]>
Commit | Line | Data |
---|---|---|
0d2bfe52 SL |
1 | /* |
2 | * Copyright 2011-2019 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 | ||
0081ce9b | 10 | /* |
85d843c8 | 11 | * All low level APIs are deprecated for public use, but still ok for internal |
0081ce9b RL |
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 | ||
0d2bfe52 SL |
17 | #include "cipher_aes_cbc_hmac_sha.h" |
18 | ||
87d3bb8e | 19 | #if !defined(AES_CBC_HMAC_SHA_CAPABLE) || !defined(AESNI_CAPABLE) |
0d2bfe52 SL |
20 | int cipher_capable_aes_cbc_hmac_sha256(void) |
21 | { | |
22 | return 0; | |
23 | } | |
87d3bb8e MC |
24 | |
25 | const PROV_CIPHER_HW_AES_HMAC_SHA *PROV_CIPHER_HW_aes_cbc_hmac_sha256(void) | |
26 | { | |
27 | return NULL; | |
28 | } | |
0d2bfe52 SL |
29 | #else |
30 | ||
993ebac9 | 31 | # include <openssl/rand.h> |
0d2bfe52 SL |
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 | ||
40 | int 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 | ||
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 | return ret < 0 ? 0 : 1; | |
66 | } | |
67 | ||
68 | void sha256_block_data_order(void *c, const void *p, size_t len); | |
69 | ||
70 | static void sha256_update(SHA256_CTX *c, const void *data, size_t len) | |
71 | { | |
72 | const unsigned char *ptr = data; | |
73 | size_t res; | |
74 | ||
75 | if ((res = c->num)) { | |
76 | res = SHA256_CBLOCK - res; | |
77 | if (len < res) | |
78 | res = len; | |
79 | SHA256_Update(c, ptr, res); | |
80 | ptr += res; | |
81 | len -= res; | |
82 | } | |
83 | ||
84 | res = len % SHA256_CBLOCK; | |
85 | len -= res; | |
86 | ||
87 | if (len) { | |
88 | sha256_block_data_order(c, ptr, len / SHA256_CBLOCK); | |
89 | ||
90 | ptr += len; | |
91 | c->Nh += len >> 29; | |
92 | c->Nl += len <<= 3; | |
93 | if (c->Nl < (unsigned int)len) | |
94 | c->Nh++; | |
95 | } | |
96 | ||
97 | if (res) | |
98 | SHA256_Update(c, ptr, res); | |
99 | } | |
100 | ||
101 | # if !defined(OPENSSL_NO_MULTIBLOCK) | |
102 | ||
103 | typedef struct { | |
104 | unsigned int A[8], B[8], C[8], D[8], E[8], F[8], G[8], H[8]; | |
105 | } SHA256_MB_CTX; | |
106 | ||
107 | typedef struct { | |
108 | const unsigned char *ptr; | |
109 | int blocks; | |
110 | } HASH_DESC; | |
111 | ||
112 | typedef struct { | |
113 | const unsigned char *inp; | |
114 | unsigned char *out; | |
115 | int blocks; | |
116 | u64 iv[2]; | |
117 | } CIPH_DESC; | |
118 | ||
119 | void sha256_multi_block(SHA256_MB_CTX *, const HASH_DESC *, int); | |
120 | void aesni_multi_cbc_encrypt(CIPH_DESC *, void *, int); | |
121 | ||
122 | static size_t tls1_multi_block_encrypt(void *vctx, | |
123 | unsigned char *out, | |
124 | const unsigned char *inp, | |
125 | size_t inp_len, int n4x) | |
126 | { /* n4x is 1 or 2 */ | |
127 | PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx; | |
128 | PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx; | |
129 | HASH_DESC hash_d[8], edges[8]; | |
130 | CIPH_DESC ciph_d[8]; | |
131 | unsigned char storage[sizeof(SHA256_MB_CTX) + 32]; | |
132 | union { | |
133 | u64 q[16]; | |
134 | u32 d[32]; | |
135 | u8 c[128]; | |
136 | } blocks[8]; | |
137 | SHA256_MB_CTX *mctx; | |
138 | unsigned int frag, last, packlen, i; | |
139 | unsigned int x4 = 4 * n4x, minblocks, processed = 0; | |
140 | size_t ret = 0; | |
141 | u8 *IVs; | |
142 | # if defined(BSWAP8) | |
143 | u64 seqnum; | |
144 | # endif | |
145 | ||
146 | /* ask for IVs in bulk */ | |
993ebac9 | 147 | if (RAND_bytes_ex(ctx->base.libctx, (IVs = blocks[0].c), 16 * x4) <= 0) |
0d2bfe52 SL |
148 | return 0; |
149 | ||
150 | mctx = (SHA256_MB_CTX *) (storage + 32 - ((size_t)storage % 32)); /* align */ | |
151 | ||
152 | frag = (unsigned int)inp_len >> (1 + n4x); | |
153 | last = (unsigned int)inp_len + frag - (frag << (1 + n4x)); | |
154 | if (last > frag && ((last + 13 + 9) % 64) < (x4 - 1)) { | |
155 | frag++; | |
156 | last -= x4 - 1; | |
157 | } | |
158 | ||
159 | packlen = 5 + 16 + ((frag + 32 + 16) & -16); | |
160 | ||
161 | /* populate descriptors with pointers and IVs */ | |
162 | hash_d[0].ptr = inp; | |
163 | ciph_d[0].inp = inp; | |
164 | /* 5+16 is place for header and explicit IV */ | |
165 | ciph_d[0].out = out + 5 + 16; | |
166 | memcpy(ciph_d[0].out - 16, IVs, 16); | |
167 | memcpy(ciph_d[0].iv, IVs, 16); | |
168 | IVs += 16; | |
169 | ||
170 | for (i = 1; i < x4; i++) { | |
171 | ciph_d[i].inp = hash_d[i].ptr = hash_d[i - 1].ptr + frag; | |
172 | ciph_d[i].out = ciph_d[i - 1].out + packlen; | |
173 | memcpy(ciph_d[i].out - 16, IVs, 16); | |
174 | memcpy(ciph_d[i].iv, IVs, 16); | |
175 | IVs += 16; | |
176 | } | |
177 | ||
178 | # if defined(BSWAP8) | |
179 | memcpy(blocks[0].c, sctx->md.data, 8); | |
180 | seqnum = BSWAP8(blocks[0].q[0]); | |
181 | # endif | |
182 | ||
183 | for (i = 0; i < x4; i++) { | |
184 | unsigned int len = (i == (x4 - 1) ? last : frag); | |
185 | # if !defined(BSWAP8) | |
186 | unsigned int carry, j; | |
187 | # endif | |
188 | ||
189 | mctx->A[i] = sctx->md.h[0]; | |
190 | mctx->B[i] = sctx->md.h[1]; | |
191 | mctx->C[i] = sctx->md.h[2]; | |
192 | mctx->D[i] = sctx->md.h[3]; | |
193 | mctx->E[i] = sctx->md.h[4]; | |
194 | mctx->F[i] = sctx->md.h[5]; | |
195 | mctx->G[i] = sctx->md.h[6]; | |
196 | mctx->H[i] = sctx->md.h[7]; | |
197 | ||
198 | /* fix seqnum */ | |
199 | # if defined(BSWAP8) | |
200 | blocks[i].q[0] = BSWAP8(seqnum + i); | |
201 | # else | |
202 | for (carry = i, j = 8; j--;) { | |
203 | blocks[i].c[j] = ((u8 *)sctx->md.data)[j] + carry; | |
204 | carry = (blocks[i].c[j] - carry) >> (sizeof(carry) * 8 - 1); | |
205 | } | |
206 | # endif | |
207 | blocks[i].c[8] = ((u8 *)sctx->md.data)[8]; | |
208 | blocks[i].c[9] = ((u8 *)sctx->md.data)[9]; | |
209 | blocks[i].c[10] = ((u8 *)sctx->md.data)[10]; | |
210 | /* fix length */ | |
211 | blocks[i].c[11] = (u8)(len >> 8); | |
212 | blocks[i].c[12] = (u8)(len); | |
213 | ||
214 | memcpy(blocks[i].c + 13, hash_d[i].ptr, 64 - 13); | |
215 | hash_d[i].ptr += 64 - 13; | |
216 | hash_d[i].blocks = (len - (64 - 13)) / 64; | |
217 | ||
218 | edges[i].ptr = blocks[i].c; | |
219 | edges[i].blocks = 1; | |
220 | } | |
221 | ||
222 | /* hash 13-byte headers and first 64-13 bytes of inputs */ | |
223 | sha256_multi_block(mctx, edges, n4x); | |
224 | /* hash bulk inputs */ | |
225 | # define MAXCHUNKSIZE 2048 | |
226 | # if MAXCHUNKSIZE%64 | |
227 | # error "MAXCHUNKSIZE is not divisible by 64" | |
228 | # elif MAXCHUNKSIZE | |
229 | /* | |
230 | * goal is to minimize pressure on L1 cache by moving in shorter steps, | |
231 | * so that hashed data is still in the cache by the time we encrypt it | |
232 | */ | |
233 | minblocks = ((frag <= last ? frag : last) - (64 - 13)) / 64; | |
234 | if (minblocks > MAXCHUNKSIZE / 64) { | |
235 | for (i = 0; i < x4; i++) { | |
236 | edges[i].ptr = hash_d[i].ptr; | |
237 | edges[i].blocks = MAXCHUNKSIZE / 64; | |
238 | ciph_d[i].blocks = MAXCHUNKSIZE / 16; | |
239 | } | |
240 | do { | |
241 | sha256_multi_block(mctx, edges, n4x); | |
242 | aesni_multi_cbc_encrypt(ciph_d, &ctx->ks, n4x); | |
243 | ||
244 | for (i = 0; i < x4; i++) { | |
245 | edges[i].ptr = hash_d[i].ptr += MAXCHUNKSIZE; | |
246 | hash_d[i].blocks -= MAXCHUNKSIZE / 64; | |
247 | edges[i].blocks = MAXCHUNKSIZE / 64; | |
248 | ciph_d[i].inp += MAXCHUNKSIZE; | |
249 | ciph_d[i].out += MAXCHUNKSIZE; | |
250 | ciph_d[i].blocks = MAXCHUNKSIZE / 16; | |
251 | memcpy(ciph_d[i].iv, ciph_d[i].out - 16, 16); | |
252 | } | |
253 | processed += MAXCHUNKSIZE; | |
254 | minblocks -= MAXCHUNKSIZE / 64; | |
255 | } while (minblocks > MAXCHUNKSIZE / 64); | |
256 | } | |
257 | # endif | |
258 | # undef MAXCHUNKSIZE | |
259 | sha256_multi_block(mctx, hash_d, n4x); | |
260 | ||
261 | memset(blocks, 0, sizeof(blocks)); | |
262 | for (i = 0; i < x4; i++) { | |
263 | unsigned int len = (i == (x4 - 1) ? last : frag), | |
264 | off = hash_d[i].blocks * 64; | |
265 | const unsigned char *ptr = hash_d[i].ptr + off; | |
266 | ||
267 | off = (len - processed) - (64 - 13) - off; /* remainder actually */ | |
268 | memcpy(blocks[i].c, ptr, off); | |
269 | blocks[i].c[off] = 0x80; | |
270 | len += 64 + 13; /* 64 is HMAC header */ | |
271 | len *= 8; /* convert to bits */ | |
272 | if (off < (64 - 8)) { | |
273 | # ifdef BSWAP4 | |
274 | blocks[i].d[15] = BSWAP4(len); | |
275 | # else | |
276 | PUTU32(blocks[i].c + 60, len); | |
277 | # endif | |
278 | edges[i].blocks = 1; | |
279 | } else { | |
280 | # ifdef BSWAP4 | |
281 | blocks[i].d[31] = BSWAP4(len); | |
282 | # else | |
283 | PUTU32(blocks[i].c + 124, len); | |
284 | # endif | |
285 | edges[i].blocks = 2; | |
286 | } | |
287 | edges[i].ptr = blocks[i].c; | |
288 | } | |
289 | ||
290 | /* hash input tails and finalize */ | |
291 | sha256_multi_block(mctx, edges, n4x); | |
292 | ||
293 | memset(blocks, 0, sizeof(blocks)); | |
294 | for (i = 0; i < x4; i++) { | |
295 | # ifdef BSWAP4 | |
296 | blocks[i].d[0] = BSWAP4(mctx->A[i]); | |
297 | mctx->A[i] = sctx->tail.h[0]; | |
298 | blocks[i].d[1] = BSWAP4(mctx->B[i]); | |
299 | mctx->B[i] = sctx->tail.h[1]; | |
300 | blocks[i].d[2] = BSWAP4(mctx->C[i]); | |
301 | mctx->C[i] = sctx->tail.h[2]; | |
302 | blocks[i].d[3] = BSWAP4(mctx->D[i]); | |
303 | mctx->D[i] = sctx->tail.h[3]; | |
304 | blocks[i].d[4] = BSWAP4(mctx->E[i]); | |
305 | mctx->E[i] = sctx->tail.h[4]; | |
306 | blocks[i].d[5] = BSWAP4(mctx->F[i]); | |
307 | mctx->F[i] = sctx->tail.h[5]; | |
308 | blocks[i].d[6] = BSWAP4(mctx->G[i]); | |
309 | mctx->G[i] = sctx->tail.h[6]; | |
310 | blocks[i].d[7] = BSWAP4(mctx->H[i]); | |
311 | mctx->H[i] = sctx->tail.h[7]; | |
312 | blocks[i].c[32] = 0x80; | |
313 | blocks[i].d[15] = BSWAP4((64 + 32) * 8); | |
314 | # else | |
315 | PUTU32(blocks[i].c + 0, mctx->A[i]); | |
316 | mctx->A[i] = sctx->tail.h[0]; | |
317 | PUTU32(blocks[i].c + 4, mctx->B[i]); | |
318 | mctx->B[i] = sctx->tail.h[1]; | |
319 | PUTU32(blocks[i].c + 8, mctx->C[i]); | |
320 | mctx->C[i] = sctx->tail.h[2]; | |
321 | PUTU32(blocks[i].c + 12, mctx->D[i]); | |
322 | mctx->D[i] = sctx->tail.h[3]; | |
323 | PUTU32(blocks[i].c + 16, mctx->E[i]); | |
324 | mctx->E[i] = sctx->tail.h[4]; | |
325 | PUTU32(blocks[i].c + 20, mctx->F[i]); | |
326 | mctx->F[i] = sctx->tail.h[5]; | |
327 | PUTU32(blocks[i].c + 24, mctx->G[i]); | |
328 | mctx->G[i] = sctx->tail.h[6]; | |
329 | PUTU32(blocks[i].c + 28, mctx->H[i]); | |
330 | mctx->H[i] = sctx->tail.h[7]; | |
331 | blocks[i].c[32] = 0x80; | |
332 | PUTU32(blocks[i].c + 60, (64 + 32) * 8); | |
333 | # endif /* BSWAP */ | |
334 | edges[i].ptr = blocks[i].c; | |
335 | edges[i].blocks = 1; | |
336 | } | |
337 | ||
338 | /* finalize MACs */ | |
339 | sha256_multi_block(mctx, edges, n4x); | |
340 | ||
341 | for (i = 0; i < x4; i++) { | |
342 | unsigned int len = (i == (x4 - 1) ? last : frag), pad, j; | |
343 | unsigned char *out0 = out; | |
344 | ||
345 | memcpy(ciph_d[i].out, ciph_d[i].inp, len - processed); | |
346 | ciph_d[i].inp = ciph_d[i].out; | |
347 | ||
