2 * Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
4 * Licensed under the Apache License 2.0 (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
11 * AES low level APIs are deprecated for public use, but still ok for internal
12 * use where we're using them to implement the higher level EVP interface, as is
15 #include "internal/deprecated.h"
19 #include <openssl/opensslconf.h>
20 #include <openssl/evp.h>
21 #include <openssl/objects.h>
22 #include <openssl/aes.h>
23 #include <openssl/sha.h>
24 #include <openssl/rand.h>
25 #include "internal/cryptlib.h"
26 #include "crypto/modes.h"
27 #include "crypto/evp.h"
28 #include "internal/constant_time.h"
32 SHA_CTX head, tail, md;
33 size_t payload_length; /* AAD length in decrypt case */
36 unsigned char tls_aad[16]; /* 13 used */
40 #define NO_PAYLOAD_LENGTH ((size_t)-1)
42 #if defined(AES_ASM) && ( \
43 defined(__x86_64) || defined(__x86_64__) || \
44 defined(_M_AMD64) || defined(_M_X64) )
46 # define AESNI_CAPABLE (1<<(57-32))
48 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
50 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
53 void aesni_cbc_encrypt(const unsigned char *in,
56 const AES_KEY *key, unsigned char *ivec, int enc);
58 void aesni_cbc_sha1_enc(const void *inp, void *out, size_t blocks,
59 const AES_KEY *key, unsigned char iv[16],
60 SHA_CTX *ctx, const void *in0);
62 void aesni256_cbc_sha1_dec(const void *inp, void *out, size_t blocks,
63 const AES_KEY *key, unsigned char iv[16],
64 SHA_CTX *ctx, const void *in0);
66 # define data(ctx) ((EVP_AES_HMAC_SHA1 *)EVP_CIPHER_CTX_get_cipher_data(ctx))
68 static int aesni_cbc_hmac_sha1_init_key(EVP_CIPHER_CTX *ctx,
69 const unsigned char *inkey,
70 const unsigned char *iv, int enc)
72 EVP_AES_HMAC_SHA1 *key = data(ctx);
76 ret = aesni_set_encrypt_key(inkey,
77 EVP_CIPHER_CTX_key_length(ctx) * 8,
80 ret = aesni_set_decrypt_key(inkey,
81 EVP_CIPHER_CTX_key_length(ctx) * 8,
84 SHA1_Init(&key->head); /* handy when benchmarking */
85 key->tail = key->head;
88 key->payload_length = NO_PAYLOAD_LENGTH;
90 return ret < 0 ? 0 : 1;
93 # define STITCHED_CALL
94 # undef STITCHED_DECRYPT_CALL
96 # if !defined(STITCHED_CALL)
100 void sha1_block_data_order(void *c, const void *p, size_t len);
102 static void sha1_update(SHA_CTX *c, const void *data, size_t len)
104 const unsigned char *ptr = data;
107 if ((res = c->num)) {
108 res = SHA_CBLOCK - res;
111 SHA1_Update(c, ptr, res);
116 res = len % SHA_CBLOCK;
120 sha1_block_data_order(c, ptr, len / SHA_CBLOCK);
125 if (c->Nl < (unsigned int)len)
130 SHA1_Update(c, ptr, res);
136 # define SHA1_Update sha1_update
138 # if !defined(OPENSSL_NO_MULTIBLOCK)
141 unsigned int A[8], B[8], C[8], D[8], E[8];
144 const unsigned char *ptr;
148 void sha1_multi_block(SHA1_MB_CTX *, const HASH_DESC *, int);
151 const unsigned char *inp;
157 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
159 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA1 *key,
161 const unsigned char *inp,
162 size_t inp_len, int n4x)
163 { /* n4x is 1 or 2 */
