[tor-talk] End-to-end correlation for fun and profit

Maxim Kammerer mk at dee.su
Mon Aug 20 07:33:29 UTC 2012

Hello gentlemen,

Here and there I see references to “global” or “state-level” powerful
adversaries when it comes to end-to-end traffic correlation — i.e.,
it's supposed to be very hard. Because Tor network has many nodes,
there are guard nodes, there is research, blog posts, CIA funding
(well, not anymore, but similar funding from EU is in the works),
useless bureaucracy, college kids playing in serious development, yada
yada — you know the drill.

Anyway, let's do some math. Below, you will find a table where left
column denotes the number of Guard+Exit+Fast+Stable Tor relays one
needs to sniff at Class-C level, and right column denotes the
probability that a given circuit will go through both intercepted
entry and exit nodes. This is slightly imprecise, because same node
can't be both entry and exit for a circuit, and there are other
ignored intricacies (e.g., port policies) that push the estimates in
the other direction — the reason is that I am better with writing
quick scripts [1] than with Excel. The consensus taken for analysis is
from a few hours ago, and I read Tor server code from current stable
version in Gentoo ( — this probably doesn't matter.

10 11.50%
11 14.56%
12 16.52%
13 16.80%
14 17.69%
15 17.98%
16 18.90%
17 19.20%
18 19.50%
19 20.46%
20 20.46%
21 21.76%
22 22.77%
23 23.43%
24 23.43%
25 24.48%
26 24.48%
27 24.82%
28 25.55%
29 25.90%
30 25.90%

As you can see, sniffing just 25 Class-C networks (or 42 individual
nodes) lets an adversary correlate ~25% of (non-.onion) circuits.
Which networks are these?

DE 31.172.30.[1-4]
NL 77.247.181.{162,164}
RO 109.163.233.{200,201,205}
US 199.48.147.{35,36,37,38,39,40,41}
FR 178.32.211.{130,140}
US 173.254.216.[66-69]
DE 88.198.100.{230,233}

All of these servers are in US/CA or EU jurisdiction, so even an
unsophisticated LE operation can issue ~20 wiretapping orders at ISP
level (many of these networks are operated by same hosting providers),
and immediately deanonymize ~25% of Tor traffic. So far for anonymity!

Oh, and if you are just into looking what sites Tor users visit, the
situation is even better — intercepting the same 25 Class-C networks
will let you see 72% of the traffic. Picking better non-Guard Exits
will improve this figure to 78%. That's right — 4/5th of Tor traffic
exits through just 25 LANs.

[1] http://pastebin.com/hgtXMSyx

Maxim Kammerer
Liberté Linux: http://dee.su/liberte

More information about the tor-talk mailing list