348 | out += 5 + 16 + len; | |
349 | ||
350 | /* write MAC */ | |
351 | PUTU32(out + 0, mctx->A[i]); | |
352 | PUTU32(out + 4, mctx->B[i]); | |
353 | PUTU32(out + 8, mctx->C[i]); | |
354 | PUTU32(out + 12, mctx->D[i]); | |
355 | PUTU32(out + 16, mctx->E[i]); | |
356 | PUTU32(out + 20, mctx->F[i]); | |
357 | PUTU32(out + 24, mctx->G[i]); | |
358 | PUTU32(out + 28, mctx->H[i]); | |
359 | out += 32; | |
360 | len += 32; | |
361 | ||
362 | /* pad */ | |
363 | pad = 15 - len % 16; | |
364 | for (j = 0; j <= pad; j++) | |
365 | *(out++) = pad; | |
366 | len += pad + 1; | |
367 | ||
368 | ciph_d[i].blocks = (len - processed) / 16; | |
369 | len += 16; /* account for explicit iv */ | |
370 | ||
371 | /* arrange header */ | |
372 | out0[0] = ((u8 *)sctx->md.data)[8]; | |
373 | out0[1] = ((u8 *)sctx->md.data)[9]; | |
374 | out0[2] = ((u8 *)sctx->md.data)[10]; | |
375 | out0[3] = (u8)(len >> 8); | |
376 | out0[4] = (u8)(len); | |
377 | ||
378 | ret += len + 5; | |
379 | inp += frag; | |
380 | } | |
381 | ||
382 | aesni_multi_cbc_encrypt(ciph_d, &ctx->ks, n4x); | |
383 | ||
384 | OPENSSL_cleanse(blocks, sizeof(blocks)); | |
385 | OPENSSL_cleanse(mctx, sizeof(*mctx)); | |
386 | ||
387 | ctx->multiblock_encrypt_len = ret; | |
388 | return ret; | |
389 | } | |
390 | # endif /* !OPENSSL_NO_MULTIBLOCK */ | |
391 | ||
392 | static int aesni_cbc_hmac_sha256_cipher(PROV_CIPHER_CTX *vctx, | |
393 | unsigned char *out, | |
394 | const unsigned char *in, size_t len) | |
395 | { | |
396 | PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx; | |
397 | PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx; | |
398 | unsigned int l; | |
399 | size_t plen = ctx->payload_length; | |
400 | size_t iv = 0; /* explicit IV in TLS 1.1 and * later */ | |
401 | size_t aes_off = 0, blocks; | |
402 | size_t sha_off = SHA256_CBLOCK - sctx->md.num; | |
403 | ||
404 | ctx->payload_length = NO_PAYLOAD_LENGTH; | |
405 | ||
406 | if (len % AES_BLOCK_SIZE) | |
407 | return 0; | |
408 | ||
409 | if (ctx->base.enc) { | |
410 | if (plen == NO_PAYLOAD_LENGTH) | |
411 | plen = len; | |
412 | else if (len != | |
413 | ((plen + SHA256_DIGEST_LENGTH + | |
414 | AES_BLOCK_SIZE) & -AES_BLOCK_SIZE)) | |
415 | return 0; | |
416 | else if (ctx->aux.tls_ver >= TLS1_1_VERSION) | |
417 | iv = AES_BLOCK_SIZE; | |
418 | ||
419 | /* | |
420 | * Assembly stitch handles AVX-capable processors, but its | |
421 | * performance is not optimal on AMD Jaguar, ~40% worse, for | |
422 | * unknown reasons. Incidentally processor in question supports | |
423 | * AVX, but not AMD-specific XOP extension, which can be used | |
424 | * to identify it and avoid stitch invocation. So that after we | |
425 | * establish that current CPU supports AVX, we even see if it's | |
426 | * either even XOP-capable Bulldozer-based or GenuineIntel one. | |
427 | * But SHAEXT-capable go ahead... | |
428 | */ | |
429 | if (((OPENSSL_ia32cap_P[2] & (1 << 29)) || /* SHAEXT? */ | |
430 | ((OPENSSL_ia32cap_P[1] & (1 << (60 - 32))) && /* AVX? */ | |
431 | ((OPENSSL_ia32cap_P[1] & (1 << (43 - 32))) /* XOP? */ | |
432 | | (OPENSSL_ia32cap_P[0] & (1 << 30))))) && /* "Intel CPU"? */ | |
433 | plen > (sha_off + iv) && | |
434 | (blocks = (plen - (sha_off + iv)) / SHA256_CBLOCK)) { | |
435 | sha256_update(&sctx->md, in + iv, sha_off); | |
436 | ||
437 | (void)aesni_cbc_sha256_enc(in, out, blocks, &ctx->ks, | |
438 | ctx->base.iv, | |
439 | &sctx->md, in + iv + sha_off); | |
440 | blocks *= SHA256_CBLOCK; | |
441 | aes_off += blocks; | |
442 | sha_off += blocks; | |
443 | sctx->md.Nh += blocks >> 29; | |
444 | sctx->md.Nl += blocks <<= 3; | |
