1 |
|
|
/* $OpenBSD: s3_cbc.c,v 1.16 2017/01/23 08:08:06 beck Exp $ */ |
2 |
|
|
/* ==================================================================== |
3 |
|
|
* Copyright (c) 2012 The OpenSSL Project. All rights reserved. |
4 |
|
|
* |
5 |
|
|
* Redistribution and use in source and binary forms, with or without |
6 |
|
|
* modification, are permitted provided that the following conditions |
7 |
|
|
* are met: |
8 |
|
|
* |
9 |
|
|
* 1. Redistributions of source code must retain the above copyright |
10 |
|
|
* notice, this list of conditions and the following disclaimer. |
11 |
|
|
* |
12 |
|
|
* 2. Redistributions in binary form must reproduce the above copyright |
13 |
|
|
* notice, this list of conditions and the following disclaimer in |
14 |
|
|
* the documentation and/or other materials provided with the |
15 |
|
|
* distribution. |
16 |
|
|
* |
17 |
|
|
* 3. All advertising materials mentioning features or use of this |
18 |
|
|
* software must display the following acknowledgment: |
19 |
|
|
* "This product includes software developed by the OpenSSL Project |
20 |
|
|
* for use in the OpenSSL Toolkit. (http://www.openssl.org/)" |
21 |
|
|
* |
22 |
|
|
* 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to |
23 |
|
|
* endorse or promote products derived from this software without |
24 |
|
|
* prior written permission. For written permission, please contact |
25 |
|
|
* openssl-core@openssl.org. |
26 |
|
|
* |
27 |
|
|
* 5. Products derived from this software may not be called "OpenSSL" |
28 |
|
|
* nor may "OpenSSL" appear in their names without prior written |
29 |
|
|
* permission of the OpenSSL Project. |
30 |
|
|
* |
31 |
|
|
* 6. Redistributions of any form whatsoever must retain the following |
32 |
|
|
* acknowledgment: |
33 |
|
|
* "This product includes software developed by the OpenSSL Project |
34 |
|
|
* for use in the OpenSSL Toolkit (http://www.openssl.org/)" |
35 |
|
|
* |
36 |
|
|
* THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY |
37 |
|
|
* EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
38 |
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
39 |
|
|
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR |
40 |
|
|
* ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
41 |
|
|
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
42 |
|
|
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
43 |
|
|
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
44 |
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, |
45 |
|
|
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
46 |
|
|
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
47 |
|
|
* OF THE POSSIBILITY OF SUCH DAMAGE. |
48 |
|
|
* ==================================================================== |
49 |
|
|
* |
50 |
|
|
* This product includes cryptographic software written by Eric Young |
51 |
|
|
* (eay@cryptsoft.com). This product includes software written by Tim |
52 |
|
|
* Hudson (tjh@cryptsoft.com). |
53 |
|
|
* |
54 |
|
|
*/ |
55 |
|
|
|
56 |
|
|
#include "ssl_locl.h" |
57 |
|
|
|
58 |
|
|
#include <openssl/md5.h> |
59 |
|
|
#include <openssl/sha.h> |
60 |
|
|
|
61 |
|
|
/* MAX_HASH_BIT_COUNT_BYTES is the maximum number of bytes in the hash's length |
62 |
|
|
* field. (SHA-384/512 have 128-bit length.) */ |
63 |
|
|
#define MAX_HASH_BIT_COUNT_BYTES 16 |
64 |
|
|
|
65 |
|
|
/* MAX_HASH_BLOCK_SIZE is the maximum hash block size that we'll support. |
66 |
|
|
* Currently SHA-384/512 has a 128-byte block size and that's the largest |
67 |
|
|
* supported by TLS.) */ |
68 |
|
|
#define MAX_HASH_BLOCK_SIZE 128 |
69 |
|
|
|
70 |
|
|
/* Some utility functions are needed: |
71 |
|
|
* |
72 |
|
|
