Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Mitigating DoS Attacks against Signature-Based Authentication in VANETs

Li He and Wen Tao Zhu


State Key Laboratory of Information Security
Graduate University of Chinese Academy of Sciences
19A Yuquan Road, Beijing 100049, P. R. China
Email: lhe@is.ac.cn & wtzhu@ieee.org

Abstract—Vehicular ad hoc networks (VANETs) are supposed concerned about their privacy, and thus it is necessary to
to improve traffic safety and drivers’ experiences. In a typical protect drivers’ privacy information. Moreover, safety mes-
VANET, a vehicle broadcasts safety messages to its neighbors. sages in VANETs must be processed with strict time con-
Since behaviors based on safety messages could be life-critical, straints due to vehicles’ high mobility.
authentication of these messages must be guaranteed. Many Lots of schemes have been proposed to achieve authenti-
signature-based schemes have been proposed for authentica- cation in VANETs. They can be classified into two catego-
tion in VANETs, but few of them have addressed the problem ries. One is based on symmetric cryptography, such as the
of denial of service (DoS) attacks against signature-based au- Timed Efficient Stream Loss-tolerant Authentication
thentication. In such a DoS attack, attackers can broadcast (TESLA) protocol [2]. The other is based on asymmetric
forged messages with invalid signatures to force the receiving
cryptography, such as the Elliptic Curve Digital Signature
vehicles to perform lots of unnecessary signature verifications,
and thus the benign vehicles cannot verify the messages from
Algorithm (ECDSA) [3], the identity-based signature algo-
other legitimate vehicles. Our scheme features a pre- rithm [4] and the group signature algorithm [5]. In this paper,
authentication process before signature verifying process to we only focus on the authentication schemes based on
deal with this kind of DoS attack. The pre-authentication asymmetric cryptography. In other words, we are concerned
process takes advantage of the one-way hash chain and a about the signature-based schemes. However, our scheme is
group rekeying scheme. Evaluations show that the proposed not limited to any specific signature algorithm.
scheme mitigates such DoS attacks effectively. An outside attacker can launch a DoS attack by repeated-
ly broadcasting forged messages with invalid signatures to
Keywords-vehicular ad hoc network; security and privacy; consume the benign vehicles’ computation resource and pre-
authentication; denial of service vent them from verifying the messages received from other
legitimate vehicles. Under such attacks, VANET loses a part
I. INTRODUCTION or all of its ability to provide services. We call this kind of
Over the past few years, vehicular ad hoc networks attack the DoS attack against signature-based broadcast au-
(VANETs) have attracted extensive attention from car manu- thentication. Since this kind of attack cannot be totally
facturers as well as wireless communication researchers. The avoided, our objective is to try to mitigate it as much as poss-
VANET is a self-organized network, which can be consi- ible.
dered as a special implementation of the mobile ad hoc net- The paper is organized as follows. Section II reviews the
work (MANET). VANET is formed with wireless communi- related work. Section III explains the models and involved
cation devices known as on-board units (OBUs) equipped in mathematics. Section IV discusses the details of the pro-
vehicles and fixed roadside units (RSUs). Vehicles can be posed scheme. Section V presents the security and perfor-
aware of the environment around them by communicating mance evaluations. Section VI concludes the paper and gives
with each other and with RSUs. VANETs are supposed to the future work.
improve traffic safety and drivers’ driving experiences. For
example, VANETs can provide some safety related services II. RELATED WORK
such as warning of cars’ collision, reporting dangerous con- In order to authenticate safety messages in VANETs, lots
ditions of the road and providing cooperative driving. Also, of schemes have been proposed.
it can provide some other services, e.g., the location-based Zhang et al. [6] introduced a novel RSU-aided message
services. authentication scheme called RAISE. According to RAISE,
According to the Dedicated Short-Range Communica- RSUs are responsible for checking the authenticity of the
tions (DSRC) standard, each vehicle broadcasts traffic re- messages by a message authentication code and notifying the
lated safety messages every 100-300 ms. The message con- results back to vehicles. Lin et al. [2] developed a timed effi-
tains information of the vehicle’s position, direction, speed, cient and secure vehicular communication (TSVC) scheme
acceleration, etc. The security and privacy issues in VANET to reduce the communication and computation overhead. The
must be carefully and properly addressed before it can be scheme made a small change to the TESLA protocol and
widely deployed [1]. Since behaviors based on the safety applied it to VANETs.
messages could be life-critical, the security of these messag-
es should be promised. Besides, people are more and more
This work was supported by the National Natural Science Foundation of
China under Grant 60970138.
___________________________________
978-1-4673-0089-6/12/$26.00 ©2012 IEEE


The two schemes above are based on symmetric crypto-
graphy, while more schemes are based on asymmetric cryp-
tography. Calandriello et al. [3] proposed that each vehicle
can generate an own set of pseudonyms and calculate a
group signature on each pseudonym to achieve the authenti-
cation in VANETs. Zhang et al. [4] introduced an efficient
batch signature verification scheme, in which an RSU can
verify a batch of identity-based signatures at the same time,
and thus the total time of verifying all the messages can be
reduced. Wasef et al. [5] employed an efficient group signa-
ture scheme supporting batch verification for VANETs,
which can also verify a large number of messages in a timely
manner. However, all these schemes have not considered the
DoS attack based on signatures.
Few of the studies have addressed the problem of DoS at-
tacks. Studer et al. [7] provided a hybrid authentication me- Figure 1. A vehicular ad hoc network formed by OBUs and RSUs
chanism that combined the advantages of ECDSA and a
modified version of TESLA to deal with DoS attacks. A trial B. Threat Model
period was suggested to reduce the DoS attacks based on
The TA and RSUs are trusted. Similar to the research ef-
signatures. Wasef et al. [8] suggested that all the vehicles in fort in [8], we only consider the outside attacker, who does
the network use the same group key to compute a hash-based
not have the secret group key shared between the TA and
message authentication code (HMAC) on the message and non-revoked vehicles. A vehicle with the current group key
send it with the message. The receiver would check the sig-
is considered to be legitimate. We assume that the legitimate
nature only after the HMAC is verified. vehicles can be trusted and always execute the protocol cor-
Some research efforts have considered the problem of
rectly. Once a vehicle is found misbehaving, it will be re-
signature-based DoS attacks in the wireless sensor network voked by the TA at once and have no way to get the new
(WSN). Ning et al. [9] presented an efficient mechanism
group key. RSUs can get the new group key through the
called message specific puzzle to mitigate such DoS attacks. communication with the TA and the details are omitted in
The mechanism added a weak authenticator in each broad-
this paper.
cast packet, which can be verified quickly by a regular sen- Since most of the proposed authentication schemes are
sor node but takes an attacker a large amount of time to forge. based on signatures, an outside adversary can forge lots of
Du et al. [10] used a sender-specific one-way key chain to bogus messages with invalid signatures and broadcast them.
defend DoS attacks against signature-based broadcast au- As a result, receivers will consume the computation resource
thentication in sensor networks. to verify the invalid signatures. If the forged messages domi-
III. PRELIMINARIES nate the messages to be verified, legitimate messages from
benign vehicles would be dropped since they cannot be veri-
A. System Model fied in time. This kind of attack is referred to as the DoS
As illustrated in Figure 1, a VANET consists of a trusted attack against signature-based authentication. We do not
authority (TA), mobile OBUs and fixed RSUs. The TA is in consider the DoS attack caused by channel jamming.
charge of a city’s OBUs and RSUs. The TA is responsible When the TA broadcasts revocation messages signed
for cryptographic operations like issuing credentials to with the TA’s private key, it first broadcasts them to the
RSUs/OBUs and when needed, revealing the real identity of RSUs (through the wired secure channel) and then RSUs
a misbehaving vehicle. An OBU (equivalently, a vehicle) broadcast them to OBUs. These messages are transmitted via
can communicate with other OBUs through vehicle-to- V2I communication and they are sent much less frequently
vehicle (V2V) communication and communicate with RSUs than the messages are sent in V2V communication. V2I
through vehicle-to-infrastructure (V2I) communication. We communication can be distinguished from V2V communica-
focus on V2V communication in this paper. In V2V commu- tion (maybe at the physical layer). As our focus in this paper
nication, we assume that an OBU broadcasts safety messages is V2V communication, we do not consider DoS attacks
to its neighboring OBUs every 300 ms and the receiving based on signatures of these revocation messages.
