2 * Copyright 2013-2016 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>
16 #include <openssl/evp.h>
17 #include <openssl/objects.h>
18 #include <openssl/aes.h>
19 #include <openssl/sha.h>
20 #include <openssl/rand.h>
21 #include "modes_lcl.h"
22 #include "internal/constant_time_locl.h"
23 #include "internal/evp_int.h"
25 #ifndef EVP_CIPH_FLAG_AEAD_CIPHER
26 # define EVP_CIPH_FLAG_AEAD_CIPHER 0x200000
27 # define EVP_CTRL_AEAD_TLS1_AAD 0x16
28 # define EVP_CTRL_AEAD_SET_MAC_KEY 0x17
31 #if !defined(EVP_CIPH_FLAG_DEFAULT_ASN1)
32 # define EVP_CIPH_FLAG_DEFAULT_ASN1 0
35 #if !defined(EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK)
36 # define EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK 0
39 #define TLS1_1_VERSION 0x0302
43 SHA256_CTX head, tail, md;
44 size_t payload_length; /* AAD length in decrypt case */
47 unsigned char tls_aad[16]; /* 13 used */
49 } EVP_AES_HMAC_SHA256;
51 # define NO_PAYLOAD_LENGTH ((size_t)-1)
53 #if defined(AES_ASM) && ( \
54 defined(__x86_64) || defined(__x86_64__) || \
55 defined(_M_AMD64) || defined(_M_X64) )
57 extern unsigned int OPENSSL_ia32cap_P[];
58 # define AESNI_CAPABLE (1<<(57-32))
60 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
62 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
65 void aesni_cbc_encrypt(const unsigned char *in,
68 const AES_KEY *key, unsigned char *ivec, int enc);
70 int aesni_cbc_sha256_enc(const void *inp, void *out, size_t blocks,
71 const AES_KEY *key, unsigned char iv[16],
72 SHA256_CTX *ctx, const void *in0);
74 # define data(ctx) ((EVP_AES_HMAC_SHA256 *)EVP_CIPHER_CTX_get_cipher_data(ctx))
76 static int aesni_cbc_hmac_sha256_init_key(EVP_CIPHER_CTX *ctx,
77 const unsigned char *inkey,
78 const unsigned char *iv, int enc)
80 EVP_AES_HMAC_SHA256 *key = data(ctx);
84 memset(&key->ks, 0, sizeof(key->ks.rd_key)),
85 ret = aesni_set_encrypt_key(inkey,
86 EVP_CIPHER_CTX_key_length(ctx) * 8,
89 ret = aesni_set_decrypt_key(inkey,
90 EVP_CIPHER_CTX_key_length(ctx) * 8,
93 SHA256_Init(&key->head); /* handy when benchmarking */
94 key->tail = key->head;
97 key->payload_length = NO_PAYLOAD_LENGTH;
99 return ret < 0 ? 0 : 1;
102 # define STITCHED_CALL
104 # if !defined(STITCHED_CALL)
108 void sha256_block_data_order(void *c, const void *p, size_t len);
110 static void sha256_update(SHA256_CTX *c, const void *data, size_t len)
112 const unsigned char *ptr = data;
115 if ((res = c->num)) {
116 res = SHA256_CBLOCK - res;
119 SHA256_Update(c, ptr, res);
124 res = len % SHA256_CBLOCK;
128 sha256_block_data_order(c, ptr, len / SHA256_CBLOCK);
133 if (c->Nl < (unsigned int)len)
138 SHA256_Update(c, ptr, res);
141 # ifdef SHA256_Update
142 # undef SHA256_Update
144 # define SHA256_Update sha256_update
146 # if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
149 unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8];
152 const unsigned char *ptr;
156 void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int);
159 const unsigned char *inp;
165 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
167 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
169 const unsigned char *inp,
170 size_t inp_len, int n4x)
171 { /* n4x is 1 or 2 */
172 HASH_DESC hash_d[8], edges[8];
174 unsigned char storage[sizeof(SHA256_MB_CTX) + 32];
181 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
189 /* ask for IVs in bulk */
190 if (RAND_bytes((IVs = blocks[0].c), 16 * x4) <= 0)
194 ctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32));
196 frag = (unsigned int)inp_len >> (1 + n4x);
