[systemd-devel] [PATCH] udev/net_id: Introduce predictable network names for Linux on System z

[Tom Gundersen]

Hi Hendrik,

On Thu, Jan 9, 2014 at 4:20 PM, Hendrik Brueckner <brueckner at redhat.com> wrote:
>> Looks good to me (assuming the names are truly stable between boots,
>> as I have no idea about how the internals of this work in the kernel).
>
> The bus-IDs do not change across reboots.

Great. I pushed the patch just now.

>> Could you also make sure that these things are supported by path_id
>> when you are at it? We use that to reference devices from
>> udev/networkd so makes sense if these things are in sync.
>
> Could you provide more information on that.  I know that the path_id creates
> the ID_PATH environment variable for udev.   There is already support in
> path_id for CCW devices used by network devices.  I am not sure if that is
> already sufficient...


It is. The only thing I meant to request was that ID_PATH is set
whenever ID_NET_NAME_PATH is. The reason for this is that ID_PATH
contains the same information, but without e.g. the length restriction
that ID_NET_NAME_PATH has. So while using ID_NET_NAME_PATH for the
interface name is the right thing to do, it is best to reference it by
ID_PATH and not by name, as there are (at least in theory) cases when
we are not able to set a persistent interface name despite all our
efforts :)

> So if you have some information, like the output of
> udevadm test-builtin path_id for a network devices, I can check how s390
> behave in this.   Anyhow, I would then address this on top of this issue.

I have:

ID_NET_NAME_PATH=wlp2s0
ID_PATH=pci-0000:02:00.0-bcma-0

but as long as ID_PATH is being set at all that should be all that's needed.

>> > See also Red Hat Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=870859
>>
>> Would be nice if you could post a summary, as a login is required.
>
> The network devices names switched between reboots and the systems where no
> longer accessible because the ip addresses were configured on the wrong
> network interface.

Thanks, added to the commit message.

Cheers,

Tom