OBUs need to authenticate the messages. All the vehicles C. One-way Hash Chain
can be time synchronized via an external time reference.
We assume that a packet sent by a node is received by a The properties of the one-way hash function are as fol-
neighboring node before a third node can replay the packet to lows (H denotes the hash function):
it [11]. It means that when a vehicle broadcasts a message, x H(x) takes a message of arbitrary length as input and
its neighbors will receive the message before they receive a produces a message of fixed length as output.
copy forwarded (maybe modified) by any other vehicle. This x When given x, it is easy to compute y = H(x). But it
is true because of the triangular inequality. is hard to compute x = H-1(y) when given y.


x When given x, it is computationally infeasible to find B. The Pre-authentication Process
x ' z x satisfying H(x') = H(x). When a vehicle Vi wants to broadcast safety messages, it
x It is computationally infeasible to find a pair of x and first generates a hash chain from a random seed Si, where Ki,n
x' satisfying x ' z x and H(x') = H(x). = Si and Ki,l = Hj-l(Ki,j) ( 0 d l  j d n ), in which Ki,j is the j-
The one-way hash chain is generated as follows: th value of the i-th vehicle’s hash chain. Each value in the
chain (except the commitment Ki,0) will be attached to a safe-
 Kn1 H( Kn ),..., Ki1 H (Ki ),..., K0 H (K1),1  i  n  ty message. The commitment is contained in the commit-
ment message broadcast with an HMAC. The format of the
Where Kn is called the seed of the chain and K0 is called commitment message is as follows:
the commitment of the chain. When the chain is used, the
elements of the chain are revealed in a reverse order. In other  PIDi | 0 | Ki ,0 | Ti,0 | HMAC kg ( PIDi , 0, K i ,0 , Ti ,0 ) 
words, with the knowledge of the commitment value K0, a
verifier can authenticate K1, K2, …, Kn in the key chain by
Where PIDi is one of the sender’s pseudo identities, the
performing a few hash functions. If some of the intermediate
next value denotes the index of the element released in the
values are missed, the verifier can still check the authenticity
hash chain and 0 means that the value revealed in the mes-
of the received value Kj based on the last authenticated value
sage is the commitment of the chain, Ki,0 is the correspond-
Ki, since Hj-i(Kj) = Ki, where i < j.
ing commitment, and Ti,0 is the timestamp. The HMAC is
IV. THE PROPOSED SCHEME calculated using the group key kg shared between the TA and
non-revoked vehicles. The commitment message should be
In this section, we present the pre-authentication process broadcast periodically, where the broadcasting period in-
and the group rekeying scheme. The pre-authentication is volves a tradeoff between security and communication cost.
similar to the pre-authenticator in [10] and the traffic authen- On the receiver side, it first verifies the HMAC with its
tication in [11]. Since safety messages must be processed in own secret group key. After verifying successfully, the re-
strict delay constraints, we do not take the scheme of mes- ceiver maintains a record for the sender, which includes the
sage specific puzzles [9] as a solution because the generation sender’s PIDi, the corresponding commitment Ki,0, and the
of a puzzle would introduce a long delay at the sender side. timestamp of the commitment message (i.e., Ti,0).
We append a chain value of the one-way hash chain to the The format of the safety messages is as follows:
message instead of appending an HMAC to the message [8]
because it is computationally more efficient to compute a
chain value (a hash over a key of a small fixed size) than to  PIDi | j | M i , j | V i , j | K i , j 
compute an HMAC over the entire message.
Where j is the index of the revealed value in the hash
A. System Setup chain, Mi,j denotes the information of the j-th safety message
The TA issues each legitimate vehicle the corresponding of the i-th vehicle,  i , j is the signature of the j-th message,
credentials used for signing the messages. For example, the
certificates for the ECDSA scheme and the identities asso- and Ki,j is the j-th value in the hash chain. The timestamp Ti,j
ciated with the secret keys for the identity-based signature is included in the message Mi,j.
scheme. Notice that all the credentials are based on pseudo- Upon receiving a message, the receiver first checks
nyms, which will be used in communication for the purpose whether the time difference between Ti,j and the instant time
of privacy protection. Mi,j was received is less than a threshold to prevent the replay
The TA has a symmetric key pool P. Each vehicle Vi (i attack. Then the receiver checks whether it is the right time
denotes the i-th vehicle) is preloaded with a set of distinct to release the value in the hash chain by checking Ti,j = Ti,l +
keys Ri from the key pool, an initial group key kg, and a 300(j-l) ms (vehicles broadcast safety messages every 300
pseudo-random function f. Keys in Ri are used as encryption ms), where Ti,l is the timestamp of the l-th message of the i-th
keys , kg is used to calculate the HMAC on the commitment vehicle buffered in the record of the receiving vehicle. If the
message, and f is used to update the keys in vehicles. Each equation holds, it means Ti,j is the right time to release the j-
key in the key pool has a unique id. The vehicles can reveal th chain value. If both of the conditions hold, it will further
some key ids in certain situations but none of them is al- check whether Ki,j is the j-th element in the hash chain of
lowed to reveal the real values of the keys loaded in them. PIDi by performing a few hash functions since Ki,l = Hj-l(Ki,j)
The TA maintains a database recording each vehicle’s real (l<j). Only if all of the checking conditions are satisfied, the
identity along with its credentials and the preloaded key set. receiver will consider the pre-authentication process is suc-
It also maintains a revocation database about the revoked cessful and decide to verify the signature of the safety mes-
vehicles along with their corresponding information. sage; otherwise it stops verifying and discards the message.
In this paper, we denote id as a symmetric key’s identity The receiver updates the records with each PID’s hash
and ID as a vehicle’s identity. We define k as the symmetric chain value and the corresponding timestamp based on its
key preloaded in vehicles and K as the value in a hash chain. verified messages. Even if some of the intermediate messag-
es are lost, the receiver can still pre-authenticate Mi,j by


checking whether Ti,j is the time to release Ki,j and authenti- x When a vehicle has the data necessary to update the
cating Ki,j based on the property of the one-way hash chain. keys, it computes the new group key k g ' f kim (0)
If a receiver does not have the sender’s commitment
when it receives some safety messages from a sender, it may and verifies it by checking if f k g ' (0) is equal to the
choose to wait for the commitment message, or it can decide corresponding value received in the revocation mes-
to verify one of the signatures if it has pre-authenticated the sage.
consistency of the chain values in the messages (i.e., the x After calculating the new group key, each vehicle al-
chain values attached to the messages satisfy the property of so updates its symmetric key set. For example, Vw
the one-way hash chain). In the second situation, once the updates every key kp in Rw as k p ' f k p (0) . However,
signature is authenticated, the receiver records the corres-
ponding PID, chain value and timestamp of the sender; oth- if kp was held by the revoked Vu, the vehicles that
erwise it will refuse to verify the signatures from the same hold k p ' further update k p ' as k p ' f kim (k p ') .
sender until it receives a correct commitment message from x Finally, non-revoked vehicles erase kim and the orig-
the sender. In both situations, the receiver can pre- inal kp’s.
authenticate the sender in the future. However, the decision The revoked vehicle cannot compute k g ' since it does
is based on the receiver’s policy, which is a tradeoff between
security and efficiency. not have kz. Furthermore, it cannot get kim from neighboring
Once a sender changes its pseudo identity, it can create a vehicles because the revocation message contains the ids of
new hash chain and then process as before. its symmetric keys, which prevents it from having shared
keys with other benign vehicles.
C. Group Rekeying Scheme If a vehicle misses a group rekeying event, it is still able
When a vehicle Vu is revoked, its ID with all its creden- to update any keys that are not revoked (i.e., k p ' f k p (0) )
tials should be added to the revocation database. Besides, all and then obtains the current group key through the shared
the keys in Ru and the current group key kg are considered to keys with other non-revoked vehicles. As long as a vehicle
be revoked, too. Therefore, the non-revoked vehicles need to possesses some keys that have not been revoked, it can al-
update their symmetric key sets and the current group key. ways get the current group key even if it has missed some
Our group rekeying scheme is inspired by [12], which is rekeying events. However, if a vehicle has missed a number
based on the research effort of [13]. The processes are as of rekeying events and all of its symmetric keys are revoked,
follows: it has to turn to the RSU for the current group key and maybe
x The TA determines z, the id of the non-revoked a set of new symmetric keys.
