2 * Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the OpenSSL license (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
10 #include <openssl/opensslconf.h>
15 #include <openssl/evp.h>
16 #include <openssl/objects.h>
17 #include <openssl/aes.h>
18 #include <openssl/sha.h>
19 #include <openssl/rand.h>
20 #include <openssl/rand_drbg.h>
21 #include "modes_lcl.h"
22 #include "internal/evp_int.h"
23 #include "internal/constant_time_locl.h"
28 SHA_CTX head, tail, md;
29 size_t payload_length; /* AAD length in decrypt case */
32 unsigned char tls_aad[16]; /* 13 used */
36 #define NO_PAYLOAD_LENGTH ((size_t)-1)
38 #if defined(AES_ASM) && ( \
39 defined(__x86_64) || defined(__x86_64__) || \
40 defined(_M_AMD64) || defined(_M_X64) )
42 extern unsigned int OPENSSL_ia32cap_P[];
43 # define AESNI_CAPABLE (1<<(57-32))
45 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
47 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
50 void aesni_cbc_encrypt(const unsigned char *in,
53 const AES_KEY *key, unsigned char *ivec, int enc);
55 void aesni_cbc_sha1_enc(const void *inp, void *out, size_t blocks,
56 const AES_KEY *key, unsigned char iv[16],
57 SHA_CTX *ctx, const void *in0);
59 void aesni256_cbc_sha1_dec(const void *inp, void *out, size_t blocks,
60 const AES_KEY *key, unsigned char iv[16],
61 SHA_CTX *ctx, const void *in0);
63 # define data(ctx) ((EVP_AES_HMAC_SHA1 *)EVP_CIPHER_CTX_get_cipher_data(ctx))
65 static int aesni_cbc_hmac_sha1_init_key(EVP_CIPHER_CTX *ctx,
66 const unsigned char *inkey,
67 const unsigned char *iv, int enc)
69 EVP_AES_HMAC_SHA1 *key = data(ctx);
73 ret = aesni_set_encrypt_key(inkey,
74 EVP_CIPHER_CTX_key_length(ctx) * 8,
77 ret = aesni_set_decrypt_key(inkey,
78 EVP_CIPHER_CTX_key_length(ctx) * 8,
81 SHA1_Init(&key->head); /* handy when benchmarking */
82 key->tail = key->head;
85 key->payload_length = NO_PAYLOAD_LENGTH;
87 return ret < 0 ? 0 : 1;
90 # define STITCHED_CALL
91 # undef STITCHED_DECRYPT_CALL
93 # if !defined(STITCHED_CALL)
97 void sha1_block_data_order(void *c, const void *p, size_t len);
99 static void sha1_update(SHA_CTX *c, const void *data, size_t len)
101 const unsigned char *ptr = data;
104 if ((res = c->num)) {
105 res = SHA_CBLOCK - res;
108 SHA1_Update(c, ptr, res);
113 res = len % SHA_CBLOCK;
117 sha1_block_data_order(c, ptr, len / SHA_CBLOCK);
122 if (c->Nl < (unsigned int)len)
127 SHA1_Update(c, ptr, res);
133 # define SHA1_Update sha1_update
135 # if !defined(OPENSSL_NO_MULTIBLOCK)
138 unsigned int A[8], B[8], C[8], D[8], E[8];
141 const unsigned char *ptr;
145 void sha1_multi_block(SHA1_MB_CTX *, const HASH_DESC *, int);
148 const unsigned char *inp;
154 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
156 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA1 *key,
158 const unsigned char *inp,
159 size_t inp_len, int n4x,
161 { /* n4x is 1 or 2 */
162 HASH_DESC hash_d[8], edges[8];
164 unsigned char storage[sizeof(SHA1_MB_CTX) + 32];
171 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
