geom_bsd.c revision 143719
1/*- 2 * Copyright (c) 2002 Poul-Henning Kamp 3 * Copyright (c) 2002 Networks Associates Technology, Inc. 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp 7 * and NAI Labs, the Security Research Division of Network Associates, Inc. 8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 9 * DARPA CHATS research program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The names of the authors may not be used to endorse or promote 20 * products derived from this software without specific prior written 21 * permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36/* 37 * This is the method for dealing with BSD disklabels. It has been 38 * extensively (by my standards at least) commented, in the vain hope that 39 * it will serve as the source in future copy&paste operations. 40 */ 41 42#include <sys/cdefs.h> 43__FBSDID("$FreeBSD: head/sys/geom/geom_bsd.c 143719 2005-03-16 20:48:13Z pjd $"); 44 45#include <sys/param.h> 46#include <sys/endian.h> 47#include <sys/systm.h> 48#include <sys/kernel.h> 49#include <sys/fcntl.h> 50#include <sys/conf.h> 51#include <sys/bio.h> 52#include <sys/malloc.h> 53#include <sys/lock.h> 54#include <sys/mutex.h> 55#include <sys/md5.h> 56#include <sys/errno.h> 57#include <sys/disklabel.h> 58#include <geom/geom.h> 59#include <geom/geom_slice.h> 60 61#define BSD_CLASS_NAME "BSD" 62 63#define ALPHA_LABEL_OFFSET 64 64 65#define LABELSIZE (148 + 16 * MAXPARTITIONS) 66 67static void g_bsd_hotwrite(void *arg, int flag); 68/* 69 * Our private data about one instance. All the rest is handled by the 70 * slice code and stored in its softc, so this is just the stuff 71 * specific to BSD disklabels. 72 */ 73struct g_bsd_softc { 74 off_t labeloffset; 75 off_t mbroffset; 76 off_t rawoffset; 77 struct disklabel ondisk; 78 u_char label[LABELSIZE]; 79 u_char labelsum[16]; 80}; 81 82/* 83 * Modify our slicer to match proposed disklabel, if possible. 84 * This is where we make sure we don't do something stupid. 85 */ 86static int 87g_bsd_modify(struct g_geom *gp, u_char *label) 88{ 89 int i, error; 90 struct partition *ppp; 91 struct g_slicer *gsp; 92 struct g_consumer *cp; 93 struct g_bsd_softc *ms; 94 u_int secsize, u; 95 off_t rawoffset, o; 96 struct disklabel dl; 97 MD5_CTX md5sum; 98 99 g_topology_assert(); 100 gsp = gp->softc; 101 ms = gsp->softc; 102 103 error = bsd_disklabel_le_dec(label, &dl, MAXPARTITIONS); 104 if (error) { 105 return (error); 106 } 107 108 /* Get dimensions of our device. */ 109 cp = LIST_FIRST(&gp->consumer); 110 secsize = cp->provider->sectorsize; 111 112 /* ... or a smaller sector size. */ 113 if (dl.d_secsize < secsize) { 114 return (EINVAL); 115 } 116 117 /* ... or a non-multiple sector size. */ 118 if (dl.d_secsize % secsize != 0) { 119 return (EINVAL); 120 } 121 122 /* Historical braindamage... */ 123 rawoffset = (off_t)dl.d_partitions[RAW_PART].p_offset * dl.d_secsize; 124 125 for (i = 0; i < dl.d_npartitions; i++) { 126 ppp = &dl.d_partitions[i]; 127 if (ppp->p_size == 0) 128 continue; 129 o = (off_t)ppp->p_offset * dl.d_secsize; 130 131 if (o < rawoffset) 132 rawoffset = 0; 133 } 134 135 if (rawoffset != 0 && (off_t)rawoffset != ms->mbroffset) 136 printf("WARNING: Expected rawoffset %jd, found %jd\n", 137 (intmax_t)ms->mbroffset/dl.d_secsize, 138 (intmax_t)rawoffset/dl.d_secsize); 139 140 /* Don't munge open partitions. */ 141 for (i = 0; i < dl.d_npartitions; i++) { 142 ppp = &dl.d_partitions[i]; 143 144 o = (off_t)ppp->p_offset * dl.d_secsize; 145 if (o == 0) 146 o = rawoffset; 147 error = g_slice_config(gp, i, G_SLICE_CONFIG_CHECK, 148 o - rawoffset, 149 (off_t)ppp->p_size * dl.d_secsize, 150 dl.d_secsize, 151 "%s%c", gp->name, 'a' + i); 152 if (error) 153 return (error); 154 } 155 156 /* Look good, go for it... */ 157 for (u = 0; u < gsp->nslice; u++) { 158 ppp = &dl.d_partitions[u]; 159 o = (off_t)ppp->p_offset * dl.d_secsize; 160 if (o == 0) 161 o = rawoffset; 162 g_slice_config(gp, u, G_SLICE_CONFIG_SET, 163 o - rawoffset, 164 (off_t)ppp->p_size * dl.d_secsize, 165 dl.d_secsize, 166 "%s%c", gp->name, 'a' + u); 167 } 168 169 /* Update our softc */ 170 ms->ondisk = dl; 171 if (label != ms->label) 172 bcopy(label, ms->label, LABELSIZE); 173 ms->rawoffset = rawoffset; 174 175 /* 176 * In order to avoid recursively attaching to the same 177 * on-disk label (it's usually visible through the 'c' 178 * partition) we calculate an MD5 and ask if other BSD's 179 * below us love that label. If they do, we don't. 180 */ 181 MD5Init(&md5sum); 182 MD5Update(&md5sum, ms->label, sizeof(ms->label)); 183 MD5Final(ms->labelsum, &md5sum); 184 185 return (0); 186} 187 188/* 189 * This is an internal helper function, called multiple times from the taste 190 * function to try to locate a disklabel on the disk. More civilized formats 191 * will not need this, as there is only one possible place on disk to look 192 * for the magic spot. 193 */ 194 195static int 196g_bsd_try(struct g_geom *gp, struct g_slicer *gsp, struct g_consumer *cp, int secsize, struct g_bsd_softc *ms, off_t offset) 197{ 198 int error; 199 u_char *buf; 200 struct disklabel *dl; 201 off_t secoff; 202 203 /* 204 * We need to read entire aligned sectors, and we assume that the 205 * disklabel does not span sectors, so one sector is enough. 206 */ 207 error = 0; 208 secoff = offset % secsize; 209 buf = g_read_data(cp, offset - secoff, secsize, &error); 210 if (buf == NULL || error != 0) 211 return (ENOENT); 212 213 /* Decode into our native format. */ 214 dl = &ms->ondisk; 215 error = bsd_disklabel_le_dec(buf + secoff, dl, MAXPARTITIONS); 216 if (!error) 217 bcopy(buf + secoff, ms->label, LABELSIZE); 218 219 /* Remember to free the buffer g_read_data() gave us. */ 220 g_free(buf); 221 222 ms->labeloffset = offset; 223 return (error); 224} 225 226/* 227 * This function writes the current label to disk, possibly updating 228 * the alpha SRM checksum. 