]> git.ipfire.org Git - thirdparty/openssl.git/blob - crypto/evp/e_aes_cbc_hmac_sha256.c
projects / thirdparty / openssl.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Tell the ciphers which DRBG to use for generating random bytes.
[thirdparty/openssl.git] / crypto / evp / e_aes_cbc_hmac_sha256.c
1 /*
2 * Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved.
3 *
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
8 */
9
10 #include <openssl/opensslconf.h>
11
12 #include <stdio.h>
13 #include <string.h>
14
15
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 <internal/rand.h>
22 #include "modes_lcl.h"
23 #include "internal/constant_time_locl.h"
24 #include "internal/evp_int.h"
25 #include "evp_locl.h"
26
27 typedef struct {
28 AES_KEY ks;
29 SHA256_CTX head, tail, md;
30 size_t payload_length; /* AAD length in decrypt case */
31 union {
32 unsigned int tls_ver;
33 unsigned char tls_aad[16]; /* 13 used */
34 } aux;
35 } EVP_AES_HMAC_SHA256;
36
37 # define NO_PAYLOAD_LENGTH ((size_t)-1)
38
39 #if defined(AES_ASM) && ( \
40 defined(__x86_64) || defined(__x86_64__) || \
41 defined(_M_AMD64) || defined(_M_X64) )
42
43 extern unsigned int OPENSSL_ia32cap_P[];
44 # define AESNI_CAPABLE (1<<(57-32))
45
46 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
47 AES_KEY *key);
48 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
49 AES_KEY *key);
50
51 void aesni_cbc_encrypt(const unsigned char *in,
52 unsigned char *out,
53 size_t length,
54 const AES_KEY *key, unsigned char *ivec, int enc);
55
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);
59
60 # define data(ctx) ((EVP_AES_HMAC_SHA256 *)EVP_CIPHER_CTX_get_cipher_data(ctx))
61
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)
65 {
66 EVP_AES_HMAC_SHA256 *key = data(ctx);
67 int ret;
68
69 if (enc)
70 ret = aesni_set_encrypt_key(inkey,
71 EVP_CIPHER_CTX_key_length(ctx) * 8,
72 &key->ks);
73 else
74 ret = aesni_set_decrypt_key(inkey,
75 EVP_CIPHER_CTX_key_length(ctx) * 8,
76 &key->ks);
77
78 SHA256_Init(&key->head); /* handy when benchmarking */
79 key->tail = key->head;
80 key->md = key->head;
81
82 key->payload_length = NO_PAYLOAD_LENGTH;
83
84 return ret < 0 ? 0 : 1;
85 }
86
87 # define STITCHED_CALL
88
89 # if !defined(STITCHED_CALL)
90 # define aes_off 0
91 # endif
92
93 void sha256_block_data_order(void *c, const void *p, size_t len);
94
95 static void sha256_update(SHA256_CTX *c, const void *data, size_t len)
96 {
97 const unsigned char *ptr = data;
98 size_t res;
99
100 if ((res = c->num)) {
101 res = SHA256_CBLOCK - res;
102 if (len < res)
103 res = len;
104 SHA256_Update(c, ptr, res);
105 ptr += res;
106 len -= res;
107 }
108
109 res = len % SHA256_CBLOCK;
110 len -= res;
111
112 if (len) {
113 sha256_block_data_order(c, ptr, len / SHA256_CBLOCK);
114
115 ptr += len;
116 c->Nh += len >> 29;
117 c->Nl += len <<= 3;
118 if (c->Nl < (unsigned int)len)
119 c->Nh++;
120 }
121
122 if (res)
123 SHA256_Update(c, ptr, res);
124 }
125
126 # ifdef SHA256_Update
127 # undef SHA256_Update
128 # endif
129 # define SHA256_Update sha256_update
130
131 # if !defined(OPENSSL_NO_MULTIBLOCK)
132
133 typedef struct {
134 unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8];
135 } SHA256_MB_CTX;
136 typedef struct {
137 const unsigned char *ptr;
138 int blocks;
139 } HASH_DESC;
140
141 void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int);
142
143 typedef struct {
144 const unsigned char *inp;
145 unsigned char *out;
146 int blocks;
147 u64 iv[2];
148 } CIPH_DESC;
149
150 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
151
152 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
153 unsigned char *out,
154 const unsigned char *inp,
155 size_t inp_len, int n4x,
156 RAND_DRBG *drbg)
