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 1.24 2007/05/19 09:46:18 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         if (info->d_media_blksize & DEV_BMASK)
 122                 return (EIO);
 123         /* Read master boot record. */
 124         wdev = dkmodpart(dkmodslice(dev, WHOLE_DISK_SLICE), WHOLE_SLICE_PART);
 125         bp = geteblk((int)info->d_media_blksize);
 126         bp->b_bio1.bio_offset = (off_t)mbr_offset * info->d_media_blksize;
 127         bp->b_bcount = info->d_media_blksize;
 128         bp->b_cmd = BUF_CMD_READ;
 129         dev_dstrategy(wdev, &bp->b_bio1);
 130         if (biowait(bp) != 0) {
 131                 diskerr(&bp->b_bio1, wdev,
 132                         "reading primary partition table: error",
 133                         LOG_PRINTF, 0);
 134                 kprintf("\n");
 135                 error = EIO;
 136                 goto done;
 137         }
 138
 139         /* Weakly verify it. */
 140         cp = bp->b_data;
 141         sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE, WHOLE_SLICE_PART, partname);
 142         if (cp[0x1FE] != 0x55 || cp[0x1FF] != 0xAA) {
 143                 if (bootverbose)
 144                         kprintf("%s: invalid primary partition table: no magic\n",
 145                                sname);
 146                 error = EINVAL;
 147                 goto done;
 148         }
 149
 150         /* Make a copy of the partition table to avoid alignment problems. */
 151         memcpy(&dpcopy[0], cp + DOSPARTOFF, sizeof(dpcopy));
 152
 153         dp0 = &dpcopy[0];
 154
 155         /* Check for "Ontrack Diskmanager". */
 156         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
 157                 if (dp->dp_typ == DOSPTYP_ONTRACK) {
 158                         if (bootverbose)
 159                                 kprintf(
 160             "%s: Found \"Ontrack Disk Manager\" on this disk.\n", sname);
 161                         bp->b_flags |= B_INVAL | B_AGE;
 162                         brelse(bp);
 163                         mbr_offset = 63;
 164                         goto reread_mbr;
 165                 }
 166         }
 167
 168         if (bcmp(dp0, historical_bogus_partition_table,
 169                  sizeof historical_bogus_partition_table) == 0 ||
 170             bcmp(dp0, historical_bogus_partition_table_fixed,
 171                  sizeof historical_bogus_partition_table_fixed) == 0) {
 172 #if 0
 173                 TRACE(("%s: invalid primary partition table: historical\n",
 174                        sname));
 175 #endif /* 0 */
 176                 if (bootverbose)
 177                         kprintf(
 178      "%s: invalid primary partition table: Dangerously Dedicated (ignored)\n",
 179                                sname);
 180                 error = EINVAL;
 181                 goto done;
 182         }
 183
 184         /* Guess the geometry. */
 185         /*
 186          * TODO:
 187          * Perhaps skip entries with 0 size.
 188          * Perhaps only look at entries of type DOSPTYP_386BSD.
 189          */
 190         max_ncyls = 0;
 191         max_nsectors = 0;
 192         max_ntracks = 0;
 193         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
 194                 int     ncyls;
 195                 int     nsectors;
 196                 int     ntracks;
 197
 198                 ncyls = DPCYL(dp->dp_ecyl, dp->dp_esect) + 1;
 199                 if (max_ncyls < ncyls)
 200                         max_ncyls = ncyls;
 201                 nsectors = DPSECT(dp->dp_esect);
 202                 if (max_nsectors < nsectors)
 203                         max_nsectors = nsectors;
 204                 ntracks = dp->dp_ehd + 1;
 205                 if (max_ntracks < ntracks)
 206                         max_ntracks = ntracks;
 207         }
 208
 209         /*
 210          * Check that we have guessed the geometry right by checking the
 211          * partition entries.
 212          */
 213         /*
 214          * TODO:
 215          * As above.
 216          * Check for overlaps.
 217          * Check against d_secperunit if the latter is reliable.
 218          */
 219         error = 0;
 220         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++) {
 221                 if (dp->dp_scyl == 0 && dp->dp_shd == 0 && dp->dp_ssect == 0
 222                     && dp->dp_start == 0 && dp->dp_size == 0)
 223                         continue;
 224                 sname = dsname(dev, dkunit(dev), BASE_SLICE + dospart,
 225                                WHOLE_SLICE_PART, partname);
 226
 227                 /*
 228                  * Temporarily ignore errors from this check.  We could
 229                  * simplify things by accepting the table eariler if we
 230                  * always ignore errors here.  Perhaps we should always
 231                  * accept the table if the magic is right but not let
 232                  * bad entries affect the geometry.
