1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2017-2019 Linaro Ltd <ard.biesheuvel@linaro.org>
4 */
5
6#include <crypto/aes.h>
7#include <linux/crypto.h>
8#include <linux/export.h>
9#include <linux/module.h>
10#include <linux/unaligned.h>
11
12/*
13 * Emit the sbox as volatile const to prevent the compiler from doing
14 * constant folding on sbox references involving fixed indexes.
15 */
16static volatile const u8 __cacheline_aligned aes_sbox[] = {
17 0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5,
18 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
19 0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0,
20 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
21 0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc,
22 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
23 0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a,
24 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
25 0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0,
26 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
27 0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b,
28 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
29 0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85,
30 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
31 0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5,
32 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
33 0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17,
34 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
35 0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88,
36 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
37 0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c,
38 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
39 0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9,
40 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
41 0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6,
42 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
43 0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e,
44 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
45 0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94,
46 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
47 0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68,
48 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16,
49};
50
51static volatile const u8 __cacheline_aligned aes_inv_sbox[] = {
52 0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38,
53 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
54 0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87,
55 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
56 0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d,
57 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
58 0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2,
59 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25,
60 0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16,
61 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92,
62 0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda,
63 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84,
64 0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a,
65 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06,
66 0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02,
67 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b,
68 0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea,
69 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73,
70 0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85,
71 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e,
72 0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89,
73 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b,
74 0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20,
75 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4,
76 0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31,
77 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f,
78 0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d,
79 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef,
80 0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0,
81 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61,
82 0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26,
83 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d,
84};
85
86extern const u8 crypto_aes_sbox[256] __alias(aes_sbox);