157 { /* n4x is 1 or 2 */
158 HASH_DESC hash_d[8], edges[8];
159 CIPH_DESC ciph_d[8];
160 unsigned char storage[sizeof(SHA256_MB_CTX) + 32];
161 union {
162 u64 q[16];
163 u32 d[32];
164 u8 c[128];
165 } blocks[8];
166 SHA256_MB_CTX *ctx;
167 unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed =
168 0;
169 size_t ret = 0;
170 u8 *IVs;
171 # if defined(BSWAP8)
172 u64 seqnum;
173 # endif
174
175 /* ask for IVs in bulk */
176 IVs = blocks[0].c;
177 if (drbg != NULL) {
178 if (RAND_DRBG_bytes(drbg, IVs, 16 * x4) == 0)
179 return 0;
180 } else if (RAND_bytes(IVs, 16 * x4) <= 0) {
181 return 0;
182 }
183
184 /* align */
185 ctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32));
186
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)) {
190 frag++;
191 last -= x4 - 1;
192 }
193
194 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
195
196 /* populate descriptors with pointers and IVs */
197 hash_d[0].ptr = inp;
198 ciph_d[0].inp = inp;
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);
203 IVs += 16;
204
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);
210 IVs += 16;
211 }
212
213 # if defined(BSWAP8)
214 memcpy(blocks[0].c, key->md.data, 8);
215 seqnum = BSWAP8(blocks[0].q[0]);
216 # endif
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;
221 # endif
222
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];
231
232 /* fix seqnum */
233 # if defined(BSWAP8)
234 blocks[i].q[0] = BSWAP8(seqnum + i);
235 # else
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);
239 }
240 # endif
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];
244 /* fix length */
245 blocks[i].c[11] = (u8)(len >> 8);
246 blocks[i].c[12] = (u8)(len);
247
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;
251
252 edges[i].ptr = blocks[i].c;
253 edges[i].blocks = 1;
254 }
255
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
260 # if MAXCHUNKSIZE%64
261 # error "MAXCHUNKSIZE is not divisible by 64"
262 # elif MAXCHUNKSIZE
263 /*
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
266 */
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;
273 }
274 do {
275 sha256_multi_block(ctx, edges, n4x);
276 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
277
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);
286 }
287 processed += MAXCHUNKSIZE;
288 minblocks -= MAXCHUNKSIZE / 64;
289 } while (minblocks > MAXCHUNKSIZE / 64);
290 }
291 # endif
292 # undef MAXCHUNKSIZE
293 sha256_multi_block(ctx, hash_d, n4x);
294
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;
300
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)) {
307 # ifdef BSWAP4
308 blocks[i].d[15] = BSWAP4(len);
309 # else
310 PUTU32(blocks[i].c + 60, len);
311 # endif
312 edges[i].blocks = 1;
313 } else {
314 # ifdef BSWAP4
315 blocks[i].d[31] = BSWAP4(len);
316 # else
317 PUTU32(blocks[i].c + 124, len);
318 # endif
319 edges[i].blocks = 2;
320 }
321 edges[i].ptr = blocks[i].c;
322 }
323
324 /* hash input tails and finalize */
325 sha256_multi_block(ctx, edges, n4x);
326
327 memset(blocks, 0, sizeof(blocks));
328 for (i = 0; i < x4; i++) {
329 # ifdef BSWAP4
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);
348 # else
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);
367 # endif
368 edges[i].ptr = blocks[i].c;
369 edges[i].blocks = 1;
370 }
371
372 /* finalize MACs */
373 sha256_multi_block(ctx, edges, n4x);
374
375 for (i = 0; i < x4; i++) {
376 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
377 unsigned char *out0 = out;
378
379 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
380 ciph_d[i].inp = ciph_d[i].out;
381
382 out += 5 + 16 + len;
383
384 /* write MAC */
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]);
393 out += 32;
394 len += 32;