197 last = (unsigned int)inp_len + frag - (frag << (1 + n4x));
198 if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) {
203 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
205 /* populate descriptors with pointers and IVs */
208 /* 5+16 is place for header and explicit IV */
209 ciph_d[0].out = out + 5 + 16;
210 memcpy(ciph_d[0].out - 16, IVs, 16);
211 memcpy(ciph_d[0].iv, IVs, 16);
214 for (i = 1; i < x4; i++) {
215 ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag;
216 ciph_d[i].out = ciph_d[i - 1].out + packlen;
217 memcpy(ciph_d[i].out - 16, IVs, 16);
218 memcpy(ciph_d[i].iv, IVs, 16);
223 memcpy(blocks[0].c, key->md.data, 8);
224 seqnum = BSWAP8(blocks[0].q[0]);
226 for (i = 0; i < x4; i++) {
227 unsigned int len = (i == (x4 - 1) ? last : frag);
228 # if !defined(BSWAP8)
229 unsigned int carry, j;
232 ctx->A[i] = key->md.h[0];
233 ctx->B[i] = key->md.h[1];
234 ctx->C[i] = key->md.h[2];
235 ctx->D[i] = key->md.h[3];
236 ctx->E[i] = key->md.h[4];
237 ctx->F[i] = key->md.h[5];
238 ctx->G[i] = key->md.h[6];
239 ctx->H[i] = key->md.h[7];
243 blocks[i].q[0] = BSWAP8(seqnum + i);
245 for (carry = i, j = 8; j--;) {
246 blocks[i].c[j] = ((u8 *)key->md.data)[j] + carry;
247 carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1);
250 blocks[i].c[8] = ((u8 *)key->md.data)[8];
251 blocks[i].c[9] = ((u8 *)key->md.data)[9];
252 blocks[i].c[10] = ((u8 *)key->md.data)[10];
254 blocks[i].c[11] = (u8)(len >> 8);
255 blocks[i].c[12] = (u8)(len);
257 memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13);
258 hash_d[i].ptr += 64 - 13;
259 hash_d[i].blocks = (len - (64 - 13)) / 64;
261 edges[i].ptr = blocks[i].c;
265 /* hash 13-byte headers and first 64-13 bytes of inputs */
266 sha256_multi_block(ctx, edges, n4x);
267 /* hash bulk inputs */
268 # define MAXCHUNKSIZE 2048
270 # error "MAXCHUNKSIZE is not divisible by 64"
273 * goal is to minimize pressure on L1 cache by moving in shorter steps,
274 * so that hashed data is still in the cache by the time we encrypt it
276 minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64;
277 if (minblocks > MAXCHUNKSIZE / 64) {
278 for (i = 0; i < x4; i++) {
279 edges[i].ptr = hash_d[i].ptr;
280 edges[i].blocks = MAXCHUNKSIZE / 64;
281 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
284 sha256_multi_block(ctx, edges, n4x);
285 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
287 for (i = 0; i < x4; i++) {
288 edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE;
289 hash_d[i].blocks -= MAXCHUNKSIZE / 64;
290 edges[i].blocks = MAXCHUNKSIZE / 64;
291 ciph_d[i].inp += MAXCHUNKSIZE;
292 ciph_d[i].out += MAXCHUNKSIZE;
293 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
294 memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16);
296 processed += MAXCHUNKSIZE;
297 minblocks -= MAXCHUNKSIZE / 64;
298 } while (minblocks > MAXCHUNKSIZE / 64);
302 sha256_multi_block(ctx, hash_d, n4x);
304 memset(blocks, 0, sizeof(blocks));
305 for (i = 0; i < x4; i++) {
306 unsigned int len = (i == (x4 - 1) ? last : frag),
307 off = hash_d[i].blocks * 64;
308 const unsigned char *ptr = hash_d[i].ptr + off;
310 off = (len - processed) - (64 - 13) - off; /* remainder actually */
311 memcpy(blocks[i].c, ptr, off);
312 blocks[i].c[off] = 0x80;
313 len += 64 + 13; /* 64 is HMAC header */
314 len *= 8; /* convert to bits */
315 if (off < (64 - 8)) {
317 blocks[i].d[15] = BSWAP4(len);
319 PUTU32(blocks[i].c + 60, len);
324 blocks[i].d[31] = BSWAP4(len);
326 PUTU32(blocks[i].c + 124, len);
330 edges[i].ptr = blocks[i].c;