* These macros return the given value with the MSB copied to all the other |
73 |
|
|
* bits. They use the fact that arithmetic shift shifts-in the sign bit. |
74 |
|
|
* However, this is not ensured by the C standard so you may need to replace |
75 |
|
|
* them with something else on odd CPUs. */ |
76 |
|
|
#define DUPLICATE_MSB_TO_ALL(x) ((unsigned)((int)(x) >> (sizeof(int) * 8 - 1))) |
77 |
|
|
#define DUPLICATE_MSB_TO_ALL_8(x) ((unsigned char)(DUPLICATE_MSB_TO_ALL(x))) |
78 |
|
|
|
79 |
|
|
/* constant_time_lt returns 0xff if a<b and 0x00 otherwise. */ |
80 |
|
|
static unsigned |
81 |
|
|
constant_time_lt(unsigned a, unsigned b) |
82 |
|
|
{ |
83 |
|
125824 |
a -= b; |
84 |
|
62912 |
return DUPLICATE_MSB_TO_ALL(a); |
85 |
|
|
} |
86 |
|
|
|
87 |
|
|
/* constant_time_ge returns 0xff if a>=b and 0x00 otherwise. */ |
88 |
|
|
static unsigned |
89 |
|
|
constant_time_ge(unsigned a, unsigned b) |
90 |
|
|
{ |
91 |
|
919552 |
a -= b; |
92 |
|
459776 |
return DUPLICATE_MSB_TO_ALL(~a); |
93 |
|
|
} |
94 |
|
|
|
95 |
|
|
/* constant_time_eq_8 returns 0xff if a==b and 0x00 otherwise. */ |
96 |
|
|
static unsigned char |
97 |
|
|
constant_time_eq_8(unsigned a, unsigned b) |
98 |
|
|
{ |
99 |
|
8960 |
unsigned c = a ^ b; |
100 |
|
4480 |
c--; |
101 |
|
4480 |
return DUPLICATE_MSB_TO_ALL_8(c); |
102 |
|
|
} |
103 |
|
|
|
104 |
|
|
/* tls1_cbc_remove_padding removes the CBC padding from the decrypted, TLS, CBC |
105 |
|
|
* record in |rec| in constant time and returns 1 if the padding is valid and |
106 |
|
|
* -1 otherwise. It also removes any explicit IV from the start of the record |
107 |
|
|
* without leaking any timing about whether there was enough space after the |
108 |
|
|
* padding was removed. |
109 |
|
|
* |
110 |
|
|
* block_size: the block size of the cipher used to encrypt the record. |
111 |
|
|
* returns: |
112 |
|
|
* 0: (in non-constant time) if the record is publicly invalid. |
113 |
|
|
* 1: if the padding was valid |
114 |
|
|
* -1: otherwise. */ |
115 |
|
|
int |
116 |
|
|
tls1_cbc_remove_padding(const SSL* s, SSL3_RECORD *rec, unsigned block_size, |
117 |
|
|
unsigned mac_size) |
118 |
|
|
{ |
119 |
|
|
unsigned padding_length, good, to_check, i; |
120 |
|
1504 |
const unsigned overhead = 1 /* padding length byte */ + mac_size; |
121 |
|
|
|
122 |
|
|
/* Check if version requires explicit IV */ |
123 |
✓✓ |
752 |
if (SSL_USE_EXPLICIT_IV(s)) { |
124 |
|
|
/* These lengths are all public so we can test them in |
125 |
|
|
* non-constant time. |
126 |
|
|
*/ |
127 |
✗✓ |
320 |
if (overhead + block_size > rec->length) |
128 |
|
|
return 0; |
129 |
|
|
/* We can now safely skip explicit IV */ |
130 |
|
320 |
rec->data += block_size; |
131 |
|
320 |
rec->input += block_size; |
132 |
|
320 |
rec->length -= block_size; |
133 |
✗✓ |
752 |
} else if (overhead > rec->length) |
134 |
|
|
return 0; |
135 |
|
|
|
136 |
|
752 |
padding_length = rec->data[rec->length - 1]; |
137 |
|
|
|
138 |
✓✓ |
752 |
if (EVP_CIPHER_flags(s->enc_read_ctx->cipher) & EVP_CIPH_FLAG_AEAD_CIPHER) { |
139 |
|
|
/* padding is already verified */ |
140 |
|
432 |
rec->length -= padding_length + 1; |
141 |
|
432 |
return 1; |
142 |
|
|
} |
143 |
|
|
|
144 |
|
320 |
good = constant_time_ge(rec->length, overhead + padding_length); |
145 |
|
|
/* The padding consists of a length byte at the end of the record and |
146 |
|
|
* then that many bytes of padding, all with the same value as the |
147 |
|
|
* length byte. Thus, with the length byte included, there are i+1 |
148 |
|
|
* bytes of padding. |
149 |
|
|
* |
150 |
|
|
* We can't check just |padding_length+1| bytes because that leaks |
151 |
|
|
* decrypted information. Therefore we always have to check the maximum |
152 |
|
|
