 = MUL32(t1 >> 32, t2); \
224 m1 = MUL32(t1, t2 >> 32); \
225 ADD128(rh, rl, MUL32(t1 >> 32, t2 >> 32), \
227 rh += (u64)(u32)(m1 >> 32) \
229 t += (u64)(u32)m1 + (u32)m2; \
231 ADD128(rh, rl, (t >> 32), (t << 32)); \
235 static void poly_step_func(u64 *ahi, u64 *alo,
236 const u64 *kh, const u64 *kl,
237 const u64 *mh, const u64 *ml)
239 #define a0 (*(((u32 *)alo)+INDEX_LOW))
240 #define a1 (*(((u32 *)alo)+INDEX_HIGH))
241 #define a2 (*(((u32 *)ahi)+INDEX_LOW))
242 #define a3 (*(((u32 *)ahi)+INDEX_HIGH))
243 #define k0 (*(((u32 *)kl)+INDEX_LOW))
244 #define k1 (*(((u32 *)kl)+INDEX_HIGH))
245 #define k2 (*(((u32 *)kh)+INDEX_LOW))
246 #define k3 (*(((u32 *)kh)+INDEX_HIGH))
263 t |= ((u64)((u32)p & 0x7fffffff)) << 32;
265 p += (u64)(((u32 *)ml)[INDEX_LOW]);
274 p += (u64)(((u32 *)ml)[INDEX_HIGH]);
281 *(u64 *)(alo) = (p << 32) | t2;
283 *(u64 *)(ahi) = p + t;
295 #define poly_step(ah, al, kh, kl, mh, ml) \
296 poly_step_func(&(ah), &(al), &(kh), &(kl), &(mh), &(ml))
298 #endif /* end of specialized NH and poly definitions */
300 /* At least nh_16 is defined. Defined others as needed here */
302 #define nh_16_2(mp, kp, nw, rh, rl, rh2, rl2) \
304 nh_16(mp, kp, nw, rh, rl); \
305 nh_16(mp, ((kp)+2), nw, rh2, rl2); \
308 #ifndef nh_vmac_nhbytes
309 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
310 nh_16(mp, kp, nw, rh, rl)
312 #ifndef nh_vmac_nhbytes_2
313 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh2, rl2) \
315 nh_vmac_nhbytes(mp, kp, nw, rh, rl); \
316 nh_vmac_nhbytes(mp, ((kp)+2), nw, rh2, rl2); \
320 static void vhash_abort(struct vmac_ctx *ctx)
322 ctx->polytmp[0] = ctx->polykey[0] ;
323 ctx->polytmp[1] = ctx->polykey[1] ;
324 ctx->first_block_processed = 0;
327 static u64 l3hash(u64 p1, u64 p2, u64 k1, u64 k2, u64 len)
329 u64 rh, rl, t, z = 0;
331 /* fully reduce (p1,p2)+(len,0) mod p127 */
334 ADD128(p1, p2, len, t);
335 /* At this point, (p1,p2) is at most 2^127+(len<<64) */
336 t = (p1 > m63) + ((p1 == m63) && (p2 == m64));
337 ADD128(p1, p2, z, t);
340 /* compute (p1,p2)/(2^64-2^32) and (p1,p2)%(2^64-2^32) */
343 t += (u32)t > 0xfffffffeu;
347 /* compute (p1+k1)%p64 and (p2+k2)%p64 */
349 p1 += (0 - (p1 < k1)) & 257;
351 p2 += (0 - (p2 < k2)) & 257;
353 /* compute (p1+k1)*(p2+k2)%p64 */
354 MUL64(rh, rl, p1, p2);
356 ADD128(t, rl, z, rh);
358 ADD128(t, rl, z, rh);
361 rl += (0 - (rl < t)) & 257;
362 rl += (0 - (rl > p64-1)) & 257;
366 static void vhash_update(const unsigned char *m,
367 unsigned int mbytes, /* Pos multiple of VMAC_NHBYTES */
368 struct vmac_ctx *ctx)
371 const u64 *kptr = (u64 *)ctx->nhkey;
374 u64 pkh = ctx->polykey[0];
375 u64 pkl = ctx->polykey[1];
378 i = mbytes / VMAC_NHBYTES; /* Must be non-zero */
380 ch = ctx->polytmp[0];
381 cl = ctx->polytmp[1];
383 if (!ctx->first_block_processed) {
384 ctx->first_block_processed = 1;
385 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
387 ADD128(ch, cl, rh, rl);
388 mptr += (VMAC_NHBYTES/sizeof(u64));
393 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
395 poly_step(ch, cl, pkh, pkl, rh, rl);
396 mptr += (VMAC_NHBYTES/sizeof(u64));
