1 | /* |
2 | * Copyright 2010-2018 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 <string.h> |
11 | #include <openssl/crypto.h> |
12 | #include "internal/cryptlib.h" |
13 | #include "crypto/modes.h" |
14 | |
15 | #if defined(BSWAP4) && defined(STRICT_ALIGNMENT) |
16 | /* redefine, because alignment is ensured */ |
17 | # undef GETU32 |
18 | # define GETU32(p) BSWAP4(*(const u32 *)(p)) |
19 | # undef PUTU32 |
20 | # define PUTU32(p,v) *(u32 *)(p) = BSWAP4(v) |
21 | #endif |
22 | |
23 | #define PACK(s) ((size_t)(s)<<(sizeof(size_t)*8-16)) |
24 | #define REDUCE1BIT(V) do { \ |
25 | if (sizeof(size_t)==8) { \ |
26 | u64 T = U64(0xe100000000000000) & (0-(V.lo&1)); \ |
27 | V.lo = (V.hi<<63)|(V.lo>>1); \ |
28 | V.hi = (V.hi>>1 )^T; \ |
29 | } \ |
30 | else { \ |
31 | u32 T = 0xe1000000U & (0-(u32)(V.lo&1)); \ |
32 | V.lo = (V.hi<<63)|(V.lo>>1); \ |
33 | V.hi = (V.hi>>1 )^((u64)T<<32); \ |
34 | } \ |
35 | } while(0) |
36 | |
37 | /*- |
38 | * Even though permitted values for TABLE_BITS are 8, 4 and 1, it should |
39 | * never be set to 8. 8 is effectively reserved for testing purposes. |
40 | * TABLE_BITS>1 are lookup-table-driven implementations referred to as |
41 | * "Shoup's" in GCM specification. In other words OpenSSL does not cover |
42 | * whole spectrum of possible table driven implementations. Why? In |
43 | * non-"Shoup's" case memory access pattern is segmented in such manner, |
44 | * that it's trivial to see that cache timing information can reveal |
45 | * fair portion of intermediate hash value. Given that ciphertext is |
46 | * always available to attacker, it's possible for him to attempt to |
47 | * deduce secret parameter H and if successful, tamper with messages |
48 | * [which is nothing but trivial in CTR mode]. In "Shoup's" case it's |
49 | * not as trivial, but there is no reason to believe that it's resistant |
50 | * to cache-timing attack. And the thing about "8-bit" implementation is |
51 | * that it consumes 16 (sixteen) times more memory, 4KB per individual |
52 | * key + 1KB shared. Well, on pros side it should be twice as fast as |
53 | * "4-bit" version. And for gcc-generated x86[_64] code, "8-bit" version |
54 | * was observed to run ~75% faster, closer to 100% for commercial |
55 | * compilers... Yet "4-bit" procedure is preferred, because it's |
56 | * believed to provide better security-performance balance and adequate |
57 | * all-round performance. "All-round" refers to things like: |
58 | * |
59 | * - shorter setup time effectively improves overall timing for |
60 | * handling short messages; |
61 | * - larger table allocation can become unbearable because of VM |
62 | * subsystem penalties (for example on Windows large enough free |
63 | * results in VM working set trimming, meaning that consequent |
64 | * malloc would immediately incur working set expansion); |
65 | * - larger table has larger cache footprint, which can affect |
66 | * performance of other code paths (not necessarily even from same |
67 | * thread in Hyper-Threading world); |
68 | * |
69 | * Value of 1 is not appropriate for performance reasons. |
70 | */ |
71 | #if TABLE_BITS==8 |
72 | |
73 | static void gcm_init_8bit(u128 Htable[256], u64 H[2]) |
74 | { |
75 | int i, j; |
76 | u128 V; |
77 | |
78 | Htable[0].hi = 0; |
79 | Htable[0].lo = 0; |
80 | V.hi = H[0]; |
81 | V.lo = H[1]; |
82 | |
83 | for (Htable[128] = V, i = 64; i > 0; i >>= 1) { |
84 | REDUCE1BIT(V); |
85 | Htable[i] = V; |
86 | } |
87 | |
88 | for (i = 2; i < 256; i <<= 1) { |
89 | u128 *Hi = Htable + i, H0 = *Hi; |
90 | for (j = 1; j < i; ++j) { |
91 | Hi[j].hi = H0.hi ^ Htable[j].hi; |
92 | Hi[j].lo = H0.lo ^ Htable[j].lo; |
93 | } |
94 | } |
95 | } |
96 | |
97 | static void gcm_gmult_8bit(u64 Xi[2], const u128 Htable[256]) |
98 | { |
99 | u128 Z = { 0, 0 }; |
100 | const u8 *xi = (const u8 *)Xi + 15; |
101 | size_t rem, n = *xi; |
102 | const union { |
103 | long one; |
104 | char little; |
105 | } is_endian = { 1 }; |
106 | static const size_t rem_8bit[256] = { |
107 | PACK(0x0000), PACK(0x01C2), PACK(0x0384), PACK(0x0246), |
108 | PACK(0x0708), PACK(0x06CA), PACK(0x048C), PACK(0x054E), |
109 | PACK(0x0E10), PACK(0x0FD2), PACK(0x0D94), PACK(0x0C56), |
110 | PACK(0x0918), PACK(0x08DA), PACK(0x0A9C), PACK(0x0B5E), |
111 | PACK(0x1C20), PACK(0x1DE2), PACK(0x1FA4), PACK(0x1E66), |
112 | PACK(0x1B28), PACK(0x1AEA), PACK(0x18AC), PACK(0x196E), |
113 | PACK(0x1230), PACK(0x13F2), PACK(0x11B4), PACK(0x1076), |
114 | PACK(0x1538), PACK(0x14FA), PACK(0x16BC), PACK(0x177E), |
115 | PACK(0x3840), PACK(0x3982), PACK(0x3BC4), PACK(0x3A06), |
116 | PACK(0x3F48), PACK(0x3E8A), PACK(0x3CCC), PACK(0x3D0E), |
117 | PACK(0x3650), PACK(0x3792), PACK(0x35D4), PACK(0x3416), |
118 | PACK(0x3158), PACK(0x309A), PACK(0x32DC), PACK(0x331E), |
119 | PACK(0x2460), PACK(0x25A2), PACK(0x27E4), PACK(0x2626), |
120 | PACK(0x2368), PACK(0x22AA), PACK(0x20EC), PACK(0x212E), |
121 | PACK(0x2A70), PACK(0x2BB2), PACK(0x29F4), PACK(0x2836), |
122 | PACK(0x2D78), PACK(0x2CBA), PACK(0x2EFC), PACK(0x2F3E), |
123 | PACK(0x7080), PACK(0x7142), PACK(0x7304), PACK(0x72C6), |
124 | PACK(0x7788), PACK(0x764A), PACK(0x740C), PACK(0x75CE), |
125 | PACK(0x7E90), PACK(0x7F52), PACK(0x7D14), PACK(0x7CD6), |
126 | PACK(0x7998), PACK(0x785A), PACK(0x7A1C), PACK(0x7BDE), |
127 | PACK(0x6CA0), PACK(0x6D62), PACK(0x6F24), PACK(0x6EE6), |
128 | PACK(0x6BA8), PACK(0x6A6A), PACK(0x682C), PACK(0x69EE), |
129 | PACK(0x62B0), PACK(0x6372), PACK(0x6134), PACK(0x60F6), |
130 | PACK(0x65B8), PACK(0x647A), PACK(0x663C), PACK(0x67FE), |
131 | PACK(0x48C0), PACK(0x4902), PACK(0x4B44), PACK(0x4A86), |
132 | PACK(0x4FC8), PACK(0x4E0A), PACK(0x4C4C), PACK(0x4D8E), |
133 | PACK(0x46D0), PACK(0x4712), PACK(0x4554), PACK(0x4496), |
134 | PACK(0x41D8), PACK(0x401A), PACK(0x425C), PACK(0x439E), |
135 | PACK(0x54E0), PACK(0x5522), PACK(0x5764), PACK(0x56A6), |
136 | PACK(0x53E8), PACK(0x522A), PACK(0x506C), PACK(0x51AE), |
137 | PACK(0x5AF0), PACK(0x5B32), PACK(0x5974), PACK(0x58B6), |
138 | PACK(0x5DF8), PACK(0x5C3A), PACK(0x5E7C), PACK(0x5FBE), |
139 | PACK(0xE100), PACK(0xE0C2), PACK(0xE284), PACK(0xE346), |
140 | PACK(0xE608), PACK(0xE7CA), PACK(0xE58C), PACK(0xE44E), |
141 | PACK(0xEF10), PACK(0xEED2), PACK(0xEC94), PACK(0xED56), |
142 | PACK(0xE818), PACK(0xE9DA), PACK(0xEB9C), PACK(0xEA5E), |
143 | PACK(0xFD20), PACK(0xFCE2), PACK(0xFEA4), PACK(0xFF66), |
144 | PACK(0xFA28), PACK(0xFBEA), PACK(0xF9AC), PACK(0xF86E), |
145 | PACK(0xF330), PACK(0xF2F2), PACK(0xF0B4), PACK(0xF176), |
146 | PACK(0xF438), PACK(0xF5FA), PACK(0xF7BC), PACK(0xF67E), |
147 | PACK(0xD940), PACK(0xD882), PACK(0xDAC4), PACK(0xDB06), |
148 | PACK(0xDE48), PACK(0xDF8A), PACK(0xDDCC), PACK(0xDC0E), |
149 | PACK(0xD750), PACK(0xD692), PACK(0xD4D4), PACK(0xD516), |
150 | PACK(0xD058), PACK(0xD19A), PACK(0xD3DC), PACK(0xD21E), |
151 | PACK(0xC560), PACK(0xC4A2), PACK(0xC6E4), PACK(0xC726), |
152 | PACK(0xC268), PACK(0xC3AA), PACK(0xC1EC), PACK(0xC02E), |
153 | PACK(0xCB70), PACK(0xCAB2), PACK(0xC8F4), PACK(0xC936), |
154 | PACK(0xCC78), PACK(0xCDBA), PACK(0xCFFC), PACK(0xCE3E), |
155 | PACK(0x9180), PACK(0x9042), PACK(0x9204), PACK(0x93C6), |
156 | PACK(0x9688), PACK(0x974A), PACK(0x950C), PACK(0x94CE), |
157 | PACK(0x9F90), PACK(0x9E52), PACK(0x9C14), PACK(0x9DD6), |
