IPv6 Implications for Network Scanning

One of the key differences between IPv4 and IPv6 is the much larger address space for IPv6, which also goes hand-in-hand with much larger subnet sizes. This change has a significant impact on the feasibility of TCP and UDP network scanning, whereby an automated process is run to detect open ports (services) on systems that may then be subject to a subsequent attack. Today many IPv4 sites are subjected to such probing on a recurring basis. Such probing is common in part due to the relatively dense population of active hosts in any given chunk of IPv4 address space. The 128 bits of IPv6 address space is considerably bigger than the 32 bits of address space in IPv4. In particular, the IPv6 subnets to which hosts attach will by default have 64 bits of host address space. As a result, traditional methods of remote TCP or UDP network scanning to discover open or running services on a host will potentially become less feasible, due to the larger search space in the subnet. Similarly, worms that rely on off-link network scanning to propagate may also potentially be more limited in impact. This document discusses this property of IPv6 and describes related issues for IPv6 site network administrators to consider, which may be useful when planning site address allocation and management strategies. For example, many worms, like Slammer, rely on such address scanning methods to propagate, whether they pick subnets numerically (and thus probably topologically) close to the current victim, or subnets in random remote networks. The nature of these worms may change, if detection of target hosts between sites or subnets is harder to achieve by traditional methods. However, there are other worms that propagate via methods such as email, for which the methods discussed in this text are not relevant. It must be remembered that the defence of a network must not rely solely on the unpredictable sparseness of the host addresses on that network. Such a feature or property is only one measure in a set of measures that may be applied. This document discusses various measures that can be used by a site to mitigate attacks as part of an overall strategy. Some of these have a lower cost to deploy than others. For example, if numbering hosts on a subnet, it may be as cheap to number hosts without any predictable pattern as it is to number them sequentially. In contrast, use of IPv6 privacy extensions may complicate network management (identifying which hosts use which addresses). This document complements the transition-centric discussion of the issues that can be found in Appendix A of "IPv6 Transition/Co-existence Security Considerations" , which takes a broad view of security issues for transitioning networks. The reader is also referred to a recent paper by Bellovin on network worm propagation strategies in IPv6 networks . This paper discusses some of the issues included in this document, from a slightly different perspective.

There are significantly different considerations for the feasibility of plain, brute-force IPv4 and IPv6 address scanning.

A typical IPv4 subnet may have 8 bits reserved for host addressing. In such a case, a remote attacker need only probe at most 256 addresses to determine if a particular service is running publicly on a host in that subnet. Even at only one probe per second, such a scan would take under 5 minutes to complete.

A typical IPv6 subnet will have 64 bits reserved for host addressing. In such a case, a remote attacker in principle needs to probe 2^64 addresses to determine if a particular open service is running on a host in that subnet. At a very conservative one probe per second, such a scan may take some 5 billion years to complete. A more rapid probe will still be limited to (effectively) infinite time for the whole address space. However, there are ways for the attacker to reduce the address search space to scan against within the target subnet, as we discuss below.

The IPv6 host address space through which an attacker may search can be reduced in at least two ways. First, the attacker may rely on the administrator conveniently numbering their hosts from [prefix]::1 upward. This makes scanning trivial, and thus should be avoided unless the host's address is readily obtainable from other sources (for example, it is the site's published primary DNS or email Mail Exchange (MX) server). Alternatively, if hosts are numbered sequentially, or using any regular scheme, knowledge of one address may expose other available addresses to scan. Second, in the case of statelessly autoconfiguring hosts, the host part of the address will usually take a well-known format that includes the Ethernet vendor prefix and the "fffe" stuffing. For such hosts, the search space can be reduced to 48 bits. Further, if the Ethernet vendor is also known, the search space may be reduced to 24 bits, with a one probe per second scan then taking a less daunting 194 days. Even where the exact vendor is not known, using a set of common vendor prefixes can reduce the search. In addition, many nodes in a site network may be procured in batches, and thus have sequential or near sequential Media Access Control (MAC) addresses; if one node's autoconfigured address is known, scanning around that address may yield results for the attacker. Again, any form of sequential host addressing should be avoided if possible.

Full advantage of the increased IPv6 address space in terms of resilience to network scanning may not be gained until IPv6-only networks and devices become more commonplace, given that most IPv6 hosts are currently dual stack, with (more readily scannable) IPv4 connectivity. However, many applications or services (e.g., new peer-to-peer applications) on the (dual-stack) hosts may emerge that are only accessible over IPv6, and that thus can only be discovered by IPv6 address scanning.

