- Chris Gordon <chris.gordon@gettyimages.com> has been watching DNS traffic at www.dshield.org and was wondering if "something was coming" and wanted to know if I had seen anything to indicate a DNS worm or virus was propagating. Chris, I have not noticed anything along those lines but all I did was actively scan DNS servers and process the responses, I did not sift through arbitrary Internet DNS traffic. - Bill Manning <bmanning@ISI.EDU> did not find the paper "particularly new or that interesting". He thought it reinforced work done over the last six years on the vulnerabilities in the installed base of DNS code. - Robert Brockway <robert@timetraveller.org> agreed with the overall statement of the paper 100%. "Somewhat OT for your discussion but it is high time for organisations to realise why they need geographically & logically seperated DNS servers. The number of organisations with 1 DNS server, all the servers on the same subnet, or lame delegations is disgraceful. In the end DNS security must rest on a properly configured DNS system." - Kurt Seifried <kurt@seifried.org> found that the paper agreed with his results: "This pretty much parallels the results I got when I did some checking into government DNS certains for a large country. I was able to do zone transfers for something like 60% of the subdomains (with some interesting results, like test-oracle-server.foo), bind versions were all over the map, and most were poorly secured if at all, to say nothing of the classic "all servers on the same subnet" for a few of the larger subdomains. I had them contacted, still no change. Sigh." - Nicholas Weaver <nweaver@CS.berkeley.edu> pointed out: "The roots really aren't vulnerable to a DDoS: Yes they are a single point, but they handle such little real traffic (mostly garbage) and the responses are cached for a long time. It is the gTLDs (eg, the .com nameservers) which are vulnerable to a DDoS, and the DDoS would probably be a traffic load related attacks." - Nuzman <nuzman@shreve.net> wrote "One thing that many corporations still overlook is diversity in DNS. Remember Microsoft getting knocked off because their DNS servers were all on one subnet (early 2001)? I did a survey recently of the largest businesses in WI (whois on domain name) and almost half had DNS all in the same subnet... even companies that I know have good multi-path Net access. Heck, even adding something like granitecanyon.com as a 3rd and/or 4th DNS server would be an improvement for some businesses. One thing I'd be interested in seeing... what's the penetration of non-BIND DNS out there? The company I work for is a MS shop and we use Win2k DNS for primary and Sprint for additional secondary." And last, but not least, David Conrad <david.conrad@nominum.com> of Nominum: "Cute title. In no particular order: 1) You appear to make a big deal out of number of lines of code implying increased vulnerability, but the data you provide shows the opposite -- BINDv9 with 300,000+ lines of code has fewer vulnerabilities than BINDv8 (v2 in particular) with half the lines of code. Note that these code estimates are most likely misleading as they appear to include the entire source tree and BINDv9 has extensive tests that BINDv8 or 4 never had. 2) Several non-BIND DNS servers respond to CHAOS TXT queries for version.bind as if they were BIND. To get an accurate assessment of the servers running, more elaborate and sophisticated fingerprinting is necessary. 3) Verisign does not run all the root servers, only two, one of which runs Atlas last I heard. The do run all the .com/.net gTLD servers. I believe two are running Atlas now. 4) There are many other DNS servers available today, not just djbdns. NSD, PowerDNS, MaraDNS, and Posadis, are 4 open source implementations. Nominum's ANS and CNS, Microsoft Win2K (and .Net or whatever it is called today) DNS, Incognito's DNS Commander, and Cisco's CNR DNS server are proprietary commercial implementations available for purchase. 5) BINDv9 has never had a arbitrary code executable buffer overflow exploit unlike BINDv8 or BINDv4. It has, however, has had denial of service vulnerabilities until the 9.2 series, most of which do not appear on ISC's web page. The 9.0 series, in particular, was susceptible to remote denial of service 'packets of death'. 6) BIND 8.2.7 has no known vulnerabilities so it should be classified as 'safe'. The difference between the 8.2 series and the 8.3 series is primarily v6 support in 8.3. 7) "Klaatu, Barada, Nikto" is actually from the 1950s movie "The Day The Earth Stood Still". Sam Raimi stole the line for "Army of Darkness" (and other projects he has done) 8) Your section title "Remediation" makes several assertions without data to back up those assertions: * "Poor programming is obviously the main issue enabling the vulnerabilities" -- you provide no data that demonstrates poor programming. An assertion along the lines of "attempts to integrate code from a wide variety of sources in the traditional open source fashion is the main issue enabling the vulnerabilities" would probably be more accurate. * "BIND ... is a perfect example of what happens when security is retrofit as opposed to designed into the product ..." -- you have not documented a basis that there was an attempt to retrofit security into the product. 9) Bill Manning at ISI runs a periodic survey of BIND versions and has been doing so since 1996 or so. Stating your report "is the first to present substantive proof quantifying just how vulnerable" the DNS infrastructure is ... a bit of a stretch. 10) You mention the root DDoS attacks but they are unrelated to BIND. The attacks didn't even use DNS packets. 11) BIND version 4 continues to get security patches. It is currently at version 4.9.11 (last I looked). 12) It is a bit misleading to say djbdns has no security vulnerabilities. While it is true that the component programs that make up djbdns have not had a known vulnerability, the design of djbdns relies on external services (Bernstein recommends rsync over ssh, I believe) to replicate data from the primary to secondaries. A vulnerability in these external services, mandatory for (the equivalent of) normal zone maintenance data replication with djbdns, would be at least as damaging as a vulnerability in the djbdns package itself. However, it makes it much easier to offer 'security guarantees' since large chunks of functionality are not covered under the warranty (so to speak). There have been vulnerabilities in ssh since djbdns was released. 13) Stating "BIND is mature" is misleading as BINDv9 was a complete, from the ground up rewrite of BIND sharing no code (except for an optionally compile backwards compatibility stub resolver library that does not link into the server) with BINDv8. BINDv4 could be called mature. BINDv8 is arguable. The large jump in lines of code for 8.2 was a result of integration of code from external parties (Intel, Checkpoint, and NAI to name three). Clearly, given the number of lines of code doubled, the maturity of the code base was reset." -- Mike Schiffman, CISSP http://www.packetfactory.net/schiffman.html
