1 /*- 2 * Copyright (c) 2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 #include <sys/param.h> 29 #include <sys/systm.h> 30 #include <sys/kernel.h> 31 #include <sys/module.h> 32 #include <sys/malloc.h> 33 #include <sys/libkern.h> 34 #include <sys/endian.h> 35 #include <sys/pcpu.h> 36 #if defined(__amd64__) || defined(__i386__) 37 #include <machine/cpufunc.h> 38 #include <machine/cputypes.h> 39 #include <machine/fpu.h> 40 #include <machine/md_var.h> 41 #include <machine/specialreg.h> 42 #endif 43 #include <machine/pcb.h> 44 45 #include <opencrypto/cryptodev.h> 46 #include <opencrypto/xform.h> 47 48 #include <crypto/via/padlock.h> 49 50 /* 51 * Implementation notes. 52 * 53 * Some VIA CPUs provides SHA1 and SHA256 acceleration. 54 * We implement all HMAC algorithms provided by crypto(9) framework, but we do 55 * the crypto work in software unless this is HMAC/SHA1 or HMAC/SHA256 and 56 * our CPU can accelerate it. 57 * 58 * Additional CPU instructions, which preform SHA1 and SHA256 are one-shot 59 * functions - we have only one chance to give the data, CPU itself will add 60 * the padding and calculate hash automatically. 61 * This means, it is not possible to implement common init(), update(), final() 62 * methods. 63 * The way I've choosen is to keep adding data to the buffer on update() 64 * (reallocating the buffer if necessary) and call XSHA{1,256} instruction on 65 * final(). 66 */ 67 68 struct padlock_sha_ctx { 69 uint8_t *psc_buf; 70 int psc_offset; 71 int psc_size; 72 }; 73 CTASSERT(sizeof(struct padlock_sha_ctx) <= sizeof(union authctx)); 74 75 static void padlock_sha_init(void *vctx); 76 static int padlock_sha_update(void *vctx, const void *buf, u_int bufsize); 77 static void padlock_sha1_final(uint8_t *hash, void *vctx); 78 static void padlock_sha256_final(uint8_t *hash, void *vctx); 79 80 static const struct auth_hash padlock_hmac_sha1 = { 81 .type = CRYPTO_SHA1_HMAC, 82 .name = "HMAC-SHA1", 83 .keysize = SHA1_BLOCK_LEN, 84 .hashsize = SHA1_HASH_LEN, 85 .ctxsize = sizeof(struct padlock_sha_ctx), 86 .blocksize = SHA1_BLOCK_LEN, 87 .Init = padlock_sha_init, 88 .Update = padlock_sha_update, 89 .Final = padlock_sha1_final, 90 }; 91 92 static const struct auth_hash padlock_hmac_sha256 = { 93 .type = CRYPTO_SHA2_256_HMAC, 94 .name = "HMAC-SHA2-256", 95 .keysize = SHA2_256_BLOCK_LEN, 96 .hashsize = SHA2_256_HASH_LEN, 97 .ctxsize = sizeof(struct padlock_sha_ctx), 98 .blocksize = SHA2_256_BLOCK_LEN, 99 .Init = padlock_sha_init, 100 .Update = padlock_sha_update, 101 .Final = padlock_sha256_final, 102 }; 103 104 MALLOC_DECLARE(M_PADLOCK); 105 106 static __inline void 107 padlock_output_block(uint32_t *src, uint32_t *dst, size_t count) 108 { 109 110 while (count-- > 0) 111 *dst++ = bswap32(*src++); 112 } 113 114 static void 115 padlock_do_sha1(const u_char *in, u_char *out, int count) 116 { 117 u_char buf[128+16]; /* PadLock needs at least 128 bytes buffer. */ 118 u_char *result = PADLOCK_ALIGN(buf); 119 120 ((uint32_t *)result)[0] = 0x67452301; 121 ((uint32_t *)result)[1] = 0xEFCDAB89; 122 ((uint32_t *)result)[2] = 0x98BADCFE; 123 ((uint32_t *)result)[3] = 0x10325476; 124 ((uint32_t *)result)[4] = 0xC3D2E1F0; 125 126 __asm __volatile( 127 ".byte 0xf3, 0x0f, 0xa6, 0xc8" /* rep xsha1 */ 128 : "+S"(in), "+D"(result) 129 : "c"(count), "a"(0) 130 ); 131 132 padlock_output_block((uint32_t *)result, (uint32_t *)out, 133 SHA1_HASH_LEN / sizeof(uint32_t)); 134 } 135 136 static void 137 padlock_do_sha256(const char *in, char *out, int count) 138 { 139 char buf[128+16]; /* PadLock needs at least 128 bytes buffer. */ 140 char *result = PADLOCK_ALIGN(buf); 141 142 ((uint32_t *)result)[0] = 