1 /*
2 * Copyright 2013-2021 The OpenSSL Project Authors. All Rights Reserved.
3 *
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
8 */
9
10 /*
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
13 * the case here.
14 */
15 #include "internal/deprecated.h"
16
17 #include <stdio.h>
18 #include <string.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 "internal/constant_time.h"
28 #include "crypto/evp.h"
29 #include "evp_local.h"
30
31 typedef struct {
32 AES_KEY ks;
33 SHA256_CTX head, tail, md;
34 size_t payload_length; /* AAD length in decrypt case */
35 union {
36 unsigned int tls_ver;
37 unsigned char tls_aad[16]; /* 13 used */
38 } aux;
39 } EVP_AES_HMAC_SHA256;
40
41 # define NO_PAYLOAD_LENGTH ((size_t)-1)
42
43 #if defined(AES_ASM) && ( \
44 defined(__x86_64) || defined(__x86_64__) || \
45 defined(_M_AMD64) || defined(_M_X64) )
46
47 # define AESNI_CAPABLE (1<<(57-32))
48
49 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
50 AES_KEY *key);
51 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
52 AES_KEY *key);
53
54 void aesni_cbc_encrypt(const unsigned char *in,
55 unsigned char *out,
56 size_t length,
57 const AES_KEY *key, unsigned char *ivec, int enc);
58
59 int aesni_cbc_sha256_enc(const void *inp, void *out, size_t blocks,
60 const AES_KEY *key, unsigned char iv[16],
61 SHA256_CTX *ctx, const void *in0);
62
63 # define data(ctx) ((EVP_AES_HMAC_SHA256 *)EVP_CIPHER_CTX_get_cipher_data(ctx))
64
aesni_cbc_hmac_sha256_init_key(EVP_CIPHER_CTX * ctx,const unsigned char * inkey,const unsigned char * iv,int enc)65 static int aesni_cbc_hmac_sha256_init_key(EVP_CIPHER_CTX *ctx,
66 const unsigned char *inkey,
67 const unsigned char *iv, int enc)
68 {
69 EVP_AES_HMAC_SHA256 *key = data(ctx);
70 int ret;
71
72 if (enc)
73 ret = aesni_set_encrypt_key(inkey,
74 EVP_CIPHER_CTX_get_key_length(ctx) * 8,
75 &key->ks);
76 else
77 ret = aesni_set_decrypt_key(inkey,
78 EVP_CIPHER_CTX_get_key_length(ctx) * 8,
79 &key->ks);
80
81 SHA256_Init(&key->head); /* handy when benchmarking */
82 key->tail = key->head;
83 key->md = key->head;
84
85 key->payload_length = NO_PAYLOAD_LENGTH;
86
87 return ret < 0 ? 0 : 1;
88 }
89
90 # define STITCHED_CALL
91
92 # if !defined(STITCHED_CALL)
93 # define aes_off 0
94 # endif
95
96 void sha256_block_data_order(void *c, const void *p, size_t len);
97
sha256_update(SHA256_CTX * c,const void * data,size_t len)98 static void sha256_update(SHA256_CTX *c, const void *data, size_t len)
99 {
100 const unsigned char *ptr = data;
101 size_t res;
102
103 if ((res = c->num)) {
104 res = SHA256_CBLOCK - res;
105 if (len < res)
106 res = len;
107 SHA256_Update(c, ptr, res);
108 ptr += res;
109 len -= res;
110 }
111
112 res = len % SHA256_CBLOCK;
113 len -= res;
114
115 if (len) {
116 sha256_block_data_order(c, ptr, len / SHA256_CBLOCK);
117
118 ptr += len;
119 c->Nh += len >> 29;
120 c->Nl += len <<= 3;
121 if (c->Nl < (unsigned int)len)
122 c->Nh++;
123 }
124
