Re: v6.1-rc1: Regression in notification of sethostname changes

From: Petr Vorel
Date: Wed Oct 19 2022 - 08:50:43 EST


> Hi Torsten,

> > Hello Petr,

> > your commit

> > commit bfca3dd3d0680fc2fc7f659a152234afbac26e4d
> > Author: Petr Vorel <pvorel@xxxxxxx>
> > Date: Thu Sep 1 21:44:03 2022 +0200

> > kernel/utsname_sysctl.c: print kernel arch

> > Print the machine hardware name (UTS_MACHINE) in /proc/sys/kernel/arch.

> > This helps people who debug kernel with initramfs with minimal environment
> > (i.e. without coreutils or even busybox) or allow to open sysfs file
> > instead of run 'uname -m' in high level languages.

> > broke the notification mechanism between the sethostname syscall and the pollers of /proc/sys/kernel/hostname.

> > The table uts_kern_table is addressed within uts_proc_notify by the enum value, however no new enum value was added in "enum uts_proc".

> > I noticed the problem when journald-systemd failed to detect hostname changes made with the sethostname syscall (as used by the hostname tool).
> > When setting the hostname through /proc/sys/kernel/hostname the poll notification was working.

> Thanks a lot for your report, working on a fix!
> Andrew, Greg, sorry for a regression.

Hi Torsten,

could you please post exact steps to reproduce the problem.
Although the required fix to add new enum into enum uts_proc is trivial,
I was not able to reproduce the problem with 6.1.0-rc1 (actually
6.1.0-rc1-4.g1d716d8-default which contains few extra patches).

# hostname; hostnamectl hostname; cat /proc/sys/kernel/hostname
opensuse-tumbleweed.20221001
opensuse-tumbleweed.20221001
opensuse-tumbleweed.20221001

# hostnamectl set-hostname foo; echo $?
0
# hostname; hostnamectl hostname; cat /proc/sys/kernel/hostname
foo
foo
foo

# hostname bar; echo $?
0
# hostname; hostnamectl hostname; cat /proc/sys/kernel/hostname
bar
bar
bar

# echo "baz" > /proc/sys/kernel/hostname
# hostname; hostnamectl hostname; cat /proc/sys/kernel/hostname
baz
baz
baz


# hostnamectl set-hostname foo; reboot
After reboot it's 'foo'.
What am I missing?

BTW I originally tested the feature only on dracut initramfs (with rapido [1]),
which obviously bypass systemd. For a fix I'm creating rpm package (binrpm-pkg).

Kind regards,
Petr

[1] https://github.com/rapido-linux/rapido

> Kind regards,
> Petr

> > Torsten