sys/kern/subr_diskmbr.c

   1 /*-
   2  * Copyright (c) 1994 Bruce D. Evans.
   3  * All rights reserved.
   4  *
   5  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  *      from: @(#)ufs_disksubr.c        7.16 (Berkeley) 5/4/91
  37  *      from: ufs_disksubr.c,v 1.8 1994/06/07 01:21:39 phk Exp $
  38  * $FreeBSD: src/sys/kern/subr_diskmbr.c,v 1.45 2000/01/28 10:22:07 bde Exp $
  39  * $DragonFly: src/sys/kern/subr_diskmbr.c,v 1.22 2007/05/16 05:20:23 dillon Exp $
  40  */
  41
  42 #include <sys/param.h>
  43 #include <sys/systm.h>
  44 #include <sys/buf.h>
  45 #include <sys/conf.h>
  46 #include <sys/diskslice.h>
  47 #define DOSPTYP_EXTENDED        5
  48 #define DOSPTYP_EXTENDEDX       15
  49 #define DOSPTYP_ONTRACK         84
  50 #include <sys/diskslice.h>
  51 #include <sys/diskmbr.h>
  52 #include <sys/disk.h>
  53 #include <sys/malloc.h>
  54 #include <sys/syslog.h>
  55 #include <sys/device.h>
  56
  57 #define TRACE(str)      do { if (dsi_debug) kprintf str; } while (0)
  58
  59 static volatile u_char dsi_debug;
  60
  61 /*
  62  * This is what we have embedded in every boot1 for supporting the bogus
  63  * "Dangerously Dedicated" mode. However, the old table is broken because
  64  * it has an illegal geometry in it - it specifies 256 heads (heads = end
  65  * head + 1) which causes nasty stuff when that wraps to zero in bios code.
  66  * eg: divide by zero etc. This caused the dead-thinkpad problem, numerous
  67  * SCSI bios crashes, EFI to crash, etc.
  68  *
  69  * We still have to recognize the old table though, even though we stopped
  70  * inflicting it upon the world.
  71  */
  72 static struct dos_partition historical_bogus_partition_table[NDOSPART] = {
  73         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
  74         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
  75         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
  76         { 0x80, 0, 1, 0, DOSPTYP_386BSD, 255, 255, 255, 0, 50000, },
  77 };
  78 static struct dos_partition historical_bogus_partition_table_fixed[NDOSPART] = {
  79         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
  80         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
  81         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },
  82         { 0x80, 0, 1, 0, DOSPTYP_386BSD, 254, 255, 255, 0, 50000, },
  83 };
  84
  85 static int check_part (char *sname, struct dos_partition *dp,
  86                            u_int64_t offset, int nsectors, int ntracks,
  87                            u_int64_t mbr_offset);
  88 static void mbr_extended (cdev_t dev, struct disk_info *info,
  89                               struct diskslices *ssp, u_int64_t ext_offset,
  90                               u_int64_t ext_size, u_int64_t base_ext_offset,
  91                               int nsectors, int ntracks, u_int64_t mbr_offset,
  92                               int level);
  93 static int mbr_setslice (char *sname, struct disk_info *info,
  94                              struct diskslice *sp, struct dos_partition *dp,
  95                              u_int64_t br_offset);
  96
  97
  98 int
  99 mbrinit(cdev_t dev, struct disk_info *info, struct diskslices **sspp)
 100 {
 101         struct buf *bp;
 102         u_char  *cp;
 103         int     dospart;
 104         struct dos_partition *dp;
 105         struct dos_partition *dp0;
 106         struct dos_partition dpcopy[NDOSPART];
 107         int     error;
 108         int     max_ncyls;
 109         int     max_nsectors;
 110         int     max_ntracks;
 111         u_int64_t mbr_offset;
 112         char    partname[2];
 113         u_long  secpercyl;
 114         char    *sname;
 115         struct diskslice *sp;
 116         struct diskslices *ssp;
 117         cdev_t wdev;
 118
 119         mbr_offset = DOSBBSECTOR;
 120 reread_mbr:
 121         /* Read master boot record. */
 122         wdev = dkmodpart(dkmodslice(dev, WHOLE_DISK_SLICE), RAW_PART);
 123         bp = geteblk((int)info->d_media_blksize);
 124         bp->b_bio1.bio_offset = (off_t)mbr_offset * info->d_media_blksize;
 125         bp->b_bcount = info->d_media_blksize;
 126         bp->b_cmd = BUF_CMD_READ;
 127         dev_dstrategy(wdev, &bp->b_bio1);
 128         if (biowait(bp) != 0) {
 129                 diskerr(&bp->b_bio1, wdev,
