Flaws in Tor anonymity network spotlighted

Fingerprinting encrypted traffic, that's it? This has been known probably since Tor was first introduced, or am I missing something?

Anyway, it's good that this receives more publicity so this can be finally fixed. There's a speed/overhead trade-off, that's probably why the devs didn't bother.

 

The key factor in the report (for home users) is that "the flaw" exists solely inside a local network, i.e. most home users of Tor need not worry about, unless they provide services of an open (unprotected) Wi-Fi local network to their neighbors, or participate in such.

However, the "other" thing for everyone to worry about when using the Internet is how ubiquitous the use of deep packet inspection becomes by ISPs - which only slows the overall use of the Internet in general (if not implemented efficiently).

-- Tom

 

What can deep packet inspection reveal if you're using Tor?

 

Say you browse a site with 23 jpgs, 3 gifs, a css, a js file all with certain sizes, how many sites do you think look exactly that way?
An attacker connects to certain web pages which he wants to monitor through tor and notes how many packages are sent, what sizes, what order and timing etc, that's what I called the fingerprinting.
It's never 100% but one can deduce with varying degree of probably whether someone is connecting to a certain website or not, despite encryption.

Mitigation: block images, adblock, scripting, trackmenot, load several pages at once, run a node.

 

Everything - that is if it can be decrypted by the ISP assuming it is saved and given a government/legal request and enough computation resources, quite possible. However, for the common user that is not in either the government's or ISP's radar it is probably unlikely that it will happen unless deep packet inspection is turned on for everyone using the ISP which IMO would be an ineffective way for the ISP to operate its resources.

-- Tom

 

Nothing because the entire stream is encrypted.

lolwut? Who exactly is going to be served the warrant?

 

Some cryptanalysis and brute forcing is certainly possible, but that's got nothing to do with "DPI". DPI means real time and you can't do that against 128bit AES and certainly not with hundreds of thousands of connections at ISP level. Not a resource problem, but _impossible_ without resorting to alien tech or conspiracy theories.

Specs:
https://gitweb.torproject.org/tor.git?a=blob_plain;hb=HEAD;f=doc/spec/tor-spec.txt

Besides, even if it was possible, IX level snooping is much cheaper and just as effective (completely breaking Tor today, against a dedicated international attacker, like your government).

 

What is the IX level, can you give a reference point?
Home user>ISP>IX>Gov

Why is it capable of breaking TOR today?

Will Tor Routers, the Buffalo/TOR partnership, affect the effectiveness of this attack (Ronjor's article)?

Apparently the router will operate as a bridge TOR node by default, who will and won't be able to snoop if this becomes widely adopted?

 

Apparently, you forgot to mention Deep Packet Capture - maybe not, but it is a part of deep packet inspection that exists to which I was referring.

-- Tom

 

How do you propose to use DPI or DPC when the stream is encrypted?

 

I was well aware of DPC but nope, it's not a typical _part_ of DPI.

DPC doesn't change the fact that breaking AES-128 is still a hard problem which even if possible with a large federal number-crunching "botnet" is still impossible to do on any larger scale.


Searching_ _ _
If an attacker sees all traffic he'll know who is connected to whom unless we add random padding, Tor doesn't. If they did they'd solve both problems...
By IX level I mean someone who controls/can observe several or all IXPs. If all 3 nodes (actually entry+exit are enough) run across monitored IXPs Tor is easily broken by correlating and timing the encrypted connections. This can be done in real time with DPI on a large scale (i.e. everyone) and it's most likely done in the US (of course affecting pretty much everyone else as so much is routed through or to them).

 

First capture the total stream - say for all Tor nodes during a set period of time. Then correlating and timing the encrypted streams would pare down the domain of inspection to several cases. Then for each case, do the following:

A large multiprocessor shared memory system with lots of cores running cracking software should be able to "eventually" crack the encryption, say using rainbow tables in shared memory would be my guess where the cracking software is modified to deal with network traffic that is encrypted. There may be other cracking software related to network traffic streams that no one but the network security specialists at NSA know about - I would expect exists, and if it does not now exist, I expect it will in the all-to-soon future.

-- Tom

 

Capturing the stream of ALL Tor nodes would be a monumental task, even for NSA.

And the timing attacks you mention are not new and the Tor evelopers are well aware of such possibilities. Moreover, these attacks are really only effective if there are "suspects" that are already being monitored.

Nah, it would take more computing power than exists in this solar system to crack a 128 bit cipher by brute force. Do the math.

Perhaps NSA has a mathematical attack against AES and/or RSA (doubtful but possible). If that's the case, then they are already reading everything anyway, so Tor wouldn't really be secure in the first place.

 

If an ISP is performing DPI, what would they see if a person was running a VPN, like Xerobank, and then connected to Tor in Firefox? I assume that they would only see the connection to the VPN and would be unable to see a connection to Tor. Is this correct?

 

Yes, in your scenario they would see you connect to the VPN node and then everything would go "dark."

 

Thanks for confirming that. I was just thinking of another interesting connection. I don't know why I think this is so funny. I guess I'm easily amused. But if a person buys one of those Tor routers, then fires up a VPN, and then connects to the internet with Tor's portable browser, everything would have to go to Tor first, then on to the VPN, and then a second Tor connection would be tunneling through the VPN. I wonder if that would work? And I wonder if the VPN service would mind? Because if I am thinking about this the correctly, the VPN would not know who was coming or where they were going......except that they were coming from one Tor connection and leaving through another, hehe! Isn't that nuts? They may not like that. But then again they really shouldn't even notice......unless they have some kind of program that alerts them to weird connections like that. But it would be interesting to try. I may some day if that Tor router becomes available.