Re: Bug#356850: initramfs-tools: upon udev upgrade, system becomes unbootable if using lilo

[Anthony DeRobertis <aderobertis@metrics.net>]

<posted & mailed>

Jonas Smedegaard wrote:

> I believe udev cintained within ramdisk needs to match udev outside -
> so if udev on the system is updated then (all!) initramfs-based
> ramdisks needs to be regenerated in order to survive _next_ boot.

Ouch! If this is true, it'd seem that somehow, the udev postinst needs
to find every intramfs image and regenerate it. First, that sounds like
a bad idea, even if it were technically possible. Second, it sounds
technically impossible (what if the admin generates one for his
hand-built kernel?)

> 
> That is a fundamental challenge to using udev in initial ramdisks. Not
> because of udev itself, but because of its lack of backwards
> compatibility (which can be a plus, as it means less ugly workarounds
> to fight with kernel maintainers about).

OTOH, it makes the system extremely fragile. If the new kernel breaks,
you're supposed to be able to boot with LinuxOLD. But it sounds like
that can't be made safe: Either there is a udev version mismatch or
there is a newly-created, potentially-broken initramfs for that kernel.

> Something like kernel-install-helper[1] perhaps?

Yeah.