333 /* hash input tails and finalize */
334 sha256_multi_block(ctx, edges, n4x);
336 memset(blocks, 0, sizeof(blocks));
337 for (i = 0; i < x4; i++) {
339 blocks[i].d[0] = BSWAP4(ctx->A[i]);
340 ctx->A[i] = key->tail.h[0];
341 blocks[i].d[1] = BSWAP4(ctx->B[i]);
342 ctx->B[i] = key->tail.h[1];
343 blocks[i].d[2] = BSWAP4(ctx->C[i]);
344 ctx->C[i] = key->tail.h[2];
345 blocks[i].d[3] = BSWAP4(ctx->D[i]);
346 ctx->D[i] = key->tail.h[3];
347 blocks[i].d[4] = BSWAP4(ctx->E[i]);
348 ctx->E[i] = key->tail.h[4];
349 blocks[i].d[5] = BSWAP4(ctx->F[i]);
350 ctx->F[i] = key->tail.h[5];
351 blocks[i].d[6] = BSWAP4(ctx->G[i]);
352 ctx->G[i] = key->tail.h[6];
353 blocks[i].d[7] = BSWAP4(ctx->H[i]);
354 ctx->H[i] = key->tail.h[7];
355 blocks[i].c[32] = 0x80;
356 blocks[i].d[15] = BSWAP4((64 + 32) * 8);
358 PUTU32(blocks[i].c + 0, ctx->A[i]);
359 ctx->A[i] = key->tail.h[0];
360 PUTU32(blocks[i].c + 4, ctx->B[i]);
361 ctx->B[i] = key->tail.h[1];
362 PUTU32(blocks[i].c + 8, ctx->C[i]);
363 ctx->C[i] = key->tail.h[2];
364 PUTU32(blocks[i].c + 12, ctx->D[i]);
365 ctx->D[i] = key->tail.h[3];
366 PUTU32(blocks[i].c + 16, ctx->E[i]);
367 ctx->E[i] = key->tail.h[4];
368 PUTU32(blocks[i].c + 20, ctx->F[i]);
369 ctx->F[i] = key->tail.h[5];
370 PUTU32(blocks[i].c + 24, ctx->G[i]);
371 ctx->G[i] = key->tail.h[6];
372 PUTU32(blocks[i].c + 28, ctx->H[i]);
373 ctx->H[i] = key->tail.h[7];
374 blocks[i].c[32] = 0x80;
375 PUTU32(blocks[i].c + 60, (64 + 32) * 8);
377 edges[i].ptr = blocks[i].c;
382 sha256_multi_block(ctx, edges, n4x);
384 for (i = 0; i < x4; i++) {
385 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
386 unsigned char *out0 = out;
388 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
389 ciph_d[i].inp = ciph_d[i].out;
394 PUTU32(out + 0, ctx->A[i]);
395 PUTU32(out + 4, ctx->B[i]);
396 PUTU32(out + 8, ctx->C[i]);
397 PUTU32(out + 12, ctx->D[i]);
398 PUTU32(out + 16, ctx->E[i]);
399 PUTU32(out + 20, ctx->F[i]);
400 PUTU32(out + 24, ctx->G[i]);
401 PUTU32(out + 28, ctx->H[i]);
407 for (j = 0; j <= pad; j++)
411 ciph_d[i].blocks = (len - processed) / 16;
412 len += 16; /* account for explicit iv */
415 out0[0] = ((u8 *)key->md.data)[8];
416 out0[1] = ((u8 *)key->md.data)[9];
417 out0[2] = ((u8 *)key->md.data)[10];
418 out0[3] = (u8)(len >> 8);
425 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
427 OPENSSL_cleanse(blocks, sizeof(blocks));
428 OPENSSL_cleanse(ctx, sizeof(*ctx));
434 static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx,
436 const unsigned char *in, size_t len)
438 EVP_AES_HMAC_SHA256 *key = data(ctx);
440 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
443 # if defined(STITCHED_CALL)
444 size_t aes_off = 0, blocks;
446 sha_off = SHA256_CBLOCK - key->md.num;
449 key->payload_length = NO_PAYLOAD_LENGTH;
451 if (len % AES_BLOCK_SIZE)
454 if (EVP_CIPHER_CTX_encrypting(ctx)) {
455 if (plen == NO_PAYLOAD_LENGTH)
458 ((plen + SHA256_DIGEST_LENGTH +
459 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
461 else if (key->aux.tls_ver >= TLS1_1_VERSION)
464 # if defined(STITCHED_CALL)
466 * Assembly stitch handles AVX-capable processors, but its
467 * performance is not optimal on AMD Jaguar, ~40% worse, for
468 * unknown reasons. Incidentally processor in question supports
469 * AVX, but not AMD-specific XOP extension, which can be used
470 * to identify it and avoid stitch invocation. So that after we
471 * establish that current CPU supports AVX, we even see if it's
472 * either even XOP-capable Bulldozer-based or GenuineIntel one.