179 /* ask for IVs in bulk */
182 if (RAND_DRBG_bytes(drbg, IVs, 16 * x4) == 0)
184 } else if (RAND_bytes(IVs, 16 * x4) <= 0) {
188 ctx = (SHA1_MB_CTX *) (storage + 32 - ((size_t)storage % 32)); /* align */
190 frag = (unsigned int)inp_len >> (1 + n4x);
191 last = (unsigned int)inp_len + frag - (frag << (1 + n4x));
192 if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) {
197 packlen = 5 + 16 + ((frag + 20 + 16) & -16);
199 /* populate descriptors with pointers and IVs */
202 /* 5+16 is place for header and explicit IV */
203 ciph_d[0].out = out + 5 + 16;
204 memcpy(ciph_d[0].out - 16, IVs, 16);
205 memcpy(ciph_d[0].iv, IVs, 16);
208 for (i = 1; i < x4; i++) {
209 ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag;
210 ciph_d[i].out = ciph_d[i - 1].out + packlen;
211 memcpy(ciph_d[i].out - 16, IVs, 16);
212 memcpy(ciph_d[i].iv, IVs, 16);
217 memcpy(blocks[0].c, key->md.data, 8);
218 seqnum = BSWAP8(blocks[0].q[0]);
220 for (i = 0; i < x4; i++) {
221 unsigned int len = (i == (x4 - 1) ? last : frag);
222 # if !defined(BSWAP8)
223 unsigned int carry, j;
226 ctx->A[i] = key->md.h0;
227 ctx->B[i] = key->md.h1;
228 ctx->C[i] = key->md.h2;
229 ctx->D[i] = key->md.h3;
230 ctx->E[i] = key->md.h4;
234 blocks[i].q[0] = BSWAP8(seqnum + i);
236 for (carry = i, j = 8; j--;) {
237 blocks[i].c[j] = ((u8 *)key->md.data)[j] + carry;
238 carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1);
241 blocks[i].c[8] = ((u8 *)key->md.data)[8];
242 blocks[i].c[9] = ((u8 *)key->md.data)[9];
243 blocks[i].c[10] = ((u8 *)key->md.data)[10];
245 blocks[i].c[11] = (u8)(len >> 8);
246 blocks[i].c[12] = (u8)(len);
248 memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13);
249 hash_d[i].ptr += 64 - 13;
250 hash_d[i].blocks = (len - (64 - 13)) / 64;
252 edges[i].ptr = blocks[i].c;
256 /* hash 13-byte headers and first 64-13 bytes of inputs */
257 sha1_multi_block(ctx, edges, n4x);
258 /* hash bulk inputs */
259 # define MAXCHUNKSIZE 2048
261 # error "MAXCHUNKSIZE is not divisible by 64"
264 * goal is to minimize pressure on L1 cache by moving in shorter steps,
265 * so that hashed data is still in the cache by the time we encrypt it
267 minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64;
268 if (minblocks > MAXCHUNKSIZE / 64) {
269 for (i = 0; i < x4; i++) {
270 edges[i].ptr = hash_d[i].ptr;
271 edges[i].blocks = MAXCHUNKSIZE / 64;
272 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
275 sha1_multi_block(ctx, edges, n4x);
276 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
278 for (i = 0; i < x4; i++) {
279 edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE;
280 hash_d[i].blocks -= MAXCHUNKSIZE / 64;
281 edges[i].blocks = MAXCHUNKSIZE / 64;
282 ciph_d[i].inp += MAXCHUNKSIZE;
283 ciph_d[i].out += MAXCHUNKSIZE;
284 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
285 memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16);
287 processed += MAXCHUNKSIZE;
288 minblocks -= MAXCHUNKSIZE / 64;
289 } while (minblocks > MAXCHUNKSIZE / 64);
