Today I started trying to write a TCP stack. In Python. We will see if this is a good idea.
So. I’ve gotten a TCP handshake working. The way this goes is you send a SYN, then get back a SYN-ACK, then send an ACK.
I tried this TCP handshake code first:
dest = "google.com" source_port += 1 # We need to set a different source port every time ip_header = IP(dst=dest) ans = sr1(ip_header / TCP(dport=80, flags="S", seq=random.randint(0, 1000))) # Send SYN, receive SYN-ACK reply = ip_header / TCP(dport=80, seq=ans.ack, ack = ans.seq + 1, flags="A") # ACK send(reply) # Send ACK
This did NOT WORK. Upon inspecting Wireshark, it turned out that my machine was the problem: it was sending out a RST (reset) packet after I got a SYN-ACK packet back from Google. What’s up with that?
Well, I already have a network stack on my machine, and it was like “what’s this SYN-ACK packet? I didn’t ask for this!“. So it would just reset the connection.
Jari (who is amazing) suggested a workaround: set up a fake IP address and tell the router using ARP (previously) that I am the person with that IP address. Here’s the fixed version:
I think I could also use iptables here to tell the kernel to ignore those packets. But I’m currently afraid of iptables. So. This code worked much better!
# Set port & MAC address FAKE_IP = "10.0.4.4" # Use something that nobody else is going to have MAC_ADDR = "60:67:20:eb:7b:bc" # My actual MAC address # Broadcast our fake IP address srp(Ether(dst="ff:ff:ff:ff:ff:ff")/ARP(psrc=FAKE_IP, hwsrc=MAC_ADDR)) source_port += 1 ip_header = IP(dst=dest, src=FAKE_IP) # Set the source port to ans = sr1(ip_header / TCP(dport=80, sport=source_port, flags="S", seq=random.randint(0, 1000))) # SYN # ans is the SYN-ACK reply = ip_header / TCP(dport=80, sport=source_port, seq=ans.ack, ack = ans.seq + 1, flags="A") # ACK send(reply) # Send ACK pkt = ip_header / TCP(dport=80, sport=source_port, seq=reply.seq, flags="AP") / "GET / HTTP/1.1\r\n\r\n" # Send our real packet send(pkt)
That is my small amount of code for the day. I also spend a ton of time reading the UDP handling code from the 4.4BSD network stack. I do not yet have anything intelligent to say about that, but it’s pretty interesting.