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 <openssl/rand_drbg.h>
22 #include "modes_lcl.h"
23 #include "internal/constant_time_locl.h"
24 #include "internal/evp_int.h"
29 SHA256_CTX head, tail, md;
30 size_t payload_length; /* AAD length in decrypt case */
33 unsigned char tls_aad[16]; /* 13 used */
35 } EVP_AES_HMAC_SHA256;
37 # define NO_PAYLOAD_LENGTH ((size_t)-1)
39 #if defined(AES_ASM) && ( \
40 defined(__x86_64) || defined(__x86_64__) || \
41 defined(_M_AMD64) || defined(_M_X64) )
43 extern unsigned int OPENSSL_ia32cap_P[];
44 # define AESNI_CAPABLE (1<<(57-32))
46 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
48 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
51 void aesni_cbc_encrypt(const unsigned char *in,
54 const AES_KEY *key, unsigned char *ivec, int enc);
56 int aesni_cbc_sha256_enc(const void *inp, void *out, size_t blocks,
57 const AES_KEY *key, unsigned char iv[16],
58 SHA256_CTX *ctx, const void *in0);
60 # define data(ctx) ((EVP_AES_HMAC_SHA256 *)EVP_CIPHER_CTX_get_cipher_data(ctx))
62 static int aesni_cbc_hmac_sha256_init_key(EVP_CIPHER_CTX *ctx,
63 const unsigned char *inkey,
64 const unsigned char *iv, int enc)
66 EVP_AES_HMAC_SHA256 *key = data(ctx);
70 ret = aesni_set_encrypt_key(inkey,
71 EVP_CIPHER_CTX_key_length(ctx) * 8,
74 ret = aesni_set_decrypt_key(inkey,
75 EVP_CIPHER_CTX_key_length(ctx) * 8,
78 SHA256_Init(&key->head); /* handy when benchmarking */
79 key->tail = key->head;
82 key->payload_length = NO_PAYLOAD_LENGTH;
84 return ret < 0 ? 0 : 1;
87 # define STITCHED_CALL
89 # if !defined(STITCHED_CALL)
93 void sha256_block_data_order(void *c, const void *p, size_t len);
95 static void sha256_update(SHA256_CTX *c, const void *data, size_t len)
97 const unsigned char *ptr = data;
100 if ((res = c->num)) {
101 res = SHA256_CBLOCK - res;
104 SHA256_Update(c, ptr, res);
109 res = len % SHA256_CBLOCK;
113 sha256_block_data_order(c, ptr, len / SHA256_CBLOCK);
118 if (c->Nl < (unsigned int)len)
123 SHA256_Update(c, ptr, res);
126 # ifdef SHA256_Update
127 # undef SHA256_Update
129 # define SHA256_Update sha256_update
131 # if !defined(OPENSSL_NO_MULTIBLOCK)
134 unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8];
137 const unsigned char *ptr;
141 void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int);
144 const unsigned char *inp;
150 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
152 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
154 const unsigned char *inp,
155 size_t inp_len, int n4x,
157 { /* n4x is 1 or 2 */
158 HASH_DESC hash_d[8], edges[8];
160 unsigned char storage[sizeof(SHA256_MB_CTX) + 32];
167 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
175 /* ask for IVs in bulk */
178 if (RAND_DRBG_bytes(drbg, IVs, 16 * x4) == 0)
180 } else if (RAND_bytes(IVs, 16 * x4) <= 0) {
185 ctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32));
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 + 32 + 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.h[0];
224 ctx->B[i] = key->md.h[1];
225 ctx->C[i] = key->md.h[2];
226 ctx->D[i] = key->md.h[3];
227 ctx->E[i] = key->md.h[4];
228 ctx->F[i] = key->md.h[5];
229 ctx->G[i] = key->md.h[6];
230 ctx->H[i] = key->md.h[7];
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 sha256_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 sha256_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 sha256_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 sha256_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.h[0];
