IPSec VPN Gateway Security Technical Implementation Guide

All network devices must present a DoD approved warning banner prior to a system administrator logging on. The banner should warn any unauthorized user not to proceed. It also should provide clear and unequivocal notice to both authorized and unauthorized personnel that access to the device is subject to monitoring to detect unauthorized usage. Failure to display the required login warning banner prior to logon attempts will limit DoD’s ability to prosecute unauthorized access and also presents the potential to give rise to criminal and civil liability for systems administrators and information systems managers. In addition, DISA’s ability to monitor the device’s usage is limited unless a proper warning banner is displayed. DoD CIO has issued new, mandatory policy standardizing the wording of “notice and consent” banners and matching user agreements for all Secret and below DoD information systems, including stand-alone systems by releasing DoD CIO Memo, “Policy on Use of Department of Defense (DoD) Information Systems Standard Consent Banner and User Agreement”, dated 9 May 2008. The banner is mandatory and deviations are not permitted except as authorized in writing by the Deputy Assistant Secretary of Defense for Information and Identity Assurance. Implementation of this banner verbiage is further directed to all DoD components for all DoD assets via USCYBERCOM CTO 08-008A. Information Assurance OfficerECWM-1

Terminating an idle session within a short time period reduces the window of opportunity for unauthorized personnel to take control of a management session enabled between the managed network element and a PC or terminal server when the later has been left unattended. In addition quickly terminating an idle session will also free up resources committed by the managed network element as well as reduce the risk of a management session from being hijacked. Setting the timeout of the session to 10 minutes or less increases the level of protection afforded critical network components.Information Assurance OfficerECSC-1

Detailed information about the network is sent across the network via SNMP. If this information is discovered by attackers it could be used to trace the network, show the networks topology, and possibly gain access to network devices.Information Assurance OfficerECSC-1

A rogue router could send a fictitious routing update to convince a site’s premise router to send traffic to an incorrect or even a rogue destination. This diverted traffic could be analyzed to learn confidential information of the site’s network, or merely used to disrupt the network’s ability to effectively communicate with other networks.Information Assurance OfficerECSC-1

Numerous vulnerabilities exist with SNMP; therefore, without unique SNMP community names, the risk of compromise is dramatically increased. This is especially true with vendors default community names which are widely known by hackers and other networking experts. If a hacker gains access to these devices and can easily guess the name, this could result in denial of service, interception of sensitive information, or other destructive actions.Information Assurance OfficerECSC-1

By not restricting authorized accounts to their proper privilege level, access to restricted functions may be allowed before authorized personell are trained or experienced enough to use those functions. Network disruptions or outages may occur due to mistakes made by inexperienced persons using accounts with greater privileges than necessary.Information Assurance OfficerECSC-1

A malicious user attempting to gain access to the network device may compromise an account that may be unauthorized for use. The unauthorized account may be a temporary or inactive account that is no longer needed to access the device. Denial of Service, interception of sensitive information, or other destructive actions could potentially take place if an unauthorized account is configured to access the network device.Information Assurance OfficerECSC-1, IAAC-1

Administration and management connections performed across a network are inherently dangerous because anyone with a packet sniffer and access to the right LAN segment can acquire the network device account and password information. With this intercepted information they could gain access to the router and cause denial of service attacks, intercept sensitive information, or perform other destructive actions.Information Assurance OfficerDCNR-1, ECSC-1

Routers can find their startup configuration either in their own NVRAM or access it over the network via TFTP or Remote Copy (rcp). Loading the image from the network is taking a security risk since the image could be intercepted by an attacker who could corrupt the image resulting in a denial of service.Information Assurance OfficerECSC-1

Source routing is a feature of IP, whereby individual packets can specify routes. This feature is used in several different network attacks by bypassing perimeter and internal defense mechanisms.Information Assurance OfficerECSC-1

The additional services that the router is enabled for increases the risk for an attack since the router will listen for these services. In addition, these services provide an unsecured method for an attacker to gain access to the router. Most recent software versions support remote configuration and monitoring using the World Wide Web's HTTP protocol. In general, HTTP access is equivalent to interactive access to the router. The authentication protocol used for HTTP is equivalent to sending a clear-text password across the network, and, unfortunately, there is no effective provision in HTTP for challenge-based or one-time passwords. This makes HTTP a relatively risky choice for use across the public Internet. Any additional services that are enabled increase the risk for an attack since the router will listen for these services.Information Assurance OfficerECSC-1