* amount of padding possible. (Again, the length of the record is |
153 |
|
|
* public information so we can use it.) */ |
154 |
|
|
to_check = 255; /* maximum amount of padding. */ |
155 |
✓✓ |
320 |
if (to_check > rec->length - 1) |
156 |
|
160 |
to_check = rec->length - 1; |
157 |
|
|
|
158 |
✓✓ |
100096 |
for (i = 0; i < to_check; i++) { |
159 |
|
49728 |
unsigned char mask = constant_time_ge(padding_length, i); |
160 |
|
49728 |
unsigned char b = rec->data[rec->length - 1 - i]; |
161 |
|
|
/* The final |padding_length+1| bytes should all have the value |
162 |
|
|
* |padding_length|. Therefore the XOR should be zero. */ |
163 |
|
49728 |
good &= ~(mask&(padding_length ^ b)); |
164 |
|
|
} |
165 |
|
|
|
166 |
|
|
/* If any of the final |padding_length+1| bytes had the wrong value, |
167 |
|
|
* one or more of the lower eight bits of |good| will be cleared. We |
168 |
|
|
* AND the bottom 8 bits together and duplicate the result to all the |
169 |
|
|
* bits. */ |
170 |
|
320 |
good &= good >> 4; |
171 |
|
320 |
good &= good >> 2; |
172 |
|
320 |
good &= good >> 1; |
173 |
|
320 |
good <<= sizeof(good)*8 - 1; |
174 |
|
320 |
good = DUPLICATE_MSB_TO_ALL(good); |
175 |
|
|
|
176 |
|
320 |
padding_length = good & (padding_length + 1); |
177 |
|
320 |
rec->length -= padding_length; |
178 |
|
320 |
rec->type |= padding_length<<8; /* kludge: pass padding length */ |
179 |
|
|
|
180 |
|
320 |
return (int)((good & 1) | (~good & -1)); |
181 |
|
752 |
} |
182 |
|
|
|
183 |
|
|
/* ssl3_cbc_copy_mac copies |md_size| bytes from the end of |rec| to |out| in |
184 |
|
|
* constant time (independent of the concrete value of rec->length, which may |
185 |
|
|
* vary within a 256-byte window). |
186 |
|
|
* |
187 |
|
|
* ssl3_cbc_remove_padding or tls1_cbc_remove_padding must be called prior to |
188 |
|
|
* this function. |
189 |
|
|
* |
190 |
|
|
* On entry: |
191 |
|
|
* rec->orig_len >= md_size |
192 |
|
|
* md_size <= EVP_MAX_MD_SIZE |
193 |
|
|
* |
194 |
|
|
* If CBC_MAC_ROTATE_IN_PLACE is defined then the rotation is performed with |
195 |
|
|
* variable accesses in a 64-byte-aligned buffer. Assuming that this fits into |
196 |
|
|
* a single or pair of cache-lines, then the variable memory accesses don't |
197 |
|
|
* actually affect the timing. CPUs with smaller cache-lines [if any] are |
198 |
|
|
* not multi-core and are not considered vulnerable to cache-timing attacks. |
199 |
|
|
*/ |
200 |
|
|
#define CBC_MAC_ROTATE_IN_PLACE |
201 |
|
|
|
202 |
|
|
void |
203 |
|
|
ssl3_cbc_copy_mac(unsigned char* out, const SSL3_RECORD *rec, |
204 |
|
|
unsigned md_size, unsigned orig_len) |
205 |
|
|
{ |
206 |
|
|
#if defined(CBC_MAC_ROTATE_IN_PLACE) |
207 |
|
640 |
unsigned char rotated_mac_buf[64 + EVP_MAX_MD_SIZE]; |
208 |
|
|
unsigned char *rotated_mac; |
209 |
|
|
#else |
210 |
|
|
unsigned char rotated_mac[EVP_MAX_MD_SIZE]; |
211 |
|
|
#endif |
212 |
|
|
|
213 |
|
|
/* mac_end is the index of |rec->data| just after the end of the MAC. */ |
214 |
|
320 |
unsigned mac_end = rec->length; |
215 |
|
320 |
unsigned mac_start = mac_end - md_size; |
216 |
|
|
/* scan_start contains the number of bytes that we can ignore because |
217 |
|
|
* the MAC's position can only vary by 255 bytes. */ |
218 |
|
|
unsigned scan_start = 0; |
219 |
|
|
unsigned i, j; |
220 |
|
|
unsigned div_spoiler; |
221 |
|
|
unsigned rotate_offset; |
222 |
|
|
|
223 |
✗✓ |
320 |
OPENSSL_assert(orig_len >= md_size); |
224 |
✗✓ |
320 |
OPENSSL_assert(md_size <= EVP_MAX_MD_SIZE); |
225 |
|
|
|
226 |
|
|
#if defined(CBC_MAC_ROTATE_IN_PLACE) |
227 |
|
320 |
rotated_mac = rotated_mac_buf + ((0 - (size_t)rotated_mac_buf)&63); |
228 |
|
|
#endif |
229 |
|
|
|
230 |
|
|
/* This information is public so it's safe to branch based on it. */ |
