1 /*- 2 * Copyright (c) 2005-2010 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 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/linker.h> 34 #include <sys/module.h> 35 #include <sys/lock.h> 36 #include <sys/mutex.h> 37 #include <sys/bio.h> 38 #include <sys/sysctl.h> 39 #include <sys/malloc.h> 40 #include <sys/kthread.h> 41 #include <sys/proc.h> 42 #include <sys/sched.h> 43 #include <sys/smp.h> 44 #include <sys/uio.h> 45 #include <sys/vnode.h> 46 47 #include <vm/uma.h> 48 49 #include <geom/geom.h> 50 #include <geom/eli/g_eli.h> 51 #include <geom/eli/pkcs5v2.h> 52 53 /* 54 * The data layout description when integrity verification is configured. 55 * 56 * One of the most important assumption here is that authenticated data and its 57 * HMAC has to be stored in the same place (namely in the same sector) to make 58 * it work reliable. 59 * The problem is that file systems work only with sectors that are multiple of 60 * 512 bytes and a power of two number. 61 * My idea to implement it is as follows. 62 * Let's store HMAC in sector. This is a must. This leaves us 480 bytes for 63 * data. We can't use that directly (ie. we can't create provider with 480 bytes 64 * sector size). We need another sector from where we take only 32 bytes of data 65 * and we store HMAC of this data as well. This takes two sectors from the 66 * original provider at the input and leaves us one sector of authenticated data 67 * at the output. Not very efficient, but you got the idea. 68 * Now, let's assume, we want to create provider with 4096 bytes sector. 69 * To output 4096 bytes of authenticated data we need 8x480 plus 1x256, so we 70 * need nine 512-bytes sectors at the input to get one 4096-bytes sector at the 71 * output. That's better. With 4096 bytes sector we can use 89% of size of the 72 * original provider. I find it as an acceptable cost. 73 * The reliability comes from the fact, that every HMAC stored inside the sector 74 * is calculated only for the data in the same sector, so its impossible to 75 * write new data and leave old HMAC or vice versa. 76 * 77 * And here is the picture: 78 * 79 * da0: +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+-----+ 80 * |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |256b | 81 * |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data | 82 * +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+-----+ 83 * |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |288 bytes | 84 * +---------+ +---------+ +---------+ +---------+ +---------+ +---------+ +---------+ +---------+ |224 unused| 85 * +----------+ 86 * da0.eli: +----+----+----+----+----+----+----+----+----+ 87 * |480b|480b|480b|480b|480b|480b|480b|480b|256b| 88 * +----+----+----+----+----+----+----+----+----+ 89 * | 4096 bytes | 90 * +--------------------------------------------+ 91 * 92 * PS. You can use any sector size with geli(8). My example is using 4kB, 93 * because it's most efficient. For 8kB sectors you need 2 extra sectors, 94 * so the cost is the same as for 4kB sectors. 95 */ 96 97 /* 98 * Code paths: 99 * BIO_READ: 100 * g_eli_start -> g_eli_auth_read -> g_io_request -> g_eli_read_done -> g_eli_auth_run -> g_eli_auth_read_done -> g_io_deliver 101 * BIO_WRITE: 102 * g_eli_start -> g_eli_auth_run -> g_eli_auth_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver 103 */ 104 105 MALLOC_DECLARE(M_ELI); 106 107 /* 108 * Here we generate key for HMAC. Every sector has its own HMAC key, so it is 109 * not possible to copy sectors. 