293 sha1_multi_block(ctx, hash_d, n4x);
295 memset(blocks, 0, sizeof(blocks));
296 for (i = 0; i < x4; i++) {
297 unsigned int len = (i == (x4 - 1) ? last : frag),
298 off = hash_d[i].blocks * 64;
299 const unsigned char *ptr = hash_d[i].ptr + off;
301 off = (len - processed) - (64 - 13) - off; /* remainder actually */
302 memcpy(blocks[i].c, ptr, off);
303 blocks[i].c[off] = 0x80;
304 len += 64 + 13; /* 64 is HMAC header */
305 len *= 8; /* convert to bits */
306 if (off < (64 - 8)) {
308 blocks[i].d[15] = BSWAP4(len);
310 PUTU32(blocks[i].c + 60, len);
315 blocks[i].d[31] = BSWAP4(len);
317 PUTU32(blocks[i].c + 124, len);
321 edges[i].ptr = blocks[i].c;
324 /* hash input tails and finalize */
325 sha1_multi_block(ctx, edges, n4x);
327 memset(blocks, 0, sizeof(blocks));
328 for (i = 0; i < x4; i++) {
330 blocks[i].d[0] = BSWAP4(ctx->A[i]);
331 ctx->A[i] = key->tail.h0;
332 blocks[i].d[1] = BSWAP4(ctx->B[i]);
333 ctx->B[i] = key->tail.h1;
334 blocks[i].d[2] = BSWAP4(ctx->C[i]);
335 ctx->C[i] = key->tail.h2;
336 blocks[i].d[3] = BSWAP4(ctx->D[i]);
337 ctx->D[i] = key->tail.h3;
338 blocks[i].d[4] = BSWAP4(ctx->E[i]);
339 ctx->E[i] = key->tail.h4;
340 blocks[i].c[20] = 0x80;
341 blocks[i].d[15] = BSWAP4((64 + 20) * 8);
343 PUTU32(blocks[i].c + 0, ctx->A[i]);
344 ctx->A[i] = key->tail.h0;
345 PUTU32(blocks[i].c + 4, ctx->B[i]);
346 ctx->B[i] = key->tail.h1;
347 PUTU32(blocks[i].c + 8, ctx->C[i]);
348 ctx->C[i] = key->tail.h2;
349 PUTU32(blocks[i].c + 12, ctx->D[i]);
350 ctx->D[i] = key->tail.h3;
351 PUTU32(blocks[i].c + 16, ctx->E[i]);
352 ctx->E[i] = key->tail.h4;
353 blocks[i].c[20] = 0x80;
354 PUTU32(blocks[i].c + 60, (64 + 20) * 8);
356 edges[i].ptr = blocks[i].c;
361 sha1_multi_block(ctx, edges, n4x);
363 for (i = 0; i < x4; i++) {
364 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
365 unsigned char *out0 = out;
367 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
368 ciph_d[i].inp = ciph_d[i].out;
373 PUTU32(out + 0, ctx->A[i]);
374 PUTU32(out + 4, ctx->B[i]);
375 PUTU32(out + 8, ctx->C[i]);
376 PUTU32(out + 12, ctx->D[i]);
377 PUTU32(out + 16, ctx->E[i]);
383 for (j = 0; j <= pad; j++)
387 ciph_d[i].blocks = (len - processed) / 16;
388 len += 16; /* account for explicit iv */
391 out0[0] = ((u8 *)key->md.data)[8];
392 out0[1] = ((u8 *)key->md.data)[9];
393 out0[2] = ((u8 *)key->md.data)[10];
394 out0[3] = (u8)(len >> 8);
401 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
403 OPENSSL_cleanse(blocks, sizeof(blocks));
404 OPENSSL_cleanse(ctx, sizeof(*ctx));
410 static int aesni_cbc_hmac_sha1_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
411 const unsigned char *in, size_t len)
413 EVP_AES_HMAC_SHA1 *key = data(ctx);
415 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
418 # if defined(STITCHED_CALL)
419 size_t aes_off = 0, blocks;
421 sha_off = SHA_CBLOCK - key->md.num;
424 key->payload_length = NO_PAYLOAD_LENGTH;
426 if (len % AES_BLOCK_SIZE)
429 if (EVP_CIPHER_CTX_encrypting(ctx)) {
430 if (plen == NO_PAYLOAD_LENGTH)