231 |
✓✓ |
320 |
if (orig_len > md_size + 255 + 1) |
232 |
|
160 |
scan_start = orig_len - (md_size + 255 + 1); |
233 |
|
|
/* div_spoiler contains a multiple of md_size that is used to cause the |
234 |
|
|
* modulo operation to be constant time. Without this, the time varies |
235 |
|
|
* based on the amount of padding when running on Intel chips at least. |
236 |
|
|
* |
237 |
|
|
* The aim of right-shifting md_size is so that the compiler doesn't |
238 |
|
|
* figure out that it can remove div_spoiler as that would require it |
239 |
|
|
* to prove that md_size is always even, which I hope is beyond it. */ |
240 |
|
320 |
div_spoiler = md_size >> 1; |
241 |
|
320 |
div_spoiler <<= (sizeof(div_spoiler) - 1) * 8; |
242 |
|
320 |
rotate_offset = (div_spoiler + mac_start - scan_start) % md_size; |
243 |
|
|
|
244 |
|
320 |
memset(rotated_mac, 0, md_size); |
245 |
✓✓ |
109312 |
for (i = scan_start, j = 0; i < orig_len; i++) { |
246 |
|
54336 |
unsigned char mac_started = constant_time_ge(i, mac_start); |
247 |
|
54336 |
unsigned char mac_ended = constant_time_ge(i, mac_end); |
248 |
|
54336 |
unsigned char b = rec->data[i]; |
249 |
|
54336 |
rotated_mac[j++] |= b & mac_started & ~mac_ended; |
250 |
|
54336 |
j &= constant_time_lt(j, md_size); |
251 |
|
|
} |
252 |
|
|
|
253 |
|
|
/* Now rotate the MAC */ |
254 |
|
|
#if defined(CBC_MAC_ROTATE_IN_PLACE) |
255 |
|
|
j = 0; |
256 |
✓✓ |
17792 |
for (i = 0; i < md_size; i++) { |
257 |
|
|
/* in case cache-line is 32 bytes, touch second line */ |
258 |
|
8576 |
((volatile unsigned char *)rotated_mac)[rotate_offset^32]; |
259 |
|
8576 |
out[j++] = rotated_mac[rotate_offset++]; |
260 |
|
8576 |
rotate_offset &= constant_time_lt(rotate_offset, md_size); |
261 |
|
|
} |
262 |
|
|
#else |
263 |
|
|
memset(out, 0, md_size); |
264 |
|
|
rotate_offset = md_size - rotate_offset; |
265 |
|
|
rotate_offset &= constant_time_lt(rotate_offset, md_size); |
266 |
|
|
for (i = 0; i < md_size; i++) { |
267 |
|
|
for (j = 0; j < md_size; j++) |
268 |
|
|
out[j] |= rotated_mac[i] & constant_time_eq_8(j, rotate_offset); |
269 |
|
|
rotate_offset++; |
270 |
|
|
rotate_offset &= constant_time_lt(rotate_offset, md_size); |
271 |
|
|
} |
272 |
|
|
#endif |
273 |
|
320 |
} |
274 |
|
|
|
275 |
|
|
/* u32toLE serialises an unsigned, 32-bit number (n) as four bytes at (p) in |
276 |
|
|
* little-endian order. The value of p is advanced by four. */ |
277 |
|
|
#define u32toLE(n, p) \ |
278 |
|
|
(*((p)++)=(unsigned char)(n), \ |
279 |
|
|
*((p)++)=(unsigned char)(n>>8), \ |
280 |
|
|
*((p)++)=(unsigned char)(n>>16), \ |
281 |
|
|
*((p)++)=(unsigned char)(n>>24)) |
282 |
|
|
|
283 |
|
|
/* These functions serialize the state of a hash and thus perform the standard |
284 |
|
|
* "final" operation without adding the padding and length that such a function |
285 |
|
|
* typically does. */ |
286 |
|
|
static void |
287 |
|
|
tls1_md5_final_raw(void* ctx, unsigned char *md_out) |
288 |
|
|
{ |
289 |
|
|
MD5_CTX *md5 = ctx; |
290 |
|
|
u32toLE(md5->A, md_out); |
291 |
|
|
u32toLE(md5->B, md_out); |
292 |
|
|
u32toLE(md5->C, md_out); |
293 |
|
|
u32toLE(md5->D, md_out); |
294 |
|
|
} |
295 |
|
|
|
296 |
|
|
static void |
297 |
|
|
tls1_sha1_final_raw(void* ctx, unsigned char *md_out) |
298 |
|
|
{ |
299 |
|
2240 |
SHA_CTX *sha1 = ctx; |
300 |
|
1120 |
l2n(sha1->h0, md_out); |
301 |
|
1120 |
l2n(sha1->h1, md_out); |
302 |
|
1120 |
l2n(sha1->h2, md_out); |
303 |
|
1120 |
l2n(sha1->h3, md_out); |
304 |
|
1120 |
l2n(sha1->h4, md_out); |
305 |
|
1120 |
} |
306 |
|
|
|
307 |
|
|
static void |
308 |
|
|
tls1_sha256_final_raw(void* ctx, unsigned char *md_out) |
309 |
|
|
{ |
310 |
|
2016 |
SHA256_CTX *sha256 = ctx; |
311 |
