1 /* ====================================================================
2 * Copyright (c) 2011-2013 The OpenSSL Project. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in
13 * the documentation and/or other materials provided with the
16 * 3. All advertising materials mentioning features or use of this
17 * software must display the following acknowledgment:
18 * "This product includes software developed by the OpenSSL Project
19 * for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
21 * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
22 * endorse or promote products derived from this software without
23 * prior written permission. For written permission, please contact
24 * licensing@OpenSSL.org.
26 * 5. Products derived from this software may not be called "OpenSSL"
27 * nor may "OpenSSL" appear in their names without prior written
28 * permission of the OpenSSL Project.
30 * 6. Redistributions of any form whatsoever must retain the following
32 * "This product includes software developed by the OpenSSL Project
33 * for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
35 * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
36 * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
38 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR
39 * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
40 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
41 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
42 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
43 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
44 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
45 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
46 * OF THE POSSIBILITY OF SUCH DAMAGE.
47 * ====================================================================
50 #include <openssl/opensslconf.h>
55 #if !defined(OPENSSL_NO_AES) && !defined(OPENSSL_NO_SHA256)
57 # include <openssl/evp.h>
58 # include <openssl/objects.h>
59 # include <openssl/aes.h>
60 # include <openssl/sha.h>
61 # include <openssl/rand.h>
62 # include "modes_lcl.h"
63 # include "constant_time_locl.h"
65 # ifndef EVP_CIPH_FLAG_AEAD_CIPHER
66 # define EVP_CIPH_FLAG_AEAD_CIPHER 0x200000
67 # define EVP_CTRL_AEAD_TLS1_AAD 0x16
68 # define EVP_CTRL_AEAD_SET_MAC_KEY 0x17
71 # if !defined(EVP_CIPH_FLAG_DEFAULT_ASN1)
72 # define EVP_CIPH_FLAG_DEFAULT_ASN1 0
75 # if !defined(EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK)
76 # define EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK 0
79 # define TLS1_1_VERSION 0x0302
83 SHA256_CTX head, tail, md;
84 size_t payload_length; /* AAD length in decrypt case */
87 unsigned char tls_aad[16]; /* 13 used */
89 } EVP_AES_HMAC_SHA256;
91 # define NO_PAYLOAD_LENGTH ((size_t)-1)
93 # if defined(AES_ASM) && ( \
94 defined(__x86_64) || defined(__x86_64__) || \
95 defined(_M_AMD64) || defined(_M_X64) || \
98 extern unsigned int OPENSSL_ia32cap_P[];
99 # define AESNI_CAPABLE (1<<(57-32))
101 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
103 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
106 void aesni_cbc_encrypt(const unsigned char *in,
109 const AES_KEY *key, unsigned char *ivec, int enc);
111 int aesni_cbc_sha256_enc(const void *inp, void *out, size_t blocks,
112 const AES_KEY *key, unsigned char iv[16],
113 SHA256_CTX *ctx, const void *in0);
115 # define data(ctx) ((EVP_AES_HMAC_SHA256 *)(ctx)->cipher_data)
117 static int aesni_cbc_hmac_sha256_init_key(EVP_CIPHER_CTX *ctx,
118 const unsigned char *inkey,
119 const unsigned char *iv, int enc)
121 EVP_AES_HMAC_SHA256 *key = data(ctx);
125 memset(&key->ks, 0, sizeof(key->ks.rd_key)),
126 ret = aesni_set_encrypt_key(inkey, ctx->key_len * 8, &key->ks);
128 ret = aesni_set_decrypt_key(inkey, ctx->key_len * 8, &key->ks);
130 SHA256_Init(&key->head); /* handy when benchmarking */
131 key->tail = key->head;
134 key->payload_length = NO_PAYLOAD_LENGTH;
136 return ret < 0 ? 0 : 1;
139 # define STITCHED_CALL
141 # if !defined(STITCHED_CALL)
145 void sha256_block_data_order(void *c, const void *p, size_t len);
147 static void sha256_update(SHA256_CTX *c, const void *data, size_t len)
149 const unsigned char *ptr = data;
152 if ((res = c->num)) {
153 res = SHA256_CBLOCK - res;
156 SHA256_Update(c, ptr, res);
161 res = len % SHA256_CBLOCK;
165 sha256_block_data_order(c, ptr, len / SHA256_CBLOCK);
170 if (c->Nl < (unsigned int)len)
175 SHA256_Update(c, ptr, res);
178 # ifdef SHA256_Update
179 # undef SHA256_Update
181 # define SHA256_Update sha256_update
183 # if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
186 unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8];
189 const unsigned char *ptr;
193 void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int);
196 const unsigned char *inp;
202 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
204 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
206 const unsigned char *inp,
207 size_t inp_len, int n4x)
208 { /* n4x is 1 or 2 */
209 HASH_DESC hash_d[8], edges[8];
211 unsigned char storage[sizeof(SHA256_MB_CTX) + 32];
218 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
226 /* ask for IVs in bulk */
227 if (RAND_bytes((IVs = blocks[0].c), 16 * x4) <= 0)
231 ctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32));
233 frag = (unsigned int)inp_len >> (1 + n4x);
234 last = (unsigned int)inp_len + frag - (frag << (1 + n4x));
235 if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) {
240 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
242 /* populate descriptors with pointers and IVs */
245 /* 5+16 is place for header and explicit IV */
246 ciph_d[0].out = out + 5 + 16;
247 memcpy(ciph_d[0].out - 16, IVs, 16);
248 memcpy(ciph_d[0].iv, IVs, 16);
251 for (i = 1; i < x4; i++) {
252 ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag;
253 ciph_d[i].out = ciph_d[i - 1].out + packlen;
254 memcpy(ciph_d[i].out - 16, IVs, 16);
255 memcpy(ciph_d[i].iv, IVs, 16);
260 memcpy(blocks[0].c, key->md.data, 8);
261 seqnum = BSWAP8(blocks[0].q[0]);
263 for (i = 0; i < x4; i++) {
264 unsigned int len = (i == (x4 - 1) ? last : frag);
265 # if !defined(BSWAP8)
266 unsigned int carry, j;
269 ctx->A[i] = key->md.h[0];
