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/module.h>
31 #include <linux/scatterlist.h>
32 #include <asm/byteorder.h>
33 #include <crypto/scatterwalk.h>
34 #include <crypto/vmac.h>
35 #include <crypto/internal/hash.h>
40 #define UINT64_C(x) x##ULL
41 static const u64 p64 = UINT64_C(0xfffffffffffffeff); /* 2^64 - 257 prime */
42 static const u64 m62 = UINT64_C(0x3fffffffffffffff); /* 62-bit mask */
43 static const u64 m63 = UINT64_C(0x7fffffffffffffff); /* 63-bit mask */
44 static const u64 m64 = UINT64_C(0xffffffffffffffff); /* 64-bit mask */
45 static const u64 mpoly = UINT64_C(0x1fffffff1fffffff); /* Poly key mask */
47 #define pe64_to_cpup le64_to_cpup /* Prefer little endian */
49 #ifdef __LITTLE_ENDIAN
58 * The following routines are used in this implementation. They are
59 * written via macros to simulate zero-overhead call-by-reference.
61 * MUL64: 64x64->128-bit multiplication
62 * PMUL64: assumes top bits cleared on inputs
63 * ADD128: 128x128->128-bit addition
66 #define ADD128(rh, rl, ih, il) \
75 #define MUL32(i1, i2) ((u64)(u32)(i1)*(u32)(i2))
77 #define PMUL64(rh, rl, i1, i2) /* Assumes m doesn't overflow */ \
79 u64 _i1 = (i1), _i2 = (i2); \
80 u64 m = MUL32(_i1, _i2>>32) + MUL32(_i1>>32, _i2); \
81 rh = MUL32(_i1>>32, _i2>>32); \
82 rl = MUL32(_i1, _i2); \
83 ADD128(rh, rl, (m >> 32), (m << 32)); \
86 #define MUL64(rh, rl, i1, i2) \
88 u64 _i1 = (i1), _i2 = (i2); \
89 u64 m1 = MUL32(_i1, _i2>>32); \
90 u64 m2 = MUL32(_i1>>32, _i2); \
91 rh = MUL32(_i1>>32, _i2>>32); \
92 rl = MUL32(_i1, _i2); \
93 ADD128(rh, rl, (m1 >> 32), (m1 << 32)); \
94 ADD128(rh, rl, (m2 >> 32), (m2 << 32)); \
98 * For highest performance the L1 NH and L2 polynomial hashes should be
99 * carefully implemented to take advantage of one's target architecture.
100 * Here these two hash functions are defined multiple time; once for
101 * 64-bit architectures, once for 32-bit SSE2 architectures, and once
102 * for the rest (32-bit) architectures.
103 * For each, nh_16 *must* be defined (works on multiples of 16 bytes).
104 * Optionally, nh_vmac_nhbytes can be defined (for multiples of
105 * VMAC_NHBYTES), and nh_16_2 and nh_vmac_nhbytes_2 (versions that do two
106 * NH computations at once).
