Cram Session: Network Access Control

The NAC labyrinth

Network access control represents the most significant change in the way that networks are secured since the invention of the firewall. But it's also contentious, confusing and -- when done right -- complicated.

With the stampede of vendors laying claim to NAC territory, IT managers are now presented with an overwhelming number of architectures promising eventual tools that will help create a strong link between users, end systems and access to network resources. In an effort to provide some insight into how each may or may not fit into your network, herein is a breakdown of their similarities and differences.

NAC is a broad new buzzword, and security and network vendors all have ideas about how best to give their products and services a place in the NAC bandwagon. The major NAC schemes we examine are Cisco's Network Admission Control, Juniper's Infranet, Microsoft's Network Access Protection and the Trusted Computing Group's (TCG) Trusted Network Connect.

Before diving into the who's who of NAC, it's important to understand its basic elements (see NAC primer).

There are three fundamental approaches to NAC based on where the access control is being enforced in the enterprise: edge control, core control and client control.

Edge control takes the principle of the firewall and pushes it to the edge of the network, where systems connect. If you are protecting a LAN, the individual switch port becomes the NAC control point. If you are working with a VPN connection, the IPSec concentrator or the SSL VPN device is in charge of enforcing access controls. In a wireless environment, the access point or wireless switch plays the NAC role.

In the core control schema, controls can be enforced anywhere in the network providing it's in deeper than the edge device. You could insert a NAC device inline, or as a passive tap, between edge switches and the core, where it would collect authentication and endpoint security information, and then enforce the appropriate access control policy. These devices (such as Lockdown Network's Enforcer) inspect traffic or control plane information passing by and reach into the network to change configuration to apply enforcement.

The client control approach focuses on the end system connecting to the network where greater attention is paid to the management and control of the end system. The Senforce Endpoint Security Suite installs a fairly heavyweight application on each end system that enforces NAC policies and local access controls, such as disabling wireless access if the VPN client isn't in use. An endpoint protected by this kind of tool inherits a strong set of security protections, such as personal firewall, USB device locking and wireless controls that might be difficult (or impossible) to assemble and manage from a slew of other NAC vendors.

While the client control approaches are attractive from a lower budget and simplistic management point of view, they don't strongly overlap with NAC approaches that integrate with the network to help to defend itself, to force user authentication or to provide identity-based access controls.

None of the NAC frameworks touted by the major network players fits neatly into any single deployment category. For example, as a manufacturer of firewalls and SSL VPN devices (but not switches), Juniper's Infranet NAC strategy is very much core-control oriented -- except when the controls are being applied to SSL VPN users, in which case the strategy looks more like an edge-control one. Similarly, Cisco's own Network Admission Control is more focused on the edge device and is designed to behave like an edge-control strategy, because Cisco controls the majority of enterprise switches in wiring closets. Cisco also includes controls upstream from those switches to handle environments where old switches that can't handle NAC are installed, so its underlying NAC architecture encompasses core-control aspects as well.

There are other distinguishing characteristics between NAC approaches. Some vendors consider the endpoint security assessment to be a one-time check at system connection, while others take a continuous approach, constantly checking and verifying the state of endpoint security. Some tightly focus on endpoint security as the key reason for implementing NAC, while others hone in on authentication and policy as the prime pieces. Some only work well in environments where their own agent is installed on the endpoint while others attempt to embrace environments where no agent is available.

Although these three implementation approaches let you start to winnow your NAC choices, you have to dive deeper into the proposed architectures to further decide what works best in your network.

The first difficulty in evaluating NAC architectures is there is a lot of paperwork but not a lot of products. For example, when Microsoft threw its hat into the NAC ring with Network Access Protection in July 2004, it used the network's DHCP server as one of the primary enforcement mechanisms. If you didn't pass the appropriate endpoint security checks, you got an IP address regulating you to the land of quarantine so that you could not disturb the rest of the network, but you couldn't fix your problems.

While that approach works great with a cooperative user community, it doesn't protect well against a malicious user looking to gain unauthorized network access. In response, Microsoft changed its architecture by adding a stronger enforcement mechanism based on 802.1X.

That was easy, because all Microsoft had to do was adjust a few white papers on its site to include this stronger enforcement. And while the current set of forecasts are that Network Access Protection will ship with the Vista version of Windows, expected late this year or early 2007, there's no promise that what comes out on those gold master CDs will include all the features in the NAC architecture.

Microsoft isn't the only vendor with a paucity of products. The TCG, a nonprofit industry-based standards organization comprising interested vendors, has been working on its Trusted Network Connect scheme since mid-2004 and still doesn't have a completed architecture. The Trusted Network Connect framework includes six separate protocols to build a complete system, but only two of these have been fully defined, making it impossible to have a fully deployed Trusted Network Connect NAC network. TCG has promised the rest of the protocols any day now.