 130                         "reading primary partition table: error",
 131                         LOG_PRINTF, 0);
 132                 kprintf("\n");
 133                 error = EIO;
 134                 goto done;
 135         }
 136
 137         /* Weakly verify it. */
 138         cp = bp->b_data;
 139         sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE, RAW_PART, partname);
 140         if (cp[0x1FE] != 0x55 || cp[0x1FF] != 0xAA) {
 141                 if (bootverbose)
 142                         kprintf("%s: invalid primary partition table: no magic\n",
 143                                sname);
 144                 error = EINVAL;
 145                 goto done;
 146         }
 147
 148         /* Make a copy of the partition table to avoid alignment problems. */
 149         memcpy(&dpcopy[0], cp + DOSPARTOFF, sizeof(dpcopy));
 150
 151         dp0 = &dpcopy[0];
 152
 153         /* Check for "Ontrack Diskmanager". */
 154         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
 155                 if (dp->dp_typ == DOSPTYP_ONTRACK) {
 156                         if (bootverbose)
 157                                 kprintf(
 158             "%s: Found \"Ontrack Disk Manager\" on this disk.\n", sname);
 159                         bp->b_flags |= B_INVAL | B_AGE;
 160                         brelse(bp);
 161                         mbr_offset = 63;
 162                         goto reread_mbr;
 163                 }
 164         }
 165
 166         if (bcmp(dp0, historical_bogus_partition_table,
 167                  sizeof historical_bogus_partition_table) == 0 ||
 168             bcmp(dp0, historical_bogus_partition_table_fixed,
 169                  sizeof historical_bogus_partition_table_fixed) == 0) {
 170 #if 0
 171                 TRACE(("%s: invalid primary partition table: historical\n",
 172                        sname));
 173 #endif /* 0 */
 174                 if (bootverbose)
 175                         kprintf(
 176      "%s: invalid primary partition table: Dangerously Dedicated (ignored)\n",
 177                                sname);
 178                 error = EINVAL;
 179                 goto done;
 180         }
 181
 182         /* Guess the geometry. */
 183         /*
 184          * TODO:
 185          * Perhaps skip entries with 0 size.
 186          * Perhaps only look at entries of type DOSPTYP_386BSD.
 187          */
 188         max_ncyls = 0;
 189         max_nsectors = 0;
 190         max_ntracks = 0;
 191         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
 192                 int     ncyls;
 193                 int     nsectors;
 194                 int     ntracks;
 195
 196                 ncyls = DPCYL(dp->dp_ecyl, dp->dp_esect) + 1;
 197                 if (max_ncyls < ncyls)
 198                         max_ncyls = ncyls;
 199                 nsectors = DPSECT(dp->dp_esect);
 200                 if (max_nsectors < nsectors)
 201                         max_nsectors = nsectors;
 202                 ntracks = dp->dp_ehd + 1;
 203                 if (max_ntracks < ntracks)
 204                         max_ntracks = ntracks;
 205         }
 206
 207         /*
 208          * Check that we have guessed the geometry right by checking the
 209          * partition entries.
 210          */
 211         /*
 212          * TODO:
 213          * As above.
 214          * Check for overlaps.
 215          * Check against d_secperunit if the latter is reliable.
 216          */
 217         error = 0;
 218         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
 219                 if (dp->dp_scyl == 0 && dp->dp_shd == 0 && dp->dp_ssect == 0
 220                     && dp->dp_start == 0 && dp->dp_size == 0)
 221                         continue;
 222                 sname = dsname(dev, dkunit(dev), BASE_SLICE + dospart,
 223                                RAW_PART, partname);
 224
 225                 /*
 226                  * Temporarily ignore errors from this check.  We could
 227                  * simplify things by accepting the table eariler if we
 228                  * always ignore errors here.  Perhaps we should always
 229                  * accept the table if the magic is right but not let
 230                  * bad entries affect the geometry.
