On May 9, 2013, at 7:32 PM, Jon Lewis <jlewis@lewis.org> wrote:
On Thu, 9 May 2013, Jared Mauch wrote:
Some interesting data: about 46% of the IPs that respond to a DNS query do not respond from port 53, meaning they are "broken" in some interesting way.
Maybe I'm not being very imaginative, but how can something from !53 be considered a DNS response to a query sent to port 53? Can you give some examples of the sorts of packets that fall into this rather large % of ill-behaved hosts? Are you sure you're not treating things like icmp port unreachable as a "!udp/53 src response"?
IP1:Port:IP-Probed:Responding-IP:time_t:RCODE:RA:CorrectAnswerInPacket Here's a sample excerpt: IP1:14474:122.177.40.2:NULL:1367712184.690540:0:1:1 IP1:10316:123.26.39.2:NULL:1367712184.690683:0:1:1 IP1:15218:5.11.41.2:NULL:1367712184.691114:0:1:1 IP1:21388:186.31.41.2:NULL:1367712184.691402:0:1:1 IP1:11161:87.21.41.2:NULL:1367712184.691693:0:1:1 IP1:23884:88.249.40.2:NULL:1367712184.692264:0:1:1 IP1:12707:77.51.41.2:NULL:1367712184.692833:0:1:1 IP1:16290:190.86.41.2:NULL:1367712184.693118:0:1:1 IP1:10169:151.48.41.2:NULL:1367712184.694703:0:1:1 IP1:20885:112.209.40.2:NULL:1367712184.694992:0:1:1 I have the raw packet data for these. They were on a UDP socket, not some tcpdump output parsing snafu… :) I have many more of these in the dataset. I'm thinking about flagging those that aren't from udp/53 and giving a pointer to things like CPE device firmware that causes problem. I've got a lot of private data on that which I can't share, either because the vendor is delivering fixed firmware or something else. - Jared