1 | /* |
2 | * Copyright 2011-2016 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 | #include <stdio.h> |
11 | #include <string.h> |
12 | #include <openssl/opensslconf.h> |
13 | #include <openssl/evp.h> |
14 | #include <openssl/objects.h> |
15 | #include <openssl/aes.h> |
16 | #include <openssl/sha.h> |
17 | #include <openssl/rand.h> |
18 | #include "internal/cryptlib.h" |
19 | #include "crypto/modes.h" |
20 | #include "crypto/evp.h" |
21 | #include "internal/constant_time.h" |
22 | |
23 | typedef struct { |
24 | AES_KEY ks; |
25 | SHA_CTX head, tail, md; |
26 | size_t payload_length; /* AAD length in decrypt case */ |
27 | union { |
28 | unsigned int tls_ver; |
29 | unsigned char tls_aad[16]; /* 13 used */ |
30 | } aux; |
31 | } EVP_AES_HMAC_SHA1; |
32 | |
33 | #define NO_PAYLOAD_LENGTH ((size_t)-1) |
34 | |
35 | #if defined(AES_ASM) && ( \ |
36 | defined(__x86_64) || defined(__x86_64__) || \ |
37 | defined(_M_AMD64) || defined(_M_X64) ) |
38 | |
39 | # define AESNI_CAPABLE (1<<(57-32)) |
40 | |
41 | int aesni_set_encrypt_key(const unsigned char *userKey, int bits, |
42 | AES_KEY *key); |
43 | int aesni_set_decrypt_key(const unsigned char *userKey, int bits, |
44 | AES_KEY *key); |
45 | |
46 | void aesni_cbc_encrypt(const unsigned char *in, |
47 | unsigned char *out, |
48 | size_t length, |
49 | const AES_KEY *key, unsigned char *ivec, int enc); |
50 | |
51 | void aesni_cbc_sha1_enc(const void *inp, void *out, size_t blocks, |
52 | const AES_KEY *key, unsigned char iv[16], |
53 | SHA_CTX *ctx, const void *in0); |
54 | |
55 | void aesni256_cbc_sha1_dec(const void *inp, void *out, size_t blocks, |
56 | const AES_KEY *key, unsigned char iv[16], |
57 | SHA_CTX *ctx, const void *in0); |
58 | |
59 | # define data(ctx) ((EVP_AES_HMAC_SHA1 *)EVP_CIPHER_CTX_get_cipher_data(ctx)) |
60 | |
61 | static int aesni_cbc_hmac_sha1_init_key(EVP_CIPHER_CTX *ctx, |
62 | const unsigned char *inkey, |
63 | const unsigned char *iv, int enc) |
64 | { |
65 | EVP_AES_HMAC_SHA1 *key = data(ctx); |
66 | int ret; |
67 | |
68 | if (enc) |
69 | ret = aesni_set_encrypt_key(inkey, |
70 | EVP_CIPHER_CTX_key_length(ctx) * 8, |
71 | &key->ks); |
72 | else |
73 | ret = aesni_set_decrypt_key(inkey, |
74 | EVP_CIPHER_CTX_key_length(ctx) * 8, |
75 | &key->ks); |
76 | |
77 | SHA1_Init(&key->head); /* handy when benchmarking */ |
78 | key->tail = key->head; |
79 | key->md = key->head; |
80 | |
81 | key->payload_length = NO_PAYLOAD_LENGTH; |
82 | |
83 | return ret < 0 ? 0 : 1; |
84 | } |
85 | |
86 | # define STITCHED_CALL |
87 | # undef STITCHED_DECRYPT_CALL |
88 | |
89 | # if !defined(STITCHED_CALL) |
90 | # define aes_off 0 |
91 | # endif |
92 | |
93 | void sha1_block_data_order(void *c, const void *p, size_t len); |
94 | |
95 | static void sha1_update(SHA_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 = SHA_CBLOCK - res; |
102 | if (len < res) |
103 | res = len; |
104 | SHA1_Update(c, ptr, res); |
105 | ptr += res; |
106 | len -= res; |
107 | } |
108 | |
109 | res = len % SHA_CBLOCK; |
110 | len -= res; |
111 | |
112 | if (len) { |
113 | sha1_block_data_order(c, ptr, len / SHA_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 | SHA1_Update(c, ptr, res); |
124 | } |
125 | |
126 | # ifdef SHA1_Update |
127 | # undef SHA1_Update |
128 | # endif |
129 | # define SHA1_Update sha1_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]; |
135 | } SHA1_MB_CTX; |
136 | typedef struct { |
137 | const unsigned char *ptr; |
138 | int blocks; |
139 | } HASH_DESC; |
140 | |
141 | void sha1_multi_block(SHA1_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_SHA1 *key, |
153 | unsigned char *out, |
154 | const unsigned char *inp, |
155 | size_t inp_len, int n4x) |
156 | { /* n4x is 1 or 2 */ |
157 | HASH_DESC hash_d[8], edges[8]; |
158 | CIPH_DESC ciph_d[8]; |