111 #define nh_16(mp, kp, nw, rh, rl) \
115 for (i = 0; i < nw; i += 2) { \
116 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
117 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
118 ADD128(rh, rl, th, tl); \
122 #define nh_16_2(mp, kp, nw, rh, rl, rh1, rl1) \
125 rh1 = rl1 = rh = rl = 0; \
126 for (i = 0; i < nw; i += 2) { \
127 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
128 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
129 ADD128(rh, rl, th, tl); \
130 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
131 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
132 ADD128(rh1, rl1, th, tl); \
136 #if (VMAC_NHBYTES >= 64) /* These versions do 64-bytes of message at a time */
137 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
141 for (i = 0; i < nw; i += 8) { \
142 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
143 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
144 ADD128(rh, rl, th, tl); \
145 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
146 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
147 ADD128(rh, rl, th, tl); \
148 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
149 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
150 ADD128(rh, rl, th, tl); \
151 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
152 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
153 ADD128(rh, rl, th, tl); \
157 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh1, rl1) \
160 rh1 = rl1 = rh = rl = 0; \
161 for (i = 0; i < nw; i += 8) { \
162 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i], \
163 pe64_to_cpup((mp)+i+1)+(kp)[i+1]); \
164 ADD128(rh, rl, th, tl); \
165 MUL64(th, tl, pe64_to_cpup((mp)+i)+(kp)[i+2], \
166 pe64_to_cpup((mp)+i+1)+(kp)[i+3]); \
167 ADD128(rh1, rl1, th, tl); \
168 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+2], \
169 pe64_to_cpup((mp)+i+3)+(kp)[i+3]); \
170 ADD128(rh, rl, th, tl); \
171 MUL64(th, tl, pe64_to_cpup((mp)+i+2)+(kp)[i+4], \
172 pe64_to_cpup((mp)+i+3)+(kp)[i+5]); \
173 ADD128(rh1, rl1, th, tl); \
174 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+4], \
175 pe64_to_cpup((mp)+i+5)+(kp)[i+5]); \
176 ADD128(rh, rl, th, tl); \
177 MUL64(th, tl, pe64_to_cpup((mp)+i+4)+(kp)[i+6], \
178 pe64_to_cpup((mp)+i+5)+(kp)[i+7]); \
179 ADD128(rh1, rl1, th, tl); \
180 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+6], \
181 pe64_to_cpup((mp)+i+7)+(kp)[i+7]); \
182 ADD128(rh, rl, th, tl); \
183 MUL64(th, tl, pe64_to_cpup((mp)+i+6)+(kp)[i+8], \
184 pe64_to_cpup((mp)+i+7)+(kp)[i+9]); \
185 ADD128(rh1, rl1, th, tl); \
190 #define poly_step(ah, al, kh, kl, mh, ml) \
192 u64 t1h, t1l, t2h, t2l, t3h, t3l, z = 0; \
193 /* compute ab*cd, put bd into result registers */ \
194 PMUL64(t3h, t3l, al, kh); \
195 PMUL64(t2h, t2l, ah, kl); \
196 PMUL64(t1h, t1l, ah, 2*kh); \
197 PMUL64(ah, al, al, kl); \
198 /* add 2 * ac to result */ \
199 ADD128(ah, al, t1h, t1l); \
200 /* add together ad + bc */ \
201 ADD128(t2h, t2l, t3h, t3l); \
202 /* now (ah,al), (t2l,2*t2h) need summing */ \
203 /* first add the high registers, carrying into t2h */ \
204 ADD128(t2h, ah, z, t2l); \
205 /* double t2h and add top bit of ah */ \
206 t2h = 2 * t2h + (ah >> 63); \