110 * We cannot depend on fact, that every sector has its own IV, because different 111 * IV doesn't change HMAC, when we use encrypt-then-authenticate method. 112 */ 113 static void 114 g_eli_auth_keygen(struct g_eli_softc *sc, off_t offset, u_char *key) 115 { 116 SHA256_CTX ctx; 117 118 /* Copy precalculated SHA256 context. */ 119 bcopy(&sc->sc_akeyctx, &ctx, sizeof(ctx)); 120 SHA256_Update(&ctx, (uint8_t *)&offset, sizeof(offset)); 121 SHA256_Final(key, &ctx); 122 } 123 124 /* 125 * The function is called after we read and decrypt data. 126 * 127 * g_eli_start -> g_eli_auth_read -> g_io_request -> g_eli_read_done -> g_eli_auth_run -> G_ELI_AUTH_READ_DONE -> g_io_deliver 128 */ 129 static int 130 g_eli_auth_read_done(struct cryptop *crp) 131 { 132 struct g_eli_softc *sc; 133 struct bio *bp; 134 135 if (crp->crp_etype == EAGAIN) { 136 if (g_eli_crypto_rerun(crp) == 0) 137 return (0); 138 } 139 bp = (struct bio *)crp->crp_opaque; 140 bp->bio_inbed++; 141 if (crp->crp_etype == 0) { 142 bp->bio_completed += crp->crp_olen; 143 G_ELI_DEBUG(3, "Crypto READ request done (%d/%d) (add=%jd completed=%jd).", 144 bp->bio_inbed, bp->bio_children, (intmax_t)crp->crp_olen, (intmax_t)bp->bio_completed); 145 } else { 146 G_ELI_DEBUG(1, "Crypto READ request failed (%d/%d) error=%d.", 147 bp->bio_inbed, bp->bio_children, crp->crp_etype); 148 if (bp->bio_error == 0) 149 bp->bio_error = crp->crp_etype; 150 } 151 /* 152 * Do we have all sectors already? 153 */ 154 if (bp->bio_inbed < bp->bio_children) 155 return (0); 156 sc = bp->bio_to->geom->softc; 157 if (bp->bio_error == 0) { 158 u_int i, lsec, nsec, data_secsize, decr_secsize, encr_secsize; 159 u_char *srcdata, *dstdata, *auth; 160 off_t coroff, corsize; 161 162 /* 163 * Verify data integrity based on calculated and read HMACs. 164 */ 165 /* Sectorsize of decrypted provider eg. 4096. */ 166 decr_secsize = bp->bio_to->sectorsize; 167 /* The real sectorsize of encrypted provider, eg. 512. */ 168 encr_secsize = LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize; 169 /* Number of data bytes in one encrypted sector, eg. 480. */ 170 data_secsize = sc->sc_data_per_sector; 171 /* Number of sectors from decrypted provider, eg. 2. */ 172 nsec = bp->bio_length / decr_secsize; 173 /* Number of sectors from encrypted provider, eg. 18. */ 174 nsec = (nsec * sc->sc_bytes_per_sector) / encr_secsize; 175 /* Last sector number in every big sector, eg. 9. */ 176 lsec = sc->sc_bytes_per_sector / encr_secsize; 177 178 srcdata = bp->bio_driver2; 179 dstdata = bp->bio_data; 180 auth = srcdata + encr_secsize * nsec; 181 coroff = -1; 182 corsize = 0; 183 184 for (i = 1; i <= nsec; i++) { 185 data_secsize = sc->sc_data_per_sector; 186 if ((i % lsec) == 0) 187 data_secsize = decr_secsize % data_secsize; 188 if (bcmp(srcdata, auth, sc->sc_alen) != 0) { 189 /* 190 * Curruption detected, remember the offset if 191 * this is the first corrupted sector and 192 * increase size. 193 */ 194 if (bp->bio_error == 0) 195 bp->bio_error = -1; 196 if (coroff == -1) { 197 coroff = bp->bio_offset + 198 (dstdata - (u_char *)bp->bio_data); 199 } 200 corsize += data_secsize; 201 } else { 202 /* 203 * No curruption, good. 204 * Report previous corruption if there was one. 205 */ 206 if (coroff != -1) { 207 G_ELI_DEBUG(0, "%s: %jd bytes " 208 "corrupted at offset %jd.", 209 sc->sc_name, (intmax_t)corsize, 210 (intmax_t)coroff); 211 coroff = -1; 212 corsize = 0; 213 } 214 bcopy(srcdata + sc->sc_alen, dstdata, 215 data_secsize); 216 } 217 srcdata += encr_secsize; 218 dstdata += data_secsize; 219 auth += sc->sc_alen; 220 } 221 /* Report previous corruption if there was one. */ 222 if (coroff != -1) { 223 G_ELI_DEBUG(0, "%s: %jd bytes corrupted at offset %jd.", 224 sc->sc_name, (intmax_t)corsize, (intmax_t)coroff); 225 } 226 } 227 free(bp->bio_driver2, M_ELI); 228 bp->bio_driver2 = NULL; 229 if (bp->bio_error != 0) { 230 if (bp->bio_error == -1) 231 bp->bio_error = EINVAL; 232 else { 233 G_ELI_LOGREQ(0, bp, 234 "Crypto READ request failed (error=%d).", 235 bp->bio_error); 236 } 237 bp->bio_completed = 0; 238 } 239 /* 240 * Read is finished, send it up. 241 */ 242 g_io_deliver(bp, bp->bio_error); 243 atomic_subtract_int(&sc->sc_inflight, 1); 244 return (0); 245 } 246 247 /* 248 * The function is called after data encryption. 249 * 250 * g_eli_start -> g_eli_auth_run -> G_ELI_AUTH_WRITE_DONE -> g_io_request -> g_eli_write_done -> g_io_deliver 251 */ 252 static int 253 g_eli_auth_write_done(struct cryptop *crp) 254 { 255 struct g_eli_softc *sc; 256 struct g_consumer *cp; 257 struct bio *bp, *cbp, *cbp2; 258 u_int nsec; 259 260 if (crp->crp_etype == EAGAIN) { 261 if (g_eli_crypto_rerun(crp) == 0) 262 return (0); 263 } 264 bp = (struct bio *)crp->crp_opaque; 265 bp->bio_inbed++; 266 if (crp->crp_etype == 0) { 267 G_ELI_DEBUG(3, "Crypto WRITE request done (%d/%d).", 268 bp->bio_inbed, bp->bio_children); 269 } else { 270 G_ELI_DEBUG(1, "Crypto WRITE request failed (%d/%d) error=%d.", 271 bp->bio_inbed, bp->bio_children, crp->crp_etype); 272 if (bp->bio_error == 0) 273 bp->bio_error = crp->crp_etype; 274 } 275 /* 276 * All sectors are already encrypted? 277 */ 278 if (bp->bio_inbed < bp->bio_children) 279 return (0); 280 sc = bp->bio_to->geom->softc; 281 if (bp->bio_error != 0) { 282 G_ELI_LOGREQ(0, bp, "Crypto WRITE request failed (error=%d).", 283 bp->bio_error); 284 free(bp->bio_driver2, M_ELI); 285 bp->bio_driver2 = NULL; 286 cbp = bp->bio_driver1; 287 bp->bio_driver1 = NULL; 288 g_destroy_bio(cbp); 289 g_io_deliver(bp, bp->bio_error); 290 atomic_subtract_int(&sc->sc_inflight, 1); 291 return (0); 292 } 293 cp = LIST_FIRST(&sc->sc_geom->consumer); 294 cbp = bp->bio_driver1; 295 bp->bio_driver1 = NULL; 296 cbp->bio_to = cp->provider; 297 cbp->bio_done = g_eli_write_done; 298 299 /* Number of sectors from decrypted provider, eg. 1. */ 300 nsec = bp->bio_length / bp->bio_to->sectorsize; 301 /* Number of sectors from encrypted provider, eg. 9. */ 302 nsec = (nsec * sc->sc_bytes_per_sector) / cp->provider->sectorsize; 303 304 cbp->bio_length = cp->provider->sectorsize * nsec; 305 cbp->bio_offset = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector; 306 cbp->bio_data = bp->bio_driver2; 307 308 /* 309 * We write more than what is requested, so we have to be ready to write 310 * more than MAXPHYS. 311 */ 312 cbp2 = NULL; 313 if (cbp->bio_length > MAXPHYS) { 314 cbp2 = g_duplicate_bio(bp); 315 cbp2->bio_length = cbp->bio_length - MAXPHYS; 316 cbp2->bio_data = cbp->bio_data + MAXPHYS; 317 cbp2->bio_offset = cbp->bio_offset + MAXPHYS; 318 cbp2->bio_to = cp->provider; 319 cbp2->bio_done = g_eli_write_done; 320 cbp->bio_length = MAXPHYS; 321 } 322 /* 323 * Send encrypted data to the provider. 324 */ 325 G_ELI_LOGREQ(2, cbp, "Sending request."); 326 bp->bio_inbed = 0; 327 bp->bio_children = (cbp2 != NULL ? 