Re: lilo-22.3.2-3 trashed my SCSI disk
Maybe a hardware fault happened, in this case unfortunately at the
same time as a rewrite of the lilo boot information. I was just about
to file a bug report... Still there are some open issues remaining.
Russell Coker writes:
> On Tue, 3 Sep 2002 08:44, Svante Signell wrote:
> > After upgrade of lilo in unstable my whole SCSI disk got trashed. It
> > could only be recovered with the use of the IBM drive fitness test
> > tool, and a complete erase disk was necessary :-( What happened?
>
> If you need to run an IBM diagnostics program to get your disk working then
> it's a hardware issue and nothing to do with LILO.
>
> LILO like all Linux programs does not get to talk to the hardware directly and
> only gets to write data to the block device (/dev/sda or /dev/sda1). If
> writing to such a device can require special IBM utilities to recover then
> it's a hardware or device driver issue (but most likely hardware).
>
> > A reinstall of Woody showed that it can only boot from the MBR
> > partition, not the root partition, i.e.
> >
> > boot=/dev/sda, works!
> > boot=/dev/sda1, does not work!
>
> Strange, /dev/hda1 in the form of /dev/md1 works for me.
As seen from the original posting, the boot sector information was
boot=/dev/sda1
root=/dev/sda1
and it did work before, but not after the crash. Explanation?
When running lilo before and after the same information is displayed:
Reading boot sector from /dev/sda1
After the crash (with 22.3.3-2):
Reading boot sector from /dev/sda
Using MENU secondary loader
Calling map_insert_data
Can someone explain (or give a pointer to) the different behaviour of
writing to the MBR vs the root partition?
>
> > What has changed for newer versions of lilo? I have been running
> > Debian stable/testing/unstable for several years now without any
> > problems before.
>
> 22.3 was one of the biggest changes to LILO in recent times that did have
> potential to cause breakage. The versions after that were minor changes.
The version upgrade was from 22.3.2-1 to 22.3.2-3. The main difference is the removal of the /boot/boot.b link.
> > Note also that I have a dual disk system, SCSI and IDE, therefore the
> > disk=, bios= statements in lilo.conf. The disk partitioning tools, such as
> > cfdisk requires both the SCSI disk and the IDE disk to have at least
> > one partition with the boot flag set. Is this really necessary?
>
> No. You don't really need any boot flags to be set.
Then why can't I write the partiton table in cfdisk without setting
the boot flag on one of the partitions of the IDE disk?
>
> > Does the install=menu stuff have anything to do with the crash?
>
> No. It's the default action anyway...
>
> > Maybe grub is better with respect to error recovery?
>
> If your hardware fails to badly that Linux programs can't fix it then grub vs
> LILO is not an issue.
>
>
> Russell Coker
>
>
> PS Let's move this to debian-user.
OK
Reply to: