Network Edge Security Concepts
The ability to classify packets by IP traffic plane helps define and enforce security policies. You can achieve improved clarity and accuracy during the classification process by considering the point in the network at which packets are observed. That is, the location of packet classification allows more intelligence to be applied when identifying good and bad traffic. In general, two distinctions are made regarding location: edge and core. Chapter 1 briefly introduced the concepts of the network edge and core, and how these differ for enterprise and SP networks. The "Network Interface Types" section earlier in this chapter introduced the concept of external and internal interfaces, which are directly related to edge and core concepts. This section extends this discussion by looking more closely at network edge and core concepts.
The network edge is your first, and sometimes best, opportunity to make decisions about trusted and untrusted packets (classification), and to apply appropriate policies. In general, both ingress and egress perspectives are important, but for different reasons. On ingress, you want to deny bad traffic and permit only good traffic. Obviously, the main question is how to determine good traffic from bad. Of course, the goal of applying security policies to ingress traffic is to protect from attack the network infrastructure itself and downstream devices and services. On egress, the same considerations should be made. On egress, bad traffic should be denied and only good traffic should be permitted to exit your network. There are several goals for egress policies, one being preventing infected or zombie internal hosts from causing damage to other internal and external networks. Once interfaces are categorized and classifications are made, policies may be applied such as: permit, deny, rate limit, recolor, tunnel, count, or others as required. Of course, distinct policies at the edge for ingress and egress traffic flows may also be applied.
Different types of networks have different definitions of trust and different security requirements. As briefly discussed in Chapter 1, and as you will see next, very different security requirements may exist even for similar networks but with differing network edge types. The Internet edge looks very different from the perspective of an enterprise than it does from the perspective of an SP, for example. These security requirements and resulting policies determine in large part just how robust the entire network is against attacks. Two types of network edges are reviewed here: the Internet edge, and the MPLS VPN edge. (Other types exist, such as the Layer 2 Ethernet edge.)
The Internet edge is always the most vulnerable of any of the network edge types. Enterprises have little control over what traffic reaches their Internet edge. SPs even have limited control as well. The only guaranteed control is the one you apply to packets as they cross this Internet edge boundary. IP packets can be sourced from anywhere and carry anything as a payload. They may be legitimate, of course, or they may have malicious intentions. There may be a single malformed or crafted packet destined to one IP address, or a flood of millions of packets per second targeting a single destination IP address. Thus, the decisions made about ingress packets at the Internet edge are the most critical to overall network security. Service providers and enterprises have vastly different security policies at the Internet edge. These can be summarized as follows:
- As introduced in Chapter 1, enterprises typically have well-defined traffic flows traversing the Internet edge from inside-to-outside and outside-to-inside. (Internal traffic flows that stay entirely within the enterprise network are not discussed here.) Also, enterprise networks should never see transit traffic; that is, packets ingressing the Internet edge should never have destination IP addresses that are not part of the enterprise network address space. This gives enterprises the opportunity to deploy well-defined security policies at the Internet edge. Generally the approach is "everything is denied unless explicitly permitted."
- Also as introduced in Chapter 1, SPs have quite different traffic flows at their Internet edge as compared with enterprises. First, it is worth identifying just exactly where the Internet edge is for SPs. For enterprises, the Internet edge is easily identifiable; it is simply their WAN connection to their SP(s). However, for SPs, their Internet edge represents all external interface Internet connections including peering interconnects, transit customer access links, and any upstream or downstream SP interconnects. These are the boundaries where SPs apply their Internet edge security policies. And in just the opposite manner as an enterprise, an SP should only see transit traffic (with the exception of some control plane and possibly management plane traffic) at these edge boundaries. This also gives the SP the opportunity to deploy well-defined security policies at their Internet edge. Generally the approach is "everything is permitted unless explicitly denied."
In looking at the most basic perspective, the Internet edge policies for enterprises and SPs are opposites from one another. The enterprise Internet edge appears as a hard boundary where nothing is permitted unless it is either return traffic from internally generated traffic, or tightly controlled externally originated traffic destined to well-defined publicly exposed services. SPs, on the other hand, build networks to allow all transit traffic to cross their Internet edge without impediment. The SP edge is designed to be generally wide open and everything is permitted except for a few explicitly forbidden destinations belonging to the SP infrastructure. These differences in philosophy are illustrated in Figure 3-4.