159 | unsigned char storage[sizeof(SHA1_MB_CTX) + 32]; |
160 | union { |
161 | u64 q[16]; |
162 | u32 d[32]; |
163 | u8 c[128]; |
164 | } blocks[8]; |
165 | SHA1_MB_CTX *ctx; |
166 | unsigned int frag, last, packlen, i, x4 = 4 * n4x, minblocks, processed = |
167 | 0; |
168 | size_t ret = 0; |
169 | u8 *IVs; |
170 | # if defined(BSWAP8) |
171 | u64 seqnum; |
172 | # endif |
173 | |
174 | /* ask for IVs in bulk */ |
175 | if (RAND_bytes((IVs = blocks[0].c), 16 * x4) <= 0) |
176 | return 0; |
177 | |
178 | ctx = (SHA1_MB_CTX *) (storage + 32 - ((size_t)storage % 32)); /* align */ |
179 | |
180 | frag = (unsigned int)inp_len >> (1 + n4x); |
181 | last = (unsigned int)inp_len + frag - (frag << (1 + n4x)); |
182 | if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) { |
183 | frag++; |
184 | last -= x4 - 1; |
185 | } |
186 | |
187 | packlen = 5 + 16 + ((frag + 20 + 16) & -16); |
188 | |
189 | /* populate descriptors with pointers and IVs */ |
190 | hash_d[0].ptr = inp; |
191 | ciph_d[0].inp = inp; |
192 | /* 5+16 is place for header and explicit IV */ |
193 | ciph_d[0].out = out + 5 + 16; |
194 | memcpy(ciph_d[0].out - 16, IVs, 16); |
195 | memcpy(ciph_d[0].iv, IVs, 16); |
196 | IVs += 16; |
197 | |
198 | for (i = 1; i < x4; i++) { |
199 | ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag; |
200 | ciph_d[i].out = ciph_d[i - 1].out + packlen; |
201 | memcpy(ciph_d[i].out - 16, IVs, 16); |
202 | memcpy(ciph_d[i].iv, IVs, 16); |
203 | IVs += 16; |
204 | } |
205 | |
206 | # if defined(BSWAP8) |
207 | memcpy(blocks[0].c, key->md.data, 8); |
208 | seqnum = BSWAP8(blocks[0].q[0]); |
209 | # endif |
210 | for (i = 0; i < x4; i++) { |
211 | unsigned int len = (i == (x4 - 1) ? last : frag); |
212 | # if !defined(BSWAP8) |
213 | unsigned int carry, j; |
214 | # endif |
215 | |
216 | ctx->A[i] = key->md.h0; |
217 | ctx->B[i] = key->md.h1; |
218 | ctx->C[i] = key->md.h2; |
219 | ctx->D[i] = key->md.h3; |
220 | ctx->E[i] = key->md.h4; |
221 | |
222 | /* fix seqnum */ |
223 | # if defined(BSWAP8) |
224 | blocks[i].q[0] = BSWAP8(seqnum + i); |
225 | # else |
226 | for (carry = i, j = 8; j--;) { |
227 | blocks[i].c[j] = ((u8 *)key->md.data)[j] + carry; |
228 | carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1); |
229 | } |
230 | # endif |
231 | blocks[i].c[8] = ((u8 *)key->md.data)[8]; |
232 | blocks[i].c[9] = ((u8 *)key->md.data)[9]; |
233 | blocks[i].c[10] = ((u8 *)key->md.data)[10]; |
234 | /* fix length */ |
235 | blocks[i].c[11] = (u8)(len >> 8); |
236 | blocks[i].c[12] = (u8)(len); |
237 | |
238 | memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13); |
239 | hash_d[i].ptr += 64 - 13; |
240 | hash_d[i].blocks = (len - (64 - 13)) / 64; |
241 | |
242 | edges[i].ptr = blocks[i].c; |
243 | edges[i].blocks = 1; |
244 | } |
245 | |
246 | /* hash 13-byte headers and first 64-13 bytes of inputs */ |
247 | sha1_multi_block(ctx, edges, n4x); |
248 | /* hash bulk inputs */ |
249 | # define MAXCHUNKSIZE 2048 |
250 | # if MAXCHUNKSIZE%64 |
251 | # error "MAXCHUNKSIZE is not divisible by 64" |
252 | # elif MAXCHUNKSIZE |
253 | /* |
254 | * goal is to minimize pressure on L1 cache by moving in shorter steps, |
255 | * so that hashed data is still in the cache by the time we encrypt it |
256 | */ |
257 | minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64; |
258 | if (minblocks > MAXCHUNKSIZE / 64) { |
259 | for (i = 0; i < x4; i++) { |
260 | edges[i].ptr = hash_d[i].ptr; |
261 | edges[i].blocks = MAXCHUNKSIZE / 64; |
262 | ciph_d[i].blocks = MAXCHUNKSIZE / 16; |
263 | } |
264 | do { |
265 | sha1_multi_block(ctx, edges, n4x); |
266 | aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x); |
267 | |
268 | for (i = 0; i < x4; i++) { |
269 | edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE; |
270 | hash_d[i].blocks -= MAXCHUNKSIZE / 64; |
271 | edges[i].blocks = MAXCHUNKSIZE / 64; |
