RFC 4942 - IPv6 Transition/Co-existence Security Considerations
Network Working Group E. Davies Request for Comments: 4942 Consultant Category: Informational S. Krishnan Ericsson P. Savola CSC/Funet September 2007 IPv6 Transition/Coexistence Security Considerations Status of This Memo This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited. Abstract The transition from a pure IPv4 network to a network where IPv4 and IPv6 coexist brings a number of extra security considerations that need to be taken into account when deploying IPv6 and operating the dual-protocol network and the associated transition mechanisms. This document attempts to give an overview of the various issues grouped into three categories: o issues due to the IPv6 protocol itself, o issues due to transition mechanisms, and o issues due to IPv6 deployment. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 2. Issues Due to IPv6 Protocol . . . . . . . . . . . . . . . . . 4 2.1. IPv6 Protocol-Specific Issues . . . . . . . . . . . . . . 5 2.1.1. Routing Headers and Hosts . . . . . . . . . . . . . . 5 2.1.2. Routing Headers for Mobile IPv6 and Other Purposes . . 6 2.1.3. Site-Scope Multicast Addresses . . . . . . . . . . . . 7 2.1.4. ICMPv6 and Multicast . . . . . . . . . . . . . . . . . 7 2.1.5. Bogus Errored Packets in ICMPv6 Error Messages . . . . 8 2.1.6. Anycast Traffic Identification and Security . . . . . 9 2.1.7. Address Privacy Extensions Interact with DDoS Defenses . . . . . . . . . . . . . . . . . . . . . . . 10 2.1.8. Dynamic DNS: Stateless Address Autoconfiguration, Privacy Extensions, and SEND . . . . . . . . . . . . . 10 2.1.9. Extension Headers . . . . . . . . . . . . . . . . . . 11 2.1.10. Fragmentation: Reassembly and Deep Packet Inspection . . . . . . . . . . . . . . . . . . . . . . 14 2.1.11. Fragmentation Related DoS Attacks . . . . . . . . . . 15 2.1.12. Link-Local Addresses and Securing Neighbor Discovery . . . . . . . . . . . . . . . . . . . . . . 16 2.1.13. Securing Router Advertisements . . . . . . . . . . . . 17 2.1.14. Host-to-Router Load Sharing . . . . . . . . . . . . . 18 2.1.15. Mobile IPv6 . . . . . . . . . . . . . . . . . . . . . 18 2.2. IPv4-Mapped IPv6 Addresses . . . . . . . . . . . . . . . . 19 2.3. Increased End-to-End Transparency . . . . . . . . . . . . 20 2.3.1. IPv6 Networks without NATs . . . . . . . . . . . . . . 20 2.3.2. Enterprise Network Security Model for IPv6 . . . . . . 21 2.4. IPv6 in IPv6 Tunnels . . . . . . . . . . . . . . . . . . . 22 3. Issues Due to Transition Mechanisms . . . . . . . . . . . . . 23 3.1. IPv6 Transition/Coexistence Mechanism-Specific Issues . . 23 3.2. Automatic Tunneling and Relays . . . . . . . . . . . . . . 23 3.3. Tunneling IPv6 through IPv4 Networks May Break IPv4 Network Security Assumptions . . . . . . . . . . . . . . . 24 4. Issues Due to IPv6 Deployment . . . . . . . . . . . . . . . . 26 4.1. Avoiding the Trap of Insecure IPv6 Service Piloting . . . 26 4.2. DNS Server Problems . . . . . . . . . . . . . . . . . . . 28 4.3. Addressing Schemes and Securing Routers . . . . . . . . . 28 4.4. Consequences of Multiple Addresses in IPv6 . . . . . . . . 28 4.5. Deploying ICMPv6 . . . . . . . . . . . . . . . . . . . . . 29 4.5.1. Problems Resulting from ICMPv6 Transparency . . . . . 30 4.6. IPsec Transport Mode . . . . . . . . . . . . . . . . . . . 30 4.7. Reduced Functionality Devices . . . . . . . . . . . . . . 31 4.8. Operational Factors when Enabling IPv6 in the Network . . 31 4.9. Security Issues Due to Neighbor Discovery Proxies . . . . 32 5. Security Considerations . . . . . . . . . . . . . . . . . . . 32 6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 32 7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 33 7.1. Normative References . . . . . . . . . . . . . . . . . . . 33 7.2. Informative References . . . . . . . . . . . . . . . . . . 34 Appendix A. IPv6 Probing/Mapping Considerations . . . . . . . . . 37 Appendix B. IPv6 Privacy Considerations . . . . . . . . . . . . . 38 B.1. Exposing MAC Addresses . . . . . . . . . . . . . . . . . . 38 B.2. Exposing Multiple Devices . . . . . . . . . . . . . . . . 39 B.3. Exposing the Site by a Stable Prefix . . . . . . . . . . . 39 1. Introduction The transition from a pure IPv4 network to a network where IPv4 and IPv6 coexist brings a number of extra security considerations that need to be taken into account when deploying IPv6 and operating the dual-protocol network with its associated transition mechanisms. This document attempts to give an overview of the various issues grouped into three categories: o issues due to the IPv6 protocol itself, o issues due to transition mechanisms, and o issues due to IPv6 deployment. It is important to understand that deployments are unlikely to be replacing IPv4 with IPv6 (in the short term), but rather will be adding IPv6 to be operated in parallel with IPv4 over a considerable period, so that security issues with transition mechanisms and dual stack networks will be of ongoing concern. This extended transition and coexistence period stems primarily from the scale of the current IPv4 network. It is unreasonable to expect that the many millions of IPv4 nodes will be converted overnight. It is more likely that it will take two or three capital equipment replacement cycles (between nine and 15 years) for IPv6 capabilities to spread through the network, and many services will remain available over IPv4 only for a significant period whilst others will be offered either just on IPv6 or on both protocols. To maintain current levels of service, enterprises and service providers will need to support IPv4 and IPv6 in parallel for some time. This document also describes two matters that have been wrongly identified as potential security concerns for IPv6 in the past and explains why they are unlikely to cause problems: considerations about probing/mapping IPv6 addresses (Appendix A) and considerations with respect to privacy in IPv6 (Appendix B). 2. Issues Due to IPv6 Protocol Administrators should be aware that some of the rules suggested in this section could potentially lead to a small amount of legitimate traffic being dropped because the source has made unusual and arguably unreasonable choices when generating the packet. The IPv6 specification [RFC2460] contains a number of areas where choices are available to packet originators that will result in packets that conform to the specification but are unlikely to be the result of a rational packet generation policy for legitimate traffic (e.g., sending a fragmented packet in a much larger than necessary number of small segments). This document highlights choices that could be made by malicious sources with the intention of damaging the target host or network, and suggests rules that try to differentiate specification-conforming packets that are legitimate traffic from conforming packets that may be trying to subvert the specification to cause damage. The differentiation tries to offer a reasonable compromise between securing the network and passing every possible conforming packet. To avoid loss of important traffic, administrators are advised to log packets dropped according to these rules and examine these logs periodically to ensure that they are having the desired effect, and are not excluding traffic inappropriately. The built-in flexibility of the IPv6 protocol may also lead to changes in the boundaries between legitimate and malicious traffic as identified by these rules. New options may be introduced in the future, and rules may need to be altered to allow the new capabilities to be (legitimately) exploited by applications. The document therefore recommends that filtering needs to be configurable to allow administrators the flexibility to update rules as new pieces of IPv6 specification are standardized. 2.1. IPv6 Protocol-Specific Issues There are significant differences between the features of IPv6 and IPv4: some of these specification changes may result in potential security issues. Several of these issues have been discussed in separate documents but are summarized here to avoid normative references that may not become RFCs. The following specification- related problems have been identified, but this is not necessarily a complete list. 2.1.1. Routing Headers and Hosts All IPv6 nodes must be able to process routing headers [RFC2460]. This RFC can be interpreted, although it is not explicitly stated, to mean that all nodes (including hosts) must have this processing enabled. The "Requirements for Internet Hosts" [RFC1122] permits implementations to perform "local source routing", that is, forwarding a packet with a routing header through the same interface on which it was received: no restrictions are placed on this operation even on hosts. In combination, these rules can result in hosts forwarding received traffic to another node if there are segments left in the Routing Header when it arrives at the host. A number of potential security issues associated with this behavior have been identified. Some of these issues have been resolved (a separate routing header (Type 2) has been standardized for Mobile IPv6 [RFC3775], and ICMP Traceback has not been standardized), but two issues remain: o Both existing types of routing header can be used to evade access controls based on destination addresses. This could be achieved by sending a packet ostensibly to a publicly accessible host address but with a routing header containing a 'forbidden' address. If the publicly accessible host is processing routing headers, it will forward the packet to the destination address in the routing header that would have been forbidden by the packet filters if the address had been in the destination field when the packet was checked. o If the packet source address can be spoofed when using a Type 0 routing header, the mechanism described in the previous bullet could be used with any host to mediate an anonymous reflection denial-of-service attack by having any publicly accessible host redirect the attack packets. (This attack cannot use Type 2 routing headers because the packet cannot be forwarded outside the host that processes the routing header, i.e., the original destination of the packet.) To counteract these threats, if a device is enforcing access controls based on destination addresses, it needs to examine both the destination address in the base IPv6 header and any waypoint destinations in a routing header that have not yet been reached by the packet at the point where it is being checked. Various forms of amplification attack on routers and firewalls using the routing header could be envisaged. A simple form involves repeating the address of a waypoint several times in the routing header. More complex forms could involve alternating waypoint addresses that would result in the packet re-transiting the router or firewall. These attacks can be counteracted by ensuring that routing headers do not contain the same waypoint address more than once, and performing ingress/egress filtering to check that the source address is appropriate to the destination: packets made to reverse their path will fail this test. 2.1.2. Routing Headers for Mobile IPv6 and Other Purposes In addition to the basic Routing Header (Type 0), which is intended to influence the trajectory of a packet through a network by specifying a sequence of router waypoints, Routing Header (Type 2) has been defined as part of the Mobile IPv6 specifications in [RFC3775]. The Type 2 Routing Header is intended for use by hosts to handle 'interface local' forwarding needed when packets are sent to the care-of address of a mobile node that is away from its home address. It is important that nodes treat the different types of routing header appropriately. It should be possible to apply separate filtering rules to the different types of Routing Header. By design, hosts must process Type 2 Routing Headers to support Mobile IPv6 but routers should not: to avoid the issues in Section 2.1.1, it may be desirable to forbid or limit the processing of Type 0 Routing Headers in hosts and some routers. Routing Headers are an extremely powerful and general capability. Alternative future uses of Routing Headers need to be carefully assessed to ensure that they do not open new avenues of attack that can be exploited. 