125 if (res)
126 SHA256_Update(c, ptr, res);
127 }
128
129 # ifdef SHA256_Update
130 # undef SHA256_Update
131 # endif
132 # define SHA256_Update sha256_update
133
134 # if !defined(OPENSSL_NO_MULTIBLOCK)
135
136 typedef struct {
137 unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8];
138 } SHA256_MB_CTX;
139 typedef struct {
140 const unsigned char *ptr;
141 int blocks;
142 } HASH_DESC;
143
144 void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int);
145
146 typedef struct {
147 const unsigned char *inp;
148 unsigned char *out;
149 int blocks;
150 u64 iv[2];
151 } CIPH_DESC;
152
153 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
154
tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 * key,unsigned char * out,const unsigned char * inp,size_t inp_len,int n4x)155 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
156 unsigned char *out,
157 const unsigned char *inp,
158 size_t inp_len, int n4x)
159 { /* n4x is 1 or 2 */
160 HASH_DESC hash_d[8], edges[8];
161 CIPH_DESC ciph_d[8];
162 unsigned char storage[sizeof(SHA256_MB_CTX) + 32];
163 union {
164 u64 q[16];
165 u32 d[32];
166 u8 c[128];
167 } blocks[8];
168 SHA256_MB_CTX *ctx;
169 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
170 0;
171 size_t ret = 0;
172 u8 *IVs;
173 # if defined(BSWAP8)
174 u64 seqnum;
175 # endif
176
177 /* ask for IVs in bulk */
178 if (RAND_bytes((IVs = blocks[0].c), 16 * x4) <= 0)
179 return 0;
180
181 /* align */
182 ctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32));
183
184 frag = (unsigned int)inp_len >> (1 + n4x);
185 last = (unsigned int)inp_len + frag - (frag << (1 + n4x));
186 if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) {
187 frag++;
188 last -= x4 - 1;
189 }
190
191 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
192
193 /* populate descriptors with pointers and IVs */
194 hash_d[0].ptr = inp;
195 ciph_d[0].inp = inp;
196 /* 5+16 is place for header and explicit IV */
197 ciph_d[0].out = out + 5 + 16;
198 memcpy(ciph_d[0].out - 16, IVs, 16);
199 memcpy(ciph_d[0].iv, IVs, 16);
200 IVs += 16;
201
202 for (i = 1; i < x4; i++) {
203 ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag;
204 ciph_d[i].out = ciph_d[i - 1].out + packlen;
205 memcpy(ciph_d[i].out - 16, IVs, 16);
206 memcpy(ciph_d[i].iv, IVs, 16);
207 IVs += 16;
208 }
209
210 # if defined(BSWAP8)
211 memcpy(blocks[0].c, key->md.data, 8);
212 seqnum = BSWAP8(blocks[0].q[0]);
213 # endif
214 for (i = 0; i < x4; i++) {
215 unsigned int len = (i == (x4 - 1) ? last : frag);
216 # if !defined(BSWAP8)
217 unsigned int carry, j;
218 # endif
219
220 ctx->A[i] = key->md.h[0];
221 ctx->B[i] = key->md.h[1];
222 ctx->C[i] = key->md.h[2];
223 ctx->D[i] = key->md.h[3];
224 ctx->E[i] = key->md.h[4];
225 ctx->F[i] = key->md.h[5];
226 ctx->G[i] = key->md.h[6];
227 ctx->H[i] = key->md.h[7];
228
229 /* fix seqnum */
230 # if defined(BSWAP8)
231 blocks[i].q[0] = BSWAP8(seqnum + i);
232 # else