2 : 1); 328 g_io_request(cbp, cp); 329 if (cbp2 != NULL) { 330 G_ELI_LOGREQ(2, cbp2, "Sending request."); 331 g_io_request(cbp2, cp); 332 } 333 return (0); 334 } 335 336 void 337 g_eli_auth_read(struct g_eli_softc *sc, struct bio *bp) 338 { 339 struct g_consumer *cp; 340 struct bio *cbp, *cbp2; 341 size_t size; 342 off_t nsec; 343 344 bp->bio_pflags = 0; 345 346 cp = LIST_FIRST(&sc->sc_geom->consumer); 347 cbp = bp->bio_driver1; 348 bp->bio_driver1 = NULL; 349 cbp->bio_to = cp->provider; 350 cbp->bio_done = g_eli_read_done; 351 352 /* Number of sectors from decrypted provider, eg. 1. */ 353 nsec = bp->bio_length / bp->bio_to->sectorsize; 354 /* Number of sectors from encrypted provider, eg. 9. */ 355 nsec = (nsec * sc->sc_bytes_per_sector) / cp->provider->sectorsize; 356 357 cbp->bio_length = cp->provider->sectorsize * nsec; 358 size = cbp->bio_length; 359 size += sc->sc_alen * nsec; 360 size += sizeof(struct cryptop) * nsec; 361 size += sizeof(struct cryptodesc) * nsec * 2; 362 size += G_ELI_AUTH_SECKEYLEN * nsec; 363 size += sizeof(struct uio) * nsec; 364 size += sizeof(struct iovec) * nsec; 365 cbp->bio_offset = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector; 366 bp->bio_driver2 = malloc(size, M_ELI, M_WAITOK); 367 cbp->bio_data = bp->bio_driver2; 368 369 /* 370 * We read more than what is requested, so we have to be ready to read 371 * more than MAXPHYS. 372 */ 373 cbp2 = NULL; 374 if (cbp->bio_length > MAXPHYS) { 375 cbp2 = g_duplicate_bio(bp); 376 cbp2->bio_length = cbp->bio_length - MAXPHYS; 377 cbp2->bio_data = cbp->bio_data + MAXPHYS; 378 cbp2->bio_offset = cbp->bio_offset + MAXPHYS; 379 cbp2->bio_to = cp->provider; 380 cbp2->bio_done = g_eli_read_done; 381 cbp->bio_length = MAXPHYS; 382 } 383 /* 384 * Read encrypted data from provider. 385 */ 386 G_ELI_LOGREQ(2, cbp, "Sending request."); 387 g_io_request(cbp, cp); 388 if (cbp2 != NULL) { 389 G_ELI_LOGREQ(2, cbp2, "Sending request."); 390 g_io_request(cbp2, cp); 391 } 392 } 393 394 /* 395 * This is the main function responsible for cryptography (ie. communication 396 * with crypto(9) subsystem). 397 * 398 * BIO_READ: 399 * g_eli_start -> g_eli_auth_read -> g_io_request -> g_eli_read_done -> G_ELI_AUTH_RUN -> g_eli_auth_read_done -> g_io_deliver 400 * BIO_WRITE: 401 * g_eli_start -> G_ELI_AUTH_RUN -> g_eli_auth_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver 402 */ 403 void 404 g_eli_auth_run(struct g_eli_worker *wr, struct bio *bp) 405 { 406 struct g_eli_softc *sc; 407 struct cryptop *crp; 408 struct cryptodesc *crde, *crda; 409 struct uio *uio; 410 struct iovec *iov; 411 u_int i, lsec, nsec, data_secsize, decr_secsize, encr_secsize; 412 off_t dstoff; 413 int err, error; 414 u_char *p, *data, *auth, *authkey, *plaindata; 415 416 G_ELI_LOGREQ(3, bp, "%s", __func__); 417 418 bp->bio_pflags = wr->w_number; 419 sc = wr->w_softc; 420 /* Sectorsize of decrypted provider eg. 4096. */ 421 decr_secsize = bp->bio_to->sectorsize; 422 /* The real sectorsize of encrypted provider, eg. 512. */ 423 encr_secsize = LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize; 424 /* Number of data bytes in one encrypted sector, eg. 480. */ 425 data_secsize = sc->sc_data_per_sector; 426 /* Number of sectors from decrypted provider, eg. 2. */ 427 nsec = bp->bio_length / decr_secsize; 428 /* Number of sectors from encrypted provider, eg. 18. */ 429 nsec = (nsec * sc->sc_bytes_per_sector) / encr_secsize; 430 /* Last sector number in every big sector, eg. 9. */ 