eth0 ipaliases disappear

eth0 ipaliases disappear

[Shaun R.]

I've now seen this problem now for years, and i've posted on the ML about it 
before too with little responce.  I have a problem when switching (soft 
reboot) from the CentOS kernel (non xen kernel) to a xen kernel (non centos 
xen kernel).  Everything comes up fine, but what happens is i start gettign 
complaints from customers that ipaliases on there DomU's are non responsive. 
The primary (eth0) ip is alive and well but if you try to ping a ip say 
bound to eth0:1 you get nothing.  The temp solution to the problem is to log 
into that customers DomU and run arping using the source ip as the eth0:1 
device and hit the router (0.0.0.0).  This brings back up the eth0:1 alias 
until the arp once again times out.  So far the only real solution i've 
found is that i have to hard reboot the host, and make sure it boots the Xen 
kernel from the start, if it boots up on a stock centos kernel it will 
happen again.  Anybody else experience this problem?  so far i've seen this 
problem back 3.0.4-1 using kernel 2.6.16.38 but am now running 3.3.1 against 
2.6.18.8 and still see the problem.

My hosts are pretty much the same, the hardware has changed alittle over the 
years but it's basically all supermicro stuff.  Network cards are intel

    04:00.0 Ethernet controller: Intel Corporation 80003ES2LAN Gigabit 
Ethernet Controller (Copper) (rev 01)
    04:00.1 Ethernet controller: Intel Corporation 80003ES2LAN Gigabit 
Ethernet Controller (Copper) (rev 01)

The router shows incompelete for the ips not working when you do a show arp 
on them.

Anybody else expereince this probelm?  I'm running domU's in bridge mode 
(xenbr0 -> vif)

~Shaun