Subject: Re: [vserver] strange network issues
From: Herbert Poetzl <herbert@13thfloor.at>
Date: Sun, 22 Feb 2009 23:38:05 +0100

On Sat, Feb 21, 2009 at 01:52:30PM +1300, Corin Langosch wrote:
> Hi all,

> I ran linux 2.6.22 with vserver 2.2.0.6 for quite a long time and
> upgraded some days ago to

> Kernel: 2.6.26-1-vserver-amd64
> VS-API: 0x00020303
> util-vserver: 0.30.216-pre2772; Dec 13 2008, 04:56:19

> I notived that sometimes the interface of a vserver gets lost and
> replaced by lo with "127.0.0.1/255.0.0.1". 

I doubt that ...

> I don't know why this happens but it's really fatal as the vserver
> isn't accessible anymore.

> When stopping the vserver outputs "RTNETLINK answers: Cannot assign
> requested address" but otherwise closes fine.

> After starting it again, everything's fine again.

make sure the address is a primary, or 'promote
secondaries' is on, if you take down the primary
(or keep the primary permanently), otherwise this
is expected (mainline) behaviour since some years
now ...

> The vserver contains only one interface configured like this:
> ls /etc/vservers/test/interfaces/0/:
> dev   ip    mask 

> contents:
> eth0:vs15746-0
~~~~~~~~~~~~~~~~ this is just wrong, unless you
		 actually have a device with that name
> 235.231.51.92
> 255.255.255.224
if you really insist on havin an alias, put the
vs15746-0 into the 'name' entry, the device entry
should read 'eth0'

> I also noticed that the server now always show a "lo" interface,
> although none is configured. Has this changed from 2.2.0.6?

depends on the kernel and guest config, but yes,
loopback isolation is available in recent patches

> Is this a know issue with the release I'm using? Anyone faced the same
> weierd problem and found a solution? May be the solution is to always
> define a "lo" interface too (is this a new requirement?)?

no, you not ever define an 'lo' interface, not
in the past and not now ...

best,
Herbert

> Thanks,
> Corin