symmetric key that is possessed by the maximum When a vehicle enters a new city (different from its reg-
number of the non-revoked vehicles. Then the TA istration city), it can get the current group key of the new city
calculates an intermediate key kim f k z (k g ) and a from the nearest RSU after it proves its legitimacy to the
new group key k g ' f kim (0) . After that, it broad- RSU (though V2I communication). As aforementioned in
section III-B, a vehicle with the correct group key is thought
casts a revocation message containing z, the ids of to be legitimate. Then the vehicle can use it to bootstrap its
the symmetric keys in the revoked vehicle, and trust with other vehicles. Here, we assume that it would not
f k g ' (0) . f k g ' (0) will be used for a vehicle to verify stay in the new city for a long time. Therefore, it just needs
to communicate with RSUs to get the updated group key but
the authenticity of the new group key k g ' that it cal-
without getting a large set of symmetric keys.
culates based on the revocation message. The mes-
sage is signed with the TA’s private key. V. EVALUATIONS
x After verifying the revocation message, the vehicles
with kz can compute the intermediate key kim inde- A. Security Evaluation
pendently. In our scheme, the HMAC in the commitment message is
x If any vehicle Vw does not have the key kz after veri- calculated with the group key shared between the TA and
fying the revocation message, it randomly selects r non-revoked vehicles. This assures that only the legitimate
keys out of its symmetric key set Rw and broadcasts vehicles can create a valid commitment message while an
the ids of the selected keys to its neighboring ve- outside attacker cannot forge a correct commitment message
hicles to request the intermediate key. It also starts a since it does not have the group key.
timer T. An attacker may try to guess the chain value that will be
x All the neighboring vehicles of Vw search their key released in the next message of a legitimate vehicle and
sets to find a shared key with Vw. If any neighboring broadcast a message with it trying to pass the pre-
vehicle finds a shared key, it uses the shared key to authentication. However, the property of one-way hash chain
encrypt kim and sends the encrypted kim to Vw along promises that it is computationally infeasible for an attacker
with the encryption key’s id. to find the next chain value. Therefore, only the sender
x If the timer T is timed out and Vw does not receive knows the key element of the chain and can release it. As a
any required data, it will select another set of r keys result, we can authenticate the legitimacy of the message by
and try again. authenticating the chain elements.


Moreover, if an attacker receives a legitimate message, it VI. CONCLUSION
may replace the actual message with some bogus message Since most of the authentication schemes for VANETs
but keep the same chain value and re-broadcast the modified are based on signatures, outside attackers can forge messages
message trying to deceive other vehicles into verifying the with invalid signatures in order to exhaust a receiver’s re-
signature. But according to the triangular inequality, a re- source to verify the invalid signatures. This would cause a
ceiver will receive a message broadcast by a sender before it DoS attack, which can prevent the benign vehicles from veri-
receives a copy forwarded (maybe modified) from an attack- fying other legitimate signatures and eventually cripples
er. Therefore, the receivers who have received the modified VANETs. We suggest adding a pre-authentication process,
message will consider it as forged because they have already which makes use of the one-way hash chain and a group
verified the same chain value before. rekeying scheme. A receiver would verify a signature only
The receiver will verify the signature of a message only after the message has passed the pre-authentication process.
after the message has passed the one-way hash chain-based Evaluations demonstrate that our scheme can mitigate such
pre-authentication; otherwise it refuses to verify the signa- DoS attacks effectively.
ture. As a result, our scheme can mitigate such DoS attacks As part of our future work, we will investigate how to
against signature-based authentication. deal with the DoS attacks from inside attackers.
B. Performance Evaluation
REFERENCES
700 [1] M. Raya and J.-P. Hubaux, “Securing vehicular ad hoc networks,”
with pre-authentication (10% invalid) Journal of Computer Security, vol. 15, pp. 39-68, Jan. 2007.
without pre-authentication (10% invalid)
[2] X. Lin, X. Sun, X. Wang, C. Zhang, P.-H. Ho, and X. Shen, “TSVC:
600 with pre-authentication (30% invalid)
without pre-authentication (30% invalid)
Timed efficient and secure vehicular communications with privacy
no attack preserving,” IEEE Transactions on Wireless Communications, vol. 7,
Authentication delay (ms)

500 pp. 4987-4998, Dec. 2008.


[3] G. Calandriello, P. Papadimitratos, J.-P. Hubaux, and A. Lioy,
“Efficient and robust pseudonymous authentication in VANET,” in
400
proc. 4th ACM International Workshop on Vehicular Ad Hoc
Networks (VANET’07), pp. 19-28, Sept. 2007.
300 [4] C. Zhang, R. Lu, X. Lin, P.-H. Ho, and X. Shen, “An efficient
240.05 identity-based batch verification scheme for vehicular sensor
networks,” in proc. 27th IEEE Conference on Computer
200
240
Communications (INFOCOM’08), pp. 816-824, Apr. 2008.
100 100.01 [5] A. Wasef and X. Shen, “Efficient group signature scheme supporting
100
batch verification for securing vehicular networks,” in proc. IEEE
60 80 100 120 140 160 180 200 International Conference on Communications (ICC) 2010, pp. 1-5,
Number of valid signatures (n) May 2010.
Figure 2. Authentication delay under different scenarios [6] C. Zhang, X. Lin, R. Lu, and P.-H. Ho, “RAISE: An efficient RSU-
aided message authentication scheme in vehicular communication
networks,” in proc. IEEE International Conference on
When we evaluate the performance of our scheme, we Communications (ICC) 2008, pp. 1451-1457, May 2008.
select ECDSA as our signature scheme. According to [8], the [7] A. Studer, F. Bai, B. Bellur, and A. Perrig, “Flexible, extensible, and
operation of the secure hash algorithm-1 (SHA-1) is 0.42 μs efficient VANET authentication,” Journal of Communications and
and verifying a signature using ECDSA takes 2.4 ms. Networks, vol. 11, pp. 574-588, Dec. 2009.
We evaluate the authentication delay under different sce- [8] A. Wasef, R. Lu, X. Lin, and X. Shen, “Complementing public key
narios in Figure 2. The black star line indicates the scenario infrastructure to secure vehicular ad hoc networks,” IEEE Wireless
where there is no DoS attack. In the scenarios where there Communications, vol. 17, pp. 22-28, Oct. 2010.
are DoS attacks, we add the number of invalid signatures to [9] P. Ning, A. Liu, and W. Du, “Mitigating DoS attacks against
broadcast authentication in wireless sensor networks,” ACM
be 10% and 30% of the number of valid signatures and show Transactions on Sensor Networks, vol. 4, no.1, Jan. 2008.
the differences between the scenarios with a pre- [10] X. Du, M. Guizani, Y. Xiao, and H.-H. Chen, “Defending DoS
authentication process and without a pre-authentication attacks on broadcast authentication in wireless sensor networks,” in
process. From the embedded small figure of Figure 2, we can proc. IEEE International Conference on Communications (ICC) 2008,
see that the pre-authentication has little effect on the authen- pp. 1653-1657, May 2008.
tication delay compared with the scenario where there is no [11] S. Zhu, S. Xu, S. Setia, and S. Jajodia, “LHAP: A lightweight hop-by-
DoS attack. On the other hand, without pre-authentication, hop authentication protocol for ad-hoc networks,” in proc. 23rd
International Conference on Distributed Computing Systems
the authentication delay would be significantly influenced if Workshops, pp. 749-755, May 2003.
there is a DoS attack. As a result, the proposed pre-
[12] A. Wasef and X. Shen, “REP: Location privacy for VANETs using
authentication process can effectively detect the invalid sig- random encryption periods,” Mobile Networks and Applications, vol.
natures. In other words, the proposed scheme can effectively 15, pp. 172-185, Feb. 2010.
mitigate such DoS attacks against signature-based authenti- [13] S. Zhu, S. Setia, S. Xu, and S. Jajodia, “GKMPAN: An efficient
cation in VANETs. group rekeying scheme for secure multicast in ad-hoc networks,”
Journal of Computer Security, vol. 14, pp. 301-325, July 2006.



You might also like