1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2002 Poul-Henning Kamp 5 * Copyright (c) 2002 Networks Associates Technology, Inc. 6 * All rights reserved. 7 * 8 * This software was developed for the FreeBSD Project by Poul-Henning Kamp 9 * and NAI Labs, the Security Research Division of Network Associates, Inc. 10 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 11 * DARPA CHATS research program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. The names of the authors may not be used to endorse or promote 22 * products derived from this software without specific prior written 23 * permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38/* 39 * This is the method for dealing with BSD disklabels. It has been 40 * extensively (by my standards at least) commented, in the vain hope that 41 * it will serve as the source in future copy&paste operations. 42 */ 43 44#include <sys/cdefs.h> 45__FBSDID("$FreeBSD$"); 46 47#include <sys/param.h> 48#include <sys/endian.h> 49#include <sys/systm.h> 50#include <sys/sysctl.h> 51#include <sys/kernel.h> 52#include <sys/fcntl.h> 53#include <sys/conf.h> 54#include <sys/bio.h> 55#include <sys/malloc.h> 56#include <sys/lock.h> 57#include <sys/mutex.h> 58#include <sys/md5.h> 59#include <sys/errno.h> 60#include <sys/disklabel.h> 61#include <sys/gpt.h> 62#include <sys/proc.h> 63#include <sys/sbuf.h> 64#include <sys/uuid.h> 65#include <geom/geom.h> 66#include <geom/geom_slice.h> 67 68FEATURE(geom_bsd, "GEOM BSD disklabels support"); 69 70#define BSD_CLASS_NAME "BSD" 71 72#define ALPHA_LABEL_OFFSET 64 73#define HISTORIC_LABEL_OFFSET 512 74 75#define LABELSIZE (148 + 16 * MAXPARTITIONS) 76 77static int g_bsd_once; 78 79static void g_bsd_hotwrite(void *arg, int flag); 80/* 81 * Our private data about one instance. All the rest is handled by the 82 * slice code and stored in its softc, so this is just the stuff 83 * specific to BSD disklabels. 84 */ 85struct g_bsd_softc { 86 off_t labeloffset; 87 off_t mbroffset; 88 off_t rawoffset; 89 struct disklabel ondisk; 90 u_char label[LABELSIZE]; 91 u_char labelsum[16]; 92}; 93 94/* 95 * Modify our slicer to match proposed disklabel, if possible. 96 * This is where we make sure we don't do something stupid. 97 */ 98static int 99g_bsd_modify(struct g_geom *gp, u_char *label) 100{ 101 int i, error; 102 struct partition *ppp; 103 struct g_slicer *gsp; 104 struct g_consumer *cp; 105 struct g_bsd_softc *ms; 106 u_int secsize, u; 107 off_t rawoffset, o; 108 struct disklabel dl; 109 MD5_CTX md5sum; 110 111 g_topology_assert(); 112 gsp = gp->softc; 113 ms = gsp->softc; 114 115 error = bsd_disklabel_le_dec(label, &dl, MAXPARTITIONS); 116 if (error) { 117 return (error); 118 } 119 120 /* Get dimensions of our device. */ 121 cp = LIST_FIRST(&gp->consumer); 122 secsize = cp->provider->sectorsize; 123 124 /* ... or a smaller sector size. */ 125 if (dl.d_secsize < secsize) { 126 return (EINVAL); 127 } 128 129 /* ... or a non-multiple sector size. */ 130 if (dl.d_secsize % secsize != 0) { 131 return (EINVAL); 132 } 133 134 /* Historical braindamage... */ 135 rawoffset = (off_t)dl.d_partitions[RAW_PART].p_offset * dl.d_secsize; 136 137 for (i = 0; i < dl.d_npartitions; i++) { 138 ppp = &dl.d_partitions[i]; 139 if (ppp->p_size == 0) 140 continue; 141 o = (off_t)ppp->p_offset * dl.d_secsize; 142 143 if (o < rawoffset) 144 rawoffset = 0; 145 } 146 147 if (rawoffset != 0 && (off_t)rawoffset != ms->mbroffset) 148 printf("WARNING: %s expected rawoffset %jd, found %jd\n", 149 gp->name, 150 (intmax_t)ms->mbroffset/dl.d_secsize, 151 (intmax_t)rawoffset/dl.d_secsize); 152 153 /* Don't munge open partitions. */ 154 for (i = 0; i < dl.d_npartitions; i++) { 155 ppp = &dl.d_partitions[i]; 156 157 o = (off_t)ppp->p_offset * dl.d_secsize; 158 if (o == 0) 159 o = rawoffset; 160 error = g_slice_config(gp, i, G_SLICE_CONFIG_CHECK, 161 o - rawoffset, 162 (off_t)ppp->p_size * dl.d_secsize, 163 dl.d_secsize, 164 "%s%c", gp->name, 'a' + i); 165 if (error) 166 return (error); 167 } 168 169 /* Look good, go for it... */ 170 for (u = 0; u < gsp->nslice; u++) { 171 ppp = &dl.d_partitions[u]; 172 o = (off_t)ppp->p_offset * dl.d_secsize; 173 if (o == 0) 174 o = rawoffset; 175 g_slice_config(gp, u, G_SLICE_CONFIG_SET, 176 o - rawoffset, 177 (off_t)ppp->p_size * dl.d_secsize, 178 dl.d_secsize, 179 "%s%c", gp->name, 'a' + u); 180 } 181 182 /* Update our softc */ 183 ms->ondisk = dl; 184 if (label != ms->label) 185 bcopy(label, ms->label, LABELSIZE); 186 ms->rawoffset = rawoffset; 187 188 /* 189 * In order to avoid recursively attaching to the same 190 * on-disk label (it's usually visible through the 'c' 191 * partition) we calculate an MD5 and ask if other BSD's 192 * below us love that label. If they do, we don't. 193 */ 194 MD5Init(&md5sum); 195 MD5Update(&md5sum, ms->label, sizeof(ms->label)); 196 MD5Final(ms->labelsum, &md5sum); 197 198 return (0); 199} 200 201/* 202 * This is an internal helper function, called multiple times from the taste 203 * function to try to locate a disklabel on the disk. More civilized formats 204 * will not need this, as there is only one possible place on disk to look 205 * for the magic spot. 206 */ 207 208static int 209g_bsd_try(struct g_geom *gp, struct g_slicer *gsp, struct g_consumer *cp, int secsize, struct g_bsd_softc *ms, off_t offset) 210{ 211 int error; 212 u_char *buf; 213 struct disklabel *dl; 214 off_t secoff; 215 216 /* 217 * We need to read entire aligned sectors, and we assume that the 218 * disklabel does not span sectors, so one sector is enough. 219 */ 220 secoff = offset % secsize; 221 buf = g_read_data(cp, offset - secoff, secsize, NULL); 222 if (buf == NULL) 223 return (ENOENT); 224 225 /* Decode into our native format. */ 226 dl = &ms->ondisk; 227 error = bsd_disklabel_le_dec(buf + secoff, dl, MAXPARTITIONS); 228 if (!error) 229 bcopy(buf + secoff, ms->label, LABELSIZE); 230 231 /* Remember to free the buffer g_read_data() gave us. */ 232 g_free(buf); 233 234 ms->labeloffset = offset; 235 return (error); 236} 237 238/* 239 * This function writes the current label to disk, possibly updating 240 * the alpha SRM checksum. 241 */ 242 243static int 244g_bsd_writelabel(struct g_geom *gp, u_char *bootcode) 245{ 246 off_t secoff; 247 u_int secsize; 248 struct g_consumer *cp; 249 struct g_slicer *gsp; 250 struct g_bsd_softc *ms; 251 u_char *buf; 252 uint64_t sum; 253 int error, i; 254 255 gsp = gp->softc; 256 ms = gsp->softc; 257 cp = LIST_FIRST(&gp->consumer); 258 /* Get sector size, we need it to read data. */ 259 secsize = cp->provider->sectorsize; 260 secoff = ms->labeloffset % secsize; 261 if (bootcode == NULL) { 262 buf = g_read_data(cp, ms->labeloffset - secoff, secsize, &error); 263 if (buf == NULL) 264 return (error); 265 bcopy(ms->label, buf + secoff, sizeof(ms->label)); 266 } else { 267 buf = bootcode; 268 bcopy(ms->label, buf + ms->labeloffset, sizeof(ms->label)); 269 } 270 if (ms->labeloffset == ALPHA_LABEL_OFFSET) { 271 sum = 0; 272 for (i = 0; i < 63; i++) 273 sum += le64dec(buf + i * 8); 274 le64enc(buf + 504, sum); 275 } 276 if (bootcode == NULL) { 277 error = g_write_data(cp, ms->labeloffset - secoff, buf, secsize); 278 g_free(buf); 279 } else { 280 error = g_write_data(cp, 0, bootcode, BBSIZE); 281 } 282 return(error); 283} 284 285/* 286 * If the user tries to overwrite our disklabel through an open partition 287 * or via a magicwrite config call, we end up here and try to prevent 288 * footshooting as best we can. 289 */ 290static void 291g_bsd_hotwrite(void *arg, int flag) 292{ 293 struct bio *bp; 294 struct g_geom *gp; 295 struct g_slicer *gsp; 296 struct g_slice *gsl; 297 struct g_bsd_softc *ms; 298 u_char *p; 299 int error; 300 301 g_topology_assert(); 302 /* 303 * We should never get canceled, because that would amount to a removal 304 * of the geom while there was outstanding I/O requests. 305 */ 306 KASSERT(flag != EV_CANCEL, ("g_bsd_hotwrite cancelled")); 307 bp = arg; 308 gp = bp->bio_to->geom; 309 gsp = gp->softc; 310 ms = gsp->softc; 311 gsl = &gsp->slices[bp->bio_to->index]; 312 p = (u_char*)bp->bio_data + ms->labeloffset - 313 (bp->bio_offset + gsl->offset); 314 error = g_bsd_modify(gp, p); 315 if (error) { 316 g_io_deliver(bp, EPERM); 317 return; 318 } 319 g_slice_finish_hot(bp); 320} 321 322static int 323g_bsd_start(struct bio *bp) 324{ 325 struct g_geom *gp; 326 struct g_bsd_softc *ms; 327 struct g_slicer *gsp; 328 329 gp = bp->bio_to->geom; 330 gsp = gp->softc; 331 ms = gsp->softc; 332 if (bp->bio_cmd == BIO_GETATTR) { 333 if (g_handleattr(bp, "BSD::labelsum", ms->labelsum, 334 sizeof(ms->labelsum))) 335 return (1); 336 } 337 return (0); 338} 339 340/* 341 * Dump configuration information in XML format. 342 * Notice that the function is called once for the geom and once for each 343 * consumer and provider. We let g_slice_dumpconf() do most of the work. 344 */ 345static void 346g_bsd_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp) 347{ 348 struct g_bsd_softc *ms; 349 struct g_slicer *gsp; 350 351 gsp = gp->softc; 352 ms = gsp->softc; 353 g_slice_dumpconf(sb, indent, gp, cp, pp); 354 if (indent != NULL && pp == NULL && cp == NULL) { 355 sbuf_printf(sb, "%s<labeloffset>%jd</labeloffset>\n", 356 indent, (intmax_t)ms->labeloffset); 357 sbuf_printf(sb, "%s<rawoffset>%jd</rawoffset>\n", 358 indent, (intmax_t)ms->rawoffset); 359 sbuf_printf(sb, "%s<mbroffset>%jd</mbroffset>\n", 360 indent, (intmax_t)ms->mbroffset); 361 } else if (pp != NULL) { 362 if (indent == NULL) 363 sbuf_printf(sb, " ty %d", 364 ms->ondisk.d_partitions[pp->index].p_fstype); 365 else 366 sbuf_printf(sb, "%s<type>%d</type>\n", indent, 367 ms->ondisk.d_partitions[pp->index].p_fstype); 368 } 369} 370 371/* 372 * The taste function is called from the event-handler, with the topology 373 * lock already held and a provider to examine. The flags are unused. 374 * 375 * If flags == G_TF_NORMAL, the idea is to take a bite of the provider and 376 * if we find valid, consistent magic on it, build a geom on it. 377 * 378 * There may be cases where the operator would like to put a BSD-geom on 379 * providers which do not meet all of the requirements. This can be done 380 * by instead passing the G_TF_INSIST flag, which will override these 381 * checks. 382 * 383 * The final flags value is G_TF_TRANSPARENT, which instructs the method 384 * to put a geom on top of the provider and configure it to be as transparent 385 * as possible. This is not really relevant to the BSD method and therefore 386 * not implemented here. 387 */ 388 389static struct uuid freebsd_slice = GPT_ENT_TYPE_FREEBSD; 390 391static struct g_geom * 392g_bsd_taste(struct g_class *mp, struct g_provider *pp, int flags) 393{ 394 struct g_geom *gp; 395 struct g_consumer *cp; 396 int error, i; 397 struct g_bsd_softc *ms; 398 u_int secsize; 399 struct g_slicer *gsp; 400 u_char hash[16]; 401 MD5_CTX md5sum; 402 struct uuid uuid; 403 404 g_trace(G_T_TOPOLOGY, "bsd_taste(%s,%s)", mp->name, pp->name); 405 g_topology_assert(); 406 407 /* We don't implement transparent inserts. */ 408 if (flags == G_TF_TRANSPARENT) 409 return (NULL); 410 411 /* 412 * BSD labels are a subclass of the general "slicing" topology so 413 * a lot of the work can be done by the common "slice" code. 414 * Create a geom with space for MAXPARTITIONS providers, one consumer 415 * and a softc structure for us. Specify the provider to attach 416 * the consumer to and our "start" routine for special requests. 417 * The provider is opened with mode (1,0,0) so we can do reads 418 * from it. 419 */ 420 gp = g_slice_new(mp, MAXPARTITIONS, pp, &cp, &ms, 421 sizeof(*ms), g_bsd_start); 422 if (gp == NULL) 423 return (NULL); 424 425 /* Get the geom_slicer softc from the geom. */ 426 gsp = gp->softc; 427 428 /* 429 * The do...while loop here allows us to have multiple escapes 430 * using a simple "break". This improves code clarity without 431 * ending up in deep nesting and without using goto or come from. 432 */ 433 do { 434 /* 435 * If the provider is an MBR we will only auto attach 436 * to type 165 slices in the G_TF_NORMAL case. We will 437 * attach to any other type. 438 */ 439 error = g_getattr("MBR::type", cp, &i); 440 if (!error) { 441 if (i != 165 && flags == G_TF_NORMAL) 442 break; 443 error = g_getattr("MBR::offset", cp, &ms->mbroffset); 444 if (error) 445 break; 446 } 447 448 /* Same thing if we are inside a GPT */ 449 error = g_getattr("GPT::type", cp, &uuid); 450 if (!error) { 451 if (memcmp(&uuid, &freebsd_slice, sizeof(uuid)) != 0 && 452 flags == G_TF_NORMAL) 453 break; 454 } 455 456 /* Get sector size, we need it to read data. */ 457 secsize = cp->provider->sectorsize; 458 if (secsize < 512) 459 break; 460 461 /* First look for a label at the start of the second sector. */ 462 error = g_bsd_try(gp, gsp, cp, secsize, ms, secsize); 463 464 /* 465 * If sector size is not 512 the label still can be at 466 * offset 512, not at the start of the second sector. At least 467 * it's true for labels created by the FreeBSD's bsdlabel(8). 468 */ 469 if (error && secsize != HISTORIC_LABEL_OFFSET) 470 error = g_bsd_try(gp, gsp, cp, secsize, ms, 471 HISTORIC_LABEL_OFFSET); 472 473 /* Next, look for alpha labels */ 474 if (error) 475 error = g_bsd_try(gp, gsp, cp, secsize, ms, 476 ALPHA_LABEL_OFFSET); 477 478 /* If we didn't find a label, punt. */ 479 if (error) 480 break; 481 482 /* 483 * In order to avoid recursively attaching to the same 484 * on-disk label (it's usually visible through the 'c' 485 * partition) we calculate an MD5 and ask if other BSD's 486 * below us love that label. If they do, we don't. 487 */ 488 MD5Init(&md5sum); 489 MD5Update(&md5sum, ms->label, sizeof(ms->label)); 490 MD5Final(ms->labelsum, &md5sum); 491 492 error = g_getattr("BSD::labelsum", cp, &hash); 493 if (!error && !bcmp(ms->labelsum, hash, sizeof(hash))) 494 break; 495 496 /* 497 * Process the found disklabel, and modify our "slice" 498 * instance to match it, if possible. 