229 */ 230 231static int 232g_bsd_writelabel(struct g_geom *gp, u_char *bootcode) 233{ 234 off_t secoff; 235 u_int secsize; 236 struct g_consumer *cp; 237 struct g_slicer *gsp; 238 struct g_bsd_softc *ms; 239 u_char *buf; 240 uint64_t sum; 241 int error, i; 242 243 gsp = gp->softc; 244 ms = gsp->softc; 245 cp = LIST_FIRST(&gp->consumer); 246 /* Get sector size, we need it to read data. */ 247 secsize = cp->provider->sectorsize; 248 secoff = ms->labeloffset % secsize; 249 if (bootcode == NULL) { 250 buf = g_read_data(cp, ms->labeloffset - secoff, secsize, &error); 251 if (buf == NULL || error != 0) 252 return (error); 253 bcopy(ms->label, buf + secoff, sizeof(ms->label)); 254 } else { 255 buf = bootcode; 256 bcopy(ms->label, buf + ms->labeloffset, sizeof(ms->label)); 257 } 258 if (ms->labeloffset == ALPHA_LABEL_OFFSET) { 259 sum = 0; 260 for (i = 0; i < 63; i++) 261 sum += le64dec(buf + i * 8); 262 le64enc(buf + 504, sum); 263 } 264 if (bootcode == NULL) { 265 error = g_write_data(cp, ms->labeloffset - secoff, buf, secsize); 266 g_free(buf); 267 } else { 268 error = g_write_data(cp, 0, bootcode, BBSIZE); 269 } 270 return(error); 271} 272 273/* 274 * If the user tries to overwrite our disklabel through an open partition 275 * or via a magicwrite config call, we end up here and try to prevent 276 * footshooting as best we can. 277 */ 278static void 279g_bsd_hotwrite(void *arg, int flag) 280{ 281 struct bio *bp; 282 struct g_geom *gp; 283 struct g_slicer *gsp; 284 struct g_slice *gsl; 285 struct g_bsd_softc *ms; 286 u_char *p; 287 int error; 288 289 g_topology_assert(); 290 /* 291 * We should never get canceled, because that would amount to a removal 292 * of the geom while there was outstanding I/O requests. 293 */ 294 KASSERT(flag != EV_CANCEL, ("g_bsd_hotwrite cancelled")); 295 bp = arg; 296 gp = bp->bio_to->geom; 297 gsp = gp->softc; 298 ms = gsp->softc; 299 gsl = &gsp->slices[bp->bio_to->index]; 300 p = (u_char*)bp->bio_data + ms->labeloffset 301 - (bp->bio_offset + gsl->offset); 302 error = g_bsd_modify(gp, p); 303 if (error) { 304 g_io_deliver(bp, EPERM); 305 return; 306 } 307 g_slice_finish_hot(bp); 308} 309 310/*- 311 * This start routine is only called for non-trivial requests, all the 312 * trivial ones are handled autonomously by the slice code. 313 * For requests we handle here, we must call the g_io_deliver() on the 314 * bio, and return non-zero to indicate to the slice code that we did so. 315 * This code executes in the "DOWN" I/O path, this means: 316 * * No sleeping. 317 * * Don't grab the topology lock. 318 * * Don't call biowait, g_getattr(), g_setattr() or g_read_data() 319 */ 320static int 321g_bsd_ioctl(struct g_provider *pp, u_long cmd, void *data, int fflag, struct thread *td) 322{ 323 struct g_geom *gp; 324 struct g_bsd_softc *ms; 325 struct g_slicer *gsp; 326 u_char *label; 327 int error; 328 329 gp = pp->geom; 330 gsp = gp->softc; 331 ms = gsp->softc; 332 333 switch(cmd) { 334 case DIOCGDINFO: 335 /* Return a copy of the disklabel to userland. */ 336 bsd_disklabel_le_dec(ms->label, data, MAXPARTITIONS); 337 return(0); 338 case DIOCBSDBB: { 339 struct g_consumer *cp; 340 u_char *buf; 341 void *p; 342 int error, i; 343 uint64_t sum; 344 345 if (!