2.1.3. Site-Scope Multicast Addresses IPv6 supports multicast addresses with site scope that can potentially allow an attacker to identify certain important resources on the site if misused. Particular examples are the 'all routers' (FF05::2) and 'all Dynamic Host Configuration Protocol (DHCP) servers' (FF05::1:3) addresses defined in [RFC2375]. An attacker that is able to infiltrate a message destined for these addresses on to the site will potentially receive in return information identifying key resources on the site. This information can then be the target of directed attacks ranging from simple flooding to more specific mechanisms designed to subvert the device. Some of these addresses have current legitimate uses within a site. The risk can be minimized by ensuring that all firewalls and site boundary routers are configured to drop packets with site-scope destination addresses. Also, nodes should not join multicast groups for which there is no legitimate use on the site, and site routers should be configured to drop packets directed to these unused addresses. 2.1.4. ICMPv6 and Multicast It is possible to launch a Denial-of-Service (DoS) attack using IPv6 that could be amplified by the multicast infrastructure. Unlike ICMP for IPv4, ICMPv6 [RFC4443] allows error notification responses to be sent when certain unprocessable packets are sent to multicast addresses. The cases in which responses are sent are: o The received packet is longer than the next link Maximum Transmission Unit (MTU): 'Packet Too Big' responses are needed to support Path MTU Discovery for multicast traffic. o The received packet contains an unrecognized option in a hop-by- hop or destination options extension header with the first two bits of the option type set to binary '10': 'Parameter Problem' responses are intended to inform the source that some or all of the recipients cannot handle the option in question. If an attacker can craft a suitable packet sent to a multicast destination, it may be possible to elicit multiple responses directed at the victim (the spoofed source of the multicast packet). On the other hand, the use of 'reverse path forwarding' checks (to eliminate loops in multicast forwarding) automatically limits the range of addresses that can be spoofed. In practice, an attack using oversize packets is unlikely to cause much amplification unless the attacker is able to carefully tune the packet size to exploit a network with smaller MTU in the edge than the core. Similarly, a packet with an unrecognized hop-by-hop option would be dropped by the first router without resulting in multiple responses. However, a packet with an unrecognized destination option could generate multiple responses. In addition to amplification, this kind of attack would potentially consume large amounts of forwarding state resources in routers on multicast-enabled networks. 2.1.5. Bogus Errored Packets in ICMPv6 Error Messages Apart from the spurious load on the network, routers, and hosts, bogus ICMPv6 error messages (types 0 to 127) containing a spoofed errored packet can impact higher-layer protocols when the alleged errored packet is referred to the higher layer at the destination of the ICMPv6 packet [RFC4443]. The potentially damaging effects on TCP connections, and some ways to mitigate the threats, are documented in [ICMP-ATT]. Specific countermeasures for particular higher-layer protocols are beyond the scope of this document, but firewalls may be able to help counter the threat by inspecting the alleged errored packet embedded in the ICMPv6 error message. Measures to mitigate the threat include: o The receiving host should test that the embedded packet is all or part of a packet that was transmitted by the host. o The firewall may be able to test that the embedded packet contains addresses that would have been legitimate (i.e., would have passed ingress/egress filtering) for a packet sent from the receiving host, but the possibility of asymmetric routing of the outgoing and returning packets may prevent this sort of test depending on the topology of the network, the location of the firewall, and whether state synchronization between firewalls is in use. o If the firewall is stateful and the test is not prevented by asymmetric routing, the firewall may also be able to check that the embedded packet is all or part of a packet that recently transited the firewall in the opposite direction. o Firewalls and destination hosts should be suspicious of ICMPv6 error messages with unnecessarily truncated errored packets (e.g., those that only carry the address fields of the IPv6 base header). The specification of ICMPv6 requires that error messages carry as much of the errored packet as possible (unlike ICMP for IPv4 which requires only a minimum amount of the errored packet) and IPv6 networks must have a guaranteed minimum MTU of 1280 octets. Accordingly, the ICMPv6 message should normally carry all the header fields of the errored packet, together with a significant amount of the payload, in order to allow robust comparison against the outgoing packet. 2.1.6. Anycast Traffic Identification and Security IPv6 introduces the notion of anycast addresses and services. Originally the IPv6 standards disallowed using an anycast address as the source address of a packet. Responses from an anycast server would therefore supply a unicast address for the responding server. To avoid exposing knowledge about the internal structure of the network, it is recommended that anycast servers now take advantage of the ability to return responses with the anycast address as the source address if possible. If the server needs to use a unicast address for any reason, it may be desirable to consider using specialized addresses for anycast servers, which are not used for any other part of the network, to restrict the information exposed. Alternatively, operators may wish to restrict the use of anycast services from outside the domain, thus requiring firewalls to filter anycast requests. For this purpose, firewalls need to know which addresses are being used for anycast services: these addresses are arbitrary and not distinguishable from any other IPv6 unicast address by structure or pattern. One particular class of anycast addresses that should be given special attention is the set of Subnet-Router anycast addresses defined in "IP Version 6 Addressing Architecture" [RFC4291]. All routers are required to support these addresses for all subnets for which they have interfaces. For most subnets using global unicast addresses, filtering anycast requests to these addresses can be achieved by dropping packets with the lower 64 bits (the Interface Identifier) set to all zeros. 2.1.7. Address Privacy Extensions Interact with DDoS Defenses The purpose of the privacy extensions for stateless address autoconfiguration [RFC4941] is to change the interface identifier (and hence the global scope addresses generated from it) from time to time. By varying the addresses used, eavesdroppers and other information collectors find it more difficult to identify which transactions actually relate to a specific node. A security issue may result from this if the frequency of node address change is sufficiently great to achieve the intended aim of the privacy extensions: with a relatively high rate of change, the observed behavior has some characteristics of a node or nodes involved in a Distributed Denial-of-Service (DDoS) attack. It should be noted, however, that addresses created in this way are topologically correct and that the other characteristics of the traffic may reveal that there is no malicious intent. This issue can be addressed in most cases by tuning the rate of change in an appropriate manner. Note that even if a node is well behaved, a change in the address could make it harder for a security administrator to define an address-based policy rule (e.g., access control list). However, nodes that employ privacy addresses do not have to use them for all communications. 2.1.8. Dynamic DNS: Stateless Address Autoconfiguration, Privacy Extensions, and SEND The introduction of Stateless Address Autoconfiguration (SLAAC) [RFC2462] with IPv6 provides an additional challenge to the security of Dynamic Domain Name System (DDNS). With manual addressing or the use of DHCP, the number of security associations that need to be maintained to secure access to the Domain Name Service (DNS) server is limited, assuming any necessary updates are carried out by the DHCP server. This is true equally for IPv4 and IPv6. Since SLAAC does not make use of a single and potentially trusted DHCP server, but depends on the node obtaining the address, securing the insertion of updates into DDNS may need a security association between each node and the DDNS server. This is discussed further in [RFC4472]. Using the Privacy Extensions to SLAAC [RFC4941] may significantly increase the rate of updates of DDNS. Even if a node using the Privacy Extensions does not publish its address for 'forward' lookup (as that would effectively compromise the privacy that it is seeking), it may still need to update the reverse DNS records. If the reverse DNS records are not updated, servers that perform reverse DNS checks will not accept connections from the node and it will not be possible to gain access to IP Security (IPsec) keying material stored in DNS [RFC4025]. If the rate of change needed to achieve real privacy has to be increased (see Section 2.1.7), the update rate for DDNS may be excessive. Similarly, the cryptographically generated addresses used by SEND [RFC3971] are expected to be periodically regenerated in line with recommendations for maximum key lifetimes. This regeneration could also impose a significant extra load on DDNS. 2.1.9. Extension Headers A number of security issues relating to IPv6 Extension headers have been identified. Several of these are a result of ambiguous or incomplete specification in the base IPv6 specification [RFC2460]. 