158 | PACK(0x9898), PACK(0x995A), PACK(0x9B1C), PACK(0x9ADE), |
159 | PACK(0x8DA0), PACK(0x8C62), PACK(0x8E24), PACK(0x8FE6), |
160 | PACK(0x8AA8), PACK(0x8B6A), PACK(0x892C), PACK(0x88EE), |
161 | PACK(0x83B0), PACK(0x8272), PACK(0x8034), PACK(0x81F6), |
162 | PACK(0x84B8), PACK(0x857A), PACK(0x873C), PACK(0x86FE), |
163 | PACK(0xA9C0), PACK(0xA802), PACK(0xAA44), PACK(0xAB86), |
164 | PACK(0xAEC8), PACK(0xAF0A), PACK(0xAD4C), PACK(0xAC8E), |
165 | PACK(0xA7D0), PACK(0xA612), PACK(0xA454), PACK(0xA596), |
166 | PACK(0xA0D8), PACK(0xA11A), PACK(0xA35C), PACK(0xA29E), |
167 | PACK(0xB5E0), PACK(0xB422), PACK(0xB664), PACK(0xB7A6), |
168 | PACK(0xB2E8), PACK(0xB32A), PACK(0xB16C), PACK(0xB0AE), |
169 | PACK(0xBBF0), PACK(0xBA32), PACK(0xB874), PACK(0xB9B6), |
170 | PACK(0xBCF8), PACK(0xBD3A), PACK(0xBF7C), PACK(0xBEBE) |
171 | }; |
172 | |
173 | while (1) { |
174 | Z.hi ^= Htable[n].hi; |
175 | Z.lo ^= Htable[n].lo; |
176 | |
177 | if ((u8 *)Xi == xi) |
178 | break; |
179 | |
180 | n = *(--xi); |
181 | |
182 | rem = (size_t)Z.lo & 0xff; |
183 | Z.lo = (Z.hi << 56) | (Z.lo >> 8); |
184 | Z.hi = (Z.hi >> 8); |
185 | if (sizeof(size_t) == 8) |
186 | Z.hi ^= rem_8bit[rem]; |
187 | else |
188 | Z.hi ^= (u64)rem_8bit[rem] << 32; |
189 | } |
190 | |
191 | if (is_endian.little) { |
192 | # ifdef BSWAP8 |
193 | Xi[0] = BSWAP8(Z.hi); |
194 | Xi[1] = BSWAP8(Z.lo); |
195 | # else |
196 | u8 *p = (u8 *)Xi; |
197 | u32 v; |
198 | v = (u32)(Z.hi >> 32); |
199 | PUTU32(p, v); |
200 | v = (u32)(Z.hi); |
201 | PUTU32(p + 4, v); |
202 | v = (u32)(Z.lo >> 32); |
203 | PUTU32(p + 8, v); |
204 | v = (u32)(Z.lo); |
205 | PUTU32(p + 12, v); |
206 | # endif |
207 | } else { |
208 | Xi[0] = Z.hi; |
209 | Xi[1] = Z.lo; |
210 | } |
211 | } |
212 | |
213 | # define GCM_MUL(ctx) gcm_gmult_8bit(ctx->Xi.u,ctx->Htable) |
214 | |
215 | #elif TABLE_BITS==4 |
216 | |
217 | static void gcm_init_4bit(u128 Htable[16], u64 H[2]) |
218 | { |
219 | u128 V; |
220 | # if defined(OPENSSL_SMALL_FOOTPRINT) |
221 | int i; |
222 | # endif |
223 | |
224 | Htable[0].hi = 0; |
225 | Htable[0].lo = 0; |
226 | V.hi = H[0]; |
227 | V.lo = H[1]; |
228 | |
229 | # if defined(OPENSSL_SMALL_FOOTPRINT) |
230 | for (Htable[8] = V, i = 4; i > 0; i >>= 1) { |
231 | REDUCE1BIT(V); |
232 | Htable[i] = V; |
233 | } |
234 | |
235 | for (i = 2; i < 16; i <<= 1) { |
236 | u128 *Hi = Htable + i; |
237 | int j; |
238 | for (V = *Hi, j = 1; j < i; ++j) { |
239 | Hi[j].hi = V.hi ^ Htable[j].hi; |
240 | Hi[j].lo = V.lo ^ Htable[j].lo; |
241 | } |
242 | } |
243 | # else |
244 | Htable[8] = V; |
245 | REDUCE1BIT(V); |
246 | Htable[4] = V; |
247 | REDUCE1BIT(V); |
248 | Htable[2] = V; |
249 | REDUCE1BIT(V); |
250 | Htable[1] = V; |
251 | Htable[3].hi = V.hi ^ Htable[2].hi, Htable[3].lo = V.lo ^ Htable[2].lo; |
252 | V = Htable[4]; |
253 | Htable[5].hi = V.hi ^ Htable[1].hi, Htable[5].lo = V.lo ^ Htable[1].lo; |
254 | Htable[6].hi = V.hi ^ Htable[2].hi, Htable[6].lo = V.lo ^ Htable[2].lo; |
255 | Htable[7].hi = V.hi ^ Htable[3].hi, Htable[7].lo = V.lo ^ Htable[3].lo; |
256 | V = Htable[8]; |
257 | Htable[9].hi = V.hi ^ Htable[1].hi, Htable[9].lo = V.lo ^ Htable[1].lo; |
258 | Htable[10].hi = V.hi ^ Htable[2].hi, Htable[10].lo = V.lo ^ Htable[2].lo; |
259 | Htable[11].hi = V.hi ^ Htable[3].hi, Htable[11].lo = V.lo ^ Htable[3].lo; |
260 | Htable[12].hi = V.hi ^ Htable[4].hi, Htable[12].lo = V.lo ^ Htable[4].lo; |
261 | Htable[13].hi = V.hi ^ Htable[5].hi, Htable[13].lo = V.lo ^ Htable[5].lo; |
262 | Htable[14].hi = V.hi ^ Htable[6].hi, Htable[14].lo = V.lo ^ Htable[6].lo; |
263 | Htable[15].hi = V.hi ^ Htable[7].hi, Htable[15].lo = V.lo ^ Htable[7].lo; |
264 | # endif |
265 | # if defined(GHASH_ASM) && (defined(__arm__) || defined(__arm)) |
266 | /* |
267 | * ARM assembler expects specific dword order in Htable. |
268 | */ |
269 | { |
270 | int j; |
271 | const union { |
272 | long one; |
273 | char little; |
274 | } is_endian = { 1 }; |
275 | |
276 | if (is_endian.little) |
277 | for (j = 0; j < 16; ++j) { |
278 | V = Htable[j]; |
279 | Htable[j].hi = V.lo; |
280 | Htable[j].lo = V.hi; |
281 | } else |
282 | for (j = 0; j < 16; ++j) { |
283 | V = Htable[j]; |
284 | Htable[j].hi = V.lo << 32 | V.lo >> 32; |
285 | Htable[j].lo = V.hi << 32 | V.hi >> 32; |
286 | } |
287 | } |
288 | # endif |
289 | } |
290 | |
291 | # ifndef GHASH_ASM |
292 | static const size_t rem_4bit[16] = { |
293 | PACK(0x0000), PACK(0x1C20), PACK(0x3840), PACK(0x2460), |
294 | PACK(0x7080), PACK(0x6CA0), PACK(0x48C0), PACK(0x54E0), |
295 | PACK(0xE100), PACK(0xFD20), PACK(0xD940), PACK(0xC560), |
296 | PACK(0x9180), PACK(0x8DA0), PACK(0xA9C0), PACK(0xB5E0) |
297 | }; |
298 | |
299 | static void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16]) |
300 | { |
301 | u128 Z; |
302 | int cnt = 15; |
303 | size_t rem, nlo, nhi; |
304 | const union { |
305 | long one; |
306 | char little; |
307 | } is_endian = { 1 }; |
308 | |
309 | nlo = ((const u8 *)Xi)[15]; |
310 | nhi = nlo >> 4; |
311 | nlo &= 0xf; |
312 | |
313 | Z.hi = Htable[nlo].hi; |
314 | Z.lo = Htable[nlo].lo; |
315 | |
316 | while (1) { |
317 | rem = (size_t)Z.lo & 0xf; |
318 | Z.lo = (Z.hi << 60) | (Z.lo >> 4); |
319 | Z.hi = (Z.hi >> 4); |
320 | if (sizeof(size_t) == 8) |
321 | Z.hi ^= rem_4bit[rem]; |
322 | else |
323 | Z.hi ^= (u64)rem_4bit[rem] << 32; |
324 | |
325 | Z.hi ^= Htable[nhi].hi; |
326 | Z.lo ^= Htable[nhi].lo; |
327 | |
328 | if (--cnt < 0) |
329 | break; |
330 | |
331 | nlo = ((const u8 *)Xi)[cnt]; |
332 | nhi = nlo >> 4; |
333 | nlo &= 0xf; |
334 | |
335 | rem = (size_t)Z.lo & 0xf; |
336 | Z.lo = (Z.hi << 60) | (Z.lo >> 4); |
337 | Z.hi = (Z.hi >> 4); |
338 | if (sizeof(size_t) == 8) |
339 | Z.hi ^= rem_4bit[rem]; |
340 | else |
341 | Z.hi ^= (u64)rem_4bit[rem] << 32; |
342 | |
343 | Z.hi ^= Htable[nlo].hi; |
344 | Z.lo ^= Htable[nlo].lo; |
345 | } |
346 | |
347 | if (is_endian.little) { |
348 | # ifdef BSWAP8 |
349 | Xi[0] = BSWAP8(Z.hi); |
350 | Xi[1] = BSWAP8(Z.lo); |
351 | # else |
352 | u8 *p = (u8 *)Xi; |
353 | u32 v; |
354 | v = (u32)(Z.hi >> 32); |
355 | PUTU32(p, v); |
356 | v = (u32)(Z.hi); |
357 | PUTU32(p + 4, v); |
358 | v = (u32)(Z.lo >> 32); |
359 | PUTU32(p + 8, v); |
360 | v = (u32)(Z.lo); |
361 | PUTU32(p + 12, v); |
362 | # endif |
363 | } else { |
364 | Xi[0] = Z.hi; |
365 | Xi[1] = Z.lo; |
366 | } |
367 | } |
368 | |
369 | # if !defined(OPENSSL_SMALL_FOOTPRINT) |
370 | /* |
371 | * Streamed gcm_mult_4bit, see CRYPTO_gcm128_[en|de]crypt for |
372 | * details... Compiler-generated code doesn't seem to give any |
373 | * performance improvement, at least not on x86[_64]. It's here |
374 | * mostly as reference and a placeholder for possible future |
375 | * non-trivial optimization[s]... |
376 | */ |
377 | static void gcm_ghash_4bit(u64 Xi[2], const u128 Htable[16], |
378 | const u8 *inp, size_t len) |
379 | { |
380 | u128 Z; |
381 | int cnt; |
382 | size_t rem, nlo, nhi; |
383 | const union { |
384 | long one; |
385 | char little; |
386 | } is_endian = { 1 }; |
387 | |
388 | # if 1 |
389 | do { |
390 | cnt = 15; |
391 | nlo = ((const u8 *)Xi)[15]; |
392 | nlo ^= inp[15]; |
393 | nhi = nlo >> 4; |
394 | nlo &= 0xf; |
395 | |
396 | Z.hi = Htable[nlo].hi; |
397 | Z.lo = Htable[nlo].lo; |
398 | |
399 | while (1) { |
400 | rem = (size_t)Z.lo & 0xf; |
401 | Z.lo = (Z.hi << 60) | (Z.lo >> 4); |
402 | Z.hi = (Z.hi >> 4); |
403 | if (sizeof(size_t) == 8) |
404 | Z.hi ^= rem_4bit[rem]; |
405 | else |
406 | Z.hi ^= (u64)rem_4bit[rem] << 32; |
407 | |
408 | Z.hi ^= Htable[nhi].hi; |
409 | Z.lo ^= Htable[nhi].lo; |
410 | |
411 | if (--cnt < 0) |
412 | break; |
413 | |
414 | nlo = ((const u8 *)Xi)[cnt]; |
415 | nlo ^= inp[cnt]; |
416 | nhi = nlo >> 4; |
417 | nlo &= 0xf; |
418 | |
419 | rem = (size_t)Z.lo & 0xf; |
420 | Z.lo = (Z.hi << 60) | (Z.lo >> 4); |
421 | Z.hi = (Z.hi >> 4); |
422 | if (sizeof(size_t) == 8) |
423 | Z.hi ^= rem_4bit[rem]; |
424 | else |
425 | Z.hi ^= (u64)rem_4bit[rem] << 32; |
426 | |
427 | Z.hi ^= Htable[nlo].hi; |
428 | Z.lo ^= Htable[nlo].lo; |
429 | } |
430 | # else |
431 | /* |
432 | * Extra 256+16 bytes per-key plus 512 bytes shared tables |
433 | * [should] give ~50% improvement... One could have PACK()-ed |
434 | * the rem_8bit even here, but the priority is to minimize |
435 | * cache footprint... |
436 | */ |
437 | u128 Hshr4[16]; /* Htable shifted right by 4 bits */ |
438 | u8 Hshl4[16]; /* Htable shifted left by 4 bits */ |
439 | static const unsigned short rem_8bit[256] = { |
440 | 0x0000, 0x01C2, 0x0384, 0x0246, 0x0708, 0x06CA, 0x048C, 0x054E, |
441 | 0x0E10, 0x0FD2, 0x0D94, 0x0C56, 0x0918, 0x08DA, 0x0A9C, 0x0B5E, |
442 | 0x1C20, 0x1DE2, 0x1FA4, 0x1E66, 0x1B28, 0x1AEA, 0x18AC, 0x196E, |
443 | 0x1230, 0x13F2, 0x11B4, 0x1076, 0x1538, 0x14FA, 0x16BC, 0x177E, |
444 | 0x3840, 0x3982, 0x3BC4, 0x3A06, 0x3F48, 0x3E8A, 0x3CCC, 0x3D0E, |
445 | 0x3650, 0x3792, 0x35D4, 0x3416, 0x3158, 0x309A, 0x32DC, 0x331E, |
446 | 0x2460, 0x25A2, 0x27E4, 0x2626, 0x2368, 0x22AA, 0x20EC, 0x212E, |
447 | 0x2A70, 0x2BB2, 0x29F4, 0x2836, 0x2D78, 0x2CBA, 0x2EFC, 0x2F3E, |
448 | 0x7080, 0x7142, 0x7304, 0x72C6, 0x7788, 0x764A, 0x740C, 0x75CE, |
449 | 0x7E90, 0x7F52, 0x7D14, 0x7CD6, 0x7998, 0x785A, 0x7A1C, 0x7BDE, |
450 | 0x6CA0, 0x6D62, 0x6F24, 0x6EE6, 0x6BA8, 0x6A6A, 0x682C, 0x69EE, |
451 | 0x62B0, 0x6372, 0x6134, 0x60F6, 0x65B8, 0x647A, 0x663C, 0x67FE, |
452 | 0x48C0, 0x4902, 0x4B44, 0x4A86, 0x4FC8, 0x4E0A, 0x4C4C, 0x4D8E, |
453 | 0x46D0, 0x4712, 0x4554, 0x4496, 0x41D8, 0x401A, 0x425C, 0x439E, |
454 | 0x54E0, 0x5522, 0x5764, 0x56A6, 0x53E8, 0x522A, 0x506C, 0x51AE, |
455 | 0x5AF0, 0x5B32, 0x5974, 0x58B6, 0x5DF8, 0x5C3A, 0x5E7C, 0x5FBE, |
456 | 0xE100, 0xE0C2, 0xE284, 0xE346, 0xE608, 0xE7CA, 0xE58C, 0xE44E, |
457 | 0xEF10, 0xEED2, 0xEC94, 0xED56, 0xE818, 0xE9DA, 0xEB9C, 0xEA5E, |
458 | 0xFD20, 0xFCE2, 0xFEA4, 0xFF66, 0xFA28, 0xFBEA, 0xF9AC, 0xF86E, |
459 | 0xF330, 0xF2F2, 0xF0B4, 0xF176, 0xF438, 0xF5FA, 0xF7BC, 0xF67E, |
460 | 0xD940, 0xD882, 0xDAC4, 0xDB06, 0xDE48, 0xDF8A, 0xDDCC, 0xDC0E, |
461 | 0xD750, 0xD692, 0xD4D4, 0xD516, 0xD058, 0xD19A, 0xD3DC, 0xD21E, |
462 | 0xC560, 0xC4A2, 0xC6E4, 0xC726, 0xC268, 0xC3AA, 0xC1EC, 0xC02E, |
463 | 0xCB70, 0xCAB2, 0xC8F4, 0xC936, 0xCC78, 0xCDBA, 0xCFFC, 0xCE3E, |
464 | 0x9180, 0x9042, 0x9204, 0x93C6, 0x9688, 0x974A, 0x950C, 0x94CE, |
465 | 0x9F90, 0x9E52, 0x9C14, 0x9DD6, 0x9898, 0x995A, 0x9B1C, 0x9ADE, |
466 | 0x8DA0, 0x8C62, 0x8E24, 0x8FE6, 0x8AA8, 0x8B6A, 0x892C, 0x88EE, |
467 | 0x83B0, 0x8272, 0x8034, 0x81F6, 0x84B8, 0x857A, 0x873C, 0x86FE, |
468 | 0xA9C0, 0xA802, 0xAA44, 0xAB86, 0xAEC8, 0xAF0A, 0xAD4C, 0xAC8E, |
469 | 0xA7D0, 0xA612, 0xA454, 0xA596, 0xA0D8, 0xA11A, 0xA35C, 0xA29E, |
470 | 0xB5E0, 0xB422, 0xB664, 0xB7A6, 0xB2E8, 0xB32A, 0xB16C, 0xB0AE, |
471 | 0xBBF0, 0xBA32, 0xB874, 0xB9B6, 0xBCF8, 0xBD3A, 0xBF7C, 0xBEBE |
472 | }; |
473 | /* |
474 | * This pre-processing phase slows down procedure by approximately |
475 | * same time as it makes each loop spin faster. In other words |
476 | * single block performance is approximately same as straightforward |
477 | * "4-bit" implementation, and then it goes only faster... |
478 | */ |
479 | for (cnt = 0; cnt < 16; ++cnt) { |
480 | Z.hi = Htable[cnt].hi; |
481 | Z.lo = Htable[cnt].lo; |
482 | Hshr4[cnt].lo = (Z.hi << 60) | (Z.lo >> 4); |
483 | Hshr4[cnt].hi = (Z.hi >> 4); |
484 | Hshl4[cnt] = (u8)(Z.lo << 4); |
485 | } |
486 | |
487 | do { |
488 | for (Z.lo = 0, Z.hi = 0, cnt = 15; cnt; --cnt) { |
489 | nlo = ((const u8 *)Xi)[cnt]; |
490 | nlo ^= inp[cnt]; |
491 | nhi = nlo >> 4; |
492 | nlo &= 0xf; |
493 | |
494 | Z.hi ^= Htable[nlo].hi; |
495 | Z.lo ^= Htable[nlo].lo; |
496 | |
497 | rem = (size_t)Z.lo & 0xff; |
498 | |
499 | Z.lo = (Z.hi << 56) | (Z.lo >> 8); |
500 | Z.hi = (Z.hi >> 8); |
501 | |
502 | Z.hi ^= Hshr4[nhi].hi; |
503 | Z.lo ^= Hshr4[nhi].lo; |
504 | Z.hi ^= (u64)rem_8bit[rem ^ Hshl4[nhi]] << 48; |
505 | } |
506 | |
507 | nlo = ((const u8 *)Xi)[0]; |
508 | nlo ^= inp[0]; |
509 | nhi = nlo >> 4; |
510 | nlo &= 0xf; |
511 | |
512 | Z.hi ^= Htable[nlo].hi; |
513 | Z.lo ^= Htable[nlo].lo; |
514 | |
515 | rem = (size_t)Z.lo & 0xf; |
516 | |
517 | Z.lo = (Z.hi << 60) | (Z.lo >> 4); |
518 | Z.hi = (Z.hi >> 4); |
519 | |
520 | Z.hi ^= Htable[nhi].hi; |
521 | Z.lo ^= Htable[nhi].lo; |
522 | Z.hi ^= ((u64)rem_8bit[rem << 4]) << 48; |
523 | # endif |
524 | |
525 | if (is_endian.little) { |
526 | # ifdef BSWAP8 |
527 | Xi[0] = BSWAP8(Z.hi); |
528 | Xi[1] = BSWAP8(Z.lo); |
529 | # else |
530 | u8 *p = (u8 *)Xi; |
531 | u32 v; |
532 | v = (u32)(Z.hi >> 32); |
533 | PUTU32(p, v); |
534 | v = (u32)(Z.hi); |
535 | PUTU32(p + 4, v); |
536 | v = (u32)(Z.lo >> 32); |
537 | PUTU32(p + 8, v); |
538 | v = (u32)(Z.lo); |
539 | PUTU32(p + 12, v); |
540 | # endif |
541 | } else { |
542 | Xi[0] = Z.hi; |
543 | Xi[1] = Z.lo; |
544 | } |
545 | } while (inp += 16, len -= 16); |
546 | } |
547 | # endif |
548 | # else |
549 | void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16]); |
550 | void gcm_ghash_4bit(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
551 | size_t len); |
552 | # endif |
553 | |
554 | # define GCM_MUL(ctx) gcm_gmult_4bit(ctx->Xi.u,ctx->Htable) |
555 | # if defined(GHASH_ASM) || !defined(OPENSSL_SMALL_FOOTPRINT) |
556 | # define GHASH(ctx,in,len) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len) |
557 | /* |
558 | * GHASH_CHUNK is "stride parameter" missioned to mitigate cache trashing |
559 | * effect. In other words idea is to hash data while it's still in L1 cache |
560 | * after encryption pass... |
561 | */ |
562 | # define GHASH_CHUNK (3*1024) |
563 | # endif |
564 | |
565 | #else /* TABLE_BITS */ |
566 | |
567 | static void gcm_gmult_1bit(u64 Xi[2], const u64 H[2]) |
568 | { |
569 | u128 V, Z = { 0, 0 }; |
570 | long X; |
571 | int i, j; |
572 | const long *xi = (const long *)Xi; |
573 | const union { |
574 | long one; |
575 | char little; |
576 | } is_endian = { 1 }; |
577 | |
578 | V.hi = H[0]; /* H is in host byte order, no byte swapping */ |
579 | V.lo = H[1]; |
580 | |
581 | for (j = 0; j < 16 / sizeof(long); ++j) { |
582 | if (is_endian.little) { |
583 | if (sizeof(long) == 8) { |
584 | # ifdef BSWAP8 |
585 | X = (long)(BSWAP8(xi[j])); |
586 | # else |
587 | const u8 *p = (const u8 *)(xi + j); |
588 | X = (long)((u64)GETU32(p) << 32 | GETU32(p + 4)); |
589 | # endif |
590 | } else { |
591 | const u8 *p = (const u8 *)(xi + j); |
592 | X = (long)GETU32(p); |
593 | } |
594 | } else |
595 | X = xi[j]; |
596 | |
597 | for (i = 0; i < 8 * sizeof(long); ++i, X <<= 1) { |
598 | u64 M = (u64)(X >> (8 * sizeof(long) - 1)); |