433 ((plen + SHA_DIGEST_LENGTH +
434 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
436 else if (key->aux.tls_ver >= TLS1_1_VERSION)
439 # if defined(STITCHED_CALL)
440 if (plen > (sha_off + iv)
441 && (blocks = (plen - (sha_off + iv)) / SHA_CBLOCK)) {
442 SHA1_Update(&key->md, in + iv, sha_off);
444 aesni_cbc_sha1_enc(in, out, blocks, &key->ks,
445 EVP_CIPHER_CTX_iv_noconst(ctx),
446 &key->md, in + iv + sha_off);
447 blocks *= SHA_CBLOCK;
450 key->md.Nh += blocks >> 29;
451 key->md.Nl += blocks <<= 3;
452 if (key->md.Nl < (unsigned int)blocks)
459 SHA1_Update(&key->md, in + sha_off, plen - sha_off);
461 if (plen != len) { /* "TLS" mode of operation */
463 memcpy(out + aes_off, in + aes_off, plen - aes_off);
465 /* calculate HMAC and append it to payload */
466 SHA1_Final(out + plen, &key->md);
468 SHA1_Update(&key->md, out + plen, SHA_DIGEST_LENGTH);
469 SHA1_Final(out + plen, &key->md);
471 /* pad the payload|hmac */
472 plen += SHA_DIGEST_LENGTH;
473 for (l = len - plen - 1; plen < len; plen++)
475 /* encrypt HMAC|padding at once */
476 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
477 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
479 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
480 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
484 unsigned int u[SHA_DIGEST_LENGTH / sizeof(unsigned int)];
485 unsigned char c[32 + SHA_DIGEST_LENGTH];
488 /* arrange cache line alignment */
489 pmac = (void *)(((size_t)mac.c + 31) & ((size_t)0 - 32));
491 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
492 size_t inp_len, mask, j, i;
493 unsigned int res, maxpad, pad, bitlen;
496 unsigned int u[SHA_LBLOCK];
497 unsigned char c[SHA_CBLOCK];
498 } *data = (void *)key->md.data;
499 # if defined(STITCHED_DECRYPT_CALL)
500 unsigned char tail_iv[AES_BLOCK_SIZE];
504 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
506 if (len < (AES_BLOCK_SIZE + SHA_DIGEST_LENGTH + 1))
509 /* omit explicit iv */
510 memcpy(EVP_CIPHER_CTX_iv_noconst(ctx), in, AES_BLOCK_SIZE);
512 in += AES_BLOCK_SIZE;
513 out += AES_BLOCK_SIZE;
514 len -= AES_BLOCK_SIZE;
515 } else if (len < (SHA_DIGEST_LENGTH + 1))
518 # if defined(STITCHED_DECRYPT_CALL)
519 if (len >= 1024 && ctx->key_len == 32) {
520 /* decrypt last block */
521 memcpy(tail_iv, in + len - 2 * AES_BLOCK_SIZE,
523 aesni_cbc_encrypt(in + len - AES_BLOCK_SIZE,
524 out + len - AES_BLOCK_SIZE, AES_BLOCK_SIZE,
525 &key->ks, tail_iv, 0);
529 /* decrypt HMAC|padding at once */
530 aesni_cbc_encrypt(in, out, len, &key->ks,
531 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
533 /* figure out payload length */
535 maxpad = len - (SHA_DIGEST_LENGTH + 1);
536 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
539 mask = constant_time_ge(maxpad, pad);
542 * If pad is invalid then we will fail the above test but we must
543 * continue anyway because we are in constant time code. However,
544 * we'll use the maxpad value instead of the supplied pad to make
545 * sure we perform well defined pointer arithmetic.
