Tor honions

HOnions: Towards Detection and Identification of
Misbehaving Tor HSDirs
Amirali Sanatinia
Northeastern University
amirali@ccs.neu.edu
Guevara Noubir
Northeastern University
noubir@ccs.neu.edu
ABSTRACT
Over the last decade privacy infrastructures such as Tor
proved to be very successful and widely used. However, Tor
remains a practical system with a variety of limitations and
open to abuse [2]. Tor’s security and anonymity is based on
the assumption that the large majority of the its relays are
honest and do not misbehave. Particularly the privacy of
the hidden services is dependent on the honest operation of
Hidden Services Directories (HSDirs). In this work we intro-
duce, the concept of honey onions (HOnions), a framework
to detect and identify misbehaving and snooping HSDirs.
After the deployment of our system and based on our ex-
perimental results during the period of 72 days, we detect
and identify at least 110 such snooping relays. Furthermore,
we reveal that more than half of them were hosted on cloud
infrastructure and delayed the use of the learned information
to prevent easy traceback.
1. HONION GENERATION & DETECTION
We introduce the concept of honey onions (honions), to
expose when a Tor relay with HSDir capability has been
modified to snoop into the hidden services that it currently
hosts. In order to automate the process of generating and
deploying honions in a way that they cover a significant frac-
tion of HSDirs, we developed several tools. A key constraint
in this process was to minimize the number of deployed ho-
nions. This derives primarily from our desire to not impact
the Tor statistics about hidden services (specially given the
recent surge anomaly). By considering the number of HS-
Dirs (approximately 3000), we could infer that we need to
generate around 1500 honions to cover all HSDirs with 0.95
probability. We used 1500 honions per batch (daily, weekly,
or monthly) and could verify that 95% of the HSDirs were
systematically covered.
HOnion back end servers: Each honion corresponds to
a process that is running locally. The server behind hidden
services, should not be running on a public IP address, to
avoid de-anonymization [1], [3]. We also log all the requests
that are made to the server programs and the time of each
visit. Recording the content of the requests allows us to
investigate the snoopers’ behavior and intent.
HOnions generation and deployment schedule: To
keep the total number of honions small, we decided on three
schedules for their generation and placement, daily, weekly,
and monthly. The three schedules allow us to detect the
malicious HSDirs who visit the honions shortly (less than 24
hours) after hosting them. Since the HSDirs for hidden ser-
vices change periodically, more sophisticated snoopers may
1. Generate honions
ho
i
ho
j
2. Place honions on HSDirs3. Build bipartite graph
On visit, mark potential HSDirs
ho
j
d
i
d
i+2
d
i+1
d
i
d
i+1
d
i+2
On visit, add to bipartite graph
Figure 1: Flow diagram of the honion system.
wait for a longer duration of time, so they can evade detec-
tion and frame other HSDirs.
Identifying snooping HSDirs: Based on the visited hid-
den service, the time of the visit, and the HSDir that have
been hosting the specific onion address prior to the visit, we
can mark the potential malicious and misbehaving HSDirs.
Then, we add the candidates to a bipartite graph, which con-
sists of edges between HSDirs and the visited honions. The
analysis of this graph allows us to infer a lower bound on
the number of malicious HSDirs as well as specific snoopers.
Figure 1 depicts the architecture of the system.
HOnion Visit Graph Formation: In the following we
first introduce a formal model and notation for the Honey
Onions system. First, HO denotes the set of honey onions
generated by the system that were visited, and HSD the set
of Tor relays with the HSDir flag (so far referred to as HS-
Dir relays). The visits of honions allow us to build a graph
G = (V, E) whose vertices are the union of HO and HSD
and edges connect a honion hoj and HSDir di iff hoj was
placed on di and subsequently experienced a visit. G is by
construction a bipartite graph. We also note that each ho-
nion periodically changes descriptors and therefore HSDirs
(approximately once a day). However, a HSDir currently a
honion ho cannot explain visits during past days. Therefore,
each time a honion changes HSDirs we clone its vertex ho
to ho and only add edges between ho and the HSDirs who
know about its existence when the visit happened.
Estimation & Set Cover: Since each honion is simultane-
ously placed on multiple HSDirs, the problem of identifying
which ones are malicious is not trivial. We first formulate
the problem of deriving a lower-bound on their number by
finding the smallest subset S of HSD that can explain all
the visits. The size s of the minimal set tells us that there
cannot be less than s malicious HSDirs who would explain
the visits.
argmin
S⊆HSD
|S : ∀(hoj, di) ∈ E∃di ∈ S ∧ (hoj, di) ∈ E| (1)

(a) Daily Visits (b) All Visits
Figure 2: Plot of the visits to the honions.
Finding the smallest set S as defined by Equation 1, is
not trivial as one can easily see that it is equivalent to the
hitting set problem, which itself is equivalent to the set cover
problem, which is well known to be NP-Complete. However,
it can also be formulated as an Integer Linear Program. Let
x1≤j≤|HSD| be binary variables taking values 0 or 1. Solving
Equation 1, consists of finding integer assignments to the xj
such that:
min(x1,...,xHSD)
|HSD|
j=1 xj
subject to ∀hoi ∈ HO ∀j:(hoi,dj )∈E xj ≥ 1
2. RESULTS & DISCUSSION
We started the daily honions on Feb 12, 2016; the weekly
and monthly experiments on February 21, 2016, which lasted
until April 24, 2016. During this period there were three
spikes in the number of hidden services, with one spike more
than tripling the average number of hidden services. There
are some theories suggesting that this was due to botnets,
ransomware, or the success of the anonymous chat service,
called Ricochet. However, none of these explanations can
definitely justify the current number of hidden services. Our
daily honions spotted snooping behavior before the spike in
the hidden services, this gives us a level of confidence that
the snoopings are not only a result of the anomaly (Figure 2).
Rather, there are entities that actively investigate hidden
services.
Snooping HSDirs Nature and Location: In total we
detected at least 110 malicious HSDir using the ILP algo-
rithm, and about 40000 visits. More than 70% of these HS-
Dirs are hosted on Cloud infrastructure. Around 25% are
exit nodes as compared to the average, 15% of all relays in
2016, that have both the HSDir and the Exit flags. This
can be interesting for further investigation, since it is known
that some Exit nodes are malicious and actively interfere
with users’ traffic and perform active MITM attacks [4].
Furthermore, 20% of the misbehaving HSDirs are, both exit
nodes and are hosted on Cloud systems, hosted in Europe
and Northern America. The top 5 countries are, USA, Ger-
many, France, UK, and Netherlands. Figure 3 depicts the
spread and the geolocation of the malicious HSDirs.
HSDirs Behavior and Intensity of the Visits: Most
of the visits were just querying the root path of the server
and were automated. However, we identified less than 20
possible manual probing, because of a query for favicon.ico,
the little icon that is shown in the browser, which the Tor
browser requests. Some snoopers kept probing for more
information even when we returned an empty page. For
Figure 3: The global map of detected misbehaving HSDirs
and their geographic origin.
example, we had queries for description.json, which is
a proposal to all HTTP servers inside Tor network to al-
low hidden services search engines such as Ahmia, to in-
dex websites. One of the snooping HSDirs (5.*.*.*:9011)
was actively querying the server every 1 hour asking for a
server-status page of Apache. It is part of the functionality
provided by mod status in Apache, which provides infor-
mation on server activity and performance. Additionally,
we detected other attack vectors, such as SQL injection,
targeting the information_schema.tables, username enu-
meration in Drupal, cross-site scripting (XSS), path traver-
sal (looking for boot.ini and /etc/passwd), targeting Ruby
on Rails framework (rails/info/properties), and PHP Easter
Eggs (?=PHP*-*-*-*-*).
3. CONCLUSION & FUTURE WORK
In this work, we introduced honey onions (HOnions), a
framework for methodically estimating and identifying Tor
HSDir nodes that are snooping on hidden services they are
hosting. We propose algorithms to both estimate the num-
ber of snooping HSDirs and identify them. Our experimen-
tal results indicate that during the period of the study (72
days) at least 110 such nodes were snooping information
about hidden services they host. Based on our observations
not all snooping HSDirs operate with the same level of so-
phistication. For example, some do not visit the hosted ho-
nions immediately to avoid detection by daily honions, our
weekly and monthly honions can detect them. We believe
that behavior of the snoopers can be modeled and studied
in more detail. Furthermore, we reveal that more than half
of them were hosted on cloud infrastructure making it dif-
ficult to detect malicious Tor nodes. Furthermore, cloud
providers such as Vultr, even accepts payments in the form
of bitcoins, which prevents the traceback and identification
of misbehaving entities.
4. REFERENCES
[1] S. Matic, P. Kotzias, and J. Caballero. Caronte:
Detecting location leaks for deanonymizing tor hidden
services. In CCS, 2015.
[2] A. Sanatinia and G. Noubir. Onionbots: Subverting
privacy infrastructure for cyber attacks. In DSN, 2015.
[3] M. Thomas and A. Mohaisen. Measuring the leakage of
onion at the root: A measurement of tor’s. onion
pseudo-tld in the global domain name system. In
WPES, 2014.
[4] P. Winter, R. Köwer, M. Mulazzani, M. Huber,
S. Schrittwieser, S. Lindskog, and E. Weippl. Spoiled
onions: Exposing malicious tor exit relays. In PETs,
2014.

Tor honions

Recommended

More Related Content

Similar to Tor honions (20)

More from The Hacker News (10)

Recently uploaded (20)

Tor honions