270 ctx->B[i] = key->md.h[1];
271 ctx->C[i] = key->md.h[2];
272 ctx->D[i] = key->md.h[3];
273 ctx->E[i] = key->md.h[4];
274 ctx->F[i] = key->md.h[5];
275 ctx->G[i] = key->md.h[6];
276 ctx->H[i] = key->md.h[7];
280 blocks[i].q[0] = BSWAP8(seqnum + i);
282 for (carry = i, j = 8; j--;) {
283 blocks[i].c[j] = ((u8 *)key->md.data)[j] + carry;
284 carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1);
287 blocks[i].c[8] = ((u8 *)key->md.data)[8];
288 blocks[i].c[9] = ((u8 *)key->md.data)[9];
289 blocks[i].c[10] = ((u8 *)key->md.data)[10];
291 blocks[i].c[11] = (u8)(len >> 8);
292 blocks[i].c[12] = (u8)(len);
294 memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13);
295 hash_d[i].ptr += 64 - 13;
296 hash_d[i].blocks = (len - (64 - 13)) / 64;
298 edges[i].ptr = blocks[i].c;
302 /* hash 13-byte headers and first 64-13 bytes of inputs */
303 sha256_multi_block(ctx, edges, n4x);
304 /* hash bulk inputs */
305 # define MAXCHUNKSIZE 2048
307 # error "MAXCHUNKSIZE is not divisible by 64"
310 * goal is to minimize pressure on L1 cache by moving in shorter steps,
311 * so that hashed data is still in the cache by the time we encrypt it
313 minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64;
314 if (minblocks > MAXCHUNKSIZE / 64) {
315 for (i = 0; i < x4; i++) {
316 edges[i].ptr = hash_d[i].ptr;
317 edges[i].blocks = MAXCHUNKSIZE / 64;
318 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
321 sha256_multi_block(ctx, edges, n4x);
322 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
324 for (i = 0; i < x4; i++) {
325 edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE;
326 hash_d[i].blocks -= MAXCHUNKSIZE / 64;
327 edges[i].blocks = MAXCHUNKSIZE / 64;
328 ciph_d[i].inp += MAXCHUNKSIZE;
329 ciph_d[i].out += MAXCHUNKSIZE;
330 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
331 memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16);
333 processed += MAXCHUNKSIZE;
334 minblocks -= MAXCHUNKSIZE / 64;
335 } while (minblocks > MAXCHUNKSIZE / 64);
339 sha256_multi_block(ctx, hash_d, n4x);
341 memset(blocks, 0, sizeof(blocks));
342 for (i = 0; i < x4; i++) {
343 unsigned int len = (i == (x4 - 1) ? last : frag),
344 off = hash_d[i].blocks * 64;
345 const unsigned char *ptr = hash_d[i].ptr + off;
347 off = (len - processed) - (64 - 13) - off; /* remainder actually */
348 memcpy(blocks[i].c, ptr, off);
349 blocks[i].c[off] = 0x80;
350 len += 64 + 13; /* 64 is HMAC header */
351 len *= 8; /* convert to bits */
352 if (off < (64 - 8)) {
354 blocks[i].d[15] = BSWAP4(len);
356 PUTU32(blocks[i].c + 60, len);
361 blocks[i].d[31] = BSWAP4(len);
363 PUTU32(blocks[i].c + 124, len);
367 edges[i].ptr = blocks[i].c;
370 /* hash input tails and finalize */
371 sha256_multi_block(ctx, edges, n4x);
373 memset(blocks, 0, sizeof(blocks));
374 for (i = 0; i < x4; i++) {
376 blocks[i].d[0] = BSWAP4(ctx->A[i]);
377 ctx->A[i] = key->tail.h[0];
378 blocks[i].d[1] = BSWAP4(ctx->B[i]);
379 ctx->B[i] = key->tail.h[1];
380 blocks[i].d[2] = BSWAP4(ctx->C[i]);
381 ctx->C[i] = key->tail.h[2];
382 blocks[i].d[3] = BSWAP4(ctx->D[i]);
383 ctx->D[i] = key->tail.h[3];