22.214.171.124. Processing Extension Headers in Middleboxes In IPv4, deep packet inspection techniques are used to implement policing and filtering both as part of routers and in middleboxes such as firewalls. Fully extending these techniques to IPv6 would require inspection of all the extension headers in a packet. This is essential to ensure that policy constraints on the use of certain headers and options are enforced and to remove, at the earliest opportunity, packets containing potentially damaging unknown options. This requirement appears to conflict with Section 4 of the IPv6 specification in [RFC2460] which requires that only hop-by-hop options are processed at any node through which the packet passes until the packet reaches the appropriate destination (either the final destination or a routing header waypoint). Also, [RFC2460] forbids processing the headers other than in the order in which they appear in the packet. A further ambiguity relates to whether an intermediate node should discard a packet that contains a header or destination option which it does not recognize. If the rules above are followed slavishly, it is not (or may not be) legitimate for the intermediate node to discard the packet because it should not be processing those headers or options. Therefore, [RFC2460] does not appear to take account of the behavior of middleboxes and other non-final destinations that may be inspecting the packet, and thereby potentially limits the security protection of these boxes. Firewall vendors and administrators may choose to ignore these rules in order to provide enhanced security as this does not appear to have any serious consequences with the currently defined set of extensions. However, administrators should be aware that future extensions might require different treatment. 126.96.36.199. Processing Extension Header Chains There is a further problem for middleboxes that want to examine the transport headers that are located at the end of the IPv6 header chain. In order to locate the transport header or other protocol data unit, the node has to parse the header chain. The IPv6 specification [RFC2460] does not mandate the use of the Type-Length-Value (TLV) format with a fixed layout for the start of each header although it is used for the majority of headers currently defined. (Only the Type field is guaranteed in size and offset.) Therefore, a middlebox cannot guarantee to be able to process header chains that may contain headers defined after the box was manufactured. As discussed further in Section 188.8.131.52, middleboxes ought not to have to know the detailed layout of all header types in use but still need to be able to skip over such headers to find the transport payload start. If this is not possible, it either limits the security policy that can be applied in firewalls or makes it difficult to deploy new extension header types. At the time of writing, only the Fragment Header does not fully conform to the TLV format used for other extension headers. In practice, many firewalls reconstruct fragmented packets before performing deep packet inspection, so this divergence is less problematic than it might have been, and is at least partially justified because the full header chain is not present in all fragments. Hop-by-hop and destination options may also contain unknown options. However, the options are required to be encoded in TLV format so that intermediate nodes can skip over them during processing, unlike the enclosing extension headers. 184.108.40.206. Unknown Headers/Destination Options and Security Policy A strict security policy might dictate that packets containing either unknown headers or destination options are discarded by firewalls or other filters. This requires the firewall to process the whole extension header chain, which may be currently in conflict with the IPv6 specification as discussed in Section 220.127.116.11. Even if the firewall does inspect the whole header chain, it may not be sensible to discard packets with items unrecognized by the firewall: the intermediate node has no knowledge of which options and headers are implemented in the destination node and IPv6 has been deliberately designed to be extensible through adding new header options. This poses a dilemma for firewall administrators. On the one hand, admitting packets with 'unknown' options is a security risk, but dropping them may disable a useful new extension. The best compromise appears to be to select firewalls that provide a configurable discard policy based on the types of the extensions. Then, if a new extension is standardized, administrators can reconfigure firewalls to pass packets with legitimate items that they would otherwise not recognize because their hardware or software is not aware of a new definition. Provided that the new extensions conform to the TLV layout followed by current extensions, the firewall would not need detailed knowledge of the function or layout of the extension header. 18.104.22.168. Excessive Hop-by-Hop Options IPv6 does not limit the number of hop-by-hop options that can be present in a hop-by-hop option header, and any option can appear multiple times. The lack of a limit and the provision of extensibility bits that force nodes to ignore classes of options that they do not understand can be used to mount denial-of-service attacks affecting all nodes on a path. A packet with large numbers of unknown hop-by-hop options will be processed at every node through which it is forwarded, consuming significant resources to determine what action should be taken for each option. Current options with the exception of Pad1 and PadN should not appear more than once so that packets with inappropriately repeated options can be dropped. However, keeping track of which options have been seen adds complexity to firewalls and may not apply to future extensions. See Section 22.214.171.124 for a discussion of the advisability of dropping packets with unknown options in firewalls. 126.96.36.199. Misuse of Pad1 and PadN Options IPv6 allows multiple padding options of arbitrary sizes to be placed in both Hop-by-Hop and Destination option headers. PadN options are required to contain zero octets as 'payload'; there is, however, no incentive for receivers to check this. It may therefore be possible to use the 'payload' of padding options as a covert channel. Firewalls and receiving hosts should actively check that PadN only has zero octets in its 'payload'. There is no legitimate reason for padding beyond the next eight octet boundary since the whole option header is aligned on an eight-octet boundary but cannot be guaranteed to be on a 16 (or higher power of two)-octet boundary. The IPv6 specification allows multiple Pad1 and PadN options to be combined in any way that the source chooses to make up the required padding. Reasonable design choices would appear to be using however many Pad1 options (i.e., zero octets) are needed or using a single PadN option of the required size (from two up to seven octets). Administrators should consider at least logging unusual padding patterns, and may consider dropping packets that contain unusual patterns if they are certain of expected source behavior. 188.8.131.52. Overuse of Router Alert Option The IPv6 router alert option specifies a hop-by-hop option that, if present, signals the router to take a closer look at the packet. This can be used for denial-of-service attacks. By sending a large number of packets containing a router alert option, an attacker can deplete the processor cycles on the routers available to legitimate traffic. 2.1.10. Fragmentation: Reassembly and Deep Packet Inspection The current specifications of IPv6 in [RFC2460] do not mandate any minimum packet size for the fragments of a packet before the last one, except for the need to carry the unfragmentable part in all fragments. The unfragmentable part does not include the transport port numbers, so it is possible that the first fragment does not contain sufficient information to carry out deep packet inspection involving the port numbers. Packets with overlapping fragments are considered to be a major security risk, but the reassembly rules for fragmented packets in [RFC2460] do not mandate behavior that would minimize the effects of overlapping fragments. In order to ensure that deep packet inspection can be carried out correctly on fragmented packets, many firewalls and other nodes that use deep packet inspection will collect the fragments and reassemble the packet before examining it. Depending on the implementation of packet reassembly and the treatment of packet fragments in these nodes, the specification issues mentioned potentially leave IPv6 open to the sort of attacks described in [RFC1858] and [RFC3128] for IPv4. The following steps can be taken to mitigate these threats: o Although permitted in [RFC2460], there is no reason for a source to generate overlapping packet fragments, and overlaps could be prohibited in a future revision of the protocol specification. Firewalls should drop all packets with overlapped fragments: certain implementations both in firewalls and other nodes already drop such packets. o Specifying a minimum size for packet fragments does not help in the same way as it does for IPv4 because IPv6 extension headers can be made to appear very long: an attacker could insert one or more undefined destination options with long lengths and the 'ignore if unknown' bit set. Given the guaranteed minimum MTU of IPv6, it seems reasonable that hosts should be able to ensure that the transport port numbers are in the first fragment in almost all cases and that deep packet inspection should be very suspicious of first fragments that do not contain them (see also the discussion of fragment sizes in Section 2.1.11). 