|
|
unsigned i; |
312 |
|
|
|
313 |
✓✓ |
18144 |
for (i = 0; i < 8; i++) { |
314 |
|
8064 |
l2n(sha256->h[i], md_out); |
315 |
|
|
} |
316 |
|
1008 |
} |
317 |
|
|
|
318 |
|
|
static void |
319 |
|
|
tls1_sha512_final_raw(void* ctx, unsigned char *md_out) |
320 |
|
|
{ |
321 |
|
224 |
SHA512_CTX *sha512 = ctx; |
322 |
|
|
unsigned i; |
323 |
|
|
|
324 |
✓✓ |
2016 |
for (i = 0; i < 8; i++) { |
325 |
|
896 |
l2n8(sha512->h[i], md_out); |
326 |
|
|
} |
327 |
|
112 |
} |
328 |
|
|
|
329 |
|
|
/* Largest hash context ever used by the functions above. */ |
330 |
|
|
#define LARGEST_DIGEST_CTX SHA512_CTX |
331 |
|
|
|
332 |
|
|
/* Type giving the alignment needed by the above */ |
333 |
|
|
#define LARGEST_DIGEST_CTX_ALIGNMENT SHA_LONG64 |
334 |
|
|
|
335 |
|
|
/* ssl3_cbc_record_digest_supported returns 1 iff |ctx| uses a hash function |
336 |
|
|
* which ssl3_cbc_digest_record supports. */ |
337 |
|
|
char |
338 |
|
|
ssl3_cbc_record_digest_supported(const EVP_MD_CTX *ctx) |
339 |
|
|
{ |
340 |
✗✗✗✗ ✗✓✗ |
640 |
switch (EVP_MD_CTX_type(ctx)) { |
341 |
|
|
case NID_md5: |
342 |
|
|
case NID_sha1: |
343 |
|
|
case NID_sha224: |
344 |
|
|
case NID_sha256: |
345 |
|
|
case NID_sha384: |
346 |
|
|
case NID_sha512: |
347 |
|
320 |
return 1; |
348 |
|
|
default: |
349 |
|
|
return 0; |
350 |
|
|
} |
351 |
|
320 |
} |
352 |
|
|
|
353 |
|
|
/* ssl3_cbc_digest_record computes the MAC of a decrypted, padded TLS |
354 |
|
|
* record. |
355 |
|
|
* |
356 |
|
|
* ctx: the EVP_MD_CTX from which we take the hash function. |
357 |
|
|
* ssl3_cbc_record_digest_supported must return true for this EVP_MD_CTX. |
358 |
|
|
* md_out: the digest output. At most EVP_MAX_MD_SIZE bytes will be written. |
359 |
|
|
* md_out_size: if non-NULL, the number of output bytes is written here. |
360 |
|
|
* header: the 13-byte, TLS record header. |
361 |
|
|
* data: the record data itself, less any preceeding explicit IV. |
362 |
|
|
* data_plus_mac_size: the secret, reported length of the data and MAC |
363 |
|
|
* once the padding has been removed. |
364 |
|
|
* data_plus_mac_plus_padding_size: the public length of the whole |
365 |
|
|
* record, including padding. |
366 |
|
|
* |
367 |
|
|
* On entry: by virtue of having been through one of the remove_padding |
368 |
|
|
* functions, above, we know that data_plus_mac_size is large enough to contain |
369 |
|
|
* a padding byte and MAC. (If the padding was invalid, it might contain the |
370 |
|
|
* padding too. ) |
371 |
|
|
*/ |
372 |
|
|
int |
373 |
|
|
ssl3_cbc_digest_record(const EVP_MD_CTX *ctx, unsigned char* md_out, |
374 |
|
|
size_t* md_out_size, const unsigned char header[13], |
375 |
|
|
const unsigned char *data, size_t data_plus_mac_size, |
376 |
|
|
size_t data_plus_mac_plus_padding_size, const unsigned char *mac_secret, |
377 |
|
|
unsigned mac_secret_length) |
378 |
|
|
{ |
379 |
|
640 |
union { |
380 |
|
|
/* |
381 |
|
|
* Alignment here is to allow this to be cast as SHA512_CTX |
382 |
|
|
* without losing alignment required by the 64-bit SHA_LONG64 |
383 |
|
|
* integer it contains. |
384 |
|
|
*/ |
385 |
|
|
LARGEST_DIGEST_CTX_ALIGNMENT align; |
386 |
|
|
unsigned char c[sizeof(LARGEST_DIGEST_CTX)]; |
387 |
|
|
} md_state; |
388 |
|
|
void (*md_final_raw)(void *ctx, unsigned char *md_out); |
389 |
|
|
void (*md_transform)(void *ctx, const unsigned char *block); |
390 |
|
|
unsigned md_size, md_block_size = 64; |
391 |
|
|
unsigned header_length, variance_blocks, |
392 |
|
|
len, max_mac_bytes, num_blocks, |
393 |
|
|
num_starting_blocks, k, mac_end_offset, c, index_a, index_b; |
394 |
|
|
unsigned int bits; /* at most 18 bits */ |
395 |
|
320 |
unsigned char length_bytes[MAX_HASH_BIT_COUNT_BYTES]; |
396 |
|
|
