Re: [PATCH kernel-helper] kernel hooks: flag pending reboot after installing a new kernel

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Am 14.10.26 um 11:37 schrieb Jonas Theisen:
> On 10/14/26 10:35, Thomas Lamprecht wrote:
>> Am 13.10.26 um 13:47 schrieb Jonas Theisen:
>>> Ship a postinst.d hook that writes /run/reboot-required and records
>>> the kernel package in /run/reboot-required.pkgs, similar to
>>> the hooks shipped by unattended-upgrades on Debian and
>>> update-notifier on Ubuntu. Tools like needrestart or monitoring checks
>>> can then notice that the newly installed kernel is not running yet,
>>> without requiring one of those packages to be installed.
>> Any reference to those docs/code/... that describe these file (paths)
>> would be appreciated here.
> 
> From https://github.com/mvo5/unattended-upgrades
> 
> unattended-upgrade:
> 
> 105 REBOOT_REQUIRED_FILE = "/var/run/reboot-required"
> [...]
> 867 def reboot_required():
> 868     # type: () -> bool
> 869     return os.path.isfile(REBOOT_REQUIRED_FILE)
> 
> kernel/postinst.d/unattended-upgrades:
> 
>  8 if [ -d /var/run ]; then
>  9     touch /var/run/reboot-required
> 10     if ! grep -q "^$DPKG_MAINTSCRIPT_PACKAGE$" /var/run/reboot-required.pkgs 2> /dev/null ; then
> 11         echo "$DPKG_MAINTSCRIPT_PACKAGE" >> /var/run/reboot-required.pkgs
> 12     fi
> 13 fi
> 
> From https://github.com/liske/needrestart
> 
> data/update-motd-reboot-required
> 
>  5 if [ -f /var/run/reboot-required ]; then
>  6         cat /var/run/reboot-required
>  7 fi
> 
> Since `/var/run` is just a symlink to `/run` i figured we should use
> the current specification.
> 
> Should i also include these refs in the commit message?

Yes, please, at least in some basic prose form describing how
this is checked/used with [X] reference to the repo/file/docs/...


>>> +echo "*** System restart required ***" > /run/reboot-required
>>
>> That's always truncating any existing reboot-required file though?
>> Or doesn't that matter because it will be either not there or contain
>> exactly this string? In any case, needs either fixing or mentioning
>> that its fine in the commit message and maybe even a short
> 
> According to my understanding of this files usage it is just checked
> for existence and printed in the motd. The content seems to be irrelevant.

Ack, then it's at least not that problematic, but potentially still not
nice to blindly override it? Did you check how others - e.g. debian or
ubuntu handles this for kernel updates? As just adding a lot of
complexity with not much gain is not really worth it either, but maybe
others came up with a sensible solution here already.