547 pad = constant_time_select(mask, pad, maxpad);
549 inp_len = len - (SHA_DIGEST_LENGTH + pad + 1);
551 key->aux.tls_aad[plen - 2] = inp_len >> 8;
552 key->aux.tls_aad[plen - 1] = inp_len;
556 SHA1_Update(&key->md, key->aux.tls_aad, plen);
558 # if defined(STITCHED_DECRYPT_CALL)
560 blocks = (len - (256 + 32 + SHA_CBLOCK)) / SHA_CBLOCK;
561 aes_off = len - AES_BLOCK_SIZE - blocks * SHA_CBLOCK;
562 sha_off = SHA_CBLOCK - plen;
564 aesni_cbc_encrypt(in, out, aes_off, &key->ks, ctx->iv, 0);
566 SHA1_Update(&key->md, out, sha_off);
567 aesni256_cbc_sha1_dec(in + aes_off,
568 out + aes_off, blocks, &key->ks,
569 ctx->iv, &key->md, out + sha_off);
571 sha_off += blocks *= SHA_CBLOCK;
576 key->md.Nl += (blocks << 3); /* at most 18 bits */
577 memcpy(ctx->iv, tail_iv, AES_BLOCK_SIZE);
581 # if 1 /* see original reference version in #else */
582 len -= SHA_DIGEST_LENGTH; /* amend mac */
583 if (len >= (256 + SHA_CBLOCK)) {
584 j = (len - (256 + SHA_CBLOCK)) & (0 - SHA_CBLOCK);
585 j += SHA_CBLOCK - key->md.num;
586 SHA1_Update(&key->md, out, j);
592 /* but pretend as if we hashed padded payload */
593 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
595 bitlen = BSWAP4(bitlen);
598 mac.c[1] = (unsigned char)(bitlen >> 16);
599 mac.c[2] = (unsigned char)(bitlen >> 8);
600 mac.c[3] = (unsigned char)bitlen;
610 for (res = key->md.num, j = 0; j < len; j++) {
612 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
614 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
615 data->c[res++] = (unsigned char)c;
617 if (res != SHA_CBLOCK)
620 /* j is not incremented yet */
621 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
622 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
623 sha1_block_data_order(&key->md, data, 1);
624 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
625 pmac->u[0] |= key->md.h0 & mask;
626 pmac->u[1] |= key->md.h1 & mask;
627 pmac->u[2] |= key->md.h2 & mask;
628 pmac->u[3] |= key->md.h3 & mask;
629 pmac->u[4] |= key->md.h4 & mask;
633 for (i = res; i < SHA_CBLOCK; i++, j++)
636 if (res > SHA_CBLOCK - 8) {
637 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
638 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
639 sha1_block_data_order(&key->md, data, 1);
640 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
641 pmac->u[0] |= key->md.h0 & mask;
642 pmac->u[1] |= key->md.h1 & mask;
643 pmac->u[2] |= key->md.h2 & mask;
644 pmac->u[3] |= key->md.h3 & mask;
645 pmac->u[4] |= key->md.h4 & mask;
647 memset(data, 0, SHA_CBLOCK);
650 data->u[SHA_LBLOCK - 1] = bitlen;
651 sha1_block_data_order(&key->md, data, 1);
652 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
653 pmac->u[0] |= key->md.h0 & mask;
654 pmac->u[1] |= key->md.h1 & mask;
655 pmac->u[2] |= key->md.h2 & mask;
656 pmac->u[3] |= key->md.h3 & mask;
657 pmac->u[4] |= key->md.h4 & mask;
660 pmac->u[0] = BSWAP4(pmac->u[0]);
661 pmac->u[1] = BSWAP4(pmac->u[1]);
662 pmac->u[2] = BSWAP4(pmac->u[2]);
663 pmac->u[3] = BSWAP4(pmac->u[3]);
664 pmac->u[4] = BSWAP4(pmac->u[4]);
666 for (i = 0; i < 5; i++) {
668 pmac->c[4 * i + 0] = (unsigned char)(res >> 24);
669 pmac->c[4 * i + 1] = (unsigned char)(res >> 16);
670 pmac->c[4 * i + 2] = (unsigned char)(res >> 8);
671 pmac->c[4 * i + 3] = (unsigned char)res;
674 len += SHA_DIGEST_LENGTH;
675 # else /* pre-lucky-13 reference version of above */
676 SHA1_Update(&key->md, out, inp_len);
