[f-nsp] Interesting DHCP/netboot problem

[George B.]

I have servers that generally run in production with two interfaces of
a stack of FCXs in a trunk running to a linux server. The linux server
has both interfaces bonded.  That works well.

The problem comes in when if the server needs to be net booted in
order to rebuild it using kickstart or any other reason it might need
to netboot.

At boot time, the interfaces are not bonded at the server.  If a DHCP
query goes out on one interface, the response might come back on the
other interface. Short of hard coding a MAC address on the kickstart
server for each server, is there any way I can tell the switch to
ignore DHCP clients on the second interface?

For example, it would seem like I would like to "config-trunk-ind" on
the trunk and then "dhcp snoop client disable <interface>" on all but
the primary port. But even that might not address the tftp problem.
It seems that trunking ports to a server for high-availability breaks
net boot unless you hard code the macs in the DHCP server and I am not
positive that even that will address the problem because traffic might
STILL go out the additional trunk ports.

The fundamental problem is that at net-boot, the two ports are trunked
on the switch but are not trunked on the server at that point and it
appears that the BIOS brings up all NICs during netboot.  Has anyone
managed to work around this?