(fflag & FWRITE)) 346 return (EPERM); 347 /* The disklabel to set is the ioctl argument. */ 348 buf = g_malloc(BBSIZE, M_WAITOK); 349 p = *(void **)data; 350 error = copyin(p, buf, BBSIZE); 351 if (!error) { 352 /* XXX: Rude, but supposedly safe */ 353 DROP_GIANT(); 354 g_topology_lock(); 355 /* Validate and modify our slice instance to match. */ 356 error = g_bsd_modify(gp, buf + ms->labeloffset); 357 if (!error) { 358 cp = LIST_FIRST(&gp->consumer); 359 if (ms->labeloffset == ALPHA_LABEL_OFFSET) { 360 sum = 0; 361 for (i = 0; i < 63; i++) 362 sum += le64dec(buf + i * 8); 363 le64enc(buf + 504, sum); 364 } 365 error = g_write_data(cp, 0, buf, BBSIZE); 366 } 367 g_topology_unlock(); 368 PICKUP_GIANT(); 369 } 370 g_free(buf); 371 return (error); 372 } 373 case DIOCSDINFO: 374 case DIOCWDINFO: { 375 if (!(fflag & FWRITE)) 376 return (EPERM); 377 label = g_malloc(LABELSIZE, M_WAITOK); 378 /* The disklabel to set is the ioctl argument. */ 379 bsd_disklabel_le_enc(label, data); 380 381 DROP_GIANT(); 382 g_topology_lock(); 383 /* Validate and modify our slice instance to match. */ 384 error = g_bsd_modify(gp, label); 385 if (error == 0 && cmd == DIOCWDINFO) 386 error = g_bsd_writelabel(gp, NULL); 387 g_topology_unlock(); 388 PICKUP_GIANT(); 389 g_free(label); 390 return(error); 391 } 392 default: 393 return (ENOIOCTL); 394 } 395} 396 397static int 398g_bsd_start(struct bio *bp) 399{ 400 struct g_geom *gp; 401 struct g_bsd_softc *ms; 402 struct g_slicer *gsp; 403 404 gp = bp->bio_to->geom; 405 gsp = gp->softc; 406 ms = gsp->softc; 407 if (bp->bio_cmd == BIO_GETATTR) { 408 if (g_handleattr(bp, "BSD::labelsum", ms->labelsum, 409 sizeof(ms->labelsum))) 410 return (1); 411 } 412 return (0); 413} 414 415/* 416 * Dump configuration information in XML format. 417 * Notice that the function is called once for the geom and once for each 418 * consumer and provider. We let g_slice_dumpconf() do most of the work. 419 */ 420static void 421g_bsd_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp) 422{ 423 struct g_bsd_softc *ms; 424 struct g_slicer *gsp; 425 426 gsp = gp->softc; 427 ms = gsp->softc; 428 g_slice_dumpconf(sb, indent, gp, cp, pp); 429 if (indent != NULL && pp == NULL && cp == NULL) { 430 sbuf_printf(sb, "%s<labeloffset>%jd</labeloffset>\n", 431 indent, (intmax_t)ms->labeloffset); 432 sbuf_printf(sb, "%s<rawoffset>%jd</rawoffset>\n", 433 indent, (intmax_t)ms->rawoffset); 434 sbuf_printf(sb, "%s<mbroffset>%jd</mbroffset>\n", 435 indent, (intmax_t)ms->mbroffset); 436 } else if (pp != NULL) { 437 if (indent == NULL) 438 sbuf_printf(sb, " ty %d", 439 ms->ondisk.d_partitions[pp->index].p_fstype); 440 else 441 sbuf_printf(sb, "%s<type>%d</type>\n", indent, 442 ms->ondisk.d_partitions[pp->index].p_fstype); 443 } 444} 445 446/* 447 * The taste function is called from the event-handler, with the topology 448 * lock already held and a provider to examine. The flags are unused. 449 * 450 * If flags == G_TF_NORMAL, the idea is to take a bite of the provider and 451 * if we find valid, consistent magic on it, build a geom on it. 452 * any magic bits which indicate that we should automatically put a BSD 453 * geom on it. 454 * 455 * There may be cases where the operator would like to put a BSD-geom on 456 * providers which do not meet all of the requirements. This can be done 457 * by instead passing the G_TF_INSIST flag, which will override these 458 * checks. 