Problem with acd on upgrade to 1.52

[Christian Schaubschläger]

Hi,

thanks for the analysis!

> First, the creation of the ACD probe fails. This shouldn't
> happen. Perhaps are you building NM with eBPF support? If so, it's
> possible that the process doesn't have the necessary capabilities to
> use the eBPF program. Anyway, the logging needs to be improved to
> indicate the exact cause for the failure.

Yes, I did compile it with eBPF support (though not entirely intentional ;-)
And I just checked my kernel: I do not have CONFIG_BPF_SYSCALL enabled. Maybe that's the problem?

However, I compiled 1.48 also with eBPF support, and I don't see the problem there...

> Second, there is a mechanism that retries the ACD probe creation and
> gives up after some seconds. Apparently this is not working and NM
> never recovers, as the log shows:
>
> Jul 10 09:08:23 acd[192.168.0.54, probing]: probing currently  stillnot possible (timeout 200 msec; failure to create nacd instance, retry probing on timeout)
> Jul 10 09:08:25 acd[192.168.0.54, probing]: probing currently  stillnot possible (timeout 200 msec; failure to create nacd instance, retry probing on timeout)
> Jul 10 09:08:29 acd[192.168.0.54, probing]: probing currently  stillnot possible (timeout 200 msec; failure to create nacd instance, retry probing on timeout)
> ...
> Jul 10 09:08:34 acd[192.168.0.54, probing]: probing currently  stillnot possible (timeout 200 msec; failure to create nacd instance, retry probing on timeout)

When NM is in the above state and I do

ip link set dev eth0 arp off

it immediately resumes and the client gets an ip address:

http://datenaustausch.s3.eu-west-1.amazonaws.com/Logs/daemon-noarp-resume.txt

(logs are from today from a different subnet).


> I will investigate this and prepare a patch. In the meantime, you can
> try disabling ACD for the connection:
>
>  nmcli connection modify "Wired connection 1" ipv4.dad-timeout 0
Can I disable ACD in general via configuration?

Thanks,
Christian