 231                  */
 232                 check_part(sname, dp, mbr_offset, max_nsectors, max_ntracks,
 233                            mbr_offset);
 234         }
 235         if (error != 0)
 236                 goto done;
 237
 238         /*
 239          * Accept the DOS partition table.
 240          *
 241          * Adjust the disk information structure with updated CHS
 242          * conversion parameters, but only use values extracted from
 243          * the primary partition table.
 244          *
 245          * NOTE!  Regardless of our having to deal with this old cruft,
 246          * we do not screw around with the info->d_media* parameters.
 247          */
 248         secpercyl = (u_long)max_nsectors * max_ntracks;
 249         if (secpercyl != 0 && mbr_offset == DOSBBSECTOR) {
 250                 info->d_secpertrack = max_nsectors;
 251                 info->d_nheads = max_ntracks;
 252                 info->d_secpercyl = secpercyl;
 253                 info->d_ncylinders = info->d_media_blocks / secpercyl;
 254         }
 255
 256         /*
 257          * We are passed a pointer to a suitably initialized minimal
 258          * slices "struct" with no dangling pointers in it.  Replace it
 259          * by a maximal one.  This usually oversizes the "struct", but
 260          * enlarging it while searching for logical drives would be
 261          * inconvenient.
 262          */
 263         kfree(*sspp, M_DEVBUF);
 264         ssp = dsmakeslicestruct(MAX_SLICES, info);
 265         *sspp = ssp;
 266
 267         /* Initialize normal slices. */
 268         sp = &ssp->dss_slices[BASE_SLICE];
 269         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++, sp++) {
 270                 sname = dsname(dev, dkunit(dev), BASE_SLICE + dospart,
 271                                RAW_PART, partname);
 272                 (void)mbr_setslice(sname, info, sp, dp, mbr_offset);
 273         }
 274         ssp->dss_nslices = BASE_SLICE + NDOSPART;
 275
 276         /* Handle extended partitions. */
 277         sp -= NDOSPART;
 278         for (dospart = 0; dospart < NDOSPART; dospart++, sp++) {
 279                 if (sp->ds_type == DOSPTYP_EXTENDED ||
 280                     sp->ds_type == DOSPTYP_EXTENDEDX) {
 281                         mbr_extended(wdev, info, ssp,
 282                                      sp->ds_offset, sp->ds_size, sp->ds_offset,
 283                                      max_nsectors, max_ntracks, mbr_offset, 1);
 284                 }
 285         }
 286
 287         /*
 288          * mbr_extended() abuses ssp->dss_nslices for the number of slices
 289          * that would be found if there were no limit on the number of slices
 290          * in *ssp.  Cut it back now.
 291          */
 292         if (ssp->dss_nslices > MAX_SLICES)
 293                 ssp->dss_nslices = MAX_SLICES;
 294
 295 done:
 296         bp->b_flags |= B_INVAL | B_AGE;
 297         brelse(bp);
 298         if (error == EINVAL)
 299                 error = 0;
 300         return (error);
 301 }
 302
 303 static int
 304 check_part(char *sname, struct dos_partition *dp, u_int64_t offset,
 305             int nsectors, int ntracks, u_int64_t mbr_offset)
 306 {
 307         int     chs_ecyl;
 308         int     chs_esect;
 309         int     chs_scyl;
 310         int     chs_ssect;
 311         int     error;
 312         u_long  secpercyl;
 313         u_int64_t esector;
 314         u_int64_t esector1;
 315         u_int64_t ssector;
 316         u_int64_t ssector1;
 317
 318         secpercyl = (u_long)nsectors * ntracks;
 319         chs_scyl = DPCYL(dp->dp_scyl, dp->dp_ssect);
 320         chs_ssect = DPSECT(dp->dp_ssect);
 321         ssector = chs_ssect - 1 + dp->dp_shd * nsectors + chs_scyl * secpercyl
 322                   + mbr_offset;
 323         ssector1 = offset + dp->dp_start;
 324
 325         /*
 326          * If ssector1 is on a cylinder >= 1024, then ssector can't be right.
 327          * Allow the C/H/S for it to be 1023/ntracks-1/nsectors, or correct
 328          * apart from the cylinder being reduced modulo 1024.  Always allow
 329          * 1023/255/63, because this is the official way to represent
 330          * pure-LBA for the starting position.
