Hi Pat, Lankswert, Patrick wrote: > Where supported IPV6_RECVPKTINFO is a nice alternative where it is supported,
Yes, but only where it is supported :-) (In my parts of the woods it's near universal, but I'm not even close to having the widest range of systems available to me). > but the last time that I looked there was not a good way to specify which > interface that a packet should be sent *from* other than having a socket > bound to each interface. For a split-horizon system (e.g., a firewall with multiple interfaces with similar routes), just binding the sockets to an address may not work: The kernel is still using the routing table to find the outgoing interface, and that may have a "better" (but still unwanted) path via an interface with an address that the socket is not bound to. Hence hacks such as SO_BINDTODEVICE (which appears to be Linux only). According to RFC 3542 [1], IPV6_PKTINFO does have an outgoing interface index. I haven't done the necessary research here to see how well that is implemented. It will probably pay to talk to the DHCP and DNS implementer community; they have lots of experience with this issue. Gr??e, Carsten [1]: http://tools.ietf.org/html/rfc3542#section-6