445 | if (sctx->md.Nl < (unsigned int)blocks) | |
446 | sctx->md.Nh++; | |
447 | } else { | |
448 | sha_off = 0; | |
449 | } | |
450 | sha_off += iv; | |
451 | sha256_update(&sctx->md, in + sha_off, plen - sha_off); | |
452 | ||
453 | if (plen != len) { /* "TLS" mode of operation */ | |
454 | if (in != out) | |
455 | memcpy(out + aes_off, in + aes_off, plen - aes_off); | |
456 | ||
457 | /* calculate HMAC and append it to payload */ | |
458 | SHA256_Final(out + plen, &sctx->md); | |
459 | sctx->md = sctx->tail; | |
460 | sha256_update(&sctx->md, out + plen, SHA256_DIGEST_LENGTH); | |
461 | SHA256_Final(out + plen, &sctx->md); | |
462 | ||
463 | /* pad the payload|hmac */ | |
464 | plen += SHA256_DIGEST_LENGTH; | |
465 | for (l = len - plen - 1; plen < len; plen++) | |
466 | out[plen] = l; | |
467 | /* encrypt HMAC|padding at once */ | |
468 | aesni_cbc_encrypt(out + aes_off, out + aes_off, len - aes_off, | |
469 | &ctx->ks, ctx->base.iv, 1); | |
470 | } else { | |
471 | aesni_cbc_encrypt(in + aes_off, out + aes_off, len - aes_off, | |
472 | &ctx->ks, ctx->base.iv, 1); | |
473 | } | |
474 | } else { | |
475 | union { | |
476 | unsigned int u[SHA256_DIGEST_LENGTH / sizeof(unsigned int)]; | |
477 | unsigned char c[64 + SHA256_DIGEST_LENGTH]; | |
478 | } mac, *pmac; | |
479 | ||
480 | /* arrange cache line alignment */ | |
481 | pmac = (void *)(((size_t)mac.c + 63) & ((size_t)0 - 64)); | |
482 | ||
483 | /* decrypt HMAC|padding at once */ | |
484 | aesni_cbc_encrypt(in, out, len, &ctx->ks, | |
485 | ctx->base.iv, 0); | |
486 | ||
487 | if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */ | |
488 | size_t inp_len, mask, j, i; | |
489 | unsigned int res, maxpad, pad, bitlen; | |
490 | int ret = 1; | |
491 | union { | |
492 | unsigned int u[SHA_LBLOCK]; | |
493 | unsigned char c[SHA256_CBLOCK]; | |
494 | } *data = (void *)sctx->md.data; | |
495 | ||
496 | if ((ctx->aux.tls_aad[plen - 4] << 8 | ctx->aux.tls_aad[plen - 3]) | |
497 | >= TLS1_1_VERSION) | |
498 | iv = AES_BLOCK_SIZE; | |
499 | ||
500 | if (len < (iv + SHA256_DIGEST_LENGTH + 1)) | |
501 | return 0; | |
502 | ||
503 | /* omit explicit iv */ | |
504 | out += iv; | |
505 | len -= iv; | |
506 | ||
507 | /* figure out payload length */ | |
508 | pad = out[len - 1]; | |
509 | maxpad = len - (SHA256_DIGEST_LENGTH + 1); | |
510 | maxpad |= (255 - maxpad) >> (sizeof(maxpad) * 8 - 8); | |
511 | maxpad &= 255; | |
512 | ||
513 | mask = constant_time_ge(maxpad, pad); | |
514 | ret &= mask; | |
515 | /* | |
516 | * If pad is invalid then we will fail the above test but we must | |
517 | * continue anyway because we are in constant time code. However, | |
518 | * we'll use the maxpad value instead of the supplied pad to make | |
519 | * sure we perform well defined pointer arithmetic. | |
520 | */ | |
521 | pad = constant_time_select(mask, pad, maxpad); | |
522 | ||
523 | inp_len = len - (SHA256_DIGEST_LENGTH + pad + 1); | |
524 | ||
525 | ctx->aux.tls_aad[plen - 2] = inp_len >> 8; | |
526 | ctx->aux.tls_aad[plen - 1] = inp_len; | |
527 | ||
528 | /* calculate HMAC */ | |
529 | sctx->md = sctx->head; | |
530 | sha256_update(&sctx->md, ctx->aux.tls_aad, plen); | |
531 | ||
532 | /* code with lucky-13 fix */ | |
533 | len -= SHA256_DIGEST_LENGTH; /* amend mac */ | |
534 | if (len >= (256 + SHA256_CBLOCK)) { | |
535 | j = (len - (256 + SHA256_CBLOCK)) & (0 - SHA256_CBLOCK); | |
536 | j += SHA256_CBLOCK - sctx->md.num; | |
537 | sha256_update(&sctx->md, out, j); | |
538 | out += j; | |
539 | len -= j; | |
540 | inp_len -= j; | |
541 | } | |
542 | ||
543 | /* but pretend as if we hashed padded payload */ | |