2.1.11. Fragmentation Related DoS Attacks Packet reassembly in IPv6 hosts also opens up the possibility of various fragment-related security attacks. Some of these are analogous to attacks identified for IPv4. Of particular concern is a DoS attack based on sending large numbers of small fragments without a terminating last fragment that would potentially overload the reconstruction buffers and consume large amounts of CPU resources. Mandating the size of packet fragments could reduce the impact of this kind of attack by limiting the rate at which fragments could arrive and limiting the number of fragments that need to be processed, but this is not currently specified by the IPv6 standard. In practice, reasonable design choices in protocol stacks are likely to either maximize the size of all fragments except the final one using the path MTU (most likely choice), or distribute the data evenly in the required minimum number of fragments. In either case, the smallest non-final fragment would be at least half the guaranteed minimum MTU (640 octets) -- the worst case arises when a payload is just too large for a single packet and is divided approximately equally between two packets. Administrators should consider configuring firewalls and hosts to drop non-final fragments smaller than 640 octets. 2.1.12. Link-Local Addresses and Securing Neighbor Discovery All IPv6 nodes are required to configure a link-local address on each interface. This address is used to communicate with other nodes directly connected to the link accessed via the interface, especially during the neighbor discovery and autoconfiguration processes. Link- local addresses are fundamental to the operation of the Neighbor Discovery Protocol (NDP) [RFC2461] and Stateless Address Autoconfiguration (SLAAC) [RFC2462]. NDP also provides the functionality of associating link-layer and IP addresses provided by the Address Resolution Protocol (ARP) in IPv4 networks. The standard version of NDP is subject to a number of security threats related to ARP spoofing attacks on IPv4. These threats are documented in [RFC3756], and mechanisms to combat them are specified in SEcure Neighbor Discovery (SEND) [RFC3971]. SEND is an optional mechanism that is particularly applicable to wireless and other environments where it is difficult to physically secure the link. Because the link-local address can, by default, be acquired without external intervention or control, it allows an attacker to commence communication on the link without needing to acquire information about the address prefixes in use or communicate with any authorities on the link. This feature gives a malicious node the opportunity to mount an attack on any other node that is attached to this link; this vulnerability exists in addition to possible direct attacks on NDP. Link-local addresses may also facilitate the unauthorized use of the link bandwidth ('bandwidth theft') to communicate with another unauthorized node on the same link. The vulnerabilities of IPv6 link-local addresses in NDP can be mitigated in several ways. A general solution will require o authenticating the link-layer connectivity, for example, by using IEEE 802.1X functionality [IEEE.802-1X] or physical security, and o using SEcure Neighbor Discovery (SEND) to create a cryptographically generated link-local address (as described in [RFC3971]) that is tied to the authenticated link-layer address. This solution would be particularly appropriate in wireless LAN deployments where it is difficult to physically secure the infrastructure, but it may not be considered necessary in wired environments where the physical infrastructure can be kept secure by other means. Limiting the potentiality for abuse of link-local addresses in general packet exchanges is more problematic because there may be circumstances, such as isolated networks, where usage is appropriate and discrimination between use and abuse requires complex filtering rules which have to be implemented on hosts. The risk of misuse may be deemed too small compared with the effort needed to control it, but special attention should be paid to tunnel end-points (see 2.4, 3.2, and 3.3). Any filtering has to be provided by a host-based or bridging firewall. In general, link-local addresses are expected to be used by applications that are written to deal with specific interfaces and links. Typically these applications are used for control and management. A node which is attached to multiple links has to deal with the potentially overlapping link-local address spaces associated with these links. IPv6 provides for this through zone identifiers that are used to discriminate between the different address scopes [RFC4007] and the scope identifier that can be associated with a socket address structure [RFC3493]. Most users are unfamiliar with these issues and general purpose applications are not intended to handle this kind of discrimination. link-local addresses are not normally used with the Domain Name System (DNS), and DNS cannot supply zone identifiers. If it is considered necessary to prevent the use of link-local addresses by means other than control and management protocols, administrators may wish to consider limiting the protocols that can be used with link-local addresses. At a minimum, ICMPv6 and any intra-domain routing protocol in use (such as Open Shortest Path First (OSPF) or Routing Information Protocol (RIP)) need to be allowed, but other protocols may also be needed. RIP illustrates the complexity of the filtering problem: its messages are encapsulated as User Datagram Protocol (UDP) payloads, and filtering needs to distinguish RIP messages addressed to UDP port 521 from other UDP messages. 2.1.13. Securing Router Advertisements As part of the Neighbor Discovery process, routers on a link advertise their capabilities in Router Advertisement messages. The version of NDP defined in [RFC2461] does not protect the integrity of these messages or validate the assertions made in the messages with the result that any node that connects to the link can maliciously claim to offer routing services that it will not fulfill, and advertise inappropriate prefixes and parameters. These threats have been documented in [RFC3756]. A malicious node may also be able to carry out a DoS attack by deprecating an established valid prefix (by advertising it with a zero lifetime). Similar DoS attacks are possible if the optional Router Selection mechanism is implemented as described in the security considerations of [RFC4191]. SEND [RFC3971] can be used to provide verification that routers are authorized to provide the services they advertise through a certificate-based mechanism. This capability of SEND is also particularly appropriate for wireless environments where clients are reliant on the assertions of the routers rather than a physically secured connection. 2.1.14. Host-to-Router Load Sharing If a host deploys the optional host-to-router load-sharing mechanism [RFC4311], a malicious application could carry out a DoS attack on one or more of the load-sharing routers if the application is able to use knowledge of the load-sharing algorithm to synthesize traffic that subverts the load-sharing algorithm and directs a large volume of bogus traffic towards a subset of the routers. The likelihood of such an attack can be reduced if the implementation uses a sufficiently sophisticated load sharing algorithm as described in the security considerations of [RFC4311]. 2.1.15. Mobile IPv6 Mobile IPv6 offers significantly enhanced security compared with Mobile IPv4 especially when using optimized routing and care-of addresses. Return routability checks are used to provide relatively robust assurance that the different addresses that a mobile node uses as it moves through the network do indeed all refer to the same node. The threats and solutions are described in [RFC3775], and a more extensive discussion of the security aspects of the design can be found in [RFC4225]. 184.108.40.206. Obsolete Home Address Option in Mobile IPv6 The Home Address option specified in early versions of Mobile IPv6 would have allowed a trivial source spoofing attack: hosts were required to substitute the source address of incoming packets with the address in the option, thereby potentially evading checks on the packet source address. The version of Mobile IPv6 as standardized in [RFC3775] has removed this issue by ensuring that the Home Address destination option is only processed if there is a corresponding binding cache entry and securing Binding Update messages. A number of pre-standard implementations of Mobile IPv6 were available that implemented this obsolete and insecure option: care should be taken to avoid running such obsolete systems. 2.2. IPv4-Mapped IPv6 Addresses Overloaded functionality is always a double-edged sword: it may yield some deployment benefits, but often also incurs the price that comes with ambiguity. One example of such is IPv4-mapped IPv6 addresses (::ffff/96): a representation of an IPv4 address as an IPv6 address inside an operating system as defined in [RFC3493]. Since the original specification, the use of IPv4-mapped addresses has been extended to a transition mechanism, Stateless IP/ICMP Translation algorithm (SIIT) [RFC2765], where they are potentially used in the addresses of packets on the wire. Therefore, it becomes difficult to unambiguously discern whether an IPv4 mapped address is really an IPv4 address represented in the IPv6 address format (basic API behavior) *or* an IPv6 address received from the wire (which may be subject to address forgery, etc.). (SIIT behavior). The security issues that arise from the ambiguous behavior when IPv4-mapped addresses are used on the wire include: o If an attacker transmits an IPv6 packet with ::ffff:127.0.0.1 in the IPv6 source address field, he might be able to bypass a node's access controls by deceiving applications into believing that the packet is from the node itself (specifically, the IPv4 loopback address, 127.0.0.1). The same attack might be performed using the node's IPv4 interface address instead. o If an attacker transmits an IPv6 packet with IPv4-mapped addresses in the IPv6 destination address field corresponding to IPv4 addresses inside a site's security perimeter (e.g., ::ffff: 10.1.1.1), he might be able to bypass IPv4 packet filtering rules and traverse a site's firewall. o If an attacker transmits an IPv6 packet with IPv4-mapped addresses in the IPv6 source and destination fields to a protocol that swaps IPv6 source and destination addresses, he might be able to use a node as a proxy for certain types of attacks. For example, this might be used to construct broadcast multiplication and proxy TCP port scan attacks. In addition, special cases like these, while giving deployment benefits in some areas, require a considerable amount of code complexity (e.g., in the implementations of bind() system calls and reverse DNS lookups) that is probably undesirable but can be managed in this case. In practice, although the packet translation mechanisms of SIIT are specified for use in "Network Address Translator - Protocol Translator (NAT-PT)" [RFC2766], NAT-PT uses a mechanism different from IPv4-mapped IPv6 addresses for communicating embedded IPv4 addresses in IPv6 addresses. Also, SIIT is not recommended for use as a standalone transition mechanism. Given the issues that have been identified, it seems appropriate that mapped addresses should not be used on the wire. However, changing application behavior by deprecating the use of mapped addresses in the operating system interface would have significant impact on application porting methods as described in [RFC4038], and it is expected that IPv4- mapped IPv6 addresses will continue to be used within the API to aid application portability. Using the basic API behavior has some security implications in that it adds additional complexity to address-based access controls. The main issue that arises is that an IPv6 (AF_INET6) socket will accept IPv4 packets even if the node has no IPv4 (AF_INET) sockets open. This has to be taken into account by application developers and may allow a malicious IPv4 peer to access a service even if there are no open IPv4 sockets. This violates the security principle of "least surprise". 