Vendors form alliance to fix DNS poisoning flaw
Giving the good guys a headstart
While Kaminsky and other internet protocol experts who discussed the issue on Tuesday would not give specific details of the flaw, a CERT vulnerability note described the issue as a combination of DNS weaknesses. While the CERT note referred to the issue with a single Common Vulnerabilities and Exposures (CVE) identifier, Microsoft, in its security bulletin, referred to the issue as two flaws: a DNS socket entropy vulnerability and a DNS cache poisoning vulnerability.
The CERT vulnerability note describing the issue lists more than 90 software developers and network equipment vendors that may be affected by the issue.
The coordinated response was anything but assured six months ago.
Kaminsky began contacting a small group of domain-name system (DNS) experts and software vendors, resulting in a brainstorming session at the end of March.
On March 31, sixteen internet and security experts met on Microsoft's campus in Redmond, Washington. The agenda was simple, according to Kaminsky: Decide if they properly understood the problem, figure out how to fix the issue, and set a timetable for release.
"We decided that the only way to do this would be a simultaneous release - Microsoft patches, Sun patches, BIND patches" all at the same time, he said.
Another problem that vexed the response team was that in many cases, researchers can use a patch to figure out the underlying vulnerability. But because the security flaw was a design issue, the group had options that could fix the problem in a way that did not spotlight the issue, Kaminsky said.
"This is the fundamental balancing act between how do we notify the good guys without bringing on the bad guys," he said. "We tried to give the good guys as much of a nonlinear advantage as possible. We think we gave them a month."
The solution implemented in the patches is to inject additional randomization into the domain name system (DNS) by randomizing the source ports used in DNS queries. The only permanent solution is to add authentication using a security-enhanced version of the protocol, such as DNSSec, but that proposal is bogged down by worries over adding costs to the name-server system.
Kaminsky's efforts to keep the issue secret until a patch appeared garnered praise from Jerry Dixon, former director of the National Cyber Security Division at the Department of Homeland Security.
"This really shows the value add of independent researchers and the research community helping to make the internet more secure," said Dixon, who is now working with Team Cymru.
Kaminsky asked for the other researchers to show good judgement and not to release additional details of the flaws, if they find them.
"I'm making a request of the open-research community," he said. "Let's see if we can get the good guys fixed (first)."
For those researchers who believe they have pinpointed the problem: Kaminsky says to send him a note, and he will buy you a beer.
If you have tips or insights on this topic, please contact SecurityFocus.
This article originally appeared in Security Focus.
Copyright © 2008, SecurityFocus
Sponsored: RAID: End of an era?