678 SHA1_Final(pmac->c, &key->md);
681 unsigned int inp_blocks, pad_blocks;
683 /* but pretend as if we hashed padded payload */
685 1 + ((SHA_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
686 res += (unsigned int)(len - inp_len);
687 pad_blocks = res / SHA_CBLOCK;
690 1 + ((SHA_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
691 for (; inp_blocks < pad_blocks; inp_blocks++)
692 sha1_block_data_order(&key->md, data, 1);
696 SHA1_Update(&key->md, pmac->c, SHA_DIGEST_LENGTH);
697 SHA1_Final(pmac->c, &key->md);
702 # if 1 /* see original reference version in #else */
704 unsigned char *p = out + len - 1 - maxpad - SHA_DIGEST_LENGTH;
705 size_t off = out - p;
706 unsigned int c, cmask;
708 maxpad += SHA_DIGEST_LENGTH;
709 for (res = 0, i = 0, j = 0; j < maxpad; j++) {
712 ((int)(j - off - SHA_DIGEST_LENGTH)) >> (sizeof(int) *
714 res |= (c ^ pad) & ~cmask; /* ... and padding */
715 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
716 res |= (c ^ pmac->c[i]) & cmask;
719 maxpad -= SHA_DIGEST_LENGTH;
721 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
724 # else /* pre-lucky-13 reference version of above */
725 for (res = 0, i = 0; i < SHA_DIGEST_LENGTH; i++)
726 res |= out[i] ^ pmac->c[i];
727 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
731 pad = (pad & ~res) | (maxpad & res);
732 out = out + len - 1 - pad;
733 for (res = 0, i = 0; i < pad; i++)
736 res = (0 - res) >> (sizeof(res) * 8 - 1);
741 # if defined(STITCHED_DECRYPT_CALL)
742 if (len >= 1024 && ctx->key_len == 32) {
743 if (sha_off %= SHA_CBLOCK)
744 blocks = (len - 3 * SHA_CBLOCK) / SHA_CBLOCK;
746 blocks = (len - 2 * SHA_CBLOCK) / SHA_CBLOCK;
747 aes_off = len - blocks * SHA_CBLOCK;
749 aesni_cbc_encrypt(in, out, aes_off, &key->ks, ctx->iv, 0);
750 SHA1_Update(&key->md, out, sha_off);
751 aesni256_cbc_sha1_dec(in + aes_off,
752 out + aes_off, blocks, &key->ks,
753 ctx->iv, &key->md, out + sha_off);
755 sha_off += blocks *= SHA_CBLOCK;
759 key->md.Nh += blocks >> 29;
760 key->md.Nl += blocks <<= 3;
761 if (key->md.Nl < (unsigned int)blocks)
765 /* decrypt HMAC|padding at once */
766 aesni_cbc_encrypt(in, out, len, &key->ks,
767 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
769 SHA1_Update(&key->md, out, len);
776 static int aesni_cbc_hmac_sha1_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
779 EVP_AES_HMAC_SHA1 *key = data(ctx);
782 case EVP_CTRL_AEAD_SET_MAC_KEY:
785 unsigned char hmac_key[64];
787 memset(hmac_key, 0, sizeof(hmac_key));
789 if (arg > (int)sizeof(hmac_key)) {
790 SHA1_Init(&key->head);
791 SHA1_Update(&key->head, ptr, arg);
792 SHA1_Final(hmac_key, &key->head);
794 memcpy(hmac_key, ptr, arg);
797 for (i = 0; i < sizeof(hmac_key); i++)
798 hmac_key[i] ^= 0x36; /* ipad */
799 SHA1_Init(&key->head);
800 SHA1_Update(&key->head, hmac_key, sizeof(hmac_key));
802 for (i = 0; i < sizeof(hmac_key); i++)
803 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
804 SHA1_Init(&key->tail);
805 SHA1_Update(&key->tail, hmac_key, sizeof(hmac_key));
807 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
811 case EVP_CTRL_AEAD_TLS1_AAD:
813 unsigned char *p = ptr;
816 if (arg != EVP_AEAD_TLS1_AAD_LEN)
819 len = p[arg - 2] << 8 | p[arg - 1];
821 if (EVP_CIPHER_CTX_encrypting(ctx)) {
822 key->payload_length = len;
823 if ((key->aux.tls_ver =
824 p[arg - 4] << 8 | p[arg - 3]) >= TLS1_1_VERSION) {
