[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#283346: Installing LILO into LVM partition fails badly



Package: debian-installer
Version: Sarge netinst RC2 

What I did: I created a LVM partition on hda6 with a root and a home 
partition inside. Because /boot is on the LVM, the installer defaults 
to LILO. I decided to install LILO into hda6 (in order to chainload it 
via another installation that uses grub). 

The problem: The installation of LILO stops at 75% and keeps hanging at 
this point. This is really a problem, because there is no way to 
continue the installation without manual intervention on the console.

The investigation: I changed to console #2 and started a shell. Using 
ps, I saw an idle lilo.real process. I killed it and the LILO 
installation failed with an error. I chrooted /target and executed lilo 
manually. Lilo complained that partition type 0x8E (LVM) "is a 
dangerous place" for the bootloader. It asked me whether it should 
proceed under this circumstances. I typed 'Y' and the installation 
finished without problem, and a reboot showed that LILO was working 
fine.

The diagnosis: I suspected that d-i can't handle the interactive 
question that LILO asks. So I tried to replace all the invocations of 
lilo.real in the /sbin/lilo-script with something like yes|lilo.real, 
in order to say "yes" to every question lilo might ask. This time the 
installation process finished without problem.

The solution: I think that there are two possibilities. First, if LILO 
is right that you should not install it into a LVM partition, then the 
installer has to warn the user that it's not possible to use this 
partition before lilo is executed. I took a look at the lilo source and 
it seems to show that lilo complains about every partition type, except 
0x83 and minix. That's why I have the feeling that the second 
possibility applies: lilo's warning and proceed-question is wrong and 
should either be ignored by d-i or be disabled in the lilo-package. I 
would prefer to do both things: stop lilo complaining about partition 
types that are safe (and thus suppressing the interactive question) and 
fix d-i in a way that it can handle all questions that lilo might ask 
(a thorough inspection of the lilo source would be a good idea).



Reply to: