Arista MLS EOS 4.2x Router Security Technical Implementation Guide

Verifying the path a route has traversed will ensure the IP core is not used as a transit network for unauthorized or possibly even internet traffic. All autonomous system boundary routers (ASBRs) must ensure updates received from eBGP peers list their AS number as the first AS in the AS_PATH attribute.

The interoperability of BGP extensions for interdomain multicast routing and MSDP enables seamless connectivity of multicast domains between autonomous systems. MP-BGP advertises the unicast prefixes of the multicast sources used by Protocol Independent Multicast (PIM) routers to perform RPF checks and build multicast distribution trees. MSDP is a mechanism used to connect multiple PIM sparse-mode domains, allowing RPs from different domains to share information about active sources. When RPs in peering multicast domains hear about active sources, they can pass on that information to their local receivers, thereby allowing multicast data to be forwarded between the domains. Configuring an import policy to block multicast advertisements for reserved, martian, single-source multicast, and any other undesirable multicast groups, as well as any source-group (S, G) states with bogon source addresses, would assist in avoiding unwanted multicast traffic from traversing the core.

To avoid global visibility of local information, there are a number of source-group (S, G) states in a PIM-SM domain that must not be leaked to another domain, such as multicast sources with private address, administratively scoped multicast addresses, and the auto-RP groups (224.0.1.39 and 224.0.1.40). Allowing a multicast distribution tree, local to the core, to extend beyond its boundary could enable local multicast traffic to leak into other autonomous systems and customer networks.

To reduce any risk of a denial-of-service (DoS) attack from a rogue or misconfigured MSDP router, the router must be configured to limit the number of source-active messages it accepts from each peer.

Verifying the path a route has traversed will ensure that the local AS is not used as a transit network for unauthorized traffic. To ensure that the local AS does not carry any prefixes that do not belong to any customers, all PE routers must be configured to reject routes with an originating AS other than that belonging to the customer.

If multicast traffic is forwarded beyond the intended boundary, it is possible that it can be intercepted by unauthorized or unintended personnel. Administrative scoped multicast addresses are locally assigned and are to be used exclusively by the enterprise network or enclave. Administrative scoped multicast traffic must not cross the enclave perimeter in either direction. Restricting multicast traffic makes it more difficult for a malicious user to access sensitive traffic. Admin-Local scope is encouraged for any multicast traffic within a network intended for network management, as well as for control plane traffic that must reach beyond link-local destinations.

An inactive interface is rarely monitored or controlled and may expose a network to an undetected attack on that interface. Unauthorized personnel with access to the communication facility could gain access to a router by connecting to a configured interface that is not in use. If an interface is no longer used, the configuration must be deleted and the interface disabled. For sub-interfaces, delete sub-interfaces that are on inactive interfaces and delete sub-interfaces that are themselves inactive. If the sub-interface is no longer necessary for authorized communications, it must be deleted.

If the static routes to the alternate gateway are being redistributed into an Exterior Gateway Protocol or Interior Gateway Protocol to a NIPRNet gateway, this could make traffic on NIPRNet flow to that particular router and not to the internet Access Point routers. This could not only wreak havoc with traffic flows on NIPRNet, but it could overwhelm the connection from the router to the NIPRNet gateway(s) and also cause traffic destined for outside of NIPRNet to bypass the defenses of the internet Access Points.

Real-time multicast traffic can entail multiple large flows of data. An attacker can flood a network segment with multicast packets, over-using the available bandwidth and thereby creating a denial-of-service (DoS) condition. Hence, it is imperative that register messages are accepted only for authorized multicast groups and sources. Satisfies: SRG-NET-000019-RTR-000013, SRG-NET-000019-RTR-000014

A compromised router introduces risk to the entire network infrastructure, as well as data resources that are accessible via the network. The perimeter defense has no oversight or control of attacks by malicious users within the network. Preventing network breaches from within is dependent on implementing a comprehensive defense-in-depth strategy, including securing each device connected to the network. This is accomplished by following and implementing all security guidance applicable for each node type. A fundamental step in securing each router is to enable only the capabilities required for operation.

RSVP-TE can be used to perform constraint-based routing when building LSP tunnels within the network core that will support QoS and traffic engineering requirements. RSVP-TE is also used to enable MPLS Fast Reroute, a network restoration mechanism that will reroute traffic onto a backup LSP in case of a node or link failure along the primary path. When there is a disruption in the MPLS core, such as a link flap or router reboot, the result is a significant amount of RSVP signaling, such as "PathErr" and "ResvErr" messages that need to be sent for every LSP using that link. When RSVP messages are sent out, they are sent either hop by hop or with the router alert bit set in the IP header. This means that every router along the path must examine the packet to determine if additional processing is required for these RSVP messages. If there is enough signaling traffic in the network, it is possible for an interface to receive more packets for its input queue than it can hold, resulting in dropped RSVP messages and hence slower RSVP convergence. Increasing the size of the interface input queue can help prevent dropping packets; however, there is still the risk of having a burst of signaling traffic that can fill the queue. Solutions to mitigate this risk are RSVP message pacing or refresh reduction to control the rate at which RSVP messages are sent. RSVP refresh reduction includes the following features: RSVP message bundling, RSVP Message ID to reduce message processing overhead, Reliable delivery of RSVP messages using Message ID, and summary refresh to reduce the amount of information transmitted every refresh interval.

Different applications have unique requirements and toleration levels for delay, jitter, bandwidth, packet loss, and availability. To manage the multitude of applications and services, a network requires a QoS framework to differentiate traffic and provide a method to manage network congestion. The Differentiated Services Model (DiffServ) is based on per-hop behavior by categorizing traffic into different classes and enabling each node to enforce a forwarding treatment to each packet as dictated by a policy. Packet markings such as IP Precedence and its successor, Differentiated Services Code Points (DSCP), were defined along with specific per-hop behaviors for key traffic types to enable a scalable QoS solution. DiffServ QoS categorizes network traffic, prioritizes it according to its relative importance, and provides priority treatment based on the classification. It is imperative that end-to-end QoS is implemented within the IP core network to provide preferred treatment for mission-critical applications.

Different applications have unique requirements and toleration levels for delay, jitter, bandwidth, packet loss, and availability. To manage the multitude of applications and services, a network requires a QoS framework to differentiate traffic and provide a method to manage network congestion. The Differentiated Services Model (DiffServ) is based on per-hop behavior by categorizing traffic into different classes and enabling each node to enforce a forwarding treatment to each packet as dictated by a policy. Packet markings such as IP Precedence and its successor, Differentiated Services Code Points (DSCP), were defined along with specific per-hop behaviors for key traffic types to enable a scalable QoS solution. DiffServ QoS categorizes network traffic, prioritizes it according to its relative importance, and provides priority treatment based on the classification. It is imperative that end-to-end QoS is implemented within the IP core network to provide preferred treatment for mission-critical applications.

An IP-directed broadcast is a datagram sent to the broadcast address of a subnet that is not directly attached to the sending machine. The directed broadcast is routed through the network as a unicast packet until it arrives at the target subnet, where it is converted into a link-layer broadcast. Because of the nature of the IP addressing architecture, only the last router in the chain, which is connected directly to the target subnet, can conclusively identify a directed broadcast. IP-directed broadcasts are used in the extremely common and popular smurf, or denial-of-service (DoS) attacks. In a smurf attack, the attacker sends Internet Control Message Protocol (ICMP) echo requests from a falsified source address to a directed broadcast address, causing all the hosts on the target subnet to send replies to the falsified source. By sending a continuous stream of such requests, the attacker can create a much larger stream of replies, which can completely inundate the host whose address is being falsified. This service should be disabled on all interfaces when not needed to prevent smurf and DoS attacks. Directed broadcast can be enabled on internal facing interfaces to support services such as Wake-On-LAN. Case scenario may also include support for legacy applications where the content server and the clients do not support multicast. The content servers send streaming data using UDP broadcast. Used in conjunction with the IP multicast helper-map feature, broadcast data can be sent across a multicast topology. The broadcast streams are converted to multicast and vice versa at the first-hop routers and last-hop routers before entering and leaving the multicast transit area respectively. The last-hop router must convert the multicast to broadcast. Hence, this interface must be configured to forward a broadcast packet (i.e., a directed broadcast address is converted to the nodes' broadcast address).

The effects of prefix de-aggregation can degrade router performance due to the size of routing tables and also result in black-holing legitimate traffic. Initiated by an attacker or a misconfigured router, prefix de-aggregation occurs when the announcement of a large prefix is fragmented into a collection of smaller prefix announcements.

