Red Hat Bugzilla – Bug 18709
RFE: Improve behavior when Unixware slices exist
Last modified: 2007-03-26 23:36:22 EDT
Firstly, this bug is not entirely the fault of RH, but more of how linux
copes with slices. However, RH may be able to provide a work around or may
wish to place appropriate comments in installation docs.
HDA is partitioned thus:
hda1 System Utilities
hda2 SysV (Contains unixware 7)
hda6 linux swap
hda7 win95 fat 32
Redhat installer correctly identifies the hda5 partition as suitable for
installation and completes a normal install. In /etc/lilo.conf, install
puts the root fs as /dev/hda5 which, again, would seem quite normal.
On reboot however, linux panics as it cannot mount the root FS.
The reason for this is that the running kernel has identified that there
are slices in hda2 and so renumbers the partitions accordingly:
hda: hda1 hda2 <unixware: hda5 hda6 hda7 hda8 hda9> hda3 hda4 <hda10 hda11
As a result, the partition specified as the root is now actually a
The resolution to this is to boot again and specify the correct partition
on the command line (i.e. CTRL+X linux root=/dev/hda11) then to
modify /etc/lilo.conf and rerun lilo.
At this stage, I have three suggestions for RedHat.
1. Removal of unixware support from their default kernel (it is, after
all, a relatively niche requirement)
2. Use the appropriate tool (prtvtoc?) at install time to probe SysV
partitions and guess what the correct partition number to use will be.
3. Document it. Obviously if the linux partitions are earlier than the
unixware ones, this problem should not occur, but it may be of use to
people to see this situation documented in the install-guide.
As an aside, I think this method of assigning partition numbers to slices
is inappropriate; there should be a seperate virtual device for these as
the current system means that this kind of error will be commonplace.
I'm changing this to an enhancement request - the bug poster has some good
suggestions for how to handle this.
Agreed - apologies for not selecting the correct definition in the first place.
If there's any further I can do to help, please let me know.
Changing to a kernel request.
Closing. Instead we switched to using volume labels which solves bigger problems too