 233                  */
 234                 check_part(sname, dp, mbr_offset, max_nsectors, max_ntracks,
 235                            mbr_offset);
 236         }
 237         if (error != 0)
 238                 goto done;
 239
 240         /*
 241          * Accept the DOS partition table.
 242          *
 243          * Adjust the disk information structure with updated CHS
 244          * conversion parameters, but only use values extracted from
 245          * the primary partition table.
 246          *
 247          * NOTE!  Regardless of our having to deal with this old cruft,
 248          * we do not screw around with the info->d_media* parameters.
 249          */
 250         secpercyl = (u_long)max_nsectors * max_ntracks;
 251         if (secpercyl != 0 && mbr_offset == DOSBBSECTOR) {
 252                 info->d_secpertrack = max_nsectors;
 253                 info->d_nheads = max_ntracks;
 254                 info->d_secpercyl = secpercyl;
 255                 info->d_ncylinders = info->d_media_blocks / secpercyl;
 256         }
 257
 258         /*
 259          * We are passed a pointer to a suitably initialized minimal
 260          * slices "struct" with no dangling pointers in it.  Replace it
 261          * by a maximal one.  This usually oversizes the "struct", but
 262          * enlarging it while searching for logical drives would be
 263          * inconvenient.
 264          */
 265         kfree(*sspp, M_DEVBUF);
 266         ssp = dsmakeslicestruct(MAX_SLICES, info);
 267         *sspp = ssp;
 268
 269         /* Initialize normal slices. */
 270         sp = &ssp->dss_slices[BASE_SLICE];
 271         for (dospart = 0, dp = dp0; dospart < NDOSPART; dospart++, dp++, sp++) {
 272                 sname = dsname(dev, dkunit(dev), BASE_SLICE + dospart,
 273                                WHOLE_SLICE_PART, partname);
 274                 (void)mbr_setslice(sname, info, sp, dp, mbr_offset);
 275         }
 276         ssp->dss_nslices = BASE_SLICE + NDOSPART;
 277
 278         /* Handle extended partitions. */
 279         sp -= NDOSPART;
 280         for (dospart = 0; dospart < NDOSPART; dospart++, sp++) {
 281                 if (sp->ds_type == DOSPTYP_EXTENDED ||
 282                     sp->ds_type == DOSPTYP_EXTENDEDX) {
 283                         mbr_extended(wdev, info, ssp,
 284                                      sp->ds_offset, sp->ds_size, sp->ds_offset,
 285                                      max_nsectors, max_ntracks, mbr_offset, 1);
 286                 }
 287         }
 288
 289         /*
 290          * mbr_extended() abuses ssp->dss_nslices for the number of slices
 291          * that would be found if there were no limit on the number of slices
 292          * in *ssp.  Cut it back now.
 293          */
 294         if (ssp->dss_nslices > MAX_SLICES)
 295                 ssp->dss_nslices = MAX_SLICES;
 296
 297 done:
 298         bp->b_flags |= B_INVAL | B_AGE;
 299         brelse(bp);
 300         if (error == EINVAL)
 301                 error = 0;
 302         return (error);
 303 }
 304
 305 static int
 306 check_part(char *sname, struct dos_partition *dp, u_int64_t offset,
 307             int nsectors, int ntracks, u_int64_t mbr_offset)
 308 {
 309         int     chs_ecyl;
 310         int     chs_esect;
 311         int     chs_scyl;
 312         int     chs_ssect;
 313         int     error;
 314         u_long  secpercyl;
 315         u_int64_t esector;
 316         u_int64_t esector1;
 317         u_int64_t ssector;
 318         u_int64_t ssector1;
 319
 320         secpercyl = (u_long)nsectors * ntracks;
 321         chs_scyl = DPCYL(dp->dp_scyl, dp->dp_ssect);
 322         chs_ssect = DPSECT(dp->dp_ssect);
 323         ssector = chs_ssect - 1 + dp->dp_shd * nsectors + chs_scyl * secpercyl
 324                   + mbr_offset;
 325         ssector1 = offset + dp->dp_start;
 326
 327         /*
 328          * If ssector1 is on a cylinder >= 1024, then ssector can't be right.
 329          * Allow the C/H/S for it to be 1023/ntracks-1/nsectors, or correct
 330          * apart from the cylinder being reduced modulo 1024.  Always allow
 331          * 1023/255/63, because this is the official way to represent
 332          * pure-LBA for the starting position.