2.3. Increased End-to-End Transparency One of the major design aims of IPv6 has been to maintain the original IP architectural concept of end-to-end transparency. Transparency can help foster technological innovation in areas such as peer-to-peer communication, but maintaining the security of the network at the same time requires some modifications in the network architecture. Ultimately, it is also likely to need changes in the security model as compared with the norms for IPv4 networks. 2.3.1. IPv6 Networks without NATs The necessity of introducing Network Address Translators (NATs) into IPv4 networks, resulting from a shortage of IPv4 addresses, has removed the end-to-end transparency of most IPv4 connections: the use of IPv6 would restore this transparency. However, the use of NATs, and the associated private addressing schemes, has become inappropriately linked to the provision of security in enterprise networks. The restored end-to-end transparency of IPv6 networks can therefore be seen as a threat by poorly informed enterprise network managers. Some seem to want to limit the end-to-end capabilities of IPv6, for example by deploying private, local addressing and translators, even when it is not necessary because of the abundance of IPv6 addresses. Recommendations for designing an IPv6 network to meet the perceived security and connectivity requirements implicit in the current usage of IPv4 NATs whilst maintaining the advantages of IPv6 end-to-end transparency are described in "IP Version 6 Network Architecture Protection" [RFC4864]. 2.3.2. Enterprise Network Security Model for IPv6 The favored model for enterprise network security in IPv4 stresses the use of a security perimeter policed by autonomous firewalls and incorporating the NATs. Both perimeter firewalls and NATs introduce asymmetry and reduce the transparency of communications through these perimeters. The symmetric bidirectionality and transparency that are extolled as virtues of IPv6 may seem to be at odds with this model. Consequently, network managers may even see them as undesirable attributes, in conflict with their need to control threats to and attacks on the networks they administer. It is worth noting that IPv6 does not *require* end-to-end connectivity. It merely provides end-to-end addressability; the connectivity can still be controlled using firewalls (or other mechanisms), and it is indeed wise to do so. A number of matters indicate that IPv6 networks should migrate towards an improved security model, which will increase the overall security of the network while at the same time facilitating end-to- end communication: o Increased usage of end-to-end security especially at the network layer. IPv6 mandates the provision of IPsec capability in all nodes, and increasing usage of end-to-end security is a challenge to current autonomous firewalls that are unable to perform deep packet inspection on encrypted packets. It is also incompatible with NATs because they modify the packets, even when packets are only authenticated rather than encrypted. o Acknowledgement that over-reliance on the perimeter model is potentially dangerous. An attacker who can penetrate today's perimeters will have free rein within the perimeter, in many cases. Also a successful attack will generally allow the attacker to capture information or resources and make use of them. o Development of mechanisms such as 'Trusted Computing' [TCGARCH] that will increase the level of trust that network managers are able to place on hosts. o Development of centralized security policy repositories and secure distribution mechanisms that, in conjunction with trusted hosts, will allow network managers to place more reliance on security mechanisms at the end-points. The mechanisms are likely to include end-node firewalling and intrusion detection systems as well as secure protocols that allow end-points to influence the behavior of perimeter security devices. o Review of the role of perimeter devices with increased emphasis on intrusion detection, and network resource protection and coordination to thwart distributed denial-of-service attacks. Several of the technologies required to support an enhanced security model are still under development, including secure protocols to allow end-points to control firewalls: the complete security model utilizing these technologies is now emerging but still requires some development. In the meantime, initial deployments will need to make use of similar firewalling and intrusion detection techniques to IPv4 that may limit end-to-end transparency temporarily, but should be prepared to use the new security model as it develops and avoid the use of NATs by the use of the architectural techniques described in [RFC4864]. In particular, using NAT-PT [RFC2766] as a general purpose transition mechanism should be avoided as it is likely to limit the exploitation of end-to-end security and other IPv6 capabilities in the future as explained in [RFC4966]. 2.4. IPv6 in IPv6 Tunnels IPv6 in IPv6 tunnels can be used to circumvent security checks, so it is essential to filter packets both at tunnel ingress and egress points (the encapsulator and decapsulator) to ensure that both the inner and outer addresses are acceptable, and the tunnel is not being used to carry inappropriate traffic. [RFC3964], which is primarily about the 6to4 transition tunneling mechanism (see Section 3.1), contains useful discussions of possible attacks and ways to counteract these threats. 3. Issues Due to Transition Mechanisms 3.1. IPv6 Transition/Coexistence Mechanism-Specific Issues The more complicated the IPv6 transition/coexistence becomes, the greater the danger that security issues will be introduced either o in the mechanisms themselves, o in the interaction between mechanisms, or o by introducing unsecured paths through multiple mechanisms. These issues may or may not be readily apparent. Hence, it would be desirable to keep the mechanisms simple (as few in number as possible and built from pieces as small as possible) to simplify analysis. One case where such security issues have been analyzed in detail is the 6to4 tunneling mechanism [RFC3964]. As tunneling has been proposed as a model for several more cases than are currently being used, its security properties should be analyzed in more detail. There are some generic dangers to tunneling: o It may be easier to avoid ingress filtering checks. o It is possible to attack the tunnel interface: several IPv6 security mechanisms depend on checking that Hop Limit equals 255 on receipt and that link-local addresses are used. Sending such packets to the tunnel interface is much easier than gaining access to a physical segment and sending them there. o Automatic tunneling mechanisms are typically particularly dangerous as there is no pre-configured association between end points. Accordingly, at the receiving end of the tunnel, packets have to be accepted and decapsulated from any source. Consequently, special care should be taken when specifying automatic tunneling techniques. 3.2. Automatic Tunneling and Relays Two mechanisms have been specified that use automatic tunneling and are intended for use outside a single domain. These mechanisms encapsulate the IPv6 packet directly in an IPv4 packet in the case of 6to4 [RFC3056] or in an IPv4 UDP packet in the case of Teredo [RFC4380]. In each case, packets can be sent and received by any similarly equipped nodes in the IPv4 Internet. As mentioned in Section 3.1, a major vulnerability in such approaches is that receiving nodes must allow decapsulation of traffic sourced from anywhere in the Internet. This kind of decapsulation function must be extremely well secured because of the wide range of potential sources. An even more difficult problem is how these mechanisms are able to establish communication with native IPv6 nodes or between the automatic tunneling mechanisms: such connectivity requires the use of some kind of "relay". These relays could be deployed in various locations such as: o all native IPv6 nodes, o native IPv6 sites, o in IPv6-enabled ISPs, or o just somewhere in the Internet. Given that a relay needs to trust all the sources (e.g., in the 6to4 case, all 6to4 routers) that are sending it traffic, there are issues in achieving this trust and at the same time scaling the relay system to avoid overloading a small number of relays. As authentication of such a relay service is very difficult to achieve, and particularly so in some of the possible deployment models, relays provide a potential vehicle for address spoofing, (reflected) denial-of-service attacks, and other threats. Threats related to 6to4 and measures to combat them are discussed in [RFC3964]. [RFC4380] incorporates extensive discussion of the threats to Teredo and measures to combat them. 3.3. Tunneling IPv6 through IPv4 Networks May Break IPv4 Network Security Assumptions NATs and firewalls have been deployed extensively in the IPv4 Internet, as discussed in Section 2.3. Operators who deploy them typically have some security/operational requirements in mind (e.g., a desire to block inbound connection attempts), which may or may not be misguided. The addition of tunneling can change the security model that such deployments are seeking to enforce. IPv6-over-IPv4 tunneling using protocol 41 is typically either explicitly allowed, or disallowed implicitly. Tunneling IPv6 over IPv4 encapsulated in UDP constitutes a more difficult problem as UDP must usually be allowed to pass through NATs and firewalls. Consequently, using UDP implies the ability to punch holes in NATs and firewalls although, depending on the implementation, this ability may be limited or only achieved in a stateful manner. In practice, the mechanisms have been explicitly designed to traverse both NATs and firewalls in a similar fashion. One possible view is that the use of tunneling is especially questionable in home and SOHO (small office/home office) environments where the level of expertise in network administration is typically not very high; in these environments, the hosts may not be as tightly managed as in others (e.g., network services might be enabled unnecessarily), leading to possible security break-ins or other vulnerabilities. Holes allowing tunneled traffic through NATs and firewalls can be punched both intentionally and unintentionally. In cases where the administrator or user makes an explicit decision to create the hole, this is less of a problem, although (for example) some enterprises might want to block IPv6 tunneling explicitly if employees were able to create such holes without reference to administrators. On the other hand, if a hole is punched transparently, it is likely that a proportion of users will not understand the consequences: this will very probably result in a serious threat sooner or later. When deploying tunneling solutions, especially tunneling solutions that are automatic and/or can be enabled easily by users who do not understand the consequences, care should be taken not to compromise the security assumptions held by the users. For example, NAT traversal should not be performed by default unless there is a firewall producing a similar by-default security policy to that provided by IPv4 NAT. IPv6-in-IPv4 (protocol 41) tunneling is less of a problem, as it is easier to block if necessary; however, if the host is protected in IPv4, the IPv6 side should be protected as well. As is shown in Appendix A, it is relatively easy to determine the IPv6 address corresponding to an IPv4 address in tunneling deployments. It is therefore vital NOT to rely on "security by obscurity", i.e., assuming that nobody is able to guess or determine the IPv6 address of the host especially when using automatic tunneling transition mechanisms. The network architecture must provide separate IPv4 and IPv6 firewalls with tunneled IPv6 traffic arriving encapsulated in IPv4 packets routed through the IPv4 firewall before being decapsulated, and then through the IPv6 firewall as shown in Figure 1. +--------+ +--------+ +--------+ Site | Native | IPv6 |v6 in v4| IPv4 | Native | Public Network <--->| IPv6 |<---->| Tunnel |<---->| IPv4 |<---> Internet |Firewall| |Endpoint| |Firewall| +--------+ +--------+ +--------+ Figure 1: Tunneled Traffic and Firewalls 4. Issues Due to IPv6 Deployment 4.1. Avoiding the Trap of Insecure IPv6 Service Piloting Because IPv6 is a new service for many networks, network managers will often opt to make a pilot deployment in a part of the network to gain experience and understand the problems as well as the benefits that may result from a full production quality IPv6 service. Unless IPv6 service piloting is done in a manner that is as secure as possible, there is a risk that if security in the pilot does not match up to what is achievable with current IPv4 production service, the comparison can adversely impact the overall assessment of the IPv6 pilot deployment. This may result in a decision to delay or even avoid deploying an IPv6 production service. For example, hosts and routers might not be protected by IPv6 firewalls, even if the corresponding IPv4 service is fully protected by firewalls. The use of tunneling transition mechanisms (see Section 3.3) and the interaction with virtual private networks also need careful attention to ensure that site security is maintained. This is particularly critical where IPv6 capabilities are turned on by default in new equipment or new releases of operating systems: network managers may not be fully aware of the security exposure that this creates. In some cases, a perceived lack of availability of IPv6 firewalls and other security capabilities, such as intrusion detection systems may have led network managers to resist any kind of IPv6 service deployment. These problems may be partly due to the relatively slow development and deployment of IPv6-capable security equipment, but the major problems appear to have been a lack of information, and more importantly a lack of documented operational experience on which managers can draw. In actual fact, at the time of writing, there are a significant number of alternative IPv6 packet filters and firewalls already in existence that could be used to provide sufficient access controls. However, there are a small number of areas where the available equipment and capabilities may still be a barrier to secure deployment as of the time of writing: o 'Personal firewalls' with support for IPv6 and intended for use on hosts are not yet widely available. o Enterprise firewalls are at an early stage of development and may not provide the full range of capabilities needed to implement the necessary IPv6 filtering rules. Network managers often expect the same devices that support and are used for IPv4 today to also become IPv6-capable -- even though this is not really required and the equipment may not have the requisite hardware capabilities to support fast packet filtering for IPv6. Suggestions for the appropriate deployment of firewalls are given in Section 3.3 -- as will be seen from this section, it is usually desirable that the firewalls are in separate boxes, and there is no necessity for them to be same the model of equipment. o A lesser factor may be that some design decisions in the IPv6 protocol make it more difficult for firewalls to be implemented and work in all cases, and to be fully future-proof (e.g., when new extension headers are used) as discussed in Section 2.1.9. It is significantly more difficult for intermediate nodes to process the IPv6 header chains than IPv4 packets. o Adequate Intrusion Detection Systems (IDS) are more difficult to construct for IPv6. IDSs are now beginning to become available but the pattern-based mechanisms used for IPv4 may not be the most appropriate for long-term development of these systems as end-to- end encryption becomes more prevalent. Future systems may be more reliant on traffic flow pattern recognition. o Implementations of high availability capabilities supporting IPv6 are also in short supply. In particular, development of the IPv6 version of the Virtual Router Redundancy Protocol (VRRP) [VRRP] has lagged the development of the main IPv6 protocol although alternatives may be available for some environments. In all these areas, developments are ongoing and they should not be considered a long-term bar to the deployment of IPv6 either as a pilot or production service. The necessary tools are now available to make a secure IPv6 deployment, and with careful selection of components and design of the network architecture, a successful pilot or production IPv6 service can be deployed. Recommendations for secure deployment and appropriate management of IPv6 networks can be found in the documentation archives of the European Union 6net project [SIXNET] and in the Deployment Guide published by the IPv6 Promotion Council of Japan [JpIPv6DC]. 4.2. DNS Server Problems Some DNS server implementations have flaws that severely affect DNS queries for IPv6 addresses as discussed in [RFC4074]. These flaws can be used for DoS attacks affecting both IPv4 and IPv6 by inducing caching DNS servers to believe that a domain is broken and causing the server to block access to all requests for the domain for a precautionary period. 4.3. Addressing Schemes and Securing Routers Whilst in general terms brute force scanning of IPv6 subnets is essentially impossible due to the enormously larger address space of IPv6 and the 64-bit interface identifiers (see Appendix A), this will be obviated if administrators do not take advantage of the large space to use unguessable interface identifiers. Because of the unmemorability of complete IPv6 addresses, there is a temptation for administrators to use small integers as interface identifiers when manually configuring them, as might happen on point- to-point links or when provisioning complete addresses from a DHCPv6 server. Such allocations make it easy for an attacker to find active nodes that they can then port scan. To make use of the larger address space properly, administrators should be very careful when entering IPv6 addresses in their configurations (e.g., access control lists), since numerical IPv6 addresses are more prone to human error than IPv4 due to their length and unmemorability. It is also essential to ensure that the management interfaces of routers are well secured (e.g., allowing remote access using Secure Shell (SSH) only and ensuring that local craft interfaces have non- default passwords) as the router will usually contain a significant cache of neighbor addresses in its neighbor cache. 4.4. Consequences of Multiple Addresses in IPv6 One positive consequence of IPv6 is that nodes that do not require global access can communicate locally just by the use of a link-local address (if very local access is sufficient) or across the site by using a Unique Local Address (ULA). In either case it is easy to ensure that access outside the assigned domain of activity can be controlled by simple filters (which should be the default for link- locals). However, the security hazards of using link-local addresses for general purposes, as documented in Section 2.1.12, should be borne in mind. On the other hand, the possibility that a node or interface can have multiple global scope addresses makes access control filtering (both on ingress and egress) more complex and requires higher maintenance levels. Vendors and network administrators need to be aware that multiple addresses are the norm rather than the exception in IPv6: when building and selecting tools for security and management, a highly desirable feature is the ability to be able to 'tokenize' access control lists and configurations in general to cater for multiple addresses and/or address prefixes. The addresses could be from the same network prefix (for example, privacy mechanisms [RFC4941] will periodically create new addresses taken from the same prefix, and two or more of these may be active at the same time), or from different prefixes (for example, when a network is multihomed, when for management purposes a node belongs to several subnets on the same link or is implementing anycast services). In all these cases, it is possible that a single host could be using several different addresses with different prefixes and/or different interface identifiers. It is desirable that the security administrator be able to identify that the same host is behind all these addresses. Some network administrators may find the mutability of addresses when privacy mechanisms are used in their network to be undesirable because of the current difficulties in maintaining access control lists and knowing the origin of traffic. In general, disabling the use of privacy addresses is only possible if the full stateful DHCPv6 mechanism [RFC3315] is used to allocate IPv6 addresses and DHCPv6 requests for privacy addresses are not honored. 4.5. Deploying ICMPv6 In IPv4 it is commonly accepted that some filtering of ICMP packets by firewalls is essential to maintain security. Because of the extended use that is made of ICMPv6 [RFC2461] with a multitude of functions, the simple set of dropping rules that are usually applied in IPv4 need to be significantly developed for IPv6. The blanket dropping of all ICMP messages that is used in some very strict environments is simply not possible for IPv6. In an IPv6 firewall, policy needs to allow some messages through the firewall but also has to permit certain messages to and from the firewall, especially those with link-local sources on links to which the firewall is attached. These messages must be permitted to ensure that Neighbor Discovery [RFC2462], Multicast Listener Discovery ([RFC2710], [RFC3810]), and Stateless Address Configuration [RFC4443] work as expected. Recommendations for filtering ICMPv6 messages can be found in [RFC4890]. 4.5.1. Problems Resulting from ICMPv6 Transparency As described in Section 4.5, certain ICMPv6 error packets need to be passed through a firewall in both directions. This means that some ICMPv6 error packets can be exchanged between inside and outside without any filtering. Using this feature, malicious users can communicate between the inside and outside of a firewall, thus bypassing the administrator's inspection (proxy, firewall, etc.). For example, it might be possible to carry out a covert conversation through the payload of ICMPv6 error messages or to tunnel inappropriate encapsulated IP packets in ICMPv6 error messages. This problem can be alleviated by filtering ICMPv6 errors using a stateful packet inspection mechanism to ensure that the packet carried as a payload is associated with legitimate traffic to or from the protected network. 4.6. IPsec Transport Mode IPsec provides security to end-to-end communications at the network layer (layer 3). The security features available include access control, connectionless integrity, data origin authentication, protection against replay attacks, confidentiality, and limited traffic flow confidentiality (see [RFC4301] Section 2.1). IPv6 mandates the implementation of IPsec in all conforming nodes, making the usage of IPsec to secure end-to-end communication possible in a way that is generally not available to IPv4. To secure IPv6 end-to-end communications, IPsec transport mode would generally be the solution of choice. However, use of these IPsec security features can result in novel problems for network administrators and decrease the effectiveness of perimeter firewalls because of the increased prevalence of encrypted packets on which the firewalls cannot perform deep packet inspection and filtering. One example of such problems is the lack of security solutions in the middlebox, including effective content-filtering, ability to provide DoS prevention based on the expected TCP protocol behavior, and intrusion detection. Future solutions to this problem are discussed in Section 2.3.2. Another example is an IPsec-based DoS (e.g., sending malformed ESP/AH packets) that can be especially detrimental to software-based IPsec implementations. 4.7. Reduced Functionality Devices With the deployment of IPv6 we can expect the attachment of a very large number of new IPv6-enabled devices with scarce resources and low computing capacity. The resource limitations are generally because of a market requirement for cost reduction. Although the [RFC4294] specifies some mandatory security capabilities for every conformant node, these do not include functions required for a node to be able to protect itself. Accordingly, some such devices may not be able even to perform the minimum set of functions required to protect themselves (e.g., 'personal' firewall, automatic firmware update, enough CPU power to endure DoS attacks). This means a different security scheme may be necessary for such reduced functionality devices. 4.8. Operational Factors when Enabling IPv6 in the Network There are a number of reasons that make it essential to take particular care when enabling IPv6 in the network equipment: Initially, IPv6-enabled router software may be less mature than current IPv4-only implementations, and there is less experience with configuring IPv6 routing, which can result in disruptions to the IPv6 routing environment and (IPv6) network outages. IPv6 processing may not happen at (near) line speed (or at a comparable performance level to IPv4 in the same equipment). A high level of IPv6 traffic (even legitimate, e.g., Network News Transport Protocol, NNTP) could easily overload IPv6 processing especially when it is software-based without the hardware support typical in high-end routers. This may potentially have deleterious knock-on effects on IPv4 processing, affecting availability of both services. Accordingly, if people don't feel confident enough in the IPv6 capabilities of their equipment, they will be reluctant to enable it in their "production" networks. Sometimes essential features may be missing from early releases of vendors' software; an example is provision of software enabling IPv6 telnet/SSH access (e.g., to the configuration application of a router), but without the ability to turn it off or limit access to it! Sometimes the default IPv6 configuration is insecure. For example, in one vendor's implementation, if you have restricted IPv4 telnet to only a few hosts in the configuration, you need to be aware that IPv6 telnet will be automatically enabled, that the configuration commands used previously do not block IPv6 telnet, that IPv6 telnet is open to the world by default, and that you have to use a separate command to also lock down the IPv6 telnet access. Many operator networks have to run interior routing protocols for both IPv4 and IPv6. It is possible to run them both in one routing protocol, or have two separate routing protocols; either approach has its tradeoffs [RFC4029]. If multiple routing protocols are used, one should note that this causes double the amount of processing when links flap or recalculation is otherwise needed -- which might more easily overload the router's CPU, causing slightly slower convergence time. 4.9. Security Issues Due to Neighbor Discovery Proxies In order to span a single subnet over multiple physical links, a new experimental capability is being trialed in IPv6 to proxy Neighbor Discovery messages. A node with this capability will be called an NDProxy (see [RFC4389]). NDProxies are susceptible to the same security issues as those faced by hosts using unsecured Neighbor Discovery or ARP. These proxies may process unsecured messages, and update the neighbor cache as a result of such processing, thus allowing a malicious node to divert or hijack traffic. This may undermine the advantages of using SEND [RFC3971]. If a form of NDProxy is standardized, SEND will need to be extended to support this capability. 5. Security Considerations This memo attempts to give an overview of security considerations of the different aspects of IPv6, particularly as they relate to the transition to a network in which IPv4- and IPv6-based communications need to coexist. 6. Acknowledgements This document draws together the work of many people who have contributed security-related documents to the IPV6 and V6OPS working groups. Alain Durand, Alain Baudot, Luc Beloeil, Sharon Chisholm, Tim Chown, Lars Eggert, Andras Kis-Szabo, Vishwas Manral, Janos Mohacsi, Mark Smith, Alvaro Vives, and Margaret Wassermann provided feedback to improve this document. Satoshi Kondo, Shinsuke Suzuki, and Alvaro Vives provided additional inputs in cooperation with the Deployment Working Group of the Japanese IPv6 Promotion Council and the Euro6IX IST co-funded project, together with inputs from Jordi Palet, Brian Carpenter, and Peter Bieringer. Michael Wittsend and Michael Cole discussed issues relating to probing/mapping and privacy. Craig Metz and Jun-ichiro itojun Hagino did the original work identifying the problems of using IPv4-mapped IPv6 addresses on the wire. Vishwas Manral made further investigations of the impact of tiny fragments on IPv6 security. Francis Dupont raised the issues relating to IPv6 Privacy Addresses. Finally, Pekka Savola wrote a number of documents on aspects IPv6 security which have been subsumed into this work. His document on "Firewalling Considerations for IPv6" (October 2003) originally identified many of the issues with the base IPv6 specification which are documented here. 7. References 7.1. Normative References [RFC1122] Braden, R., "Requirements for Internet Hosts - Communication Layers", STD 3, RFC 1122, October 1989. [RFC2375] Hinden, R. and S. Deering, "IPv6 Multicast Address Assignments", RFC 2375, July 1998. [RFC2460] Deering, S. and R. Hinden, "Internet Protocol, Version 6 (IPv6) Specification", RFC 2460, December 1998. [RFC2461] Narten, T., Nordmark, E., and W. Simpson, "Neighbor Discovery for IP Version 6 (IPv6)", RFC 2461, December 1998. [RFC2462] Thomson, S. and T. Narten, "IPv6 Stateless Address Autoconfiguration", RFC 2462, December 1998. [RFC2710] Deering, S., Fenner, W., and B. Haberman, "Multicast Listener Discovery (MLD) for IPv6", RFC 2710, October 1999. [RFC3056] Carpenter, B. and K. Moore, "Connection of IPv6 Domains via IPv4 Clouds", RFC 3056, February 2001. [RFC3775] Johnson, D., Perkins, C., and J. Arkko, "Mobility Support in IPv6", RFC 3775, June 2004. [RFC3810] Vida, R. and L. Costa, "Multicast Listener Discovery Version 2 (MLDv2) for IPv6", RFC 3810, June 2004. [RFC3964] Savola, P. and C. Patel, "Security Considerations for 6to4", RFC 3964, December 2004. [RFC4007] Deering, S., Haberman, B., Jinmei, T., Nordmark, E., and B. Zill, "IPv6 Scoped Address Architecture", RFC 4007, March 2005. [RFC4291] Hinden, R. and S. Deering, "IP Version 6 Addressing Architecture", RFC 4291, February 2006. [RFC4380] Huitema, C., "Teredo: Tunneling IPv6 over UDP through Network Address Translations (NATs)", RFC 4380, February 2006. [RFC4443] Conta, A., Deering, S., and M. Gupta, "Internet Control Message Protocol (ICMPv6) for the Internet Protocol Version 6 (IPv6) Specification", RFC 4443, March 2006. [RFC4941] Narten, T., Draves, R., and S. Krishnan, "Privacy Extensions for Stateless Address Autoconfiguration in IPv6", RFC 4941, September 2007. 