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);
236 }
237 # endif
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];
241 /* fix length */
242 blocks[i].c[11] = (u8)(len >> 8);
243 blocks[i].c[12] = (u8)(len);
244
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;
248
249 edges[i].ptr = blocks[i].c;
250 edges[i].blocks = 1;
251 }
252
253 /* hash 13-byte headers and first 64-13 bytes of inputs */
254 sha256_multi_block(ctx, edges, n4x);
255 /* hash bulk inputs */
256 # define MAXCHUNKSIZE 2048
257 # if MAXCHUNKSIZE%64
258 # error "MAXCHUNKSIZE is not divisible by 64"
259 # elif MAXCHUNKSIZE
260 /*
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
263 */
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;
270 }
271 do {
272 sha256_multi_block(ctx, edges, n4x);
273 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
274
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);
283 }
284 processed += MAXCHUNKSIZE;
285 minblocks -= MAXCHUNKSIZE / 64;
286 } while (minblocks > MAXCHUNKSIZE / 64);
287 }
288 # endif
289 # undef MAXCHUNKSIZE
290 sha256_multi_block(ctx, hash_d, n4x);
291
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;
297
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)) {
304 # ifdef BSWAP4
305 blocks[i].d[15] = BSWAP4(len);
306 # else
307 PUTU32(blocks[i].c + 60, len);
308 # endif
309 edges[i].blocks = 1;
310 } else {
311 # ifdef BSWAP4
312 blocks[i].d[31] = BSWAP4(len);
313 # else
314 PUTU32(blocks[i].c + 124, len);
315 # endif
316 edges[i].blocks = 2;
317 }
318 edges[i].ptr = blocks[i].c;
319 }
320
321 /* hash input tails and finalize */
322 sha256_multi_block(ctx, edges, n4x);
323
324 memset(blocks, 0, sizeof(blocks));
325 for (i = 0; i < x4; i++) {
326 # ifdef BSWAP4
327 blocks[i].d[0] = BSWAP4(ctx->A[i]);
328 ctx->A[i] = key->tail.h[0];
329 blocks[i].d[1] = BSWAP4(ctx->B[i]);
330 ctx->B[i] = key->tail.h[1];
331 blocks[i].d[2] = BSWAP4(ctx->C[i]);
332 ctx->C[i] = key->tail.h[2];
333 blocks[i].d[3] = BSWAP4(ctx->D[i]);
334 ctx->D[i] = key->tail.h[3];
335 blocks[i].d[4] = BSWAP4(ctx->E[i]);
336 ctx->E[i] = key->tail.h[4];
337 blocks[i].d[5] = BSWAP4(ctx->F[i]);
338 ctx->F[i] = key->tail.h[5];
339 blocks[i].d[6] = BSWAP4(ctx->G[i]);
340 ctx->G[i] = key->tail.h[6];
341 blocks[i].d[7] = BSWAP4(ctx->H[i]);
342 ctx->H[i] = key->tail.h[7];
343 blocks[i].c[32] = 0x80;
344 blocks[i].d[15] = BSWAP4((64 + 32) * 8);
345 # else
346 PUTU32(blocks[i].c + 0, ctx->A[i]);
347 ctx->A[i] = key->tail.h[0];
348 PUTU32(blocks[i].c + 4, ctx->B[i]);
349 ctx->B[i] = key->tail.h[1];
350 PUTU32(blocks[i].c + 8, ctx->C[i]);
351 ctx->C[i] = key->tail.h[2];
352 PUTU32(blocks[i].c + 12, ctx->D[i]);
353 ctx->D[i] = key->tail.h[3];
354 PUTU32(blocks[i].c + 16, ctx->E[i]);
355 ctx->E[i] = key->tail.h[4];
356 PUTU32(blocks[i].c + 20, ctx->F[i]);
357 ctx->F[i] = key->tail.h[5];
358 PUTU32(blocks[i].c + 24, ctx->G[i]);
359 ctx->G[i] = key->tail.h[6];
360 PUTU32(blocks[i].c + 28, ctx->H[i]);