208 /* now add the low registers */ \
209 ADD128(ah, al, mh, ml); \
210 ADD128(ah, al, z, t2h); \
213 #else /* ! CONFIG_64BIT */
216 #define nh_16(mp, kp, nw, rh, rl) \
218 u64 t1, t2, m1, m2, t; \
221 for (i = 0; i < nw; i += 2) { \
222 t1 = pe64_to_cpup(mp+i) + kp[i]; \
223 t2 = pe64_to_cpup(mp+i+1) + kp[i+1]; \
224 m2 = MUL32(t1 >> 32, t2); \
225 m1 = MUL32(t1, t2 >> 32); \
226 ADD128(rh, rl, MUL32(t1 >> 32, t2 >> 32), \
228 rh += (u64)(u32)(m1 >> 32) \
230 t += (u64)(u32)m1 + (u32)m2; \
232 ADD128(rh, rl, (t >> 32), (t << 32)); \
236 static void poly_step_func(u64 *ahi, u64 *alo,
237 const u64 *kh, const u64 *kl,
238 const u64 *mh, const u64 *ml)
240 #define a0 (*(((u32 *)alo)+INDEX_LOW))
241 #define a1 (*(((u32 *)alo)+INDEX_HIGH))
242 #define a2 (*(((u32 *)ahi)+INDEX_LOW))
243 #define a3 (*(((u32 *)ahi)+INDEX_HIGH))
244 #define k0 (*(((u32 *)kl)+INDEX_LOW))
245 #define k1 (*(((u32 *)kl)+INDEX_HIGH))
246 #define k2 (*(((u32 *)kh)+INDEX_LOW))
247 #define k3 (*(((u32 *)kh)+INDEX_HIGH))
264 t |= ((u64)((u32)p & 0x7fffffff)) << 32;
266 p += (u64)(((u32 *)ml)[INDEX_LOW]);
275 p += (u64)(((u32 *)ml)[INDEX_HIGH]);
282 *(u64 *)(alo) = (p << 32) | t2;
284 *(u64 *)(ahi) = p + t;
296 #define poly_step(ah, al, kh, kl, mh, ml) \
297 poly_step_func(&(ah), &(al), &(kh), &(kl), &(mh), &(ml))
299 #endif /* end of specialized NH and poly definitions */
301 /* At least nh_16 is defined. Defined others as needed here */
303 #define nh_16_2(mp, kp, nw, rh, rl, rh2, rl2) \
305 nh_16(mp, kp, nw, rh, rl); \
306 nh_16(mp, ((kp)+2), nw, rh2, rl2); \
309 #ifndef nh_vmac_nhbytes
310 #define nh_vmac_nhbytes(mp, kp, nw, rh, rl) \
311 nh_16(mp, kp, nw, rh, rl)
313 #ifndef nh_vmac_nhbytes_2
314 #define nh_vmac_nhbytes_2(mp, kp, nw, rh, rl, rh2, rl2) \
316 nh_vmac_nhbytes(mp, kp, nw, rh, rl); \
317 nh_vmac_nhbytes(mp, ((kp)+2), nw, rh2, rl2); \
321 static void vhash_abort(struct vmac_ctx *ctx)
323 ctx->polytmp[0] = ctx->polykey[0] ;
324 ctx->polytmp[1] = ctx->polykey[1] ;
325 ctx->first_block_processed = 0;
328 static u64 l3hash(u64 p1, u64 p2, u64 k1, u64 k2, u64 len)
330 u64 rh, rl, t, z = 0;
332 /* fully reduce (p1,p2)+(len,0) mod p127 */
335 ADD128(p1, p2, len, t);
336 /* At this point, (p1,p2) is at most 2^127+(len<<64) */
337 t = (p1 > m63) + ((p1 == m63) && (p2 == m64));
338 ADD128(p1, p2, z, t);
341 /* compute (p1,p2)/(2^64-2^32) and (p1,p2)%(2^64-2^32) */
344 t += (u32)t > 0xfffffffeu;
348 /* compute (p1+k1)%p64 and (p2+k2)%p64 */
350 p1 += (0 - (p1 < k1)) & 257;
352 p2 += (0 - (p2 < k2)) & 257;
354 /* compute (p1+k1)*(p2+k2)%p64 */
355 MUL64(rh, rl, p1, p2);
357 ADD128(t, rl, z, rh);
359 ADD128(t, rl, z, rh);
362 rl += (0 - (rl < t)) & 257;
363 rl += (0 - (rl > p64-1)) & 257;
367 static void vhash_update(const unsigned char *m,
368 unsigned int mbytes, /* Pos multiple of VMAC_NHBYTES */
369 struct vmac_ctx *ctx)
372 const u64 *kptr = (u64 *)ctx->nhkey;
375 u64 pkh = ctx->polykey[0];
376 u64 pkl = ctx->polykey[1];
379 i = mbytes / VMAC_NHBYTES; /* Must be non-zero */
381 ch = ctx->polytmp[0];
382 cl = ctx->polytmp[1];
384 if (!ctx->first_block_processed) {
385 ctx->first_block_processed = 1;
386 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