0x6A09E667; 143 ((uint32_t *)result)[1] = 0xBB67AE85; 144 ((uint32_t *)result)[2] = 0x3C6EF372; 145 ((uint32_t *)result)[3] = 0xA54FF53A; 146 ((uint32_t *)result)[4] = 0x510E527F; 147 ((uint32_t *)result)[5] = 0x9B05688C; 148 ((uint32_t *)result)[6] = 0x1F83D9AB; 149 ((uint32_t *)result)[7] = 0x5BE0CD19; 150 151 __asm __volatile( 152 ".byte 0xf3, 0x0f, 0xa6, 0xd0" /* rep xsha256 */ 153 : "+S"(in), "+D"(result) 154 : "c"(count), "a"(0) 155 ); 156 157 padlock_output_block((uint32_t *)result, (uint32_t *)out, 158 SHA2_256_HASH_LEN / sizeof(uint32_t)); 159 } 160 161 static void 162 padlock_sha_init(void *vctx) 163 { 164 struct padlock_sha_ctx *ctx; 165 166 ctx = vctx; 167 ctx->psc_buf = NULL; 168 ctx->psc_offset = 0; 169 ctx->psc_size = 0; 170 } 171 172 static int 173 padlock_sha_update(void *vctx, const void *buf, u_int bufsize) 174 { 175 struct padlock_sha_ctx *ctx; 176 177 ctx = vctx; 178 if (ctx->psc_size - ctx->psc_offset < bufsize) { 179 ctx->psc_size = MAX(ctx->psc_size * 2, ctx->psc_size + bufsize); 180 ctx->psc_buf = realloc(ctx->psc_buf, ctx->psc_size, M_PADLOCK, 181 M_NOWAIT); 182 if(ctx->psc_buf == NULL) 183 return (ENOMEM); 184 } 185 bcopy(buf, ctx->psc_buf + ctx->psc_offset, bufsize); 186 ctx->psc_offset += bufsize; 187 return (0); 188 } 189 190 static void 191 padlock_sha_free(void *vctx) 192 { 193 struct padlock_sha_ctx *ctx; 194 195 ctx = vctx; 196 if (ctx->psc_buf != NULL) { 197 zfree(ctx->psc_buf, M_PADLOCK); 198 ctx->psc_buf = NULL; 199 ctx->psc_offset = 0; 200 ctx->psc_size = 0; 201 } 202 } 203 204 static void 205 padlock_sha1_final(uint8_t *hash, void *vctx) 206 { 207 struct padlock_sha_ctx *ctx; 208 209 ctx = vctx; 210 padlock_do_sha1(ctx->psc_buf, hash, ctx->psc_offset); 211 padlock_sha_free(ctx); 212 } 213 214 static void 215 padlock_sha256_final(uint8_t *hash, void *vctx) 216 { 217 struct padlock_sha_ctx *ctx; 218 219 ctx = vctx; 220 padlock_do_sha256(ctx->psc_buf, hash, ctx->psc_offset); 221 padlock_sha_free(ctx); 222 } 223 224 static void 225 padlock_copy_ctx(const struct auth_hash *axf, void *sctx, void *dctx) 226 { 227 228 if ((via_feature_xcrypt & VIA_HAS_SHA) != 0 && 229 (axf->type == CRYPTO_SHA1_HMAC || 230 axf->type == CRYPTO_SHA2_256_HMAC)) { 231 struct padlock_sha_ctx *spctx = sctx, *dpctx = dctx; 232 233 dpctx->psc_offset = spctx->psc_offset; 234 dpctx->psc_size = spctx->psc_size; 235 dpctx->psc_buf = malloc(dpctx->psc_size, M_PADLOCK, M_WAITOK); 236 bcopy(spctx->psc_buf, dpctx->psc_buf, dpctx->psc_size); 237 } else { 238 bcopy(sctx, dctx, axf->ctxsize); 239 } 240 } 241 242 static void 243 padlock_free_ctx(const struct auth_hash *axf, void *ctx) 244 { 245 246 if ((via_feature_xcrypt & VIA_HAS_SHA) != 0 && 247 (axf->type == CRYPTO_SHA1_HMAC || 248 axf->type == CRYPTO_SHA2_256_HMAC)) { 249 padlock_sha_free(ctx); 250 } 251 } 252 253 static void 254 padlock_hash_key_setup(struct padlock_session *ses, const uint8_t *key, 255 int klen) 256 { 257 const struct auth_hash *axf; 258 259 axf = ses->ses_axf; 260 261 /* 262 * Try to free contexts before using them, because 263 * padlock_hash_key_setup() can be called twice - once from 264 * padlock_newsession() and again from padlock_process(). 265 */ 266 padlock_free_ctx(axf, ses->ses_ictx); 267 padlock_free_ctx(axf, ses->ses_octx); 268 269 hmac_init_ipad(axf, key, klen, ses->ses_ictx); 270 hmac_init_opad(axf, key, klen, ses->ses_octx); 271 } 272 273 /* 274 * Compute keyed-hash authenticator. 