[lvs-users] LVS stops TFTP Server from sendign UDP packets

Richard Scott lvs-users at pointb.co.uk
Fri Sep 18 11:03:25 BST 2015


Hi,

I am unable to get TFTP to continue to work after activating LVS on a 
kernel that is 3.8.x or newer.
Oh, and before you ask... this is not an ARP problem ;-)

Im my current setup I have a PXE Boot Server running Scientific Linux 
7.1 with LVS on a 3.10.x kernel. LVS distributes traffic for port 25 to 
some PXE booted nodes. This all runs fine with older kernels and kernel 
3.6.10 was my last successful version used. I have two interfaces on the 
server, one for LVS listening for traffic for port 25, and a 2nd 
interface just for PXE booting and the LVS to send traffic to the real 
servers.

LVS is not broken, but it breaks TFTP and drops UDP packets leaving the 
server once activated but ONLY in kernels 3.8.x or newer.

To test, setup a PXE server and follow these steps.

1. boot server with selinux disabled and no iptables or ipvsadm rules in 
place.
   -- test boot PXE node = OK

2. iptables -t mangle -A PREROUTING -i enp0s8 -p tcp -s 0/0 -d 
192.168.0.253 --dport smtp -j MARK --set-mark 1
   -- test boot PXE node = OK

3. ipvsadm -A -f 1 -s rr
   -- test boot PXE node = FAIL

4. reboot to restore pxe boot service.

Now, if you boot a PXE node during each step above, you will notice that 
at step 3 your PXE nodes will start failing to boot and you need to 
reboot your PXE server to restore PXE boot service.

I currently have this working fine in Scientific Linux 6 running a 
2.6.32 kernel, but kernels newer than 3.6.10 all seem to have this a 
problem ;-(
I did raise a bug a while back with RedHat as I was testing with Fedora 
at the time but it got ignored - 
https://bugzilla.redhat.com/show_bug.cgi?id=988941

Is this a known bug/limitation of LVS?

Thanks,

Richard



More information about the lvs-users mailing list