| 1 | |
| 2 | |
| 3 | * SHA-1 and HMAC-SHA1 library functions
|
| 4 |
|
| 5 |
|
| 6 | #include <crypto/hmac.h>
|
| 7 | #include <crypto/sha1.h>
|
| 8 | #include <linux/bitops.h>
|
| 9 | #include <linux/export.h>
|
| 10 | #include <linux/kernel.h>
|
| 11 | #include <linux/module.h>
|
| 12 | #include <linux/string.h>
|
| 13 | #include <linux/unaligned.h>
|
| 14 | #include <linux/wordpart.h>
|
| 15 | #include "fips-sha.h"
|
| 16 |
|
| 17 | static const struct sha1_block_state sha1_iv = {
|
| 18 | .h = { SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 },
|
| 19 | };
|
| 20 |
|
| 21 | |
| 22 | * If you have 32 registers or more, the compiler can (and should)
|
| 23 | * try to change the array[] accesses into registers. However, on
|
| 24 | * machines with less than ~25 registers, that won't really work,
|
| 25 | * and at least gcc will make an unholy mess of it.
|
| 26 | *
|
| 27 | * So to avoid that mess which just slows things down, we force
|
| 28 | * the stores to memory to actually happen (we might be better off
|
| 29 | * with a 'W(t)=(val);asm("":"+m" (W(t))' there instead, as
|
| 30 | * suggested by Artur Skawina - that will also make gcc unable to
|
| 31 | * try to do the silly "optimize away loads" part because it won't
|
| 32 | * see what the value will be).
|
| 33 | *
|
| 34 | * Ben Herrenschmidt reports that on PPC, the C version comes close
|
| 35 | * to the optimized asm with this (ie on PPC you don't want that
|
| 36 | * 'volatile', since there are lots of registers).
|
| 37 | *
|
| 38 | * On ARM we get the best code generation by forcing a full memory barrier
|
| 39 | * between each SHA_ROUND, otherwise gcc happily get wild with spilling and
|
| 40 | * the stack frame size simply explode and performance goes down the drain.
|
| 41 |
|
| 42 |
|
| 43 | #ifdef CONFIG_X86
|
| 44 | #define setW(x, val) (*(volatile __u32 *)&W(x) = (val))
|
| 45 | #elif defined(CONFIG_ARM)
|
| 46 | #define setW(x, val) do { W(x) = (val); __asm__("":::"memory"); } while (0)
|
| 47 | #else
|
| 48 | #define setW(x, val) (W(x) = (val))
|
| 49 | #endif
|
| 50 |
|
| 51 |
|
| 52 | #define W(x) (workspace[(x)&15])
|
| 53 |
|
| 54 | |
| 55 | * Where do we get the source from? The first 16 iterations get it from
|
| 56 | * the input data, the next mix it from the 512-bit array.
|
| 57 |
|
| 58 | #define SHA_SRC(t) get_unaligned_be32((__u32 *)data + t)
|
| 59 | #define SHA_MIX(t) rol32(W(t+13) ^ W(t+8) ^ W(t+2) ^ W(t), 1)
|
| 60 |
|
| 61 | #define SHA_ROUND(t, input, fn, constant, A, B, C, D, E) do { \
|
| 62 | __u32 TEMP = input(t); setW(t, TEMP); \
|
| 63 | E += TEMP + rol32(A,5) + (fn) + (constant); \
|
| 64 | B = ror32(B, 2); \
|
| 65 | TEMP = E; E = D; D = C; C = B; B = A; A = TEMP; } while (0)
|
| 66 |
|
| 67 | #define T_0_15(t, A, B, C, D, E) SHA_ROUND(t, SHA_SRC, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
|
| 68 | #define T_16_19(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
|
| 69 | #define T_20_39(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0x6ed9eba1, A, B, C, D, E )