395
396 /* pad */
397 pad = 15 - len % 16;
398 for (j = 0; j <= pad; j++)
399 *(out++) = pad;
400 len += pad + 1;
401
402 ciph_d[i].blocks = (len - processed) / 16;
403 len += 16; /* account for explicit iv */
404
405 /* arrange header */
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);
410 out0[4] = (u8)(len);
411
412 ret += len + 5;
413 inp += frag;
414 }
415
416 aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x);
417
418 OPENSSL_cleanse(blocks, sizeof(blocks));
419 OPENSSL_cleanse(ctx, sizeof(*ctx));
420
421 return ret;
422 }
423 # endif
424
425 static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx,
426 unsigned char *out,
427 const unsigned char *in, size_t len)
428 {
429 EVP_AES_HMAC_SHA256 *key = data(ctx);
430 unsigned int l;
431 size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and
432 * later */
433 sha_off = 0;
434 # if defined(STITCHED_CALL)
435 size_t aes_off = 0, blocks;
436
437 sha_off = SHA256_CBLOCK - key->md.num;
438 # endif
439
440 key->payload_length = NO_PAYLOAD_LENGTH;
441
442 if (len % AES_BLOCK_SIZE)
443 return 0;
444
445 if (EVP_CIPHER_CTX_encrypting(ctx)) {
446 if (plen == NO_PAYLOAD_LENGTH)
447 plen = len;
448 else if (len !=
449 ((plen + SHA256_DIGEST_LENGTH +
450 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
451 return 0;
452 else if (key->aux.tls_ver >= TLS1_1_VERSION)
453 iv = AES_BLOCK_SIZE;
454
455 # if defined(STITCHED_CALL)
456 /*
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...
465 */
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);
473
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;
478 aes_off += blocks;
479 sha_off += blocks;
480 key->md.Nh += blocks >> 29;
481 key->md.Nl += blocks <<= 3;
482 if (key->md.Nl < (unsigned int)blocks)
483 key->md.Nh++;
484 } else {
485 sha_off = 0;
486 }
487 # endif
488 sha_off += iv;
489 SHA256_Update(&key->md, in + sha_off, plen - sha_off);
490
491 if (plen != len) { /* "TLS" mode of operation */
492 if (in != out)
493 memcpy(out + aes_off, in + aes_off, plen - aes_off);
494
495 /* calculate HMAC and append it to payload */
496 SHA256_Final(out + plen, &key->md);
497 key->md = key->tail;
498 SHA256_Update(&key->md, out + plen, SHA256_DIGEST_LENGTH);
499 SHA256_Final(out + plen, &key->md);
500
501 /* pad the payload|hmac */
502 plen += SHA256_DIGEST_LENGTH;
503 for (l = len - plen - 1; plen < len; plen++)
504 out[plen] = l;
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);
508 } else {
509 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
510 &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1);
511 }
512 } else {
513 union {
514 unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)];
515 unsigned char c[64 + SHA256_DIGEST_LENGTH];
516 } mac, *pmac;
517
518 /* arrange cache line alignment */
519 pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64));
520
521 /* decrypt HMAC|padding at once */
522 aesni_cbc_encrypt(in, out, len, &key->ks,
523 EVP_CIPHER_CTX_iv_noconst(ctx), 0);
524
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;
528 int ret = 1;
529 union {
530 unsigned int u[SHA_LBLOCK];
531 unsigned char c[SHA256_CBLOCK];
532 } *data = (void *)key->md.data;
533
534 if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3])
535 >= TLS1_1_VERSION)
536 iv = AES_BLOCK_SIZE;
537
538 if (len < (iv + SHA256_DIGEST_LENGTH + 1))
539 return 0;
540
541 /* omit explicit iv */
542 out += iv;
543 len -= iv;
544
545 /* figure out payload length */
546 pad = out[len - 1];
547 maxpad = len - (SHA256_DIGEST_LENGTH + 1);
548 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
549 maxpad &= 255;
550
551 mask = constant_time_ge(maxpad, pad);
552 ret &= mask;
553 /*
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.