384 blocks[i].d[4] = BSWAP4(ctx->E[i]);
385 ctx->E[i] = key->tail.h[4];
386 blocks[i].d[5] = BSWAP4(ctx->F[i]);
387 ctx->F[i] = key->tail.h[5];
388 blocks[i].d[6] = BSWAP4(ctx->G[i]);
389 ctx->G[i] = key->tail.h[6];
390 blocks[i].d[7] = BSWAP4(ctx->H[i]);
391 ctx->H[i] = key->tail.h[7];
392 blocks[i].c[32] = 0x80;
393 blocks[i].d[15] = BSWAP4((64 + 32) * 8);
395 PUTU32(blocks[i].c + 0, ctx->A[i]);
396 ctx->A[i] = key->tail.h[0];
397 PUTU32(blocks[i].c + 4, ctx->B[i]);
398 ctx->B[i] = key->tail.h[1];
399 PUTU32(blocks[i].c + 8, ctx->C[i]);
400 ctx->C[i] = key->tail.h[2];
401 PUTU32(blocks[i].c + 12, ctx->D[i]);
402 ctx->D[i] = key->tail.h[3];
403 PUTU32(blocks[i].c + 16, ctx->E[i]);
404 ctx->E[i] = key->tail.h[4];
405 PUTU32(blocks[i].c + 20, ctx->F[i]);
406 ctx->F[i] = key->tail.h[5];
407 PUTU32(blocks[i].c + 24, ctx->G[i]);
408 ctx->G[i] = key->tail.h[6];
409 PUTU32(blocks[i].c + 28, ctx->H[i]);
410 ctx->H[i] = key->tail.h[7];
411 blocks[i].c[32] = 0x80;
412 PUTU32(blocks[i].c + 60, (64 + 32) * 8);
414 edges[i].ptr = blocks[i].c;
419 sha256_multi_block(ctx, edges, n4x);
421 for (i = 0; i < x4; i++) {
422 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
423 unsigned char *out0 = out;
425 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
426 ciph_d[i].inp = ciph_d[i].out;
431 PUTU32(out + 0, ctx->A[i]);
432 PUTU32(out + 4, ctx->B[i]);
433 PUTU32(out + 8, ctx->C[i]);
434 PUTU32(out + 12, ctx->D[i]);
435 PUTU32(out + 16, ctx->E[i]);
436 PUTU32(out + 20, ctx->F[i]);
437 PUTU32(out + 24, ctx->G[i]);
438 PUTU32(out + 28, ctx->H[i]);
444 for (j = 0; j <= pad; j++)
448 ciph_d[i].blocks = (len - processed) / 16;
449 len += 16; /* account for explicit iv */
452 out0[0] = ((u8 *)key->md.data)[8];
453 out0[1] = ((u8 *)key->md.data)[9];
454 out0[2] = ((u8 *)key->md.data)[10];
455 out0[3] = (u8)(len >> 8);
462 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
464 OPENSSL_cleanse(blocks, sizeof(blocks));
465 OPENSSL_cleanse(ctx, sizeof(*ctx));
471 static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx,
473 const unsigned char *in, size_t len)
475 EVP_AES_HMAC_SHA256 *key = data(ctx);
477 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
480 # if defined(STITCHED_CALL)
481 size_t aes_off = 0, blocks;
483 sha_off = SHA256_CBLOCK - key->md.num;
486 key->payload_length = NO_PAYLOAD_LENGTH;
488 if (len % AES_BLOCK_SIZE)
492 if (plen == NO_PAYLOAD_LENGTH)
495 ((plen + SHA256_DIGEST_LENGTH +
496 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
498 else if (key->aux.tls_ver >= TLS1_1_VERSION)
501 # if defined(STITCHED_CALL)
503 * Assembly stitch handles AVX-capable processors, but its
504 * performance is not optimal on AMD Jaguar, ~40% worse, for
505 * unknown reasons. Incidentally processor in question supports
506 * AVX, but not AMD-specific XOP extension, which can be used
507 * to identify it and avoid stitch invocation. So that after we
508 * establish that current CPU supports AVX, we even see if it's
509 * either even XOP-capable Bulldozer-based or GenuineIntel one.
510 * But SHAEXT-capable go ahead...