Start with a little TCG

That said, the best starting point for evaluating NAC architecture is with the TCG's Trusted Network Connect because its specifications are created in an open, vendor-neutral environment and can be used as a good model just to get the terminology straight. Every proposed NAC strategy can be mapped to the Trusted Network Connect architecture, but that doesn't mean Trusted Network Connect is a superset of other products. Many NAC vendors add features not explicitly discussed by TCG, such as control of personal firewall or continuous rechecking of endpoint security status. Other NAC vendors handle cases that are not discussed in the TCG architecture, such as how to provide access controls when the end system is a guest laptop and doesn't have all the necessary software installed.

The Trusted Network Connect architecture divides the NAC problem into three entities: the Access Requestor, the Policy Enforcement Point and the Policy Decision Point.

TCG's Access Requestor is a combination of the entity trying to gain access to the network, such as a laptop or desktop computer, and the software and drivers that implement authentication and endpoint security assessment processes. TCG divides the Access Requestor into three smaller pieces. At the bottom is a Network Access Requestor, software used by the client to connect to the network, request access and provide authentication. For example, an 802.1X supplicant or an IPSec VPN client could serve as a network access requester.

Integrity Measurement Collectors, software components responsible for evaluating the security posture of the end system, are on top of the Network Access Requestor, still on the client system and part of the access. If your policy is defined such that everyone has to be running anti-virus software, then your anti-virus vendor would provide a plug-in that serves up status information on its software. TCG divides this task into two pieces: the Integrity Measurement Collectors, and the Trusted Network Connect client that collects information from the Integrity Measurement Collectors and helps package it up for policy evaluation.

The Trusted Network Connect NAC Policy Enforcement Point is exactly as it sounds: the point at which policy is enforced. TCG's NAC doesn't describe what kinds of policy-based enforcement mechanisms are available, probably to remain as vendor neutral as possible, although the architectural documents describe how quarantine and remediation might be part of policy enforcement.

The Trusted Network Connect NAC Policy Decision Point also is divided into three parts. The bottom piece, which is in charge of talking to the authentication server and communicating decisions to the Policy Enforcement Point, is the Network Access Authority. In a typical network, this would likely be an AAA (authentication, authorization and accounting) server.

Behind the Network Access Authority are Integrity Measurement Verifiers. These are the counterparts to the Integrity Measurement Collectors on the client. They receive the reports the Client Integrity Measurement collectors send and provide verification information back to the Network Access Authority. The verifiers and the collectors are a matched set. They can talk to each other through a tunnel provided by all the other pieces of the NAC architecture, using whatever proprietary vendor-specific protocol vendors want.

The Trusted Network Connect architecture layers a thin server piece, called the TNC server, in the Policy Decision Point that is the interface between the Integrity Measurement Verifiers and the Network Access Authority.

The Trusted Network Connect NAC architecture is designed to work within an existing 802.1X authentication and authorization system. If you rename the client-side network Access Requestor to "802.1X supplicant"; the Policy Enforcement Point to "802.1X-compatible switch or access point"; and the Network Access Authority Policy Decision Point to "802.1X RADIUS server", then the Trusted Network Connect NAC plan is bits of software that sit on top of an existing 802.1X deployment to add endpoint security assessment to the mix.

This becomes obvious if you look at the protocols the Trusted Network Connect chose to publish first. There are those that let the vendor-supplied integrity measurement collectors talk to a vendor-neutral Trusted Network Connect client on the client system and those that let the vendor-supplied Integrity Measurement Verifier talk to the vendor-neutral Trusted Network Connect server on the Policy Decision Point end. As prototype implementations started showing up, Trusted Network Connect relied heavily on existing 802.1X mechanisms such as authentication and tunneling to get the other pieces to work, although none of this is laid out explicitly in the architecture documents.

Focusing on 802.1X does not mean that the Trusted Network Connect architecture won't support other kinds of Policy Enforcement Points, such as firewalls, VPN concentrators or core switches. But it does mean that anyone trying to go from architecture to implementation using the Trusted Network Connect documents will quickly find even more missing pieces, at least at this stage of the architecture. Important parts of the big picture, such as how the Trusted Network Connect client and server talk to the network access requestor and network access authority, aren't called out for future discussion -- which means even when the architecture is completed as planned, it won't be fully complete.

While TCG's Trusted Network Connect architecture is a well-constructed way to think about NAC, refer to NAC components and compare NAC solutions, it doesn't yet represent an architecture that's complete enough to be used to begin implementation.

continued on page 2

Back to top

Submit A StoryClick here to submit a story for consideration by Cram Session Editor, stories@cramsessionnac.com