87extern const u8 crypto_aes_inv_sbox[256] __alias(aes_inv_sbox);
88
89EXPORT_SYMBOL(crypto_aes_sbox);
90EXPORT_SYMBOL(crypto_aes_inv_sbox);
91
92static u32 mul_by_x(u32 w)
93{
94 u32 x = w & 0x7f7f7f7f;
95 u32 y = w & 0x80808080;
96
97 /* multiply by polynomial 'x' (0b10) in GF(2^8) */
98 return (x << 1) ^ (y >> 7) * 0x1b;
99}
100
101static u32 mul_by_x2(u32 w)
102{
103 u32 x = w & 0x3f3f3f3f;
104 u32 y = w & 0x80808080;
105 u32 z = w & 0x40404040;
106
107 /* multiply by polynomial 'x^2' (0b100) in GF(2^8) */
108 return (x << 2) ^ (y >> 7) * 0x36 ^ (z >> 6) * 0x1b;
109}
110
111static u32 mix_columns(u32 x)
112{
113 /*
114 * Perform the following matrix multiplication in GF(2^8)
115 *
116 * | 0x2 0x3 0x1 0x1 | | x[0] |
117 * | 0x1 0x2 0x3 0x1 | | x[1] |
118 * | 0x1 0x1 0x2 0x3 | x | x[2] |
119 * | 0x3 0x1 0x1 0x2 | | x[3] |
120 */
121 u32 y = mul_by_x(w: x) ^ ror32(word: x, shift: 16);
122
123 return y ^ ror32(word: x ^ y, shift: 8);
124}
125
126static u32 inv_mix_columns(u32 x)
127{
128 /*
129 * Perform the following matrix multiplication in GF(2^8)
130 *
131 * | 0xe 0xb 0xd 0x9 | | x[0] |
132 * | 0x9 0xe 0xb 0xd | | x[1] |
133 * | 0xd 0x9 0xe 0xb | x | x[2] |
134 * | 0xb 0xd 0x9 0xe | | x[3] |
135 *
136 * which can conveniently be reduced to
137 *
138 * | 0x2 0x3 0x1 0x1 | | 0x5 0x0 0x4 0x0 | | x[0] |
139 * | 0x1 0x2 0x3 0x1 | | 0x0 0x5 0x0 0x4 | | x[1] |
140 * | 0x1 0x1 0x2 0x3 | x | 0x4 0x0 0x5 0x0 | x | x[2] |
141 * | 0x3 0x1 0x1 0x2 | | 0x0 0x4 0x0 0x5 | | x[3] |
142 */
143 u32 y = mul_by_x2(w: x);
144
145 return mix_columns(x: x ^ y ^ ror32(word: y, shift: 16));
146}
147
148static __always_inline u32 subshift(u32 in[], int pos)
149{
150 return (aes_sbox[in[pos] & 0xff]) ^
151 (aes_sbox[(in[(pos + 1) % 4] >> 8) & 0xff] << 8) ^
152 (aes_sbox[(in[(pos + 2) % 4] >> 16) & 0xff] << 16) ^
153 (aes_sbox[(in[(pos + 3) % 4] >> 24) & 0xff] << 24);
154}
155
156static __always_inline u32 inv_subshift(u32 in[], int pos)
157{
158 return (aes_inv_sbox[in[pos] & 0xff]) ^
159 (aes_inv_sbox[(in[(pos + 3) % 4] >> 8) & 0xff] << 8) ^
160 (aes_inv_sbox[(in[(pos + 2) % 4] >> 16) & 0xff] << 16) ^
161 (aes_inv_sbox[(in[(pos + 1) % 4] >> 24) & 0xff] << 24);
162}
163
164static u32 subw(u32 in)
165{
166 return (aes_sbox[in & 0xff]) ^
167 (aes_sbox[(in >> 8) & 0xff] << 8) ^
168 (aes_sbox[(in >> 16) & 0xff] << 16) ^
169 (aes_sbox[(in >> 24) & 0xff] << 24);
170}
171
172/**
173 * aes_expandkey - Expands the AES key as described in FIPS-197
174 * @ctx: The location where the computed key will be stored.
175 * @in_key: The supplied key.
176 * @key_len: The length of the supplied key.
177 *
178 * Returns 0 on success. The function fails only if an invalid key size (or
179 * pointer) is supplied.
180 * The expanded key size is 240 bytes (max of 14 rounds with a unique 16 bytes
181 * key schedule plus a 16 bytes key which is used before the first round).
182 * The decryption key is prepared for the "Equivalent Inverse Cipher" as
183 * described in FIPS-197. The first slot (16 bytes) of each key (enc or dec) is
184 * for the initial combination, the second slot for the first round and so on.
185 */
186int aes_expandkey(struct crypto_aes_ctx *ctx, const u8 *in_key,
187 unsigned int key_len)
188{
189 u32 kwords = key_len / sizeof(u32);
190 u32 rc, i, j;
191 int err;
192
193 err = aes_check_keylen(keylen: key_len);
194 if (err)
195 return err;
196
197 ctx->key_length = key_len;
198
199 for (i = 0; i < kwords; i++)
200 ctx->key_enc[i] = get_unaligned_le32(p: in_key + i * sizeof(u32));
201
202 for (i = 0, rc = 1; i < 10; i++, rc = mul_by_x(w: rc)) {
203 u32 *rki = ctx->key_enc + (i * kwords);
204 u32 *rko = rki + kwords;
205
206 rko[0] = ror32(word: subw(in: rki[kwords - 1]), shift: 8) ^ rc ^ rki[0];
207 rko[1] = rko[0] ^ rki[1];
208 rko[2] = rko[1] ^ rki[2];
209 rko[3] = rko[2] ^ rki[3];
210
211 if (key_len == AES_KEYSIZE_192) {
212 if (i >= 7)
213 break;
214 rko[4] = rko[3] ^ rki[4];
215 rko[5] = rko[4] ^ rki[5];
216 } else if (key_len == AES_KEYSIZE_256) {
217 if (i >= 6)
218 break;
219 rko[4] = subw(in: rko[3]) ^ rki[4];
220 rko[5] = rko[4] ^ rki[5];
221 rko[6] = rko[5] ^ rki[6];
222 rko[7] = rko[6] ^ rki[7];
223 }
224 }
225
226 /*
227 * Generate the decryption keys for the Equivalent Inverse Cipher.
228 * This involves reversing the order of the round keys, and applying
229 * the Inverse Mix Columns transformation to all but the first and
230 * the last one.