Network devices not running the latest tested and approved versions of software are vulnerable to network attacks. Running the most current, approved version of system and device software helps the site maintain a stable base of security fixes and patches, as well as enhancements to IP security. Viruses, denial of service attacks, system weaknesses, back doors and other potentially harmful situations could render a system vulnerable, allowing unauthorized access to DoD assets.Information Assurance OfficerECSC-1

Authentication for administrative access to the device is required at all times. A single account can be created on the device's local database for use in an emergency such as when the authentication server is down or connectivity between the device and the authentication server is not operable. The console or emergency account logon credentials must be stored in a sealed envelope and kept in a safe.Information Assurance OfficerECSC-1

Terminating an idle session within a short time period reduces the window of opportunity for unauthorized personnel to take control of a management session enabled on the console or console port that has been left unattended. In addition quickly terminating an idle session will also free up resources committed by the managed network element. Setting the timeout of the session to 10 minutes or less increases the level of protection afforded critical network components.Information Assurance OfficerECSC-1

Enabling write access to the device via SNMP provides a mechanism that can be exploited by an attacker to set configuration variables that can disrupt network operations.Information Assurance OfficerECSC-1

Remote administration is inherently dangerous because anyone with a sniffer and access to the right LAN segment, could acquire the device account and password information. With this intercepted information they could gain access to the infrastructure and cause denial of service attacks, intercept sensitive information, or perform other destructive actions.Information Assurance OfficerECSC-1

An attacker may attempt to connect to the device using SSH by guessing the authentication method, encryption algorithm, and keys. Limiting the amount of time allowed for authenticating and negotiating the SSH session reduces the window of opportunity for the malicious user attempting to make a connection to the network element.Information Assurance OfficerECSC-1

An attacker may attempt to connect to the device using SSH by guessing the authentication method and authentication key or shared secret. Setting the authentication retry to 3 or less strengthens against a Brute Force attack.Information Assurance OfficerECSC-1

A gratuitous ARP is an ARP broadcast in which the source and destination MAC addresses are the same. It is used to inform the network about a hosts IP address. A spoofed gratuitous ARP message can cause network mapping information to be stored incorrectly, causing network malfunction.Information Assurance OfficerECSC-1

A TCP connection consists of a three-way handshake message sequence. A connection request is transmitted by the originator, an acknowledgement is returned from the receiver, and then an acceptance of that acknowledgement is sent by the originator. An attacker’s goal in this scenario is to cause a denial of service to the network or device by initiating a high volume of TCP packets, then never sending an acknowledgement, leaving connections in a half-opened state. Without the device having a connection or time threshold for these half-opened sessions, the device risks being a victim of a denial of service attack. Setting a TCP timeout threshold will instruct the device to shut down any incomplete connections. Services such as SSH, BGP, SNMP, LDP, etc. are some services that may be prone to these types of denial of service attacks. If the router does not have any BGP connections with BGP neighbors across WAN links, values could be set to even tighter constraints.Information Assurance OfficerECSC-1

Berkeley Software Distribution (BSD) “r” commands allow users to execute commands on remote systems using a variety of protocols. The BSD "r" commands (e.g., rsh, rlogin, rcp, rdump, rrestore, and rdist) are designed to provide convenient remote access without passwords to services such as remote command execution (rsh), remote login (rlogin), and remote file copy (rcp and rdist). The difficulty with these commands is they use address-based authentication. An attacker who convinces a server that he is coming from a "trusted" machine can essentially get complete and unrestricted access to a system. The attacker can convince the server by impersonating a trusted machine and using IP address, by confusing DNS so that DNS thinks that the attacker's IP address maps to a trusted machine's name, or by any of a number of other methods.Information Assurance OfficerECSC-1

Since NTP is used to ensure accurate log file timestamp information, NTP could pose a security risk if a malicious user were able to falsify NTP information. To launch an attack on the NTP infrastructure, a hacker could inject time that would be accepted by NTP clients by spoofing the IP address of a valid NTP server. To mitigate this risk, the time messages must be authenticated by the client before accepting them as a time source. Two NTP-enabled devices can communicate in either client-server mode or peer-to-peer mode (aka “symmetric mode”). The peering mode is configured manually on the device and indicated in the outgoing NTP packets. The fundamental difference is the synchronization behavior: an NTP server can synchronize to a peer with better stratum, whereas it will never synchronize to its client regardless of the client’s stratum. From a protocol perspective, NTP clients are no different from the NTP servers. The NTP client can synchronize to multiple NTP servers, select the best server and synchronize with it, or synchronize to the averaged value returned by the servers. A hierarchical model can be used to improve scalability. With this implementation, an NTP client can also become an NTP server providing time to downstream clients at a higher stratum level and of decreasing accuracy than that of its upstream server. To increase availability, NTP peering can be used between NTP servers. In the event the device looses connectivity to it upstream NTP server, it will be able to choose time from one of its peers. The NTP authentication model is opposite of the typical client-server authentication model. NTP authentication enables an NTP client or peer to authenticate time received from their servers and peers. It’s not used to authenticate NTP clients because NTP servers don’t care about the authenticity of their clients, as they never accept any time from them. Information Assurance OfficerECSC-1

