1 /* Copyright (c) 2019, Google Inc.
2  *
3  * Permission to use, copy, modify, and/or distribute this software for any
4  * purpose with or without fee is hereby granted, provided that the above
5  * copyright notice and this permission notice appear in all copies.
6  *
7  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
8  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
9  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
10  * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
11  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
12  * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
13  * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
14 
15 #include <openssl/base.h>
16 
17 #include "../../internal.h"
18 #include "internal.h"
19 
20 #if !defined(BORINGSSL_HAS_UINT128) && defined(OPENSSL_SSE2)
21 #include <emmintrin.h>
22 #endif
23 
24 
25 // This file contains a constant-time implementation of GHASH based on the notes
26 // in https://bearssl.org/constanttime.html#ghash-for-gcm and the reduction
27 // algorithm described in
28 // https://crypto.stanford.edu/RealWorldCrypto/slides/gueron.pdf.
29 //
30 // Unlike the BearSSL notes, we use uint128_t in the 64-bit implementation. Our
31 // primary compilers (clang, clang-cl, and gcc) all support it. MSVC will run
32 // the 32-bit implementation, but we can use its intrinsics if necessary.
33 
34 #if defined(BORINGSSL_HAS_UINT128)
35 
gcm_mul64_nohw(uint64_t * out_lo,uint64_t * out_hi,uint64_t a,uint64_t b)36 static void gcm_mul64_nohw(uint64_t *out_lo, uint64_t *out_hi, uint64_t a,
37                            uint64_t b) {
38   // One term every four bits means the largest term is 64/4 = 16, which barely
39   // overflows into the next term. Using one term every five bits would cost 25
40   // multiplications instead of 16. It is faster to mask off the bottom four
41   // bits of |a|, giving a largest term of 60/4 = 15, and apply the bottom bits
42   // separately.
43   uint64_t a0 = a & UINT64_C(0x1111111111111110);
44   uint64_t a1 = a & UINT64_C(0x2222222222222220);
45   uint64_t a2 = a & UINT64_C(0x4444444444444440);
46   uint64_t a3 = a & UINT64_C(0x8888888888888880);
47 
48   uint64_t b0 = b & UINT64_C(0x1111111111111111);
49   uint64_t b1 = b & UINT64_C(0x2222222222222222);
50   uint64_t b2 = b & UINT64_C(0x4444444444444444);
51   uint64_t b3 = b & UINT64_C(0x8888888888888888);
52 
53   uint128_t c0 = (a0 * (uint128_t)b0) ^ (a1 * (uint128_t)b3) ^
54                  (a2 * (uint128_t)b2) ^ (a3 * (uint128_t)b1);
55   uint128_t c1 = (a0 * (uint128_t)b1) ^ (a1 * (uint128_t)b0) ^
56                  (a2 * (uint128_t)b3) ^ (a3 * (uint128_t)b2);
57   uint128_t c2 = (a0 * (uint128_t)b2) ^ (a1 * (uint128_t)b1) ^
58                  (a2 * (uint128_t)b0) ^ (a3 * (uint128_t)b3);
59   uint128_t c3 = (a0 * (uint128_t)b3) ^ (a1 * (uint128_t)b2) ^
60                  (a2 * (uint128_t)b1) ^ (a3 * (uint128_t)b0);
61 
62   // Multiply the bottom four bits of |a| with |b|.
63   uint64_t a0_mask = UINT64_C(0) - (a & 1);
64   uint64_t a1_mask = UINT64_C(0) - ((a >> 1) & 1);
65   uint64_t a2_mask = UINT64_C(0) - ((a >> 2) & 1);
66   uint64_t a3_mask = UINT64_C(0) - ((a >> 3) & 1);
67   uint128_t extra = (a0_mask & b) ^ ((uint128_t)(a1_mask & b) << 1) ^
68                     ((uint128_t)(a2_mask & b) << 2) ^
69                     ((uint128_t)(a3_mask & b) << 3);
70 
71   *out_lo = (((uint64_t)c0) & UINT64_C(0x1111111111111111)) ^
72             (((uint64_t)c1) & UINT64_C(0x2222222222222222)) ^
73             (((uint64_t)c2) & UINT64_C(0x4444444444444444)) ^
74             (((uint64_t)c3) & UINT64_C(0x8888888888888888)) ^ ((uint64_t)extra);
75   *out_hi = (((uint64_t)(c0 >> 64)) & UINT64_C(0x1111111111111111)) ^
76             (((uint64_t)(c1 >> 64)) & UINT64_C(0x2222222222222222)) ^
77             (((uint64_t)(c2 >> 64)) & UINT64_C(0x4444444444444444)) ^
78             (((uint64_t)(c3 >> 64)) & UINT64_C(0x8888888888888888)) ^
79             ((uint64_t)(extra >> 64));
80 }
81 
82 #elif defined(OPENSSL_SSE2)
83 
gcm_mul32_nohw(uint32_t a,uint32_t b)84 static __m128i gcm_mul32_nohw(uint32_t a, uint32_t b) {
85   // One term every four bits means the largest term is 32/4 = 8, which does not
86   // overflow into the next term.