599 | Z.hi ^= V.hi & M; |
600 | Z.lo ^= V.lo & M; |
601 | |
602 | REDUCE1BIT(V); |
603 | } |
604 | } |
605 | |
606 | if (is_endian.little) { |
607 | # ifdef BSWAP8 |
608 | Xi[0] = BSWAP8(Z.hi); |
609 | Xi[1] = BSWAP8(Z.lo); |
610 | # else |
611 | u8 *p = (u8 *)Xi; |
612 | u32 v; |
613 | v = (u32)(Z.hi >> 32); |
614 | PUTU32(p, v); |
615 | v = (u32)(Z.hi); |
616 | PUTU32(p + 4, v); |
617 | v = (u32)(Z.lo >> 32); |
618 | PUTU32(p + 8, v); |
619 | v = (u32)(Z.lo); |
620 | PUTU32(p + 12, v); |
621 | # endif |
622 | } else { |
623 | Xi[0] = Z.hi; |
624 | Xi[1] = Z.lo; |
625 | } |
626 | } |
627 | |
628 | # define GCM_MUL(ctx) gcm_gmult_1bit(ctx->Xi.u,ctx->H.u) |
629 | |
630 | #endif |
631 | |
632 | #if TABLE_BITS==4 && (defined(GHASH_ASM) || defined(OPENSSL_CPUID_OBJ)) |
633 | # if !defined(I386_ONLY) && \ |
634 | (defined(__i386) || defined(__i386__) || \ |
635 | defined(__x86_64) || defined(__x86_64__) || \ |
636 | defined(_M_IX86) || defined(_M_AMD64) || defined(_M_X64)) |
637 | # define GHASH_ASM_X86_OR_64 |
638 | # define GCM_FUNCREF_4BIT |
639 | |
640 | void gcm_init_clmul(u128 Htable[16], const u64 Xi[2]); |
641 | void gcm_gmult_clmul(u64 Xi[2], const u128 Htable[16]); |
642 | void gcm_ghash_clmul(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
643 | size_t len); |
644 | |
645 | # if defined(__i386) || defined(__i386__) || defined(_M_IX86) |
646 | # define gcm_init_avx gcm_init_clmul |
647 | # define gcm_gmult_avx gcm_gmult_clmul |
648 | # define gcm_ghash_avx gcm_ghash_clmul |
649 | # else |
650 | void gcm_init_avx(u128 Htable[16], const u64 Xi[2]); |
651 | void gcm_gmult_avx(u64 Xi[2], const u128 Htable[16]); |
652 | void gcm_ghash_avx(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
653 | size_t len); |
654 | # endif |
655 | |
656 | # if defined(__i386) || defined(__i386__) || defined(_M_IX86) |
657 | # define GHASH_ASM_X86 |
658 | void gcm_gmult_4bit_mmx(u64 Xi[2], const u128 Htable[16]); |
659 | void gcm_ghash_4bit_mmx(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
660 | size_t len); |
661 | |
662 | void gcm_gmult_4bit_x86(u64 Xi[2], const u128 Htable[16]); |
663 | void gcm_ghash_4bit_x86(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
664 | size_t len); |
665 | # endif |
666 | # elif defined(__arm__) || defined(__arm) || defined(__aarch64__) |
667 | # include "arm_arch.h" |
668 | # if __ARM_MAX_ARCH__>=7 |
669 | # define GHASH_ASM_ARM |
670 | # define GCM_FUNCREF_4BIT |
671 | # define PMULL_CAPABLE (OPENSSL_armcap_P & ARMV8_PMULL) |
672 | # if defined(__arm__) || defined(__arm) |
673 | # define NEON_CAPABLE (OPENSSL_armcap_P & ARMV7_NEON) |
674 | # endif |
675 | void gcm_init_neon(u128 Htable[16], const u64 Xi[2]); |
676 | void gcm_gmult_neon(u64 Xi[2], const u128 Htable[16]); |
677 | void gcm_ghash_neon(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
678 | size_t len); |
679 | void gcm_init_v8(u128 Htable[16], const u64 Xi[2]); |
680 | void gcm_gmult_v8(u64 Xi[2], const u128 Htable[16]); |
681 | void gcm_ghash_v8(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
682 | size_t len); |
683 | # endif |
684 | # elif defined(__sparc__) || defined(__sparc) |
685 | # include "sparc_arch.h" |
686 | # define GHASH_ASM_SPARC |
687 | # define GCM_FUNCREF_4BIT |
688 | extern unsigned int OPENSSL_sparcv9cap_P[]; |
689 | void gcm_init_vis3(u128 Htable[16], const u64 Xi[2]); |
690 | void gcm_gmult_vis3(u64 Xi[2], const u128 Htable[16]); |
691 | void gcm_ghash_vis3(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
692 | size_t len); |
693 | # elif defined(OPENSSL_CPUID_OBJ) && (defined(__powerpc__) || defined(__ppc__) || defined(_ARCH_PPC)) |
694 | # include "ppc_arch.h" |
695 | # define GHASH_ASM_PPC |
696 | # define GCM_FUNCREF_4BIT |
697 | void gcm_init_p8(u128 Htable[16], const u64 Xi[2]); |
698 | void gcm_gmult_p8(u64 Xi[2], const u128 Htable[16]); |
699 | void gcm_ghash_p8(u64 Xi[2], const u128 Htable[16], const u8 *inp, |
700 | size_t len); |
701 | # endif |
702 | #endif |
703 | |
704 | #ifdef GCM_FUNCREF_4BIT |
705 | # undef GCM_MUL |
706 | # define GCM_MUL(ctx) (*gcm_gmult_p)(ctx->Xi.u,ctx->Htable) |
707 | # ifdef GHASH |
708 | # undef GHASH |
709 | # define GHASH(ctx,in,len) (*gcm_ghash_p)(ctx->Xi.u,ctx->Htable,in,len) |
710 | # endif |
711 | #endif |
712 | |
713 | void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx, void *key, block128_f block) |
714 | { |
715 | const union { |
716 | long one; |
717 | char little; |
718 | } is_endian = { 1 }; |
719 | |
720 | memset(ctx, 0, sizeof(*ctx)); |
721 | ctx->block = block; |
722 | ctx->key = key; |
723 | |
724 | (*block) (ctx->H.c, ctx->H.c, key); |
725 | |
726 | if (is_endian.little) { |
727 | /* H is stored in host byte order */ |
728 | #ifdef BSWAP8 |
729 | ctx->H.u[0] = BSWAP8(ctx->H.u[0]); |
730 | ctx->H.u[1] = BSWAP8(ctx->H.u[1]); |
731 | #else |
732 | u8 *p = ctx->H.c; |
733 | u64 hi, lo; |
734 | hi = (u64)GETU32(p) << 32 | GETU32(p + 4); |
735 | lo = (u64)GETU32(p + 8) << 32 | GETU32(p + 12); |
736 | ctx->H.u[0] = hi; |
737 | ctx->H.u[1] = lo; |
738 | #endif |
739 | } |
740 | #if TABLE_BITS==8 |
741 | gcm_init_8bit(ctx->Htable, ctx->H.u); |
742 | #elif TABLE_BITS==4 |
743 | # if defined(GHASH) |
744 | # define CTX__GHASH(f) (ctx->ghash = (f)) |
745 | # else |
746 | # define CTX__GHASH(f) (ctx->ghash = NULL) |
747 | # endif |
748 | # if defined(GHASH_ASM_X86_OR_64) |
749 | # if !defined(GHASH_ASM_X86) || defined(OPENSSL_IA32_SSE2) |
750 | if (OPENSSL_ia32cap_P[1] & (1 << 1)) { /* check PCLMULQDQ bit */ |
751 | if (((OPENSSL_ia32cap_P[1] >> 22) & 0x41) == 0x41) { /* AVX+MOVBE */ |
752 | gcm_init_avx(ctx->Htable, ctx->H.u); |
753 | ctx->gmult = gcm_gmult_avx; |
754 | CTX__GHASH(gcm_ghash_avx); |
755 | } else { |
756 | gcm_init_clmul(ctx->Htable, ctx->H.u); |
757 | ctx->gmult = gcm_gmult_clmul; |
758 | CTX__GHASH(gcm_ghash_clmul); |
759 | } |
760 | return; |
761 | } |
762 | # endif |
763 | gcm_init_4bit(ctx->Htable, ctx->H.u); |
764 | # if defined(GHASH_ASM_X86) /* x86 only */ |
765 | # if defined(OPENSSL_IA32_SSE2) |
766 | if (OPENSSL_ia32cap_P[0] & (1 << 25)) { /* check SSE bit */ |
767 | # else |
768 | if (OPENSSL_ia32cap_P[0] & (1 << 23)) { /* check MMX bit */ |
769 | # endif |
770 | ctx->gmult = gcm_gmult_4bit_mmx; |
771 | CTX__GHASH(gcm_ghash_4bit_mmx); |
772 | } else { |
773 | ctx->gmult = gcm_gmult_4bit_x86; |
774 | CTX__GHASH(gcm_ghash_4bit_x86); |
775 | } |
776 | # else |
777 | ctx->gmult = gcm_gmult_4bit; |
778 | CTX__GHASH(gcm_ghash_4bit); |
779 | # endif |
780 | # elif defined(GHASH_ASM_ARM) |
781 | # ifdef PMULL_CAPABLE |
782 | if (PMULL_CAPABLE) { |
783 | gcm_init_v8(ctx->Htable, ctx->H.u); |
784 | ctx->gmult = gcm_gmult_v8; |
785 | CTX__GHASH(gcm_ghash_v8); |
786 | } else |
787 | # endif |
788 | # ifdef NEON_CAPABLE |
789 | if (NEON_CAPABLE) { |
790 | gcm_init_neon(ctx->Htable, ctx->H.u); |
791 | ctx->gmult = gcm_gmult_neon; |
792 | CTX__GHASH(gcm_ghash_neon); |
793 | } else |
794 | # endif |
795 | { |
796 | gcm_init_4bit(ctx->Htable, ctx->H.u); |
797 | ctx->gmult = gcm_gmult_4bit; |
798 | CTX__GHASH(gcm_ghash_4bit); |
799 | } |
800 | # elif defined(GHASH_ASM_SPARC) |
801 | if (OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3) { |
802 | gcm_init_vis3(ctx->Htable, ctx->H.u); |
