I have a multi-homed application server that serves multiple networks that are not connected with each other (other than being all connected to the server, each one on its dedicated network interface). Some of these networks have duplicate addressing, ie. there may be two completely different networks that use 10.10.0.0/16 address space. We cannot change addressing of these networks, so in that case we connect one of them to the server via static one-to-one NAT, so that from the server point of view the network is for example 10.90.0.0/16. Because the server should have a fixed, predefined address within the network (say, 10.10.0.100), both destination and source NAT is used, so the NATed network can access the server using address 10.10.0.100 like there was no NAT.
Until now, everything works fine using just iptables on the NATing machine. But we need to add the functionality to network boot some machines in the NATed network via PXE. The boot/DHCP server will be of course the muti-homed server in question, so I guess a DHCP relay needs to be added on the NAT machine.
I don't know how to configure the relay so that NAT will be taken into account as well, ie. a machine requesting DHCP will be pointed to 10.10.0.100 as DHCP server and boot source, but will in fact communicate over NAT with the server on the other side.
The machine doing NAT is a VM running OpenWRT, so a solution using dnsmasq (which is installed by default in OpenWRT) would be preferred. I know dnsmasq can act as DHCP relay, but I don't quite understand the parameters to configure this feature.
Of course if a specific configuration of the DHCP server itself, that for example involves sending to clients already transformed (NATed or unNATed) IP addresses, is required, that's acceptable as well.
Is it possible, and if yes, how?
