Comparison to ATM/FR Security
Many enterprises have been using VPN services based on ATM or Frame Relay (FR) in the past and are considering moving to MPLS VPNs. Unfortunately, the discussion about this topic has often been emotional and unbalanced.
New MPLS users are often concerned about the fact that an MPLS VPN service has a control plane on Layer 3. However, as shown in the previous sections, Layer 3 services can also be correctly secured and are fit to provide VPN services.
ATM/FR might be perceived as more secure because they are mostly not vulnerable to Layer 3 attacks (also ATM/FR switches typically have a Layer 3 control plane such as telnet). However, the security of Layer 2 in those technologies is typically assumed rather than actually proven. As we discuss in various parts of this book, Layer 2 has its own security issues that have to be considered. Many ATM/FR users are asking very hard questions about MPLS VPN security, while never having questioned whether a flood of signaling packets to an ATM switch might not affect that switch. It is good to discuss security of a technology, but it should be discussed in a balanced way.
This section discusses the features of both technologies and compares them.
VPN Separation
A VPN user requires his VPN to be separate from other VPNs and the core. In Layer 2 technologies, this is achieved implicitly by layering: the core exclusively uses Layer 2, so that the Layer 3 information of a VPN is separate. In MPLS VPNs, separation is achieved logically, by maintaining separate contexts on a provider router. Both ways are different, but both achieve the same result: each VPN can use the entire IP address space in their VPN, and it is impossible to send packets into other VPNs on the same core.
Misconfigurations are a problem in both technologies: an ATM circuit can be misconfigured, connecting a VPN router to a router from another company. Because many topologies are hub-and-spoke with default routing, this might lead to serious security breaches.
On the MPLS side, misconfigurations can equally break security. For example, a wrong route target on a PE router can bring a CE into a wrong VPN. You can argue about which misconfigurations are worse or more likely, but the fact is that if the core is misconfigured, VPN separation might be broken—in any VPN technology.
Robustness Against Attacks
VPN users demand a stable service, and most of all a service that cannot be attacked from the outside. For many VPN users, it would not be acceptable if a VPN service could be affected by a DoS attack from the outside. Even worse, an attacker gaining control of a network element could control any VPN. Therefore, any VPN technology must be resistant against attacks.
MPLS VPNs have been heavily scrutinized for the Layer 3 control plane and their frequent accessibility from the Internet. The issue was raised that given enough time, a good hacker would get access to a PE router over the Internet.
As shown previously in this chapter, an MPLS core has few and well-defined interface points to the outside. An MPLS core is not at all comparable to a traditional IP core, where every router was accessible (assuming the MPLS core has no global interfaces to the outside, only VRF interfaces). Rather, only single interfaces can be reached, and those can be very well secured. Therefore, it is very difficult to attack an MPLS network directly. An attack using transit traffic is the only possibility, and it might lead to a DoS condition. However, this can be controlled through appropriate dimensioning of the routers and architectural decisions, as is discussed in Chapter 4, in the section "Internet Access."
ATM or Frame Relay networks are also resistant against attack, assuming correct implementation. However, ATM switches and Frame Relay switches also have Layer 3 control planes (for example, telnet), and can be attacked if not appropriately secured. Frame Relay links depend on correct dimensioning of parameters such as Committed and Extended Information Rate (CIR/EIR). Misconfigurations of any of the above protocols or parameters can also result in a degradation or loss of service.
As long as both types of VPN technologies are configured correctly, they cannot be easily attacked.
Hiding the Core Infrastructure
In Layer 2 networks, the core is usually hidden because the VPN user works on Layer 3. Also, MPLS VPN cores are hidden to the VPN user, although using a different method: most addresses are hidden by architecture; the only visible part is the peering PE address. This address is, however, part of the VPN address space, so that in reality no core information is visible to the outside.
The fact that the PE router is reachable on this single interface is an exception to this rule. However, this usually is not a problem per se, but only in connection with attacks against the PE. As shown previously, this is very difficult if the PE is properly secured.
Impossibility of VPN Spoofing
As shown earlier in this chapter, it is impossible for an outsider to spoof another VPN, or the core, because a VPN user is always treated in his own context. Also, in ATM or Frame Relay, there are no known ways to spoof VPN signaling mechanisms such as the Virtual Path/Circuit Identifier (VPI/VCI) to spoof another VPN.
CE-CE Visibility
There is one area where ATM/Frame Relay point-to-point services do have an advantage over MPLS IP VPNs: Because the former are Layer 2 services, CEs can establish a direct Layer 3 adjacency and "see" the other CE. For example, the Cisco Discovery Protocol (CDP) can be used to find out basic properties of the peer router. This includes addressing of the Layer 3 link, so that a CE is able to verify to some extent the identity of the CE on the other end of the point-to-point link.
This is not possible in MPLS IP VPNs, and a given CE has no direct visibility of other CEs in his VPN. The reason for this is the connection model of MPLS IP VPNs: Although ATM and Frame Relay provide mostly point-to-point connections, where such a check is possible, MPLS IP VPNs provide connectivity from a CE to a "cloud." This avoids the overlay issue of having to establish a tunnel between all CEs (the so-called n2 issue), but it has the disadvantage of losing the direct peering information.
This problem is not only theoretical. There is a real issue when a service provider accidentally or maliciously adds a CE to a wrong VPN by configuring wrong route targets (RTs). The VPN to which this CE has been added has no easy means to find the bogus CE. It can only monitor traffic, control routing, and watch the used IP address space. This must be controlled by the service provider operationally. See Chapter 8, "Secure Operation and Maintenance of an MPLS Core," for more information on this important issue.
Other MPLS services, such as the Pseudo Wire Emulation (PWE), also implement point-to-point services based on Layer 2, where direct CE-CE visibility is possible.
Comparison of VPN Security Technologies
Table 3-2 compares all the aspects of VPN security for the different VPN technologies.
Table 3-2. Security Comparison Between MPLS and ATM/Frame Relay
|
MPLS |
ATM/Frame Relay |
|
|
VPN separation |
Yes |
Yes |
|
Robustness against attacks |
Yes |
Yes |
|
Hiding of the core infrastructure |
Yes |
Yes |
|
Impossibility of VPN spoofing |
Yes |
Yes |
|
CE-CE visibility |
Not in MLPS IP VPNs Yes for MPLS pseudo wire emulation |
Yes |
Overall, at the time of writing this book, the industry had mostly accepted that both MPLS and ATM/Frame Relay can be operated securely. It was also common understanding that operational issues such as misconfigurations are an issue for any VPN technology.