Xen users - Jan 2008 - xen 3.2 network problems

Hi,

I try to get networking running with xen 3.2, but it seems I fail miserably.

I am on sles10sp1, with xen 3.2.0, and kernel rpm''s installed from SP2.

I configured in /etc/sysconfig/network the NETWORK_LIST:

NETWORK_LIST="bridge,0,default,00:DE:AD:BE:EF,192.168.8.13/24,dhcp-off"

but an ifconfig xenbr0 still shows the mac address of peth0 
fe:ff:ff:ff:ff:ff.

Initially, I had a bond interface on top of eth0 and eth1, the default route 
was via the bond interface. after rebooting the xen kernel, it saw a message 
that veth0 was missing, it said maybe because of not enough available 
loopback devices. I think more that this had to to with the bond0 interface.
However, I followed the advice, and created some files below /etc/modprobe.d 
with following contents:

cat /etc/modprobe.d/*loop
options loop max_loop=64
options netloop nloopbacks=64

Also, with the bond interface, when the ordinary network started, it created 
the bond0 interface, and I was able to ping it for some seconds, until xend 
was started, and it tried to add the bond0 to the xenbr0.

when I now restart xend, I have veth0-veth63 and I have vif0.0-vif0.63. Is 
this the way it should be? I thought it would only create virtual interfaces 
for the amount of physical interfaces on the dom0 available, at least this 
was how I have it seen on xen 3.0.4 and 3.1.0. 

Then I removed the bond0, and only had a eth0 interface, with IP configured, 
the default route of the dom0 now via the eth0 interface. After a reboot, 
this was working well. 

So fine, then I added a vlan on top of the eth0, and rebooted. Without 
starting xend, I was able to communicate via eth0 and the vlan, then I 
started xen, and the tcpdump on the vlan interface stopped seeing packets on 
that interface. The eth0 was still working. On xen 3.0.4 and xen3.1. I had 
no such problem with vlan interfaces, so is there anything I am missing?


kind regards
Sebastian



_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Sebastian Reitenbach schrieb:
> Hi,
>   
Hi,
> I try to get networking running with xen 3.2, but it seems I fail
miserably.
>
> I am on sles10sp1, with xen 3.2.0, and kernel rpm''s installed from
SP2.
>
>   
> I configured in /etc/sysconfig/network the NETWORK_LIST:
>   
I guess you meant /etc/sysconfig/xend?
>
NETWORK_LIST="bridge,0,default,00:DE:AD:BE:EF,192.168.8.13/24,dhcp-off"
>
> but an ifconfig xenbr0 still shows the mac address of peth0 
> fe:ff:ff:ff:ff:ff.
>
> Initially, I had a bond interface on top of eth0 and eth1, the default
route
> was via the bond interface. after rebooting the xen kernel, it saw a
message
> that veth0 was missing, it said maybe because of not enough available 
> loopback devices. I think more that this had to to with the bond0
interface.
> However, I followed the advice, and created some files below
/etc/modprobe.d
> with following contents:
>
> cat /etc/modprobe.d/*loop
> options loop max_loop=64
> options netloop nloopbacks=64
>
> Also, with the bond interface, when the ordinary network started, it
created
> the bond0 interface, and I was able to ping it for some seconds, until xend
> was started, and it tried to add the bond0 to the xenbr0.
>
> when I now restart xend, I have veth0-veth63 and I have vif0.0-vif0.63. Is 
> this the way it should be? I thought it would only create virtual
interfaces
> for the amount of physical interfaces on the dom0 available, at least this 
> was how I have it seen on xen 3.0.4 and 3.1.0. 
>
> Then I removed the bond0, and only had a eth0 interface, with IP
configured,
> the default route of the dom0 now via the eth0 interface. After a reboot, 
> this was working well. 
>
> So fine, then I added a vlan on top of the eth0, and rebooted. Without 
> starting xend, I was able to communicate via eth0 and the vlan, then I 
> started xen, and the tcpdump on the vlan interface stopped seeing packets
on
> that interface. The eth0 was still working. On xen 3.0.4 and xen3.1. I had 
> no such problem with vlan interfaces, so is there anything I am missing?
>   
I guess not. I''ve tried the same with SLES 10 SP2 (Beta1!). I think I
tracked down
the problem: SuSE changed the network script to "network-multinet" in
3.2.0.
It''s really cool, because it''s quite easy to setup several
different
networks (and
you don''t need to start scripting in order to get this up and running).
However
the script has some problems with vlans (and maybe bonds, didn''t test
that).
The problem is the usage of ifup/ifdown do (de-)activate interfaces.
That works
fine on real (eth*) devices, but running ifdown on a vlan makes the
device disappear
(and that''s what breaks the Xen magic).