87   __m128i aa = _mm_setr_epi32(a, 0, a, 0);
88   __m128i bb = _mm_setr_epi32(b, 0, b, 0);
89 
90   __m128i a0a0 =
91       _mm_and_si128(aa, _mm_setr_epi32(0x11111111, 0, 0x11111111, 0));
92   __m128i a2a2 =
93       _mm_and_si128(aa, _mm_setr_epi32(0x44444444, 0, 0x44444444, 0));
94   __m128i b0b1 =
95       _mm_and_si128(bb, _mm_setr_epi32(0x11111111, 0, 0x22222222, 0));
96   __m128i b2b3 =
97       _mm_and_si128(bb, _mm_setr_epi32(0x44444444, 0, 0x88888888, 0));
98 
99   __m128i c0c1 =
100       _mm_xor_si128(_mm_mul_epu32(a0a0, b0b1), _mm_mul_epu32(a2a2, b2b3));
101   __m128i c2c3 =
102       _mm_xor_si128(_mm_mul_epu32(a2a2, b0b1), _mm_mul_epu32(a0a0, b2b3));
103 
104   __m128i a1a1 =
105       _mm_and_si128(aa, _mm_setr_epi32(0x22222222, 0, 0x22222222, 0));
106   __m128i a3a3 =
107       _mm_and_si128(aa, _mm_setr_epi32(0x88888888, 0, 0x88888888, 0));
108   __m128i b3b0 =
109       _mm_and_si128(bb, _mm_setr_epi32(0x88888888, 0, 0x11111111, 0));
110   __m128i b1b2 =
111       _mm_and_si128(bb, _mm_setr_epi32(0x22222222, 0, 0x44444444, 0));
112 
113   c0c1 = _mm_xor_si128(c0c1, _mm_mul_epu32(a1a1, b3b0));
114   c0c1 = _mm_xor_si128(c0c1, _mm_mul_epu32(a3a3, b1b2));
115   c2c3 = _mm_xor_si128(c2c3, _mm_mul_epu32(a3a3, b3b0));
116   c2c3 = _mm_xor_si128(c2c3, _mm_mul_epu32(a1a1, b1b2));
117 
118   c0c1 = _mm_and_si128(
119       c0c1, _mm_setr_epi32(0x11111111, 0x11111111, 0x22222222, 0x22222222));
120   c2c3 = _mm_and_si128(
121       c2c3, _mm_setr_epi32(0x44444444, 0x44444444, 0x88888888, 0x88888888));
122 
123   c0c1 = _mm_xor_si128(c0c1, c2c3);
124   // c0 ^= c1
125   c0c1 = _mm_xor_si128(c0c1, _mm_srli_si128(c0c1, 8));
126   return c0c1;
127 }
128 
gcm_mul64_nohw(uint64_t * out_lo,uint64_t * out_hi,uint64_t a,uint64_t b)129 static void gcm_mul64_nohw(uint64_t *out_lo, uint64_t *out_hi, uint64_t a,
130                            uint64_t b) {
131   uint32_t a0 = a & 0xffffffff;
132   uint32_t a1 = a >> 32;
133   uint32_t b0 = b & 0xffffffff;
134   uint32_t b1 = b >> 32;
135   // Karatsuba multiplication.
136   __m128i lo = gcm_mul32_nohw(a0, b0);
137   __m128i hi = gcm_mul32_nohw(a1, b1);
138   __m128i mid = gcm_mul32_nohw(a0 ^ a1, b0 ^ b1);
139   mid = _mm_xor_si128(mid, lo);
140   mid = _mm_xor_si128(mid, hi);
141   __m128i ret = _mm_unpacklo_epi64(lo, hi);
142   mid = _mm_slli_si128(mid, 4);
143   mid = _mm_and_si128(mid, _mm_setr_epi32(0, 0xffffffff, 0xffffffff, 0));
144   ret = _mm_xor_si128(ret, mid);
145   memcpy(out_lo, &ret, 8);
146   memcpy(out_hi, ((char*)&ret) + 8, 8);
147 }
148 
149 #else  // !BORINGSSL_HAS_UINT128 && !OPENSSL_SSE2
150 
gcm_mul32_nohw(uint32_t a,uint32_t b)151 static uint64_t gcm_mul32_nohw(uint32_t a, uint32_t b) {
152   // One term every four bits means the largest term is 32/4 = 8, which does not
153   // overflow into the next term.
154   uint32_t a0 = a & 0x11111111;
155   uint32_t a1 = a & 0x22222222;
156   uint32_t a2 = a & 0x44444444;
157   uint32_t a3 = a & 0x88888888;
158 
159   uint32_t b0 = b & 0x11111111;
160   uint32_t b1 = b & 0x22222222;
161   uint32_t b2 = b & 0x44444444;
162   uint32_t b3 = b & 0x88888888;
163 
164   uint64_t c0 = (a0 * (uint64_t)b0) ^ (a1 * (uint64_t)b3) ^
165                 (a2 * (uint64_t)b2) ^ (a3 * (uint64_t)b1);
166   uint64_t c1 = (a0 * (uint64_t)b1) ^ (a1 * (uint64_t)b0) ^
167                 (a2 * (uint64_t)b3) ^ (a3 * (uint64_t)b2);
168   uint64_t c2 = (a0 * (uint64_t)b2) ^ (a1 * (uint64_t)b1) ^
169                 (a2 * (uint64_t)b0) ^ (a3 * (uint64_t)b3);
170   uint64_t c3 = (a0 * (uint64_t)b3) ^ (a1 * (uint64_t)b2) ^
171                 (a2 * (uint64_t)b1) ^ (a3 * (uint64_t)b0);
172 
173   return (c0 & UINT64_C(0x1111111111111111)) |
174          (c1 & UINT64_C(0x2222222222222222)) |
175          (c2 & UINT64_C(0x4444444444444444)) |
176          (c3 & UINT64_C(0x8888888888888888));
177 }
178 
gcm_mul64_nohw(uint64_t * out_lo,uint64_t * out_hi,uint64_t a,uint64_t b)179 static void gcm_mul64_nohw(uint64_t *out_lo, uint64_t *out_hi, uint64_t a,
180                            uint64_t b) {
181   uint32_t a0 = a & 0xffffffff;
182   uint32_t a1 = a >> 32;
183   uint32_t b0 = b & 0xffffffff;
184   uint32_t b1 = b >> 32;
185   // Karatsuba multiplication.
186   uint64_t lo = gcm_mul32_nohw(a0, b0);
187   uint64_t hi = gcm_mul32_nohw(a1, b1);
188   uint64_t mid = gcm_mul32_nohw(a0 ^ a1, b0 ^ b1) ^ lo ^ hi;
189   *out_lo = lo ^ (mid << 32);
190   *out_hi = hi ^ (mid >> 32);
191 }
192 
193 #endif  // BORINGSSL_HAS_UINT128
194 
gcm_init_nohw(u128 Htable[16],const uint64_t Xi[2])195 void gcm_init_nohw(u128 Htable[16], const uint64_t Xi[2]) {
196   // We implement GHASH in terms of POLYVAL, as described in RFC8452. This
197   // avoids a shift by 1 in the multiplication, needed to account for bit
198   // reversal losing a bit after multiplication, that is,
199   // rev128(X) * rev128(Y) = rev255(X*Y).
200   //
201   // Per Appendix A, we run mulX_POLYVAL. Note this is the same transformation
202   // applied by |gcm_init_clmul|, etc. Note |Xi| has already been byteswapped.
203   //
204   // See also slide 16 of
205   // https://crypto.stanford.edu/RealWorldCrypto/slides/gueron.pdf
206   Htable[0].lo = Xi[1];
207   Htable[0].hi = Xi[0];
208 
209   uint64_t carry = Htable[0].hi >> 63;
210   carry = 0u - carry;
211 
212   Htable[0].hi <<= 1;
213   Htable[0].hi |= Htable[0].lo >> 63;
214   Htable[0].lo <<= 1;
215 
216   // The irreducible polynomial is 1 + x^121 + x^126 + x^127 + x^128, so we
217   // conditionally add 0xc200...0001.
218   Htable[0].lo ^= carry & 1;
219   Htable[0].hi ^= carry & UINT64_C(0xc200000000000000);
220 
221   // This implementation does not use the rest of |Htable|.
222 }
223 
gcm_polyval_nohw(uint64_t Xi[2],const u128 * H)224 static void gcm_polyval_nohw(uint64_t Xi[2], const u128 *H) {
225   // Karatsuba multiplication. The product of |Xi| and |H| is stored in |r0|
226   // through |r3|. Note there is no byte or bit reversal because we are
227   // evaluating POLYVAL.
228   uint64_t r0, r1;
229   gcm_mul64_nohw(&r0, &r1, Xi[0], H->lo);
230   uint64_t r2, r3;
231   gcm_mul64_nohw(&r2, &r3, Xi[1], H->hi);
232   uint64_t mid0, mid1;
233   gcm_mul64_nohw(&mid0, &mid1, Xi[0] ^ Xi[1], H->hi ^ H->lo);
234   mid0 ^= r0 ^ r2;
235   mid1 ^= r1 ^ r3;
236   r2 ^= mid1;
237   r1 ^= mid0;
238 
239   // Now we multiply our 256-bit result by x^-128 and reduce. |r2| and
240   // |r3| shifts into position and we must multiply |r0| and |r1| by x^-128. We
241   // have:
242   //
243   //       1 = x^121 + x^126 + x^127 + x^128
244   //  x^-128 = x^-7 + x^-2 + x^-1 + 1
245   //
246   // This is the GHASH reduction step, but with bits flowing in reverse.
247 
248   // The x^-7, x^-2, and x^-1 terms shift bits past x^0, which would require
249   // another reduction steps. Instead, we gather the excess bits, incorporate
250   // them into |r0| and |r1| and reduce once. See slides 17-19
251   // of https://crypto.stanford.edu/RealWorldCrypto/slides/gueron.pdf.
252   r1 ^= (r0 << 63) ^ (r0 << 62) ^ (r0 << 57);
253 
254   // 1
255   r2 ^= r0;
256   r3 ^= r1;
257 
258   // x^-1
259   r2 ^= r0 >> 1;
260   r2 ^= r1 << 63;
261   r3 ^= r1 >> 1;
262 
263   // x^-2
264   r2 ^= r0 >> 2;
265   r2 ^= r1 << 62;
266   r3 ^= r1 >> 2;
267 
268   // x^-7
269   r2 ^= r0 >> 7;
270   r2 ^= r1 << 57;
271   r3 ^= r1 >> 7;
272 
273   Xi[0] = r2;
274   Xi[1] = r3;
275 }
276 
gcm_gmult_nohw(uint64_t Xi[2],const u128 Htable[16])277 void gcm_gmult_nohw(uint64_t Xi[2], const u128 Htable[16]) {
278   uint64_t swapped[2];
279   swapped[0] = CRYPTO_bswap8(Xi[1]);
280   swapped[1] = CRYPTO_bswap8(Xi[0]);
281   gcm_polyval_nohw(swapped, &Htable[0]);
282   Xi[0] = CRYPTO_bswap8(swapped[1]);
283   Xi[1] = CRYPTO_bswap8(swapped[0]);
284 }
285 
gcm_ghash_nohw(uint64_t Xi[2],const u128 Htable[16],const uint8_t * inp,size_t len)286 void gcm_ghash_nohw(uint64_t Xi[2], const u128 Htable[16], const uint8_t *inp,
287                     size_t len) {
288   uint64_t swapped[2];
289   swapped[0] = CRYPTO_bswap8(Xi[1]);
290   swapped[1] = CRYPTO_bswap8(Xi[0]);
291 
292   while (len >= 16) {
293     uint64_t block[2];
294     OPENSSL_memcpy(block, inp, 16);
295     swapped[0] ^= CRYPTO_bswap8(block[1]);
296     swapped[1] ^= CRYPTO_bswap8(block[0]);
297     gcm_polyval_nohw(swapped, &Htable[0]);
298     inp += 16;
299     len -= 16;
300   }
301 
302   Xi[0] = CRYPTO_bswap8(swapped[1]);
303   Xi[1] = CRYPTO_bswap8(swapped[0]);
304 }
305