1 /*
2 * Copyright 2011-2021 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 * All 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 "cipher_aes_cbc_hmac_sha.h"
18
19 #if !defined(AES_CBC_HMAC_SHA_CAPABLE) || !defined(AESNI_CAPABLE)
ossl_cipher_capable_aes_cbc_hmac_sha256(void)20 int ossl_cipher_capable_aes_cbc_hmac_sha256(void)
21 {
22 return 0;
23 }
24
ossl_prov_cipher_hw_aes_cbc_hmac_sha256(void)25 const PROV_CIPHER_HW_AES_HMAC_SHA *ossl_prov_cipher_hw_aes_cbc_hmac_sha256(void)
26 {
27 return NULL;
28 }
29 #else
30
31 # include <openssl/rand.h>
32 # include "crypto/evp.h"
33 # include "internal/constant_time.h"
34
35 void sha256_block_data_order(void *c, const void *p, size_t len);
36 int aesni_cbc_sha256_enc(const void *inp, void *out, size_t blocks,
37 const AES_KEY *key, unsigned char iv[16],
38 SHA256_CTX *ctx, const void *in0);
39
ossl_cipher_capable_aes_cbc_hmac_sha256(void)40 int ossl_cipher_capable_aes_cbc_hmac_sha256(void)
41 {
42 return AESNI_CBC_HMAC_SHA_CAPABLE
43 && aesni_cbc_sha256_enc(NULL, NULL, 0, NULL, NULL, NULL, NULL);
44 }
45
aesni_cbc_hmac_sha256_init_key(PROV_CIPHER_CTX * vctx,const unsigned char * key,size_t keylen)46 static int aesni_cbc_hmac_sha256_init_key(PROV_CIPHER_CTX *vctx,
47 const unsigned char *key,
48 size_t keylen)
49 {
50 int ret;
51 PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx;
52 PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx;
53
54 if (ctx->base.enc)
55 ret = aesni_set_encrypt_key(key, ctx->base.keylen * 8, &ctx->ks);
56 else
57 ret = aesni_set_decrypt_key(key, ctx->base.keylen * 8, &ctx->ks);
58
59 SHA256_Init(&sctx->head); /* handy when benchmarking */
60 sctx->tail = sctx->head;
61 sctx->md = sctx->head;
62
63 ctx->payload_length = NO_PAYLOAD_LENGTH;
64
65 vctx->removetlspad = 1;
66 vctx->removetlsfixed = SHA256_DIGEST_LENGTH + AES_BLOCK_SIZE;
67
68 return ret < 0 ? 0 : 1;
69 }
70
71 void sha256_block_data_order(void *c, const void *p, size_t len);
72
sha256_update(SHA256_CTX * c,const void * data,size_t len)73 static void sha256_update(SHA256_CTX *c, const void *data, size_t len)
74 {
75 const unsigned char *ptr = data;
76 size_t res;
77
78 if ((res = c->num)) {
79 res = SHA256_CBLOCK - res;
80 if (len < res)
81 res = len;
82 SHA256_Update(c, ptr, res);
83 ptr += res;
84 len -= res;
85 }
86
87 res = len % SHA256_CBLOCK;
88 len -= res;
89
90 if (len) {
91 sha256_block_data_order(c, ptr, len / SHA256_CBLOCK);
92
93 ptr += len;
94 c->Nh += len >> 29;
95 c->Nl += len <<= 3;
96 if (c->Nl < (unsigned int)len)
97 c->Nh++;
98 }
99
100 if (res)
101 SHA256_Update(c, ptr, res);
102 }
103
104 # if !defined(OPENSSL_NO_MULTIBLOCK)
105
106 typedef struct {
107 unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8];
108 } SHA256_MB_CTX;
109
110 typedef struct {
111 const unsigned char *ptr;
112 int blocks;
113 } HASH_DESC;
114
115 typedef struct {
116 const unsigned char *inp;
117 unsigned char *out;
118 int blocks;
119 u64 iv[2];
120 } CIPH_DESC;
121
122 void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int);
123 void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int);
124
tls1_multi_block_encrypt(void * vctx,unsigned char * out,const unsigned char * inp,size_t inp_len,int n4x)125 static size_t tls1_multi_block_encrypt(void *vctx,