459 * 460 * The final flags value is G_TF_TRANSPARENT, which instructs the method 461 * to put a geom on top of the provider and configure it to be as transparent 462 * as possible. This is not really relevant to the BSD method and therefore 463 * not implemented here. 464 */ 465 466static struct g_geom * 467g_bsd_taste(struct g_class *mp, struct g_provider *pp, int flags) 468{ 469 struct g_geom *gp; 470 struct g_consumer *cp; 471 int error, i; 472 struct g_bsd_softc *ms; 473 u_int secsize; 474 struct g_slicer *gsp; 475 u_char hash[16]; 476 MD5_CTX md5sum; 477 478 g_trace(G_T_TOPOLOGY, "bsd_taste(%s,%s)", mp->name, pp->name); 479 g_topology_assert(); 480 481 /* We don't implement transparent inserts. */ 482 if (flags == G_TF_TRANSPARENT) 483 return (NULL); 484 485 /* 486 * BSD labels are a subclass of the general "slicing" topology so 487 * a lot of the work can be done by the common "slice" code. 488 * Create a geom with space for MAXPARTITIONS providers, one consumer 489 * and a softc structure for us. Specify the provider to attach 490 * the consumer to and our "start" routine for special requests. 491 * The provider is opened with mode (1,0,0) so we can do reads 492 * from it. 493 */ 494 gp = g_slice_new(mp, MAXPARTITIONS, pp, &cp, &ms, 495 sizeof(*ms), g_bsd_start); 496 if (gp == NULL) 497 return (NULL); 498 499 /* Get the geom_slicer softc from the geom. */ 500 gsp = gp->softc; 501 502 /* 503 * The do...while loop here allows us to have multiple escapes 504 * using a simple "break". This improves code clarity without 505 * ending up in deep nesting and without using goto or come from. 506 */ 507 do { 508 /* 509 * If the provider is an MBR we will only auto attach 510 * to type 165 slices in the G_TF_NORMAL case. We will 511 * attach to any other type. 512 */ 513 error = g_getattr("MBR::type", cp, &i); 514 if (!error) { 515 if (i != 165 && flags == G_TF_NORMAL) 516 break; 517 error = g_getattr("MBR::offset", cp, &ms->mbroffset); 518 if (error) 519 break; 520 } 521 522 /* Same thing if we are inside a PC98 */ 523 error = g_getattr("PC98::type", cp, &i); 524 if (!error) { 525 if (i != 0xc494 && flags == G_TF_NORMAL) 526 break; 527 error = g_getattr("PC98::offset", cp, &ms->mbroffset); 528 if (error) 529 break; 530 } 531 532 /* Get sector size, we need it to read data. */ 533 secsize = cp->provider->sectorsize; 534 if (secsize < 512) 535 break; 536 537 /* First look for a label at the start of the second sector. */ 538 error = g_bsd_try(gp, gsp, cp, secsize, ms, secsize); 539 540 /* Next, look for alpha labels */ 541 if (error) 542 error = g_bsd_try(gp, gsp, cp, secsize, ms, 543 ALPHA_LABEL_OFFSET); 544 545 /* If we didn't find a label, punt. */ 546 if (error) 547 break; 548 549 /* 550 * In order to avoid recursively attaching to the same 551 * on-disk label (it's usually visible through the 'c' 552 * partition) we calculate an MD5 and ask if other BSD's 553 * below us love that label. If they do, we don't. 554 */ 555 MD5Init(&md5sum); 556 MD5Update(&md5sum, ms->label, sizeof(ms->label)); 557 MD5Final(ms->labelsum, &md5sum); 558 559 error = g_getattr("BSD::labelsum", cp, &hash); 560 if (!error && !bcmp(ms->labelsum, hash, sizeof(hash))) 561 break; 562 563 /* 564 * Process the found disklabel, and modify our "slice" 565 * instance to match it, if possible. 