164 HASH_DESC hash_d[8], edges[8];
166 unsigned char storage[sizeof(SHA1_MB_CTX) + 32];
173 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
181 /* ask for IVs in bulk */
182 if (RAND_bytes((IVs = blocks[0].c), 16 * x4) <= 0)
185 ctx = (SHA1_MB_CTX *) (storage + 32 - ((size_t)storage % 32)); /* align */
187 frag = (unsigned int)inp_len >> (1 + n4x);
188 last = (unsigned int)inp_len + frag - (frag << (1 + n4x));
189 if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) {
194 packlen = 5 + 16 + ((frag + 20 + 16) & -16);
196 /* populate descriptors with pointers and IVs */
199 /* 5+16 is place for header and explicit IV */
200 ciph_d[0].out = out + 5 + 16;
201 memcpy(ciph_d[0].out - 16, IVs, 16);
202 memcpy(ciph_d[0].iv, IVs, 16);
205 for (i = 1; i < x4; i++) {
206 ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag;
207 ciph_d[i].out = ciph_d[i - 1].out + packlen;
208 memcpy(ciph_d[i].out - 16, IVs, 16);
209 memcpy(ciph_d[i].iv, IVs, 16);
214 memcpy(blocks[0].c, key->md.data, 8);
215 seqnum = BSWAP8(blocks[0].q[0]);
217 for (i = 0; i < x4; i++) {
218 unsigned int len = (i == (x4 - 1) ? last : frag);
219 # if !defined(BSWAP8)
220 unsigned int carry, j;
223 ctx->A[i] = key->md.h0;
224 ctx->B[i] = key->md.h1;
225 ctx->C[i] = key->md.h2;
226 ctx->D[i] = key->md.h3;
227 ctx->E[i] = key->md.h4;
231 blocks[i].q[0] = BSWAP8(seqnum + i);
233 for (carry = i, j = 8; j--;) {
234 blocks[i].c[j] = ((u8 *)key->md.data)[j] + carry;
235 carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1);
238 blocks[i].c[8] = ((u8 *)key->md.data)[8];
239 blocks[i].c[9] = ((u8 *)key->md.data)[9];
240 blocks[i].c[10] = ((u8 *)key->md.data)[10];
242 blocks[i].c[11] = (u8)(len >> 8);
243 blocks[i].c[12] = (u8)(len);
245 memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13);
246 hash_d[i].ptr += 64 - 13;
247 hash_d[i].blocks = (len - (64 - 13)) / 64;
249 edges[i].ptr = blocks[i].c;
253 /* hash 13-byte headers and first 64-13 bytes of inputs */
254 sha1_multi_block(ctx, edges, n4x);
255 /* hash bulk inputs */
256 # define MAXCHUNKSIZE 2048
258 # error "MAXCHUNKSIZE is not divisible by 64"
261 * goal is to minimize pressure on L1 cache by moving in shorter steps,
262 * so that hashed data is still in the cache by the time we encrypt it
264 minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64;
265 if (minblocks > MAXCHUNKSIZE / 64) {
266 for (i = 0; i < x4; i++) {
267 edges[i].ptr = hash_d[i].ptr;
268 edges[i].blocks = MAXCHUNKSIZE / 64;
269 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
272 sha1_multi_block(ctx, edges, n4x);
273 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
275 for (i = 0; i < x4; i++) {
276 edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE;
277 hash_d[i].blocks -= MAXCHUNKSIZE / 64;
278 edges[i].blocks = MAXCHUNKSIZE / 64;
279 ciph_d[i].inp += MAXCHUNKSIZE;
280 ciph_d[i].out += MAXCHUNKSIZE;
281 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
282 memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16);
284 processed += MAXCHUNKSIZE;
285 minblocks -= MAXCHUNKSIZE / 64;
