PacketFu::Capture / PacketFu::Inject / PacketFu::Packet#to_w

[lupine]

Bit of an architecturing one, this. I'm using PacketFu to implement a DHCP server on a system that should technically be a mere bridge (so a UDP/IP socket isn't an option).

I was using PacketFu::Capture to receive ethernet frames with DHCP in them, building a response frame, and calling frame#to_w("iface_name") to inject a response, and it works quite nicely. However, there are issues when the server it's running on is loaded - PacketFu::UDPPacket#to_w was taking perhaps three seconds to return. This is because it's creating a temporary PacketFu::Inject instance, which itself creates a brand-new Pcap instance, which is slow, and seems to scale with the number of open captures, or network interfaces, or something.

Suggestion: PacketFu::Session, wrapped around a Pcap instance which it passes into PacketFu::Capture and PacketFu::Inject instances? I'm working around it for now by making my capture instance more widely available, and using #stream.inject(frame.to_s) - so maybe it just needs documentation on #to_w that it's slow?

Personally, I find the capture/inject separation to be a bit artificial; if you're interested in a new interface that conserves Pcap sessions, let me know and I'll see if I can slap a pull request together.

[todb]

I find the capture/inject separation to be a bit artificial

It kind of is, and that piece does need a good, long look, especially under load. Most of the things I and others have used PacketFu for are injecting maybe one or two packets, once.

if you're interested in a new interface that conserves Pcap sessions

I love getting pull requests. Thanks! The whole notion of a "session" is kind of a big missing piece; PacketFu is packet-centric, after all, but it's high time to at least offer some kind of reference implementation of a session manager.