388 ADD128(ch, cl, rh, rl);
389 mptr += (VMAC_NHBYTES/sizeof(u64));
394 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
396 poly_step(ch, cl, pkh, pkl, rh, rl);
397 mptr += (VMAC_NHBYTES/sizeof(u64));
400 ctx->polytmp[0] = ch;
401 ctx->polytmp[1] = cl;
404 static u64 vhash(unsigned char m[], unsigned int mbytes,
405 u64 *tagl, struct vmac_ctx *ctx)
408 const u64 *kptr = (u64 *)ctx->nhkey;
411 u64 pkh = ctx->polykey[0];
412 u64 pkl = ctx->polykey[1];
415 i = mbytes / VMAC_NHBYTES;
416 remaining = mbytes % VMAC_NHBYTES;
418 if (ctx->first_block_processed) {
419 ch = ctx->polytmp[0];
420 cl = ctx->polytmp[1];
422 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, ch, cl);
424 ADD128(ch, cl, pkh, pkl);
425 mptr += (VMAC_NHBYTES/sizeof(u64));
427 } else if (remaining) {
428 nh_16(mptr, kptr, 2*((remaining+15)/16), ch, cl);
430 ADD128(ch, cl, pkh, pkl);
431 mptr += (VMAC_NHBYTES/sizeof(u64));
433 } else {/* Empty String */
439 nh_vmac_nhbytes(mptr, kptr, VMAC_NHBYTES/8, rh, rl);
441 poly_step(ch, cl, pkh, pkl, rh, rl);
442 mptr += (VMAC_NHBYTES/sizeof(u64));
445 nh_16(mptr, kptr, 2*((remaining+15)/16), rh, rl);
447 poly_step(ch, cl, pkh, pkl, rh, rl);
453 return l3hash(ch, cl, ctx->l3key[0], ctx->l3key[1], remaining);
456 static u64 vmac(unsigned char m[], unsigned int mbytes,
457 unsigned char n[16], u64 *tagl,
458 struct vmac_ctx_t *ctx)
464 in_n = ctx->__vmac_ctx.cached_nonce;
465 out_p = ctx->__vmac_ctx.cached_aes;
468 if ((*(u64 *)(n+8) != in_n[1]) || (*(u64 *)(n) != in_n[0])) {
469 in_n[0] = *(u64 *)(n);
470 in_n[1] = *(u64 *)(n+8);
471 ((unsigned char *)in_n)[15] &= 0xFE;
472 crypto_cipher_encrypt_one(ctx->child,
473 (unsigned char *)out_p, (unsigned char *)in_n);
475 ((unsigned char *)in_n)[15] |= (unsigned char)(1-i);
477 p = be64_to_cpup(out_p + i);
478 h = vhash(m, mbytes, (u64 *)0, &ctx->__vmac_ctx);
479 return le64_to_cpu(p + h);
482 static int vmac_set_key(unsigned char user_key[], struct vmac_ctx_t *ctx)
484 u64 in[2] = {0}, out[2];
488 err = crypto_cipher_setkey(ctx->child, user_key, VMAC_KEY_LEN);
493 ((unsigned char *)in)[0] = 0x80;
494 for (i = 0; i < sizeof(ctx->__vmac_ctx.nhkey)/8; i += 2) {
495 crypto_cipher_encrypt_one(ctx->child,
496 (unsigned char *)out, (unsigned char *)in);
497 ctx->__vmac_ctx.nhkey[i] = be64_to_cpup(out);
498 ctx->__vmac_ctx.nhkey[i+1] = be64_to_cpup(out+1);
499 ((unsigned char *)in)[15] += 1;
503 ((unsigned char *)in)[0] = 0xC0;
505 for (i = 0; i < sizeof(ctx->__vmac_ctx.polykey)/8; i += 2) {
506 crypto_cipher_encrypt_one(ctx->child,
507 (unsigned char *)out, (unsigned char *)in);
508 ctx->__vmac_ctx.polytmp[i] =
509 ctx->__vmac_ctx.polykey[i] =
510 be64_to_cpup(out) & mpoly;
511 ctx->__vmac_ctx.polytmp[i+1] =
512 ctx->__vmac_ctx.polykey[i+1] =
513 be64_to_cpup(out+1) & mpoly;
514 ((unsigned char *)in)[15] += 1;
518 ((unsigned char *)in)[0] = 0xE0;
520 for (i = 0; i < sizeof(ctx->__vmac_ctx.l3key)/8; i += 2) {
522 crypto_cipher_encrypt_one(ctx->child,
523 (unsigned char *)out, (unsigned char *)in);
524 ctx->__vmac_ctx.l3key[i] = be64_to_cpup(out);
525 ctx->__vmac_ctx.l3key[i+1] = be64_to_cpup(out+1);
526 ((unsigned char *)in)[15] += 1;
527 } while (ctx->__vmac_ctx.l3key[i] >= p64