558 */
559 pad = constant_time_select(mask, pad, maxpad);
560
561 inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1);
562
563 key->aux.tls_aad[plen - 2] = inp_len >> 8;
564 key->aux.tls_aad[plen - 1] = inp_len;
565
566 /* calculate HMAC */
567 key->md = key->head;
568 SHA256_Update(&key->md, key->aux.tls_aad, plen);
569
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);
576 out += j;
577 len -= j;
578 inp_len -= j;
579 }
580
581 /* but pretend as if we hashed padded payload */
582 bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */
583 # ifdef BSWAP4
584 bitlen = BSWAP4(bitlen);
585 # else
586 mac.c[0] = 0;
587 mac.c[1] = (unsigned char)(bitlen >> 16);
588 mac.c[2] = (unsigned char)(bitlen >> 8);
589 mac.c[3] = (unsigned char)bitlen;
590 bitlen = mac.u[0];
591 # endif
592
593 pmac->u[0] = 0;
594 pmac->u[1] = 0;
595 pmac->u[2] = 0;
596 pmac->u[3] = 0;
597 pmac->u[4] = 0;
598 pmac->u[5] = 0;
599 pmac->u[6] = 0;
600 pmac->u[7] = 0;
601
602 for (res = key->md.num, j = 0; j < len; j++) {
603 size_t c = out[j];
604 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
605 c &= mask;
606 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
607 data->c[res++] = (unsigned char)c;
608
609 if (res != SHA256_CBLOCK)
610 continue;
611
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;
625 res = 0;
626 }
627
628 for (i = res; i < SHA256_CBLOCK; i++, j++)
629 data->c[i] = 0;
630
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;
644
645 memset(data, 0, SHA256_CBLOCK);
646 j += 64;
647 }
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;
659
660 # ifdef BSWAP4
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]);
669 # else
670 for (i = 0; i < 8; i++) {
671 res = pmac->u[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;
676 }
677 # endif
678 len += SHA256_DIGEST_LENGTH;
679 # else
680 SHA256_Update(&key->md, out, inp_len);
681 res = key->md.num;
682 SHA256_Final(pmac->c, &key->md);
683
684 {
685 unsigned int inp_blocks, pad_blocks;
686
687 /* but pretend as if we hashed padded payload */
688 inp_blocks =
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;
693 pad_blocks +=
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);
697 }
698 # endif /* pre-lucky-13 reference version of above */
699 key->md = key->tail;
700 SHA256_Update(&key->md, pmac->c, SHA256_DIGEST_LENGTH);
701 SHA256_Final(pmac->c, &key->md);
702
703 /* verify HMAC */
704 out += inp_len;
705 len -= inp_len;
706 # if 1 /* see original reference version in #else */
707 {
708 unsigned char *p =
709 out + len - 1 - maxpad - SHA256_DIGEST_LENGTH;
710 size_t off = out - p;
711 unsigned int c, cmask;
712
713 maxpad += SHA256_DIGEST_LENGTH;
714 for (res = 0, i = 0, j = 0; j < maxpad; j++) {
715 c = p[j];
716 cmask =
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;
722 i += 1 & cmask;
723 }
724 maxpad -= SHA256_DIGEST_LENGTH;
725
726 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
727 ret &= (int)~res;
728 }
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));
733 ret &= (int)~res;
734
735 /* verify padding */
736 pad = (pad & ~res) | (maxpad & res);
737 out = out + len - 1 - pad;
738 for (res = 0, i = 0; i < pad; i++)
739 res |= out[i] ^ pad;
740
741 res = (0 - res) >> (sizeof(res) * 8 - 1);
742 ret &= (int)~res;
743 # endif
744 return ret;
745 } else {
746 SHA256_Update(&key->md, out, len);
747 }
748 }
749
750 return 1;
751 }
752
753 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg,
754 void *ptr)
755 {
756 EVP_AES_HMAC_SHA256 *key = data(ctx);
757 unsigned int u_arg = (unsigned int)arg;
758
759 switch (type) {
760 case EVP_CTRL_AEAD_SET_MAC_KEY:
761 {
762 unsigned int i;
763 unsigned char hmac_key[64];
764
