Understanding SAFE IP Telephony Axioms
SAFE IP telephony assumes conformance to the original SAFE axioms, as discussed in the "SAFE: A Security Blueprint for Enterprise Networks" whitepaper (refer to Chapter 3, "SAFE Design Concepts"). In addition to these, the SAFE IP telephony work introduces other axioms to the design that are specific to IP telephony networks:
Voice networks are targets.
Data and voice segmentation is key.
Telephony devices do not support confidentiality.
IP phones provide access to the data-voice segments.
PC-based IP phones require open access.
PC-based IP phones are especially susceptible to attack.
Controlling the voice-to-data segment interaction is key.
Establishing identity is key.
Rogue devices pose serious threats.
Secure and monitor all voice servers and segments.
Each of these axioms is described in greater detail next.
Voice Networks Are Targets
Voice networks increasingly represent high-value targets for attacks. Attacks can range from a practical joke on company employees through a company-wide voice-mail recording telling all employees to take a day off, to eavesdropping on the chief financial officer's conversations with analysts discussing the company's earnings before being announced, to eavesdropping on internal calls regarding customers. Voice networks today represent a greater risk to security than any other technology; it is imperative that these networks be secured as tightly as possible to reduce the impact that an attack can have on both the voice network and the data network.
Data and Voice Segmentation Is Key
Although IP-based telephony traffic can share the same physical network as data traffic, it should be segmented to a separate virtual LAN (VLAN) to provide additional QoS, scalability, manageability, and security, as shown in Figure 19-1. Segmenting telephony traffic from data traffic greatly enhances the security of the IP-based telephony traffic and allows for the same physical infrastructure to be leveraged.
Figure 19-1 Data and Voice Segmentation
Telephony Devices Do Not Support Confidentiality
IP-based telephony uses the same underlying physical infrastructure as the data network. As such, it is possible for an attacker to gain access to the telephony stream using a variety of attack tools. One of the most popular of these tools is called VOMIT. This tool reconstructs the data stream of the voice traffic captured using another tool, such as TCPdump or snoop; reconstructs the voice traffic; and outputs a WAV sound file. Although the phone is not actually misconfigured, this example reinforces the need to segment the voice and data traffic on the network. The use of a switched infrastructure is critical to that effort and becomes significantly advantageous in the capability to tune network intrusion detection systems (NIDS). However, even a switched infrastructure can be defeated by tools such as dsniff. dsniff can turn the switched medium into a shared medium, thus defeating the benefits of the switch technology. Another way that an attacker can defeat a switched medium is to plug a workstation into a network port in place of an IP phone.
IP Phones Provide Access to the Data-Voice Segments
IP phones typically provide a second network port so that a PC or workstation can plug into the phone, which then plugs into the network port. This provides the simplicity of a single cable for network connectivity. When this is the case, it is critical that you follow the data/voice segmentation principle. Some IP phones provide for simple Layer 2 connectivity, in which the phone acts as a hub; others provide switched infrastructure capabilities and can understand VLAN technology such as 802.1q tags. The phones that are VLAN capable support the segmentation of the data and voice segments through the use of 802.1q tags. However, your security design should not be based solely on VLAN segmentation; it should implement layered security best practices and Layer 3 access control in the distribution layer of the design.
PC-Based IP Phones Require Open Access
In addition to standalone IP phones, you have the option of PC-based IP phones. However, because these are software-only IP telephony devices, they reside on the data segment of the network but require access to the voice segment, thus violating the second axiom: Data and voice segmentation is key. As such, using PC-based IP phones is not recommended without the presence of a stateful firewall to broker the data-voice interaction. IP-based telephony devices typically use UDP port numbers greater than 16384. Without a stateful firewall in place to broker the connections between the data and voice segments, a wide range of UDP ports would have to be permitted through a filter. As a result, securing all connections between the two segments would be impossible. A stateful firewall is required to prevent an attack from one segment to the other.
PC-Based IP Phones Are Especially Susceptible to Attack
PC-based IP phones represent a significant difficulty in an IP telephony deployment. Unlike their standalone IP phone brethren, PC phones run on top of standard operating systems such as Microsoft Windows, which leaves them vulnerable to many of the same application, service, and OS attacks. Another difficulty is that PC-based IP phones reside in the data segment of the network and thus are susceptible to attacks such as Code-Red, Nimda, and SQL Slammer. In these examples, the worms bog down the PC-based IP phone user systems and the segments they reside in to such an extent that they are unusable.
Controlling the Voice-to-Data Segment Interaction Is Key
Controlling the voice-to-data segment interaction is critical to successfully deploying and securing an IP telephony system. The best way to accomplish this task is to use a stateful firewall. This type of firewall provides denial-of-service (DoS) protection against connection starvation and fragmentation attacks; allows dynamic, per-port access to the network; and provides spoof mitigation and general packet filtering. The placement of the stateful firewall is limited to areas of the network where the voice and data segments interact. These legitimate connections should be allowed:
Communication between the voice-mail system and the call-processing manager if one is located in the data segment.
Call establishment and configuration traffic between IP phones in a voice segment connecting to the call-processing manager in another voice segment.
Connections from IP phones in the voice segment and the voice-mail system, if it is located in the data segment.
IP phones in the voice segment browsing resources outside the voice segment through the proxy server. This requires that the proxy server be capable of accessing resources in the data segment or another voice segment through the firewall. Additionally, the firewall should broker users in the data segment browsing the call-processing manager in the voice segment.
If PC-based IP phones are deployed, the firewall must broker connections from the PC-based IP phones in the data segment connecting to the call-processing manager in the voice segment. In addition, if the voice-mail system is in the voice segment, connections from the PC-based IP phones to this system must be brokered by the stateful firewall.
It is recommended that you use RFC 1918 address spaces for all IP telephony devices, to reduce the possibility of voice traffic traversing outside the network. The added benefit to using RFC 1918 addresses is that attackers will not be able to easily scan for vulnerabilities because NAT will be configured on the firewall. If possible, use different RFC 1918 addresses for both the voice and data segments.
Establishing Identity Is Key
Device authentication in an IP telephony deployment typically is based on the MAC address of the phone. The IP telephone tries to retrieve its network configuration from the call-processing manager using the MAC address as the identification string. If the call-processing manager has no knowledge of a specific MAC address being provided by an IP telephony device (whether it is an IP phone or PC-based IP phone), it will not provide the network configuration to the device.
When possible, it is recommended that you apply user authentication in addition to device authentication. With user authentication, the user must log into a phone with a password or PIN before telephony services are provided. This feature originally was designed for shared office spaces and enables you to provide a custom configuration based on user identity. Although a slight incon-venience factor is associated with user authentication, it helps to further mitigate the placement of rogue phones into the network and the placing of a call. The next section describes the threats associated with rogue devices in more detail.
Rogue Devices Pose Serious Threats
To mitigate the impact of rogue devices, it is recommended that you lock down the switch ports, network segments, and services in the network. Best practices, including disabling unused ports, discussed throughout the SAFE designs apply to IP telephony. In addition, the following four best practices provide mitigation details that are specific to IP telephony:
Statically assign IP addresses to known MAC addresses in DHCP networks with IP phones deployed.
Turn off the common temporary automatic phone registration feature that many call-processing managers have available. In addition, configure the call-processing managers to deny config-uration information to unknown PC-based IP phones.
Consider using a utility such as ARPwatch to monitor MAC addresses in the voice segment. ARPwatch is available at http://www-nrg.ee.lbl.gov/nrg.html.
Filter in all network segments to restrict which devices can connect to the call-processing manager or the voice-mail system.
Secure and Monitor All Voice Servers and Segments
The same attacks that can cripple servers in the data segment can affect key voice servers in the voice segment. It is recommended that the same considerations given to production servers in the data segment be provided to the voice servers in the voice segment. These considerations include the following:
Turn off all unneeded services.
Update the operating system with the latest security patches.
Harden the OS configuration.
Disable unnecessary or unused features in the voice system.
Do not run unnecessary applications on the voice servers.
In addition, deploy NIDS in front of the call-processing managers, to detect attacks sourced from the data segment, and host-based IPS on the call-processing managers themselves. NIDS also can be deployed between the voice and data segments, to detect any DoS attacks targeted against the voice segment specifically. Finally, it is recommended that the management axioms discussed in Chapter 3 be used when managing the voice servers.