MSDP peering between networks enables sharing of multicast source information. Enclaves with an existing multicast topology using PIM-SM can configure their RP routers to peer with MSDP routers. As a first step of defense against a denial-of-service (DoS) attack, all RP routers must limit the multicast forwarding cache to ensure that router resources are not saturated managing an overwhelming number of PIM and MSDP source-active entries.

GTSM is designed to protect a router's IP-based control plane from DoS attacks. Many attacks focused on CPU load and line-card overload can be prevented by implementing GTSM on all Exterior Border Gateway Protocol speaking routers. GTSM is based on the fact that the vast majority of control plane peering is established between adjacent routers; that is, the Exterior Border Gateway Protocol peers are either between connecting interfaces or between loopback interfaces. Since TTL spoofing is considered nearly impossible, a mechanism based on an expected TTL value provides a simple and reasonably robust defense from infrastructure attacks based on forged control plane traffic.

LLDPs are primarily used to obtain protocol addresses of neighboring devices and discover platform capabilities of those devices. Use of SNMP with the LLDP Management Information Base (MIB) allows network management applications to learn the device type and the SNMP agent address of neighboring devices, thereby enabling the application to send SNMP queries to those devices. LLDPs are also media- and protocol-independent as they run over the data link layer; therefore, two systems that support different network-layer protocols can still learn about each other. Allowing LLDP messages to reach external network nodes is dangerous as it provides an attacker a method to obtain information of the network infrastructure that can be useful to plan an attack.

Real-time multicast traffic can entail multiple large flows of data. Large unicast flows tend to be fairly isolated (i.e., someone doing a file download here or there), whereas multicast can have broader impact on bandwidth consumption, resulting in extreme network congestion. Hence, it is imperative that there is multicast admission control to restrict which multicast groups hosts are allowed to join via IGMP or MLD.

Using a loopback address as the source address offers a multitude of uses for security, access, management, and scalability of the BGP routers. It is easier to construct appropriate ingress filters for router management plane traffic destined to the network management subnet since the source addresses will be from the range used for loopback interfaces instead of a larger range of addresses used for physical interfaces. Log information recorded by authentication and syslog servers will record the router's loopback address instead of the numerous physical interface addresses. When the loopback address is used as the source for eBGP peering, the BGP session will be harder to hijack since the source address to be used is not known globally—making it more difficult for a hacker to spoof an eBGP neighbor. By using traceroute, a hacker can easily determine the addresses for an eBGP speaker when the IP address of an external interface is used as the source address. The routers within the iBGP domain should also use loopback addresses as the source address when establishing BGP sessions.

Using a loopback address as the source address offers a multitude of uses for security, access, management, and scalability of backbone routers. It is easier to construct appropriate ingress filters for router management plane traffic destined to the network management subnet since the source addresses will be from the range used for loopback interfaces instead of from a larger range of addresses used for physical interfaces. Log information recorded by authentication and syslog servers will record the router's loopback address instead of the numerous physical interface addresses.

Packet loss can occur when an IGP adjacency is established and the router begins forwarding packets using the new adjacency before the LDP label exchange completes between the peers on that link. Packet loss can also occur if an LDP session closes and the router continues to forward traffic using the link associated with the LDP peer rather than an alternate pathway with a fully synchronized LDP session. The MPLS LDP-IGP Synchronization feature provides a means to synchronize LDP with OSPF or IS-IS to minimize MPLS packet loss. When an IGP adjacency is established on a link but LDP-IGP synchronization is not yet achieved or is lost, the IGP will advertise the max-metric on that link.

Using a loopback address as the source address offers a multitude of uses for security, access, management, and scalability of MSDP routers. It is easier to construct appropriate ingress filters for router management plane traffic destined to the network management subnet since the source addresses will be from the range used for loopback interfaces instead of a larger range of addresses used for physical interfaces. Log information recorded by authentication and syslog servers will record the router's loopback address instead of the numerous physical interface addresses.

The Neighbor Discovery protocol allows a hop limit value to be advertised by routers in a Router Advertisement message being used by hosts instead of the standardized default value. If a very small value was configured and advertised to hosts on the LAN segment, communications would fail due to the hop limit reaching zero before the packets sent by a host reached its destination.