7.2. Informative References [FNAT] Bellovin, S., "Technique for Counting NATted Hosts", Proc. Second Internet Measurement Workshop , November 2002, <http://www.research.att.com/~smb/papers/fnat.pdf>. [ICMP-ATT] Gont, F., "ICMP attacks against TCP", Work in Progress, May 2007. [IEEE.802-1X] Institute of Electrical and Electronics Engineers, "Port-Based Network Access Control", IEEE Standard for Local and Metropolitan Area Networks 802.1X-2004, December 2004. [JpIPv6DC] Deployment WG, "IPv6 Deployment Guideline (2005 Edition)", IPv6 Promotion Council (Japan) Deployment Working Group, 2005, <http://www.v6pc.jp/>. [RFC1858] Ziemba, G., Reed, D., and P. Traina, "Security Considerations for IP Fragment Filtering", RFC 1858, October 1995. [RFC2765] Nordmark, E., "Stateless IP/ICMP Translation Algorithm (SIIT)", RFC 2765, February 2000. [RFC2766] Tsirtsis, G. and P. Srisuresh, "Network Address Translation - Protocol Translation (NAT-PT)", RFC 2766, February 2000. [RFC3128] Miller, I., "Protection Against a Variant of the Tiny Fragment Attack (RFC 1858)", RFC 3128, June 2001. [RFC3315] Droms, R., Bound, J., Volz, B., Lemon, T., Perkins, C., and M. Carney, "Dynamic Host Configuration Protocol for IPv6 (DHCPv6)", RFC 3315, July 2003. [RFC3493] Gilligan, R., Thomson, S., Bound, J., McCann, J., and W. Stevens, "Basic Socket Interface Extensions for IPv6", RFC 3493, February 2003. [RFC3756] Nikander, P., Kempf, J., and E. Nordmark, "IPv6 Neighbor Discovery (ND) Trust Models and Threats", RFC 3756, May 2004. [RFC3971] Arkko, J., Kempf, J., Zill, B., and P. Nikander, "SEcure Neighbor Discovery (SEND)", RFC 3971, March 2005. [RFC4025] Richardson, M., "A Method for Storing IPsec Keying Material in DNS", RFC 4025, March 2005. [RFC4029] Lind, M., Ksinant, V., Park, S., Baudot, A., and P. Savola, "Scenarios and Analysis for Introducing IPv6 into ISP Networks", RFC 4029, March 2005. [RFC4038] Shin, M-K., Hong, Y-G., Hagino, J., Savola, P., and E. Castro, "Application Aspects of IPv6 Transition", RFC 4038, March 2005. [RFC4074] Morishita, Y. and T. Jinmei, "Common Misbehavior Against DNS Queries for IPv6 Addresses", RFC 4074, May 2005. [RFC4191] Draves, R. and D. Thaler, "Default Router Preferences and More-Specific Routes", RFC 4191, November 2005. [RFC4225] Nikander, P., Arkko, J., Aura, T., Montenegro, G., and E. Nordmark, "Mobile IP Version 6 Route Optimization Security Design Background", RFC 4225, December 2005. [RFC4294] Loughney, J., "IPv6 Node Requirements", RFC 4294, April 2006. [RFC4301] Kent, S. and K. Seo, "Security Architecture for the Internet Protocol", RFC 4301, December 2005. [RFC4311] Hinden, R. and D. Thaler, "IPv6 Host-to-Router Load Sharing", RFC 4311, November 2005. [RFC4389] Thaler, D., Talwar, M., and C. Patel, "Neighbor Discovery Proxies (ND Proxy)", RFC 4389, April 2006. [RFC4472] Durand, A., Ihren, J., and P. Savola, "Operational Considerations and Issues with IPv6 DNS", RFC 4472, April 2006. [RFC4864] Van de Velde, G., Hain, T., Droms, R., Carpenter, B., and E. Klein, "Local Network Protection for IPv6", RFC 4864, May 2007. [RFC4890] Davies, E. and J. Mohacsi, "Recommendations for Filtering ICMPv6 Messages in Firewalls", RFC 4890, May 2007. [RFC4966] Aoun, C. and E. Davies, "Reasons to Move NAT-PT to Historic Status", RFC 4966, July 2007. [SCAN-IMP] Chown, T., "IPv6 Implications for Network Scanning", Work in Progress, March 2007. [SIXNET] 6Net, "Large Scale International IPv6 Pilot Network", EU Information Society Technologies Project , 2005, <http://www.6net.org/>. [TCGARCH] The Trusted Computing Group, "TCG Specification Architecture Overview", April 2004, <https:// www.trustedcomputinggroup.org/groups/ TCG_1_0_Architecture_Overview.pdf>. [VRRP] Hinden, R. and J. Cruz, "Virtual Router Redundancy Protocol for IPv6", Work in Progress, March 2007. Appendix A. IPv6 Probing/Mapping Considerations One school of thought wanted the IPv6 numbering topology (either at network or node level) to match IPv4 as exactly as possible, whereas others see IPv6 as giving more flexibility to the address plans, not wanting to constrain the design of IPv6 addressing. Mirroring the address plans is now generally seen as a security threat because an IPv6 deployment may have different security properties from IPv4. Given the relatively immature state of IPv6 network security, if an attacker knows the IPv4 address of the node and believes it to be dual-stacked with IPv4 and IPv6, he might want to try to probe the corresponding IPv6 address, based on the assumption that the security defenses might be lower. This might be the case particularly for nodes which are behind a NAT in IPv4, but globally addressable in IPv6. Naturally, this is not a concern if similar and adequate security policies are in place. On the other hand, brute-force scanning or probing of addresses is computationally infeasible due to the large search space of interface identifiers on most IPv6 subnets (somewhat less than 64 bits wide, depending on how identifiers are chosen), always provided that identifiers are chosen at random out of the available space, as discussed in [SCAN-IMP]. For example, automatic tunneling mechanisms typically use deterministic methods for generating IPv6 addresses, so probing/ port-scanning an IPv6 node is simplified. The IPv4 address is embedded at least in 6to4, Teredo, and ISATAP addresses. Additionally, it is possible (in the case of 6to4 in particular) to learn the address behind the prefix; for example, Microsoft 6to4 implementation uses the address 2002:V4ADDR::V4ADDR while older Linux and FreeBSD implementations default to 2002:V4ADDR::1. This could also be used as one way to identify an implementation and hence target any specific weaknesses. One proposal has been to randomize the addresses or subnet identifier in the address of the 6to4 router. This does not really help, as the 6to4 router (whether a host or a router) will return an ICMPv6 Hop Limit Exceeded message, revealing the IP address. Hosts behind the 6to4 router can use methods such as privacy addresses [RFC4941] to conceal themselves, provided that they are not meant to be reachable by sessions started from elsewhere; they would still require a globally accessible static address if they wish to receive communications initiated elsewhere. To conclude, it seems that when an automatic tunneling mechanism is being used, given an IPv4 address, the corresponding IPv6 address could possibly be guessed with relative ease. This has significant implications if the IPv6 security policy is less adequate than that for IPv4. Appendix B. IPv6 Privacy Considerations The generation of IPv6 addresses from MAC addresses potentially allows the behavior of users to be tracked in a way which may infringe their privacy. [RFC4941] specifies mechanisms which can be used to reduce the risk of infringement. It has also been claimed that IPv6 harms the privacy of the user, either by exposing the MAC address, or by exposing the number of nodes connected to a site. Additional discussion of privacy issues can be found in [RFC4864]. B.1. Exposing MAC Addresses Using stateless address autoconfiguration results in the MAC address being incorporated in an EUI64 that exposes the model of network card. The concern has been that a user might not want to expose the details of the system to outsiders, e.g., fearing a resulting burglary if a thief identifies expensive equipment from the vendor identifier embedded in MAC addresses, or allowing the type of equipment in use to be identified, thus facilitating an attack on specific security weaknesses. In most cases, this seems completely unfounded. First, such an address must be learned somehow -- this is a non-trivial process; the addresses are visible, e.g., in Web site access logs, but the chances that a random Web site owner is collecting this kind of information (or whether it would be of any use) are quite slim. Being able to eavesdrop the traffic to learn such addresses (e.g., by the compromise of DSL (Digital Subscriber Line) or Cable modem physical media) seems also quite far-fetched. Further, using statically configured interface identifiers or privacy addresses [RFC4941] for such purposes is straightforward if worried about the risk. Second, the burglar would have to be able to map the IP address to the physical location; typically this would only be possible with information from the private customer database of the Internet Service Provider (ISP) and, for large sites, the administrative records of the site, although some physical address information may be available from the WHOIS database of Internet registries. B.2. Exposing Multiple Devices Another concern that has been aired involves the user wanting to conceal the presence of a large number of computers or other devices connected to a network; NAT can "hide" all this equipment behind a single address, but it is not perfect either [FNAT]. One practical reason why some administrators may find this desirable is being able to thwart certain ISPs' business models. These models require payment based on the number of connected computers, rather than the connectivity as a whole. Similar feasibility issues as described above apply. To a degree, the number of machines present could be obscured by the sufficiently frequent reuse of privacy addresses [RFC4941] -- that is, if during a short period, dozens of generated addresses seem to be in use, it's difficult to estimate whether they are generated by just one host or multiple hosts. B.3. Exposing the Site by a Stable Prefix When an ISP provides IPv6 connectivity to its customers, including home or consumer users, it delegates a fixed global routing prefix (usually a /48) to them. This is in contrast to the typical IPv4 situation where home users typically receive a dynamically allocated address that may be stable only for a period of hours. Due to this fixed allocation, it is easier to correlate the global routing prefix to a network site. With consumer users, this correlation leads to a privacy issue, since a site is often equivalent to an individual or a family in such a case. Consequently some users might be concerned about being able to be tracked based on their /48 allocation if it is static [RFC4941]. On the other hand, many users may find having a static allocation desirable as it allows them to offer services hosted in their network more easily. This situation is not affected even if a user changes his/her interface ID or subnet ID, because malicious users can still discover this binding. On larger sites, the situation can be mitigated by using "untraceable" IPv6 addresses as described in [RFC4864], and it is possible that in the future ISPs might be prepared to offer untraceable addresses to their consumer customers to minimize the privacy issues. This privacy issue is common to both IPv4 and IPv6 and is inherent in the use of IP addresses as both identifiers for node interfaces and locators for the nodes. Authors' Addresses Elwyn B. Davies Consultant Soham, Cambs UK Phone: +44 7889 488 335 EMail: email@example.com Suresh Krishnan Ericsson 8400 Decarie Blvd. Town of Mount Royal, QC H4P 2N2 Canada Phone: +1 514-345-7900 EMail: firstname.lastname@example.org Pekka Savola CSC/Funet EMail: email@example.com Full Copyright Statement Copyright (C) The IETF Trust (2007). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Intellectual Property The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79. Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at firstname.lastname@example.org.