272 | ciph_d[i].inp += MAXCHUNKSIZE; |
273 | ciph_d[i].out += MAXCHUNKSIZE; |
274 | ciph_d[i].blocks = MAXCHUNKSIZE / 16; |
275 | memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16); |
276 | } |
277 | processed += MAXCHUNKSIZE; |
278 | minblocks -= MAXCHUNKSIZE / 64; |
279 | } while (minblocks > MAXCHUNKSIZE / 64); |
280 | } |
281 | # endif |
282 | # undef MAXCHUNKSIZE |
283 | sha1_multi_block(ctx, hash_d, n4x); |
284 | |
285 | memset(blocks, 0, sizeof(blocks)); |
286 | for (i = 0; i < x4; i++) { |
287 | unsigned int len = (i == (x4 - 1) ? last : frag), |
288 | off = hash_d[i].blocks * 64; |
289 | const unsigned char *ptr = hash_d[i].ptr + off; |
290 | |
291 | off = (len - processed) - (64 - 13) - off; /* remainder actually */ |
292 | memcpy(blocks[i].c, ptr, off); |
293 | blocks[i].c[off] = 0x80; |
294 | len += 64 + 13; /* 64 is HMAC header */ |
295 | len *= 8; /* convert to bits */ |
296 | if (off < (64 - 8)) { |
297 | # ifdef BSWAP4 |
298 | blocks[i].d[15] = BSWAP4(len); |
299 | # else |
300 | PUTU32(blocks[i].c + 60, len); |
301 | # endif |
302 | edges[i].blocks = 1; |
303 | } else { |
304 | # ifdef BSWAP4 |
305 | blocks[i].d[31] = BSWAP4(len); |
306 | # else |
307 | PUTU32(blocks[i].c + 124, len); |
308 | # endif |
309 | edges[i].blocks = 2; |
310 | } |
311 | edges[i].ptr = blocks[i].c; |
312 | } |
313 | |
314 | /* hash input tails and finalize */ |
315 | sha1_multi_block(ctx, edges, n4x); |
316 | |
317 | memset(blocks, 0, sizeof(blocks)); |
318 | for (i = 0; i < x4; i++) { |
319 | # ifdef BSWAP4 |
320 | blocks[i].d[0] = BSWAP4(ctx->A[i]); |
321 | ctx->A[i] = key->tail.h0; |
322 | blocks[i].d[1] = BSWAP4(ctx->B[i]); |
323 | ctx->B[i] = key->tail.h1; |
324 | blocks[i].d[2] = BSWAP4(ctx->C[i]); |
325 | ctx->C[i] = key->tail.h2; |
326 | blocks[i].d[3] = BSWAP4(ctx->D[i]); |
327 | ctx->D[i] = key->tail.h3; |
328 | blocks[i].d[4] = BSWAP4(ctx->E[i]); |
329 | ctx->E[i] = key->tail.h4; |
330 | blocks[i].c[20] = 0x80; |
331 | blocks[i].d[15] = BSWAP4((64 + 20) * 8); |
332 | # else |
333 | PUTU32(blocks[i].c + 0, ctx->A[i]); |
334 | ctx->A[i] = key->tail.h0; |
335 | PUTU32(blocks[i].c + 4, ctx->B[i]); |
336 | ctx->B[i] = key->tail.h1; |
337 | PUTU32(blocks[i].c + 8, ctx->C[i]); |
338 | ctx->C[i] = key->tail.h2; |
339 | PUTU32(blocks[i].c + 12, ctx->D[i]); |
340 | ctx->D[i] = key->tail.h3; |
341 | PUTU32(blocks[i].c + 16, ctx->E[i]); |
342 | ctx->E[i] = key->tail.h4; |
343 | blocks[i].c[20] = 0x80; |
344 | PUTU32(blocks[i].c + 60, (64 + 20) * 8); |
345 | # endif |
346 | edges[i].ptr = blocks[i].c; |
347 | edges[i].blocks = 1; |
348 | } |
349 | |
350 | /* finalize MACs */ |
351 | sha1_multi_block(ctx, edges, n4x); |
352 | |
353 | for (i = 0; i < x4; i++) { |
354 | unsigned int len = (i == (x4 - 1) ? last : frag), pad, j; |
355 | unsigned char *out0 = out; |
356 | |
357 | memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed); |
358 | ciph_d[i].inp = ciph_d[i].out; |
359 | |
360 | out += 5 + 16 + len; |
361 | |
362 | /* write MAC */ |
363 | PUTU32(out + 0, ctx->A[i]); |
364 | PUTU32(out + 4, ctx->B[i]); |
365 | PUTU32(out + 8, ctx->C[i]); |
366 | PUTU32(out + 12, ctx->D[i]); |
367 | PUTU32(out + 16, ctx->E[i]); |
368 | out += 20; |
369 | len += 20; |
370 | |
371 | /* pad */ |
372 | pad = 15 - len % 16; |
373 | for (j = 0; j <= pad; j++) |
374 | *(out++) = pad; |
375 | len += pad + 1; |
376 | |
377 | ciph_d[i].blocks = (len - processed) / 16; |
378 | len += 16; /* account for explicit iv */ |
379 | |
380 | /* arrange header */ |
381 | out0[0] = ((u8 *)key->md.data)[8]; |
382 | out0[1] = ((u8 *)key->md.data)[9]; |
383 | out0[2] = ((u8 *)key->md.data)[10]; |
384 | out0[3] = (u8)(len >> 8); |
385 | out0[4] = (u8)(len); |
386 | |
387 | ret += len + 5; |
388 | inp += frag; |
389 | } |
390 | |
391 | aesni_multi_cbc_encrypt(ciph_d, &key->ks, n4x); |