The problem faced by the attacker for an IPv6 network is also faced by a site administrator looking for vulnerabilities in their own network's systems. The administrator should have the advantage of being on-link for scanning purposes though.

If IPv6 hosts in subnets are allocated addresses 'randomly', and as a result IPv6 network scanning becomes relatively infeasible, attackers will need to find new methods to identify IPv6 addresses for subsequent scanning. In this section, we discuss some possible paths attackers may take. In these cases, the attacker will attempt to identify specific IPv6 addresses for subsequent targeted probes.

Note that in IPv6, an attacker would not be able to search across the entire IPv6 address space as they might in IPv4. An attacker may learn general prefixes to focus their efforts on by observing route view information (e.g., from public looking-glass services) or information on allocated address space from Regional Internet Registries (RIRs). In general, this would only yield information at most at the /48 prefix granularity, but specific /64 prefixes may be observed from route views on some parts of some networks.

Any servers that are DNS listed, e.g., MX mail relays, or web servers, will remain open to probing from the very fact that their IPv6 addresses will be published in the DNS. While servers are relatively easy to find because they are DNS-published, any systems that are not DNS-published will be much harder to locate via traditional scanning than is the case for IPv4 networks. It is worth noting that where a site uses sequential host numbering, publishing just one address may lead to a threat upon the other hosts.

In the IPv6 world, a DNS zone transfer is much more likely to narrow the number of hosts an attacker needs to target. This implies that restricting zone transfers is (more) important for IPv6, even if it is already good practice to restrict them in the IPv4 world. There are some projects that provide Internet mapping data from access to such transfers. Administrators may of course agree to provide such transfers where they choose to do so.

IPv6 addresses may be harvested from recorded logs, such as web site logs. Anywhere else where IPv6 addresses are explicitly recorded may prove a useful channel for an attacker, e.g., by inspection of the (many) Received from: or other header lines in archived email or Usenet news messages.

More recent peer-to-peer applications often include some centralised server that coordinates the transfer of data between peers. The BitTorrent application builds swarms of nodes that exchange chunks of files, with a tracker passing information about peers with available chunks of data between the peers. Such applications may offer an attacker a source of peer IP addresses to probe.

Where an Embedded Rendezvous Point (RP) multicast group address is known, the unicast address of the RP is implied by the group address. Where unicast-prefix-based multicast group addresses are used, specific /64 link prefixes may also be disclosed in traffic that goes off-site. An administrator may thus choose to put aside /64 bit prefixes for multicast group addresses that are not in use for normal unicast routing and addressing. Alternatively, a site may simply use their /48 site prefix allocation to generate RFC3306 multicast group addresses.

Specific knowledge of the target network may be gleaned if that attacker knows it is using 6to4, ISATAP, Teredo, or other techniques that derive low-order bits from IPv4 addresses (though in this case, unless they are using IPv4 NAT, the IPv4 addresses may be probed anyway). For example, the current Microsoft 6to4 implementation uses the address 2002:V4ADDR::V4ADDR while older Linux and FreeBSD implementations default to 2002:V4ADDR::1. This leads to specific knowledge of specific hosts in the network. Given one host in the network is observed as using a given transition technique, it is likely that there are more. In the case of Teredo, the 64-bit node identifier is generated from the IPv4 address observed at a Teredo server along with a UDP port number. The Teredo specification also allows for discovery of other Teredo clients on the same IPv4 subnet via a well-known IPv4 multicast address (see Section 2.17 of RFC 4380).

The main thrust of this text is considerations for off-link attackers or probing of a network. In general, once one host on a link is compromised, others on the link can be very readily discovered.

If the attacker already has access to a system on the current subnet, then traffic on that subnet, be it Neighbour Discovery or application-based traffic, can invariably be observed, and active node addresses within the local subnet learnt. In addition to making observations of traffic on the link, IPv6-enabled hosts on local subnets may be discovered through probing the "all hosts" link-local multicast address. Likewise, any routers on the subnet may be found via the "all routers" link-local multicast address. An attacker may choose to probe in a slightly more obfuscated way by probing the solicited node multicast address of a potential target host. Where a host has already been compromised, its Neighbour Discovery cache is also likely to include information about active nodes on the current subnet, just as an ARP cache would do for IPv4. Also, depending on the node, traffic to or from other nodes (in particular, server systems) is likely to show up if an attacker can gain a presence on a node in any one subnet in a site's network.

A site may also have site- or organisational-scope multicast configured, in which case application traffic, or service discovery, may be exposed site wide. An attacker may also choose to use any other service discovery methods supported by the site.