/* hmac_pad is the masked HMAC key. */ |
397 |
|
320 |
unsigned char hmac_pad[MAX_HASH_BLOCK_SIZE]; |
398 |
|
320 |
unsigned char first_block[MAX_HASH_BLOCK_SIZE]; |
399 |
|
320 |
unsigned char mac_out[EVP_MAX_MD_SIZE]; |
400 |
|
320 |
unsigned i, j, md_out_size_u; |
401 |
|
320 |
EVP_MD_CTX md_ctx; |
402 |
|
|
/* mdLengthSize is the number of bytes in the length field that terminates |
403 |
|
|
* the hash. */ |
404 |
|
|
unsigned md_length_size = 8; |
405 |
|
|
char length_is_big_endian = 1; |
406 |
|
|
|
407 |
|
|
/* This is a, hopefully redundant, check that allows us to forget about |
408 |
|
|
* many possible overflows later in this function. */ |
409 |
✗✓ |
320 |
OPENSSL_assert(data_plus_mac_plus_padding_size < 1024*1024); |
410 |
|
|
|
411 |
✗✓✗✓ ✓✗✗ |
320 |
switch (EVP_MD_CTX_type(ctx)) { |
412 |
|
|
case NID_md5: |
413 |
|
|
MD5_Init((MD5_CTX*)md_state.c); |
414 |
|
|
md_final_raw = tls1_md5_final_raw; |
415 |
|
|
md_transform = (void(*)(void *ctx, const unsigned char *block)) MD5_Transform; |
416 |
|
|
md_size = 16; |
417 |
|
|
length_is_big_endian = 0; |
418 |
|
|
break; |
419 |
|
|
case NID_sha1: |
420 |
|
160 |
SHA1_Init((SHA_CTX*)md_state.c); |
421 |
|
|
md_final_raw = tls1_sha1_final_raw; |
422 |
|
|
md_transform = (void(*)(void *ctx, const unsigned char *block)) SHA1_Transform; |
423 |
|
|
md_size = 20; |
424 |
|
160 |
break; |
425 |
|
|
case NID_sha224: |
426 |
|
|
SHA224_Init((SHA256_CTX*)md_state.c); |
427 |
|
|
md_final_raw = tls1_sha256_final_raw; |
428 |
|
|
md_transform = (void(*)(void *ctx, const unsigned char *block)) SHA256_Transform; |
429 |
|
|
md_size = 224/8; |
430 |
|
|
break; |
431 |
|
|
case NID_sha256: |
432 |
|
144 |
SHA256_Init((SHA256_CTX*)md_state.c); |
433 |
|
|
md_final_raw = tls1_sha256_final_raw; |
434 |
|
|
md_transform = (void(*)(void *ctx, const unsigned char *block)) SHA256_Transform; |
435 |
|
|
md_size = 32; |
436 |
|
144 |
break; |
437 |
|
|
case NID_sha384: |
438 |
|
16 |
SHA384_Init((SHA512_CTX*)md_state.c); |
439 |
|
|
md_final_raw = tls1_sha512_final_raw; |
440 |
|
|
md_transform = (void(*)(void *ctx, const unsigned char *block)) SHA512_Transform; |
441 |
|
|
md_size = 384/8; |
442 |
|
|
md_block_size = 128; |
443 |
|
|
md_length_size = 16; |
444 |
|
16 |
break; |
445 |
|
|
case NID_sha512: |
446 |
|
|
SHA512_Init((SHA512_CTX*)md_state.c); |
447 |
|
|
md_final_raw = tls1_sha512_final_raw; |
448 |
|
|
md_transform = (void(*)(void *ctx, const unsigned char *block)) SHA512_Transform; |
449 |
|
|
md_size = 64; |
450 |
|
|
md_block_size = 128; |
451 |
|
|
md_length_size = 16; |
452 |
|
|
break; |
453 |
|
|
default: |
454 |
|
|
/* ssl3_cbc_record_digest_supported should have been |
455 |
|
|
* called first to check that the hash function is |
456 |
|
|
* supported. */ |
457 |
|
|
OPENSSL_assert(0); |
458 |
|
|
if (md_out_size) |
459 |
|
|
*md_out_size = 0; |
460 |
|
|
return 0; |
461 |
|
|
} |
462 |
|
|
|
463 |
✗✓ |
320 |
OPENSSL_assert(md_length_size <= MAX_HASH_BIT_COUNT_BYTES); |
464 |
✗✓ |
320 |
OPENSSL_assert(md_block_size <= MAX_HASH_BLOCK_SIZE); |
465 |
✗✓ |
320 |
OPENSSL_assert(md_size <= EVP_MAX_MD_SIZE); |
466 |
|
|
|
467 |
|
|
header_length = 13; |
468 |
|
|
|
469 |
|
|
/* variance_blocks is the number of blocks of the hash that we have to |
470 |
|
|
* calculate in constant time because they could be altered by the |
471 |
|
|
* padding value. |
472 |
|
|
* |
473 |
|
|
* TLSv1 has MACs up to 48 bytes long (SHA-384) and the padding is not |
474 |
|
|
* required to be minimal. Therefore we say that the final six blocks |
475 |
|
|
* can vary based on the padding. |
476 |
|
|
* |
477 |
|
|
* Later in the function, if the message is short and there obviously |
478 |
|
|