286 } while (minblocks > MAXCHUNKSIZE / 64);
290 sha1_multi_block(ctx, hash_d, n4x);
292 memset(blocks, 0, sizeof(blocks));
293 for (i = 0; i < x4; i++) {
294 unsigned int len = (i == (x4 - 1) ? last : frag),
295 off = hash_d[i].blocks * 64;
296 const unsigned char *ptr = hash_d[i].ptr + off;
298 off = (len - processed) - (64 - 13) - off; /* remainder actually */
299 memcpy(blocks[i].c, ptr, off);
300 blocks[i].c[off] = 0x80;
301 len += 64 + 13; /* 64 is HMAC header */
302 len *= 8; /* convert to bits */
303 if (off < (64 - 8)) {
305 blocks[i].d[15] = BSWAP4(len);
307 PUTU32(blocks[i].c + 60, len);
312 blocks[i].d[31] = BSWAP4(len);
314 PUTU32(blocks[i].c + 124, len);
318 edges[i].ptr = blocks[i].c;
321 /* hash input tails and finalize */
322 sha1_multi_block(ctx, edges, n4x);
324 memset(blocks, 0, sizeof(blocks));
325 for (i = 0; i < x4; i++) {
327 blocks[i].d[0] = BSWAP4(ctx->A[i]);
328 ctx->A[i] = key->tail.h0;
329 blocks[i].d[1] = BSWAP4(ctx->B[i]);
330 ctx->B[i] = key->tail.h1;
331 blocks[i].d[2] = BSWAP4(ctx->C[i]);
332 ctx->C[i] = key->tail.h2;
333 blocks[i].d[3] = BSWAP4(ctx->D[i]);
334 ctx->D[i] = key->tail.h3;
335 blocks[i].d[4] = BSWAP4(ctx->E[i]);
336 ctx->E[i] = key->tail.h4;
337 blocks[i].c[20] = 0x80;
338 blocks[i].d[15] = BSWAP4((64 + 20) * 8);
340 PUTU32(blocks[i].c + 0, ctx->A[i]);
341 ctx->A[i] = key->tail.h0;
342 PUTU32(blocks[i].c + 4, ctx->B[i]);
343 ctx->B[i] = key->tail.h1;
344 PUTU32(blocks[i].c + 8, ctx->C[i]);
345 ctx->C[i] = key->tail.h2;
346 PUTU32(blocks[i].c + 12, ctx->D[i]);
347 ctx->D[i] = key->tail.h3;
348 PUTU32(blocks[i].c + 16, ctx->E[i]);
349 ctx->E[i] = key->tail.h4;
350 blocks[i].c[20] = 0x80;
351 PUTU32(blocks[i].c + 60, (64 + 20) * 8);
353 edges[i].ptr = blocks[i].c;
358 sha1_multi_block(ctx, edges, n4x);
360 for (i = 0; i < x4; i++) {
361 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
362 unsigned char *out0 = out;
364 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
365 ciph_d[i].inp = ciph_d[i].out;
370 PUTU32(out + 0, ctx->A[i]);
371 PUTU32(out + 4, ctx->B[i]);
372 PUTU32(out + 8, ctx->C[i]);
373 PUTU32(out + 12, ctx->D[i]);
374 PUTU32(out + 16, ctx->E[i]);
380 for (j = 0; j <= pad; j++)
384 ciph_d[i].blocks = (len - processed) / 16;
385 len += 16; /* account for explicit iv */
388 out0[0] = ((u8 *)key->md.data)[8];
389 out0[1] = ((u8 *)key->md.data)[9];
390 out0[2] = ((u8 *)key->md.data)[10];
391 out0[3] = (u8)(len >> 8);
398 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
400 OPENSSL_cleanse(blocks, sizeof(blocks));
401 OPENSSL_cleanse(ctx, sizeof(*ctx));
407 static int aesni_cbc_hmac_sha1_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
408 const unsigned char *in, size_t len)
410 EVP_AES_HMAC_SHA1 *key = data(ctx);
412 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
415 # if defined(STITCHED_CALL)
416 size_t aes_off = 0, blocks;
418 sha_off = SHA_CBLOCK - key->md.num;
421 key->payload_length = NO_PAYLOAD_LENGTH;
423 if (len % AES_BLOCK_SIZE)
426 if (EVP_CIPHER_CTX_encrypting(ctx)) {
