> I have finally gotten some time to get back to my project of making
> the installer install on an LVM root (and boot it afterward :-) and I
> am pleased to report that my mods work! With a caveat or two that I
> am hoping I can get help working out here.
>
Congratulations! :-) Still, have you seen my post with possible caveats
in using LVM for root?
http://www.mail-archive.com/[email protected]/msg44013.html
>
> The changes in the initrd for an LVM root are:
>
> echo 0x3a00 > /proc/sys/kernel/real-root-dev
This is not needed at all. Currently initrd mounts kernel and leaves
real-root-dev to be RAM disk and system is happy with it:
[root@cooker root]# cat /proc/sys/kernel/real-root-dev
256
Besides, as I already wrote, you do not have fixed relation between
device number and volume name. What you need in initrd is the *name* of
root device, not its number.
> /sbin/vgscan
Well, I ask ones more - what is the size of config database that is
created by vgscan? In my test one PV, one VG, one LV takes about 130K.
In case of several groups it looks a bit too much for RAM disk.
> /sbin/vgchange -a y
>
> The 0x3a00 as a root device relates to using /dev/VG/lvol1 (on my
> system here anyway)
Exacly! And even this relation may break if you create another LVM
group.
>
> Also, in the installer, when installing on an LVM "device", when it
> comes time to populating the lilo.conf file, for some reason the fully
> pathed root device is prefixed with a "/dev" so that it writes
> "root=/dev/dev/vg/lvol1".
>
> I'm afraid my perl is not up to snuff enough to suss out in the
> installer why this is happening. The install goes on fine if I choose
> to edit the image and change the Root field to remove the duplicate
> "/dev". Any ideas why this is happening? Is there some logic
> somewhere that is supposed to be stripping the /dev from the root
> device path that is failing in the case of an LVM device being used?
>
If you can make your mods available I'll get a look on weekend.
-andrej