803 | ctx->gmult = gcm_gmult_vis3; |
804 | CTX__GHASH(gcm_ghash_vis3); |
805 | } else { |
806 | gcm_init_4bit(ctx->Htable, ctx->H.u); |
807 | ctx->gmult = gcm_gmult_4bit; |
808 | CTX__GHASH(gcm_ghash_4bit); |
809 | } |
810 | # elif defined(GHASH_ASM_PPC) |
811 | if (OPENSSL_ppccap_P & PPC_CRYPTO207) { |
812 | gcm_init_p8(ctx->Htable, ctx->H.u); |
813 | ctx->gmult = gcm_gmult_p8; |
814 | CTX__GHASH(gcm_ghash_p8); |
815 | } else { |
816 | gcm_init_4bit(ctx->Htable, ctx->H.u); |
817 | ctx->gmult = gcm_gmult_4bit; |
818 | CTX__GHASH(gcm_ghash_4bit); |
819 | } |
820 | # else |
821 | gcm_init_4bit(ctx->Htable, ctx->H.u); |
822 | # endif |
823 | # undef CTX__GHASH |
824 | #endif |
825 | } |
826 | |
827 | void CRYPTO_gcm128_setiv(GCM128_CONTEXT *ctx, const unsigned char *iv, |
828 | size_t len) |
829 | { |
830 | const union { |
831 | long one; |
832 | char little; |
833 | } is_endian = { 1 }; |
834 | unsigned int ctr; |
835 | #ifdef GCM_FUNCREF_4BIT |
836 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
837 | #endif |
838 | |
839 | ctx->len.u[0] = 0; /* AAD length */ |
840 | ctx->len.u[1] = 0; /* message length */ |
841 | ctx->ares = 0; |
842 | ctx->mres = 0; |
843 | |
844 | if (len == 12) { |
845 | memcpy(ctx->Yi.c, iv, 12); |
846 | ctx->Yi.c[12] = 0; |
847 | ctx->Yi.c[13] = 0; |
848 | ctx->Yi.c[14] = 0; |
849 | ctx->Yi.c[15] = 1; |
850 | ctr = 1; |
851 | } else { |
852 | size_t i; |
853 | u64 len0 = len; |
854 | |
855 | /* Borrow ctx->Xi to calculate initial Yi */ |
856 | ctx->Xi.u[0] = 0; |
857 | ctx->Xi.u[1] = 0; |
858 | |
859 | while (len >= 16) { |
860 | for (i = 0; i < 16; ++i) |
861 | ctx->Xi.c[i] ^= iv[i]; |
862 | GCM_MUL(ctx); |
863 | iv += 16; |
864 | len -= 16; |
865 | } |
866 | if (len) { |
867 | for (i = 0; i < len; ++i) |
868 | ctx->Xi.c[i] ^= iv[i]; |
869 | GCM_MUL(ctx); |
870 | } |
871 | len0 <<= 3; |
872 | if (is_endian.little) { |
873 | #ifdef BSWAP8 |
874 | ctx->Xi.u[1] ^= BSWAP8(len0); |
875 | #else |
876 | ctx->Xi.c[8] ^= (u8)(len0 >> 56); |
877 | ctx->Xi.c[9] ^= (u8)(len0 >> 48); |
878 | ctx->Xi.c[10] ^= (u8)(len0 >> 40); |
879 | ctx->Xi.c[11] ^= (u8)(len0 >> 32); |
880 | ctx->Xi.c[12] ^= (u8)(len0 >> 24); |
881 | ctx->Xi.c[13] ^= (u8)(len0 >> 16); |
882 | ctx->Xi.c[14] ^= (u8)(len0 >> 8); |
883 | ctx->Xi.c[15] ^= (u8)(len0); |
884 | #endif |
885 | } else { |
886 | ctx->Xi.u[1] ^= len0; |
887 | } |
888 | |
889 | GCM_MUL(ctx); |
890 | |
891 | if (is_endian.little) |
892 | #ifdef BSWAP4 |
893 | ctr = BSWAP4(ctx->Xi.d[3]); |
894 | #else |
895 | ctr = GETU32(ctx->Xi.c + 12); |
896 | #endif |
897 | else |
898 | ctr = ctx->Xi.d[3]; |
899 | |
900 | /* Copy borrowed Xi to Yi */ |
901 | ctx->Yi.u[0] = ctx->Xi.u[0]; |
902 | ctx->Yi.u[1] = ctx->Xi.u[1]; |
903 | } |
904 | |
905 | ctx->Xi.u[0] = 0; |
906 | ctx->Xi.u[1] = 0; |
907 | |
908 | (*ctx->block) (ctx->Yi.c, ctx->EK0.c, ctx->key); |
909 | ++ctr; |
910 | if (is_endian.little) |
911 | #ifdef BSWAP4 |
912 | ctx->Yi.d[3] = BSWAP4(ctr); |
913 | #else |
914 | PUTU32(ctx->Yi.c + 12, ctr); |
915 | #endif |
916 | else |
917 | ctx->Yi.d[3] = ctr; |
918 | } |
919 | |
920 | int CRYPTO_gcm128_aad(GCM128_CONTEXT *ctx, const unsigned char *aad, |
921 | size_t len) |
922 | { |
923 | size_t i; |
924 | unsigned int n; |
925 | u64 alen = ctx->len.u[0]; |
926 | #ifdef GCM_FUNCREF_4BIT |
927 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
928 | # ifdef GHASH |
929 | void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16], |
930 | const u8 *inp, size_t len) = ctx->ghash; |
931 | # endif |
932 | #endif |
933 | |
934 | if (ctx->len.u[1]) |
935 | return -2; |
936 | |
937 | alen += len; |
938 | if (alen > (U64(1) << 61) || (sizeof(len) == 8 && alen < len)) |
939 | return -1; |
940 | ctx->len.u[0] = alen; |
941 | |
942 | n = ctx->ares; |
943 | if (n) { |
944 | while (n && len) { |
945 | ctx->Xi.c[n] ^= *(aad++); |
946 | --len; |
947 | n = (n + 1) % 16; |
948 | } |
949 | if (n == 0) |
950 | GCM_MUL(ctx); |
951 | else { |
952 | ctx->ares = n; |
953 | return 0; |
954 | } |
955 | } |
956 | #ifdef GHASH |
957 | if ((i = (len & (size_t)-16))) { |
958 | GHASH(ctx, aad, i); |
959 | aad += i; |
960 | len -= i; |
961 | } |
962 | #else |
963 | while (len >= 16) { |
964 | for (i = 0; i < 16; ++i) |
965 | ctx->Xi.c[i] ^= aad[i]; |
966 | GCM_MUL(ctx); |
967 | aad += 16; |
968 | len -= 16; |
969 | } |
970 | #endif |
971 | if (len) { |
972 | n = (unsigned int)len; |
973 | for (i = 0; i < len; ++i) |
974 | ctx->Xi.c[i] ^= aad[i]; |
975 | } |
976 | |
977 | ctx->ares = n; |
978 | return 0; |
979 | } |
980 | |
981 | int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx, |
982 | const unsigned char *in, unsigned char *out, |
983 | size_t len) |
984 | { |
985 | const union { |
986 | long one; |
987 | char little; |
988 | } is_endian = { 1 }; |
989 | unsigned int n, ctr, mres; |
990 | size_t i; |
991 | u64 mlen = ctx->len.u[1]; |
992 | block128_f block = ctx->block; |
993 | void *key = ctx->key; |
994 | #ifdef GCM_FUNCREF_4BIT |
995 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
996 | # if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
997 | void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16], |
998 | const u8 *inp, size_t len) = ctx->ghash; |
999 | # endif |
1000 | #endif |
1001 | |
1002 | mlen += len; |
1003 | if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len)) |
1004 | return -1; |
1005 | ctx->len.u[1] = mlen; |
1006 | |
1007 | mres = ctx->mres; |
1008 | |
1009 | if (ctx->ares) { |
1010 | /* First call to encrypt finalizes GHASH(AAD) */ |
1011 | #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1012 | if (len == 0) { |
1013 | GCM_MUL(ctx); |
1014 | ctx->ares = 0; |
1015 | return 0; |
1016 | } |
1017 | memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi)); |
1018 | ctx->Xi.u[0] = 0; |
1019 | ctx->Xi.u[1] = 0; |
1020 | mres = sizeof(ctx->Xi); |
1021 | #else |
1022 | GCM_MUL(ctx); |
1023 | #endif |
1024 | ctx->ares = 0; |
1025 | } |
1026 | |
1027 | if (is_endian.little) |
1028 | #ifdef BSWAP4 |
1029 | ctr = BSWAP4(ctx->Yi.d[3]); |
1030 | #else |
1031 | ctr = GETU32(ctx->Yi.c + 12); |
1032 | #endif |
1033 | else |
1034 | ctr = ctx->Yi.d[3]; |
1035 | |
1036 | n = mres % 16; |
1037 | #if !defined(OPENSSL_SMALL_FOOTPRINT) |
1038 | if (16 % sizeof(size_t) == 0) { /* always true actually */ |
1039 | do { |
1040 | if (n) { |
1041 | # if defined(GHASH) |
1042 | while (n && len) { |
1043 | ctx->Xn[mres++] = *(out++) = *(in++) ^ ctx->EKi.c[n]; |
1044 | --len; |
1045 | n = (n + 1) % 16; |
1046 | } |
1047 | if (n == 0) { |
1048 | GHASH(ctx, ctx->Xn, mres); |
1049 | mres = 0; |
1050 | } else { |
1051 | ctx->mres = mres; |
1052 | return 0; |
1053 | } |
1054 | # else |
1055 | while (n && len) { |
1056 | ctx->Xi.c[n] ^= *(out++) = *(in++) ^ ctx->EKi.c[n]; |
1057 | --len; |
1058 | n = (n + 1) % 16; |
1059 | } |
1060 | if (n == 0) { |
1061 | GCM_MUL(ctx); |
1062 | mres = 0; |
1063 | } else { |
1064 | ctx->mres = n; |
1065 | return 0; |
1066 | } |
1067 | # endif |
1068 | } |
1069 | # if defined(STRICT_ALIGNMENT) |
1070 | if (((size_t)in | (size_t)out) % sizeof(size_t) != 0) |
1071 | break; |
1072 | # endif |
1073 | # if defined(GHASH) |
1074 | if (len >= 16 && mres) { |
1075 | GHASH(ctx, ctx->Xn, mres); |
1076 | mres = 0; |
1077 | } |
1078 | # if defined(GHASH_CHUNK) |
