Incoming interface would be useful for multicast UDP

[pusateri]

In C, you use recvmsg() to get the incoming interface. It's a bit complicated for differences between unix and linux. Not sure about Windows. Sample C code is in this gist:

https://gist.github.com/pusateri/58d4ea943cfbfb7e2b3af955293227fe

recvfrom() won't give you the incoming interface, just the source address.

[alexcrichton]

Is there perhaps a missing function to bind for this library? Or could you narrow it down a bit what needs to happen in this library?

[pusateri]

recvfrom() doesn't provide the incoming interface for a packet. code that runs on multiple interfaces (like mDNS) needs the incoming interface (usually an index).

[pusateri]

This is important for multicast UDP because the destination address is always the multicast IP address and the source address can't be trusted because it's UDP and no 3-way handshake occurs to ensure return reachability. So the incoming interface is necessary to prevent spoofing.

[alexcrichton]

@pusateri er right yeah makes sense, but I'm curious what API needs to be added to this library to expose the incoming interface? Or put another way, what would a PR to solve this look like?

[pusateri]

Sorry, I misunderstood the question. Similar to the existing recv_from:

pub fn recv_from(&self, buf: &mut [u8]) -> Result<(usize, SocketAddr)>

but with the new if_index. Maybe something like:

pub fn recv_from_intf(&self, buf: &mut [u8]) -> Result<(usize, SocketAddr, usize)>

But any function name is fine if you prefer something like recv_from_if() or recv_if()

[alexcrichton]

Hm the implementation here is basically switching to recvmsg, right? If that's the case then I think we'd probably want to expose the lowest level API, and leave all the various parsing of the result to calling code (although we could also add conveniences as well)

[pusateri]

Sure. I can try documenting a Rust version of recvmsg(). It does a lot more than incoming interface which others will probably appreciate.

[alexcrichton]

FWIW I think it'd be fine to basically add that API to this crate at any time, and we could probably just use the same raw pointers as the underlying C APIs and just push the safety to users, making it convenient to at least call the C API

[pusateri]

It looks like nix has already done this.
https://github.com/nix-rust/nix/blob/master/src/sys/socket/mod.rs
Test here:
https://github.com/nix-rust/nix/blob/master/test/sys/test_socket.rs (test_scm_rights)

[pusateri]

nix-rust/nix#990 includes support for incoming interface index.