512 if (((OPENSSL_ia32cap_P[2] & (1 << 29)) || /* SHAEXT? */
513 ((OPENSSL_ia32cap_P[1] & (1 << (60 - 32))) && /* AVX? */
514 ((OPENSSL_ia32cap_P[1] & (1 << (43 - 32))) /* XOP? */
515 | (OPENSSL_ia32cap_P[0] & (1 << 30))))) && /* "Intel CPU"? */
516 plen > (sha_off + iv) &&
517 (blocks = (plen - (sha_off + iv)) / SHA256_CBLOCK)) {
518 SHA256_Update(&key->md, in + iv, sha_off);
520 (void)aesni_cbc_sha256_enc(in, out, blocks, &key->ks,
521 ctx->iv, &key->md, in + iv + sha_off);
522 blocks *= SHA256_CBLOCK;
525 key->md.Nh += blocks >> 29;
526 key->md.Nl += blocks <<= 3;
527 if (key->md.Nl < (unsigned int)blocks)
534 SHA256_Update(&key->md, in + sha_off, plen - sha_off);
536 if (plen != len) { /* "TLS" mode of operation */
538 memcpy(out + aes_off, in + aes_off, plen - aes_off);
540 /* calculate HMAC and append it to payload */
541 SHA256_Final(out + plen, &key->md);
543 SHA256_Update(&key->md, out + plen, SHA256_DIGEST_LENGTH);
544 SHA256_Final(out + plen, &key->md);
546 /* pad the payload|hmac */
547 plen += SHA256_DIGEST_LENGTH;
548 for (l = len - plen - 1; plen < len; plen++)
550 /* encrypt HMAC|padding at once */
551 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
552 &key->ks, ctx->iv, 1);
554 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
555 &key->ks, ctx->iv, 1);
559 unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)];
560 unsigned char c[64 + SHA256_DIGEST_LENGTH];
563 /* arrange cache line alignment */
564 pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64));
566 /* decrypt HMAC|padding at once */
567 aesni_cbc_encrypt(in, out, len, &key->ks, ctx->iv, 0);
569 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
570 size_t inp_len, mask, j, i;
571 unsigned int res, maxpad, pad, bitlen;
574 unsigned int u[SHA_LBLOCK];
575 unsigned char c[SHA256_CBLOCK];
576 } *data = (void *)key->md.data;
578 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
582 if (len < (iv + SHA256_DIGEST_LENGTH + 1))
585 /* omit explicit iv */
589 /* figure out payload length */
591 maxpad = len - (SHA256_DIGEST_LENGTH + 1);
592 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
595 mask = constant_time_ge(maxpad, pad);
598 * If pad is invalid then we will fail the above test but we must
599 * continue anyway because we are in constant time code. However,
600 * we'll use the maxpad value instead of the supplied pad to make
601 * sure we perform well defined pointer arithmetic.
603 pad = constant_time_select(mask, pad, maxpad);
605 inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1);
607 key->aux.tls_aad[plen - 2] = inp_len >> 8;
608 key->aux.tls_aad[plen - 1] = inp_len;
612 SHA256_Update(&key->md, key->aux.tls_aad, plen);
615 len -= SHA256_DIGEST_LENGTH; /* amend mac */
616 if (len >= (256 + SHA256_CBLOCK)) {
617 j = (len - (256 + SHA256_CBLOCK)) & (0 - SHA256_CBLOCK);
618 j += SHA256_CBLOCK - key->md.num;
619 SHA256_Update(&key->md, out, j);
625 /* but pretend as if we hashed padded payload */
626 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
628 bitlen = BSWAP4(bitlen);
631 mac.c[1] = (unsigned char)(bitlen >> 16);
632 mac.c[2] = (unsigned char)(bitlen >> 8);
633 mac.c[3] = (unsigned char)bitlen;
646 for (res = key->md.num, j = 0; j < len; j++) {
648 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
650 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
651 data->c[res++] = (unsigned char)c;
653 if (res != SHA256_CBLOCK)
656 /* j is not incremented yet */
657 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
658 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
659 sha256_block_data_order(&key->md, data, 1);
660 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
661 pmac->u[0] |= key->md.h[0] & mask;
662 pmac->u[1] |= key->md.h[1] & mask;
663 pmac->u[2] |= key->md.h[2] & mask;
664 pmac->u[3] |= key->md.h[3] & mask;
665 pmac->u[4] |= key->md.h[4] & mask;
