VAvasilia·Aug 10, 2019
No, it's not full implementation. The main difference between this and Zalewski implementation is using direct mapping data between kernel space and userspace. Zalewski implementation uses pcap for capturing packets. If you have huge workloads you can try this https://gist.github.com/xpahos/4bc215aa5fe237161bc076f247643... to try multithreaded capturing packets(https://github.com/the-tcpdump-group/libpcap/pull/674).
My implementation creates a hash map between kernel space and userspace where eBPF(kernel space, directly from sk_buff) code stores parsed data in this map. After accepting a connection in userspace application could check this hash map and get data from kernel space to userspace. So it's possible to add p0f extra data to request headers directly inside load balancer/web server and pass it to underlying levels.
> Does anybody
Read the rest
have an p0f3 fingerprint files later than those that Zalewski put in the distribution? They're 3+ years out of date at this point,
I don't know. Another team uses fingerprinting to check if it's a robot.