I have a number of Ethernet interfaces on a box running a 2.6 kernel and want to be able to configure IP addresses arbitrarily on these interfaces and talk to arbitrary remote hosts using them. My basic approach is to bind TCP sockets to different source IP addresses (based on the interface I want to go out) and use source-based routing and NAT. This almost works -- I can have the same subnet on all interfaces and remote hosts and it all works properly. I am having issues when the IP address on one of the interfaces collides with an IP address I want to talk to on another interface, however. What I see is that the TCP SYN gets an immediate RST back. The frame is never leaving the box so I presume there is some sort of local routing issue. Am I having issues because of the "local" routing table? The output of "ip rule list" makes me think that the local table has priority over my source-route tables: 0: from all lookup 255 32764: from 198.18.0.101 lookup TAP1 32765: from 198.18.0.100 lookup TAP0 32766: from all lookup main 32767: from all lookup default Looking at the local table I see the following (among other things): broadcast 10.10.10.0 dev tap0 proto kernel scope link src 10.10.10.10 broadcast 10.10.10.0 dev tap1 proto kernel scope link src 10.10.10.20 broadcast 10.10.10.255 dev tap0 proto kernel scope link src 10.10.10.10 broadcast 10.10.10.255 dev tap1 proto kernel scope link src 10.10.10.20 local 10.10.10.10 dev tap0 proto kernel scope host src 10.10.10.10 local 10.10.10.20 dev tap1 proto kernel scope host src 10.10.10.20 If I want to talk to 10.10.10.10 using the 10.10.10.20 interface, is my issue that the kernel is seeing 10.10.10.10 in the local table and never going any further? Joe Buehler -- To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
