On Mon, Oct 7, 2013 at 6:01 PM, Yasha Karant <[log in to unmask]> wrote:


> Two interesting observations from a colleague who now must support a system
> using systemd --


> /usr is now required to be in the boot partition because the boot process
> requires access to /usr and the executables/configuration files therein
> before the system actually is fully booted (even in the equivalent of single
> user mode)

For Fedora, that's incorrect. Dracut has two modules to mount non-"/"
partitions, fstab-sys, which has been available for a long time, and
usrmount, which has been in Fedora since v17 (IIRC). The latter's
specifically to mount a separate "/usr" partition after fsck-ing, just
like the "/" partition.

For Arch, that's incorrect. Mkinitcpio has a module for fsck-ing and
mounting a separate "/usr" partition.

For Gentoo, that's incorrect. I don't use genkernel but I know that it
can mount a separate "/usr" partition; but you have to have an
initrd/initramfs if you want to have a separate "/usr" partition.

For Debian, that's correct. There's an mkinitamfs patch to mount a
separate "/usr" partition but it hasn't been integrated into a release
even though some test debs were created and made available by the
patch-writer.

I'm not familiar with other distributions, but I'd be surprised if a
recent OpenSUSE mkinitrd couldn't mount a separate "/usr" partition.


> when the system generates a diagnostic due to a condition with systemd
> (e.g., some state issue for which the issue normally should not cause a
> machine to crash/become unstable), this fills the message file and then
> crashes the system because there is no space left on the partition that is
> required by systemd

"partition that is required by systemd"? Anyway, your colleague should
report a bug.