* cannot be this many blocks then variance_blocks can be reduced. */ |
479 |
|
|
variance_blocks = 6; |
480 |
|
|
/* From now on we're dealing with the MAC, which conceptually has 13 |
481 |
|
|
* bytes of `header' before the start of the data (TLS) */ |
482 |
|
320 |
len = data_plus_mac_plus_padding_size + header_length; |
483 |
|
|
/* max_mac_bytes contains the maximum bytes of bytes in the MAC, including |
484 |
|
|
* |header|, assuming that there's no padding. */ |
485 |
|
320 |
max_mac_bytes = len - md_size - 1; |
486 |
|
|
/* num_blocks is the maximum number of hash blocks. */ |
487 |
|
320 |
num_blocks = (max_mac_bytes + 1 + md_length_size + md_block_size - 1) / md_block_size; |
488 |
|
|
/* In order to calculate the MAC in constant time we have to handle |
489 |
|
|
* the final blocks specially because the padding value could cause the |
490 |
|
|
* end to appear somewhere in the final |variance_blocks| blocks and we |
491 |
|
|
* can't leak where. However, |num_starting_blocks| worth of data can |
492 |
|
|
* be hashed right away because no padding value can affect whether |
493 |
|
|
* they are plaintext. */ |
494 |
|
|
num_starting_blocks = 0; |
495 |
|
|
/* k is the starting byte offset into the conceptual header||data where |
496 |
|
|
* we start processing. */ |
497 |
|
|
k = 0; |
498 |
|
|
/* mac_end_offset is the index just past the end of the data to be |
499 |
|
|
* MACed. */ |
500 |
|
320 |
mac_end_offset = data_plus_mac_size + header_length - md_size; |
501 |
|
|
/* c is the index of the 0x80 byte in the final hash block that |
502 |
|
|
* contains application data. */ |
503 |
|
320 |
c = mac_end_offset % md_block_size; |
504 |
|
|
/* index_a is the hash block number that contains the 0x80 terminating |
505 |
|
|
* value. */ |
506 |
|
320 |
index_a = mac_end_offset / md_block_size; |
507 |
|
|
/* index_b is the hash block number that contains the 64-bit hash |
508 |
|
|
* length, in bits. */ |
509 |
|
320 |
index_b = (mac_end_offset + md_length_size) / md_block_size; |
510 |
|
|
/* bits is the hash-length in bits. It includes the additional hash |
511 |
|
|
* block for the masked HMAC key. */ |
512 |
|
|
|
513 |
✗✓ |
320 |
if (num_blocks > variance_blocks) { |
514 |
|
|
num_starting_blocks = num_blocks - variance_blocks; |
515 |
|
|
k = md_block_size*num_starting_blocks; |
516 |
|
|
} |
517 |
|
|
|
518 |
|
320 |
bits = 8*mac_end_offset; |
519 |
|
|
/* Compute the initial HMAC block. */ |
520 |
|
320 |
bits += 8*md_block_size; |
521 |
|
320 |
memset(hmac_pad, 0, md_block_size); |
522 |
✗✓ |
320 |
OPENSSL_assert(mac_secret_length <= sizeof(hmac_pad)); |
523 |
|
320 |
memcpy(hmac_pad, mac_secret, mac_secret_length); |
524 |
✓✓ |
43648 |
for (i = 0; i < md_block_size; i++) |
525 |
|
21504 |
hmac_pad[i] ^= 0x36; |
526 |
|
|
|
527 |
|
320 |
md_transform(md_state.c, hmac_pad); |
528 |
|
|
|
529 |
✓✗ |
320 |
if (length_is_big_endian) { |
530 |
|
320 |
memset(length_bytes, 0, md_length_size - 4); |
531 |
|
320 |
length_bytes[md_length_size - 4] = (unsigned char)(bits >> 24); |
532 |
|
320 |
length_bytes[md_length_size - 3] = (unsigned char)(bits >> 16); |
533 |
|
320 |
length_bytes[md_length_size - 2] = (unsigned char)(bits >> 8); |
534 |
|
320 |
length_bytes[md_length_size - 1] = (unsigned char)bits; |
535 |
|
320 |
} else { |
536 |
|
|
memset(length_bytes, 0, md_length_size); |
537 |
|
|
length_bytes[md_length_size - 5] = (unsigned char)(bits >> 24); |
538 |
|
|
length_bytes[md_length_size - 6] = (unsigned char)(bits >> 16); |
539 |
|
|
length_bytes[md_length_size - 7] = (unsigned char)(bits >> 8); |
540 |
|
|
length_bytes[md_length_size - 8] = (unsigned char)bits; |
541 |
|
|
} |
542 |
|
|
|
543 |
✗✓ |
320 |
if (k > 0) { |
544 |
|
|
/* k is a multiple of md_block_size. */ |