361 ctx->H[i] = key->tail.h[7];
362 blocks[i].c[32] = 0x80;
363 PUTU32(blocks[i].c + 60, (64 + 32) * 8);
364 # endif
365 edges[i].ptr = blocks[i].c;
366 edges[i].blocks = 1;
367 }
368
369 /* finalize MACs */
370 sha256_multi_block(ctx, edges, n4x);
371
372 for (i = 0; i < x4; i++) {
373 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
374 unsigned char *out0 = out;
375
376 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
377 ciph_d[i].inp = ciph_d[i].out;
378
379 out += 5 + 16 + len;
380
381 /* write MAC */
382 PUTU32(out + 0, ctx->A[i]);
383 PUTU32(out + 4, ctx->B[i]);
384 PUTU32(out + 8, ctx->C[i]);
385 PUTU32(out + 12, ctx->D[i]);
386 PUTU32(out + 16, ctx->E[i]);
387 PUTU32(out + 20, ctx->F[i]);
388 PUTU32(out + 24, ctx->G[i]);
389 PUTU32(out + 28, ctx->H[i]);
390 out += 32;
391 len += 32;
392
393 /* pad */
394 pad = 15 - len % 16;
395 for (j = 0; j <= pad; j++)
396 *(out++) = pad;
397 len += pad + 1;
398
399 ciph_d[i].blocks = (len - processed) / 16;
400 len += 16; /* account for explicit iv */
401
402 /* arrange header */
403 out0[0] = ((u8 *)key->md.data)[8];
404 out0[1] = ((u8 *)key->md.data)[9];
405 out0[2] = ((u8 *)key->md.data)[10];
406 out0[3] = (u8)(len >> 8);
407 out0[4] = (u8)(len);
408
409 ret += len + 5;
410 inp += frag;
411 }
412
413 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
414
415 OPENSSL_cleanse(blocks, sizeof(blocks));
416 OPENSSL_cleanse(ctx, sizeof(*ctx));
417
418 return ret;
419 }
420 # endif
421
aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX * ctx,unsigned char * out,const unsigned char * in,size_t len)422 static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx,
423 unsigned char *out,
424 const unsigned char *in, size_t len)
425 {
426 EVP_AES_HMAC_SHA256 *key = data(ctx);
427 unsigned int l;
428 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
429 * later */
430 sha_off = 0;
431 # if defined(STITCHED_CALL)
432 size_t aes_off = 0, blocks;
433
434 sha_off = SHA256_CBLOCK - key->md.num;
435 # endif
436
437 key->payload_length = NO_PAYLOAD_LENGTH;
438
439 if (len % AES_BLOCK_SIZE)
440 return 0;
441
442 if (EVP_CIPHER_CTX_is_encrypting(ctx)) {
443 if (plen == NO_PAYLOAD_LENGTH)
444 plen = len;
445 else if (len !=
446 ((plen + SHA256_DIGEST_LENGTH +
447 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
448 return 0;
449 else if (key->aux.tls_ver >= TLS1_1_VERSION)
450 iv = AES_BLOCK_SIZE;
451
452 # if defined(STITCHED_CALL)
453 /*
454 * Assembly stitch handles AVX-capable processors, but its
455 * performance is not optimal on AMD Jaguar, ~40% worse, for
456 * unknown reasons. Incidentally processor in question supports
457 * AVX, but not AMD-specific XOP extension, which can be used
458 * to identify it and avoid stitch invocation. So that after we
459 * establish that current CPU supports AVX, we even see if it's
460 * either even XOP-capable Bulldozer-based or GenuineIntel one.
461 * But SHAEXT-capable go ahead...