474 if (OPENSSL_ia32cap_P[1] & (1 << (60 - 32)) && /* AVX? */
475 ((OPENSSL_ia32cap_P[1] & (1 << (43 - 32))) /* XOP? */
476 | (OPENSSL_ia32cap_P[0] & (1<<30))) && /* "Intel CPU"? */
477 plen > (sha_off + iv) &&
478 (blocks = (plen - (sha_off + iv)) / SHA256_CBLOCK)) {
479 SHA256_Update(&key->md, in + iv, sha_off);
481 (void)aesni_cbc_sha256_enc(in, out, blocks, &key->ks,
482 EVP_CIPHER_CTX_iv_noconst(ctx),
483 &key->md, in + iv + sha_off);
484 blocks *= SHA256_CBLOCK;
487 key->md.Nh += blocks >> 29;
488 key->md.Nl += blocks <<= 3;
489 if (key->md.Nl < (unsigned int)blocks)
496 SHA256_Update(&key->md, in + sha_off, plen - sha_off);
498 if (plen != len) { /* "TLS" mode of operation */
500 memcpy(out + aes_off, in + aes_off, plen - aes_off);
502 /* calculate HMAC and append it to payload */
503 SHA256_Final(out + plen, &key->md);
505 SHA256_Update(&key->md, out + plen, SHA256_DIGEST_LENGTH);
506 SHA256_Final(out + plen, &key->md);
508 /* pad the payload|hmac */
509 plen += SHA256_DIGEST_LENGTH;
510 for (l = len - plen - 1; plen < len; plen++)
512 /* encrypt HMAC|padding at once */
513 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
514 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
516 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
517 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
521 unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)];
522 unsigned char c[64 + SHA256_DIGEST_LENGTH];
525 /* arrange cache line alignment */
526 pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64));
528 /* decrypt HMAC|padding at once */
529 aesni_cbc_encrypt(in, out, len, &key->ks,
530 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
532 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
533 size_t inp_len, mask, j, i;
534 unsigned int res, maxpad, pad, bitlen;
537 unsigned int u[SHA_LBLOCK];
538 unsigned char c[SHA256_CBLOCK];
539 } *data = (void *)key->md.data;
541 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
545 if (len < (iv + SHA256_DIGEST_LENGTH + 1))
548 /* omit explicit iv */
552 /* figure out payload length */
554 maxpad = len - (SHA256_DIGEST_LENGTH + 1);
555 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
558 ret &= constant_time_ge(maxpad, pad);
560 inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1);
561 mask = (0 - ((inp_len - len) >> (sizeof(inp_len) * 8 - 1)));
565 key->aux.tls_aad[plen - 2] = inp_len >> 8;
566 key->aux.tls_aad[plen - 1] = inp_len;
570 SHA256_Update(&key->md, key->aux.tls_aad, plen);
573 len -= SHA256_DIGEST_LENGTH; /* amend mac */
574 if (len >= (256 + SHA256_CBLOCK)) {
575 j = (len - (256 + SHA256_CBLOCK)) & (0 - SHA256_CBLOCK);
576 j += SHA256_CBLOCK - key->md.num;
577 SHA256_Update(&key->md, out, j);
583 /* but pretend as if we hashed padded payload */
584 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
586 bitlen = BSWAP4(bitlen);
589 mac.c[1] = (unsigned char)(bitlen >> 16);
590 mac.c[2] = (unsigned char)(bitlen >> 8);
591 mac.c[3] = (unsigned char)bitlen;
604 for (res = key->md.num, j = 0; j < len; j++) {
606 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
608 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
609 data->c[res++] = (unsigned char)c;
611 if (res != SHA256_CBLOCK)
614 /* j is not incremented yet */
615 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
616 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
617 sha256_block_data_order(&key->md, data, 1);
618 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
619 pmac->u[0] |= key->md.h[0] & mask;
620 pmac->u[1] |= key->md.h[1] & mask;
621 pmac->u[2] |= key->md.h[2] & mask;
622 pmac->u[3] |= key->md.h[3] & mask;
623 pmac->u[4] |= key->md.h[4] & mask;
624 pmac->u[5] |= key->md.h[5] & mask;
625 pmac->u[6] |= key->md.h[6] & mask;
626 pmac->u[7] |= key->md.h[7] & mask;