231 */
232 ctx->key_dec[0] = ctx->key_enc[key_len + 24];
233 ctx->key_dec[1] = ctx->key_enc[key_len + 25];
234 ctx->key_dec[2] = ctx->key_enc[key_len + 26];
235 ctx->key_dec[3] = ctx->key_enc[key_len + 27];
236
237 for (i = 4, j = key_len + 20; j > 0; i += 4, j -= 4) {
238 ctx->key_dec[i] = inv_mix_columns(x: ctx->key_enc[j]);
239 ctx->key_dec[i + 1] = inv_mix_columns(x: ctx->key_enc[j + 1]);
240 ctx->key_dec[i + 2] = inv_mix_columns(x: ctx->key_enc[j + 2]);
241 ctx->key_dec[i + 3] = inv_mix_columns(x: ctx->key_enc[j + 3]);
242 }
243
244 ctx->key_dec[i] = ctx->key_enc[0];
245 ctx->key_dec[i + 1] = ctx->key_enc[1];
246 ctx->key_dec[i + 2] = ctx->key_enc[2];
247 ctx->key_dec[i + 3] = ctx->key_enc[3];
248
249 return 0;
250}
251EXPORT_SYMBOL(aes_expandkey);
252
253/**
254 * aes_encrypt - Encrypt a single AES block
255 * @ctx: Context struct containing the key schedule
256 * @out: Buffer to store the ciphertext
257 * @in: Buffer containing the plaintext
258 */
259void aes_encrypt(const struct crypto_aes_ctx *ctx, u8 *out, const u8 *in)
260{
261 const u32 *rkp = ctx->key_enc + 4;
262 int rounds = 6 + ctx->key_length / 4;
263 u32 st0[4], st1[4];
264 int round;
265
266 st0[0] = ctx->key_enc[0] ^ get_unaligned_le32(p: in);
267 st0[1] = ctx->key_enc[1] ^ get_unaligned_le32(p: in + 4);
268 st0[2] = ctx->key_enc[2] ^ get_unaligned_le32(p: in + 8);
269 st0[3] = ctx->key_enc[3] ^ get_unaligned_le32(p: in + 12);
270
271 /*
272 * Force the compiler to emit data independent Sbox references,
273 * by xoring the input with Sbox values that are known to add up
274 * to zero. This pulls the entire Sbox into the D-cache before any
275 * data dependent lookups are done.
276 */
277 st0[0] ^= aes_sbox[ 0] ^ aes_sbox[ 64] ^ aes_sbox[134] ^ aes_sbox[195];
278 st0[1] ^= aes_sbox[16] ^ aes_sbox[ 82] ^ aes_sbox[158] ^ aes_sbox[221];
279 st0[2] ^= aes_sbox[32] ^ aes_sbox[ 96] ^ aes_sbox[160] ^ aes_sbox[234];
280 st0[3] ^= aes_sbox[48] ^ aes_sbox[112] ^ aes_sbox[186] ^ aes_sbox[241];
281
282 for (round = 0;; round += 2, rkp += 8) {
283 st1[0] = mix_columns(x: subshift(in: st0, pos: 0)) ^ rkp[0];
284 st1[1] = mix_columns(x: subshift(in: st0, pos: 1)) ^ rkp[1];
285 st1[2] = mix_columns(x: subshift(in: st0, pos: 2)) ^ rkp[2];
286 st1[3] = mix_columns(x: subshift(in: st0, pos: 3)) ^ rkp[3];
287
288 if (round == rounds - 2)
289 break;
290
291 st0[0] = mix_columns(x: subshift(in: st1, pos: 0)) ^ rkp[4];
292 st0[1] = mix_columns(x: subshift(in: st1, pos: 1)) ^ rkp[5];
293 st0[2] = mix_columns(x: subshift(in: st1, pos: 2)) ^ rkp[6];
294 st0[3] = mix_columns(x: subshift(in: st1, pos: 3)) ^ rkp[7];
295 }
296
297 put_unaligned_le32(val: subshift(in: st1, pos: 0) ^ rkp[4], p: out);