427 if (plen == NO_PAYLOAD_LENGTH)
430 ((plen + SHA_DIGEST_LENGTH +
431 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
433 else if (key->aux.tls_ver >= TLS1_1_VERSION)
436 # if defined(STITCHED_CALL)
437 if (plen > (sha_off + iv)
438 && (blocks = (plen - (sha_off + iv)) / SHA_CBLOCK)) {
439 SHA1_Update(&key->md, in + iv, sha_off);
441 aesni_cbc_sha1_enc(in, out, blocks, &key->ks,
442 EVP_CIPHER_CTX_iv_noconst(ctx),
443 &key->md, in + iv + sha_off);
444 blocks *= SHA_CBLOCK;
447 key->md.Nh += blocks >> 29;
448 key->md.Nl += blocks <<= 3;
449 if (key->md.Nl < (unsigned int)blocks)
456 SHA1_Update(&key->md, in + sha_off, plen - sha_off);
458 if (plen != len) { /* "TLS" mode of operation */
460 memcpy(out + aes_off, in + aes_off, plen - aes_off);
462 /* calculate HMAC and append it to payload */
463 SHA1_Final(out + plen, &key->md);
465 SHA1_Update(&key->md, out + plen, SHA_DIGEST_LENGTH);
466 SHA1_Final(out + plen, &key->md);
468 /* pad the payload|hmac */
469 plen += SHA_DIGEST_LENGTH;
470 for (l = len - plen - 1; plen < len; plen++)
472 /* encrypt HMAC|padding at once */
473 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
474 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
476 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
477 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
481 unsigned int u[SHA_DIGEST_LENGTH / sizeof(unsigned int)];
482 unsigned char c[32 + SHA_DIGEST_LENGTH];
485 /* arrange cache line alignment */
486 pmac = (void *)(((size_t)mac.c + 31) & ((size_t)0 - 32));
488 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
489 size_t inp_len, mask, j, i;
490 unsigned int res, maxpad, pad, bitlen;
493 unsigned int u[SHA_LBLOCK];
494 unsigned char c[SHA_CBLOCK];
495 } *data = (void *)key->md.data;
496 # if defined(STITCHED_DECRYPT_CALL)
497 unsigned char tail_iv[AES_BLOCK_SIZE];
501 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
503 if (len < (AES_BLOCK_SIZE + SHA_DIGEST_LENGTH + 1))
506 /* omit explicit iv */
507 memcpy(EVP_CIPHER_CTX_iv_noconst(ctx), in, AES_BLOCK_SIZE);
509 in += AES_BLOCK_SIZE;
510 out += AES_BLOCK_SIZE;
511 len -= AES_BLOCK_SIZE;
512 } else if (len < (SHA_DIGEST_LENGTH + 1))
515 # if defined(STITCHED_DECRYPT_CALL)
516 if (len >= 1024 && ctx->key_len == 32) {
517 /* decrypt last block */
518 memcpy(tail_iv, in + len - 2 * AES_BLOCK_SIZE,
520 aesni_cbc_encrypt(in + len - AES_BLOCK_SIZE,
521 out + len - AES_BLOCK_SIZE, AES_BLOCK_SIZE,
522 &key->ks, tail_iv, 0);
526 /* decrypt HMAC|padding at once */
527 aesni_cbc_encrypt(in, out, len, &key->ks,
528 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
530 /* figure out payload length */
532 maxpad = len - (SHA_DIGEST_LENGTH + 1);
533 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
536 mask = constant_time_ge(maxpad, pad);
539 * If pad is invalid then we will fail the above test but we must
540 * continue anyway because we are in constant time code. However,
541 * we'll use the maxpad value instead of the supplied pad to make
542 * sure we perform well defined pointer arithmetic.