666 pmac->u[5] |= key->md.h[5] & mask;
667 pmac->u[6] |= key->md.h[6] & mask;
668 pmac->u[7] |= key->md.h[7] & mask;
672 for (i = res; i < SHA256_CBLOCK; i++, j++)
675 if (res > SHA256_CBLOCK - 8) {
676 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
677 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
678 sha256_block_data_order(&key->md, data, 1);
679 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
680 pmac->u[0] |= key->md.h[0] & mask;
681 pmac->u[1] |= key->md.h[1] & mask;
682 pmac->u[2] |= key->md.h[2] & mask;
683 pmac->u[3] |= key->md.h[3] & mask;
684 pmac->u[4] |= key->md.h[4] & mask;
685 pmac->u[5] |= key->md.h[5] & mask;
686 pmac->u[6] |= key->md.h[6] & mask;
687 pmac->u[7] |= key->md.h[7] & mask;
689 memset(data, 0, SHA256_CBLOCK);
692 data->u[SHA_LBLOCK - 1] = bitlen;
693 sha256_block_data_order(&key->md, data, 1);
694 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
695 pmac->u[0] |= key->md.h[0] & mask;
696 pmac->u[1] |= key->md.h[1] & mask;
697 pmac->u[2] |= key->md.h[2] & mask;
698 pmac->u[3] |= key->md.h[3] & mask;
699 pmac->u[4] |= key->md.h[4] & mask;
700 pmac->u[5] |= key->md.h[5] & mask;
701 pmac->u[6] |= key->md.h[6] & mask;
702 pmac->u[7] |= key->md.h[7] & mask;
705 pmac->u[0] = BSWAP4(pmac->u[0]);
706 pmac->u[1] = BSWAP4(pmac->u[1]);
707 pmac->u[2] = BSWAP4(pmac->u[2]);
708 pmac->u[3] = BSWAP4(pmac->u[3]);
709 pmac->u[4] = BSWAP4(pmac->u[4]);
710 pmac->u[5] = BSWAP4(pmac->u[5]);
711 pmac->u[6] = BSWAP4(pmac->u[6]);
712 pmac->u[7] = BSWAP4(pmac->u[7]);
714 for (i = 0; i < 8; i++) {
716 pmac->c[4 * i + 0] = (unsigned char)(res >> 24);
717 pmac->c[4 * i + 1] = (unsigned char)(res >> 16);
718 pmac->c[4 * i + 2] = (unsigned char)(res >> 8);
719 pmac->c[4 * i + 3] = (unsigned char)res;
722 len += SHA256_DIGEST_LENGTH;
724 SHA256_Update(&key->md, out, inp_len);
726 SHA256_Final(pmac->c, &key->md);
729 unsigned int inp_blocks, pad_blocks;
731 /* but pretend as if we hashed padded payload */
733 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
734 res += (unsigned int)(len - inp_len);
735 pad_blocks = res / SHA256_CBLOCK;
736 res %= SHA256_CBLOCK;
738 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
739 for (; inp_blocks < pad_blocks; inp_blocks++)
740 sha1_block_data_order(&key->md, data, 1);
744 SHA256_Update(&key->md, pmac->c, SHA256_DIGEST_LENGTH);
745 SHA256_Final(pmac->c, &key->md);
753 out + len - 1 - maxpad - SHA256_DIGEST_LENGTH;
754 size_t off = out - p;
755 unsigned int c, cmask;
757 maxpad += SHA256_DIGEST_LENGTH;
758 for (res = 0, i = 0, j = 0; j < maxpad; j++) {
761 ((int)(j - off - SHA256_DIGEST_LENGTH)) >>
762 (sizeof(int) * 8 - 1);
763 res |= (c ^ pad) & ~cmask; /* ... and padding */
764 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
765 res |= (c ^ pmac->c[i]) & cmask;
768 maxpad -= SHA256_DIGEST_LENGTH;
770 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
774 for (res = 0, i = 0; i < SHA256_DIGEST_LENGTH; i++)
775 res |= out[i] ^ pmac->c[i];
776 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
780 pad = (pad & ~res) | (maxpad & res);
781 out = out + len - 1 - pad;
782 for (res = 0, i = 0; i < pad; i++)
785 res = (0 - res) >> (sizeof(res) * 8 - 1);
790 SHA256_Update(&key->md, out, len);
797 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
800 EVP_AES_HMAC_SHA256 *key = data(ctx);
803 case EVP_CTRL_AEAD_SET_MAC_KEY:
806 unsigned char hmac_key[64];
808 memset(hmac_key, 0, sizeof(hmac_key));
810 if (arg > (int)sizeof(hmac_key)) {
811 SHA256_Init(&key->head);
812 SHA256_Update(&key->head, ptr, arg);
813 SHA256_Final(hmac_key, &key->head);
815 memcpy(hmac_key, ptr, arg);
818 for (i = 0; i < sizeof(hmac_key); i++)
819 hmac_key[i] ^= 0x36; /* ipad */
820 SHA256_Init(&key->head);
821 SHA256_Update(&key->head, hmac_key, sizeof(hmac_key));
823 for (i = 0; i < sizeof(hmac_key); i++)