392 | |
393 | OPENSSL_cleanse(blocks, sizeof(blocks)); |
394 | OPENSSL_cleanse(ctx, sizeof(*ctx)); |
395 | |
396 | return ret; |
397 | } |
398 | # endif |
399 | |
400 | static int aesni_cbc_hmac_sha1_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out, |
401 | const unsigned char *in, size_t len) |
402 | { |
403 | EVP_AES_HMAC_SHA1 *key = data(ctx); |
404 | unsigned int l; |
405 | size_t plen = key->payload_length, iv = 0, /* explicit IV in TLS 1.1 and |
406 | * later */ |
407 | sha_off = 0; |
408 | # if defined(STITCHED_CALL) |
409 | size_t aes_off = 0, blocks; |
410 | |
411 | sha_off = SHA_CBLOCK - key->md.num; |
412 | # endif |
413 | |
414 | key->payload_length = NO_PAYLOAD_LENGTH; |
415 | |
416 | if (len % AES_BLOCK_SIZE) |
417 | return 0; |
418 | |
419 | if (EVP_CIPHER_CTX_encrypting(ctx)) { |
420 | if (plen == NO_PAYLOAD_LENGTH) |
421 | plen = len; |
422 | else if (len != |
423 | ((plen + SHA_DIGEST_LENGTH + |
424 | AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)) |
425 | return 0; |
426 | else if (key->aux.tls_ver >= TLS1_1_VERSION) |
427 | iv = AES_BLOCK_SIZE; |
428 | |
429 | # if defined(STITCHED_CALL) |
430 | if (plen > (sha_off + iv) |
431 | && (blocks = (plen - (sha_off + iv)) / SHA_CBLOCK)) { |
432 | SHA1_Update(&key->md, in + iv, sha_off); |
433 | |
434 | aesni_cbc_sha1_enc(in, out, blocks, &key->ks, |
435 | EVP_CIPHER_CTX_iv_noconst(ctx), |
436 | &key->md, in + iv + sha_off); |
437 | blocks *= SHA_CBLOCK; |
438 | aes_off += blocks; |
439 | sha_off += blocks; |
440 | key->md.Nh += blocks >> 29; |
441 | key->md.Nl += blocks <<= 3; |
442 | if (key->md.Nl < (unsigned int)blocks) |
443 | key->md.Nh++; |
444 | } else { |
445 | sha_off = 0; |
446 | } |
447 | # endif |
448 | sha_off += iv; |
449 | SHA1_Update(&key->md, in + sha_off, plen - sha_off); |
450 | |
451 | if (plen != len) { /* "TLS" mode of operation */ |
452 | if (in != out) |
453 | memcpy(out + aes_off, in + aes_off, plen - aes_off); |
454 | |
455 | /* calculate HMAC and append it to payload */ |
456 | SHA1_Final(out + plen, &key->md); |
457 | key->md = key->tail; |
458 | SHA1_Update(&key->md, out + plen, SHA_DIGEST_LENGTH); |
459 | SHA1_Final(out + plen, &key->md); |
460 | |
461 | /* pad the payload|hmac */ |
462 | plen += SHA_DIGEST_LENGTH; |
463 | for (l = len - plen - 1; plen < len; plen++) |
464 | out[plen] = l; |
465 | /* encrypt HMAC|padding at once */ |
466 | aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off, |
467 | &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1); |
468 | } else { |
469 | aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off, |
470 | &key->ks, EVP_CIPHER_CTX_iv_noconst(ctx), 1); |
471 | } |
472 | } else { |
473 | union { |
474 | unsigned int u[SHA_DIGEST_LENGTH / sizeof(unsigned int)]; |
475 | unsigned char c[32 + SHA_DIGEST_LENGTH]; |
476 | } mac, *pmac; |
477 | |
478 | /* arrange cache line alignment */ |
479 | pmac = (void *)(((size_t)mac.c + 31) & ((size_t)0 - 32)); |
480 | |
481 | if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */ |
482 | size_t inp_len, mask, j, i; |
483 | unsigned int res, maxpad, pad, bitlen; |
484 | int ret = 1; |
485 | union { |
486 | unsigned int u[SHA_LBLOCK]; |
487 | unsigned char c[SHA_CBLOCK]; |
488 | } *data = (void *)key->md.data; |
489 | # if defined(STITCHED_DECRYPT_CALL) |
490 | unsigned char tail_iv[AES_BLOCK_SIZE]; |
491 | int stitch = 0; |
492 | # endif |
493 | |
494 | if ((key->aux.tls_aad[plen - 4] << 8 | key->aux.tls_aad[plen - 3]) |
495 | >= TLS1_1_VERSION) { |
496 | if (len < (AES_BLOCK_SIZE + SHA_DIGEST_LENGTH + 1)) |
497 | return 0; |
498 | |
499 | /* omit explicit iv */ |
500 | memcpy(EVP_CIPHER_CTX_iv_noconst(ctx), in, AES_BLOCK_SIZE); |
501 | |
502 | in += AES_BLOCK_SIZE; |
503 | out += AES_BLOCK_SIZE; |
504 | len -= AES_BLOCK_SIZE; |