544 pad = constant_time_select(mask, pad, maxpad);
546 inp_len = len - (SHA_DIGEST_LENGTH + pad + 1);
548 key->aux.tls_aad[plen - 2] = inp_len >> 8;
549 key->aux.tls_aad[plen - 1] = inp_len;
553 SHA1_Update(&key->md, key->aux.tls_aad, plen);
555 # if defined(STITCHED_DECRYPT_CALL)
557 blocks = (len - (256 + 32 + SHA_CBLOCK)) / SHA_CBLOCK;
558 aes_off = len - AES_BLOCK_SIZE - blocks * SHA_CBLOCK;
559 sha_off = SHA_CBLOCK - plen;
561 aesni_cbc_encrypt(in, out, aes_off, &key->ks, ctx->iv, 0);
563 SHA1_Update(&key->md, out, sha_off);
564 aesni256_cbc_sha1_dec(in + aes_off,
565 out + aes_off, blocks, &key->ks,
566 ctx->iv, &key->md, out + sha_off);
568 sha_off += blocks *= SHA_CBLOCK;
573 key->md.Nl += (blocks << 3); /* at most 18 bits */
574 memcpy(ctx->iv, tail_iv, AES_BLOCK_SIZE);
578 # if 1 /* see original reference version in #else */
579 len -= SHA_DIGEST_LENGTH; /* amend mac */
580 if (len >= (256 + SHA_CBLOCK)) {
581 j = (len - (256 + SHA_CBLOCK)) & (0 - SHA_CBLOCK);
582 j += SHA_CBLOCK - key->md.num;
583 SHA1_Update(&key->md, out, j);
589 /* but pretend as if we hashed padded payload */
590 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
592 bitlen = BSWAP4(bitlen);
595 mac.c[1] = (unsigned char)(bitlen >> 16);
596 mac.c[2] = (unsigned char)(bitlen >> 8);
597 mac.c[3] = (unsigned char)bitlen;
607 for (res = key->md.num, j = 0; j < len; j++) {
609 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
611 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
612 data->c[res++] = (unsigned char)c;
614 if (res != SHA_CBLOCK)
617 /* j is not incremented yet */
618 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
619 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
620 sha1_block_data_order(&key->md, data, 1);
621 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
622 pmac->u[0] |= key->md.h0 & mask;
623 pmac->u[1] |= key->md.h1 & mask;
624 pmac->u[2] |= key->md.h2 & mask;
625 pmac->u[3] |= key->md.h3 & mask;
626 pmac->u[4] |= key->md.h4 & mask;
630 for (i = res; i < SHA_CBLOCK; i++, j++)
633 if (res > SHA_CBLOCK - 8) {
634 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
635 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
636 sha1_block_data_order(&key->md, data, 1);
637 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
638 pmac->u[0] |= key->md.h0 & mask;
639 pmac->u[1] |= key->md.h1 & mask;
640 pmac->u[2] |= key->md.h2 & mask;
641 pmac->u[3] |= key->md.h3 & mask;
642 pmac->u[4] |= key->md.h4 & mask;
644 memset(data, 0, SHA_CBLOCK);
647 data->u[SHA_LBLOCK - 1] = bitlen;
648 sha1_block_data_order(&key->md, data, 1);
649 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
650 pmac->u[0] |= key->md.h0 & mask;
651 pmac->u[1] |= key->md.h1 & mask;
652 pmac->u[2] |= key->md.h2 & mask;
653 pmac->u[3] |= key->md.h3 & mask;
654 pmac->u[4] |= key->md.h4 & mask;
657 pmac->u[0] = BSWAP4(pmac->u[0]);
658 pmac->u[1] = BSWAP4(pmac->u[1]);
659 pmac->u[2] = BSWAP4(pmac->u[2]);
660 pmac->u[3] = BSWAP4(pmac->u[3]);
661 pmac->u[4] = BSWAP4(pmac->u[4]);
663 for (i = 0; i < 5; i++) {
665 pmac->c[4 * i + 0] = (unsigned char)(res >> 24);
666 pmac->c[4 * i + 1] = (unsigned char)(res >> 16);
667 pmac->c[4 * i + 2] = (unsigned char)(res >> 8);
668 pmac->c[4 * i + 3] = (unsigned char)res;
671 len += SHA_DIGEST_LENGTH;
672 # else /* pre-lucky-13 reference version of above */
673 SHA1_Update(&key->md, out, inp_len);
675 SHA1_Final(pmac->c, &key->md);
678 unsigned int inp_blocks, pad_blocks;
680 /* but pretend as if we hashed padded payload */
682 1 + ((SHA_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
683 res += (unsigned int)(len - inp_len);
684 pad_blocks = res / SHA_CBLOCK;
687 1 + ((SHA_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
688 for (; inp_blocks < pad_blocks; inp_blocks++)
689 sha1_block_data_order(&key->md, data, 1);
693 SHA1_Update(&key->md, pmac->c, SHA_DIGEST_LENGTH);
694 SHA1_Final(pmac->c, &key->md);
699 # if 1 /* see original reference version in #else */
701 unsigned char *p = out + len - 1 - maxpad - SHA_DIGEST_LENGTH;
702 size_t off = out - p;
703 unsigned int c, cmask;
705 maxpad += SHA_DIGEST_LENGTH;
706 for (res = 0, i = 0, j = 0; j < maxpad; j++) {
709 ((int)(j - off - SHA_DIGEST_LENGTH)) >> (sizeof(int) *
711 res |= (c ^ pad) & ~cmask; /* ... and padding */
712 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
713 res |= (c ^ pmac->c[i]) & cmask;
716 maxpad -= SHA_DIGEST_LENGTH;
718 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
721 # else /* pre-lucky-13 reference version of above */
722 for (res = 0, i = 0; i < SHA_DIGEST_LENGTH; i++)
723 res |= out[i] ^ pmac->c[i];
724 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
728 pad = (pad & ~res) | (maxpad & res);
729 out = out + len - 1 - pad;
730 for (res = 0, i = 0; i < pad; i++)
733 res = (0 - res) >> (sizeof(res) * 8 - 1);
738 # if defined(STITCHED_DECRYPT_CALL)
739 if (len >= 1024 && ctx->key_len == 32) {
740 if (sha_off %= SHA_CBLOCK)
741 blocks = (len - 3 * SHA_CBLOCK) / SHA_CBLOCK;
743 blocks = (len - 2 * SHA_CBLOCK) / SHA_CBLOCK;
744 aes_off = len - blocks * SHA_CBLOCK;
746 aesni_cbc_encrypt(in, out, aes_off, &key->ks, ctx->iv, 0);
747 SHA1_Update(&key->md, out, sha_off);
748 aesni256_cbc_sha1_dec(in + aes_off,
749 out + aes_off, blocks, &key->ks,
750 ctx->iv, &key->md, out + sha_off);
752 sha_off += blocks *= SHA_CBLOCK;
756 key->md.Nh += blocks >> 29;
757 key->md.Nl += blocks <<= 3;
758 if (key->md.Nl < (unsigned int)blocks)
762 /* decrypt HMAC|padding at once */
763 aesni_cbc_encrypt(in, out, len, &key->ks,
764 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
766 SHA1_Update(&key->md, out, len);
773 static int aesni_cbc_hmac_sha1_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
776 EVP_AES_HMAC_SHA1 *key = data(ctx);
779 case EVP_CTRL_AEAD_SET_MAC_KEY:
782 unsigned char hmac_key[64];
784 memset(hmac_key, 0, sizeof(hmac_key));
786 if (arg > (int)sizeof(hmac_key)) {
787 SHA1_Init(&key->head);
788 SHA1_Update(&key->head, ptr, arg);
789 SHA1_Final(hmac_key, &key->head);
791 memcpy(hmac_key, ptr, arg);
794 for (i = 0; i < sizeof(hmac_key); i++)
795 hmac_key[i] ^= 0x36; /* ipad */
796 SHA1_Init(&key->head);
797 SHA1_Update(&key->head, hmac_key, sizeof(hmac_key));
799 for (i = 0; i < sizeof(hmac_key); i++)
800 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
801 SHA1_Init(&key->tail);
802 SHA1_Update(&key->tail, hmac_key, sizeof(hmac_key));
804 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
808 case EVP_CTRL_AEAD_TLS1_AAD:
810 unsigned char *p = ptr;
813 if (arg != EVP_AEAD_TLS1_AAD_LEN)
816 len = p[arg - 2] << 8 | p[arg - 1];
818 if (EVP_CIPHER_CTX_encrypting(ctx)) {
819 key->payload_length = len;
820 if ((key->aux.tls_ver =
821 p[arg - 4] << 8 | p[arg - 3]) >= TLS1_1_VERSION) {
