Patent Application
20060230450
Denial of Service (DoS) attacks continue to present a significant challenge. In fact, the frequency and magnitude of attacks in the Internet have been steadily increasing. There have been a number of well-publicized attacks such as the February, 2000 attack on popular Web sites that included Yahoo, CNN, EBay, etc., and the recent attacks on root DNS servers.
DoS attacks typically involve sending a large volume of traffic to a node that exceeds its processing capability, in effect knocking the afflicted node out of the network for the duration of the attack. A more sophisticated attack is a Distributed DoS attack (DDoS).
An attacker intending to launch a DDoS attack begins by subverting a number of nodes, using well-known security loopholes. These compromised nodes essentially become slaves of the attacker and act as launch points to inject traffic into the network. By summoning a reasonable number of compromised nodes, an attacker can potentially launch a large-scale, network wide attack by cascading the traffic from multiple launch points.
A large variety of solutions have been proposed to counter DoS attacks. The current state-of-the-art in defending against DoS attacks include firewalls (Cisco's PIX router, Netscreen, Checkpoint's Firewall-1 are some examples), router modifications to support pushback, traceback mechanisms that attempt to detect the source of the attack, and related intrusion detection mechanisms that look for anomalies or signatures in arriving traffic. Some of these approaches require significant changes to existing network elements, and are costly to deploy, while others require collaboration across ISPs, and thus may be impractical. Nonetheless, these schemes do reduce the threat of wireline DoS attacks. For example, most firewalls do not allow connections to be initiated from outside, thus preventing DoS flooding attacks.
Wireless networks on the other hand are significantly more fragile than wireline networks. There are several vulnerabilities in wireless networks that can be exploited by novel DoS attacks:
An attacker launching a wireless-specific DoS attack can easily exploit these vulnerabilities.
Another by-product of a wireless attack is that once the attack reaches a mobile it is too late. In a wireline DoS attack, it takes a certain amount of time for a server to be disabled because such machines have a larger processing capacity than a wireless endpoint (mobile). In contrast, a mobile has limited processing and battery lifetime. In addition, a wireless link is severely bandwidth-constrained when compared to a wireline network. If the traffic from an attack reaches a mobile, the attack has already succeeded in wasting critical resources of the wireless link, the wireless infrastructure, and the battery power of the mobile. This is in contrast to typical wireline DoS attacks that must overload processing resources at a server in order to succeed.
It is desirable, therefore, to provide methods and devices for detecting, preventing and defending against 3G wireless network DoS-like attacks. More particularly, it is desirable to provide methods and devices for detecting, preventing and defending against wireless DoS-like attacks launched against UMTS, CDMA2000 and other 3G wireless networks.
a and 3b depict exemplary simulations of a possible impact of a signaling attack.
A 3G wireless network requires the establishment (i.e., set-up) of a dedicated channel between a mobile and the associated wireless infrastructure for data to be transmitted. In order to set up the channel, signaling messages need to be transmitted between the mobile and elements of the wireless infrastructure. Signaling attacks seek to exploit the nature of this signaling to overload the wireless infrastructure.
Signaling Attacks in UMTS Networks
First, we begin with a brief introduction of the UMTS architecture. Then, we present an overview of the signaling required to set up a data channel and the vulnerability of a UMTS network to signaling attacks.
The SGSN is responsible for sending data to and from mobile stations, in addition to maintaining information about the location and authentication of a mobile. Typically, there are multiple SGSNs, each of which serves the GPRS users physically located in their serving area.
A Radio Network Controller (RNC), also known as a Base Station Controller (BSC) is the point where wireless link layer protocols terminate. The RNC provides an interface between wireless devices communicating through Base Stations (BS) and the network edge. This includes controlling and managing the radio transceivers in BS equipment, including radio resource control, admission control, channel allocation, as well as management tasks like handoff between BS's and deciding power control parameters. The BS functionality includes wireless link transmission/reception, modulation/demodulation, physical channel coding, error handling, and power control. In this hierarchical architecture, multiple mobiles communicate with a BS, multiple BS's communicate with an RNC, and multiple RNC's talk to the GGSN/SGSN.
The UMTS signaling control flow for setting up channels for resource allocation is now described. Essentially, when data arrives for a subscriber, a UMTS RNC needs to establish a radio access bearer (RAB) with the base station of the subscriber. The RAB is a channel for data transfer and is released after a timeout period for inactivity. For establishing and releasing an RAB, a significant number of messages are exchanged between an RNC, mobile and base station. This is a significant amount of overhead for the RNC and causes severe processing overhead during signaling time intervals.
An example of a message flow between an RNC, base station(s) and a mobile is shown in
Radio Resource Control (RRC) Messages
RRC-like messages are used to establish/release radio channels for mobile power measurements, to transport paging messages and to broadcast information. An RRC setup leads to the creation of a dedicated signaling channel, which is the first step in enabling data transmission to/from a mobile. This requires 6 messages exchanged between the RNC and the mobile. Following this, there is a series of messages exchanged between the mobile and the core network for the purposes of authentication, and establishment of context, including a PPP connection with a PDSN, and assignment of an IP address. A further 4 messages are exchanged between the RNC, the core network and the mobile for a control security mode which results in the exchange of ciphers to secure the context for a given subscriber. Finally, an RAB is set up requiring an additional 8 messages exchanged at the RNC (2 with a core network, 3 with a mobile and 3 with a base station), resulting in a total of 24 messages to establish a single RAB with the RNC (not including messages exchanged between different elements within the RNC). It should be noted that subsequent RAB establishments do not need to perform an RRC signaling channel setup, though use of a control security mode may be required to re-authenticate a mobile.
Depending on the location of a mobile, a so-called soft handoff requires adding “legs” or base stations to the primary node with which the mobile has the strongest signal. This requires an additional 4 messages exchanged at the RNC (2 with the mobile and 2 with the primary base station). Finally, after a transfer, a mobile can initiate a teardown of RAB and RRC connections. This is initiated by an IU release function. This requires a total of 11 messages exchanged at the RNC (2 with the core network, 4 with the mobile and 5 with the base station). This number again does not include intra-RNC processing messages that contribute to the overall load at the RNC.
If there is no data exchanged between a mobile and its core network, the mobile is placed in a suspended state. In such a state, air link resources are released and assigned to other active mobiles, which requires 4 messages (2 with its core network and 2 with a base station). However, the context is still maintained at the RNC, and the mobile retains its IP address obtained during RRC/RAB establishment. A mobile can be reactivated (2 messages exchanged with RNC) using a packet call context resumption message (additional 2 messages with its core network) as long as an idle timeout period of 5 seconds has not occurred.
Vulnerability and Nature of the Attack
An attacker exploits the heavy-duty signaling overhead required for setup of RABs by essentially triggering an excessive signaling message exchange between the RNC and BS. This may be achieved by sending a low volume burst at appropriately timed periods so that immediately after an RAB is torn down due to inactivity, a burst arrives from an attacker to trigger an establishment of a new RAB. This frequent setup/release can easily overload the RNC by requiring an excessive amount of signaling messages.
It should be noted that it is relatively easy for an attacker to obtain the IP addresses of mobiles. Wireless service providers typically assign chunks of contiguous IP addresses making it easy for an attacker to guess a valid range of addresses simply by posing as a valid subscriber.
Two salient features of an attack are worthy of note. First, as noted before, the low average transmission rate of the attack (small bursts are sent periodically), make it hard for any existing detection mechanism to classify the traffic as malicious. The low volume also makes it easier for an attacker to launch an attack as opposed to conventional DDoS attacks requiring the attacker to compromise thousands of hosts in order to even launch an attack. Furthermore, only one packet needs to be sent per mobile allowing the attacker to have a widespread, diffused impact further complicating detection.
Damage Due to the Attack
The damage due to such an attack may be so severe that valid traffic may not receive an allocation of resources causing it to be dropped by an RNC. The RNC can also become overloaded, effectively denying service to a significant number of subscribers. In addition, RNC's are engineered to handle a certain amount of simultaneously active mobiles/users (in practice, 10%). It is easy for an attacker to exceed this number due to the low-volume nature of an attack.
Another side effect of a signaling attack is the potential for draining a mobile's battery. Normally, to conserve power, a mobile switches to a low-power idle or dormant state when there are no packets being sent or received. Because low volume bursts are sent periodically, mobiles would be forced to stay active longer than necessary. In a worst case scenario, a mobile may never be allowed to enter a dormant state causing rapid draining of its battery.
The present inventors performed simulations of the impact of an attack; some of the results are given by the graphs in
The results from
Changing gears, the following is a brief description of the 3G1x architecture (e.g., CDMA2000) and signaling.
3G1x Architecture
3G1x Signaling
In a 3G1x network, there is an analogous problem for resource allocation. The equivalent of an RAB in UMTS networks is the fundamental channel (FCH) that needs to be setup to transfer data. Typical steps used to set-up an FCH are shown in
When a PDSN receives data for a mobile, it pages the mobile. Once a successful response to a paging message has been received (3 messages exchanged), a base station initiates the setup of a FCH or Traffic Channel with the mobile (exchanging 8 messages). In parallel, a service request is made between the base station and an RNC requiring 4 messages. The RNC is also expected to forward messages to the core network, which in this case is required to authenticate the user. This results in an additional 6 messages. Finally, 2 additional messages are exchanged between the RNC and the PDSN for accounting purposes. Once this is done, an active channel exists for transmitting data to and from the mobile.
Call release(s) follow the reverse procedure and require 7 messages from the RNC to the PDSN and base station, 8 messages between the base station and the RNC, PDSN and mobile. Overall, 29 messages are generated or received by the BS and 13 messages by the RNC in addition to 9 more messages that the RNC is responsible for forwarding to the core network.
Every subscriber typically has an primary leg assigned, which originally acts as the forwarding base station when a call is first setup. In certain cases, an additional base-station, called an anchor leg, is established for long-lived connections. This leg could be distinct from the primary leg and is defined as the base station with the strongest signal to the mobile. The anchor leg takes charge of deciding when to allocate supplemental channels (SCH), which are used when there is insufficient capacity on a 9.6 Kbps FCH.
If there is excess data in the form of bursts that exceed the capacity of the FCH, then resources are allocated on demand to create a SCH for each user. There are an additional 16 messages exchanged between the anchor leg, the RNC and other base stations (for soft handoffs) leading to a similar vulnerability of the anchor leg for supplemental channel allocation and release.
While an attack on the anchor leg potentially has less widespread damage than an attack on the RNC, the impact is still severe enough to lead to a significant loss of revenue to a wireless service provider. If an attacker can impact multiple anchor legs, then the damage is magnified even more. An attacker may even be able to target the subscribers that belong to a particular anchor leg.
It should be noted that if an attacker simply sends randomly addressed bursts the damage could still be severe. Even if a single anchor leg is not in charge of the subscribers that are receiving the bursts, the RNC can still be overloaded due to its interaction with multiple anchor legs.
To prevent and defend against an attack, the present inventors recognized that knowledge of so-called wireless states, in particular the signaling cost(s) as traffic traverses through a wireless network was needed. This allows malicious traffic to be identified as soon as it begins to introduce excessive signaling cost. Signaling cost can be obtained in various ways depending on the wireless infrastructure involved. Ideally, an exact signaling cost can be obtained by querying wireless elements, such as an RNC and base station, when these elements provide an interface for such queries. However, current 3G wireless networks do not have such an interface, thus requiring modifications in order to support such queries. Changing an existing infrastructure, however, is not a viable solution given the amount of investment already expended by network owners/operators.
The present inventors discovered a simple, yet novel mechanism of estimating the signaling cost from traffic arrival patterns assuming knowledge of signaling call(s). The present inventions provide for methods and devices for detecting an attack using the so estimated signaling cost.
As mentioned above, if a wireless element provides an interface for querying, the signaling cost can be obtained by a simple query. Absent that, the challenge is to obtain the cost without the assistance of such wireless elements. In one embodiment of the present invention, the signaling cost is estimated from traffic arrival patterns. This requires knowledge of the signaling protocols inside wireless elements. Table 1 shows an example of a technique for estimating signaling cost (due to RAB establish/release) in a UMTS network according to one embodiment of the present invention. Similar techniques within the scope of the present invention can be used to estimate other types of signaling costs for CDMA2000 networks according to alternative embodiments of the present invention.
The present invention makes use of the fact that, upon arrival of a packet, if the destination's RAB has been released, the destination mobile has to reestablish the RAB. This reestablishment creates an added cost to establish a new RAB and release the previous RAB due to the expiration of an idle timer; a cost which can be detected by techniques provided by the present invention.
A reliable sign of an ongoing signaling attack is the detection of excessive or additional signaling costs even though the volume of actual, transmitted data is low. Before continuing, it should be noted that an attacker can flood a network using huge amounts of traffic that would also introduce excessive signaling costs. However, this can be, relatively speaking, easily detected by existing firewall or intrusion detection mechanisms. With a low-volume attack, there needs to be a more accurate metric for detection. In accordance with the present invention, a statistical measure referred to as a signaling cost to data ratio is used as a metric. If the ratio exceeds a certain profiled threshold, a signaling attack is detected and malicious traffic/packets from the source of the attack are blocked. In a further embodiment of the present invention, if multiple attacks from the same source are detected, malicious traffic/packets, etc. from the source of the attack is blocked while allowing traffic from other sources through, for example. Other intrusion detection mechanisms may also be used to reduce the chance of false alarms.
In accordance with an exemplary method of the present invention, the first step in determining if traffic is part of an attack is to define a threshold for later comparison. This should be user/application specific. The value of this threshold may be chosen by profiling user/applications during a pre-processing time period.
During such a time period, a profile for each user may be created based on a statistical signaling cost to data ratio. Information used in building the profile includes packet arrival times, IP addresses and port numbers of source(s) and destination(s).
One novel aspect of the profiling mechanism provided by the present invention is the ability to aggregate (user, application, as well as server) related profiles. By user profile, we refer to the statistics for an individual user. This division can be further categorized by individual application. For instance, web surfing is the most frequently used service by most users. Similarly, a video-on-demand server may use RTP packets to broadcast video to users. Statistics on an individual web server basis can also be compiled by logging the arrival of HTTP/RTP packets.
To enable scalability, the profiles can be aggregated across users with similar behaviors. Current traffic can then be compared to the aggregated profile to detect inconsistencies. Aggregated profiles can analogously be maintained for popular servers and also for popular applications. The flexibility of using different classification approaches allows a more comprehensive and accurate characterization of what is considered as normal traffic. This profile is key to detecting abnormal and malicious traffic, while also minimizing the probability of false positives (incorrect classification of valid traffic as malicious traffic).
Table 2 above sets forth an example of a method or algorithm for defending against a DoS Signaling attack. This method may be executed by a device or devices capable of either estimating or collecting user state statistical information (e.g., actual traffic flow) from the wireless infrastructure. In general this method operates as follows.
In an initial step, a current measure, such as a signaling cost to data ratio is generated or derived. This ratio can be obtained through either directly querying the infrastructure or by using an estimation technique.
Next, the signaling cost to data ratio is compared against a threshold, reference ratio. If the derived ratio exceeds the threshold for a user, i.e., S/DTHRESH, determined from a pre-processing step that builds a profile for the user, subsequent traffic from a sender “s” is flagged.
Lastly, if a sufficient number of packets (when compared to a threshold, VOTETHRESH) have been flagged as suspicious from the sender s, OR the suspicious behavior lasts for an extended period of time (INVALIDTIMER), then a filter may be applied at a firewall to block future traffic from that sender.
It should be noted that the user profile generated during the pre-processing step is adaptable to user behavior in order to minimize false positives and accurately detect when a violation occurs. More specifically, after the initial profile is created during pre-processing, it may be updated regularly based on changes to a user's behavior.
The method utilized in Table 2 may be implemented in a number of different ways, one of which may be referred to as an Architecture for Wireless Attack REsistance (AWARE). AWARE-enabled devices may be modular and can support upgrades in order to counter future attacks.
In one embodiment of the present invention, an AWARE architecture (e.g., enabled device) may comprise two components: a learning database and profiler, and/or a detection engine or detector. The learning database is operable to capture and store information about a user during the pre-processing step. The profiler is operable to generate a traffic profile for a given user under normal (i.e., non-attack) conditions. The database and profiler may be one and the same and may be correlated with other user databases and profilers for cross-mobile correlation. The information in these databases is fed to a detector/detection engine. In one embodiment of the present invention, the detection engine is operable to maintain a threshold for each user and verify if current traffic for a user or set of users violates the corresponding threshold. Depending on the wireless elements that are capable of communicating with an AWARE-enabled device (one that contains a database, profiler and/or a detector), the location of the AWARE-enabled device may be varied:
In a “least-invasive” design, an AWARE-enabled device simply looks at IP packets that are passed on from a firewall before they reach the PDSN. All of the necessary information is contained in the application, TCP and IP headers and the payload itself. Information needed to build a profile can be extracted from the above headers and payload.
An AWARE-enabled device should be able to communicate with existing firewalls or IPsec gateways. Ideally, an AWARE-enabled device may be co-located at these entities acting as a filter to block suspected traffic. If a AWARE-enabled device is not co-located with the IPsec gateway, there needs to be a so-called security association with the gateway so it can decrypt and process ESP-encapsulated packets in a tunnel mode. Even if an AWARE-enabled device is not co-located with the firewall, there typically is an interface with most commercial firewalls such as Checkpoint's Firewall-1 that allows the configuration of filters.
If an AWARE-enabled device is placed between the PDSN and an RNC, more user-specific state information may be gathered (i.e., which RNC a user belongs to, and potentially other information that can be obtained from an RLP frame). Also, an AWARE-enabled device may obtain mobility related information because a mobile may cross over from one RNC to another. The impact of mobility information on the detection heuristic is worth analyzing, because highly mobile end-users can contribute significantly to the load of an infrastructure. Launching a wireless DoS attack against highly mobile users requires additional tasks, such as more frequent paging, that can add substantially to processing overhead. Also, a mobile may initiate a PPP connection with the PDSN before initiating a transfer. An AWARE-enabled device may also query the PDSN to obtain a PPP state history.
In addition to AWARE-enabled devices, the present inventors also recognized that an AWARE related architecture may also require additional devices.
For example, an AWARE-compatible interface is provided by the present invention. In one embodiment of the present invention, such an interface is operable to query wireless user/mobile state(s). Such an interface also allows an AWARE-enabled device (or devices) to communicate in a secure manner with the wireless infrastructure in order to obtain mobile/user-specific information. It should be noted that such an interface may be included as an option in infrastructures other than a PDSN, because at a minimum, packet arrivals need to be known in order to estimate state information.
In addition to an interface, the present invention provides for a plug-in detector.
Snort is an open-source IDS mechanism that emulates the functionality of an AWARE-enabled device. Snort is modular and allows new plug-ins to be installed, thus allowing the detection mechanism to be customized and enhanced for defense against current and future attacks. By “plug-in” is meant a generic term that refers to modules that can be added dynamically to alter the behavior of Snort.
In an additional embodiment of the present invention, a Snort compatible plug-in incorporating the detection heuristic functions of the present invention is provided.
Other plug-ins are also provided by the present invention. Again, during their experiments the present inventors utilized Snortsam to interface with a firewall and react to attacks.
In an additional embodiment of the present invention, a Snortsam compatible plug-in that allows such interfacing is provided by the present invention. Such a plug-in may be operable to act as filter(s) to block malicious traffic.
Alternatively, this plug-in may be interfaced with a wireless packet scheduler to reduce the priority of malicious traffic.
It should be understood that the methods of the present invention, the AWARE-enabled devices, interfaces, and any subcomponents (e.g., learning database, profiler, detector, etc.) may be realized in hardware, software, firmware or some combination of the three. For example, one or more programmable or programmed controllers, processors, etc. may be operable to store one or more programs or code (and data) that, in turn, is operable to carry out the features and functions of the present invention described above and in the claims that follow.