332 blocks[i].d[1] = BSWAP4(ctx->B[i]);
333 ctx->B[i] = key->tail.h[1];
334 blocks[i].d[2] = BSWAP4(ctx->C[i]);
335 ctx->C[i] = key->tail.h[2];
336 blocks[i].d[3] = BSWAP4(ctx->D[i]);
337 ctx->D[i] = key->tail.h[3];
338 blocks[i].d[4] = BSWAP4(ctx->E[i]);
339 ctx->E[i] = key->tail.h[4];
340 blocks[i].d[5] = BSWAP4(ctx->F[i]);
341 ctx->F[i] = key->tail.h[5];
342 blocks[i].d[6] = BSWAP4(ctx->G[i]);
343 ctx->G[i] = key->tail.h[6];
344 blocks[i].d[7] = BSWAP4(ctx->H[i]);
345 ctx->H[i] = key->tail.h[7];
346 blocks[i].c[32] = 0x80;
347 blocks[i].d[15] = BSWAP4((64 + 32) * 8);
349 PUTU32(blocks[i].c + 0, ctx->A[i]);
350 ctx->A[i] = key->tail.h[0];
351 PUTU32(blocks[i].c + 4, ctx->B[i]);
352 ctx->B[i] = key->tail.h[1];
353 PUTU32(blocks[i].c + 8, ctx->C[i]);
354 ctx->C[i] = key->tail.h[2];
355 PUTU32(blocks[i].c + 12, ctx->D[i]);
356 ctx->D[i] = key->tail.h[3];
357 PUTU32(blocks[i].c + 16, ctx->E[i]);
358 ctx->E[i] = key->tail.h[4];
359 PUTU32(blocks[i].c + 20, ctx->F[i]);
360 ctx->F[i] = key->tail.h[5];
361 PUTU32(blocks[i].c + 24, ctx->G[i]);
362 ctx->G[i] = key->tail.h[6];
363 PUTU32(blocks[i].c + 28, ctx->H[i]);
364 ctx->H[i] = key->tail.h[7];
365 blocks[i].c[32] = 0x80;
366 PUTU32(blocks[i].c + 60, (64 + 32) * 8);
368 edges[i].ptr = blocks[i].c;
373 sha256_multi_block(ctx, edges, n4x);
375 for (i = 0; i < x4; i++) {
376 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
377 unsigned char *out0 = out;
379 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
380 ciph_d[i].inp = ciph_d[i].out;
385 PUTU32(out + 0, ctx->A[i]);
386 PUTU32(out + 4, ctx->B[i]);
387 PUTU32(out + 8, ctx->C[i]);
388 PUTU32(out + 12, ctx->D[i]);
389 PUTU32(out + 16, ctx->E[i]);
390 PUTU32(out + 20, ctx->F[i]);
391 PUTU32(out + 24, ctx->G[i]);
392 PUTU32(out + 28, ctx->H[i]);
398 for (j = 0; j <= pad; j++)
402 ciph_d[i].blocks = (len - processed) / 16;
403 len += 16; /* account for explicit iv */
406 out0[0] = ((u8 *)key->md.data)[8];
407 out0[1] = ((u8 *)key->md.data)[9];
408 out0[2] = ((u8 *)key->md.data)[10];
409 out0[3] = (u8)(len >> 8);
416 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
418 OPENSSL_cleanse(blocks, sizeof(blocks));
419 OPENSSL_cleanse(ctx, sizeof(*ctx));
425 static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx,
427 const unsigned char *in, size_t len)
429 EVP_AES_HMAC_SHA256 *key = data(ctx);
431 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
434 # if defined(STITCHED_CALL)
435 size_t aes_off = 0, blocks;
437 sha_off = SHA256_CBLOCK - key->md.num;
440 key->payload_length = NO_PAYLOAD_LENGTH;
442 if (len % AES_BLOCK_SIZE)
445 if (EVP_CIPHER_CTX_encrypting(ctx)) {
446 if (plen == NO_PAYLOAD_LENGTH)
449 ((plen + SHA256_DIGEST_LENGTH +
450 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
452 else if (key->aux.tls_ver >= TLS1_1_VERSION)
455 # if defined(STITCHED_CALL)
457 * Assembly stitch handles AVX-capable processors, but its
458 * performance is not optimal on AMD Jaguar, ~40% worse, for
459 * unknown reasons. Incidentally processor in question supports
460 * AVX, but not AMD-specific XOP extension, which can be used
461 * to identify it and avoid stitch invocation. So that after we
462 * establish that current CPU supports AVX, we even see if it's
463 * either even XOP-capable Bulldozer-based or GenuineIntel one.
464 * But SHAEXT-capable go ahead...
