Null-stuffing attack on SSL certificates

There is much news lately about the Null-Stuffing attack to SSL presented at BlackHat by Dan Kaminsky, Len Sassaman and Moxie Marlinspike. Quick answer: Our current assessment is that we were “probably not” vulnerable to this particular attack.

Bug: there is a theoretical possibility to create a certificate for example:

A CA might check the domain, and end up accepting a POSITIVE on the second part only. Meanwhile, a browser might show the first part, because it is written to stop processing when it sees the NULL . This is because NULLs are special characters that might be interpreted as the end of string, or might not be, and a browser might mistake it one way while a CA another way.

Analysis: At one level this is “just” input validation, and both browsers and CAs should reject immediately. At another level, the code that manages this input is very complicated, because of the way certificates are built. Too many standards, layouts, encodings. Hence the comments by many that this bug is actually indicative of systemic weaknesses in SSL. Technically, implementing SSL properly means this isn’t possible, but the system is so complex that it isn’t easy to rule out these sorts of issues. But, we don’t win anything if we pass the blame onto someone else, because we’ve still got the bug. Basic technical conclusion is that we need to check our inputs carefully, and hope that others do the same.

CAcert. So where are we at? Is CAcert vulnerable? This boils down to whether CAcert can issue a certificate with a NULL stuffed into a domain name. Which is in two parts: adding a domain name to your account, and sending in a request for a certificate (CSR). Investigations are on-going, but here is a status update:

  • Adding domain names is now covered with a quick fix that was patched in Friday.
  • CSRs were already being filtered on NULLs, so someone was alert back in the earlier years!

What is outstanding is checking that the database copy of the domain name is used instead of the CSR, and scanning the database for any NULLs. It’s still not entirely clear if there was a way to sneak a NULL through before that patch in, but it’s covered now. This work is ongoing, and updates will be reported here.

Leave a Reply