 331          */
 332         if ((ssector < ssector1
 333              && ((chs_ssect == nsectors && dp->dp_shd == ntracks - 1
 334                   && chs_scyl == 1023)
 335                  || (secpercyl != 0
 336                      && (ssector1 - ssector) % (1024 * secpercyl) == 0)))
 337             || (dp->dp_scyl == 255 && dp->dp_shd == 255
 338                 && dp->dp_ssect == 255)) {
 339                 TRACE(("%s: C/H/S start %d/%d/%d, start %llu: allow\n",
 340                        sname, chs_scyl, dp->dp_shd, chs_ssect, ssector1));
 341                 ssector = ssector1;
 342         }
 343
 344         chs_ecyl = DPCYL(dp->dp_ecyl, dp->dp_esect);
 345         chs_esect = DPSECT(dp->dp_esect);
 346         esector = chs_esect - 1 + dp->dp_ehd * nsectors + chs_ecyl * secpercyl
 347                   + mbr_offset;
 348         esector1 = ssector1 + dp->dp_size - 1;
 349
 350         /*
 351          * Allow certain bogus C/H/S values for esector, as above. However,
 352          * heads == 255 isn't really legal and causes some BIOS crashes. The
 353          * correct value to indicate a pure-LBA end is 1023/heads-1/sectors -
 354          * usually 1023/254/63. "heads" is base 0, "sectors" is base 1.
 355          */
 356         if ((esector < esector1
 357              && ((chs_esect == nsectors && dp->dp_ehd == ntracks - 1
 358                   && chs_ecyl == 1023)
 359                  || (secpercyl != 0
 360                      && (esector1 - esector) % (1024 * secpercyl) == 0)))
 361             || (dp->dp_ecyl == 255 && dp->dp_ehd == 255
 362                 && dp->dp_esect == 255)) {
 363                 TRACE(("%s: C/H/S end %d/%d/%d, end %llu: allow\n",
 364                        sname, chs_ecyl, dp->dp_ehd, chs_esect, esector1));
 365                 esector = esector1;
 366         }
 367
 368         error = (ssector == ssector1 && esector == esector1) ? 0 : EINVAL;
 369         if (bootverbose)
 370                 kprintf("%s: type 0x%x, start %llu, end = %llu, size %lu %s\n",
 371                        sname, dp->dp_typ, ssector1, esector1,
 372                        (u_long)dp->dp_size, error ? "" : ": OK");
 373         if (ssector != ssector1 && bootverbose)
 374                 kprintf("%s: C/H/S start %d/%d/%d (%llu) != start %llu: invalid\n",
 375                        sname, chs_scyl, dp->dp_shd, chs_ssect,
 376                        ssector, ssector1);
 377         if (esector != esector1 && bootverbose)
 378                 kprintf("%s: C/H/S end %d/%d/%d (%llu) != end %llu: invalid\n",
 379                        sname, chs_ecyl, dp->dp_ehd, chs_esect,
 380                        esector, esector1);
 381         return (error);
 382 }
 383
 384 static
 385 void
 386 mbr_extended(cdev_t dev, struct disk_info *info, struct diskslices *ssp,
 387             u_int64_t ext_offset, u_int64_t ext_size, u_int64_t base_ext_offset,
 388             int nsectors, int ntracks, u_int64_t mbr_offset, int level)
 389 {
 390         struct buf *bp;
 391         u_char  *cp;
 392         int     dospart;
 393         struct dos_partition *dp;
 394         struct dos_partition dpcopy[NDOSPART];
 395         u_int64_t ext_offsets[NDOSPART];
 396         u_int64_t ext_sizes[NDOSPART];
 397         char    partname[2];
 398         int     slice;
 399         char    *sname;
 400         struct diskslice *sp;
 401
 402         if (level >= 16) {
 403                 kprintf(
 404         "%s: excessive recursion in search for slices; aborting search\n",
 405                        devtoname(dev));
 406                 return;
 407         }
 408
 409         /* Read extended boot record. */
 410         bp = geteblk((int)info->d_media_blksize);
 411         bp->b_bio1.bio_offset = (off_t)ext_offset * info->d_media_blksize;
 412         bp->b_bcount = info->d_media_blksize;
 413         bp->b_cmd = BUF_CMD_READ;
 414         dev_dstrategy(dev, &bp->b_bio1);
 415         if (biowait(bp) != 0) {
 416                 diskerr(&bp->b_bio1, dev,
 417                         "reading extended partition table: error",
 418                         LOG_PRINTF, 0);
 419                 kprintf("\n");
 420                 goto done;
 421         }
 422
 423         /* Weakly verify it. */
 424         cp = bp->b_data;
 425         if (cp[0x1FE] != 0x55 || cp[0x1FF] != 0xAA) {