126 unsigned char *out,
127 const unsigned char *inp,
128 size_t inp_len, int n4x)
129 { /* n4x is 1 or 2 */
130 PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx;
131 PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx;
132 HASH_DESC hash_d[8], edges[8];
133 CIPH_DESC ciph_d[8];
134 unsigned char storage[sizeof(SHA256_MB_CTX) + 32];
135 union {
136 u64 q[16];
137 u32 d[32];
138 u8 c[128];
139 } blocks[8];
140 SHA256_MB_CTX *mctx;
141 unsigned int frag, last, packlen, i;
142 unsigned int x4 = 4 * n4x, minblocks, processed = 0;
143 size_t ret = 0;
144 u8 *IVs;
145 # if defined(BSWAP8)
146 u64 seqnum;
147 # endif
148
149 /* ask for IVs in bulk */
150 if (RAND_bytes_ex(ctx->base.libctx, (IVs = blocks[0].c), 16 * x4, 0) <= 0)
151 return 0;
152
153 mctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32)); /* align */
154
155 frag = (unsigned int)inp_len >> (1 + n4x);
156 last = (unsigned int)inp_len + frag - (frag << (1 + n4x));
157 if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) {
158 frag++;
159 last -= x4 - 1;
160 }
161
162 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
163
164 /* populate descriptors with pointers and IVs */
165 hash_d[0].ptr = inp;
166 ciph_d[0].inp = inp;
167 /* 5+16 is place for header and explicit IV */
168 ciph_d[0].out = out + 5 + 16;
169 memcpy(ciph_d[0].out - 16, IVs, 16);
170 memcpy(ciph_d[0].iv, IVs, 16);
171 IVs += 16;
172
173 for (i = 1; i < x4; i++) {
174 ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag;
175 ciph_d[i].out = ciph_d[i - 1].out + packlen;
176 memcpy(ciph_d[i].out - 16, IVs, 16);
177 memcpy(ciph_d[i].iv, IVs, 16);
178 IVs += 16;
179 }
180
181 # if defined(BSWAP8)
182 memcpy(blocks[0].c, sctx->md.data, 8);
183 seqnum = BSWAP8(blocks[0].q[0]);
184 # endif
185
186 for (i = 0; i < x4; i++) {
187 unsigned int len = (i == (x4 - 1) ? last : frag);
188 # if !defined(BSWAP8)
189 unsigned int carry, j;
190 # endif
191
192 mctx->A[i] = sctx->md.h[0];
193 mctx->B[i] = sctx->md.h[1];
194 mctx->C[i] = sctx->md.h[2];
195 mctx->D[i] = sctx->md.h[3];
196 mctx->E[i] = sctx->md.h[4];
197 mctx->F[i] = sctx->md.h[5];
198 mctx->G[i] = sctx->md.h[6];
199 mctx->H[i] = sctx->md.h[7];
200
201 /* fix seqnum */
202 # if defined(BSWAP8)
203 blocks[i].q[0] = BSWAP8(seqnum + i);
204 # else
205 for (carry = i, j = 8; j--;) {
206 blocks[i].c[j] = ((u8 *)sctx->md.data)[j] + carry;
207 carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1);
208 }
209 # endif
210 blocks[i].c[8] = ((u8 *)sctx->md.data)[8];
211 blocks[i].c[9] = ((u8 *)sctx->md.data)[9];
212 blocks[i].c[10] = ((u8 *)sctx->md.data)[10];
213 /* fix length */
214 blocks[i].c[11] = (u8)(len >> 8);
215 blocks[i].c[12] = (u8)(len);
216
217 memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13);
218 hash_d[i].ptr += 64 - 13;
219 hash_d[i].blocks = (len - (64 - 13)) / 64;
220
221 edges[i].ptr = blocks[i].c;
222 edges[i].blocks = 1;
223 }
224
225 /* hash 13-byte headers and first 64-13 bytes of inputs */
226 sha256_multi_block(mctx, edges, n4x);
227 /* hash bulk inputs */
228 # define MAXCHUNKSIZE 2048
229 # if MAXCHUNKSIZE%64
230 # error "MAXCHUNKSIZE is not divisible by 64"