824 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
825 SHA256_Init(&key->tail);
826 SHA256_Update(&key->tail, hmac_key, sizeof(hmac_key));
828 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
832 case EVP_CTRL_AEAD_TLS1_AAD:
834 unsigned char *p = ptr;
837 if (arg != EVP_AEAD_TLS1_AAD_LEN)
840 len = p[arg - 2] << 8 | p[arg - 1];
843 key->payload_length = len;
844 if ((key->aux.tls_ver =
845 p[arg - 4] << 8 | p[arg - 3]) >= TLS1_1_VERSION) {
846 if (len < AES_BLOCK_SIZE)
848 len -= AES_BLOCK_SIZE;
849 p[arg - 2] = len >> 8;
853 SHA256_Update(&key->md, p, arg);
855 return (int)(((len + SHA256_DIGEST_LENGTH +
856 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
859 memcpy(key->aux.tls_aad, ptr, arg);
860 key->payload_length = arg;
862 return SHA256_DIGEST_LENGTH;
865 # if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
866 case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
867 return (int)(5 + 16 + ((arg + 32 + 16) & -16));
868 case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
870 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
871 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
872 unsigned int n4x = 1, x4;
873 unsigned int frag, last, packlen, inp_len;
875 if (arg < (int)sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
878 inp_len = param->inp[11] << 8 | param->inp[12];
881 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
886 return 0; /* too short */
888 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
890 } else if ((n4x = param->interleave / 4) && n4x <= 2)
891 inp_len = param->len;
896 SHA256_Update(&key->md, param->inp, 13);
901 frag = inp_len >> n4x;
902 last = inp_len + frag - (frag << n4x);
903 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
908 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
909 packlen = (packlen << n4x) - packlen;
910 packlen += 5 + 16 + ((last + 32 + 16) & -16);
912 param->interleave = x4;
916 return -1; /* not yet */
918 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
920 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
921 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
923 return (int)tls1_1_multi_block_encrypt(key, param->out,
924 param->inp, param->len,
925 param->interleave / 4);
927 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
934 static EVP_CIPHER aesni_128_cbc_hmac_sha256_cipher = {
935 # ifdef NID_aes_128_cbc_hmac_sha256
936 NID_aes_128_cbc_hmac_sha256,
941 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
942 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
943 aesni_cbc_hmac_sha256_init_key,
944 aesni_cbc_hmac_sha256_cipher,
946 sizeof(EVP_AES_HMAC_SHA256),
947 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
948 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
949 aesni_cbc_hmac_sha256_ctrl,
953 static EVP_CIPHER aesni_256_cbc_hmac_sha256_cipher = {
954 # ifdef NID_aes_256_cbc_hmac_sha256
955 NID_aes_256_cbc_hmac_sha256,
960 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
961 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
962 aesni_cbc_hmac_sha256_init_key,
963 aesni_cbc_hmac_sha256_cipher,
965 sizeof(EVP_AES_HMAC_SHA256),
966 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
967 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
968 aesni_cbc_hmac_sha256_ctrl,
972 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
974 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
975 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
976 &aesni_128_cbc_hmac_sha256_cipher : NULL);
979 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
981 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
982 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
983 &aesni_256_cbc_hmac_sha256_cipher : NULL);
986 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
991 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)