I don't expect to can see a perfect DoS solution in my lifetime;-) As we known, there are tons of commercial gateway-level boxes can mitigate the DoS attack in some ways. But I prefer the combination of x86+GNU/Linux, like the most old school guys. Why? My answer is simple: It's the fuc*ing cheapest solution we have. SYNPROXY is one of new features of linux kernel 3.13. It's based on netfilter framework and connection tracking. If I understand correctly, SYPROXY should mark the initial SYN packet as UNTRACKED and redirecting them into iptables's action "SYNPROXY"( like ACCEPT, DROP, NF_QUEUE, etc). SYNPROXY would be acting like a network gateway device( router?) to performing the regular TCP x-way handshakes. The original packet will be passing into the dst when handshake process is finished. The contributor Jesper Dangaard Brouer gave us a free speech at DEVCON last month. According to his slide's test result, these numbers are really looking good. I did a little test with my colleague today.
Platform: Debian, SLES-12-beta2
Hardware: Laptop, Server, 100Mbps Switch
Tools: hping3, metasploit
root@d6-test:/home/shawn# iptables -t raw -A PREROUTING -i eth0 -p tcp --dport 8888 --syn -j NOTRACK
root@d6-test:/home/shawn# iptables -A INPUT -i eth0 -p tcp --dport 8888 -m state --state UNTRACKED,INVALID -j SYNPROXY --sack-perm --timestamp --mss 1480 --wscale 7 --ecn
echo 0 > /proc/sys/net/netfilter/nf_conntrack_tcp_loose
Without SYNPROXY: ksoftirq is around 8%-9%
With SYNPROXY: ksoftirq is less than 3%
btw: This result may not be very accurate. Anyway, SYNPROXY works.