231 # elif MAXCHUNKSIZE
232 /*
233 * goal is to minimize pressure on L1 cache by moving in shorter steps,
234 * so that hashed data is still in the cache by the time we encrypt it
235 */
236 minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64;
237 if (minblocks > MAXCHUNKSIZE / 64) {
238 for (i = 0; i < x4; i++) {
239 edges[i].ptr = hash_d[i].ptr;
240 edges[i].blocks = MAXCHUNKSIZE / 64;
241 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
242 }
243 do {
244 sha256_multi_block(mctx, edges, n4x);
245 aesni_multi_cbc_encrypt(ciph_d, &ctx->ks, n4x);
246
247 for (i = 0; i < x4; i++) {
248 edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE;
249 hash_d[i].blocks -= MAXCHUNKSIZE / 64;
250 edges[i].blocks = MAXCHUNKSIZE / 64;
251 ciph_d[i].inp += MAXCHUNKSIZE;
252 ciph_d[i].out += MAXCHUNKSIZE;
253 ciph_d[i].blocks = MAXCHUNKSIZE / 16;
254 memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16);
255 }
256 processed += MAXCHUNKSIZE;
257 minblocks -= MAXCHUNKSIZE / 64;
258 } while (minblocks > MAXCHUNKSIZE / 64);
259 }
260 # endif
261 # undef MAXCHUNKSIZE
262 sha256_multi_block(mctx, hash_d, n4x);
263
264 memset(blocks, 0, sizeof(blocks));
265 for (i = 0; i < x4; i++) {
266 unsigned int len = (i == (x4 - 1) ? last : frag),
267 off = hash_d[i].blocks * 64;
268 const unsigned char *ptr = hash_d[i].ptr + off;
269
270 off = (len - processed) - (64 - 13) - off; /* remainder actually */
271 memcpy(blocks[i].c, ptr, off);
272 blocks[i].c[off] = 0x80;
273 len += 64 + 13; /* 64 is HMAC header */
274 len *= 8; /* convert to bits */
275 if (off < (64 - 8)) {
276 # ifdef BSWAP4
277 blocks[i].d[15] = BSWAP4(len);
278 # else
279 PUTU32(blocks[i].c + 60, len);
280 # endif
281 edges[i].blocks = 1;
282 } else {
283 # ifdef BSWAP4
284 blocks[i].d[31] = BSWAP4(len);
285 # else
286 PUTU32(blocks[i].c + 124, len);
287 # endif
288 edges[i].blocks = 2;
289 }
290 edges[i].ptr = blocks[i].c;
291 }
292
293 /* hash input tails and finalize */
294 sha256_multi_block(mctx, edges, n4x);
295
296 memset(blocks, 0, sizeof(blocks));
297 for (i = 0; i < x4; i++) {
298 # ifdef BSWAP4
299 blocks[i].d[0] = BSWAP4(mctx->A[i]);
300 mctx->A[i] = sctx->tail.h[0];
301 blocks[i].d[1] = BSWAP4(mctx->B[i]);
302 mctx->B[i] = sctx->tail.h[1];
303 blocks[i].d[2] = BSWAP4(mctx->C[i]);
304 mctx->C[i] = sctx->tail.h[2];
305 blocks[i].d[3] = BSWAP4(mctx->D[i]);
306 mctx->D[i] = sctx->tail.h[3];
307 blocks[i].d[4] = BSWAP4(mctx->E[i]);
308 mctx->E[i] = sctx->tail.h[4];
309 blocks[i].d[5] = BSWAP4(mctx->F[i]);
310 mctx->F[i] = sctx->tail.h[5];
311 blocks[i].d[6] = BSWAP4(mctx->G[i]);
312 mctx->G[i] = sctx->tail.h[6];
313 blocks[i].d[7] = BSWAP4(mctx->H[i]);
314 mctx->H[i] = sctx->tail.h[7];
315 blocks[i].c[32] = 0x80;
316 blocks[i].d[15] = BSWAP4((64 + 32) * 8);
317 # else
318 PUTU32(blocks[i].c + 0, mctx->A[i]);
319 mctx->A[i] = sctx->tail.h[0];
320 PUTU32(blocks[i].c + 4, mctx->B[i]);
321 mctx->B[i] = sctx->tail.h[1];
322 PUTU32(blocks[i].c + 8, mctx->C[i]);
323 mctx->C[i] = sctx->tail.h[2];
324 PUTU32(blocks[i].c + 12, mctx->D[i]);
325 mctx->D[i] = sctx->tail.h[3];
326 PUTU32(blocks[i].c + 16, mctx->E[i]);
327 mctx->E[i] = sctx->tail.h[4];
328 PUTU32(blocks[i].c + 20, mctx->F[i]);