505 | } else if (len < (SHA_DIGEST_LENGTH + 1)) |
506 | return 0; |
507 | |
508 | # if defined(STITCHED_DECRYPT_CALL) |
509 | if (len >= 1024 && ctx->key_len == 32) { |
510 | /* decrypt last block */ |
511 | memcpy(tail_iv, in + len - 2 * AES_BLOCK_SIZE, |
512 | AES_BLOCK_SIZE); |
513 | aesni_cbc_encrypt(in + len - AES_BLOCK_SIZE, |
514 | out + len - AES_BLOCK_SIZE, AES_BLOCK_SIZE, |
515 | &key->ks, tail_iv, 0); |
516 | stitch = 1; |
517 | } else |
518 | # endif |
519 | /* decrypt HMAC|padding at once */ |
520 | aesni_cbc_encrypt(in, out, len, &key->ks, |
521 | EVP_CIPHER_CTX_iv_noconst(ctx), 0); |
522 | |
523 | /* figure out payload length */ |
524 | pad = out[len - 1]; |
525 | maxpad = len - (SHA_DIGEST_LENGTH + 1); |
526 | maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8); |
527 | maxpad &= 255; |
528 | |
529 | mask = constant_time_ge(maxpad, pad); |
530 | ret &= mask; |
531 | /* |
532 | * If pad is invalid then we will fail the above test but we must |
533 | * continue anyway because we are in constant time code. However, |
534 | * we'll use the maxpad value instead of the supplied pad to make |
535 | * sure we perform well defined pointer arithmetic. |
536 | */ |
537 | pad = constant_time_select(mask, pad, maxpad); |
538 | |
539 | inp_len = len - (SHA_DIGEST_LENGTH + pad + 1); |
540 | |
541 | key->aux.tls_aad[plen - 2] = inp_len >> 8; |
542 | key->aux.tls_aad[plen - 1] = inp_len; |
543 | |
544 | /* calculate HMAC */ |
545 | key->md = key->head; |
546 | SHA1_Update(&key->md, key->aux.tls_aad, plen); |
547 | |
548 | # if defined(STITCHED_DECRYPT_CALL) |
549 | if (stitch) { |
550 | blocks = (len - (256 + 32 + SHA_CBLOCK)) / SHA_CBLOCK; |
551 | aes_off = len - AES_BLOCK_SIZE - blocks * SHA_CBLOCK; |
552 | sha_off = SHA_CBLOCK - plen; |
553 | |
554 | aesni_cbc_encrypt(in, out, aes_off, &key->ks, ctx->iv, 0); |
555 | |
556 | SHA1_Update(&key->md, out, sha_off); |
557 | aesni256_cbc_sha1_dec(in + aes_off, |
558 | out + aes_off, blocks, &key->ks, |
559 | ctx->iv, &key->md, out + sha_off); |
560 | |
561 | sha_off += blocks *= SHA_CBLOCK; |
562 | out += sha_off; |
563 | len -= sha_off; |
564 | inp_len -= sha_off; |
565 | |
566 | key->md.Nl += (blocks << 3); /* at most 18 bits */ |
567 | memcpy(ctx->iv, tail_iv, AES_BLOCK_SIZE); |
568 | } |
569 | # endif |
570 | |
571 | # if 1 /* see original reference version in #else */ |
572 | len -= SHA_DIGEST_LENGTH; /* amend mac */ |
573 | if (len >= (256 + SHA_CBLOCK)) { |
574 | j = (len - (256 + SHA_CBLOCK)) & (0 - SHA_CBLOCK); |
575 | j += SHA_CBLOCK - key->md.num; |
576 | SHA1_Update(&key->md, out, j); |
577 | out += j; |
578 | len -= j; |
579 | inp_len -= j; |
580 | } |
581 | |
582 | /* but pretend as if we hashed padded payload */ |
583 | bitlen = key->md.Nl + (inp_len << 3); /* at most 18 bits */ |
584 | # ifdef BSWAP4 |
585 | bitlen = BSWAP4(bitlen); |
586 | # else |
587 | mac.c[0] = 0; |
588 | mac.c[1] = (unsigned char)(bitlen >> 16); |
589 | mac.c[2] = (unsigned char)(bitlen >> 8); |
590 | mac.c[3] = (unsigned char)bitlen; |
591 | bitlen = mac.u[0]; |
592 | # endif |
593 | |
594 | pmac->u[0] = 0; |
595 | pmac->u[1] = 0; |
596 | pmac->u[2] = 0; |
597 | pmac->u[3] = 0; |
598 | pmac->u[4] = 0; |
599 | |
600 | for (res = key->md.num, j = 0; j < len; j++) { |
601 | size_t c = out[j]; |
602 | mask = (j - inp_len) >> (sizeof(j) * 8 - 8); |
603 | c &= mask; |
604 | c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8)); |
605 | data->c[res++] = (unsigned char)c; |
606 | |
607 | if (res != SHA_CBLOCK) |
608 | continue; |
609 | |
610 | /* j is not incremented yet */ |
611 | mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1)); |
612 | data->u[SHA_LBLOCK - 1] |= bitlen & mask; |
613 | sha1_block_data_order(&key->md, data, 1); |
614 | mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1)); |
