Re: division by zero in winword

[Vincent Béron]

Le mer 22/09/2004 à 15:39, Joris Huizer a écrit :
> Robbert Xerox wrote:
> > Hi, in quite a few programs i can not browse the
> > filesystem when opening a "listbox"; to be more clear:
> > for example when i do in winword Open and the listbox
> > appears, and i try to open the pull down menu winword
> > just quits(without an error message in wine).  In
> > another program like "Camel join" it quits as soon as
> > i go up to the level of "my computer" with a message
> > "division by zero in Volme_ReadFATSuperblock. I was
> > able to
> > track the problem down to the following lines in
> > volume.c (starting from line 502)
> >   nsect -= GETWORD(buff, 0x0e) + buff[0x10] * sz +
> >         (GETWORD(buff, 0x11) * 32 + (GETWORD(buff,
> >
> > 0x0b) - 1)) / GETWORD(buff, 0x0b);
> >         nclust = nsect / buff[0x0d];
>
> if (GETWORD(buff, 0x0b) == 0)
> {
>      /* what to do about this?? */
> }
> else
> {
>      nsect -= GETWORD(buff, 0x0e) + buff[0x10] * sz +
>              (GETWORD(buff, 0x11) * 32 + (GETWORD(buff,
>               0x0b) - 1)) / GETWORD(buff, 0x0b);
>      if (buff[0x0d] == 0)
>      {
>          /* what to do about this one?? */
>      }
>      else
>      {
>          nclust = nsect / buff[0x0d];
>      }
> }
>
> Now, how come those values become zero? I don't know, so I can't fill in 
> the missing stuff there
> From an IRC session with the original user (or at least somebody pasting

the exact same question), he was using Wine as root, with a z: drive
pointing to /. The permissions on his / partition allowed Wine to read
the device directly.

Running Wine as a normal user fixed it for him.

Now onto what I think is a proper fix:
Note that the following check
    if (buff[0] == 0xE9 || (buff[0] == 0xEB && buff[2] == 0x90))
did pass before the above code snippet is executed, and doesn't
discriminate against other types of bootloaders (as that's what's
usually in the first sector, even a FAT). Indeed, my /dev/hde (boot
drive) MBR does pass the test as grub is installed in /dev/hde. If grub
had been installed in /dev/hde2 (/boot), the test would fail, but if I
had a single partition for /boot and / (rest of tree doesn't matter),
and installed grub in it's first sector rather than MBR, it'd pass.

So maybe we should actually check for the FAT signature at buff[0x36] or
buff[0x52] first, and then do the calculations if we actually do find
one of them.

Changelog:
Make sure we're really looking at a FAT before doing some calculations
based on the contents of the first sector of a partition.

Vincent