Figure 3-4 Internet Edge Security Policy Comparisons for Enterprise and Service Provider Networks
Chapters 4 through 7 describe in detail the many security techniques that may be used on the Internet edge to mitigate the risk of attacks. The case studies in Chapters 8 and 9 present additional details on how these and other features may be deployed and how they complement one another.
MPLS VPN Edge
Multiprotocol Label Switching (MPLS) Virtual Private Networks (VPN) provide addressing and routing separation to create virtual IP VPN networks, typically as replacements for classic SP-based Frame Relay or ATM-based networks. MPLS-based Layer 3 VPNs combine Multiprotocol BGP using extended community attributes and VPN address families, LDP (RFC 3036) or RSVP-TE (RFC 3209) for label distribution, and router support for Virtual Routing and Forwarding (VRF) instances to create these virtual IP networks. The MPLS VPN edge, illustrated in Figure 3-5, includes the portion of the network encompassing the provider edge (PE) router(s), the customer edge (CE) router(s), and the CE-PE links between these routers.
Figure 3-5 Conceptual MPLS VPN Network Topology
As illustrated in Figure 3-5, CE routers sit physically at each customer premises location (typically) and are logically part of the customer VPN. CE routers use only IP routing (not MPLS) to forward traffic associated with the customer's VPN network. IP traffic destined to remote customer VPN sites is forwarded downstream toward the PE routers, exactly like any other IP router would. The MPLS VPN functions implemented on the PE routers provide IP reachability to remote customer VPN sites as well as isolation between different customer VPNs. As such, CE routers and internal customer VPN networks are reachable only from within the assigned customer VPN. Therefore, by default, CE routers are not susceptible to attacks sourced from outside the assigned VPN. Internal attacks sourced from within the VPN remain possible just as with any enterprise or SP network. For example, a malware infected host within one customer VPN site may attack other hosts within the same VPN (locally or remotely connected). Thus, security mechanisms appropriate for internal deployment within the enterprise network remain appropriate, even for managed MPLS VPN–based services.
Each CE router is connected to one or more PE routers via some data link layer interface. This CE-PE link belongs logically to the assigned customer VPN as well, and includes the IP addresses used on the CE and associated PE interfaces. These interface addresses are typically provided by the SP, because MPLS VPNs are often offered as a managed service, and the management functions used by the SP network operations center (NOC) require unique CE addressing for proper management connectivity. Refer to Chapter 6 for a detailed review of the Management VPN used for MPLS VPNs.
PE routers are logically part of the SP's network and peer at Layer 3 with both directly connected CE routers and SP core (P) routers. SP core (P) routers are not directly reachable by VPN customer traffic given the addressing and routing separation provided by RFC 4364, although indirect attacks are plausible. However, PE routers (the PE side of each CE-PE link) are often reachable from within a customer VPN and thus must be protected from internal attacks. In the Internet edge case, CE routers may be attacked from the wider Internet if reachable via the wider Internet. In the general MPLS VPN case, however, each VPN is logically isolated from one another as well as from the global Internet routing table. Thus, CE and PE routers are only susceptible to attacks sourced from inside a customer VPN. Note, even though CE and PE routers are reachable internally within the configured customer VPN(s), it is not possible for a host in one VPN to directly attack the CE router or PE router interfaces associated with another customer VPN given the isolation provided by RFC 4364. However, an attack against the PE from within one customer VPN may have an adverse impact on other VPNs configured on the same PE if the attack is able to disrupt a shared PE resource such as CPU, packet memory, and so forth. This is referred to as collateral damage, as described in Chapter 2, and is considered the most significant threat against MPLS VPNs.
Thus, similar to the Internet edge, SPs may also consider deploying security mechanisms on MPLS VPN PE routers to protect their own infrastructure from attack. Although not generally susceptible to Internet-based attacks, internal attacks sourced from inside a customer VPN may adversely affect other VPN customers as outlined previously in this chapter. Chapter 7 describes the security techniques applicable to MPLS VPN networks.