822 if (len < AES_BLOCK_SIZE)
824 len -= AES_BLOCK_SIZE;
825 p[arg - 2] = len >> 8;
829 SHA1_Update(&key->md, p, arg);
831 return (int)(((len + SHA_DIGEST_LENGTH +
832 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
835 memcpy(key->aux.tls_aad, ptr, arg);
836 key->payload_length = arg;
838 return SHA_DIGEST_LENGTH;
841 # if !defined(OPENSSL_NO_MULTIBLOCK)
842 case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
843 return (int)(5 + 16 + ((arg + 20 + 16) & -16));
844 case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
846 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
847 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
848 unsigned int n4x = 1, x4;
849 unsigned int frag, last, packlen, inp_len;
851 if (arg < (int)sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
854 inp_len = param->inp[11] << 8 | param->inp[12];
856 if (EVP_CIPHER_CTX_encrypting(ctx)) {
857 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
862 return 0; /* too short */
864 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
866 } else if ((n4x = param->interleave / 4) && n4x <= 2)
867 inp_len = param->len;
872 SHA1_Update(&key->md, param->inp, 13);
877 frag = inp_len >> n4x;
878 last = inp_len + frag - (frag << n4x);
879 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
884 packlen = 5 + 16 + ((frag + 20 + 16) & -16);
885 packlen = (packlen << n4x) - packlen;
886 packlen += 5 + 16 + ((last + 20 + 16) & -16);
888 param->interleave = x4;
892 return -1; /* not yet */
894 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
896 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
897 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
899 return (int)tls1_1_multi_block_encrypt(key, param->out,
900 param->inp, param->len,
901 param->interleave / 4);
903 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
910 static EVP_CIPHER aesni_128_cbc_hmac_sha1_cipher = {
911 # ifdef NID_aes_128_cbc_hmac_sha1
912 NID_aes_128_cbc_hmac_sha1,
916 AES_BLOCK_SIZE, 16, AES_BLOCK_SIZE,
917 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
918 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
919 aesni_cbc_hmac_sha1_init_key,
920 aesni_cbc_hmac_sha1_cipher,
922 sizeof(EVP_AES_HMAC_SHA1),
923 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
924 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
925 aesni_cbc_hmac_sha1_ctrl,
929 static EVP_CIPHER aesni_256_cbc_hmac_sha1_cipher = {
930 # ifdef NID_aes_256_cbc_hmac_sha1
931 NID_aes_256_cbc_hmac_sha1,
935 AES_BLOCK_SIZE, 32, AES_BLOCK_SIZE,
936 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
937 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
938 aesni_cbc_hmac_sha1_init_key,
939 aesni_cbc_hmac_sha1_cipher,
941 sizeof(EVP_AES_HMAC_SHA1),
942 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
943 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
944 aesni_cbc_hmac_sha1_ctrl,
948 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha1(void)
950 return (OPENSSL_ia32cap_P[1] & AESNI_CAPABLE ?
951 &aesni_128_cbc_hmac_sha1_cipher : NULL);
954 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha1(void)
956 return (OPENSSL_ia32cap_P[1] & AESNI_CAPABLE ?
957 &aesni_256_cbc_hmac_sha1_cipher : NULL);
960 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha1(void)
965 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha1(void)