615 | pmac->u[0] |= key->md.h0 & mask; |
616 | pmac->u[1] |= key->md.h1 & mask; |
617 | pmac->u[2] |= key->md.h2 & mask; |
618 | pmac->u[3] |= key->md.h3 & mask; |
619 | pmac->u[4] |= key->md.h4 & mask; |
620 | res = 0; |
621 | } |
622 | |
623 | for (i = res; i < SHA_CBLOCK; i++, j++) |
624 | data->c[i] = 0; |
625 | |
626 | if (res > SHA_CBLOCK - 8) { |
627 | mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1)); |
628 | data->u[SHA_LBLOCK - 1] |= bitlen & mask; |
629 | sha1_block_data_order(&key->md, data, 1); |
630 | mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1)); |
631 | pmac->u[0] |= key->md.h0 & mask; |
632 | pmac->u[1] |= key->md.h1 & mask; |
633 | pmac->u[2] |= key->md.h2 & mask; |
634 | pmac->u[3] |= key->md.h3 & mask; |
635 | pmac->u[4] |= key->md.h4 & mask; |
636 | |
637 | memset(data, 0, SHA_CBLOCK); |
638 | j += 64; |
639 | } |
640 | data->u[SHA_LBLOCK - 1] = bitlen; |
641 | sha1_block_data_order(&key->md, data, 1); |
642 | mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1)); |
643 | pmac->u[0] |= key->md.h0 & mask; |
644 | pmac->u[1] |= key->md.h1 & mask; |
645 | pmac->u[2] |= key->md.h2 & mask; |
646 | pmac->u[3] |= key->md.h3 & mask; |
647 | pmac->u[4] |= key->md.h4 & mask; |
648 | |
649 | # ifdef BSWAP4 |
650 | pmac->u[0] = BSWAP4(pmac->u[0]); |
651 | pmac->u[1] = BSWAP4(pmac->u[1]); |
652 | pmac->u[2] = BSWAP4(pmac->u[2]); |
653 | pmac->u[3] = BSWAP4(pmac->u[3]); |
654 | pmac->u[4] = BSWAP4(pmac->u[4]); |
655 | # else |
656 | for (i = 0; i < 5; i++) { |
657 | res = pmac->u[i]; |
658 | pmac->c[4 * i + 0] = (unsigned char)(res >> 24); |
659 | pmac->c[4 * i + 1] = (unsigned char)(res >> 16); |
660 | pmac->c[4 * i + 2] = (unsigned char)(res >> 8); |
661 | pmac->c[4 * i + 3] = (unsigned char)res; |
662 | } |
663 | # endif |
664 | len += SHA_DIGEST_LENGTH; |
665 | # else /* pre-lucky-13 reference version of above */ |
666 | SHA1_Update(&key->md, out, inp_len); |
667 | res = key->md.num; |
668 | SHA1_Final(pmac->c, &key->md); |
669 | |
670 | { |
671 | unsigned int inp_blocks, pad_blocks; |
672 | |
673 | /* but pretend as if we hashed padded payload */ |
674 | inp_blocks = |
675 | 1 + ((SHA_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1)); |
676 | res += (unsigned int)(len - inp_len); |
677 | pad_blocks = res / SHA_CBLOCK; |
678 | res %= SHA_CBLOCK; |
679 | pad_blocks += |
680 | 1 + ((SHA_CBLOCK - 9 - res) >> (sizeof(res) * 8 - 1)); |
681 | for (; inp_blocks < pad_blocks; inp_blocks++) |
682 | sha1_block_data_order(&key->md, data, 1); |
683 | } |
684 | # endif |
685 | key->md = key->tail; |
686 | SHA1_Update(&key->md, pmac->c, SHA_DIGEST_LENGTH); |
687 | SHA1_Final(pmac->c, &key->md); |
688 | |
689 | /* verify HMAC */ |
690 | out += inp_len; |
691 | len -= inp_len; |
692 | # if 1 /* see original reference version in #else */ |
693 | { |
694 | unsigned char *p = out + len - 1 - maxpad - SHA_DIGEST_LENGTH; |
695 | size_t off = out - p; |
696 | unsigned int c, cmask; |
697 | |
698 | maxpad += SHA_DIGEST_LENGTH; |
699 | for (res = 0, i = 0, j = 0; j < maxpad; j++) { |
700 | c = p[j]; |
701 | cmask = |
702 | ((int)(j - off - SHA_DIGEST_LENGTH)) >> (sizeof(int) * |
703 | 8 - 1); |
704 | res |= (c ^ pad) & ~cmask; /* ... and padding */ |
705 | cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1); |
706 | res |= (c ^ pmac->c[i]) & cmask; |
707 | i += 1 & cmask; |
708 | } |
709 | maxpad -= SHA_DIGEST_LENGTH; |
710 | |
711 | res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1)); |
712 | ret &= (int)~res; |
713 | } |
714 | # else /* pre-lucky-13 reference version of above */ |
715 | for (res = 0, i = 0; i < SHA_DIGEST_LENGTH; i++) |
716 | res |= out[i] ^ pmac->c[i]; |
717 | res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1)); |
718 | ret &= (int)~res; |
719 | |
720 | /* verify padding */ |
721 | pad = (pad & ~res) | (maxpad & res); |