630 for (i = res; i < SHA256_CBLOCK; i++, j++)
633 if (res > SHA256_CBLOCK - 8) {
634 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
635 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
636 sha256_block_data_order(&key->md, data, 1);
637 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
638 pmac->u[0] |= key->md.h[0] & mask;
639 pmac->u[1] |= key->md.h[1] & mask;
640 pmac->u[2] |= key->md.h[2] & mask;
641 pmac->u[3] |= key->md.h[3] & mask;
642 pmac->u[4] |= key->md.h[4] & mask;
643 pmac->u[5] |= key->md.h[5] & mask;
644 pmac->u[6] |= key->md.h[6] & mask;
645 pmac->u[7] |= key->md.h[7] & mask;
647 memset(data, 0, SHA256_CBLOCK);
650 data->u[SHA_LBLOCK - 1] = bitlen;
651 sha256_block_data_order(&key->md, data, 1);
652 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
653 pmac->u[0] |= key->md.h[0] & mask;
654 pmac->u[1] |= key->md.h[1] & mask;
655 pmac->u[2] |= key->md.h[2] & mask;
656 pmac->u[3] |= key->md.h[3] & mask;
657 pmac->u[4] |= key->md.h[4] & mask;
658 pmac->u[5] |= key->md.h[5] & mask;
659 pmac->u[6] |= key->md.h[6] & mask;
660 pmac->u[7] |= key->md.h[7] & mask;
663 pmac->u[0] = BSWAP4(pmac->u[0]);
664 pmac->u[1] = BSWAP4(pmac->u[1]);
665 pmac->u[2] = BSWAP4(pmac->u[2]);
666 pmac->u[3] = BSWAP4(pmac->u[3]);
667 pmac->u[4] = BSWAP4(pmac->u[4]);
668 pmac->u[5] = BSWAP4(pmac->u[5]);
669 pmac->u[6] = BSWAP4(pmac->u[6]);
670 pmac->u[7] = BSWAP4(pmac->u[7]);
672 for (i = 0; i < 8; i++) {
674 pmac->c[4 * i + 0] = (unsigned char)(res >> 24);
675 pmac->c[4 * i + 1] = (unsigned char)(res >> 16);
676 pmac->c[4 * i + 2] = (unsigned char)(res >> 8);
677 pmac->c[4 * i + 3] = (unsigned char)res;
680 len += SHA256_DIGEST_LENGTH;
682 SHA256_Update(&key->md, out, inp_len);
684 SHA256_Final(pmac->c, &key->md);
687 unsigned int inp_blocks, pad_blocks;
689 /* but pretend as if we hashed padded payload */
691 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
692 res += (unsigned int)(len - inp_len);
693 pad_blocks = res / SHA256_CBLOCK;
694 res %= SHA256_CBLOCK;
696 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
697 for (; inp_blocks < pad_blocks; inp_blocks++)
698 sha1_block_data_order(&key->md, data, 1);
702 SHA256_Update(&key->md, pmac->c, SHA256_DIGEST_LENGTH);
703 SHA256_Final(pmac->c, &key->md);
711 out + len - 1 - maxpad - SHA256_DIGEST_LENGTH;
712 size_t off = out - p;
713 unsigned int c, cmask;
715 maxpad += SHA256_DIGEST_LENGTH;
716 for (res = 0, i = 0, j = 0; j < maxpad; j++) {
719 ((int)(j - off - SHA256_DIGEST_LENGTH)) >>
720 (sizeof(int) * 8 - 1);
721 res |= (c ^ pad) & ~cmask; /* ... and padding */
722 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
723 res |= (c ^ pmac->c[i]) & cmask;
726 maxpad -= SHA256_DIGEST_LENGTH;
728 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
732 for (res = 0, i = 0; i < SHA256_DIGEST_LENGTH; i++)
733 res |= out[i] ^ pmac->c[i];
734 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
738 pad = (pad & ~res) | (maxpad & res);
739 out = out + len - 1 - pad;
740 for (res = 0, i = 0; i < pad; i++)
743 res = (0 - res) >> (sizeof(res) * 8 - 1);
748 SHA256_Update(&key->md, out, len);
755 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
758 EVP_AES_HMAC_SHA256 *key = data(ctx);
759 unsigned int u_arg = (unsigned int)arg;
762 case EVP_CTRL_AEAD_SET_MAC_KEY:
765 unsigned char hmac_key[64];
767 memset(hmac_key, 0, sizeof(hmac_key));
772 if (u_arg > sizeof(hmac_key)) {
773 SHA256_Init(&key->head);
774 SHA256_Update(&key->head, ptr, arg);
775 SHA256_Final(hmac_key, &key->head);
777 memcpy(hmac_key, ptr, arg);
780 for (i = 0; i < sizeof(hmac_key); i++)
781 hmac_key[i] ^= 0x36; /* ipad */
782 SHA256_Init(&key->head);
783 SHA256_Update(&key->head, hmac_key, sizeof(hmac_key));
785 for (i = 0; i < sizeof(hmac_key); i++)
786 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