566 */ 567 error = g_bsd_modify(gp, ms->label); 568 } while (0); 569 570 /* Success or failure, we can close our provider now. */ 571 g_access(cp, -1, 0, 0); 572 573 /* If we have configured any providers, return the new geom. */ 574 if (gsp->nprovider > 0) { 575 g_slice_conf_hot(gp, 0, ms->labeloffset, LABELSIZE, 576 G_SLICE_HOT_ALLOW, G_SLICE_HOT_DENY, G_SLICE_HOT_CALL); 577 gsp->hot = g_bsd_hotwrite; 578 return (gp); 579 } 580 /* 581 * ...else push the "self-destruct" button, by spoiling our own 582 * consumer. This triggers a call to g_slice_spoiled which will 583 * dismantle what was setup. 584 */ 585 g_slice_spoiled(cp); 586 return (NULL); 587} 588 589struct h0h0 { 590 struct g_geom *gp; 591 struct g_bsd_softc *ms; 592 u_char *label; 593 int error; 594}; 595 596static void 597g_bsd_callconfig(void *arg, int flag) 598{ 599 struct h0h0 *hp; 600 601 hp = arg; 602 hp->error = g_bsd_modify(hp->gp, hp->label); 603 if (!hp->error) 604 hp->error = g_bsd_writelabel(hp->gp, NULL); 605} 606 607/* 608 * NB! curthread is user process which GCTL'ed. 609 */ 610static void 611g_bsd_config(struct gctl_req *req, struct g_class *mp, char const *verb) 612{ 613 u_char *label; 614 int error; 615 struct h0h0 h0h0; 616 struct g_geom *gp; 617 struct g_slicer *gsp; 618 struct g_consumer *cp; 619 struct g_bsd_softc *ms; 620 621 g_topology_assert(); 622 gp = gctl_get_geom(req, mp, "geom"); 623 if (gp == NULL) 624 return; 625 cp = LIST_FIRST(&gp->consumer); 626 gsp = gp->softc; 627 ms = gsp->softc; 628 if (!strcmp(verb, "read mbroffset")) { 629 gctl_set_param(req, "mbroffset", 630 &ms->mbroffset, sizeof(ms->mbroffset)); 631 return; 632 } else if (!strcmp(verb, "write label")) { 633 label = gctl_get_paraml(req, "label", LABELSIZE); 634 if (label == NULL) 635 return; 636 h0h0.gp = gp; 637 h0h0.ms = gsp->softc; 638 h0h0.label = label; 639 h0h0.error = -1; 640 /* XXX: Does this reference register with our selfdestruct code ? */ 641 error = g_access(cp, 1, 1, 1); 642 if (error) { 643 gctl_error(req, "could not access consumer"); 644 return; 645 } 646 g_bsd_callconfig(&h0h0, 0); 647 error = h0h0.error; 648 g_access(cp, -1, -1, -1); 649 } else if (!strcmp(verb, "write bootcode")) { 650 label = gctl_get_paraml(req, "bootcode", BBSIZE); 651 if (label == NULL) 652 return; 653 /* XXX: Does this reference register with our selfdestruct code ? */ 654 error = g_access(cp, 1, 1, 1); 655 if (error) { 656 gctl_error(req, "could not access consumer"); 657 return; 658 } 659 error = g_bsd_writelabel(gp, label); 660 g_access(cp, -1, -1, -1); 661 } else { 662 gctl_error(req, "Unknown verb parameter"); 663 } 664 665 return; 666} 667 668/* Finally, register with GEOM infrastructure. */ 669static struct g_class g_bsd_class = { 670 .name = BSD_CLASS_NAME, 671 .version = G_VERSION, 672 .taste = g_bsd_taste, 673 .ctlreq = g_bsd_config, 674 .dumpconf = g_bsd_dumpconf, 675 .ioctl = g_bsd_ioctl, 676}; 677 678DECLARE_GEOM_CLASS(g_bsd_class, g_bsd); 679