275 */ 276 static int 277 padlock_authcompute(struct padlock_session *ses, struct cryptop *crp) 278 { 279 u_char hash[HASH_MAX_LEN], hash2[HASH_MAX_LEN]; 280 const struct auth_hash *axf; 281 union authctx ctx; 282 int error; 283 284 axf = ses->ses_axf; 285 286 padlock_copy_ctx(axf, ses->ses_ictx, &ctx); 287 error = crypto_apply(crp, crp->crp_aad_start, crp->crp_aad_length, 288 axf->Update, &ctx); 289 if (error != 0) { 290 padlock_free_ctx(axf, &ctx); 291 return (error); 292 } 293 error = crypto_apply(crp, crp->crp_payload_start, 294 crp->crp_payload_length, axf->Update, &ctx); 295 if (error != 0) { 296 padlock_free_ctx(axf, &ctx); 297 return (error); 298 } 299 axf->Final(hash, &ctx); 300 301 padlock_copy_ctx(axf, ses->ses_octx, &ctx); 302 axf->Update(&ctx, hash, axf->hashsize); 303 axf->Final(hash, &ctx); 304 305 if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) { 306 crypto_copydata(crp, crp->crp_digest_start, ses->ses_mlen, 307 hash2); 308 if (timingsafe_bcmp(hash, hash2, ses->ses_mlen) != 0) 309 return (EBADMSG); 310 } else 311 crypto_copyback(crp, crp->crp_digest_start, ses->ses_mlen, 312 hash); 313 return (0); 314 } 315 316 /* Find software structure which describes HMAC algorithm. */ 317 static const struct auth_hash * 318 padlock_hash_lookup(int alg) 319 { 320 const struct auth_hash *axf; 321 322 switch (alg) { 323 case CRYPTO_NULL_HMAC: 324 axf = &auth_hash_null; 325 break; 326 case CRYPTO_SHA1_HMAC: 327 if ((via_feature_xcrypt & VIA_HAS_SHA) != 0) 328 axf = &padlock_hmac_sha1; 329 else 330 axf = &auth_hash_hmac_sha1; 331 break; 332 case CRYPTO_RIPEMD160_HMAC: 333 axf = &auth_hash_hmac_ripemd_160; 334 break; 335 case CRYPTO_SHA2_256_HMAC: 336 if ((via_feature_xcrypt & VIA_HAS_SHA) != 0) 337 axf = &padlock_hmac_sha256; 338 else 339 axf = &auth_hash_hmac_sha2_256; 340 break; 341 case CRYPTO_SHA2_384_HMAC: 342 axf = &auth_hash_hmac_sha2_384; 343 break; 344 case CRYPTO_SHA2_512_HMAC: 345 axf = &auth_hash_hmac_sha2_512; 346 break; 347 default: 348 axf = NULL; 349 break; 350 } 351 return (axf); 352 } 353 354 bool 355 padlock_hash_check(const struct crypto_session_params *csp) 356 { 357 358 return (padlock_hash_lookup(csp->csp_auth_alg) != NULL); 359 } 360 361 int 362 padlock_hash_setup(struct padlock_session *ses, 363 const struct crypto_session_params *csp) 364 { 365 366 ses->ses_axf = padlock_hash_lookup(csp->csp_auth_alg); 367 if (csp->csp_auth_mlen == 0) 368 ses->ses_mlen = ses->ses_axf->hashsize; 369 else 370 ses->ses_mlen = csp->csp_auth_mlen; 371 372 /* Allocate memory for HMAC inner and outer contexts. */ 373 ses->ses_ictx = malloc(ses->ses_axf->ctxsize, M_PADLOCK, 374 M_ZERO | M_NOWAIT); 375 ses->ses_octx = malloc(ses->ses_axf->ctxsize, M_PADLOCK, 376 M_ZERO | M_NOWAIT); 377 if (ses->ses_ictx == NULL || ses->ses_octx == NULL) 378 return (ENOMEM); 379 380 /* Setup key if given. */ 381 if (csp->csp_auth_key != NULL) { 382 padlock_hash_key_setup(ses, csp->csp_auth_key, 383 csp->csp_auth_klen); 384 } 385 return (0); 386 } 387 388 int 389 padlock_hash_process(struct padlock_session *ses, struct cryptop *crp, 390 const struct crypto_session_params *csp) 391 { 392 struct thread *td; 393 int error; 394 395 td = curthread; 396 fpu_kern_enter(td, NULL, FPU_KERN_NORMAL | FPU_KERN_NOCTX); 397 if (crp->crp_auth_key != NULL) 398 padlock_hash_key_setup(ses, crp->crp_auth_key, 399 csp->csp_auth_klen); 400 401 error = padlock_authcompute(ses, crp); 402 fpu_kern_leave(td, NULL); 403 return (error); 404 } 405 406 void 407 padlock_hash_free(struct padlock_session *ses) 408 { 409 410 if (ses->ses_ictx != NULL) { 411 padlock_free_ctx(ses->ses_axf, ses->ses_ictx); 412 zfree(ses->ses_ictx, M_PADLOCK); 413 ses->ses_ictx = NULL; 414 } 415 if (ses->ses_octx != NULL) { 416 padlock_free_ctx(ses->ses_axf, ses->ses_octx); 417 zfree(ses->ses_octx, M_PADLOCK); 418 ses->ses_octx = NULL; 419 } 420 } 421