News that two Carnegie-Mellon CERT researchers have developed an inexpensive way to breach the Tor network has the project, privacy advocates, and probably criminals who use the network equally concerned.
The Tor Project posted has advised relays to upgrade to Tor 0.2.r.23e or 0.2.5.6-alpha to close the protocol vulnerability used by the researchers, but It warned that preventing traffic confirmation in general “remains an open research problem.”
Hidden service operators should consider changing the location of their service, the Tor Project said.
“So much for being secure,” remarked Jim McGregor, principal analyst at TIrias Research.
“If you were using Tor for classified communications and data, this could be very serious,” he told TechNewsWorld.
What the Tor Project Found
On July 4, the Tor Project found a group of relays that were trying to deanonymize people who operate or access Tor hidden services by modifying Tor protocol headers to conduct traffic confirmation attacks.
The attack also probably tried to learn who9 published hidden service descriptors, Tor said.
This would let the attackers learn the location of hidden services and, in theory, link users to their destinations on normal Tor circuits, although this was unlikely because the operators did not operate any exit relays.
The attack might aid other attackers in deanonymizing Tor users, the project cautioned.
Technical Details of the Attack
The attackers are believed to have used a combination of a traffic confirmation attack and a Sybil attack.
In a traffic confirmation attack, the attacker controls or observes the relays on both ends of a Tor circuit and compares traffic timing, volume or other characteristics to discover whether the two relays are on the same circuit.
If the first relay in the circuit, also known as the “entry guard,” knows a user’s IP address and the last relay knows the resource or destination being accessed, the user can be deanonymized.
There are several varieties of confirmation attacks; the one used consisted of the attackers injecting a signal into the Tor protocol headers at the relay on one end and having the relay on the other end read the signal.
That let the attackers obtain the HSDir (“suitable for hidden service directory”) and Guard (“suitable for being an entry guard”). The attackers then injected the signal whenever it was used as a hidden service directory and looked for an injected signal whenever it was used as an entry guard.
The Sybil attack was standard. The attackers signed up 115 fast non-exit relays running on either of two IP addresses: 220.127.116.11/16 or 18.104.22.168/16. These added up to about 6.4 percent of the Tor network’s Guard capacity, and they became entry guards for “a significant chunk” of Tor users over the five months they were in operation, Tor said.
Was the NSA Involved?
They were scheduled to present their findings at the Black Hat security conference, to be held in Las Vegas in August, but they canceled the presentation.
Richard Lynch of the Carnegie Mellon Software Engineering Institute, which runs CERT, demurred when approached for comment.
“Sorry, but we’re not able to comment on Tor,” he told TechNewsWorld.
Carnegie-Mellon CERT boasts of partnering regularly with government and law enforcement, which has given rise to speculation that the NSA or U.S. law enforcement agencies may have been behind the attack on Tor.
“That was the first thing that came to mind,” McGregor said. “Who better than the government to attack Tor?”
On the other hand, disclosure of the attack would have worked against the interests of law enforcement and the NSA, Rob Enderle, principal analyst of the Enderle Group, told TechNewsWorld.
Announcing the breach “leads to people putting resources into monitoring this kind of attack,” he said, “improving the response time, and moving to something that could be more difficult to penetrate or that’s less well known.”