SSH Version 1 is a protocol that has never been defined in a standard. Since SSH-1 has inherent design flaws which make it vulnerable to attacks, e.g., man-in-the-middle attacks, it is now generally considered obsolete and should be avoided by explicitly disabling fallback to SSH-1. Information Assurance OfficerECSC-1

The use of Authentication, Authorization, and Accounting (AAA) affords the best methods for controlling user access, authorization levels, and activity logging. By enabling AAA on the routers in conjunction with an authentication server such as TACACS+ or RADIUS, the administrators can easily add or remove user accounts, add or remove command authorizations, and maintain a log of user activity. The use of an authentication server provides the capability to assign router administrators to tiered groups that contain their privilege level that is used for authorization of specific commands. For example, user mode would be authorized for all authenticated administrators while configuration or edit mode should only be granted to those administrators that are permitted to implement router configuration changes. Information Assurance OfficerIAIA-1

The OOBM access switch will connect to the management interface of the managed network elements. The management interface of the managed network element will be directly connected to the OOBM network. An OOBM interface does not forward transit traffic; thereby, providing complete separation of production and management traffic. Since all management traffic is immediately forwarded into the management network, it is not exposed to possible tampering. The separation also ensures that congestion or failures in the managed network do not affect the management of the device. If the OOBM interface does not have an IP address from the managed network address space, it will not have reachability from the NOC using scalable and normal control plane and forwarding mechanisms.System AdministratorInformation Assurance OfficerECSC-1

The OOBM access switch will connect to the management interface of the managed network elements. The management interface can be a true OOBM interface or a standard interface functioning as the management interface. In either case, the management interface of the managed network element will be directly connected to the OOBM network. An OOBM interface does not forward transit traffic; thereby, providing complete separation of production and management traffic. Since all management traffic is immediately forwarded into the management network, it is not exposed to possible tampering. The separation also ensures that congestion or failures in the managed network do not affect the management of the device. If the device does not have an OOBM port, the interface functioning as the management interface must be configured so that management traffic does not leak into the managed network and that production traffic does not leak into the management network.System AdministratorInformation Assurance OfficerECSC-1

The Route Processor (RP) is critical to all network operations as it is the component used to build all forwarding paths for the data plane via control plane processes. It is also instrumental with ongoing network management functions that keep the routers and links available for providing network services. Hence, any disruption to the RP or the control and management planes can result in mission critical network outages. In addition to control plane and management plane traffic that is in the router’s receive path, the RP must also handle other traffic that must be punted to the RP—that is, the traffic must be fast or process switched. This is the result of packets that must be fragmented, require an ICMP response (TTL expiration, unreachable, etc.) have IP options, etc. A DoS attack targeting the RP can be perpetrated either inadvertently or maliciously involving high rates of punted traffic resulting in excessive RP CPU and memory utilization. To maintain network stability, the router must be able to securely handle specific control plane and management plane traffic that is destined to it, as well as other punted traffic. Using the ingress filter on forwarding interfaces is a method that has been used in the past to filter both forwarding path and receiving path traffic. However, this method does not scale well as the number of interfaces grows and the size of the ingress filters grow. Control plane policing can be used to increase security of routers and multilayer switches by protecting the RP from unnecessary or malicious traffic. Filtering and rate limiting the traffic flow of control plane packets can be implemented to protect routers against reconnaissance and DoS attacks allowing the control plane to maintain packet forwarding and protocol states despite an attack or heavy load on the router or multilayer switch. System AdministratorInformation Assurance OfficerECSC-1

Using shared secrets between two IPSec endpoints is easy to implement but are also easy to compromise. Regardless of the strength of the password, they can be cracked using software tools that are readily available. Furthermore, implementation using shared secrets is not scalable since all VPN gateways and software clients would need to be configured with the shared secrets. In addition, there cannot be a preshared key for every user because the VPN gateway server does not know the client’s identity (the IP address is commonly used). Hence, remote users must use a group-based preshared key for authentication. When an individual leaves the group, changing the key must be coordinated with the other users of the group. PKI mitigates the risk involved with group passwords because each user has a certificate. PKI offers a scalable way to authenticate all IPSec endpoints in a secure manner. Every VPN gateway or remote client that needs to participate in IPSec VPN is issued a digital certificate by the Certification Authority (CA). The digital certificate binds the identity information of a VPN gateway (e.g., hostname or IP address) to the device’s public key by means of digital signature. This involves the use of public key cryptography algorithms, such as RSA. Based on this binding, any device that trusts the CA certificate, i.e., trusts the signature of the CA, would accept the identity inside the signed certificate. This model enables all VPN gateways and clients that trust the same CA to authenticate each other. Information Assurance OfficerECSC-1

When using digital certificates, Internet Key Exchange (IKE) negotiation between peers is restricted by either manually configuring each peer with the public key for each peer to which it is allowed to connect, or enrolling each peer with a Certificate Authority (CA). All peers to which the peer is allowed to connect must enroll with the same CA server and belong to the same organization. Certificates are issued and signed by a CA. Hence, the signature on a certificate identifies the particular CA that issued a certificate. The CA in turn has a certificate that binds its identity to its public key, so the CA’s identity can be verified. The primary role of the CA is to digitally sign and publish the public key bound to a given user or device via a digital certificate. This is done using the CA's own private key, so that trust in the user’s key relies on trust in the validity of the CA's key. Hence, to establish trust in the certificate of the remote client or peer, the VPN gateway must be configured to validate the peer’s certificate with the DoD-approved CA, as well as validate the identity of the DoD-approved CA. If the peer’s certificate is not validated, there is a risk of establishing an IPSec Security Association with a malicious user or a remote client that is not authorized. Information Assurance OfficerECSC-1

Enabling the password save function requires users to only enter their password once when establishing the VPN tunnel. After that the software client will automatically re-enter the password when prompted for credentials by the VPN gateway.Information Assurance OfficerECSC-1

All software remote clients must present a DoD approved warning banner prior allowing access to VPN. The banner should warn any unauthorized user not to proceed. It also should provide clear and unequivocal notice to both authorized and unauthorized personnel that access to the network is subject to monitoring to detect unauthorized usage. Failure to display the required warning banner prior to logon attempts will limit the ability to prosecute unauthorized access and also presents the potential to give rise to criminal and civil liability for systems administrators and information systems managers. DoD CIO has issued new, mandatory policy standardizing the wording of “notice and consent” banners and matching user agreements for all Secret and below DoD information systems, including stand-alone systems by releasing DoD CIO Memo, “Policy on Use of Department of Defense (DoD) Information Systems Standard Consent Banner and User Agreement”, dated 9 May 2008. The banner is mandatory and deviations are not permitted except as authorized in writing by the Deputy Assistant Secretary of Defense for Information and Identity Assurance. Implementation of this banner verbiage is further directed to all DoD components for all DoD assets via USCYBERCOM CTO 08-008A.Information Assurance OfficerECSC-1

Situations may arise in which the certificate issued by a Certificate Authority (CA) may need to be revoked before the lifetime of the certificate expires. For example, the certificate is known to have been compromised. To achieve this, a list of certificates that have been revoked, known as a Certificate Revocation List (CRL), is sent periodically from the CA to the IPSec gateway. When an incoming Internet Key Exchange (IKE) session is initiated for a remote client or peer whose certificate is revoked, the CRL will be checked to see if the certificate is valid; if the certificate is revoked, IKE will fail and an IPSec security association will not be established for the remote end-point.Information Assurance OfficerECSC-1

The security posture of the remote PC connecting to the enclave via VPN is vital to the overall security of the enclave. While on-site hosts are behind the enclave’s perimeter defense, a remote PC is not and therefore is exposed to many vulnerabilities existing in the Internet when connected to a service provider via dial-up or broadband connection. Though it is policy to have a firewall installed on the remote PC according to the Secure Remote Computing Endpoint STIG (SRC-EPT-405), it is imperative the VPN gateway enforce the policy to the software client to verify the firewall is active prior to enabling access to the VPN.Information Assurance OfficerECSC-1

A VPN hardware or software client with split tunneling enabled provides an unsecured backdoor to the enclave from the Internet. With split tunneling enabled, a remote client has access to the Internet while at the same time has established a secured path to the enclave via an IPSec tunnel. A remote client connected to the Internet that has been compromised by an attacker in the Internet, provides an attack base to the enclave’s private network via the IPSec tunnel. Hence, it is imperative that the VPN gateway enforces a no split-tunneling policy to all remote clients.Information Assurance OfficerECSC-1