399 ctx->polytmp[0] = ch;
400 ctx->polytmp[1] = cl;
403 static u64 vhash(unsigned char m[], unsigned int mbytes,
404 u64 *tagl, struct vmac_ctx *ctx)
407 const u64 *kptr = (u64 *)ctx->nhkey;
410 u64 pkh = ctx->polykey[0];
411 u64 pkl = ctx->polykey[1];
414 i = mbytes / VMAC_NHBYTES;
415 remaining = mbytes % VMAC_NHBYTES;
417 if (ctx->first_block_processed) {
418 ch = ctx->polytmp[0];
419 cl = ctx->polytmp[1];
421 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, ch, cl);
423 ADD128(ch, cl, pkh, pkl);
424 mptr += (VMAC_NHBYTES/sizeof(u64));
426 } else if (remaining) {
427 nh_16(mptr, kptr, 2*((remaining+15)/16), ch, cl);
429 ADD128(ch, cl, pkh, pkl);
430 mptr += (VMAC_NHBYTES/sizeof(u64));
432 } else {/* Empty String */
438 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
440 poly_step(ch, cl, pkh, pkl, rh, rl);
441 mptr += (VMAC_NHBYTES/sizeof(u64));
444 nh_16(mptr, kptr, 2*((remaining+15)/16), rh, rl);
446 poly_step(ch, cl, pkh, pkl, rh, rl);
452 return l3hash(ch, cl, ctx->l3key[0], ctx->l3key[1], remaining);
455 static u64 vmac(unsigned char m[], unsigned int mbytes,
456 unsigned char n[16], u64 *tagl,
457 struct vmac_ctx_t *ctx)
463 in_n = ctx->__vmac_ctx.cached_nonce;
464 out_p = ctx->__vmac_ctx.cached_aes;
467 if ((*(u64 *)(n+8) != in_n[1]) || (*(u64 *)(n) != in_n[0])) {
468 in_n[0] = *(u64 *)(n);
469 in_n[1] = *(u64 *)(n+8);
470 ((unsigned char *)in_n)[15] &= 0xFE;
471 crypto_cipher_encrypt_one(ctx->child,
472 (unsigned char *)out_p, (unsigned char *)in_n);
474 ((unsigned char *)in_n)[15] |= (unsigned char)(1-i);
476 p = be64_to_cpup(out_p + i);
477 h = vhash(m, mbytes, (u64 *)0, &ctx->__vmac_ctx);
478 return le64_to_cpu(p + h);
481 static int vmac_set_key(unsigned char user_key[], struct vmac_ctx_t *ctx)
483 u64 in[2] = {0}, out[2];
487 err = crypto_cipher_setkey(ctx->child, user_key, VMAC_KEY_LEN);
492 ((unsigned char *)in)[0] = 0x80;
493 for (i = 0; i < sizeof(ctx->__vmac_ctx.nhkey)/8; i += 2) {
494 crypto_cipher_encrypt_one(ctx->child,
495 (unsigned char *)out, (unsigned char *)in);
496 ctx->__vmac_ctx.nhkey[i] = be64_to_cpup(out);
497 ctx->__vmac_ctx.nhkey[i+1] = be64_to_cpup(out+1);
498 ((unsigned char *)in)[15] += 1;
502 ((unsigned char *)in)[0] = 0xC0;
504 for (i = 0; i < sizeof(ctx->__vmac_ctx.polykey)/8; i += 2) {
505 crypto_cipher_encrypt_one(ctx->child,
506 (unsigned char *)out, (unsigned char *)in);
507 ctx->__vmac_ctx.polytmp[i] =
508 ctx->__vmac_ctx.polykey[i] =
509 be64_to_cpup(out) & mpoly;
510 ctx->__vmac_ctx.polytmp[i+1] =
511 ctx->__vmac_ctx.polykey[i+1] =
512 be64_to_cpup(out+1) & mpoly;
513 ((unsigned char *)in)[15] += 1;
517 ((unsigned char *)in)[0] = 0xE0;
519 for (i = 0; i < sizeof(ctx->__vmac_ctx.l3key)/8; i += 2) {
521 crypto_cipher_encrypt_one(ctx->child,
522 (unsigned char *)out, (unsigned char *)in);
523 ctx->__vmac_ctx.l3key[i] = be64_to_cpup(out);
524 ctx->__vmac_ctx.l3key[i+1] = be64_to_cpup(out+1);
525 ((unsigned char *)in)[15] += 1;
526 } while (ctx->__vmac_ctx.l3key[i] >= p64
527 || ctx->__vmac_ctx.l3key[i+1] >= p64);
530 /* Invalidate nonce/aes cache and reset other elements */