So I replaced some calls to ifdown with "ip link set dev $dev down"
and
vlans
started working. I''ve attached my version of multinet-common.sh (just
copy to
/etc/xen/scripts/) - my modifications are only in
create_normal_bridge(), maybe
to get your setup working you need to do the same in create_bonded_bridge().

I''ve submitted a service request to Novell, so if you could test the
bonding stuff,
and give some feedback, I''d update / extend this service call. And
maybe
we''ll
get a working version on SLES 10 SP2 final ;-)
> kind regards
> Sebastian
>   
Marcel

-- 
----
Dipl.-Inf. Marcel Ritter
Linux/Novell
Regionales Rechenzentrum Erlangen
----
Unix _IS_ user friendly... It''s just selective about who its friends
are.




_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Sebastian Reitenbach <sebastia@l00-bugdead-prods.de> wrote:
> Hi,
> 
> I try to get networking running with xen 3.2, but it seems I fail 
miserably.
> 
> I am on sles10sp1, with xen 3.2.0, and kernel rpm''s installed from
SP2.
> 
> I configured in /etc/sysconfig/network the NETWORK_LIST:
> 
>
NETWORK_LIST="bridge,0,default,00:DE:AD:BE:EF,192.168.8.13/24,dhcp-off"
> 
> but an ifconfig xenbr0 still shows the mac address of peth0 
> fe:ff:ff:ff:ff:ff.
> 
> Initially, I had a bond interface on top of eth0 and eth1, the default 
route 
> was via the bond interface. after rebooting the xen kernel, it saw a 
message 
> that veth0 was missing, it said maybe because of not enough available 
> loopback devices. I think more that this had to to with the bond0 
interface.
> However, I followed the advice, and created some files 
below /etc/modprobe.d 
> with following contents:
> 
> cat /etc/modprobe.d/*loop
> options loop max_loop=64
> options netloop nloopbacks=64
> 
> Also, with the bond interface, when the ordinary network started, it 
created 
> the bond0 interface, and I was able to ping it for some seconds, until 
xend 
> was started, and it tried to add the bond0 to the xenbr0.
> 
> when I now restart xend, I have veth0-veth63 and I have vif0.0-vif0.63. Is 
> this the way it should be? I thought it would only create virtual 
interfaces 
> for the amount of physical interfaces on the dom0 available, at least this 
> was how I have it seen on xen 3.0.4 and 3.1.0. 
> 
> Then I removed the bond0, and only had a eth0 interface, with IP 
configured, 
> the default route of the dom0 now via the eth0 interface. After a reboot, 
> this was working well. 
> 
> So fine, then I added a vlan on top of the eth0, and rebooted. Without 
> starting xend, I was able to communicate via eth0 and the vlan, then I 
> started xen, and the tcpdump on the vlan interface stopped seeing packets 
on 
> that interface. The eth0 was still working. On xen 3.0.4 and xen3.1. I had 
> no such problem with vlan interfaces, so is there anything I am missing?
> 
The first thing was my fault, the default network-script changed from 
network-bridge to network-multinet, but after changing this back, it did not 
helped much.

Nevertheless, after some more hours fiddling around, it seems I got it 
working as I want it to work. I had to take the network-bridge, 
xen-network-common.sh, xen-script-common.sh, and the vif-bridge from a xen 
3.0.4 version. 

With the 3.2 network scripts, it throwed the vlan interfaces on top of my 
bond0 interfaces away, but I need these later. With the 3.0.4 scripts, these 
interfaces stay alive, and they work as expected after a xend start.
I had to define this in the xend-config.sxp file to be able to use the bond0 
interface, instead of the eth0, as I wanted:
(network-script ''network-bridge netdev=bond0'')
The default route was set via the bond0 interface therefore I wonder why 
xend start tried to fiddle with eth0 instead of bond0.

Can anybody explain me the reason why the vlan interfaces on top of bond0 
are thrown away on xend start in the xen 3.2 version?


cheers
Sebastian


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi,
Marcel Ritter <Marcel.Ritter@rrze.uni-erlangen.de> wrote:
> Sebastian Reitenbach schrieb:
> > Hi,
> >   
> Hi,
> > I try to get networking running with xen 3.2, but it seems I fail 
miserably.
> >
> > I am on sles10sp1, with xen 3.2.0, and kernel rpm''s installed
from SP2.
> >
> >   
> > I configured in /etc/sysconfig/network the NETWORK_LIST:
> >   
> I guess you meant /etc/sysconfig/xend?
> >
NETWORK_LIST="bridge,0,default,00:DE:AD:BE:EF,192.168.8.13/24,dhcp-off"
> >
> > but an ifconfig xenbr0 still shows the mac address of peth0 
> > fe:ff:ff:ff:ff:ff.
> >
> > Initially, I had a bond interface on top of eth0 and eth1, the default
route 
> > was via the bond interface. after rebooting the xen kernel, it saw a 
message 
> > that veth0 was missing, it said maybe because of not enough available 
> > loopback devices. I think more that this had to to with the bond0 
interface.
> > However, I followed the advice, and created some files 
below /etc/modprobe.d 
> > with following contents:
> >
> > cat /etc/modprobe.d/*loop
> > options loop max_loop=64
> > options netloop nloopbacks=64
> >
> > Also, with the bond interface, when the ordinary network started, it 
created 
> > the bond0 interface, and I was able to ping it for some seconds, until
xend 
> > was started, and it tried to add the bond0 to the xenbr0.
> >
> > when I now restart xend, I have veth0-veth63 and I have
vif0.0-vif0.63.
Is 
> > this the way it should be? I thought it would only create virtual 
interfaces 
> > for the amount of physical interfaces on the dom0 available, at least 
this 
> > was how I have it seen on xen 3.0.4 and 3.1.0. 
> >
> > Then I removed the bond0, and only had a eth0 interface, with IP 
configured, 
> > the default route of the dom0 now via the eth0 interface. After a 
reboot, 
> > this was working well. 
> >
> > So fine, then I added a vlan on top of the eth0, and rebooted. Without
> > starting xend, I was able to communicate via eth0 and the vlan, then I
> > started xen, and the tcpdump on the vlan interface stopped seeing 
packets on 
> > that interface. The eth0 was still working. On xen 3.0.4 and xen3.1. I
had 
> > no such problem with vlan interfaces, so is there anything I am
missing?
> >   
> I guess not. I''ve tried the same with SLES 10 SP2 (Beta1!). I
think I
> tracked down
> the problem: SuSE changed the network script to
"network-multinet" in
3.2.0.
> It''s really cool, because it''s quite easy to setup
several different
> networks (and
> you don''t need to start scripting in order to get this up and
running).
> However
> the script has some problems with vlans (and maybe bonds, didn''t
test
that).
> The problem is the usage of ifup/ifdown do (de-)activate interfaces.
> That works
> fine on real (eth*) devices, but running ifdown on a vlan makes the
> device disappear
> (and that''s what breaks the Xen magic).
> 
> So I replaced some calls to ifdown with "ip link set dev $dev
down" and
> vlans
> started working. I''ve attached my version of multinet-common.sh
(just
> copy to
> /etc/xen/scripts/) - my modifications are only in
> create_normal_bridge(), maybe
> to get your setup working you need to do the same in 
create_bonded_bridge().
> 
> I''ve submitted a service request to Novell, so if you could test
the
> bonding stuff,
> and give some feedback, I''d update / extend this service call. And
maybe
> we''ll
> get a working version on SLES 10 SP2 final ;-)
I''m ready for today, see my other mail ;) I''ll find some time
tomorrow to
try the bonding and vlan stuff.

thanks
Sebastian


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users