Two researchers have developed a new attack on TLS 1.0/SSL 3.0 that enables them to decrypt client requests on the fly and hijack supposedly confidential sessions with sensitive sites such as online banking, e-commerce and payment sites. The attack breaks the confidentiality model of the protocol and is the first known exploitation of a long-known flaw in TLS, potentially affecting the security of transactions on millions of sites.
The attack, developed by Juliano Rizzo and Thai Duong, will be presented at the Ekoparty conference in Argentina on Friday, and, unlike many other attacks on TLS and SSL, it has nothing to do with the certificate trust model in the protocol. Instead, the researchers have developed a tool called BEAST that enables them to grab and decrypt HTTPS cookies from active user sessions. The attack can even decrypt cookies that are marked HTTPS only from sites that use HTTP Strict Transport Security, which forces browsers to communicate over TLS/SSL when it’s available.
The researchers use what’s known as a block-wise chosen-plaintext attack against the AES encryption algorithm that’s used in TLS/SSL. In order to execute their attack, Rizzo and Duong use BEAST (Browser Exploit Against SSL/TLS) against a victim who is on a network on which they have a man-in-the-middle position. Once a victim visits a high-value site, such as PayPal, that uses TLS 1.0, and logs in and receives a cookie, they inject the client-side BEAST code into the victim’s browser. This can be done through the use of an iframe ad or just loading the BEAST JavaScript into the victim’s browser.
After the BEAST agent is loaded, the second part of the tool, a network sniffer, looks for active TLS connections and then grabs and decrypts the HTTPS cookie, enabling the attacker to hijack the victim’s session with that site. Once that encrypted connection with the site is established, the victim can move off to another tab or do other things on the machine and the attack will still work. The attack forces the browser to load pages from the target site, and the tool then decrypts the first part of the request to the Web server, which includes the secure cookie. The researchers have the ability to decrypt those cookies from within SSL sessions, which essentially negates the confidentiality promise of the protocol.
The decryption process is fast enough that it’s likely imperceptible users, and the researchers said that in a targeted attack, they likely could steal the cookie from a specific site within five minutes of loading the tool. Rizzo and Duong said that their attack exploits a vulnerability in the TLS 1.0 protocol that has been known for quite some time, but was thought to be unexploitable.
“It is worth noting that the vulnerability that BEAST exploits has been presented since the very first version of SSL. Most people in the crypto and security community have concluded that it is non-exploitable, that’s why it has been largely ignored for many years. Our work has two contributions,” Duong said in an email interview. “We introduce a practical and efficient plaintext-recovery attack for that vulnerability. It’s an enhancement of something crypto people call ‘block-wise chosen-plaintext attack’. We present one application the attack: BEAST. BEAST focuses on SSL implementations on browsers which is HTTPS. BEAST works for most major browsers and websites.”
The researchers said that the browser-based attack is just one variant. They said they also could implement a similar attack against other services that use SSL, such as instant-messaging clients or VPNs.
“While other attacks focus on the authenticity property of SSL, BEAST attacks the confidentiality of the protocol. As far as we know, BEAST implements the first attack that actually decrypts HTTPS requests,” Duong said. “While fixing the authenticity vulnerabilities may require a new trust model, fixing the vulnerability that BEAST exploits may require a major change to the protocol itself. Actually we have worked with browser and SSL vendors since early May, and every single proposed fix is incompatible with some existing SSL applications.”
Rizzo and Duong are well-known in the security world for the research last year, also presented at Ekoparty, on the padding oracle attack on ASP.NET applications. That research prompted an emergency out-of-band patch from Microsoft. Opera already has implemented a fix for the TLS attack, and the researchers have been in touch with the other major browser vendors, but it’s not clear when their fixes will be available.
“Browser vendors are implementing a workaround to stop this attack but the real solution is switching to a new protocol,” Rizzo said.
A spokesman for Opera said that the gorup has found that the patch to fix this problem breaks a small number of sites and most users employ the browser’s default configuration, which is not vulnerable to this attack.
“In fact, we do have a patch as you alluded to, but we have not put it in the final release of the browser because, according to our testing, it breaks some sites. Since the vast majority of Opera’s users use the default security configuration and are thus not vulnerable, we decided to pull it at this time rather than risk site breakage,” Opera’s Thomas Ford said in an email.
