[systemd-devel] Fwd: systemd-nspawn network interface name collisions

[Florian Koch]

forgot the list....


---------- Forwarded message ----------
From: Florian Koch <florian.koch1981 at gmail.com>
Date: 2015-06-18 22:03 GMT+02:00
Subject: Re: [systemd-devel] systemd-nspawn network interface name collisions
To: Lennart Poettering <lennart at poettering.net>


2015-06-18 18:42 GMT+02:00 Lennart Poettering <lennart at poettering.net>:
> On Thu, 18.06.15 18:27, Florian Koch (florian.koch1981 at gmail.com) wrote:
>
>> Hi,
>>
>> if i understnd this correct, the network interface names (veth and
>> macvlan) are created with the frist 11 Caracters from the
>> Containername (Machinename).
>
> IFNAMSIZ emposed by the Linux kernel is 16, and we need three chars
> for the prefix "ve-" and one for the trailing NUL byte. makes 12 chars.

Thank you for the clarification


>> Now if you use similar names for conatiners, like
>>
>> com.$company.$devision.$name1
>> com.$company.$devision.$name2
>> com.$company.$devision.$name3
>>
>> the network device name handling is broken.
>>
>> I tryed to prefix the machinename with a uuid (uuidgen) but the the
>> names are to long.
>>
>> Why not using a 11 Caracter uuid / random  for network interface
>> names, and avoid all the naming trouble?
>
> Well, because we want to keep things easy to grok for users. If you
> type "ip link" and see the container names for the veth links, then
> that's certainly a lot more useful than seeing some random
> gibberish....

that is totally understandable, but what is with macvlan interfaces?
these are not shown in ip link (they are moved to the container
namespace)
The macvlan are my main Problem , we do not use veth interfaces.

But i think there is another Thread about this Problem too, with a proposed fix
http://lists.freedesktop.org/archives/systemd-devel/2015-June/033142.html

> I'd be willing to make this configurable:
>
>     --network-veth                → as it is now, host is called
>                                     ve-<container-name> and container
>                                     side is called host0
>
>     --network-veth=foo            → creates a veth link with both
>                                     sides named "foo"
>
>     --network-veth=foo:bar        → host side called "foo", container
>                                     side called "bar".
>
> At the same time we should open this up so that multiple links can be
> created, not just one.

sounds good.

> Happy to take a patch for that!

unfortunately I have no C programming experience, so i can not patch it.

regards Florian


> Lennart
>
> --
> Lennart Poettering, Red Hat