462 */
463 if (((OPENSSL_ia32cap_P[2] & (1 << 29)) || /* SHAEXT? */
464 ((OPENSSL_ia32cap_P[1] & (1 << (60 - 32))) && /* AVX? */
465 ((OPENSSL_ia32cap_P[1] & (1 << (43 - 32))) /* XOP? */
466 | (OPENSSL_ia32cap_P[0] & (1 << 30))))) && /* "Intel CPU"? */
467 plen > (sha_off + iv) &&
468 (blocks = (plen - (sha_off + iv)) / SHA256_CBLOCK)) {
469 SHA256_Update(&key->md, in + iv, sha_off);
470
471 (void)aesni_cbc_sha256_enc(in, out, blocks, &key->ks,
472 ctx->iv, &key->md, in + iv + sha_off);
473 blocks *= SHA256_CBLOCK;
474 aes_off += blocks;
475 sha_off += blocks;
476 key->md.Nh += blocks >> 29;
477 key->md.Nl += blocks <<= 3;
478 if (key->md.Nl < (unsigned int)blocks)
479 key->md.Nh++;
480 } else {
481 sha_off = 0;
482 }
483 # endif
484 sha_off += iv;
485 SHA256_Update(&key->md, in + sha_off, plen - sha_off);
486
487 if (plen != len) { /* "TLS" mode of operation */
488 if (in != out)
489 memcpy(out + aes_off, in + aes_off, plen - aes_off);
490
491 /* calculate HMAC and append it to payload */
492 SHA256_Final(out + plen, &key->md);
493 key->md = key->tail;
494 SHA256_Update(&key->md, out + plen, SHA256_DIGEST_LENGTH);
495 SHA256_Final(out + plen, &key->md);
496
497 /* pad the payload|hmac */
498 plen += SHA256_DIGEST_LENGTH;
499 for (l = len - plen - 1; plen < len; plen++)
500 out[plen] = l;
501 /* encrypt HMAC|padding at once */
502 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
503 &key->ks, ctx->iv, 1);
504 } else {
505 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
506 &key->ks, ctx->iv, 1);
507 }
508 } else {
509 union {
510 unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)];
511 unsigned char c[64 + SHA256_DIGEST_LENGTH];
512 } mac, *pmac;
513
514 /* arrange cache line alignment */
515 pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64));
516
517 /* decrypt HMAC|padding at once */
518 aesni_cbc_encrypt(in, out, len, &key->ks,
519 ctx->iv, 0);
520
521 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
522 size_t inp_len, mask, j, i;
523 unsigned int res, maxpad, pad, bitlen;
524 int ret = 1;
525 union {
526 unsigned int u[SHA_LBLOCK];
527 unsigned char c[SHA256_CBLOCK];
528 } *data = (void *)key->md.data;
529
530 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
531 >= TLS1_1_VERSION)
532 iv = AES_BLOCK_SIZE;
533
534 if (len < (iv + SHA256_DIGEST_LENGTH + 1))
535 return 0;
536
537 /* omit explicit iv */
538 out += iv;
539 len -= iv;
540
541 /* figure out payload length */
542 pad = out[len - 1];
543 maxpad = len - (SHA256_DIGEST_LENGTH + 1);
544 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
545 maxpad &= 255;
546
547 mask = constant_time_ge(maxpad, pad);
548 ret &= mask;
549 /*
550 * If pad is invalid then we will fail the above test but we must
551 * continue anyway because we are in constant time code. However,
552 * we'll use the maxpad value instead of the supplied pad to make
553 * sure we perform well defined pointer arithmetic.
554 */
555 pad = constant_time_select(mask, pad, maxpad);
556
557 inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1);
558
559 key->aux.tls_aad[plen - 2] = inp_len >> 8;
560 key->aux.tls_aad[plen - 1] = inp_len;
561
562 /* calculate HMAC */
563 key->md = key->head;
564 SHA256_Update(&key->md, key->aux.tls_aad, plen);
565
566 # if 1 /* see original reference version in #else */
567 len -= SHA256_DIGEST_LENGTH; /* amend mac */
568 if (len >= (256 + SHA256_CBLOCK)) {
569 j = (len - (256 + SHA256_CBLOCK)) & (0 - SHA256_CBLOCK);
570 j += SHA256_CBLOCK - key->md.num;
571 SHA256_Update(&key->md, out, j);
572 out += j;
573 len -= j;
574 inp_len -= j;
575 }
576
577 /* but pretend as if we hashed padded payload */