544 | bitlen = sctx->md.Nl + (inp_len << 3); /* at most 18 bits */ | |
545 | # ifdef BSWAP4 | |
546 | bitlen = BSWAP4(bitlen); | |
547 | # else | |
548 | mac.c[0] = 0; | |
549 | mac.c[1] = (unsigned char)(bitlen >> 16); | |
550 | mac.c[2] = (unsigned char)(bitlen >> 8); | |
551 | mac.c[3] = (unsigned char)bitlen; | |
552 | bitlen = mac.u[0]; | |
553 | # endif /* BSWAP */ | |
554 | ||
555 | pmac->u[0] = 0; | |
556 | pmac->u[1] = 0; | |
557 | pmac->u[2] = 0; | |
558 | pmac->u[3] = 0; | |
559 | pmac->u[4] = 0; | |
560 | pmac->u[5] = 0; | |
561 | pmac->u[6] = 0; | |
562 | pmac->u[7] = 0; | |
563 | ||
564 | for (res = sctx->md.num, j = 0; j < len; j++) { | |
565 | size_t c = out[j]; | |
566 | mask = (j - inp_len) >> (sizeof(j) * 8 - 8); | |
567 | c &= mask; | |
568 | c |= 0x80 & ~mask & ~((inp_len - j) >> (sizeof(j) * 8 - 8)); | |
569 | data->c[res++] = (unsigned char)c; | |
570 | ||
571 | if (res != SHA256_CBLOCK) | |
572 | continue; | |
573 | ||
574 | /* j is not incremented yet */ | |
575 | mask = 0 - ((inp_len + 7 - j) >> (sizeof(j) * 8 - 1)); | |
576 | data->u[SHA_LBLOCK - 1] |= bitlen & mask; | |
577 | sha256_block_data_order(&sctx->md, data, 1); | |
578 | mask &= 0 - ((j - inp_len - 72) >> (sizeof(j) * 8 - 1)); | |
579 | pmac->u[0] |= sctx->md.h[0] & mask; | |
580 | pmac->u[1] |= sctx->md.h[1] & mask; | |
581 | pmac->u[2] |= sctx->md.h[2] & mask; | |
582 | pmac->u[3] |= sctx->md.h[3] & mask; | |
583 | pmac->u[4] |= sctx->md.h[4] & mask; | |
584 | pmac->u[5] |= sctx->md.h[5] & mask; | |
585 | pmac->u[6] |= sctx->md.h[6] & mask; | |
586 | pmac->u[7] |= sctx->md.h[7] & mask; | |
587 | res = 0; | |
588 | } | |
589 | ||
590 | for (i = res; i < SHA256_CBLOCK; i++, j++) | |
591 | data->c[i] = 0; | |
592 | ||
593 | if (res > SHA256_CBLOCK - 8) { | |
594 | mask = 0 - ((inp_len + 8 - j) >> (sizeof(j) * 8 - 1)); | |
595 | data->u[SHA_LBLOCK - 1] |= bitlen & mask; | |
596 | sha256_block_data_order(&sctx->md, data, 1); | |
597 | mask &= 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1)); | |
598 | pmac->u[0] |= sctx->md.h[0] & mask; | |
599 | pmac->u[1] |= sctx->md.h[1] & mask; | |
600 | pmac->u[2] |= sctx->md.h[2] & mask; | |
601 | pmac->u[3] |= sctx->md.h[3] & mask; | |
602 | pmac->u[4] |= sctx->md.h[4] & mask; | |
603 | pmac->u[5] |= sctx->md.h[5] & mask; | |
604 | pmac->u[6] |= sctx->md.h[6] & mask; | |
605 | pmac->u[7] |= sctx->md.h[7] & mask; | |
606 | ||
607 | memset(data, 0, SHA256_CBLOCK); | |
608 | j += 64; | |
609 | } | |
610 | data->u[SHA_LBLOCK - 1] = bitlen; | |
611 | sha256_block_data_order(&sctx->md, data, 1); | |
612 | mask = 0 - ((j - inp_len - 73) >> (sizeof(j) * 8 - 1)); | |
613 | pmac->u[0] |= sctx->md.h[0] & mask; | |
614 | pmac->u[1] |= sctx->md.h[1] & mask; | |
615 | pmac->u[2] |= sctx->md.h[2] & mask; | |
616 | pmac->u[3] |= sctx->md.h[3] & mask; | |
617 | pmac->u[4] |= sctx->md.h[4] & mask; | |
618 | pmac->u[5] |= sctx->md.h[5] & mask; | |
619 | pmac->u[6] |= sctx->md.h[6] & mask; | |
620 | pmac->u[7] |= sctx->md.h[7] & mask; | |
621 | ||
622 | # ifdef BSWAP4 | |
623 | pmac->u[0] = BSWAP4(pmac->u[0]); | |
624 | pmac->u[1] = BSWAP4(pmac->u[1]); | |
625 | pmac->u[2] = BSWAP4(pmac->u[2]); | |
626 | pmac->u[3] = BSWAP4(pmac->u[3]); | |
627 | pmac->u[4] = BSWAP4(pmac->u[4]); | |
628 | pmac->u[5] = BSWAP4(pmac->u[5]); | |
629 | pmac->u[6] = BSWAP4(pmac->u[6]); | |
630 | pmac->u[7] = BSWAP4(pmac->u[7]); | |
631 | # else | |
632 | for (i = 0; i < 8; i++) { | |
633 | res = pmac->u[i]; | |
634 | pmac->c[4 * i + 0] = (unsigned char)(res >> 24); | |
635 | pmac->c[4 * i + 1] = (unsigned char)(res >> 16); | |
636 | pmac->c[4 * i + 2] = (unsigned char)(res >> 8); | |
