Xen users - Feb 2007 - Network problem: packets are lost in domU

Hello,

I am having big problems with Xen virtual network interfaces. First
I tried typical bridge support, no luck. The packets sent from dom0
to domU seemed to disappear somewhere. Then, to debug the problem
I have set up a simple point-to-point link and still have the same
problem.

I am using Xen 3.0.4-1 and Linux 2.6.16.x in both dom0 and domU.
I have tried kernel 2.6.16.35 and 2.6.16.41. 
I have compiled the kernel and xen with PAE support and without it,
using gcc 3.3 and gcc 4.2 (snapshot). I have played with various kernel
compilation options. No luck: the same problem in every case.

My system configuration:

in domain config file:

        vif = [ ''ip=192.168.16.2'' ]

in xend-config.sxp:

        (network-script ''network-route netdev=dummy0'')
        (vif-script     ''vif-route netdev=dummy0'')

dom0 interface setup:

# ip addr show dev dummy0
10: dummy0: <BROADCAST,NOARP,UP> mtu 1500 qdisc noqueue
    link/ether 1a:aa:f1:3f:c5:4e brd ff:ff:ff:ff:ff:ff
    inet 192.168.16.1/32 scope global dummy0
# ip addr show dev vif2.0
14: vif2.0: <BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast qlen 32
    link/ether fe:ff:ff:ff:ff:ff brd ff:ff:ff:ff:ff:ff
    inet 192.168.16.1/32 brd 192.168.16.255 scope global vif2.0
# ip route show 192.168.16.2
192.168.16.2 dev vif2.0  scope link  src 192.168.16.1

domU interface setup:

# ip addr show dev eth0
2: eth0: <BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast qlen 1000
    link/ether 00:16:3e:32:a3:70 brd ff:ff:ff:ff:ff:ff
    inet 192.168.16.2/24 brd 192.168.16.255 scope global eth0
# ip route
192.168.16.0/24 dev eth0  proto kernel  scope link  src 192.168.16.2

And "ping" does not work in either direction. Tcpdump shows that
frames
sent from dom0 do not reach domU. So does "ip":

# ip -s link show dev eth0
2: eth0: <BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast qlen 1000
    link/ether 00:16:3e:32:a3:70 brd ff:ff:ff:ff:ff:ff
    RX: bytes  packets  errors  dropped overrun mcast
    0          0        72      0       0       0
    TX: bytes  packets  errors  dropped carrier collsns
    2394       57       0       0       0       0

And there are errors in dmesg of domU for each package lost this way:

netfront: rx->offset: 12, size: 4294967295
netfront: rx->offset: 12, size: 4294967295
netfront: rx->offset: 12, size: 4294967295
netfront: rx->offset: 12, size: 4294967295
[...]

No other suspicious messages in dmesg of domU, nothing worying in dmesg
of dom0.

Any ideas or hints?

Greets,
        Jacek

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

On Thu, 2007-02-22 at 15:09 +0100, Jacek Konieczny
wrote:
> My system configuration:
> 
> in domain config file:
> 
>         vif = [ ''ip=192.168.16.2'' ]
> 
> in xend-config.sxp:
> 
>         (network-script ''network-route netdev=dummy0'')
>         (vif-script     ''vif-route netdev=dummy0'')
> 
> dom0 interface setup:
> 
> # ip addr show dev dummy0
> 10: dummy0: <BROADCAST,NOARP,UP> mtu 1500 qdisc noqueue
>     link/ether 1a:aa:f1:3f:c5:4e brd ff:ff:ff:ff:ff:ff
>     inet 192.168.16.1/32 scope global dummy0 
I''m not sure if the config you shows is supposed to be for bridging or
PtP, but this response assumes bridging.  If that''s not the case then
just ignore.

For bridging your network-script should be ''network-bridge''
instead of
''network-route''.

Also for bridging dummy0 should be in promiscuous mode.

I really don''t know how the Xen networking scripts work, I usually set
up the host-side networking manually (using the host OS''s networking
tools) and let Xen handle the domU side.

Probably doesn''t help but maybe this provides some clues.

-m


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

On Thu, Feb 22, 2007 at 08:24:23AM -0600, Marduk wrote:
> I''m not sure if the config you shows is supposed to be for
bridging
The setup included in my mail was for the "PtP" setup. I wanted to
exclude possible bridging problems during debugging.
> For bridging your network-script should be
''network-bridge'' instead of
> ''network-route''.
I know. That setup didn''t work either.

Greets,
        Jacek

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

Hi,
> I am using Xen 3.0.4-1 and Linux 2.6.16.x in both dom0 and domU.
> I have tried kernel 2.6.16.35 and 2.6.16.41. 
> I have compiled the kernel and xen with PAE support and without it,
> using gcc 3.3 and gcc 4.2 (snapshot). I have played with various kernel
> compilation options. No luck: the same problem in every case.
> 
> [...]
>
> And there are errors in dmesg of domU for each package lost this way:
> 
> netfront: rx->offset: 12, size: 4294967295
> netfront: rx->offset: 12, size: 4294967295
> netfront: rx->offset: 12, size: 4294967295
> netfront: rx->offset: 12, size: 4294967295
> [...]
> 
> No other suspicious messages in dmesg of domU, nothing worying in dmesg
> of dom0.
I have the exact same problem. I just wanted to upgrade to 2.6.19 +
3.0.4 from the Fedora development and boom, same problem.

It works when I use my old Dom0 (2.6.18 + 3.0.3) with 3.0.4 guests. The
only visible difference is that with the new (broken) Dom0, the guests
are using the "copying receive path" while with the working old Dom0
they are using the "flipping receive path".

This smells like a serious bug in the netback or netfront driver in the
copying receive path.

I don''t know what can trigger this, but my configuration is:

Dom0: xenbr0 with one netloop attached (int0) and several vif#.0 for
DomU''s (no physical network card attached, done via routing on Dom0).

4294967295 is -1 writted as unsigned, which indicates some sort of error
state in the code, but I don''t know from where that one might come.



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