499 */ 500 error = g_bsd_modify(gp, ms->label); 501 } while (0); 502 503 /* Success or failure, we can close our provider now. */ 504 g_access(cp, -1, 0, 0); 505 506 /* If we have configured any providers, return the new geom. */ 507 if (gsp->nprovider > 0) { 508 g_slice_conf_hot(gp, 0, ms->labeloffset, LABELSIZE, 509 G_SLICE_HOT_ALLOW, G_SLICE_HOT_DENY, G_SLICE_HOT_CALL); 510 gsp->hot = g_bsd_hotwrite; 511 if (!g_bsd_once) { 512 g_bsd_once = 1; 513 printf( 514 "WARNING: geom_bsd (geom %s) is deprecated, " 515 "use gpart instead.\n", gp->name); 516 } 517 return (gp); 518 } 519 /* 520 * ...else push the "self-destruct" button, by spoiling our own 521 * consumer. This triggers a call to g_slice_spoiled which will 522 * dismantle what was setup. 523 */ 524 g_slice_spoiled(cp); 525 return (NULL); 526} 527 528struct h0h0 { 529 struct g_geom *gp; 530 struct g_bsd_softc *ms; 531 u_char *label; 532 int error; 533}; 534 535static void 536g_bsd_callconfig(void *arg, int flag) 537{ 538 struct h0h0 *hp; 539 540 hp = arg; 541 hp->error = g_bsd_modify(hp->gp, hp->label); 542 if (!hp->error) 543 hp->error = g_bsd_writelabel(hp->gp, NULL); 544} 545 546/* 547 * NB! curthread is user process which GCTL'ed. 548 */ 549static void 550g_bsd_config(struct gctl_req *req, struct g_class *mp, char const *verb) 551{ 552 u_char *label; 553 int error; 554 struct h0h0 h0h0; 555 struct g_geom *gp; 556 struct g_slicer *gsp; 557 struct g_consumer *cp; 558 struct g_bsd_softc *ms; 559 560 g_topology_assert(); 561 gp = gctl_get_geom(req, mp, "geom"); 562 if (gp == NULL) 563 return; 564 cp = LIST_FIRST(&gp->consumer); 565 gsp = gp->softc; 566 ms = gsp->softc; 567 if (!strcmp(verb, "read mbroffset")) { 568 gctl_set_param_err(req, "mbroffset", &ms->mbroffset, 569 sizeof(ms->mbroffset)); 570 return; 571 } else if (!strcmp(verb, "write label")) { 572 label = gctl_get_paraml(req, "label", LABELSIZE); 573 if (label == NULL) 574 return; 575 h0h0.gp = gp; 576 h0h0.ms = gsp->softc; 577 h0h0.label = label; 578 h0h0.error = -1; 579 /* XXX: Does this reference register with our selfdestruct code ? */ 580 error = g_access(cp, 1, 1, 1); 581 if (error) { 582 gctl_error(req, "could not access consumer"); 583 return; 584 } 585 g_bsd_callconfig(&h0h0, 0); 586 error = h0h0.error; 587 g_access(cp, -1, -1, -1); 588 } else if (!strcmp(verb, "write bootcode")) { 589 label = gctl_get_paraml(req, "bootcode", BBSIZE); 590 if (label == NULL) 591 return; 592 /* XXX: Does this reference register with our selfdestruct code ? */ 593 error = g_access(cp, 1, 1, 1); 594 if (error) { 595 gctl_error(req, "could not access consumer"); 596 return; 597 } 598 error = g_bsd_writelabel(gp, label); 599 g_access(cp, -1, -1, -1); 600 } else { 601 gctl_error(req, "Unknown verb parameter"); 602 } 603 604 return; 605} 606 607/* Finally, register with GEOM infrastructure. */ 608static struct g_class g_bsd_class = { 609 .name = BSD_CLASS_NAME, 610 .version = G_VERSION, 611 .taste = g_bsd_taste, 612 .ctlreq = g_bsd_config, 613 .dumpconf = g_bsd_dumpconf, 614}; 615 616DECLARE_GEOM_CLASS(g_bsd_class, g_bsd); 617MODULE_VERSION(geom_bsd, 0); 618