There are some tools that site administrators can apply to make the task for IPv6 network scanning attackers harder. These methods arise from the considerations in the previous section. The author notes that at his current (university) site, there is no evidence of general network scanning running across subnets. However, there is network scanning over IPv6 connections to systems whose IPv6 addresses are advertised (DNS servers, MX relays, web servers, etc.), which are presumably looking for other open ports on these hosts to probe. At the time of writing, DHCPv6 is not yet in use, and clients use stateless autoconfiguration. Therefore, the author's site does not yet have sequentially numbered client hosts deployed as may typically be seen in today's IPv4 DHCP-served networks.

Hosts in a network using IPv6 privacy extensions will typically only connect to external systems using their current (temporary) privacy address. The precise behaviour of a host with a stable global address and one or more dynamic privacy address(es) when selecting a source address to use may be operating-system-specific, or configurable, but typical behaviour when initiating a connection is use of a privacy address when available. While an attacker may be able to port scan a privacy address, if they do so quickly upon observing or otherwise learning of the address, the threat or risk is reduced due to the time-constrained value of the address. One implementation of RFC 4941 already deployed has privacy addresses active (used by the node) for one day, with such addresses reachable for seven days. Note that an RFC 4941 host will usually also have a separate static global IPv6 address by which it can also be reached, and that may be DNS-advertised if an externally reachable service is running on it. DHCPv6 can be used to serve normal global addresses and IPv6 privacy addresses. The implication is that while privacy addresses can mitigate the long-term value of harvested addresses, an attacker creating an IPv6 application server to which clients connect will still be able to probe the clients by their privacy address when they visit that server. The duration for which privacy addresses are valid will impact the usefulness of such observed addresses to an external attacker. For example, a worm that may spread using such observed addresses may be less effective if it relies on harvested privacy addresses. The frequency with which such address get recycled could be increased, though this may increase the complexity of local network management for the administrator, since doing so will cause more addresses to be used over time in the site. A further option here may be to consider using different addresses for specific applications, or even each new application instance, which may reduce exposure to other services running on the same host when such an address is observed externally.

The use of Cryptographically Generated Addresses (CGAs) may also cause the search space to be increased from that presented by default use of stateless autoconfiguration. Such addresses would be seen where Secure Neighbour Discovery (SEND) is in use.

The EUI-64 identifier format does not require the use of MAC addresses for identifier construction. At least one well known operating system currently defaults to generation of the 64-bit interface identifier by use of random bits, and thus does not embed the MAC address. Where such a method exists, an administrator may wish to consider using that option.

One option open to an administrator is to configure DHCPv6, if possible, so that the first addresses allocated from the pool begins much higher in the address space than at [prefix]::1. Further, it is desirable that allocated addresses are not sequential and do not have any predictable pattern to them. Unpredictable sparseness in the allocated addresses is a desirable property. DHCPv6 implementers could reduce the cost for administrators to deploy such 'random' addressing by supporting configuration options to allow such behaviour. DHCPv6 also includes an option to use privacy extension addresses, i.e., temporary addresses, as described in Section 12 of the DHCPv6 specification.

Due to the much larger size of IPv6 subnets in comparison to IPv4, it will become less feasible for traditional network scanning methods to detect open services for subsequent attacks, assuming the attackers are off-site and services are not listed in the DNS. If administrators number their IPv6 subnets in 'random', non-predictable ways, attackers, whether they be in the form of automated network scanners or dynamic worm propagation, will need to make wider use of new methods to determine IPv6 host addresses to target (e.g., looking to obtain logs of activity from a site and scanning addresses around the ones observed). Such numbering schemes may be very low cost to deploy in comparison to conventional sequential numbering, and thus, a useful part of an overall defence-in-depth strategy. Of course, if those systems are dual-stack, and have open IPv4 services running, they will remain exposed to traditional probes over IPv4 transport.

There are no specific security considerations in this document outside of the topic of discussion itself. However, it must be noted that the 'security through obscurity' discussions and commentary within this text must be noted in their proper context. Relying purely on obscurity of a node address is not prudent, rather the advice here should be considered as part of a 'defence-in-depth' approach to security for a site or network. This also implies that these measures require coordination between network administrators and those who maintain DNS services, though this is common in most scenarios.

Thanks are due to people in the 6NET project (www.6net.org) for discussion of this topic, including Pekka Savola, Christian Strauf, and Martin Dunmore, as well as other contributors from the IETF v6ops and other mailing lists, including Tony Finch, David Malone, Bernie Volz, Fred Baker, Andrew Sullivan, Tony Hain, Dave Thaler, and Alex Petrescu. Thanks are also due for editorial feedback from Brian Carpenter, Lars Eggert, and Jonne Soininen amongst others.