637 | pmac->c[4 * i + 3] = (unsigned char)res; | |
638 | } | |
639 | # endif /* BSWAP */ | |
640 | len += SHA256_DIGEST_LENGTH; | |
641 | sctx->md = sctx->tail; | |
642 | sha256_update(&sctx->md, pmac->c, SHA256_DIGEST_LENGTH); | |
643 | SHA256_Final(pmac->c, &sctx->md); | |
644 | ||
645 | /* verify HMAC */ | |
646 | out += inp_len; | |
647 | len -= inp_len; | |
648 | /* code containing lucky-13 fix */ | |
649 | { | |
650 | unsigned char *p = | |
651 | out + len - 1 - maxpad - SHA256_DIGEST_LENGTH; | |
652 | size_t off = out - p; | |
653 | unsigned int c, cmask; | |
654 | ||
655 | maxpad += SHA256_DIGEST_LENGTH; | |
656 | for (res = 0, i = 0, j = 0; j < maxpad; j++) { | |
657 | c = p[j]; | |
658 | cmask = | |
659 | ((int)(j - off - SHA256_DIGEST_LENGTH)) >> | |
660 | (sizeof(int) * 8 - 1); | |
661 | res |= (c ^ pad) & ~cmask; /* ... and padding */ | |
662 | cmask &= ((int)(off - 1 - j)) >> (sizeof(int) * 8 - 1); | |
663 | res |= (c ^ pmac->c[i]) & cmask; | |
664 | i += 1 & cmask; | |
665 | } | |
666 | maxpad -= SHA256_DIGEST_LENGTH; | |
667 | ||
668 | res = 0 - ((0 - res) >> (sizeof(res) * 8 - 1)); | |
669 | ret &= (int)~res; | |
670 | } | |
671 | return ret; | |
672 | } else { | |
673 | sha256_update(&sctx->md, out, len); | |
674 | } | |
675 | } | |
676 | ||
677 | return 1; | |
678 | } | |
679 | ||
680 | /* EVP_CTRL_AEAD_SET_MAC_KEY */ | |
681 | static void aesni_cbc_hmac_sha256_set_mac_key(void *vctx, | |
682 | const unsigned char *mackey, | |
683 | size_t len) | |
684 | { | |
685 | PROV_AES_HMAC_SHA256_CTX *ctx = (PROV_AES_HMAC_SHA256_CTX *)vctx; | |
686 | unsigned int i; | |
687 | unsigned char hmac_key[64]; | |
688 | ||
689 | memset(hmac_key, 0, sizeof(hmac_key)); | |
690 | ||
691 | if (len > sizeof(hmac_key)) { | |
692 | SHA256_Init(&ctx->head); | |
693 | sha256_update(&ctx->head, mackey, len); | |
694 | SHA256_Final(hmac_key, &ctx->head); | |
695 | } else { | |
696 | memcpy(hmac_key, mackey, len); | |
697 | } | |
698 | ||
699 | for (i = 0; i < sizeof(hmac_key); i++) | |
700 | hmac_key[i] ^= 0x36; /* ipad */ | |
701 | SHA256_Init(&ctx->head); | |
702 | sha256_update(&ctx->head, hmac_key, sizeof(hmac_key)); | |
703 | ||
704 | for (i = 0; i < sizeof(hmac_key); i++) | |
705 | hmac_key[i] ^= 0x36 ^ 0x5c; /* opad */ | |
706 | SHA256_Init(&ctx->tail); | |
707 | sha256_update(&ctx->tail, hmac_key, sizeof(hmac_key)); | |
708 | ||
709 | OPENSSL_cleanse(hmac_key, sizeof(hmac_key)); | |
710 | } | |
711 | ||
712 | /* EVP_CTRL_AEAD_TLS1_AAD */ | |
713 | static int aesni_cbc_hmac_sha256_set_tls1_aad(void *vctx, | |
714 | unsigned char *aad_rec, int aad_len) | |
715 | { | |
716 | PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx; | |
717 | PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx; | |
718 | unsigned char *p = aad_rec; | |
719 | unsigned int len; | |
720 | ||
721 | if (aad_len != EVP_AEAD_TLS1_AAD_LEN) | |
722 | return -1; | |
723 | ||
724 | len = p[aad_len - 2] << 8 | p[aad_len - 1]; | |
725 | ||
726 | if (ctx->base.enc) { | |
727 | ctx->payload_length = len; | |
728 | if ((ctx->aux.tls_ver = | |
729 | p[aad_len - 4] << 8 | p[aad_len - 3]) >= TLS1_1_VERSION) { | |
730 | if (len < AES_BLOCK_SIZE) | |
731 | return 0; | |
732 | len -= AES_BLOCK_SIZE; | |
733 | p[aad_len] = len >> 8; | |
734 | p[aad_len - 1] = len; | |
735 | } | |
736 | sctx->md = sctx->head; | |
737 | sha256_update(&sctx->md, p, aad_len); | |
738 | ctx->tls_aad_pad = (int)(((len + SHA256_DIGEST_LENGTH + | |
739 | AES_BLOCK_SIZE) & -AES_BLOCK_SIZE) | |
740 | - len); | |
741 | return 1; | |
742 | } else { | |
743 | memcpy(ctx->aux.tls_aad, p, aad_len); | |
744 | ctx->payload_length = aad_len; | |