531 ctx->__vmac_ctx.cached_nonce[0] = (u64)-1; /* Ensure illegal nonce */
532 ctx->__vmac_ctx.cached_nonce[1] = (u64)0; /* Ensure illegal nonce */
533 ctx->__vmac_ctx.first_block_processed = 0;
538 static int vmac_setkey(struct crypto_shash *parent,
539 const u8 *key, unsigned int keylen)
541 struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
543 if (keylen != VMAC_KEY_LEN) {
544 crypto_shash_set_flags(parent, CRYPTO_TFM_RES_BAD_KEY_LEN);
548 return vmac_set_key((u8 *)key, ctx);
551 static int vmac_init(struct shash_desc *pdesc)
556 static int vmac_update(struct shash_desc *pdesc, const u8 *p,
559 struct crypto_shash *parent = pdesc->tfm;
560 struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
562 vhash_update(p, len, &ctx->__vmac_ctx);
567 static int vmac_final(struct shash_desc *pdesc, u8 *out)
569 struct crypto_shash *parent = pdesc->tfm;
570 struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
574 mac = vmac(NULL, 0, nonce, NULL, ctx);
575 memcpy(out, &mac, sizeof(vmac_t));
576 memset(&mac, 0, sizeof(vmac_t));
577 memset(&ctx->__vmac_ctx, 0, sizeof(struct vmac_ctx));
581 static int vmac_init_tfm(struct crypto_tfm *tfm)
583 struct crypto_cipher *cipher;
584 struct crypto_instance *inst = (void *)tfm->__crt_alg;
585 struct crypto_spawn *spawn = crypto_instance_ctx(inst);
586 struct vmac_ctx_t *ctx = crypto_tfm_ctx(tfm);
588 cipher = crypto_spawn_cipher(spawn);
590 return PTR_ERR(cipher);
596 static void vmac_exit_tfm(struct crypto_tfm *tfm)
598 struct vmac_ctx_t *ctx = crypto_tfm_ctx(tfm);
599 crypto_free_cipher(ctx->child);
602 static int vmac_create(struct crypto_template *tmpl, struct rtattr **tb)
604 struct shash_instance *inst;
605 struct crypto_alg *alg;
608 err = crypto_check_attr_type(tb, CRYPTO_ALG_TYPE_SHASH);
612 alg = crypto_get_attr_alg(tb, CRYPTO_ALG_TYPE_CIPHER,
613 CRYPTO_ALG_TYPE_MASK);
617 inst = shash_alloc_instance("vmac", alg);
622 err = crypto_init_spawn(shash_instance_ctx(inst), alg,
623 shash_crypto_instance(inst),
624 CRYPTO_ALG_TYPE_MASK);
628 inst->alg.base.cra_priority = alg->cra_priority;
629 inst->alg.base.cra_blocksize = alg->cra_blocksize;
630 inst->alg.base.cra_alignmask = alg->cra_alignmask;
632 inst->alg.digestsize = sizeof(vmac_t);
633 inst->alg.base.cra_ctxsize = sizeof(struct vmac_ctx_t);
634 inst->alg.base.cra_init = vmac_init_tfm;
635 inst->alg.base.cra_exit = vmac_exit_tfm;
637 inst->alg.init = vmac_init;
638 inst->alg.update = vmac_update;
639 inst->alg.final = vmac_final;
640 inst->alg.setkey = vmac_setkey;
642 err = shash_register_instance(tmpl, inst);
645 shash_free_instance(shash_crypto_instance(inst));
653 static struct crypto_template vmac_tmpl = {
655 .create = vmac_create,
656 .free = shash_free_instance,
657 .module = THIS_MODULE,
660 static int __init vmac_module_init(void)
662 return crypto_register_template(&vmac_tmpl);
665 static void __exit vmac_module_exit(void)
667 crypto_unregister_template(&vmac_tmpl);
670 module_init(vmac_module_init);
671 module_exit(vmac_module_exit);
673 MODULE_LICENSE("GPL");
674 MODULE_DESCRIPTION("VMAC hash algorithm");