329 mctx->F[i] = sctx->tail.h[5];
330 PUTU32(blocks[i].c + 24, mctx->G[i]);
331 mctx->G[i] = sctx->tail.h[6];
332 PUTU32(blocks[i].c + 28, mctx->H[i]);
333 mctx->H[i] = sctx->tail.h[7];
334 blocks[i].c[32] = 0x80;
335 PUTU32(blocks[i].c + 60, (64 + 32) * 8);
336 # endif /* BSWAP */
337 edges[i].ptr = blocks[i].c;
338 edges[i].blocks = 1;
339 }
340
341 /* finalize MACs */
342 sha256_multi_block(mctx, edges, n4x);
343
344 for (i = 0; i < x4; i++) {
345 unsigned int len = (i == (x4 - 1) ? last : frag), pad, j;
346 unsigned char *out0 = out;
347
348 memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed);
349 ciph_d[i].inp = ciph_d[i].out;
350
351 out += 5 + 16 + len;
352
353 /* write MAC */
354 PUTU32(out + 0, mctx->A[i]);
355 PUTU32(out + 4, mctx->B[i]);
356 PUTU32(out + 8, mctx->C[i]);
357 PUTU32(out + 12, mctx->D[i]);
358 PUTU32(out + 16, mctx->E[i]);
359 PUTU32(out + 20, mctx->F[i]);
360 PUTU32(out + 24, mctx->G[i]);
361 PUTU32(out + 28, mctx->H[i]);
362 out += 32;
363 len += 32;
364
365 /* pad */
366 pad = 15 - len % 16;
367 for (j = 0; j <= pad; j++)
368 *(out++) = pad;
369 len += pad + 1;
370
371 ciph_d[i].blocks = (len - processed) / 16;
372 len += 16; /* account for explicit iv */
373
374 /* arrange header */
375 out0[0] = ((u8 *)sctx->md.data)[8];
376 out0[1] = ((u8 *)sctx->md.data)[9];
377 out0[2] = ((u8 *)sctx->md.data)[10];
378 out0[3] = (u8)(len >> 8);
379 out0[4] = (u8)(len);
380
381 ret += len + 5;
382 inp += frag;
383 }
384
385 aesni_multi_cbc_encrypt(ciph_d, &ctx->ks, n4x);
386
387 OPENSSL_cleanse(blocks, sizeof(blocks));
388 OPENSSL_cleanse(mctx, sizeof(*mctx));
389
390 ctx->multiblock_encrypt_len = ret;
391 return ret;
392 }
393 # endif /* !OPENSSL_NO_MULTIBLOCK */
394
aesni_cbc_hmac_sha256_cipher(PROV_CIPHER_CTX * vctx,unsigned char * out,const unsigned char * in,size_t len)395 static int aesni_cbc_hmac_sha256_cipher(PROV_CIPHER_CTX *vctx,
396 unsigned char *out,
397 const unsigned char *in, size_t len)
398 {
399 PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx;
400 PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx;
401 unsigned int l;
402 size_t plen = ctx->payload_length;
403 size_t iv = 0; /* explicit IV in TLS 1.1 and * later */
404 size_t aes_off = 0, blocks;
405 size_t sha_off = SHA256_CBLOCK - sctx->md.num;
406
407 ctx->payload_length = NO_PAYLOAD_LENGTH;
408
409 if (len % AES_BLOCK_SIZE)
410 return 0;
411
412 if (ctx->base.enc) {
413 if (plen == NO_PAYLOAD_LENGTH)
414 plen = len;
415 else if (len !=
416 ((plen + SHA256_DIGEST_LENGTH +
417 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE))
418 return 0;
419 else if (ctx->aux.tls_ver >= TLS1_1_VERSION)
420 iv = AES_BLOCK_SIZE;
421
422 /*
423 * Assembly stitch handles AVX-capable processors, but its
424 * performance is not optimal on AMD Jaguar, ~40% worse, for
425 * unknown reasons. Incidentally processor in question supports
426 * AVX, but not AMD-specific XOP extension, which can be used
427 * to identify it and avoid stitch invocation. So that after we
428 * establish that current CPU supports AVX, we even see if it's
429 * either even XOP-capable Bulldozer-based or GenuineIntel one.
430 * But SHAEXT-capable go ahead...