1079 | while (len >= GHASH_CHUNK) { |
1080 | size_t j = GHASH_CHUNK; |
1081 | |
1082 | while (j) { |
1083 | size_t *out_t = (size_t *)out; |
1084 | const size_t *in_t = (const size_t *)in; |
1085 | |
1086 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1087 | ++ctr; |
1088 | if (is_endian.little) |
1089 | # ifdef BSWAP4 |
1090 | ctx->Yi.d[3] = BSWAP4(ctr); |
1091 | # else |
1092 | PUTU32(ctx->Yi.c + 12, ctr); |
1093 | # endif |
1094 | else |
1095 | ctx->Yi.d[3] = ctr; |
1096 | for (i = 0; i < 16 / sizeof(size_t); ++i) |
1097 | out_t[i] = in_t[i] ^ ctx->EKi.t[i]; |
1098 | out += 16; |
1099 | in += 16; |
1100 | j -= 16; |
1101 | } |
1102 | GHASH(ctx, out - GHASH_CHUNK, GHASH_CHUNK); |
1103 | len -= GHASH_CHUNK; |
1104 | } |
1105 | # endif |
1106 | if ((i = (len & (size_t)-16))) { |
1107 | size_t j = i; |
1108 | |
1109 | while (len >= 16) { |
1110 | size_t *out_t = (size_t *)out; |
1111 | const size_t *in_t = (const size_t *)in; |
1112 | |
1113 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1114 | ++ctr; |
1115 | if (is_endian.little) |
1116 | # ifdef BSWAP4 |
1117 | ctx->Yi.d[3] = BSWAP4(ctr); |
1118 | # else |
1119 | PUTU32(ctx->Yi.c + 12, ctr); |
1120 | # endif |
1121 | else |
1122 | ctx->Yi.d[3] = ctr; |
1123 | for (i = 0; i < 16 / sizeof(size_t); ++i) |
1124 | out_t[i] = in_t[i] ^ ctx->EKi.t[i]; |
1125 | out += 16; |
1126 | in += 16; |
1127 | len -= 16; |
1128 | } |
1129 | GHASH(ctx, out - j, j); |
1130 | } |
1131 | # else |
1132 | while (len >= 16) { |
1133 | size_t *out_t = (size_t *)out; |
1134 | const size_t *in_t = (const size_t *)in; |
1135 | |
1136 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1137 | ++ctr; |
1138 | if (is_endian.little) |
1139 | # ifdef BSWAP4 |
1140 | ctx->Yi.d[3] = BSWAP4(ctr); |
1141 | # else |
1142 | PUTU32(ctx->Yi.c + 12, ctr); |
1143 | # endif |
1144 | else |
1145 | ctx->Yi.d[3] = ctr; |
1146 | for (i = 0; i < 16 / sizeof(size_t); ++i) |
1147 | ctx->Xi.t[i] ^= out_t[i] = in_t[i] ^ ctx->EKi.t[i]; |
1148 | GCM_MUL(ctx); |
1149 | out += 16; |
1150 | in += 16; |
1151 | len -= 16; |
1152 | } |
1153 | # endif |
1154 | if (len) { |
1155 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1156 | ++ctr; |
1157 | if (is_endian.little) |
1158 | # ifdef BSWAP4 |
1159 | ctx->Yi.d[3] = BSWAP4(ctr); |
1160 | # else |
1161 | PUTU32(ctx->Yi.c + 12, ctr); |
1162 | # endif |
1163 | else |
1164 | ctx->Yi.d[3] = ctr; |
1165 | # if defined(GHASH) |
1166 | while (len--) { |
1167 | ctx->Xn[mres++] = out[n] = in[n] ^ ctx->EKi.c[n]; |
1168 | ++n; |
1169 | } |
1170 | # else |
1171 | while (len--) { |
1172 | ctx->Xi.c[n] ^= out[n] = in[n] ^ ctx->EKi.c[n]; |
1173 | ++n; |
1174 | } |
1175 | mres = n; |
1176 | # endif |
1177 | } |
1178 | |
1179 | ctx->mres = mres; |
1180 | return 0; |
1181 | } while (0); |
1182 | } |
1183 | #endif |
1184 | for (i = 0; i < len; ++i) { |
1185 | if (n == 0) { |
1186 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1187 | ++ctr; |
1188 | if (is_endian.little) |
1189 | #ifdef BSWAP4 |
1190 | ctx->Yi.d[3] = BSWAP4(ctr); |
1191 | #else |
1192 | PUTU32(ctx->Yi.c + 12, ctr); |
1193 | #endif |
1194 | else |
1195 | ctx->Yi.d[3] = ctr; |
1196 | } |
1197 | #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1198 | ctx->Xn[mres++] = out[i] = in[i] ^ ctx->EKi.c[n]; |
1199 | n = (n + 1) % 16; |
1200 | if (mres == sizeof(ctx->Xn)) { |
1201 | GHASH(ctx,ctx->Xn,sizeof(ctx->Xn)); |
1202 | mres = 0; |
1203 | } |
1204 | #else |
1205 | ctx->Xi.c[n] ^= out[i] = in[i] ^ ctx->EKi.c[n]; |
1206 | mres = n = (n + 1) % 16; |
1207 | if (n == 0) |
1208 | GCM_MUL(ctx); |
1209 | #endif |
1210 | } |
1211 | |
1212 | ctx->mres = mres; |
1213 | return 0; |
1214 | } |
1215 | |
1216 | int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx, |
1217 | const unsigned char *in, unsigned char *out, |
1218 | size_t len) |
1219 | { |
1220 | const union { |
1221 | long one; |
1222 | char little; |
1223 | } is_endian = { 1 }; |
1224 | unsigned int n, ctr, mres; |
1225 | size_t i; |
1226 | u64 mlen = ctx->len.u[1]; |
1227 | block128_f block = ctx->block; |
1228 | void *key = ctx->key; |
1229 | #ifdef GCM_FUNCREF_4BIT |
1230 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
1231 | # if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1232 | void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16], |
1233 | const u8 *inp, size_t len) = ctx->ghash; |
1234 | # endif |
1235 | #endif |
1236 | |
1237 | mlen += len; |
1238 | if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len)) |
1239 | return -1; |
1240 | ctx->len.u[1] = mlen; |
1241 | |
1242 | mres = ctx->mres; |
1243 | |
1244 | if (ctx->ares) { |
1245 | /* First call to decrypt finalizes GHASH(AAD) */ |
1246 | #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1247 | if (len == 0) { |
1248 | GCM_MUL(ctx); |
1249 | ctx->ares = 0; |
1250 | return 0; |
1251 | } |
1252 | memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi)); |
1253 | ctx->Xi.u[0] = 0; |
1254 | ctx->Xi.u[1] = 0; |
1255 | mres = sizeof(ctx->Xi); |
1256 | #else |
1257 | GCM_MUL(ctx); |
1258 | #endif |
1259 | ctx->ares = 0; |
1260 | } |
1261 | |
1262 | if (is_endian.little) |
1263 | #ifdef BSWAP4 |
1264 | ctr = BSWAP4(ctx->Yi.d[3]); |
1265 | #else |
1266 | ctr = GETU32(ctx->Yi.c + 12); |
1267 | #endif |
1268 | else |
1269 | ctr = ctx->Yi.d[3]; |
1270 | |
1271 | n = mres % 16; |
1272 | #if !defined(OPENSSL_SMALL_FOOTPRINT) |
1273 | if (16 % sizeof(size_t) == 0) { /* always true actually */ |
1274 | do { |
1275 | if (n) { |
1276 | # if defined(GHASH) |
1277 | while (n && len) { |
1278 | *(out++) = (ctx->Xn[mres++] = *(in++)) ^ ctx->EKi.c[n]; |
1279 | --len; |
1280 | n = (n + 1) % 16; |
1281 | } |
1282 | if (n == 0) { |
1283 | GHASH(ctx, ctx->Xn, mres); |
1284 | mres = 0; |
1285 | } else { |
1286 | ctx->mres = mres; |
1287 | return 0; |
1288 | } |
1289 | # else |
1290 | while (n && len) { |
1291 | u8 c = *(in++); |
1292 | *(out++) = c ^ ctx->EKi.c[n]; |
1293 | ctx->Xi.c[n] ^= c; |
1294 | --len; |
1295 | n = (n + 1) % 16; |
1296 | } |
1297 | if (n == 0) { |
1298 | GCM_MUL(ctx); |
1299 | mres = 0; |
1300 | } else { |
1301 | ctx->mres = n; |
1302 | return 0; |
1303 | } |
1304 | # endif |
1305 | } |
1306 | # if defined(STRICT_ALIGNMENT) |
1307 | if (((size_t)in | (size_t)out) % sizeof(size_t) != 0) |
1308 | break; |
1309 | # endif |
1310 | # if defined(GHASH) |
1311 | if (len >= 16 && mres) { |
1312 | GHASH(ctx, ctx->Xn, mres); |
1313 | mres = 0; |
1314 | } |
1315 | # if defined(GHASH_CHUNK) |
1316 | while (len >= GHASH_CHUNK) { |
1317 | size_t j = GHASH_CHUNK; |
1318 | |
1319 | GHASH(ctx, in, GHASH_CHUNK); |
1320 | while (j) { |
1321 | size_t *out_t = (size_t *)out; |
1322 | const size_t *in_t = (const size_t *)in; |
1323 | |
1324 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1325 | ++ctr; |
1326 | if (is_endian.little) |
1327 | # ifdef BSWAP4 |
1328 | ctx->Yi.d[3] = BSWAP4(ctr); |
1329 | # else |
1330 | PUTU32(ctx->Yi.c + 12, ctr); |
1331 | # endif |
1332 | else |
1333 | ctx->Yi.d[3] = ctr; |
1334 | for (i = 0; i < 16 / sizeof(size_t); ++i) |
1335 | out_t[i] = in_t[i] ^ ctx->EKi.t[i]; |
1336 | out += 16; |
1337 | in += 16; |
1338 | j -= 16; |
1339 | } |
1340 | len -= GHASH_CHUNK; |
1341 | } |
1342 | # endif |
1343 | if ((i = (len & (size_t)-16))) { |
1344 | GHASH(ctx, in, i); |
1345 | while (len >= 16) { |
1346 | size_t *out_t = (size_t *)out; |
1347 | const size_t *in_t = (const size_t *)in; |
1348 | |
1349 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1350 | ++ctr; |
1351 | if (is_endian.little) |
1352 | # ifdef BSWAP4 |
1353 | ctx->Yi.d[3] = BSWAP4(ctr); |
1354 | # else |
1355 | PUTU32(ctx->Yi.c + 12, ctr); |
1356 | # endif |
1357 | else |
1358 | ctx->Yi.d[3] = ctr; |
1359 | for (i = 0; i < 16 / sizeof(size_t); ++i) |
1360 | out_t[i] = in_t[i] ^ ctx->EKi.t[i]; |
1361 | out += 16; |
1362 | in += 16; |
1363 | len -= 16; |
1364 | } |
1365 | } |
1366 | # else |
1367 | while (len >= 16) { |
1368 | size_t *out_t = (size_t *)out; |
1369 | const size_t *in_t = (const size_t *)in; |
1370 | |
1371 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1372 | ++ctr; |
1373 | if (is_endian.little) |
1374 | # ifdef BSWAP4 |
1375 | ctx->Yi.d[3] = BSWAP4(ctr); |
1376 | # else |
1377 | PUTU32(ctx->Yi.c + 12, ctr); |
1378 | # endif |
1379 | else |
1380 | ctx->Yi.d[3] = ctr; |
1381 | for (i = 0; i < 16 / sizeof(size_t); ++i) { |
1382 | size_t c = in[i]; |
1383 | out[i] = c ^ ctx->EKi.t[i]; |
1384 | ctx->Xi.t[i] ^= c; |
1385 | } |
1386 | GCM_MUL(ctx); |
1387 | out += 16; |
1388 | in += 16; |
1389 | len -= 16; |
1390 | } |
1391 | # endif |
1392 | if (len) { |
1393 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1394 | ++ctr; |
1395 | if (is_endian.little) |
1396 | # ifdef BSWAP4 |
1397 | ctx->Yi.d[3] = BSWAP4(ctr); |
1398 | # else |
1399 | PUTU32(ctx->Yi.c + 12, ctr); |
1400 | # endif |
1401 | else |
1402 | ctx->Yi.d[3] = ctr; |
1403 | # if defined(GHASH) |
1404 | while (len--) { |
1405 | out[n] = (ctx->Xn[mres++] = in[n]) ^ ctx->EKi.c[n]; |
1406 | ++n; |
1407 | } |
1408 | # else |
1409 | while (len--) { |
1410 | u8 c = in[n]; |
1411 | ctx->Xi.c[n] ^= c; |
1412 | out[n] = c ^ ctx->EKi.c[n]; |
1413 | ++n; |
1414 | } |
1415 | mres = n; |
1416 | # endif |
1417 | } |
1418 | |
1419 | ctx->mres = mres; |
1420 | return 0; |
1421 | } while (0); |
1422 | } |
1423 | #endif |
1424 | for (i = 0; i < len; ++i) { |
1425 | u8 c; |
1426 | if (n == 0) { |
1427 | (*block) (ctx->Yi.c, ctx->EKi.c, key); |
1428 | ++ctr; |
1429 | if (is_endian.little) |
1430 | #ifdef BSWAP4 |
1431 | ctx->Yi.d[3] = BSWAP4(ctr); |
1432 | #else |
1433 | PUTU32(ctx->Yi.c + 12, ctr); |
1434 | #endif |
1435 | else |
1436 | ctx->Yi.d[3] = ctr; |
1437 | } |
1438 | #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1439 | out[i] = (ctx->Xn[mres++] = c = in[i]) ^ ctx->EKi.c[n]; |
1440 | n = (n + 1) % 16; |
1441 | if (mres == sizeof(ctx->Xn)) { |
1442 | GHASH(ctx,ctx->Xn,sizeof(ctx->Xn)); |
1443 | mres = 0; |
1444 | } |
1445 | #else |
1446 | c = in[i]; |
1447 | out[i] = c ^ ctx->EKi.c[n]; |
1448 | ctx->Xi.c[n] ^= c; |
1449 | mres = n = (n + 1) % 16; |
1450 | if (n == 0) |
1451 | GCM_MUL(ctx); |
1452 | #endif |
1453 | } |
1454 | |
1455 | ctx->mres = mres; |
1456 | return 0; |
1457 | } |
1458 | |
1459 | int CRYPTO_gcm128_encrypt_ctr32(GCM128_CONTEXT *ctx, |
1460 | const unsigned char *in, unsigned char *out, |
1461 | size_t len, ctr128_f stream) |
1462 | { |
1463 | #if defined(OPENSSL_SMALL_FOOTPRINT) |
1464 | return CRYPTO_gcm128_encrypt(ctx, in, out, len); |
1465 | #else |
1466 | const union { |
1467 | long one; |
1468 | char little; |
1469 | } is_endian = { 1 }; |
1470 | unsigned int n, ctr, mres; |
1471 | size_t i; |
1472 | u64 mlen = ctx->len.u[1]; |
1473 | void *key = ctx->key; |
1474 | # ifdef GCM_FUNCREF_4BIT |
1475 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
1476 | # ifdef GHASH |
1477 | void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16], |
1478 | const u8 *inp, size_t len) = ctx->ghash; |
1479 | # endif |
1480 | # endif |
1481 | |
1482 | mlen += len; |
1483 | if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len)) |
1484 | return -1; |
1485 | ctx->len.u[1] = mlen; |
1486 | |
1487 | mres = ctx->mres; |
1488 | |
1489 | if (ctx->ares) { |
1490 | /* First call to encrypt finalizes GHASH(AAD) */ |
1491 | #if defined(GHASH) |
1492 | if (len == 0) { |
1493 | GCM_MUL(ctx); |
1494 | ctx->ares = 0; |
1495 | return 0; |
1496 | } |
1497 | memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi)); |
1498 | ctx->Xi.u[0] = 0; |
1499 | ctx->Xi.u[1] = 0; |
1500 | mres = sizeof(ctx->Xi); |
1501 | #else |
1502 | GCM_MUL(ctx); |
1503 | #endif |
1504 | ctx->ares = 0; |
1505 | } |
1506 | |
1507 | if (is_endian.little) |
1508 | # ifdef BSWAP4 |
1509 | ctr = BSWAP4(ctx->Yi.d[3]); |
1510 | # else |
1511 | ctr = GETU32(ctx->Yi.c + 12); |
1512 | # endif |
1513 | else |
1514 | ctr = ctx->Yi.d[3]; |
1515 | |
1516 | n = mres % 16; |
1517 | if (n) { |
1518 | # if defined(GHASH) |
1519 | while (n && len) { |
1520 | ctx->Xn[mres++] = *(out++) = *(in++) ^ ctx->EKi.c[n]; |
1521 | --len; |
1522 | n = (n + 1) % 16; |
1523 | } |
1524 | if (n == 0) { |
1525 | GHASH(ctx, ctx->Xn, mres); |
1526 | mres = 0; |
1527 | } else { |
1528 | ctx->mres = mres; |
1529 | return 0; |
1530 | } |
1531 | # else |
1532 | while (n && len) { |
1533 | ctx->Xi.c[n] ^= *(out++) = *(in++) ^ ctx->EKi.c[n]; |
1534 | --len; |
1535 | n = (n + 1) % 16; |
1536 | } |
1537 | if (n == 0) { |
1538 | GCM_MUL(ctx); |
1539 | mres = 0; |
1540 | } else { |
1541 | ctx->mres = n; |
1542 | return 0; |
1543 | } |
1544 | # endif |
1545 | } |
1546 | # if defined(GHASH) |
1547 | if (len >= 16 && mres) { |
1548 | GHASH(ctx, ctx->Xn, mres); |
1549 | mres = 0; |
1550 | } |
1551 | # if defined(GHASH_CHUNK) |
1552 | while (len >= GHASH_CHUNK) { |
1553 | (*stream) (in, out, GHASH_CHUNK / 16, key, ctx->Yi.c); |
1554 | ctr += GHASH_CHUNK / 16; |
1555 | if (is_endian.little) |
1556 | # ifdef BSWAP4 |
1557 | ctx->Yi.d[3] = BSWAP4(ctr); |
1558 | # else |
1559 | PUTU32(ctx->Yi.c + 12, ctr); |
1560 | # endif |
1561 | else |
1562 | ctx->Yi.d[3] = ctr; |
1563 | GHASH(ctx, out, GHASH_CHUNK); |
1564 | out += GHASH_CHUNK; |
1565 | in += GHASH_CHUNK; |
1566 | len -= GHASH_CHUNK; |
1567 | } |
1568 | # endif |
1569 | # endif |
1570 | if ((i = (len & (size_t)-16))) { |
1571 | size_t j = i / 16; |
1572 | |
1573 | (*stream) (in, out, j, key, ctx->Yi.c); |
1574 | ctr += (unsigned int)j; |
1575 | if (is_endian.little) |
1576 | # ifdef BSWAP4 |
1577 | ctx->Yi.d[3] = BSWAP4(ctr); |
1578 | # else |
1579 | PUTU32(ctx->Yi.c + 12, ctr); |
1580 | # endif |
1581 | else |
1582 | ctx->Yi.d[3] = ctr; |
1583 | in += i; |
1584 | len -= i; |
1585 | # if defined(GHASH) |
1586 | GHASH(ctx, out, i); |
1587 | out += i; |
1588 | # else |
1589 | while (j--) { |
1590 | for (i = 0; i < 16; ++i) |
1591 | ctx->Xi.c[i] ^= out[i]; |
1592 | GCM_MUL(ctx); |
1593 | out += 16; |
1594 | } |
1595 | # endif |
1596 | } |
1597 | if (len) { |
1598 | (*ctx->block) (ctx->Yi.c, ctx->EKi.c, key); |
1599 | ++ctr; |
1600 | if (is_endian.little) |
1601 | # ifdef BSWAP4 |
1602 | ctx->Yi.d[3] = BSWAP4(ctr); |
1603 | # else |
1604 | PUTU32(ctx->Yi.c + 12, ctr); |
1605 | # endif |
1606 | else |
1607 | ctx->Yi.d[3] = ctr; |
1608 | while (len--) { |
1609 | # if defined(GHASH) |
1610 | ctx->Xn[mres++] = out[n] = in[n] ^ ctx->EKi.c[n]; |
1611 | # else |
1612 | ctx->Xi.c[mres++] ^= out[n] = in[n] ^ ctx->EKi.c[n]; |
1613 | # endif |
1614 | ++n; |
1615 | } |
1616 | } |
1617 | |
1618 | ctx->mres = mres; |
1619 | return 0; |
1620 | #endif |
1621 | } |
1622 | |
1623 | int CRYPTO_gcm128_decrypt_ctr32(GCM128_CONTEXT *ctx, |
1624 | const unsigned char *in, unsigned char *out, |
1625 | size_t len, ctr128_f stream) |
1626 | { |
1627 | #if defined(OPENSSL_SMALL_FOOTPRINT) |
1628 | return CRYPTO_gcm128_decrypt(ctx, in, out, len); |
1629 | #else |
1630 | const union { |
1631 | long one; |
1632 | char little; |
1633 | } is_endian = { 1 }; |
1634 | unsigned int n, ctr, mres; |
1635 | size_t i; |
1636 | u64 mlen = ctx->len.u[1]; |
1637 | void *key = ctx->key; |
1638 | # ifdef GCM_FUNCREF_4BIT |
1639 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
1640 | # ifdef GHASH |
1641 | void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16], |
1642 | const u8 *inp, size_t len) = ctx->ghash; |
1643 | # endif |
1644 | # endif |
1645 | |
1646 | mlen += len; |
1647 | if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len)) |
1648 | return -1; |
1649 | ctx->len.u[1] = mlen; |
1650 | |
1651 | mres = ctx->mres; |
1652 | |
1653 | if (ctx->ares) { |
1654 | /* First call to decrypt finalizes GHASH(AAD) */ |
1655 | # if defined(GHASH) |
1656 | if (len == 0) { |
1657 | GCM_MUL(ctx); |
1658 | ctx->ares = 0; |
1659 | return 0; |
1660 | } |
1661 | memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi)); |
1662 | ctx->Xi.u[0] = 0; |
1663 | ctx->Xi.u[1] = 0; |
1664 | mres = sizeof(ctx->Xi); |
1665 | # else |
1666 | GCM_MUL(ctx); |
1667 | # endif |
1668 | ctx->ares = 0; |
1669 | } |
1670 | |
1671 | if (is_endian.little) |
1672 | # ifdef BSWAP4 |
1673 | ctr = BSWAP4(ctx->Yi.d[3]); |
1674 | # else |
1675 | ctr = GETU32(ctx->Yi.c + 12); |
1676 | # endif |
1677 | else |
1678 | ctr = ctx->Yi.d[3]; |
1679 | |
1680 | n = mres % 16; |
1681 | if (n) { |
1682 | # if defined(GHASH) |
1683 | while (n && len) { |
1684 | *(out++) = (ctx->Xn[mres++] = *(in++)) ^ ctx->EKi.c[n]; |
1685 | --len; |
1686 | n = (n + 1) % 16; |
1687 | } |
1688 | if (n == 0) { |
1689 | GHASH(ctx, ctx->Xn, mres); |
1690 | mres = 0; |
1691 | } else { |
1692 | ctx->mres = mres; |
1693 | return 0; |
1694 | } |
1695 | # else |
1696 | while (n && len) { |
1697 | u8 c = *(in++); |
1698 | *(out++) = c ^ ctx->EKi.c[n]; |
1699 | ctx->Xi.c[n] ^= c; |
1700 | --len; |
1701 | n = (n + 1) % 16; |
1702 | } |
1703 | if (n == 0) { |
1704 | GCM_MUL(ctx); |
1705 | mres = 0; |
1706 | } else { |
1707 | ctx->mres = n; |
1708 | return 0; |
1709 | } |
1710 | # endif |
1711 | } |
1712 | # if defined(GHASH) |
1713 | if (len >= 16 && mres) { |
1714 | GHASH(ctx, ctx->Xn, mres); |
1715 | mres = 0; |
1716 | } |
1717 | # if defined(GHASH_CHUNK) |
1718 | while (len >= GHASH_CHUNK) { |
1719 | GHASH(ctx, in, GHASH_CHUNK); |
1720 | (*stream) (in, out, GHASH_CHUNK / 16, key, ctx->Yi.c); |
1721 | ctr += GHASH_CHUNK / 16; |
1722 | if (is_endian.little) |
1723 | # ifdef BSWAP4 |
1724 | ctx->Yi.d[3] = BSWAP4(ctr); |
1725 | # else |
1726 | PUTU32(ctx->Yi.c + 12, ctr); |
1727 | # endif |
1728 | else |
1729 | ctx->Yi.d[3] = ctr; |
1730 | out += GHASH_CHUNK; |
1731 | in += GHASH_CHUNK; |
1732 | len -= GHASH_CHUNK; |
1733 | } |
1734 | # endif |
1735 | # endif |
1736 | if ((i = (len & (size_t)-16))) { |
1737 | size_t j = i / 16; |
1738 | |
1739 | # if defined(GHASH) |
1740 | GHASH(ctx, in, i); |
1741 | # else |
1742 | while (j--) { |
1743 | size_t k; |
1744 | for (k = 0; k < 16; ++k) |
1745 | ctx->Xi.c[k] ^= in[k]; |
1746 | GCM_MUL(ctx); |
1747 | in += 16; |
1748 | } |
1749 | j = i / 16; |
1750 | in -= i; |
1751 | # endif |
1752 | (*stream) (in, out, j, key, ctx->Yi.c); |
1753 | ctr += (unsigned int)j; |
1754 | if (is_endian.little) |
1755 | # ifdef BSWAP4 |
1756 | ctx->Yi.d[3] = BSWAP4(ctr); |
1757 | # else |
1758 | PUTU32(ctx->Yi.c + 12, ctr); |
1759 | # endif |
1760 | else |
1761 | ctx->Yi.d[3] = ctr; |
1762 | out += i; |
1763 | in += i; |
1764 | len -= i; |
1765 | } |
1766 | if (len) { |
1767 | (*ctx->block) (ctx->Yi.c, ctx->EKi.c, key); |
1768 | ++ctr; |
1769 | if (is_endian.little) |
1770 | # ifdef BSWAP4 |
1771 | ctx->Yi.d[3] = BSWAP4(ctr); |
1772 | # else |
1773 | PUTU32(ctx->Yi.c + 12, ctr); |
1774 | # endif |
1775 | else |
1776 | ctx->Yi.d[3] = ctr; |
1777 | while (len--) { |
1778 | # if defined(GHASH) |
1779 | out[n] = (ctx->Xn[mres++] = in[n]) ^ ctx->EKi.c[n]; |
1780 | # else |
1781 | u8 c = in[n]; |
1782 | ctx->Xi.c[mres++] ^= c; |
1783 | out[n] = c ^ ctx->EKi.c[n]; |
1784 | # endif |
1785 | ++n; |
1786 | } |
1787 | } |
1788 | |
1789 | ctx->mres = mres; |
1790 | return 0; |
1791 | #endif |
1792 | } |
1793 | |
1794 | int CRYPTO_gcm128_finish(GCM128_CONTEXT *ctx, const unsigned char *tag, |
1795 | size_t len) |
1796 | { |
1797 | const union { |
1798 | long one; |
1799 | char little; |
1800 | } is_endian = { 1 }; |
1801 | u64 alen = ctx->len.u[0] << 3; |
1802 | u64 clen = ctx->len.u[1] << 3; |
1803 | #ifdef GCM_FUNCREF_4BIT |
1804 | void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult; |
1805 | # if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1806 | void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16], |
1807 | const u8 *inp, size_t len) = ctx->ghash; |
1808 | # endif |
1809 | #endif |
1810 | |
1811 | #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1812 | u128 bitlen; |
1813 | unsigned int mres = ctx->mres; |
1814 | |
1815 | if (mres) { |
1816 | unsigned blocks = (mres + 15) & -16; |
1817 | |
1818 | memset(ctx->Xn + mres, 0, blocks - mres); |
1819 | mres = blocks; |
1820 | if (mres == sizeof(ctx->Xn)) { |
1821 | GHASH(ctx, ctx->Xn, mres); |
1822 | mres = 0; |
1823 | } |
1824 | } else if (ctx->ares) { |
1825 | GCM_MUL(ctx); |
1826 | } |
1827 | #else |
1828 | if (ctx->mres || ctx->ares) |
1829 | GCM_MUL(ctx); |
1830 | #endif |
1831 | |
1832 | if (is_endian.little) { |
1833 | #ifdef BSWAP8 |
1834 | alen = BSWAP8(alen); |
1835 | clen = BSWAP8(clen); |
1836 | #else |
1837 | u8 *p = ctx->len.c; |
1838 | |
1839 | ctx->len.u[0] = alen; |
1840 | ctx->len.u[1] = clen; |
1841 | |
1842 | alen = (u64)GETU32(p) << 32 | GETU32(p + 4); |
1843 | clen = (u64)GETU32(p + 8) << 32 | GETU32(p + 12); |
1844 | #endif |
1845 | } |
1846 | |
1847 | #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) |
1848 | bitlen.hi = alen; |
1849 | bitlen.lo = clen; |
1850 | memcpy(ctx->Xn + mres, &bitlen, sizeof(bitlen)); |
1851 | mres += sizeof(bitlen); |
1852 | GHASH(ctx, ctx->Xn, mres); |
1853 | #else |
1854 | ctx->Xi.u[0] ^= alen; |
1855 | ctx->Xi.u[1] ^= clen; |
1856 | GCM_MUL(ctx); |
1857 | #endif |
1858 | |
1859 | ctx->Xi.u[0] ^= ctx->EK0.u[0]; |
1860 | ctx->Xi.u[1] ^= ctx->EK0.u[1]; |
1861 | |
1862 | if (tag && len <= sizeof(ctx->Xi)) |
1863 | return CRYPTO_memcmp(ctx->Xi.c, tag, len); |
1864 | else |
1865 | return -1; |
1866 | } |
1867 | |
1868 | void CRYPTO_gcm128_tag(GCM128_CONTEXT *ctx, unsigned char *tag, size_t len) |
1869 | { |
1870 | CRYPTO_gcm128_finish(ctx, NULL, 0); |
1871 | memcpy(tag, ctx->Xi.c, |
1872 | len <= sizeof(ctx->Xi.c) ? len : sizeof(ctx->Xi.c)); |
1873 | } |
1874 | |
1875 | GCM128_CONTEXT *CRYPTO_gcm128_new(void *key, block128_f block) |
1876 | { |
1877 | GCM128_CONTEXT *ret; |
1878 | |
1879 | if ((ret = OPENSSL_malloc(sizeof(*ret))) != NULL) |
1880 | CRYPTO_gcm128_init(ret, key, block); |
1881 | |
1882 | return ret; |
1883 | } |
1884 | |
1885 | void CRYPTO_gcm128_release(GCM128_CONTEXT *ctx) |
1886 | { |
1887 | OPENSSL_clear_free(ctx, sizeof(*ctx)); |
1888 | } |
1889 | |