 426                 sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE, RAW_PART,
 427                                partname);
 428                 if (bootverbose)
 429                         kprintf("%s: invalid extended partition table: no magic\n",
 430                                sname);
 431                 goto done;
 432         }
 433
 434         /* Make a copy of the partition table to avoid alignment problems. */
 435         memcpy(&dpcopy[0], cp + DOSPARTOFF, sizeof(dpcopy));
 436
 437         slice = ssp->dss_nslices;
 438         for (dospart = 0, dp = &dpcopy[0]; dospart < NDOSPART;
 439             dospart++, dp++) {
 440                 ext_sizes[dospart] = 0;
 441                 if (dp->dp_scyl == 0 && dp->dp_shd == 0 && dp->dp_ssect == 0
 442                     && dp->dp_start == 0 && dp->dp_size == 0)
 443                         continue;
 444                 if (dp->dp_typ == DOSPTYP_EXTENDED ||
 445                     dp->dp_typ == DOSPTYP_EXTENDEDX) {
 446                         static char buf[32];
 447
 448                         sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE,
 449                                        RAW_PART, partname);
 450                         ksnprintf(buf, sizeof(buf), "%s", sname);
 451                         if (strlen(buf) < sizeof buf - 11)
 452                                 strcat(buf, "<extended>");
 453                         check_part(buf, dp, base_ext_offset, nsectors,
 454                                    ntracks, mbr_offset);
 455                         ext_offsets[dospart] = base_ext_offset + dp->dp_start;
 456                         ext_sizes[dospart] = dp->dp_size;
 457                 } else {
 458                         sname = dsname(dev, dkunit(dev), slice, RAW_PART,
 459                                        partname);
 460                         check_part(sname, dp, ext_offset, nsectors, ntracks,
 461                                    mbr_offset);
 462                         if (slice >= MAX_SLICES) {
 463                                 kprintf("%s: too many slices\n", sname);
 464                                 slice++;
 465                                 continue;
 466                         }
 467                         sp = &ssp->dss_slices[slice];
 468                         if (mbr_setslice(sname, info, sp, dp, ext_offset) != 0)
 469                                 continue;
 470                         slice++;
 471                 }
 472         }
 473         ssp->dss_nslices = slice;
 474
 475         /* If we found any more slices, recursively find all the subslices. */
 476         for (dospart = 0; dospart < NDOSPART; dospart++) {
 477                 if (ext_sizes[dospart] != 0) {
 478                         mbr_extended(dev, info, ssp, ext_offsets[dospart],
 479                                      ext_sizes[dospart], base_ext_offset,
 480                                      nsectors, ntracks, mbr_offset, ++level);
 481                 }
 482         }
 483
 484 done:
 485         bp->b_flags |= B_INVAL | B_AGE;
 486         brelse(bp);
 487 }
 488
 489 static int
 490 mbr_setslice(char *sname, struct disk_info *info, struct diskslice *sp,
 491             struct dos_partition *dp, u_int64_t br_offset)
 492 {
 493         u_int64_t       offset;
 494         u_int64_t       size;
 495
 496         offset = br_offset + dp->dp_start;
 497         if (offset > info->d_media_blocks || offset < br_offset) {
 498                 kprintf(
 499                 "%s: slice starts beyond end of the disk: rejecting it\n",
 500                        sname);
 501                 return (1);
 502         }
 503         size = info->d_media_blocks - offset;
 504         if (size >= dp->dp_size)
 505                 size = dp->dp_size;
 506         else
 507                 kprintf(
 508 "%s: slice extends beyond end of disk: truncating from %lu to %llu sectors\n",
 509                        sname, (u_long)dp->dp_size, size);
 510         sp->ds_offset = offset;
 511         sp->ds_size = size;
 512         sp->ds_type = dp->dp_typ;
 513
 514         /*
 515          * The first sector in each slice is reserved for a system boot
 516          * sector.
 517          */
 518         sp->ds_skip_platform = 1;
 519         sp->ds_skip_bsdlabel = 0;
 520         return (0);
 521 }