Update: The video of [Moxie]‘s presentation is now online.

[Moxie Marlinspike] appeared on our radar back in February when he showed sslstrip at Black Hat DC. It was an amazing piece of software that could hijack and rewrite all SSL connections. The differences between a legitimate site and the hijacked ones were very hard to notice. He recently stumbled across something thing that makes the attack even more effective.

If you apply for a certificate, the certificate authority looks at the common name on the form and contacts the domain owner. The CA ignores the subdomain. The trick is to drop in a null character in the subdomain. If you register, www.paypal.com[null character].thoughtcrime.org, the CA will contact the owner of thoughtcrime.org and issue the cert. When clients like Firefox use NSS to verify the cert, the null character causes them to think the certficate is valid for www.paypal.com because they stop at the null character. Even if the person examines the cert in their browser, it will show www.paypal.com.

Wildcards work as well. You could get a certificate for *[null character].thoughtcrime.org and appear as any site you want. [Moxie] has worked out ways to prevent certificate revocation and browser updates too. This new code will be part of sslsniff 0.6.

[Apologies for the odd notation. WordPress apparently strips null characters...]

A team of security researchers and academics has broken a core piece of internet technology. They made their work public at the 25th Chaos Communication Congress in Berlin today. The team was able to create a rogue certificate authority and use it to issue valid SSL certificates for any site they want. The user would have no indication that their HTTPS connection was being monitored/modified.

This attack is possible because of a flaw in MD5. MD5 is a hashing algorithm; each unique file has a unique hash. In 2004, a team of Chinese researchers demonstrated creating two different files that had the same MD5 hash. In 2007, another team showed theoretical attacks that took advantage of these collisions. The team focused on SSL certificates signed with MD5 for their exploit.

The first step was doing some broad scans to see what certificate authorities (CA) were issuing MD5 signed certs. They collected 30K certs from Firefox trusted CAs. 9K of them were MD5 signed. 97% of those came from RapidSSL.

Having selected their target, the team needed to generate their rogue certificate to transfer the signature to. They employed the processing power of 200 Playstation 3s to get the job done. For this task, it’s the equivalent of 8000 standard CPU cores or $20K of Amazon EC2 time. The task takes ~1-2 days to calculate. The tricky part was knowing the content of the certificate that would be issued by RapidSSL. They needed to predict two variables: the serial number and the timestamp. RapidSSL’s serial numbers were all sequential. From testing, they knew that RapidSSL would always sign six seconds after the order was acknowledged. Knowing these two facts they were able to generate a certificate in advance and then purchase the exact certificate they wanted. They’d purchase certificates to advance the serial number and then buy on the exact time they calculated.

The cert was issued to their particular domain, but since they controlled the content, they changed the flags to make themselves an intermediate certificate authority. That gave them authority to issue any certificate they wanted. All of these ‘valid’ certs were signed using SHA-1.

If you set your clock back to before August 2004, you can try out their live demo site. This time is just a security measure for the example and this would work identically with a certificate that hasn’t expired. There’s a project site and a much more detailed writeup than this.

To fix this vulnerability, all CAs are now using SHA-1 for signing and Microsoft and Firefox will be blacklisting the team’s rogue CA in their browser products.