|
| 70 | #define T_40_59(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, ((B&C)+(D&(B^C))) , 0x8f1bbcdc, A, B, C, D, E )
|
| 71 | #define T_60_79(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0xca62c1d6, A, B, C, D, E )
|
| 72 |
|
| 73 | #define SHA1_WORKSPACE_WORDS 16
|
| 74 |
|
| 75 | static void sha1_block_generic(struct sha1_block_state *state,
|
| 76 | const u8 data[SHA1_BLOCK_SIZE],
|
| 77 | u32 workspace[SHA1_WORKSPACE_WORDS])
|
| 78 | {
|
| 79 | __u32 A, B, C, D, E;
|
| 80 | unsigned int i = 0;
|
| 81 |
|
| 82 | A = state->h[0];
|
| 83 | B = state->h[1];
|
| 84 | C = state->h[2];
|
| 85 | D = state->h[3];
|
| 86 | E = state->h[4];
|
| 87 |
|
| 88 |
|
| 89 | for (; i < 16; ++i)
|
| 90 | T_0_15(i, A, B, C, D, E);
|
| 91 |
|
| 92 |
|
| 93 | for (; i < 20; ++i)
|
| 94 | T_16_19(i, A, B, C, D, E);
|
| 95 |
|
| 96 |
|
| 97 | for (; i < 40; ++i)
|
| 98 | T_20_39(i, A, B, C, D, E);
|
| 99 |
|
| 100 |
|
| 101 | for (; i < 60; ++i)
|
| 102 | T_40_59(i, A, B, C, D, E);
|
| 103 |
|
| 104 |
|
| 105 | for (; i < 80; ++i)
|
| 106 | T_60_79(i, A, B, C, D, E);
|
| 107 |
|
| 108 | state->h[0] += A;
|
| 109 | state->h[1] += B;
|
| 110 | state->h[2] += C;
|
| 111 | state->h[3] += D;
|
| 112 | state->h[4] += E;
|
| 113 | }
|
| 114 |
|
| 115 | static void __maybe_unused sha1_blocks_generic(struct sha1_block_state *state,
|
| 116 | const u8 *data, size_t nblocks)
|
| 117 | {
|
| 118 | u32 workspace[SHA1_WORKSPACE_WORDS];
|
| 119 |
|
| 120 | do {
|
| 121 | sha1_block_generic(state, data, workspace);
|
| 122 | data += SHA1_BLOCK_SIZE;
|
| 123 | } while (--nblocks);
|
| 124 |
|
| 125 | memzero_explicit(workspace, sizeof(workspace));
|
| 126 | }
|
| 127 |
|
| 128 | #ifdef CONFIG_CRYPTO_LIB_SHA1_ARCH
|
| 129 | #include "sha1.h"
|
| 130 | #else
|
| 131 | #define sha1_blocks sha1_blocks_generic
|
| 132 | #endif
|
| 133 |
|
| 134 | void sha1_init(struct sha1_ctx *ctx)
|
| 135 | {
|
| 136 | ctx->state = sha1_iv;
|
| 137 | ctx->bytecount = 0;
|
| 138 | }
|
| 139 | EXPORT_SYMBOL_GPL(sha1_init);
|
| 140 |
|
| 141 | void sha1_update(struct sha1_ctx *ctx, const u8 *data, size_t len)
|
| 142 | {
|
| 143 | size_t partial = ctx->bytecount % SHA1_BLOCK_SIZE;
|
| 144 |
|
| 145 | ctx->bytecount += len;
|
| 146 |
|
| 147 | if (partial + len >= SHA1_BLOCK_SIZE) {
|
| 148 | size_t nblocks;
|
| 149 |
|
| 150 | if (partial) {
|
| 151 | size_t l = SHA1_BLOCK_SIZE - partial;
|
| 152 |
|
| 153 | memcpy(&ctx->buf[partial], data, l);
|
| 154 | data += l;
|
| 155 | len -= l;
|
| 156 |
|
| 157 | sha1_blocks(&ctx->state, ctx->buf, 1);
|
| 158 | }
|
| 159 |
|
| 160 | nblocks = len / SHA1_BLOCK_SIZE;
|
| 161 | len %= SHA1_BLOCK_SIZE;
|
| 162 |
|
| 163 | if (nblocks) {
|
| 164 | sha1_blocks(&ctx->state, data, nblocks);
|
| 165 | data += nblocks * SHA1_BLOCK_SIZE;
|
| 166 | }
|
| 167 | partial = 0;
|
| 168 | }
|
| 169 | if (len)
|
| 170 | memcpy(&ctx->buf[partial], data, len);
|
| 171 | }
|
| 172 | EXPORT_SYMBOL_GPL(sha1_update);
|
| 173 |
|