 333          */
 334         if ((ssector < ssector1
 335              && ((chs_ssect == nsectors && dp->dp_shd == ntracks - 1
 336                   && chs_scyl == 1023)
 337                  || (secpercyl != 0
 338                      && (ssector1 - ssector) % (1024 * secpercyl) == 0)))
 339             || (dp->dp_scyl == 255 && dp->dp_shd == 255
 340                 && dp->dp_ssect == 255)) {
 341                 TRACE(("%s: C/H/S start %d/%d/%d, start %llu: allow\n",
 342                        sname, chs_scyl, dp->dp_shd, chs_ssect, ssector1));
 343                 ssector = ssector1;
 344         }
 345
 346         chs_ecyl = DPCYL(dp->dp_ecyl, dp->dp_esect);
 347         chs_esect = DPSECT(dp->dp_esect);
 348         esector = chs_esect - 1 + dp->dp_ehd * nsectors + chs_ecyl * secpercyl
 349                   + mbr_offset;
 350         esector1 = ssector1 + dp->dp_size - 1;
 351
 352         /*
 353          * Allow certain bogus C/H/S values for esector, as above. However,
 354          * heads == 255 isn't really legal and causes some BIOS crashes. The
 355          * correct value to indicate a pure-LBA end is 1023/heads-1/sectors -
 356          * usually 1023/254/63. "heads" is base 0, "sectors" is base 1.
 357          */
 358         if ((esector < esector1
 359              && ((chs_esect == nsectors && dp->dp_ehd == ntracks - 1
 360                   && chs_ecyl == 1023)
 361                  || (secpercyl != 0
 362                      && (esector1 - esector) % (1024 * secpercyl) == 0)))
 363             || (dp->dp_ecyl == 255 && dp->dp_ehd == 255
 364                 && dp->dp_esect == 255)) {
 365                 TRACE(("%s: C/H/S end %d/%d/%d, end %llu: allow\n",
 366                        sname, chs_ecyl, dp->dp_ehd, chs_esect, esector1));
 367                 esector = esector1;
 368         }
 369
 370         error = (ssector == ssector1 && esector == esector1) ? 0 : EINVAL;
 371         if (bootverbose)
 372                 kprintf("%s: type 0x%x, start %llu, end = %llu, size %lu %s\n",
 373                        sname, dp->dp_typ, ssector1, esector1,
 374                        (u_long)dp->dp_size, error ? "" : ": OK");
 375         if (ssector != ssector1 && bootverbose)
 376                 kprintf("%s: C/H/S start %d/%d/%d (%llu) != start %llu: invalid\n",
 377                        sname, chs_scyl, dp->dp_shd, chs_ssect,
 378                        ssector, ssector1);
 379         if (esector != esector1 && bootverbose)
 380                 kprintf("%s: C/H/S end %d/%d/%d (%llu) != end %llu: invalid\n",
 381                        sname, chs_ecyl, dp->dp_ehd, chs_esect,
 382                        esector, esector1);
 383         return (error);
 384 }
 385
 386 static
 387 void
 388 mbr_extended(cdev_t dev, struct disk_info *info, struct diskslices *ssp,
 389             u_int64_t ext_offset, u_int64_t ext_size, u_int64_t base_ext_offset,
 390             int nsectors, int ntracks, u_int64_t mbr_offset, int level)
 391 {
 392         struct buf *bp;
 393         u_char  *cp;
 394         int     dospart;
 395         struct dos_partition *dp;
 396         struct dos_partition dpcopy[NDOSPART];
 397         u_int64_t ext_offsets[NDOSPART];
 398         u_int64_t ext_sizes[NDOSPART];
 399         char    partname[2];
 400         int     slice;
 401         char    *sname;
 402         struct diskslice *sp;
 403
 404         if (level >= 16) {
 405                 kprintf(
 406         "%s: excessive recursion in search for slices; aborting search\n",
 407                        devtoname(dev));
 408                 return;
 409         }
 410
 411         /* Read extended boot record. */
 412         bp = geteblk((int)info->d_media_blksize);
 413         bp->b_bio1.bio_offset = (off_t)ext_offset * info->d_media_blksize;
 414         bp->b_bcount = info->d_media_blksize;
 415         bp->b_cmd = BUF_CMD_READ;
 416         dev_dstrategy(dev, &bp->b_bio1);
 417         if (biowait(bp) != 0) {
 418                 diskerr(&bp->b_bio1, dev,
 419                         "reading extended partition table: error",
 420                         LOG_PRINTF, 0);
 421                 kprintf("\n");
 422                 goto done;
 423         }
 424
 425         /* Weakly verify it. */
 426         cp = bp->b_data;
 427         if (cp[0x1FE] != 0x55 || cp[0x1FF] != 0xAA) {
 428                 sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE, WHOLE_SLICE_PART,