722 | out = out + len - 1 - pad; |
723 | for (res = 0, i = 0; i < pad; i++) |
724 | res |= out[i] ^ pad; |
725 | |
726 | res = (0 - res) >> (sizeof(res) * 8 - 1); |
727 | ret &= (int)~res; |
728 | # endif |
729 | return ret; |
730 | } else { |
731 | # if defined(STITCHED_DECRYPT_CALL) |
732 | if (len >= 1024 && ctx->key_len == 32) { |
733 | if (sha_off %= SHA_CBLOCK) |
734 | blocks = (len - 3 * SHA_CBLOCK) / SHA_CBLOCK; |
735 | else |
736 | blocks = (len - 2 * SHA_CBLOCK) / SHA_CBLOCK; |
737 | aes_off = len - blocks * SHA_CBLOCK; |
738 | |
739 | aesni_cbc_encrypt(in, out, aes_off, &key->ks, ctx->iv, 0); |
740 | SHA1_Update(&key->md, out, sha_off); |
741 | aesni256_cbc_sha1_dec(in + aes_off, |
742 | out + aes_off, blocks, &key->ks, |
743 | ctx->iv, &key->md, out + sha_off); |
744 | |
745 | sha_off += blocks *= SHA_CBLOCK; |
746 | out += sha_off; |
747 | len -= sha_off; |
748 | |
749 | key->md.Nh += blocks >> 29; |
750 | key->md.Nl += blocks <<= 3; |
751 | if (key->md.Nl < (unsigned int)blocks) |
752 | key->md.Nh++; |
753 | } else |
754 | # endif |
755 | /* decrypt HMAC|padding at once */ |
756 | aesni_cbc_encrypt(in, out, len, &key->ks, |
757 | EVP_CIPHER_CTX_iv_noconst(ctx), 0); |
758 | |
759 | SHA1_Update(&key->md, out, len); |
760 | } |
761 | } |
762 | |
763 | return 1; |
764 | } |
765 | |
766 | static int aesni_cbc_hmac_sha1_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg, |
767 | void *ptr) |
768 | { |
769 | EVP_AES_HMAC_SHA1 *key = data(ctx); |
770 | |
771 | switch (type) { |
772 | case EVP_CTRL_AEAD_SET_MAC_KEY: |
773 | { |
774 | unsigned int i; |
775 | unsigned char hmac_key[64]; |
776 | |
777 | memset(hmac_key, 0, sizeof(hmac_key)); |
778 | |
779 | if (arg > (int)sizeof(hmac_key)) { |
780 | SHA1_Init(&key->head); |
781 | SHA1_Update(&key->head, ptr, arg); |
782 | SHA1_Final(hmac_key, &key->head); |
783 | } else { |
784 | memcpy(hmac_key, ptr, arg); |
785 | } |
786 | |
787 | for (i = 0; i < sizeof(hmac_key); i++) |
788 | hmac_key[i] ^= 0x36; /* ipad */ |
789 | SHA1_Init(&key->head); |
790 | SHA1_Update(&key->head, hmac_key, sizeof(hmac_key)); |
791 | |
792 | for (i = 0; i < sizeof(hmac_key); i++) |
793 | hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */ |
794 | SHA1_Init(&key->tail); |
795 | SHA1_Update(&key->tail, hmac_key, sizeof(hmac_key)); |
796 | |
797 | OPENSSL_cleanse(hmac_key, sizeof(hmac_key)); |
798 | |
799 | return 1; |
800 | } |
801 | case EVP_CTRL_AEAD_TLS1_AAD: |
802 | { |
803 | unsigned char *p = ptr; |
804 | unsigned int len; |
805 | |
806 | if (arg != EVP_AEAD_TLS1_AAD_LEN) |
807 | return -1; |
808 | |
809 | len = p[arg - 2] << 8 | p[arg - 1]; |
810 | |
811 | if (EVP_CIPHER_CTX_encrypting(ctx)) { |
812 | key->payload_length = len; |
813 | if ((key->aux.tls_ver = |
814 | p[arg - 4] << 8 | p[arg - 3]) >= TLS1_1_VERSION) { |
815 | if (len < AES_BLOCK_SIZE) |
816 | return 0; |
817 | len -= AES_BLOCK_SIZE; |
818 | p[arg - 2] = len >> 8; |
819 | p[arg - 1] = len; |
820 | } |
821 | key->md = key->head; |
822 | SHA1_Update(&key->md, p, arg); |
823 | |
824 | return (int)(((len + SHA_DIGEST_LENGTH + |
825 | AES_BLOCK_SIZE) & -AES_BLOCK_SIZE) |
826 | - len); |
827 | } else { |
828 | memcpy(key->aux.tls_aad, ptr, arg); |
829 | key->payload_length = arg; |
830 | |
831 | return SHA_DIGEST_LENGTH; |
832 | } |
833 | } |
834 | # if !defined(OPENSSL_NO_MULTIBLOCK) |
835 | case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE: |
836 | return (int)(5 + 16 + ((arg + 20 + 16) & -16)); |
837 | case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD: |
838 | { |
839 | EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param = |
840 | (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr; |
841 | unsigned int n4x = 1, x4; |
842 | unsigned int frag, last, packlen, inp_len; |
843 | |