| 174 | static void __sha1_final(struct sha1_ctx *ctx, u8 out[SHA1_DIGEST_SIZE])
|
| 175 | {
|
| 176 | u64 bitcount = ctx->bytecount << 3;
|
| 177 | size_t partial = ctx->bytecount % SHA1_BLOCK_SIZE;
|
| 178 |
|
| 179 | ctx->buf[partial++] = 0x80;
|
| 180 | if (partial > SHA1_BLOCK_SIZE - 8) {
|
| 181 | memset(&ctx->buf[partial], 0, SHA1_BLOCK_SIZE - partial);
|
| 182 | sha1_blocks(&ctx->state, ctx->buf, 1);
|
| 183 | partial = 0;
|
| 184 | }
|
| 185 | memset(&ctx->buf[partial], 0, SHA1_BLOCK_SIZE - 8 - partial);
|
| 186 | *(__be64 *)&ctx->buf[SHA1_BLOCK_SIZE - 8] = cpu_to_be64(bitcount);
|
| 187 | sha1_blocks(&ctx->state, ctx->buf, 1);
|
| 188 |
|
| 189 | for (size_t i = 0; i < SHA1_DIGEST_SIZE; i += 4)
|
| 190 | put_unaligned_be32(ctx->state.h[i / 4], out + i);
|
| 191 | }
|
| 192 |
|
| 193 | void sha1_final(struct sha1_ctx *ctx, u8 out[SHA1_DIGEST_SIZE])
|
| 194 | {
|
| 195 | __sha1_final(ctx, out);
|
| 196 | memzero_explicit(ctx, sizeof(*ctx));
|
| 197 | }
|
| 198 | EXPORT_SYMBOL_GPL(sha1_final);
|
| 199 |
|
| 200 | void sha1(const u8 *data, size_t len, u8 out[SHA1_DIGEST_SIZE])
|
| 201 | {
|
| 202 | struct sha1_ctx ctx;
|
| 203 |
|
| 204 | sha1_init(&ctx);
|
| 205 | sha1_update(&ctx, data, len);
|
| 206 | sha1_final(&ctx, out);
|
| 207 | }
|
| 208 | EXPORT_SYMBOL_GPL(sha1);
|
| 209 |
|
| 210 | static void __hmac_sha1_preparekey(struct sha1_block_state *istate,
|
| 211 | struct sha1_block_state *ostate,
|
| 212 | const u8 *raw_key, size_t raw_key_len)
|
| 213 | {
|
| 214 | union {
|
| 215 | u8 b[SHA1_BLOCK_SIZE];
|
| 216 | unsigned long w[SHA1_BLOCK_SIZE / sizeof(unsigned long)];
|
| 217 | } derived_key = { 0 };
|
| 218 |
|
| 219 | if (unlikely(raw_key_len > SHA1_BLOCK_SIZE))
|
| 220 | sha1(raw_key, raw_key_len, derived_key.b);
|
| 221 | else
|
| 222 | memcpy(derived_key.b, raw_key, raw_key_len);
|
| 223 |
|
| 224 | for (size_t i = 0; i < ARRAY_SIZE(derived_key.w); i++)
|
| 225 | derived_key.w[i] ^= REPEAT_BYTE(HMAC_IPAD_VALUE);
|
| 226 | *istate = sha1_iv;
|
| 227 | sha1_blocks(istate, derived_key.b, 1);
|
| 228 |
|
| 229 | for (size_t i = 0; i < ARRAY_SIZE(derived_key.w); i++)
|
| 230 | derived_key.w[i] ^= REPEAT_BYTE(HMAC_OPAD_VALUE ^
|
| 231 | HMAC_IPAD_VALUE);
|
| 232 | *ostate = sha1_iv;
|
| 233 | sha1_blocks(ostate, derived_key.b, 1);
|
| 234 |
|
| 235 | memzero_explicit(&derived_key, sizeof(derived_key));
|
| 236 | }
|
| 237 |
|
| 238 | void hmac_sha1_preparekey(struct hmac_sha1_key *key,
|
| 239 | const u8 *raw_key, size_t raw_key_len)
|
| 240 | {
|
| 241 | __hmac_sha1_preparekey(&key->istate, &key->ostate,
|
| 242 | raw_key, raw_key_len);
|
| 243 | }
|
| 244 | EXPORT_SYMBOL_GPL(hmac_sha1_preparekey);
|
| 245 |
|
| 246 | void hmac_sha1_init(struct hmac_sha1_ctx *ctx, const struct hmac_sha1_key *key)
|
| 247 | {
|
| 248 | ctx->sha_ctx.state = key->istate;
|
| 249 | ctx->sha_ctx.bytecount = SHA1_BLOCK_SIZE;
|
| 250 | ctx->ostate = key->ostate;
|
| 251 | }
|
| 252 | EXPORT_SYMBOL_GPL(hmac_sha1_init);
|
| 253 |
|
| 254 | void hmac_sha1_init_usingrawkey(struct hmac_sha1_ctx *ctx,
|
| 255 | const u8 *raw_key, size_t raw_key_len)
|
| 256 | {
|
| 257 | __hmac_sha1_preparekey(&ctx->sha_ctx.state, &ctx->ostate,
|
| 258 | raw_key, raw_key_len);
|
| 259 | ctx->sha_ctx.bytecount = SHA1_BLOCK_SIZE;
|
| 260 | }
|
| 261 | EXPORT_SYMBOL_GPL(hmac_sha1_init_usingrawkey);
|
| 262 |
|
| 263 | void hmac_sha1_final(struct hmac_sha1_ctx *ctx, u8 out[SHA1_DIGEST_SIZE])
|
| 264 | {
|
| 265 |
|
| 266 | __sha1_final(&ctx->sha_ctx, ctx->sha_ctx.buf);
|
| 267 | memset(&ctx->sha_ctx.buf[SHA1_DIGEST_SIZE], 0,
|
| 268 | SHA1_BLOCK_SIZE - SHA1_DIGEST_SIZE);
|
| 269 | ctx->sha_ctx.buf[SHA1_DIGEST_SIZE] = 0x80;
|
| 270 | *(__be32 *)&ctx->sha_ctx.buf[SHA1_BLOCK_SIZE - 4] =
|
| 271 | cpu_to_be32(8 * (SHA1_BLOCK_SIZE + SHA1_DIGEST_SIZE));
|
| 272 |
|
| 273 |
|
| 274 | sha1_blocks(&ctx->ostate, ctx->sha_ctx.buf, 1);
|
| 275 | for (size_t i = 0; i < SHA1_DIGEST_SIZE; i += 4)
|
| 276 | put_unaligned_be32(ctx->ostate.h[i / 4], out + i);
|
| 277 |
|
| 278 | hmac_sha1_zeroize_ctx(ctx);
|
| 279 | }
|
| 280 | EXPORT_SYMBOL_GPL(hmac_sha1_final);
|
| 281 |
|
| 282 | void hmac_sha1(const struct hmac_sha1_key *key,
|
| 283 | const u8 *data, size_t data_len, u8 out[SHA1_DIGEST_SIZE])
|
| 284 | {
|
| 285 | struct hmac_sha1_ctx ctx;
|
| 286 |
|
| 287 | hmac_sha1_init(&ctx, key);
|
| 288 | hmac_sha1_update(&ctx, data, data_len);
|
| 289 | hmac_sha1_final(&ctx, out);
|
| 290 | }
|
| 291 | EXPORT_SYMBOL_GPL(hmac_sha1);
|
| 292 |
|
| 293 | void hmac_sha1_usingrawkey(const u8 *raw_key, size_t raw_key_len,
|
| 294 | const u8 *data, size_t data_len,
|
| 295 | u8 out[SHA1_DIGEST_SIZE])
|
| 296 | {
|
| 297 | struct hmac_sha1_ctx ctx;
|
| 298 |
|
| 299 | hmac_sha1_init_usingrawkey(&ctx, raw_key, raw_key_len);
|
| 300 | hmac_sha1_update(&ctx, data, data_len);
|
| 301 | hmac_sha1_final(&ctx, out);
|
| 302 | }
|
| 303 | EXPORT_SYMBOL_GPL(hmac_sha1_usingrawkey);
|
| 304 |
|
| 305 | #if defined(sha1_mod_init_arch) || defined(CONFIG_CRYPTO_FIPS)
|
| 306 | static int __init sha1_mod_init(void)
|
| 307 | {
|
| 308 | #ifdef sha1_mod_init_arch
|
| 309 | sha1_mod_init_arch();
|
| 310 | #endif
|
| 311 | if (fips_enabled) {
|
| 312 | |
| 313 | * FIPS cryptographic algorithm self-test. As per the FIPS
|
| 314 | * Implementation Guidance, testing HMAC-SHA1 satisfies the test
|
| 315 | * requirement for SHA-1 too.
|
| 316 |
|
| 317 | u8 mac[SHA1_DIGEST_SIZE];
|
| 318 |
|
| 319 | hmac_sha1_usingrawkey(fips_test_key, sizeof(fips_test_key),
|
| 320 | fips_test_data, sizeof(fips_test_data),
|
| 321 | mac);
|
| 322 | if (memcmp(fips_test_hmac_sha1_value, mac, sizeof(mac)) != 0)
|
| 323 | panic("sha1: FIPS self-test failed\n");
|
| 324 | }
|
| 325 | return 0;
|
| 326 | }
|
| 327 | subsys_initcall(sha1_mod_init);
|
| 328 |
|
| 329 | static void __exit sha1_mod_exit(void)
|
| 330 | {
|
| 331 | }
|
| 332 | module_exit(sha1_mod_exit);
|
| 333 | #endif
|
| 334 |
|
| 335 | MODULE_DESCRIPTION("SHA-1 and HMAC-SHA1 library functions");
|
| 336 | MODULE_LICENSE("GPL");
|