578 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
579 # ifdef BSWAP4
580 bitlen = BSWAP4(bitlen);
581 # else
582 mac.c[0] = 0;
583 mac.c[1] = (unsigned char)(bitlen >> 16);
584 mac.c[2] = (unsigned char)(bitlen >> 8);
585 mac.c[3] = (unsigned char)bitlen;
586 bitlen = mac.u[0];
587 # endif
588
589 pmac->u[0] = 0;
590 pmac->u[1] = 0;
591 pmac->u[2] = 0;
592 pmac->u[3] = 0;
593 pmac->u[4] = 0;
594 pmac->u[5] = 0;
595 pmac->u[6] = 0;
596 pmac->u[7] = 0;
597
598 for (res = key->md.num, j = 0; j < len; j++) {
599 size_t c = out[j];
600 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
601 c &= mask;
602 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
603 data->c[res++] = (unsigned char)c;
604
605 if (res != SHA256_CBLOCK)
606 continue;
607
608 /* j is not incremented yet */
609 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
610 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
611 sha256_block_data_order(&key->md, data, 1);
612 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
613 pmac->u[0] |= key->md.h[0] & mask;
614 pmac->u[1] |= key->md.h[1] & mask;
615 pmac->u[2] |= key->md.h[2] & mask;
616 pmac->u[3] |= key->md.h[3] & mask;
617 pmac->u[4] |= key->md.h[4] & mask;
618 pmac->u[5] |= key->md.h[5] & mask;
619 pmac->u[6] |= key->md.h[6] & mask;
620 pmac->u[7] |= key->md.h[7] & mask;
621 res = 0;
622 }
623
624 for (i = res; i < SHA256_CBLOCK; i++, j++)
625 data->c[i] = 0;
626
627 if (res > SHA256_CBLOCK - 8) {
628 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
629 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
630 sha256_block_data_order(&key->md, data, 1);
631 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
632 pmac->u[0] |= key->md.h[0] & mask;
633 pmac->u[1] |= key->md.h[1] & mask;
634 pmac->u[2] |= key->md.h[2] & mask;
635 pmac->u[3] |= key->md.h[3] & mask;
636 pmac->u[4] |= key->md.h[4] & mask;
637 pmac->u[5] |= key->md.h[5] & mask;
638 pmac->u[6] |= key->md.h[6] & mask;
639 pmac->u[7] |= key->md.h[7] & mask;
640
641 memset(data, 0, SHA256_CBLOCK);
642 j += 64;
643 }
644 data->u[SHA_LBLOCK - 1] = bitlen;
645 sha256_block_data_order(&key->md, data, 1);
646 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
647 pmac->u[0] |= key->md.h[0] & mask;
648 pmac->u[1] |= key->md.h[1] & mask;
649 pmac->u[2] |= key->md.h[2] & mask;
650 pmac->u[3] |= key->md.h[3] & mask;
651 pmac->u[4] |= key->md.h[4] & mask;
652 pmac->u[5] |= key->md.h[5] & mask;
653 pmac->u[6] |= key->md.h[6] & mask;
654 pmac->u[7] |= key->md.h[7] & mask;
655
656 # ifdef BSWAP4
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]);
662 pmac->u[5] = BSWAP4(pmac->u[5]);
663 pmac->u[6] = BSWAP4(pmac->u[6]);
664 pmac->u[7] = BSWAP4(pmac->u[7]);
665 # else
666 for (i = 0; i < 8; i++) {
667 res = pmac->u[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;
672 }
673 # endif
674 len += SHA256_DIGEST_LENGTH;
675 # else
676 SHA256_Update(&key->md, out, inp_len);
677 res = key->md.num;
678 SHA256_Final(pmac->c, &key->md);
679
680 {
681 unsigned int inp_blocks, pad_blocks;
682
683 /* but pretend as if we hashed padded payload */
684 inp_blocks =
685 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
686 res += (unsigned int)(len - inp_len);
687 pad_blocks = res / SHA256_CBLOCK;
688 res %= SHA256_CBLOCK;
689 pad_blocks +=
690 1 + ((SHA256_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1));
691 for (; inp_blocks < pad_blocks; inp_blocks++)
692 sha1_block_data_order(&key->md, data, 1);
693 }
694 # endif /* pre-lucky-13 reference version of above */
695 key->md = key->tail;
696 SHA256_Update(&key->md, pmac->c, SHA256_DIGEST_LENGTH);