765 memset(hmac_key, 0, sizeof(hmac_key));
766
767 if (arg < 0)
768 return -1;
769
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);
774 } else {
775 memcpy(hmac_key, ptr, arg);
776 }
777
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));
782
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));
787
788 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
789
790 return 1;
791 }
792 case EVP_CTRL_AEAD_TLS1_AAD:
793 {
794 unsigned char *p = ptr;
795 unsigned int len;
796
797 if (arg != EVP_AEAD_TLS1_AAD_LEN)
798 return -1;
799
800 len = p[arg - 2] << 8 | p[arg - 1];
801
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)
807 return 0;
808 len -= AES_BLOCK_SIZE;
809 p[arg - 2] = len >> 8;
810 p[arg - 1] = len;
811 }
812 key->md = key->head;
813 SHA256_Update(&key->md, p, arg);
814
815 return (int)(((len + SHA256_DIGEST_LENGTH +
816 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
817 - len);
818 } else {
819 memcpy(key->aux.tls_aad, ptr, arg);
820 key->payload_length = arg;
821
822 return SHA256_DIGEST_LENGTH;
823 }
824 }
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:
829 {
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;
834
835 if (arg < 0)
836 return -1;
837
838 if (u_arg < sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM))
839 return -1;
840
841 inp_len = param->inp[11] << 8 | param->inp[12];
842
843 if (EVP_CIPHER_CTX_encrypting(ctx)) {
844 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
845 return -1;
846
847 if (inp_len) {
848 if (inp_len < 4096)
849 return 0; /* too short */
850
851 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
852 n4x = 2; /* AVX2 */
853 } else if ((n4x = param->interleave / 4) && n4x <= 2)
854 inp_len = param->len;
855 else
856 return -1;
857
858 key->md = key->head;
859 SHA256_Update(&key->md, param->inp, 13);
860
861 x4 = 4 * n4x;
862 n4x += 1;
863
864 frag = inp_len >> n4x;
865 last = inp_len + frag - (frag << n4x);
866 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
867 frag++;
868 last -= x4 - 1;
869 }
870
871 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
872 packlen = (packlen << n4x) - packlen;
873 packlen += 5 + 16 + ((last + 32 + 16) & -16);
874
875 param->interleave = x4;
876
877 return (int)packlen;
878 } else
879 return -1; /* not yet */
880 }
881 case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
882 {
883 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
884 (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr;
885
886 return (int)tls1_1_multi_block_encrypt(key, param->out,
887 param->inp, param->len,
888 param->interleave / 4,
889 ctx->drbg);
890 }
891 case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
892 # endif
893 default:
894 return -1;
895 }
896 }
897
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,
901 # else
902 NID_undef,
903 # endif
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,
909 NULL,
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,
914 NULL
915 };
916
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,
920 # else
921 NID_undef,
922 # endif
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,
928 NULL,
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,
933 NULL
934 };
935
936 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
937 {
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);
941 }
942
943 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
944 {
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);
948 }
949 #else
950 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
951 {
952 return NULL;
953 }
954
955 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
956 {
957 return NULL;
958 }
959 #endif
A mirror of the official openssl repository