528 || ctx->__vmac_ctx.l3key[i+1] >= p64);
531 /* Invalidate nonce/aes cache and reset other elements */
532 ctx->__vmac_ctx.cached_nonce[0] = (u64)-1; /* Ensure illegal nonce */
533 ctx->__vmac_ctx.cached_nonce[1] = (u64)0; /* Ensure illegal nonce */
534 ctx->__vmac_ctx.first_block_processed = 0;
539 static int vmac_setkey(struct crypto_shash *parent,
540 const u8 *key, unsigned int keylen)
542 struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
544 if (keylen != VMAC_KEY_LEN) {
545 crypto_shash_set_flags(parent, CRYPTO_TFM_RES_BAD_KEY_LEN);
549 return vmac_set_key((u8 *)key, ctx);
552 static int vmac_init(struct shash_desc *pdesc)
557 static int vmac_update(struct shash_desc *pdesc, const u8 *p,
560 struct crypto_shash *parent = pdesc->tfm;
561 struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
563 vhash_update(p, len, &ctx->__vmac_ctx);
568 static int vmac_final(struct shash_desc *pdesc, u8 *out)
570 struct crypto_shash *parent = pdesc->tfm;
571 struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
575 mac = vmac(NULL, 0, nonce, NULL, ctx);
576 memcpy(out, &mac, sizeof(vmac_t));
577 memset(&mac, 0, sizeof(vmac_t));
578 memset(&ctx->__vmac_ctx, 0, sizeof(struct vmac_ctx));
582 static int vmac_init_tfm(struct crypto_tfm *tfm)
584 struct crypto_cipher *cipher;
585 struct crypto_instance *inst = (void *)tfm->__crt_alg;
586 struct crypto_spawn *spawn = crypto_instance_ctx(inst);
587 struct vmac_ctx_t *ctx = crypto_tfm_ctx(tfm);
589 cipher = crypto_spawn_cipher(spawn);
591 return PTR_ERR(cipher);
597 static void vmac_exit_tfm(struct crypto_tfm *tfm)
599 struct vmac_ctx_t *ctx = crypto_tfm_ctx(tfm);
600 crypto_free_cipher(ctx->child);
603 static int vmac_create(struct crypto_template *tmpl, struct rtattr **tb)
605 struct shash_instance *inst;
606 struct crypto_alg *alg;
609 err = crypto_check_attr_type(tb, CRYPTO_ALG_TYPE_SHASH);
613 alg = crypto_get_attr_alg(tb, CRYPTO_ALG_TYPE_CIPHER,
614 CRYPTO_ALG_TYPE_MASK);
618 inst = shash_alloc_instance("vmac", alg);
623 err = crypto_init_spawn(shash_instance_ctx(inst), alg,
624 shash_crypto_instance(inst),
625 CRYPTO_ALG_TYPE_MASK);
629 inst->alg.base.cra_priority = alg->cra_priority;
630 inst->alg.base.cra_blocksize = alg->cra_blocksize;
631 inst->alg.base.cra_alignmask = alg->cra_alignmask;
633 inst->alg.digestsize = sizeof(vmac_t);
634 inst->alg.base.cra_ctxsize = sizeof(struct vmac_ctx_t);
635 inst->alg.base.cra_init = vmac_init_tfm;
636 inst->alg.base.cra_exit = vmac_exit_tfm;
638 inst->alg.init = vmac_init;
639 inst->alg.update = vmac_update;
640 inst->alg.final = vmac_final;
641 inst->alg.setkey = vmac_setkey;
643 err = shash_register_instance(tmpl, inst);
646 shash_free_instance(shash_crypto_instance(inst));
654 static struct crypto_template vmac_tmpl = {
656 .create = vmac_create,
657 .free = shash_free_instance,
658 .module = THIS_MODULE,
661 static int __init vmac_module_init(void)
663 return crypto_register_template(&vmac_tmpl);
666 static void __exit vmac_module_exit(void)
668 crypto_unregister_template(&vmac_tmpl);
671 module_init(vmac_module_init);
672 module_exit(vmac_module_exit);
674 MODULE_LICENSE("GPL");
675 MODULE_DESCRIPTION("VMAC hash algorithm");