697 SHA256_Final(pmac->c, &key->md);
698
699 /* verify HMAC */
700 out += inp_len;
701 len -= inp_len;
702 # if 1 /* see original reference version in #else */
703 {
704 unsigned char *p =
705 out + len - 1 - maxpad - SHA256_DIGEST_LENGTH;
706 size_t off = out - p;
707 unsigned int c, cmask;
708
709 for (res = 0, i = 0, j = 0; j < maxpad + SHA256_DIGEST_LENGTH;
710 j++) {
711 c = p[j];
712 cmask =
713 ((int)(j - off - SHA256_DIGEST_LENGTH)) >>
714 (sizeof(int) * 8 - 1);
715 res |= (c ^ pad) & ~cmask; /* ... and padding */
716 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
717 res |= (c ^ pmac->c[i]) & cmask;
718 i += 1 & cmask;
719 }
720
721 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
722 ret &= (int)~res;
723 }
724 # else /* pre-lucky-13 reference version of above */
725 for (res = 0, i = 0; i < SHA256_DIGEST_LENGTH; i++)
726 res |= out[i] ^ pmac->c[i];
727 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
728 ret &= (int)~res;
729
730 /* verify padding */
731 pad = (pad & ~res) | (maxpad & res);
732 out = out + len - 1 - pad;
733 for (res = 0, i = 0; i < pad; i++)
734 res |= out[i] ^ pad;
735
736 res = (0 - res) >> (sizeof(res) * 8 - 1);
737 ret &= (int)~res;
738 # endif
739 return ret;
740 } else {
741 SHA256_Update(&key->md, out, len);
742 }
743 }
744
745 return 1;
746 }
747
aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX * ctx,int type,int arg,void * ptr)748 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
749 void *ptr)
750 {
751 EVP_AES_HMAC_SHA256 *key = data(ctx);
752 unsigned int u_arg = (unsigned int)arg;
753
754 switch (type) {
755 case EVP_CTRL_AEAD_SET_MAC_KEY:
756 {
757 unsigned int i;
758 unsigned char hmac_key[64];
759
760 memset(hmac_key, 0, sizeof(hmac_key));
761
762 if (arg < 0)
763 return -1;
764
765 if (u_arg > sizeof(hmac_key)) {
766 SHA256_Init(&key->head);
767 SHA256_Update(&key->head, ptr, arg);
768 SHA256_Final(hmac_key, &key->head);
769 } else {
770 memcpy(hmac_key, ptr, arg);
771 }
772
773 for (i = 0; i < sizeof(hmac_key); i++)
774 hmac_key[i] ^= 0x36; /* ipad */
775 SHA256_Init(&key->head);
776 SHA256_Update(&key->head, hmac_key, sizeof(hmac_key));
777
778 for (i = 0; i < sizeof(hmac_key); i++)
779 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
780 SHA256_Init(&key->tail);
781 SHA256_Update(&key->tail, hmac_key, sizeof(hmac_key));
782
783 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
784
785 return 1;
786 }
787 case EVP_CTRL_AEAD_TLS1_AAD:
788 {
789 unsigned char *p = ptr;
790 unsigned int len;
791
792 if (arg != EVP_AEAD_TLS1_AAD_LEN)
793 return -1;
794
795 len = p[arg - 2] << 8 | p[arg - 1];
796
797 if (EVP_CIPHER_CTX_is_encrypting(ctx)) {
798 key->payload_length = len;
799 if ((key->aux.tls_ver =
800 p[arg - 4] << 8 | p[arg - 3]) >= TLS1_1_VERSION) {
801 if (len < AES_BLOCK_SIZE)
802 return 0;
803 len -= AES_BLOCK_SIZE;
804 p[arg - 2] = len >> 8;
805 p[arg - 1] = len;
806 }
807 key->md = key->head;
808 SHA256_Update(&key->md, p, arg);
809
810 return (int)(((len + SHA256_DIGEST_LENGTH +
811 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
812 - len);
813 } else {
814 memcpy(key->aux.tls_aad, ptr, arg);
815 key->payload_length = arg;
816
817 return SHA256_DIGEST_LENGTH;
818 }
819 }
820 # if !defined(OPENSSL_NO_MULTIBLOCK)
821 case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
822 return (int)(5 + 16 + ((arg + 32 + 16) & -16));
823 case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
824 {
825 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
826 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
827 unsigned int n4x = 1, x4;
828 unsigned int frag, last, packlen, inp_len;