545 |
|
|
memcpy(first_block, header, 13); |
546 |
|
|
memcpy(first_block + 13, data, md_block_size - 13); |
547 |
|
|
md_transform(md_state.c, first_block); |
548 |
|
|
for (i = 1; i < k/md_block_size; i++) |
549 |
|
|
md_transform(md_state.c, data + md_block_size*i - 13); |
550 |
|
|
} |
551 |
|
|
|
552 |
|
320 |
memset(mac_out, 0, sizeof(mac_out)); |
553 |
|
|
|
554 |
|
|
/* We now process the final hash blocks. For each block, we construct |
555 |
|
|
* it in constant time. If the |i==index_a| then we'll include the 0x80 |
556 |
|
|
* bytes and zero pad etc. For each block we selectively copy it, in |
557 |
|
|
* constant time, to |mac_out|. */ |
558 |
✓✓ |
5120 |
for (i = num_starting_blocks; i <= num_starting_blocks + variance_blocks; i++) { |
559 |
|
2240 |
unsigned char block[MAX_HASH_BLOCK_SIZE]; |
560 |
|
2240 |
unsigned char is_block_a = constant_time_eq_8(i, index_a); |
561 |
|
2240 |
unsigned char is_block_b = constant_time_eq_8(i, index_b); |
562 |
✓✓ |
305536 |
for (j = 0; j < md_block_size; j++) { |
563 |
|
|
unsigned char b = 0, is_past_c, is_past_cp1; |
564 |
✓✓ |
150528 |
if (k < header_length) |
565 |
|
4160 |
b = header[k]; |
566 |
✓✓ |
146368 |
else if (k < data_plus_mac_plus_padding_size + header_length) |
567 |
|
56448 |
b = data[k - header_length]; |
568 |
|
150528 |
k++; |
569 |
|
|
|
570 |
|
150528 |
is_past_c = is_block_a & constant_time_ge(j, c); |
571 |
|
150528 |
is_past_cp1 = is_block_a & constant_time_ge(j, c + 1); |
572 |
|
|
/* If this is the block containing the end of the |
573 |
|
|
* application data, and we are at the offset for the |
574 |
|
|
* 0x80 value, then overwrite b with 0x80. */ |
575 |
|
150528 |
b = (b&~is_past_c) | (0x80&is_past_c); |
576 |
|
|
/* If this is the block containing the end of the |
577 |
|
|
* application data and we're past the 0x80 value then |
578 |
|
|
* just write zero. */ |
579 |
|
150528 |
b = b&~is_past_cp1; |
580 |
|
|
/* If this is index_b (the final block), but not |
581 |
|
|
* index_a (the end of the data), then the 64-bit |
582 |
|
|
* length didn't fit into index_a and we're having to |
583 |
|
|
* add an extra block of zeros. */ |
584 |
|
150528 |
b &= ~is_block_b | is_block_a; |
585 |
|
|
|
586 |
|
|
/* The final bytes of one of the blocks contains the |
587 |
|
|
* length. */ |
588 |
✓✓ |
150528 |
if (j >= md_block_size - md_length_size) { |
589 |
|
|
/* If this is index_b, write a length byte. */ |
590 |
|
18816 |
b = (b&~is_block_b) | (is_block_b&length_bytes[j - (md_block_size - md_length_size)]); |
591 |
|
18816 |
} |
592 |
|
150528 |
block[j] = b; |
593 |
|
|
} |
594 |
|
|
|
595 |
|
2240 |
md_transform(md_state.c, block); |
596 |
|
2240 |
md_final_raw(md_state.c, block); |
597 |
|
|
/* If this is index_b, copy the hash value to |mac_out|. */ |
598 |
✓✓ |
124544 |
for (j = 0; j < md_size; j++) |
599 |
|
60032 |
mac_out[j] |= block[j]&is_block_b; |
600 |
|
2240 |
} |
601 |
|
|
|
602 |
|
320 |
EVP_MD_CTX_init(&md_ctx); |
603 |
✗✓ |
320 |
if (!EVP_DigestInit_ex(&md_ctx, ctx->digest, NULL /* engine */)) { |
604 |
|
|
EVP_MD_CTX_cleanup(&md_ctx); |
605 |
|
|
return 0; |
606 |
|
|
} |
607 |
|
|
|
608 |
|
|
/* Complete the HMAC in the standard manner. */ |
609 |
✓✓ |
43648 |
for (i = 0; i < md_block_size; i++) |
610 |
|
21504 |
hmac_pad[i] ^= 0x6a; |
611 |
|
|
|
612 |
|
320 |
EVP_DigestUpdate(&md_ctx, hmac_pad, md_block_size); |
613 |
|
320 |
EVP_DigestUpdate(&md_ctx, mac_out, md_size); |
614 |
|
|
|
615 |
|
320 |
EVP_DigestFinal(&md_ctx, md_out, &md_out_size_u); |
616 |
✓✗ |
320 |
if (md_out_size) |
617 |
|
320 |
*md_out_size = md_out_size_u; |
618 |
|
320 |
EVP_MD_CTX_cleanup(&md_ctx); |
619 |
|
|
|
620 |
|
320 |
return 1; |
621 |
|
320 |
} |