CentOS - Aug 2020 - NFSroot over wifi in CentOS 7 - graceful shutdown problem

Hi,

I've got a task to have a small number of laptops netboot Linux over 
WiFi. The kernel is loaded off the USB stick of cource, it's off topic 
for now.

The WPA-supplicant daemon is started early by dracut off initrd. It 
works. Mostly.

The problem is that upon shutdown systemd terminates all the processes 
FIRST and unmounts filesystems NEXT.

Guess what? Upon termination, wpa-supplicant brings the wireless 
interface down and the system hangs being unable to unmount now-defunct 
NFSroot.

There were some discussions regarding similar matter and there's even 
the RH Errata:
https://access.redhat.com/errata/RHBA-2018:2447
But I'm woking on a fully updated CentOS Linux release 7.8.2003 and 
nevertheless I've got the problem.

I don't have rights to see the BZ.
https://bugzilla.redhat.com/show_bug.cgi?id=1593649

I'm deciding to use a quick and dirty hack to do a totally ungraceful 
shutdown/reboot:
https://unix.stackexchange.com/questions/533307/systemd-fails-to-umount-manually-mounted-nfs-shares-in-initramfs

Any better ideas?

Best regards,
	Dmitry Mikhailov.

> Hi,
>
> I've got a task to have a small number of laptops netboot Linux over
> WiFi. The kernel is loaded off the USB stick of cource, it's off topic
> for now.
>
> The WPA-supplicant daemon is started early by dracut off initrd. It
> works. Mostly.
>
> The problem is that upon shutdown systemd terminates all the processes
> FIRST and unmounts filesystems NEXT.
>
> Guess what? Upon termination, wpa-supplicant brings the wireless
> interface down and the system hangs being unable to unmount now-defunct
> NFSroot.
>
> There were some discussions regarding similar matter and there's even
> the RH Errata:
> https://access.redhat.com/errata/RHBA-2018:2447
> But I'm woking on a fully updated CentOS Linux release 7.8.2003 and
> nevertheless I've got the problem.
>
> I don't have rights to see the BZ.
> https://bugzilla.redhat.com/show_bug.cgi?id=1593649
>
> I'm deciding to use a quick and dirty hack to do a totally ungraceful
> shutdown/reboot:
>
https://unix.stackexchange.com/questions/533307/systemd-fails-to-umount-manually-mounted-nfs-shares-in-initramfs
>
> Any better ideas?
Short answer:

Looks like this should be fixed in systemd or its configuration.

Long answer is the one I already gave you on centos-devel. I post it here
because someone may have deeper knowledge of what exactly goes on:

What I'm wondering here is, do you see the hangs because of wpa_supplicant
was terminated and therefore also the wireless interface is down, or do
you see the hangs because networking is already completely down.

If it is because wpa_supplicant gets terminated, then maybe it would be
possible to terminate wpa_supplicant in a way that it lets the wireless
interface still up and functional at least for a certain amount of time.

I don't know wpa_supplicant good enough but I think maybe it is not always
needed to be alive for the wireless interface to work.

Otherwise, I guess it's more a question for the systemd developers than
for CentOS or Linux in general.

Regards,
Simon

On 8/3/20 6:50 PM, Simon Matter via CentOS wrote:
>> Hi,
>>
>> I've got a task to have a small number of laptops netboot Linux
over
>> WiFi. The kernel is loaded off the USB stick of cource, it's off
topic
>> for now.
>>
>> The WPA-supplicant daemon is started early by dracut off initrd. It
>> works. Mostly.
>>
>> The problem is that upon shutdown systemd terminates all the processes
>> FIRST and unmounts filesystems NEXT.
>>
>> Guess what? Upon termination, wpa-supplicant brings the wireless
>> interface down and the system hangs being unable to unmount now-defunct
>> NFSroot.
> Short answer:
> 
> Looks like this should be fixed in systemd or its configuration.
> 
> Long answer is the one I already gave you on centos-devel. I post it here
> because someone may have deeper knowledge of what exactly goes on:
> 
> What I'm wondering here is, do you see the hangs because of
wpa_supplicant
> was terminated and therefore also the wireless interface is down, or do
> you see the hangs because networking is already completely down.
It's because of wpa_supplicant. When I run the same config over wired 
network, it shuts down gracefully.

A dirty solution I found is to kill (-9) wpa_supplicant daemon right 
before shutdown. WiFi keeps running for a second it requires to do a 
graceful NFS unmount.

Best regards,
	Dmitry Mikhailov