298 put_unaligned_le32(val: subshift(in: st1, pos: 1) ^ rkp[5], p: out + 4);
299 put_unaligned_le32(val: subshift(in: st1, pos: 2) ^ rkp[6], p: out + 8);
300 put_unaligned_le32(val: subshift(in: st1, pos: 3) ^ rkp[7], p: out + 12);
301}
302EXPORT_SYMBOL(aes_encrypt);
303
304/**
305 * aes_decrypt - Decrypt a single AES block
306 * @ctx: Context struct containing the key schedule
307 * @out: Buffer to store the plaintext
308 * @in: Buffer containing the ciphertext
309 */
310void aes_decrypt(const struct crypto_aes_ctx *ctx, u8 *out, const u8 *in)
311{
312 const u32 *rkp = ctx->key_dec + 4;
313 int rounds = 6 + ctx->key_length / 4;
314 u32 st0[4], st1[4];
315 int round;
316
317 st0[0] = ctx->key_dec[0] ^ get_unaligned_le32(p: in);
318 st0[1] = ctx->key_dec[1] ^ get_unaligned_le32(p: in + 4);
319 st0[2] = ctx->key_dec[2] ^ get_unaligned_le32(p: in + 8);
320 st0[3] = ctx->key_dec[3] ^ get_unaligned_le32(p: in + 12);
321
322 /*
323 * Force the compiler to emit data independent Sbox references,
324 * by xoring the input with Sbox values that are known to add up
325 * to zero. This pulls the entire Sbox into the D-cache before any
326 * data dependent lookups are done.
327 */
328 st0[0] ^= aes_inv_sbox[ 0] ^ aes_inv_sbox[ 64] ^ aes_inv_sbox[129] ^ aes_inv_sbox[200];
329 st0[1] ^= aes_inv_sbox[16] ^ aes_inv_sbox[ 83] ^ aes_inv_sbox[150] ^ aes_inv_sbox[212];
330 st0[2] ^= aes_inv_sbox[32] ^ aes_inv_sbox[ 96] ^ aes_inv_sbox[160] ^ aes_inv_sbox[236];
331 st0[3] ^= aes_inv_sbox[48] ^ aes_inv_sbox[112] ^ aes_inv_sbox[187] ^ aes_inv_sbox[247];
332
333 for (round = 0;; round += 2, rkp += 8) {
334 st1[0] = inv_mix_columns(x: inv_subshift(in: st0, pos: 0)) ^ rkp[0];
335 st1[1] = inv_mix_columns(x: inv_subshift(in: st0, pos: 1)) ^ rkp[1];
336 st1[2] = inv_mix_columns(x: inv_subshift(in: st0, pos: 2)) ^ rkp[2];
337 st1[3] = inv_mix_columns(x: inv_subshift(in: st0, pos: 3)) ^ rkp[3];
338
339 if (round == rounds - 2)
340 break;
341
342 st0[0] = inv_mix_columns(x: inv_subshift(in: st1, pos: 0)) ^ rkp[4];
343 st0[1] = inv_mix_columns(x: inv_subshift(in: st1, pos: 1)) ^ rkp[5];
344 st0[2] = inv_mix_columns(x: inv_subshift(in: st1, pos: 2)) ^ rkp[6];
345 st0[3] = inv_mix_columns(x: inv_subshift(in: st1, pos: 3)) ^ rkp[7];
346 }
347
348 put_unaligned_le32(val: inv_subshift(in: st1, pos: 0) ^ rkp[4], p: out);
349 put_unaligned_le32(val: inv_subshift(in: st1, pos: 1) ^ rkp[5], p: out + 4);
350 put_unaligned_le32(val: inv_subshift(in: st1, pos: 2) ^ rkp[6], p: out + 8);
351 put_unaligned_le32(val: inv_subshift(in: st1, pos: 3) ^ rkp[7], p: out + 12);
352}
353EXPORT_SYMBOL(aes_decrypt);
354
355MODULE_DESCRIPTION("Generic AES library");
356MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
357MODULE_LICENSE("GPL v2");
358