431 lsec = sc->sc_bytes_per_sector / encr_secsize; 432 /* Destination offset, used for IV generation. */ 433 dstoff = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector; 434 435 auth = NULL; /* Silence compiler warning. */ 436 plaindata = bp->bio_data; 437 if (bp->bio_cmd == BIO_READ) { 438 data = bp->bio_driver2; 439 auth = data + encr_secsize * nsec; 440 p = auth + sc->sc_alen * nsec; 441 } else { 442 size_t size; 443 444 size = encr_secsize * nsec; 445 size += sizeof(*crp) * nsec; 446 size += sizeof(*crde) * nsec; 447 size += sizeof(*crda) * nsec; 448 size += G_ELI_AUTH_SECKEYLEN * nsec; 449 size += sizeof(*uio) * nsec; 450 size += sizeof(*iov) * nsec; 451 data = malloc(size, M_ELI, M_WAITOK); 452 bp->bio_driver2 = data; 453 p = data + encr_secsize * nsec; 454 } 455 bp->bio_inbed = 0; 456 bp->bio_children = nsec; 457 458 error = 0; 459 for (i = 1; i <= nsec; i++, dstoff += encr_secsize) { 460 crp = (struct cryptop *)p; p += sizeof(*crp); 461 crde = (struct cryptodesc *)p; p += sizeof(*crde); 462 crda = (struct cryptodesc *)p; p += sizeof(*crda); 463 authkey = (u_char *)p; p += G_ELI_AUTH_SECKEYLEN; 464 uio = (struct uio *)p; p += sizeof(*uio); 465 iov = (struct iovec *)p; p += sizeof(*iov); 466 467 data_secsize = sc->sc_data_per_sector; 468 if ((i % lsec) == 0) 469 data_secsize = decr_secsize % data_secsize; 470 471 if (bp->bio_cmd == BIO_READ) { 472 /* Remember read HMAC. */ 473 bcopy(data, auth, sc->sc_alen); 474 auth += sc->sc_alen; 475 /* TODO: bzero(9) can be commented out later. */ 476 bzero(data, sc->sc_alen); 477 } else { 478 bcopy(plaindata, data + sc->sc_alen, data_secsize); 479 plaindata += data_secsize; 480 } 481 482 iov->iov_len = sc->sc_alen + data_secsize; 483 iov->iov_base = data; 484 data += encr_secsize; 485 486 uio->uio_iov = iov; 487 uio->uio_iovcnt = 1; 488 uio->uio_segflg = UIO_SYSSPACE; 489 uio->uio_resid = iov->iov_len; 490 491 crp->crp_sid = wr->w_sid; 492 crp->crp_ilen = uio->uio_resid; 493 crp->crp_olen = data_secsize; 494 crp->crp_opaque = (void *)bp; 495 crp->crp_buf = (void *)uio; 496 crp->crp_flags = CRYPTO_F_IOV | CRYPTO_F_CBIFSYNC | CRYPTO_F_REL; 497 if (g_eli_batch) 498 crp->crp_flags |= CRYPTO_F_BATCH; 499 if (bp->bio_cmd == BIO_WRITE) { 500 crp->crp_callback = g_eli_auth_write_done; 501 crp->crp_desc = crde; 502 crde->crd_next = crda; 503 crda->crd_next = NULL; 504 } else { 505 crp->crp_callback = g_eli_auth_read_done; 506 crp->crp_desc = crda; 507 crda->crd_next = crde; 508 crde->crd_next = NULL; 509 } 510 511 crde->crd_skip = sc->sc_alen; 512 crde->crd_len = data_secsize; 513 crde->crd_flags = CRD_F_IV_EXPLICIT | CRD_F_IV_PRESENT; 514 if (bp->bio_cmd == BIO_WRITE) 515 crde->crd_flags |= CRD_F_ENCRYPT; 516 crde->crd_alg = sc->sc_ealgo; 517 crde->crd_key = g_eli_crypto_key(sc, dstoff, encr_secsize); 518 crde->crd_klen = sc->sc_ekeylen; 519 if (sc->sc_ealgo == CRYPTO_AES_XTS) 520 crde->crd_klen <<= 1; 521 g_eli_crypto_ivgen(sc, dstoff, crde->crd_iv, 522 sizeof(crde->crd_iv)); 523 524 crda->crd_skip = sc->sc_alen; 525 crda->crd_len = data_secsize; 526 crda->crd_inject = 0; 527 crda->crd_flags = CRD_F_KEY_EXPLICIT; 528 crda->crd_alg = sc->sc_aalgo; 529 g_eli_auth_keygen(sc, dstoff, authkey); 530 crda->crd_key = authkey; 531 crda->crd_klen = G_ELI_AUTH_SECKEYLEN * 8; 532 533 crp->crp_etype = 0; 534 err = crypto_dispatch(crp); 535 if (err != 0 && error == 0) 536 error = err; 537 } 538 if (bp->bio_error == 0) 539 bp->bio_error = error; 540 } 541