825 if (len < AES_BLOCK_SIZE)
827 len -= AES_BLOCK_SIZE;
828 p[arg - 2] = len >> 8;
832 SHA1_Update(&key->md, p, arg);
834 return (int)(((len + SHA_DIGEST_LENGTH +
835 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
838 memcpy(key->aux.tls_aad, ptr, arg);
839 key->payload_length = arg;
841 return SHA_DIGEST_LENGTH;
844 # if !defined(OPENSSL_NO_MULTIBLOCK)
845 case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
846 return (int)(5 + 16 + ((arg + 20 + 16) & -16));
847 case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
849 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
850 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
851 unsigned int n4x = 1, x4;
852 unsigned int frag, last, packlen, inp_len;
854 if (arg < (int)sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
857 inp_len = param->inp[11] << 8 | param->inp[12];
859 if (EVP_CIPHER_CTX_encrypting(ctx)) {
860 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
865 return 0; /* too short */
867 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
869 } else if ((n4x = param->interleave / 4) && n4x <= 2)
870 inp_len = param->len;
875 SHA1_Update(&key->md, param->inp, 13);
880 frag = inp_len >> n4x;
881 last = inp_len + frag - (frag << n4x);
882 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
887 packlen = 5 + 16 + ((frag + 20 + 16) & -16);
888 packlen = (packlen << n4x) - packlen;
889 packlen += 5 + 16 + ((last + 20 + 16) & -16);
891 param->interleave = x4;
895 return -1; /* not yet */
897 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
899 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
900 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
902 return (int)tls1_1_multi_block_encrypt(key, param->out,
903 param->inp, param->len,
904 param->interleave / 4,
907 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
914 static EVP_CIPHER aesni_128_cbc_hmac_sha1_cipher = {
915 # ifdef NID_aes_128_cbc_hmac_sha1
916 NID_aes_128_cbc_hmac_sha1,
920 AES_BLOCK_SIZE, 16, AES_BLOCK_SIZE,
921 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
922 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
923 aesni_cbc_hmac_sha1_init_key,
924 aesni_cbc_hmac_sha1_cipher,
926 sizeof(EVP_AES_HMAC_SHA1),
927 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
928 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
929 aesni_cbc_hmac_sha1_ctrl,
933 static EVP_CIPHER aesni_256_cbc_hmac_sha1_cipher = {
934 # ifdef NID_aes_256_cbc_hmac_sha1
935 NID_aes_256_cbc_hmac_sha1,
939 AES_BLOCK_SIZE, 32, AES_BLOCK_SIZE,
940 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
941 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
942 aesni_cbc_hmac_sha1_init_key,
943 aesni_cbc_hmac_sha1_cipher,
945 sizeof(EVP_AES_HMAC_SHA1),
946 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
947 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
948 aesni_cbc_hmac_sha1_ctrl,
952 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha1(void)
954 return (OPENSSL_ia32cap_P[1] & AESNI_CAPABLE ?
955 &aesni_128_cbc_hmac_sha1_cipher : NULL);
958 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha1(void)
960 return (OPENSSL_ia32cap_P[1] & AESNI_CAPABLE ?
961 &aesni_256_cbc_hmac_sha1_cipher : NULL);
964 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha1(void)
969 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha1(void)