Silk Road forums

Did you do anything illegal on that site?

Did you do anything that could link you to illegal activity (eg, mentioning Silk Road)?

If no, you'll be fine.

Actually I think it is trivial for introduction points to know which hidden services they are intro points for, they just need to as clients connect to the hidden service and see if they are the introductory point selected. It is the correlation that I am worried about not MITM. It makes the hidden service less safe actually, because now not only can the correlation attack happen if the attacker owns the hidden services entry guard(s), but it can also happen if it owns the hidden services introductory points. And the introductory points of popular hidden services change rapidly because they become DDOSed.

Do you have evidence for this? Intro points are used temporarily to establish a connection, but the bulk of the bandwidth is distributed through the rendezvous points. That's one of the reasons to use rendezvous points.

So if you compare introductory points to exit nodes to clearnet sites, in this way they would be roughly equivalent in anonymity if we assume churn time is roughly the same (new introduction point selected once every ten minutes or so).

That seems far fetched. Take this forum as an example. There are 100-150 concurrent users, but how many are accessing the forum for the first time at any moment? Probably not more than one every 10 seconds.  6 new users per minute, 60 in 10 minutes. That seems like a fair estimate. I'm sure three intro points can handle 6 or even 20 users per minute. Even if the main site has 3 times the traffic, it doesn't seem like they would get DDOSed.

But with connection to clearnet site, if a thousand users access it there are a thousand different possible exit node selections, and if one exit node selected is bad it only effects the client that selected it, most clients will not be using that exit node. But with introductory nodes there are only a handful of them that ALL clients connecting to the hidden service with in the same ten minute period have to pick from. Now if one of the clients is using a bad introductory node they are likely sharing that bad node with hundreds of other clients all trying to connect to the hidden service.

I don't think that's going to be more than a few hundred people, even on the busiest hidden service.

Of course, the attacker could fetch the descriptors of the popular hidden services and see if he connects to himself. In that case it would be provable.

I love discussions like this. They really get you thinking.

Thank you for the clarification. In the end of the day, How big is the risk for a TOR user to be identified? Assuming you are not giving personal info.

That's hard to quantify. What I can tell you is that Tor with TBB in its default configuration is the safest and most anonymous way to access clearnet sites, of all the options available on the Internet.

Accessing hidden services is safer than clearnet sites. Some argue that Freenet is safer than hidden services, but I have reservations about using Freenet and I2P because they are so small. It's easier to enumerate all IP addresses, and I'd rather not have my IP on some list, even if they can't prove what I'm doing. I'd rather mix in with the millions of Tor users and access hidden services instead.

Say you were to enter a site controlled by the attacker. Is that alone enough to reveal the true IP? Out of 1000 visits from TOR users, in how many cases would the attacker be able to reveal the true ip ?

From simply logging into the site? Assuming you don't provide identifying info, that alone won't deanonymize you. Again, there would have to be some specific attack, like the adversary tricking you into running malware, or selecting his relays for entry guards.

Quote from: astor on April 03, 2013, 05:09 pm

Quote from: kmfkewm on April 03, 2013, 04:42 pm

Although you also need to take into consideration that the attacker probably just passively watches destinations of interest, so in practice they don't likely need to own your exit node, only your entry guard. If there are 900 entry guards and the attacker owns 100 of them, they own 1/9 entry guards. You select three entry guards every month to two months, and every time you do the chances the attacker owns one of them is 1/3.

True, true. Although you will only be accessing the destination web site 1/3 of the time through that one entry guard, if the purpose of the attack is to identity person X on the destination web site, you only need to go through the bad guard once.

Sounds like it is actually quite easy for LE to identify a TOR user? How many TOR users in % do you think has been identified by LE?

Well, that attack isn't trivial. It would be incredibly difficult to spin up 100 relays without getting noticed. If LEA are running relays, it's most likely fewer than 10. Also, watching certain sites, like Google and Facebook, would be really hard, because they use distributed content delivery networks. A user in Seattle accesses a different server than a user in New York when they go to the same site. It might be useful for specific smaller sites, but if the attacker runs 10 relays, then the chances of picking 1 of them as 1 of your 3 entry guards is 1/30, or only 3%. And if you don't pick any of them, the attacker would have to wait 1-2 months for you to new pick news, with again only a 3% chance of pwning you. It's not an effective attack for identifying a specific person. More like, "among all the users of this site, I can find a few of them every few months".

Also, none of this applies to hidden services, since the attacker can't watch the other end, although there are different potential attacks there.

Although you also need to take into consideration that the attacker probably just passively watches destinations of interest, so in practice they don't likely need to own your exit node, only your entry guard. If there are 900 entry guards and the attacker owns 100 of them, they own 1/9 entry guards. You select three entry guards every month to two months, and every time you do the chances the attacker owns one of them is 1/3.

True, true. Although you will only be accessing the destination web site 1/3 of the time through that one entry guard, if the purpose of the attack is to identity person X on the destination web site, you only need to go through the bad guard once.

. The attacker hacks you and gets your IP address without breaking Tor
- I guess for this to happen the attacker would have to find and hack you outside of TOR, and therefor are not hacking you due to your SR activities, correct?

An attacker could hack you through a browser exploit. Even with transparent proxying of all connections, if he can get root privileges, he can disable Tor and your firewall. The best defense against this is an anonymizing middle box, a separate physical device from your main computer that runs Tor and transparently proxies all connections over the Tor network.

(What is zero days?)

Unpublished exploits.

B. The attacker is able to see your traffic enter Tor and arrive at its final destination
- What would it take for an attacker to accomplish this? And how would that work?

Run many malicious relays. The probability of sending your circuits through the attacker's nodes is roughly

Centry / Nentry * Cexit / Nexit

where

Centry = number of entry nodes run by attacker
Nentry = total number of entry nodes
Cexit = number of exit nodes run by attacker
Nexit = total number of exit nodes

Let's say the attacker spins up 100 entry nodes and 100 exit nodes. For the sake of simplicity, let's say no entry nodes are exits and vice versa. Currently there are about 900 total entry nodes and 900 exit nodes. Then the probability of getting pwned by the attacker is 100/1000 * 100/1000 = 1/100, or 1%.

That doesn't sound so bad, but consider that your Tor client builds new circuits every 10 minutes. If it chose from all entry and exit nodes, there would be a 50% probability of getting pwned after 8.3 hours of Tor use. That threat is mitigated by using entry guards. Instead of changing entry nodes every 10 minutes, your client changes them about every 2 months. So it takes 8600 times longer to accomplish this attack.