745 | ctx->tls_aad_pad = SHA256_DIGEST_LENGTH; | |
746 | return 1; | |
747 | } | |
748 | } | |
749 | ||
750 | # if !defined(OPENSSL_NO_MULTIBLOCK) | |
751 | /* EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE */ | |
752 | static int aesni_cbc_hmac_sha256_tls1_multiblock_max_bufsize( | |
753 | void *vctx) | |
754 | { | |
755 | PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx; | |
756 | ||
757 | OPENSSL_assert(ctx->multiblock_max_send_fragment != 0); | |
758 | return (int)(5 + 16 | |
759 | + (((int)ctx->multiblock_max_send_fragment + 32 + 16) & -16)); | |
760 | } | |
761 | ||
762 | /* EVP_CTRL_TLS1_1_MULTIBLOCK_AAD */ | |
763 | static int aesni_cbc_hmac_sha256_tls1_multiblock_aad( | |
764 | void *vctx, EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param) | |
765 | { | |
766 | PROV_AES_HMAC_SHA_CTX *ctx = (PROV_AES_HMAC_SHA_CTX *)vctx; | |
767 | PROV_AES_HMAC_SHA256_CTX *sctx = (PROV_AES_HMAC_SHA256_CTX *)vctx; | |
768 | unsigned int n4x = 1, x4; | |
769 | unsigned int frag, last, packlen, inp_len; | |
770 | ||
771 | inp_len = param->inp[11] << 8 | param->inp[12]; | |
772 | ||
773 | if (ctx->base.enc) { | |
774 | if ((param->inp[9] << 8 | param->inp[10]) < TLS1_1_VERSION) | |
775 | return -1; | |
776 | ||
777 | if (inp_len) { | |
778 | if (inp_len < 4096) | |
779 | return 0; /* too short */ | |
780 | ||
781 | if (inp_len >= 8192 && OPENSSL_ia32cap_P[2] & (1 << 5)) | |
782 | n4x = 2; /* AVX2 */ | |
783 | } else if ((n4x = param->interleave / 4) && n4x <= 2) | |
784 | inp_len = param->len; | |
785 | else | |
786 | return -1; | |
787 | ||
788 | sctx->md = sctx->head; | |
789 | sha256_update(&sctx->md, param->inp, 13); | |
790 | ||
791 | x4 = 4 * n4x; | |
792 | n4x += 1; | |
793 | ||
794 | frag = inp_len >> n4x; | |
795 | last = inp_len + frag - (frag << n4x); | |
796 | if (last > frag && ((last + 13 + 9) % 64 < (x4 - 1))) { | |
797 | frag++; | |
798 | last -= x4 - 1; | |
799 | } | |
800 | ||
801 | packlen = 5 + 16 + ((frag + 32 + 16) & -16); | |
802 | packlen = (packlen << n4x) - packlen; | |
803 | packlen += 5 + 16 + ((last + 32 + 16) & -16); | |
804 | ||
805 | param->interleave = x4; | |
806 | /* The returned values used by get need to be stored */ | |
807 | ctx->multiblock_interleave = x4; | |
808 | ctx->multiblock_aad_packlen = packlen; | |
809 | return 1; | |
810 | } | |
811 | return -1; /* not yet */ | |
812 | } | |
813 | ||
814 | /* EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT */ | |
815 | static int aesni_cbc_hmac_sha256_tls1_multiblock_encrypt( | |
816 | void *ctx, EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param) | |
817 | { | |
818 | return (int)tls1_multi_block_encrypt(ctx, param->out, | |
819 | param->inp, param->len, | |
820 | param->interleave / 4); | |
821 | } | |
87d3bb8e | 822 | # endif |
0d2bfe52 SL |
823 | |
824 | static const PROV_CIPHER_HW_AES_HMAC_SHA cipher_hw_aes_hmac_sha256 = { | |
825 | { | |
826 | aesni_cbc_hmac_sha256_init_key, | |
827 | aesni_cbc_hmac_sha256_cipher | |
828 | }, | |
829 | aesni_cbc_hmac_sha256_set_mac_key, | |
830 | aesni_cbc_hmac_sha256_set_tls1_aad, | |
831 | # if !defined(OPENSSL_NO_MULTIBLOCK) | |
832 | aesni_cbc_hmac_sha256_tls1_multiblock_max_bufsize, | |
833 | aesni_cbc_hmac_sha256_tls1_multiblock_aad, | |
834 | aesni_cbc_hmac_sha256_tls1_multiblock_encrypt | |
835 | # endif | |
836 | }; | |
837 | ||
838 | const PROV_CIPHER_HW_AES_HMAC_SHA *PROV_CIPHER_HW_aes_cbc_hmac_sha256(void) | |
839 | { | |
840 | return &cipher_hw_aes_hmac_sha256; | |
841 | } | |
842 | ||
87d3bb8e | 843 | #endif /* !defined(AES_CBC_HMAC_SHA_CAPABLE) || !defined(AESNI_CAPABLE) */ |