787 SHA256_Init(&key->tail);
788 SHA256_Update(&key->tail, hmac_key, sizeof(hmac_key));
790 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
794 case EVP_CTRL_AEAD_TLS1_AAD:
796 unsigned char *p = ptr;
797 unsigned int len = p[arg - 2] << 8 | p[arg - 1];
799 if (arg != EVP_AEAD_TLS1_AAD_LEN)
802 if (EVP_CIPHER_CTX_encrypting(ctx)) {
803 key->payload_length = len;
804 if ((key->aux.tls_ver =
805 p[arg - 4] << 8 | p[arg - 3]) >= TLS1_1_VERSION) {
806 len -= AES_BLOCK_SIZE;
807 p[arg - 2] = len >> 8;
811 SHA256_Update(&key->md, p, arg);
813 return (int)(((len + SHA256_DIGEST_LENGTH +
814 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
817 memcpy(key->aux.tls_aad, ptr, arg);
818 key->payload_length = arg;
820 return SHA256_DIGEST_LENGTH;
823 # if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
824 case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
825 return (int)(5 + 16 + ((arg + 32 + 16) & -16));
826 case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
828 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
829 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
830 unsigned int n4x = 1, x4;
831 unsigned int frag, last, packlen, inp_len;
836 if (u_arg < sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
839 inp_len = param->inp[11] << 8 | param->inp[12];
841 if (EVP_CIPHER_CTX_encrypting(ctx)) {
842 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
847 return 0; /* too short */
849 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
851 } else if ((n4x = param->interleave / 4) && n4x <= 2)
852 inp_len = param->len;
857 SHA256_Update(&key->md, param->inp, 13);
862 frag = inp_len >> n4x;
863 last = inp_len + frag - (frag << n4x);
864 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
869 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
870 packlen = (packlen << n4x) - packlen;
871 packlen += 5 + 16 + ((last + 32 + 16) & -16);
873 param->interleave = x4;
877 return -1; /* not yet */
879 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
881 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
882 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
884 return (int)tls1_1_multi_block_encrypt(key, param->out,
885 param->inp, param->len,
886 param->interleave / 4);
888 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
895 static EVP_CIPHER aesni_128_cbc_hmac_sha256_cipher = {
896 # ifdef NID_aes_128_cbc_hmac_sha256
897 NID_aes_128_cbc_hmac_sha256,
901 AES_BLOCK_SIZE, 16, AES_BLOCK_SIZE,
902 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
903 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
904 aesni_cbc_hmac_sha256_init_key,
905 aesni_cbc_hmac_sha256_cipher,
907 sizeof(EVP_AES_HMAC_SHA256),
908 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
909 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
910 aesni_cbc_hmac_sha256_ctrl,
914 static EVP_CIPHER aesni_256_cbc_hmac_sha256_cipher = {
915 # ifdef NID_aes_256_cbc_hmac_sha256
916 NID_aes_256_cbc_hmac_sha256,
920 AES_BLOCK_SIZE, 32, 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_sha256_init_key,
924 aesni_cbc_hmac_sha256_cipher,
926 sizeof(EVP_AES_HMAC_SHA256),
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_sha256_ctrl,
933 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
935 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
936 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
937 &aesni_128_cbc_hmac_sha256_cipher : NULL);
940 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
942 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
943 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
944 &aesni_256_cbc_hmac_sha256_cipher : NULL);
947 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
952 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)