844 | if (arg < (int)sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM)) |
845 | return -1; |
846 | |
847 | inp_len = param->inp[11] << 8 | param->inp[12]; |
848 | |
849 | if (EVP_CIPHER_CTX_encrypting(ctx)) { |
850 | if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION) |
851 | return -1; |
852 | |
853 | if (inp_len) { |
854 | if (inp_len < 4096) |
855 | return 0; /* too short */ |
856 | |
857 | if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5)) |
858 | n4x = 2; /* AVX2 */ |
859 | } else if ((n4x = param->interleave / 4) && n4x <= 2) |
860 | inp_len = param->len; |
861 | else |
862 | return -1; |
863 | |
864 | key->md = key->head; |
865 | SHA1_Update(&key->md, param->inp, 13); |
866 | |
867 | x4 = 4 * n4x; |
868 | n4x += 1; |
869 | |
870 | frag = inp_len >> n4x; |
871 | last = inp_len + frag - (frag << n4x); |
872 | if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) { |
873 | frag++; |
874 | last -= x4 - 1; |
875 | } |
876 | |
877 | packlen = 5 + 16 + ((frag + 20 + 16) & -16); |
878 | packlen = (packlen << n4x) - packlen; |
879 | packlen += 5 + 16 + ((last + 20 + 16) & -16); |
880 | |
881 | param->interleave = x4; |
882 | |
883 | return (int)packlen; |
884 | } else |
885 | return -1; /* not yet */ |
886 | } |
887 | case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT: |
888 | { |
889 | EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param = |
890 | (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *) ptr; |
891 | |
892 | return (int)tls1_1_multi_block_encrypt(key, param->out, |
893 | param->inp, param->len, |
894 | param->interleave / 4); |
895 | } |
896 | case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT: |
897 | # endif |
898 | default: |
899 | return -1; |
900 | } |
901 | } |
902 | |
903 | static EVP_CIPHER aesni_128_cbc_hmac_sha1_cipher = { |
904 | # ifdef NID_aes_128_cbc_hmac_sha1 |
905 | NID_aes_128_cbc_hmac_sha1, |
906 | # else |
907 | NID_undef, |
908 | # endif |
909 | AES_BLOCK_SIZE, 16, AES_BLOCK_SIZE, |
910 | EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 | |
911 | EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK, |
912 | aesni_cbc_hmac_sha1_init_key, |
913 | aesni_cbc_hmac_sha1_cipher, |
914 | NULL, |
915 | sizeof(EVP_AES_HMAC_SHA1), |
916 | EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv, |
917 | EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv, |
918 | aesni_cbc_hmac_sha1_ctrl, |
919 | NULL |
920 | }; |
921 | |
922 | static EVP_CIPHER aesni_256_cbc_hmac_sha1_cipher = { |
923 | # ifdef NID_aes_256_cbc_hmac_sha1 |
924 | NID_aes_256_cbc_hmac_sha1, |
925 | # else |
926 | NID_undef, |
927 | # endif |
928 | AES_BLOCK_SIZE, 32, AES_BLOCK_SIZE, |
929 | EVP_CIPH_CBC_MODE | EVP_CIPH_FLAG_DEFAULT_ASN1 | |
930 | EVP_CIPH_FLAG_AEAD_CIPHER | EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK, |
931 | aesni_cbc_hmac_sha1_init_key, |
932 | aesni_cbc_hmac_sha1_cipher, |
933 | NULL, |
934 | sizeof(EVP_AES_HMAC_SHA1), |
935 | EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_set_asn1_iv, |
936 | EVP_CIPH_FLAG_DEFAULT_ASN1 ? NULL : EVP_CIPHER_get_asn1_iv, |
937 | aesni_cbc_hmac_sha1_ctrl, |
938 | NULL |
939 | }; |
940 | |
941 | const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha1(void) |
942 | { |
943 | return (OPENSSL_ia32cap_P[1] & AESNI_CAPABLE ? |
944 | &aesni_128_cbc_hmac_sha1_cipher : NULL); |
945 | } |
946 | |
947 | const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha1(void) |
948 | { |
949 | return (OPENSSL_ia32cap_P[1] & AESNI_CAPABLE ? |
950 | &aesni_256_cbc_hmac_sha1_cipher : NULL); |
951 | } |
952 | #else |
953 | const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha1(void) |
954 | { |
955 | return NULL; |
956 | } |
957 | |
958 | const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha1(void) |
959 | { |
960 | return NULL; |
961 | } |
962 | #endif |
963 | |