829
830 if (arg < 0)
831 return -1;
832
833 if (u_arg < sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
834 return -1;
835
836 inp_len = param->inp[11] << 8 | param->inp[12];
837
838 if (EVP_CIPHER_CTX_is_encrypting(ctx)) {
839 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
840 return -1;
841
842 if (inp_len) {
843 if (inp_len < 4096)
844 return 0; /* too short */
845
846 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
847 n4x = 2; /* AVX2 */
848 } else if ((n4x = param->interleave / 4) && n4x <= 2)
849 inp_len = param->len;
850 else
851 return -1;
852
853 key->md = key->head;
854 SHA256_Update(&key->md, param->inp, 13);
855
856 x4 = 4 * n4x;
857 n4x += 1;
858
859 frag = inp_len >> n4x;
860 last = inp_len + frag - (frag << n4x);
861 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
862 frag++;
863 last -= x4 - 1;
864 }
865
866 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
867 packlen = (packlen << n4x) - packlen;
868 packlen += 5 + 16 + ((last + 32 + 16) & -16);
869
870 param->interleave = x4;
871
872 return (int)packlen;
873 } else
874 return -1; /* not yet */
875 }
876 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
877 {
878 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
879 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
880
881 return (int)tls1_1_multi_block_encrypt(key, param->out,
882 param->inp, param->len,
883 param->interleave / 4);
884 }
885 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
886 # endif
887 default:
888 return -1;
889 }
890 }
891
892 static EVP_CIPHER aesni_128_cbc_hmac_sha256_cipher = {
893 # ifdef NID_aes_128_cbc_hmac_sha256
894 NID_aes_128_cbc_hmac_sha256,
895 # else
896 NID_undef,
897 # endif
898 AES_BLOCK_SIZE, 16, AES_BLOCK_SIZE,
899 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
900 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
901 EVP_ORIG_GLOBAL,
902 aesni_cbc_hmac_sha256_init_key,
903 aesni_cbc_hmac_sha256_cipher,
904 NULL,
905 sizeof(EVP_AES_HMAC_SHA256),
906 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
907 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
908 aesni_cbc_hmac_sha256_ctrl,
909 NULL
910 };
911
912 static EVP_CIPHER aesni_256_cbc_hmac_sha256_cipher = {
913 # ifdef NID_aes_256_cbc_hmac_sha256
914 NID_aes_256_cbc_hmac_sha256,
915 # else
916 NID_undef,
917 # endif
918 AES_BLOCK_SIZE, 32, AES_BLOCK_SIZE,
919 EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 |
920 EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
921 EVP_ORIG_GLOBAL,
922 aesni_cbc_hmac_sha256_init_key,
923 aesni_cbc_hmac_sha256_cipher,
924 NULL,
925 sizeof(EVP_AES_HMAC_SHA256),
926 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv,
927 EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv,
928 aesni_cbc_hmac_sha256_ctrl,
929 NULL
930 };
931
EVP_aes_128_cbc_hmac_sha256(void)932 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
933 {
934 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
935 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
936 &aesni_128_cbc_hmac_sha256_cipher : NULL);
937 }
938
EVP_aes_256_cbc_hmac_sha256(void)939 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
940 {
941 return ((OPENSSL_ia32cap_P[1] & AESNI_CAPABLE) &&
942 aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL) ?
943 &aesni_256_cbc_hmac_sha256_cipher : NULL);
944 }
945 #else
EVP_aes_128_cbc_hmac_sha256(void)946 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
947 {
948 return NULL;
949 }
950
EVP_aes_256_cbc_hmac_sha256(void)951 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
952 {
953 return NULL;
954 }
955 #endif
956
