Re: bin/37656: fdisk segfault on invalid MBR entries on non-x86 geometry detection

[jnemeth%victoria.tc.ca@localhost (John Nemeth)]

The following reply was made to PR bin/37656; it has been noted by GNATS.

From: jnemeth%victoria.tc.ca@localhost (John Nemeth)
To: Eric Auge <eau%phear.org@localhost>, gnats-bugs%NetBSD.org@localhost
Cc: netbsd-bugs%NetBSD.org@localhost
Subject: Re: bin/37656: fdisk segfault on invalid MBR entries on non-x86 
geometry detection
Date: Wed, 2 Jan 2008 02:13:59 -0800

 On May 24,  8:17pm, Eric Auge wrote:
 } 
 } John Nemeth wrote:
 } > The following reply was made to PR bin/37656; it has been noted by GNATS.
 } >
 } > From: jnemeth%victoria.tc.ca@localhost (John Nemeth)
 } > Date: Tue, 1 Jan 2008 12:35:29 -0800
 } >
 } >  On Apr 18,  4:27am, eau%phear.org@localhost wrote:
 } >  }
 } >  } >Number:         37656
 } >  } >Synopsis:       fdisk segfault on invalid MBR entries on non-x86 
geometry detection
 } >  } >State:          open
 } >  } >Class:          sw-bug
 } >  } >Arrival-Date:   Tue Jan 01 16:20:00 +0000 2008
 } >  } >Originator:     Eric Auge
 } >  } >Release:        netbsd 4.0
 } >  } >Environment:
 } >  } NetBSD client 4.0_STABLE NetBSD 4.0_STABLE (GENERIC) #0: Fri Dec 28
 } >  } 13:23:05 CET 2007
 } >  } 
root%bearcave.no.phear.org@localhost:/usr/obj/sys/arch/cobalt/compile/GENERIC 
cobalt
 } >  } >Description:
 } >  } On a cobalt, after few install tests, I got the following
 } >  }
 } >  } # fdisk -S /dev/wd0
 } >  } [1]   Trace/BPT trap          fdisk -S /dev/wd0
 } >  }
 } >  } I was unable to partition the hard drive in order to have a correct
 } partitioning scheme for installing my cobalt.
 } >  }
 } >  } fdisk seems to make an almost correct sanity check on
 } >  } get_mapping function except in some cases where head returned in the
 }  MBR is 0 on 2 consecutives mappings while these values are trusted to make 
division, fdisk crash with a divide by zero problem.
 } >  }
 } >  } initial sanity check was the following line 1525 in fdisk.c:
 } >  } [...]
 } >  }     /* Sanity check the data against all zeroes */
 } >  }     if ((*cylinder == 0) && (*sector == 0) && (*head == 0))
 } >  }         return -1;
 } >  } [...]
 } >  } >How-To-Repeat:
 } >  } have a fucked up MBR on your hardrive and try to install port-cobalt
 } >  } >Fix:
 } >  } --- fdisk.c     2008-01-01 11:18:07.000000000 +0100
 } >  } +++ fdisk-new.c 2008-01-01 11:18:46.000000000 +0100
 } >  } @@ -1541,8 +1541,8 @@
 } >  }                 *absolute =3D le32toh(part->mbrp_start)
 } >  }                     + le32toh(part->mbrp_size) - 1;
 } >  }         }
 } >  } -       /* Sanity check the data against all zeroes */
 } >  } -       if ((*cylinder =3D=3D 0) && (*sector =3D=3D 0) && (*head == 0))
 } >  } +       /* Sanity check the data against cyl/sec/head zeroes */
 } >  } +       if ((*cylinder =3D=3D 0) || (*sector =3D=3D 0) || (*head == 0))
 } >  }                 return -1;
 } >  }         /* Sanity check the data against max values */
 } >  }         if ((((*cylinder * MAXHEAD) + *head) * MAXSECTOR + *sector) < 
*absolute)
 } >
 } >       This change is wrong.  It is perfectly legal for cylinder and head
 } >  to be 0.  This change would exclude any partition that starts on
 } >  cylinder 0 from consideration, which is where the first partition
 } >  normally starts.  It would also exclude any partition that starts with
 } >  head 0 which is where primary partitions other then the first one
 } >  normally start.  The effect is that you would most likely exclude all
 } >  partitions.  The bug is in the code that tries to deduce the number of
 } >  heads.  I'll have to think about how to fix this properly.
 } 
 } good to know, It was a quick hack in order to have fdisk working on
 } port-cobalt while the installation was failing because the MBR
 } was not correct apparently.
 } 
 } my understanding is that on non-x86 arch you check the MBR to guess the
 
      It does on x86 as well (see below and note that none of the head
 numbers makes sense).
 
 } geometry ? what about brand new harddrives, MBR must always be null no ?
 
      Not sure how this is handled, but yes the MBR would be all 0s.
 
 P4-3679GHz: {1} fdisk
 Disk: /dev/rwd1d
 NetBSD disklabel disk geometry:
 cylinders: 1453521, heads: 16, sectors/track: 63 (1008 sectors/cylinder)
 total sectors: 1465149168
 
 BIOS disk geometry:
 cylinders: 1023, heads: 81, sectors/track: 63 (5103 sectors/cylinder)
 total sectors: 1465149168
 
 Partition table:
 0: NetBSD (sysid 169)
     start 63, size 1465149105 (715405 MB, Cyls 0-287115/20/63), Active
 1: <UNUSED>
 2: <UNUSED>
 3: <UNUSED>
 Bootselector disabled.
 Drive serial number: 2140322 (0x0020a8a2)
 
 }-- End of excerpt from Eric Auge