466 if (((OPENSSL_ia32cap_P[2] & (1 << 29)) || /* SHAEXT? */
467 ((OPENSSL_ia32cap_P[1] & (1 << (60 - 32))) && /* AVX? */
468 ((OPENSSL_ia32cap_P[1] & (1 << (43 - 32))) /* XOP? */
469 | (OPENSSL_ia32cap_P[0] & (1 << 30))))) && /* "Intel CPU"? */
470 plen > (sha_off + iv) &&
471 (blocks = (plen - (sha_off + iv)) / SHA256_CBLOCK)) {
472 SHA256_Update(&key->md, in + iv, sha_off);
474 (void)aesni_cbc_sha256_enc(in, out, blocks, &key->ks,
475 EVP_CIPHER_CTX_iv_noconst(ctx),
476 &key->md, in + iv + sha_off);
477 blocks *= SHA256_CBLOCK;
480 key->md.Nh += blocks >> 29;
481 key->md.Nl += blocks <<= 3;
482 if (key->md.Nl < (unsigned int)blocks)
489 SHA256_Update(&key->md, in + sha_off, plen - sha_off);
491 if (plen != len) { /* "TLS" mode of operation */
493 memcpy(out + aes_off, in + aes_off, plen - aes_off);
495 /* calculate HMAC and append it to payload */
496 SHA256_Final(out + plen, &key->md);
498 SHA256_Update(&key->md, out + plen, SHA256_DIGEST_LENGTH);
499 SHA256_Final(out + plen, &key->md);
501 /* pad the payload|hmac */
502 plen += SHA256_DIGEST_LENGTH;
503 for (l = len - plen - 1; plen < len; plen++)
505 /* encrypt HMAC|padding at once */
506 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
507 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
509 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
510 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
514 unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)];
515 unsigned char c[64 + SHA256_DIGEST_LENGTH];
518 /* arrange cache line alignment */
519 pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64));
521 /* decrypt HMAC|padding at once */
522 aesni_cbc_encrypt(in, out, len, &key->ks,
523 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
525 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
526 size_t inp_len, mask, j, i;
527 unsigned int res, maxpad, pad, bitlen;
530 unsigned int u[SHA_LBLOCK];
531 unsigned char c[SHA256_CBLOCK];
532 } *data = (void *)key->md.data;
534 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
538 if (len < (iv + SHA256_DIGEST_LENGTH + 1))
541 /* omit explicit iv */
545 /* figure out payload length */
547 maxpad = len - (SHA256_DIGEST_LENGTH + 1);
548 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
551 mask = constant_time_ge(maxpad, pad);
554 * If pad is invalid then we will fail the above test but we must
555 * continue anyway because we are in constant time code. However,
556 * we'll use the maxpad value instead of the supplied pad to make
557 * sure we perform well defined pointer arithmetic.
559 pad = constant_time_select(mask, pad, maxpad);
561 inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1);
563 key->aux.tls_aad[plen - 2] = inp_len >> 8;
564 key->aux.tls_aad[plen - 1] = inp_len;
568 SHA256_Update(&key->md, key->aux.tls_aad, plen);
570 # if 1 /* see original reference version in #else */
571 len -= SHA256_DIGEST_LENGTH; /* amend mac */
572 if (len >= (256 + SHA256_CBLOCK)) {
573 j = (len - (256 + SHA256_CBLOCK)) & (0 - SHA256_CBLOCK);
574 j += SHA256_CBLOCK - key->md.num;
575 SHA256_Update(&key->md, out, j);
581 /* but pretend as if we hashed padded payload */
582 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
584 bitlen = BSWAP4(bitlen);
587 mac.c[1] = (unsigned char)(bitlen >> 16);
588 mac.c[2] = (unsigned char)(bitlen >> 8);
589 mac.c[3] = (unsigned char)bitlen;
602 for (res = key->md.num, j = 0; j < len; j++) {
604 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
606 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
607 data->c[res++] = (unsigned char)c;
609 if (res != SHA256_CBLOCK)
612 /* j is not incremented yet */
613 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
614 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
615 sha256_block_data_order(&key->md, data, 1);
616 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
617 pmac->u[0] |= key->md.h[0] & mask;
618 pmac->u[1] |= key->md.h[1] & mask;
619 pmac->u[2] |= key->md.h[2] & mask;