A VPN gateway peer at the remote site provides connectivity to the central or other remote sites belonging to the enclave via an IPSec tunnel across an IP backbone network such as the NIPRNet. This creates an extension or Intranet for the enclave using IPSec tunnels in lieu of traditional or legacy WAN services (T carrier, ATM, frame relay, etc). Unless the remote site has the required enclave perimeter defense (firewall, IPS, deny by default, etc), it is imperative that all inbound and outbound traffic traverse only the IPSec tunnels or other provisioned WAN links connecting the remote site to other sites belonging to the enclave. In other words, no packets can leak out an external-facing interface as “native” IP traffic into an IP backbone (i.e. NIPRNet, Internet). In addition, the external interface must not receive any traffic that is not secured by an IPSec tunnel or other provisioned WAN links connected to the central or remote site. This not only ensures that inbound and outbound traffic does not bypass the enclave’s perimeter defense, but also eliminates any backdoor connection. Information Assurance OfficerECSC-1

Packets from a remote client destined outbound must be inspected and proxied the same as any other traffic that will egress the enclave. Otherwise, there is the risk that the return traffic that will ingress the IPSec tunnel could compromise the remote client and possibly the remote LAN. This scenario can exist with a VPN-on-a-stick implementation that allows traffic to u-turn—that is, traffic from the remote site that traverses the IPSec tunnel is immediately forwarded out the same interface towards the NIPRNet and Internet with no upstream firewall. If a remote LAN is breached, the entire enclave could be exposed via the secured tunnel or any other provisioned link between the compromised remote LAN and other remote sites and the central site. Hence, it is imperative that traffic from the remote site that is destined outbound does not bypass the applicable inspection and proxy services deployed for the enclave’s perimeter defense. Information Assurance OfficerECSC-1

Replay attack is a type of injection attack when an IPSec packet is captured by an attacker and re-inserts it into the legitimate flow to disrupt service or create undesired behavior. IPSec anti-replay service can mitigate a replay attack by running sequence numbers for each end of the tunnel and incrementing it for each packet sent. If a packet that is received does not have the expected sequence number, it is dropped. Information Assurance OfficerECSC-1

The Internet Key Exchange (IKE) Phase-2 (Quick Mode) Security Association (SA) is used to create an IPSec session key. Hence, its rekey or key regeneration procedure is very important. The Phase-2 rekey can be performed with or without Perfect Forward Secrecy (PFS). With PFS, every time a new IPSec Security Association is negotiated during the Quick Mode, a new Diffie-Hellman (DH) exchange occurs. The new DH shared secret will be included with original keying material (SYKEID_d, initiator nonce, and responder nonce from Phase 1) for generating a new IPSec session key. If PFS is not used, the IPSec session key will always be completely dependent on the original keying material from the Phase-1. Hence, if an older key is compromised at any time, it is possible that all new keys may be compromised. The DH exchange is performed in the same manner as was done in Phase 1 (Main or Aggressive Mode). However, the Phase-2 exchange is protected by encrypting the Phase-2 packets with the key derived from the Phase-1 negotiation. Because DH negotiations during Phase-2 are encrypted, the new IPSec session key has an added element of secrecy. Information Assurance OfficerECSC-1

As specified in RFC 793, TCP utilizes sequence checking to ensure proper ordering of received packets. RFC 793 also specifies that RST (reset) control flags should be processed immediately, without waiting for out of sequence packets to arrive. RFC 793 also requires that sequence numbers are checked against the window size before accepting data or control flags as valid. A router receiving an RST segment will close the TCP session with the BGP peer that is being spoofed; thereby, purging all routes learned from that BGP neighbor. A RST segment is valid as long as the sequence number is within the window. The TCP reset attack is made possible due to the requirements that Reset flags should be processed immediately and that a TCP endpoint must accept out of order packets that are within the range of a window size. This reduces the number of sequence number guesses the attack must make by a factor equivalent to the active window size. Each sequence number guess made by the attacker can be simply incremented by the receiving connections window size. The BGP peering session can protect itself against such an attack by authenticating each TCP segment. The TCP header options include an MD5 signature in every packet and are checked prior to the acceptance and processing of any TCP packet—including RST flags. One way to create havoc in a network is to advertise bogus routes to a network. A rogue router could send a fictitious routing update to convince a BGP router to send traffic to an incorrect or rogue destination. This diverted traffic could be analyzed to learn confidential information of the site’s network, or merely used to disrupt the network’s ability to effectively communicate with other networks. An autonomous system can advertise incorrect information by sending BGP updates messages to routers in a neighboring AS. A malicious AS can advertise a prefix originated from another AS and claim that it is the originator (prefix hijacking). Neighboring autonomous systems receiving this announcement will believe that the malicious AS is the prefix owner and route packets to it.ECSC-1