431 */
432 if (((OPENSSL_ia32cap_P[2] & (1 << 29)) || /* SHAEXT? */
433 ((OPENSSL_ia32cap_P[1] & (1 << (60 - 32))) && /* AVX? */
434 ((OPENSSL_ia32cap_P[1] & (1 << (43 - 32))) /* XOP? */
435 | (OPENSSL_ia32cap_P[0] & (1 << 30))))) && /* "Intel CPU"? */
436 plen > (sha_off + iv) &&
437 (blocks = (plen - (sha_off + iv)) / SHA256_CBLOCK)) {
438 sha256_update(&sctx->md, in + iv, sha_off);
439
440 (void)aesni_cbc_sha256_enc(in, out, blocks, &ctx->ks,
441 ctx->base.iv,
442 &sctx->md, in + iv + sha_off);
443 blocks *= SHA256_CBLOCK;
444 aes_off += blocks;
445 sha_off += blocks;
446 sctx->md.Nh += blocks >> 29;
447 sctx->md.Nl += blocks <<= 3;
448 if (sctx->md.Nl < (unsigned int)blocks)
449 sctx->md.Nh++;
450 } else {
451 sha_off = 0;
452 }
453 sha_off += iv;
454 sha256_update(&sctx->md, in + sha_off, plen - sha_off);
455
456 if (plen != len) { /* "TLS" mode of operation */
457 if (in != out)
458 memcpy(out + aes_off, in + aes_off, plen - aes_off);
459
460 /* calculate HMAC and append it to payload */
461 SHA256_Final(out + plen, &sctx->md);
462 sctx->md = sctx->tail;
463 sha256_update(&sctx->md, out + plen, SHA256_DIGEST_LENGTH);
464 SHA256_Final(out + plen, &sctx->md);
465
466 /* pad the payload|hmac */
467 plen += SHA256_DIGEST_LENGTH;
468 for (l = len - plen - 1; plen < len; plen++)
469 out[plen] = l;
470 /* encrypt HMAC|padding at once */
471 aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off,
472 &ctx->ks, ctx->base.iv, 1);
473 } else {
474 aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off,
475 &ctx->ks, ctx->base.iv, 1);
476 }
477 } else {
478 union {
479 unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)];
480 unsigned char c[64 + SHA256_DIGEST_LENGTH];
481 } mac, *pmac;
482
483 /* arrange cache line alignment */
484 pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64));
485
486 /* decrypt HMAC|padding at once */
487 aesni_cbc_encrypt(in, out, len, &ctx->ks,
488 ctx->base.iv, 0);
489
490 if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
491 size_t inp_len, mask, j, i;
492 unsigned int res, maxpad, pad, bitlen;
493 int ret = 1;
494 union {
495 unsigned int u[SHA_LBLOCK];
496 unsigned char c[SHA256_CBLOCK];
497 } *data = (void *)sctx->md.data;
498
499 if ((ctx->aux.tls_aad[plen - 4] << 8 | ctx->aux.tls_aad[plen - 3])
500 >= TLS1_1_VERSION)
501 iv = AES_BLOCK_SIZE;
502
503 if (len < (iv + SHA256_DIGEST_LENGTH + 1))
504 return 0;
505
506 /* omit explicit iv */
507 out += iv;
508 len -= iv;
509
510 /* figure out payload length */
511 pad = out[len - 1];
512 maxpad = len - (SHA256_DIGEST_LENGTH + 1);
513 maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8);
514 maxpad &= 255;
515
516 mask = constant_time_ge(maxpad, pad);
517 ret &= mask;
518 /*
519 * If pad is invalid then we will fail the above test but we must
520 * continue anyway because we are in constant time code. However,
521 * we'll use the maxpad value instead of the supplied pad to make
522 * sure we perform well defined pointer arithmetic.