620 pmac->u[3] |= key->md.h[3] & mask;
621 pmac->u[4] |= key->md.h[4] & mask;
622 pmac->u[5] |= key->md.h[5] & mask;
623 pmac->u[6] |= key->md.h[6] & mask;
624 pmac->u[7] |= key->md.h[7] & mask;
628 for (i = res; i < SHA256_CBLOCK; i++, j++)
631 if (res > SHA256_CBLOCK - 8) {
632 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
633 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
634 sha256_block_data_order(&key->md, data, 1);
635 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
636 pmac->u[0] |= key->md.h[0] & mask;
637 pmac->u[1] |= key->md.h[1] & mask;
638 pmac->u[2] |= key->md.h[2] & mask;
639 pmac->u[3] |= key->md.h[3] & mask;
640 pmac->u[4] |= key->md.h[4] & mask;
641 pmac->u[5] |= key->md.h[5] & mask;
642 pmac->u[6] |= key->md.h[6] & mask;
643 pmac->u[7] |= key->md.h[7] & mask;
645 memset(data, 0, SHA256_CBLOCK);
648 data->u[SHA_LBLOCK - 1] = bitlen;
649 sha256_block_data_order(&key->md, data, 1);
650 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
651 pmac->u[0] |= key->md.h[0] & mask;
652 pmac->u[1] |= key->md.h[1] & mask;
653 pmac->u[2] |= key->md.h[2] & mask;
654 pmac->u[3] |= key->md.h[3] & mask;
655 pmac->u[4] |= key->md.h[4] & mask;
656 pmac->u[5] |= key->md.h[5] & mask;
657 pmac->u[6] |= key->md.h[6] & mask;
658 pmac->u[7] |= key->md.h[7] & mask;
661 pmac->u[0] = BSWAP4(pmac->u[0]);
662 pmac->u[1] = BSWAP4(pmac->u[1]);
663 pmac->u[2] = BSWAP4(pmac->u[2]);
664 pmac->u[3] = BSWAP4(pmac->u[3]);
665 pmac->u[4] = BSWAP4(pmac->u[4]);
666 pmac->u[5] = BSWAP4(pmac->u[5]);
667 pmac->u[6] = BSWAP4(pmac->u[6]);
668 pmac->u[7] = BSWAP4(pmac->u[7]);
670 for (i = 0; i < 8; i++) {
672 pmac->c[4 * i + 0] = (unsigned char)(res >> 24);
673 pmac->c[4 * i + 1] = (unsigned char)(res >> 16);
674 pmac->c[4 * i + 2] = (unsigned char)(res >> 8);
675 pmac->c[4 * i + 3] = (unsigned char)res;
678 len += SHA256_DIGEST_LENGTH;
680 SHA256_Update(&key->md, out, inp_len);
682 SHA256_Final(pmac->c, &key->md);
685 unsigned int inp_blocks, pad_blocks;
687 /* but pretend as if we hashed padded payload */
689 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
690 res += (unsigned int)(len - inp_len);
691 pad_blocks = res / SHA256_CBLOCK;
692 res %= SHA256_CBLOCK;
694 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
695 for (; inp_blocks < pad_blocks; inp_blocks++)
696 sha1_block_data_order(&key->md, data, 1);
698 # endif /* pre-lucky-13 reference version of above */
700 SHA256_Update(&key->md, pmac->c, SHA256_DIGEST_LENGTH);
701 SHA256_Final(pmac->c, &key->md);
706 # if 1 /* see original reference version in #else */
709 out + len - 1 - maxpad - SHA256_DIGEST_LENGTH;
710 size_t off = out - p;
711 unsigned int c, cmask;
713 maxpad += SHA256_DIGEST_LENGTH;
714 for (res = 0, i = 0, j = 0; j < maxpad; j++) {
717 ((int)(j - off - SHA256_DIGEST_LENGTH)) >>
718 (sizeof(int) * 8 - 1);
719 res |= (c ^ pad) & ~cmask; /* ... and padding */
720 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
721 res |= (c ^ pmac->c[i]) & cmask;
724 maxpad -= SHA256_DIGEST_LENGTH;
726 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
729 # else /* pre-lucky-13 reference version of above */
730 for (res = 0, i = 0; i < SHA256_DIGEST_LENGTH; i++)
731 res |= out[i] ^ pmac->c[i];
732 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
736 pad = (pad & ~res) | (maxpad & res);
737 out = out + len - 1 - pad;
738 for (res = 0, i = 0; i < pad; i++)
741 res = (0 - res) >> (sizeof(res) * 8 - 1);
746 SHA256_Update(&key->md, out, len);
753 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
756 EVP_AES_HMAC_SHA256 *key = data(ctx);
757 unsigned int u_arg = (unsigned int)arg;
760 case EVP_CTRL_AEAD_SET_MAC_KEY:
763 unsigned char hmac_key[64];
765 memset(hmac_key, 0, sizeof(hmac_key));
770 if (u_arg > sizeof(hmac_key)) {
771 SHA256_Init(&key->head);
772 SHA256_Update(&key->head, ptr, arg);
773 SHA256_Final(hmac_key, &key->head);
775 memcpy(hmac_key, ptr, arg);
778 for (i = 0; i < sizeof(hmac_key); i++)
779 hmac_key[i] ^= 0x36; /* ipad */
780 SHA256_Init(&key->head);
