DNS Flaw Underscores Danger of Taking Web Security for Granted

LAS VEGAS—Perhaps more than any other flaw in the last several years, the DNS protocol vulnerability discovered by security researcher Dan Kaminsky has shown that the circle of trust on the Internet can be broken more easily than we feared.

After listening to Kaminsky’s talk Aug. 6 at the Black Hat conference here, it is clear the flaw extends far beyond DNS (Domain Name System) cache poisoning. As he explained later, it is a game of dominoes—one domino could be redirecting Web traffic to malicious sites, the next could be interception of sensitive corporate e-mail. The possibilities are numerous and problematic.

“I spent the last month terrified of large companies having all their e-mail stolen because of a bug that I found,” Kaminsky, director of penetration testing at IOActive, told a group of journalists after his session.

Vendors worked together to coordinate a release of a patch in July. If the figures offered by Kaminsky are any indication, the number of companies now protected is significant.

But fundamentally, the flaw means the level of security we have traditionally taken for granted on the Internet may not always be there. It is possible for an attacker to be the man-in-the-middle. In total, there are 15 ways of running the attack that Kaminsky and others admitted knowing about, but the researcher added there were likely others as well.

Though details of the flaw leaked out some time ago, the scope of the vulnerability was shown by Kaminsky to be much larger than first indicated. The flaw basically allows hackers to poison the cache of any vulnerable recursive DNS server, which in turn allows attackers to send users to malicious Web sites. But as Kaminsky revealed in his talk, attackers can use the flaw to target a number of applications and protocols, including the FTP and SSL (Secure Sockets Layer) certificates validated via the Web.

Many argued during the past few weeks that SSL certificates offered a level of protection against the issue.

"SSL's not the panacea we would like it to be," he said during his session.

The stopgap answer to all this is port randomization, which decreases the odds of successfully guessing the transaction ID of any request to a domain server for a particular domain from one in roughly 65,000 to between one in 163 million and one in 2.1 billion. But Kaminsky himself said that is not the final solution.

“If I had the answer for what the correct fix would be, we probably would have gone out with it,” he said, adding there are lots of proposed solutions such as DNSSEC (Domain Name System Security Extensions) that fall short.  “There’s lots of proposals on the table, all of which have side effects that ... if we had gone out with on day one, would not have worked on day one."

Note: A previous version of this story stated incorrectly port randomization increased the odds of guessing the transaction ID. Port randomization decreases the odds.