Hi,
Michael Biebl <biebl@debian.org> (2017-05-19):
> In jessie we handled this slightly differently [1]. We had a dedicated
> service unit which checked if /etc/mtab was a symlink. So we didn't run
> into the issue there, that the symlink can be absolute or relative and
> point to either /proc/mounts or /proc/self/mounts.
>
> We chose ../proc/self/mounts in debian.conf since that's also what's
> used by systemd upstream [2], i.e. we are consistent with other distros
> in that aspect.
>
> Maybe we can change debootstrap to use ../proc/self/mounts or is there a
> good reason why it should point to ../proc/mounts?
>
> CCed the debootstrap maintainers for their input.
I'm not exactly sure about the situation you describe, d-i does this
through a finish-install script (finish-install.d/70mtab):
| #! /bin/sh
|
| # some things inside d-i will make an /etc/mtab file, but it shouldn't
| # be there in the installed system. Systemd will be desperately unhappy.
| if [ -f /target/etc/mtab ]; then
| ln -sf /proc/self/mounts /target/etc/mtab
| fi
debootstrap itself does this:
| clear_mtab () {
| if [ -f "$TARGET/etc/mtab" ] && [ ! -h "$TARGET/etc/mtab" ]; then
| rm -f "$TARGET/etc/mtab"
| fi
| }
[ which matches its changelog entry → “On Linux, clear out /etc/mtab on
exit if it's not a symlink.” ]
and there's no /proc/*mounts in debootstrap's code.
Back to the original bug report:
Just performed a stretch debootststrap from a jessie system (with 1.0.67
version), no /etc/mtab afterwards. Did the same with debootstrap upgraded
to 1.0.90, same story.
Both times, with this command:
sudo debootstrap stretch /scratch/stretch http://localhost/debian
with /scratch being a tmpfs and localhost being my local, partial mirror.
Maximilian: If you're seeing a /etc/mtab inside a debootstrap'd
environment, you'll have to be more specific about the way you're
generating it.
KiBi.
Attachment:
signature.asc
Description: Digital signature