[Pkg-cryptsetup-devel] Bug#428329: Bug#428329: Bug in debian-Installer/initramfs when I use "GUID Partition Table (GPT)" with a crypted root filesystem

[David Härdeman]

On Sun, December 9, 2007 18:43, Frans Pop wrote:
> On Monday 03 December 2007, David Härdeman wrote:
>> What happens if you don't? This sounds like a separate bug report should
>> be filed against the appropriate part of debian-installer (discuss on
>> debian-boot at lists.debian.org if you're unsure which component to report
>> it against).
>
> To David:
> Why a separate BR? What is the "other issue" in this one?
> Even if it is a separate issue, this report should be cloned rather than a
> new one filed, and that generally is the job of maintainers, not users.

There are two issues in the report:

One confusing description of not being able to select lilo in the
installation menu, I didn't quite understand what the problem was there
which is why I said that a separate BR is needed *if* indeed he
experienced some bug when trying to choose lilo.

The other problem (the real problem in this BR) is lilo's choice to use
major/minor notation for device-mapper devices which have no stable
major/minor numbers.

> The problem here seems to be that either lilo or lilo-installer fails to
> set root to the correct path (/dev/mapper/*) and falls back to the
> major/minor representation.

Yes

>> Root is on /dev/mapper/sda3_crypt so lilo encoded root=fe01, but the
>> next time you boot, /dev/mapper/sda3_crypt is brought up first, which
>> gives it major 254, minor 0, which is why it can't be found.
>>
>> I'm pretty certain that everything would have worked if the root device
>> would have been created first during installation
>> (making it /dev/mapper/sda2_crypt with major 254, minor 0) and swap
>> second.
>
> IMO it would be good to test if grub has the same issue or that providing
> the correct device path in the root= boot parameter somehow makes the
> initramfs scripts do the right thing.

The correct root=<path> boot parameter makes initramfs do the right thing,
and grub doesn't have the same issue since it always seem to use
root=<path> notation.

>> So this is not really a bug in cryptsetup, but a problem in the
>> interaction between device-mapper devices and lilo in debian-installer.
>
> Isn't providing consistent mapping across reboots (or between installation
> and reboot) something that cryptsetup (and its initramfs hooks) should
> provide?

No, major/minor are not stable (and they will get even less so over time,
the kernel has been moving more and more towards dynamic major/minor for
different systems), lilo is wrong for assuming they are. Grub gets it
right by using path, lilo doesn't.

So I'd still suggest that this BR be reassigned to lilo-installer, but I
do agree that fixing this is likely low priority and that the reporter
should use Grub unless there is a very good reason for not doing so.

-- 
David Härdeman