523 */
524 pad = constant_time_select(mask, pad, maxpad);
525
526 inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1);
527
528 ctx->aux.tls_aad[plen - 2] = inp_len >> 8;
529 ctx->aux.tls_aad[plen - 1] = inp_len;
530
531 /* calculate HMAC */
532 sctx->md = sctx->head;
533 sha256_update(&sctx->md, ctx->aux.tls_aad, plen);
534
535 /* code with lucky-13 fix */
536 len -= SHA256_DIGEST_LENGTH; /* amend mac */
537 if (len >= (256 + SHA256_CBLOCK)) {
538 j = (len - (256 + SHA256_CBLOCK)) & (0 - SHA256_CBLOCK);
539 j += SHA256_CBLOCK - sctx->md.num;
540 sha256_update(&sctx->md, out, j);
541 out += j;
542 len -= j;
543 inp_len -= j;
544 }
545
546 /* but pretend as if we hashed padded payload */
547 bitlen = sctx->md.Nl + (inp_len << 3); /* at most 18 bits */
548 # ifdef BSWAP4
549 bitlen = BSWAP4(bitlen);
550 # else
551 mac.c[0] = 0;
552 mac.c[1] = (unsigned char)(bitlen >> 16);
553 mac.c[2] = (unsigned char)(bitlen >> 8);
554 mac.c[3] = (unsigned char)bitlen;
555 bitlen = mac.u[0];
556 # endif /* BSWAP */
557
558 pmac->u[0] = 0;
559 pmac->u[1] = 0;
560 pmac->u[2] = 0;
561 pmac->u[3] = 0;
562 pmac->u[4] = 0;
563 pmac->u[5] = 0;
564 pmac->u[6] = 0;
565 pmac->u[7] = 0;
566
567 for (res = sctx->md.num, j = 0; j < len; j++) {
568 size_t c = out[j];
569 mask = (j - inp_len) >> (sizeof(j) * 8 - 8);
570 c &= mask;
571 c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8));
572 data->c[res++] = (unsigned char)c;
573
574 if (res != SHA256_CBLOCK)
575 continue;
576
577 /* j is not incremented yet */
578 mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1));
579 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
580 sha256_block_data_order(&sctx->md, data, 1);
581 mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1));
582 pmac->u[0] |= sctx->md.h[0] & mask;
583 pmac->u[1] |= sctx->md.h[1] & mask;
584 pmac->u[2] |= sctx->md.h[2] & mask;
585 pmac->u[3] |= sctx->md.h[3] & mask;
586 pmac->u[4] |= sctx->md.h[4] & mask;
587 pmac->u[5] |= sctx->md.h[5] & mask;
588 pmac->u[6] |= sctx->md.h[6] & mask;
589 pmac->u[7] |= sctx->md.h[7] & mask;
590 res = 0;
591 }
592
593 for (i = res; i < SHA256_CBLOCK; i++, j++)
594 data->c[i] = 0;
595
596 if (res > SHA256_CBLOCK - 8) {
597 mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1));
598 data->u[SHA_LBLOCK - 1] |= bitlen & mask;
599 sha256_block_data_order(&sctx->md, data, 1);
600 mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
601 pmac->u[0] |= sctx->md.h[0] & mask;
602 pmac->u[1] |= sctx->md.h[1] & mask;
603 pmac->u[2] |= sctx->md.h[2] & mask;
604 pmac->u[3] |= sctx->md.h[3] & mask;
605 pmac->u[4] |= sctx->md.h[4] & mask;
606 pmac->u[5] |= sctx->md.h[5] & mask;
607 pmac->u[6] |= sctx->md.h[6] & mask;
608 pmac->u[7] |= sctx->md.h[7] & mask;
609
610 memset(data, 0, SHA256_CBLOCK);
611 j += 64;
612 }
613 data->u[SHA_LBLOCK - 1] = bitlen;
614 sha256_block_data_order(&sctx->md, data, 1);
615 mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1));
616 pmac->u[0] |= sctx->md.h[0] & mask;
617 pmac->u[1] |= sctx->md.h[1] & mask;
618 pmac->u[2] |= sctx->md.h[2] & mask;
619 pmac->u[3] |= sctx->md.h[3] & mask;
620 pmac->u[4] |= sctx->md.h[4] & mask;
621 pmac->u[5] |= sctx->md.h[5] & mask;
622 pmac->u[6] |= sctx->md.h[6] & mask;
623 pmac->u[7] |= sctx->md.h[7] & mask;
624
625 # ifdef BSWAP4
626 pmac->u[0] = BSWAP4(pmac->u[0]);
627 pmac->u[1] = BSWAP4(pmac->u[1]);
628 pmac->u[2] = BSWAP4(pmac->u[2]);
629 pmac->u[3] = BSWAP4(pmac->u[3]);
630 pmac->u[4] = BSWAP4(pmac->u[4]);
631 pmac->u[5] = BSWAP4(pmac->u[5]);
632 pmac->u[6] = BSWAP4(pmac->u[6]);
633 pmac->u[7] = BSWAP4(pmac->u[7]);
634 # else
635 for (i = 0; i < 8; i++) {
636 res = pmac->u[i];
637 pmac->c[4 * i + 0] = (unsigned char)(res >> 24);
638 pmac->c[4 * i + 1] = (unsigned char)(res >> 16);
639 pmac->c[4 * i + 2] = (unsigned char)(res >> 8);
640 pmac->c[4 * i + 3] = (unsigned char)res;
641 }
642 # endif /* BSWAP */
643 len += SHA256_DIGEST_LENGTH;
644 sctx->md = sctx->tail;
645 sha256_update(&sctx->md, pmac->c, SHA256_DIGEST_LENGTH);
646 SHA256_Final(pmac->c, &sctx->md);
647
648 /* verify HMAC */
649 out += inp_len;
650 len -= inp_len;
651 /* code containing lucky-13 fix */
652 {
653 unsigned char *p =
654 out + len - 1 - maxpad - SHA256_DIGEST_LENGTH;
655 size_t off = out - p;
656 unsigned int c, cmask;
657
658 for (res = 0, i = 0, j = 0;
659 j < maxpad + SHA256_DIGEST_LENGTH;
660 j++) {
661 c = p[j];
662 cmask =
663 ((int)(j - off - SHA256_DIGEST_LENGTH)) >>
664 (sizeof(int) * 8 - 1);
665 res |= (c ^ pad) & ~cmask; /* ... and padding */
666 cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1);
667 res |= (c ^ pmac->c[i]) & cmask;
668 i += 1 & cmask;
669 }
670
671 res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1));
672 ret &= (int)~res;
673 }
674 return ret;
675 } else {
676 sha256_update(&sctx->md, out, len);
677 }
678 }
679
680 return 1;
681 }
682
683 /* EVP_CTRL_AEAD_SET_MAC_KEY */
aesni_cbc_hmac_sha256_set_mac_key(void * vctx,const unsigned char * mackey,size_t len)684 static void aesni_cbc_hmac_sha256_set_mac_key(void *vctx,
685 const unsigned char *mackey,
686 size_t len)
687 {
688 PROV_AES_HMAC_SHA256_CTX *ctx = (PROV_AES_HMAC_SHA256_CTX *)vctx;
689 unsigned int i;
690 unsigned char hmac_key[64];
691
692 memset(hmac_key, 0, sizeof(hmac_key));
693
694 if (len > sizeof(hmac_key)) {
695 SHA256_Init(&ctx->head);
696 sha256_update(&ctx->head, mackey, len);
697 SHA256_Final(hmac_key, &ctx->head);
698 } else {
699 memcpy(hmac_key, mackey, len);
700 }
701
702 for (i = 0; i < sizeof(hmac_key); i++)
703 hmac_key[i] ^= 0x36; /* ipad */
704 SHA256_Init(&ctx->head);
705 sha256_update(&ctx->head, hmac_key, sizeof(hmac_key));
706
707 for (i = 0; i < sizeof(hmac_key); i++)
708 hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */
709 SHA256_Init(&ctx->tail);
710 sha256_update(&ctx->tail, hmac_key, sizeof(hmac_key));
711
712 OPENSSL_cleanse(hmac_key, sizeof(hmac_key));
713 }
714
715 /* EVP_CTRL_AEAD_TLS1_AAD */
aesni_cbc_hmac_sha256_set_tls1_aad(void * vctx,unsigned char * aad_rec,int aad_len)716 static int aesni_cbc_hmac_sha256_set_tls1_aad(void *vctx,
717 unsigned char *aad_rec, int aad_len)
718 {
719 PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx;
720 PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx;
721 unsigned char *p = aad_rec;
722 unsigned int len;
723
724 if (aad_len != EVP_AEAD_TLS1_AAD_LEN)
725 return -1;
726
727 len = p[aad_len - 2] << 8 | p[aad_len - 1];
728
729 if (ctx->base.enc) {
730 ctx->payload_length = len;
731 if ((ctx->aux.tls_ver =
732 p[aad_len - 4] << 8 | p[aad_len - 3]) >= TLS1_1_VERSION) {
733 if (len < AES_BLOCK_SIZE)
734 return 0;
735 len -= AES_BLOCK_SIZE;
736 p[aad_len - 2] = len >> 8;
737 p[aad_len - 1] = len;
738 }
739 sctx->md = sctx->head;
740 sha256_update(&sctx->md, p, aad_len);
741 ctx->tls_aad_pad = (int)(((len + SHA256_DIGEST_LENGTH +
742 AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)
743 - len);
744 return 1;
745 } else {
746 memcpy(ctx->aux.tls_aad, p, aad_len);
747 ctx->payload_length = aad_len;
748 ctx->tls_aad_pad = SHA256_DIGEST_LENGTH;
749 return 1;
750 }
751 }
752
753 # if !defined(OPENSSL_NO_MULTIBLOCK)
754 /* EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE */
aesni_cbc_hmac_sha256_tls1_multiblock_max_bufsize(void * vctx)755 static int aesni_cbc_hmac_sha256_tls1_multiblock_max_bufsize(
756 void *vctx)
757 {
758 PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx;
759
760 OPENSSL_assert(ctx->multiblock_max_send_fragment != 0);
761 return (int)(5 + 16
762 + (((int)ctx->multiblock_max_send_fragment + 32 + 16) & -16));
763 }
764
765 /* EVP_CTRL_TLS1_1_MULTIBLOCK_AAD */
aesni_cbc_hmac_sha256_tls1_multiblock_aad(void * vctx,EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM * param)766 static int aesni_cbc_hmac_sha256_tls1_multiblock_aad(
767 void *vctx, EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param)
768 {
769 PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx;
770 PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx;
771 unsigned int n4x = 1, x4;
772 unsigned int frag, last, packlen, inp_len;
773
774 inp_len = param->inp[11] << 8 | param->inp[12];
775
776 if (ctx->base.enc) {
777 if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION)
778 return -1;
779
780 if (inp_len) {
781 if (inp_len < 4096)
782 return 0; /* too short */
783
784 if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5))
785 n4x = 2; /* AVX2 */
786 } else if ((n4x = param->interleave / 4) && n4x <= 2)
787 inp_len = param->len;
788 else
789 return -1;
790
791 sctx->md = sctx->head;
792 sha256_update(&sctx->md, param->inp, 13);
793
794 x4 = 4 * n4x;
795 n4x += 1;
796
797 frag = inp_len >> n4x;
798 last = inp_len + frag - (frag << n4x);
799 if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) {
800 frag++;
801 last -= x4 - 1;
802 }
803
804 packlen = 5 + 16 + ((frag + 32 + 16) & -16);
805 packlen = (packlen << n4x) - packlen;
806 packlen += 5 + 16 + ((last + 32 + 16) & -16);
807
808 param->interleave = x4;
809 /* The returned values used by get need to be stored */
810 ctx->multiblock_interleave = x4;
811 ctx->multiblock_aad_packlen = packlen;
812 return 1;
813 }
814 return -1; /* not yet */
815 }
816
817 /* EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT */
aesni_cbc_hmac_sha256_tls1_multiblock_encrypt(void * ctx,EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM * param)818 static int aesni_cbc_hmac_sha256_tls1_multiblock_encrypt(
819 void *ctx, EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param)
820 {
821 return (int)tls1_multi_block_encrypt(ctx, param->out,
822 param->inp, param->len,
823 param->interleave / 4);
824 }
825 # endif
826
827 static const PROV_CIPHER_HW_AES_HMAC_SHA cipher_hw_aes_hmac_sha256 = {
828 {
829 aesni_cbc_hmac_sha256_init_key,
830 aesni_cbc_hmac_sha256_cipher
831 },
832 aesni_cbc_hmac_sha256_set_mac_key,
833 aesni_cbc_hmac_sha256_set_tls1_aad,
834 # if !defined(OPENSSL_NO_MULTIBLOCK)
835 aesni_cbc_hmac_sha256_tls1_multiblock_max_bufsize,
836 aesni_cbc_hmac_sha256_tls1_multiblock_aad,
837 aesni_cbc_hmac_sha256_tls1_multiblock_encrypt
838 # endif
839 };
840
ossl_prov_cipher_hw_aes_cbc_hmac_sha256(void)841 const PROV_CIPHER_HW_AES_HMAC_SHA *ossl_prov_cipher_hw_aes_cbc_hmac_sha256(void)
842 {
843 return &cipher_hw_aes_hmac_sha256;
844 }
845
846 #endif /* !defined(AES_CBC_HMAC_SHA_CAPABLE) || !defined(AESNI_CAPABLE) */
847