 429                                partname);
 430                 if (bootverbose)
 431                         kprintf("%s: invalid extended partition table: no magic\n",
 432                                sname);
 433                 goto done;
 434         }
 435
 436         /* Make a copy of the partition table to avoid alignment problems. */
 437         memcpy(&dpcopy[0], cp + DOSPARTOFF, sizeof(dpcopy));
 438
 439         slice = ssp->dss_nslices;
 440         for (dospart = 0, dp = &dpcopy[0]; dospart < NDOSPART;
 441             dospart++, dp++) {
 442                 ext_sizes[dospart] = 0;
 443                 if (dp->dp_scyl == 0 && dp->dp_shd == 0 && dp->dp_ssect == 0
 444                     && dp->dp_start == 0 && dp->dp_size == 0)
 445                         continue;
 446                 if (dp->dp_typ == DOSPTYP_EXTENDED ||
 447                     dp->dp_typ == DOSPTYP_EXTENDEDX) {
 448                         static char buf[32];
 449
 450                         sname = dsname(dev, dkunit(dev), WHOLE_DISK_SLICE,
 451                                        WHOLE_SLICE_PART, partname);
 452                         ksnprintf(buf, sizeof(buf), "%s", sname);
 453                         if (strlen(buf) < sizeof buf - 11)
 454                                 strcat(buf, "<extended>");
 455                         check_part(buf, dp, base_ext_offset, nsectors,
 456                                    ntracks, mbr_offset);
 457                         ext_offsets[dospart] = base_ext_offset + dp->dp_start;
 458                         ext_sizes[dospart] = dp->dp_size;
 459                 } else {
 460                         sname = dsname(dev, dkunit(dev), slice, WHOLE_SLICE_PART,
 461                                        partname);
 462                         check_part(sname, dp, ext_offset, nsectors, ntracks,
 463                                    mbr_offset);
 464                         if (slice >= MAX_SLICES) {
 465                                 kprintf("%s: too many slices\n", sname);
 466                                 slice++;
 467                                 continue;
 468                         }
 469                         sp = &ssp->dss_slices[slice];
 470                         if (mbr_setslice(sname, info, sp, dp, ext_offset) != 0)
 471                                 continue;
 472                         slice++;
 473                 }
 474         }
 475         ssp->dss_nslices = slice;
 476
 477         /* If we found any more slices, recursively find all the subslices. */
 478         for (dospart = 0; dospart < NDOSPART; dospart++) {
 479                 if (ext_sizes[dospart] != 0) {
 480                         mbr_extended(dev, info, ssp, ext_offsets[dospart],
 481                                      ext_sizes[dospart], base_ext_offset,
 482                                      nsectors, ntracks, mbr_offset, ++level);
 483                 }
 484         }
 485
 486 done:
 487         bp->b_flags |= B_INVAL | B_AGE;
 488         brelse(bp);
 489 }
 490
 491 static int
 492 mbr_setslice(char *sname, struct disk_info *info, struct diskslice *sp,
 493             struct dos_partition *dp, u_int64_t br_offset)
 494 {
 495         u_int64_t       offset;
 496         u_int64_t       size;
 497
 498         offset = br_offset + dp->dp_start;
 499         if (offset > info->d_media_blocks || offset < br_offset) {
 500                 kprintf(
 501                 "%s: slice starts beyond end of the disk: rejecting it\n",
 502                        sname);
 503                 return (1);
 504         }
 505         size = info->d_media_blocks - offset;
 506         if (size >= dp->dp_size)
 507                 size = dp->dp_size;
 508         else
 509                 kprintf(
 510 "%s: slice extends beyond end of disk: truncating from %lu to %llu sectors\n",
 511                        sname, (u_long)dp->dp_size, size);
 512         sp->ds_offset = offset;
 513         sp->ds_size = size;
 514         sp->ds_type = dp->dp_typ;
 515
 516         /*
 517          * The first sector in each slice is reserved for a system boot
 518          * sector.  ds_skip_bsdlabel is always inclusive of ds_skip_platform,
 519          * if they are the same then there is no label present (or yet
 520          * loaded).
 521          */
 522         sp->ds_skip_platform = 1;
 523         sp->ds_skip_bsdlabel = 1;
 524         return (0);
 525 }