781 SHA256_Update(&key->head, hmac_key, sizeof(hmac_key));
783 for (i = 0; i < sizeof(hmac_key); i++)
784 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
785 SHA256_Init(&key->tail);
786 SHA256_Update(&key->tail, hmac_key, sizeof(hmac_key));
788 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
792 case EVP_CTRL_AEAD_TLS1_AAD:
794 unsigned char *p = ptr;
797 if (arg != EVP_AEAD_TLS1_AAD_LEN)
800 len = p[arg - 2] << 8 | p[arg - 1];
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 if (len < AES_BLOCK_SIZE)
808 len -= AES_BLOCK_SIZE;
809 p[arg - 2] = len >> 8;
813 SHA256_Update(&key->md, p, arg);
815 return (int)(((len + SHA256_DIGEST_LENGTH +
816 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
819 memcpy(key->aux.tls_aad, ptr, arg);
820 key->payload_length = arg;
822 return SHA256_DIGEST_LENGTH;
825 # if !defined(OPENSSL_NO_MULTIBLOCK)
826 case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
827 return (int)(5 + 16 + ((arg + 32 + 16) & -16));
828 case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
830 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
831 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
832 unsigned int n4x = 1, x4;
833 unsigned int frag, last, packlen, inp_len;
838 if (u_arg < sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
841 inp_len = param->inp[11] << 8 | param->inp[12];
843 if (EVP_CIPHER_CTX_encrypting(ctx)) {
844 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
849 return 0; /* too short */
851 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
853 } else if ((n4x = param->interleave / 4) && n4x <= 2)
854 inp_len = param->len;
859 SHA256_Update(&key->md, param->inp, 13);
864 frag = inp_len >> n4x;
865 last = inp_len + frag - (frag << n4x);
866 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
871 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
872 packlen = (packlen << n4x) - packlen;
873 packlen += 5 + 16 + ((last + 32 + 16) & -16);
875 param->interleave = x4;
879 return -1; /* not yet */
881 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
883 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
884 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
886 return (int)tls1_1_multi_block_encrypt(key, param->out,
887 param->inp, param->len,
888 param->interleave / 4,
891 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
898 static EVP_CIPHER aesni_128_cbc_hmac_sha256_cipher = {
899 # ifdef NID_aes_128_cbc_hmac_sha256
900 NID_aes_128_cbc_hmac_sha256,
904 AES_BLOCK_SIZE, 16, AES_BLOCK_SIZE,
905 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
906 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
907 aesni_cbc_hmac_sha256_init_key,
908 aesni_cbc_hmac_sha256_cipher,
910 sizeof(EVP_AES_HMAC_SHA256),
911 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
912 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
913 aesni_cbc_hmac_sha256_ctrl,
917 static EVP_CIPHER aesni_256_cbc_hmac_sha256_cipher = {
918 # ifdef NID_aes_256_cbc_hmac_sha256
919 NID_aes_256_cbc_hmac_sha256,
923 AES_BLOCK_SIZE, 32, AES_BLOCK_SIZE,
924 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
925 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
926 aesni_cbc_hmac_sha256_init_key,
927 aesni_cbc_hmac_sha256_cipher,
929 sizeof(EVP_AES_HMAC_SHA256),
930 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
931 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
932 aesni_cbc_hmac_sha256_ctrl,
936 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
938 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
939 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
940 &aesni_128_cbc_hmac_sha256_cipher : NULL);
943 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
945 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
946 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
947 &aesni_256_cbc_hmac_sha256_cipher : NULL);
950 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
955 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)