Hash-Based Addresses (HBA)

In order to preserve inter-domain routing system scalability, IPv6 sites obtain addresses from their Internet Service Providers (ISPs). Such an addressing strategy significantly reduces the amount of routes in the global routing tables, since each ISP only announces routes to its own address blocks, rather than announcing one route per customer site. However, this addressing scheme implies that multihomed sites will obtain multiple prefixes, one per ISP. Moreover, since each ISP only announces its own address block, a multihomed site will be reachable through a given ISP if the ISP prefix is contained in the destination address of the packets. This means that, if an established communication needs to be routed through different ISPs during its lifetime, addresses with different prefixes will have to be used. Changing the address used to carry packets of an established communication exposes the communication to numerous attacks, as described in , so security mechanisms are required to provide the required protection to the involved parties. This memo describes a tool that can be used to provide protection against some of the potential attacks, in particular against future/premeditated attacks (aka time shifting attacks in ). This memo describes a mechanism to provide a secure binding between the multiple addresses with different prefixes available to a host within a multihomed site. It should be noted that, as opposed to the mobility case where the addresses that will be used by the mobile node are not known a priori, the multiple addresses available to a host within the multihomed site are pre-defined and known in advance in most of the cases. The mechanism proposed in this memo employs either Cryptographically Generated Addresses (CGAs) or a new variant of the same theme that uses the same format in the addresses. The new variant, Hash-Based Address (HBA), takes advantage of the address set stability. In either case, a secure binding between the addresses of a node in a multihomed site can be provided. CGAs employ public key cryptography and can deal with changing address sets. HBAs employ only symmetric key cryptography, and have smaller computational requirements. For the purposes of the Shim6 protocol, the other characteristics of the CGAs and HBAs are similar. Both can be generated by the host itself without any reliance on external infrastructure. Both employ the same format of addresses and same format of data fed to generate the addresses. It is not required that all interface identifiers of a node's addresses be equal, preserving some degree of privacy through changes in the addresses used during the communications. The main idea in HBAs is that information about the multiple prefixes is included within the addresses themselves. This is achieved by generating the interface identifiers of the addresses of a host as hashes of the available prefixes and a random number. Then, the multiple addresses are obtained by prepending the different prefixes to the generated interface identifiers. The result is a set of addresses that are inherently bound. A cost-efficient mechanism is available to determine if two addresses belong to the same set, since given the prefix set and the additional parameters used to generate the HBA, a single hash operation is enough to verify if an HBA belongs to a given HBA set.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 .

The threat analysis for the multihoming problem is described in . This analysis basically identifies attacks based on redirection of packets by a malicious attacker towards addresses that do not belong to the multihomed node. There are essentially two types of redirection attacks: communication hijacking and flooding attacks. Communication hijacking attacks are about an attacker stealing on-going and/or future communications from a victim. Flooding attacks are about redirecting the traffic generated by a legitimate source towards a third party, flooding it. The HBA solution provides full protection against the communication hijacking attacks. The Shim6 protocol protects against flooding attacks. Residual threats are described in the "Security Considerations" section.

The basic goal of the HBA mechanism is to securely bind together multiple IPv6 addresses that belong to the same multihomed host. This allows rerouting of traffic without worrying that the communication is being redirected to an attacker. The technique that is used is to include a hash of the permitted prefixes in the low&nbhy;order bits of the IPv6 address. So, eliding some details, say the available prefixes are A, B, C, and D, the host would generate a prefix list P consisting of (A,B,C,D) and a random number called Modifier M. Then it would generate the new addresses: A || H(M || A || P) B || H(M || B || P) C || H(M || C || P) D || H(M || D || P) Thus, given one valid address out of the group and the prefix list P and the random Modifier M it is possible to determine whether another address is part of the group by computing the hash and checking against the low-order bits.

The design of the HBA technique was driven by the following considerations: First of all, the goal of HBA is to provide a secure binding between the IPv6 address used as an identifier by the upper-layer protocols and the alternative locators available in the multihomed node so that redirection attacks are prevented. Second, in order to achieve such protection, the selected approach was to include security information in the identifier itself, instead of relying on third trusted parties to secure the binding, such as the ones based on repositories or Public Key Infrastructure. This decision was driven by deployment considerations, i.e., the cost of deploying the trusted third-party infrastructure. Third, application support considerations described in resulted in selecting routable IPv6 addresses to be used as identifiers. Hence, security information is stuffed within the interface identifier part of the IPv6 address. Fourth, performance considerations as described in motivated the usage of a hash-based approach as opposed to a public-key-based approach based on pure Cryptographic Generated Addresses (CGA), in order to avoid imposing the performance of public key operations for every communication in multihomed environments. The HBA approach presented in this document presents a cheaper alternative that is attractive to many common usage cases. Note that the HBA approach and the CGA approaches are not mutually exclusive and that it is possible to generate addresses that are both valid CGA and HBA addresses providing the benefits of both approaches if needed.

As described in the previous section, the HBA technique uses the interface identifier part of the IPv6 address to encode information about the multiple prefixes available to a multihomed host. However, the interface identifier is also used to carry cryptographic information when Cryptographic Generated Addresses (CGAs) are used. Therefore, conflicting usages of the interface identifier bits may result if this is not taken into account during the HBA design. There are at least two valid reasons to provide CGA-HBA compatibility: First, the current Secure Neighbor Discovery (SeND) specification uses the CGAs defined in to prove address ownership. If HBAs are not compatible with CGAs, then nodes using HBAs for multihoming wouldn't be able to do Secure Neighbor Discovery using the same addresses (at least the parts of SeND that require CGAs). This would imply that nodes would have to choose between security (from SeND) and fault tolerance (from IPv6 multihoming support provided by the Shim6 protocol ). In addition to SeND, there are other protocols that are considered to benefit from the advantages offered by the CGA scheme, such as mobility support protocols . Those protocols could not be used with HBAs if HBAs are not compatible with CGAs. Second, CGAs provide additional features that cannot be achieved using only HBAs. In particular, because of its own nature, the HBA technique only supports a predetermined prefix set that is known at the time of the generation of the HBA set. No additions of new prefixes to this original set are supported after the HBA set generation. In most of the cases relevant for site multihoming, this is not a problem because the prefix set available to a multihomed set is not very dynamic. New prefixes may be added in a multihomed site when a new ISP is available, but the timing of those events are rarely in the same time scale as the lifetime of established communications. It is then enough for many situations that the new prefix is not available for established communications and that only new communications benefit from it. However, in the case that such functionality is required, it is possible to use CGAs to provide it. This approach clearly requires that HBA and CGA approaches be compatible. If this is the case, it then would be possible to create HBA/CGA addresses that support CGA and HBA functionality simultaneously. The inputs to the HBA/CGA generation process will be both a prefix set and a public key. In this way, a node that has established a communication using one address of the CGA/HBA set can tell its peer to use the HBA verification when one of the addresses of its HBA/CGA set is used as locator in the communication or to use CGA (public-/private-key-based) verification when a new address that does not belong to the HBA/CGA set is used as locator in the communication. So, because of the aforementioned reasons, it is a goal of the HBA design to define HBAs in such a way that they are compatible with CGAs as defined in and their usages described in (consequently, to understand the rest of this note, the reader should be familiar with the CGA specification defined in ). This means that it must be possible to generate addresses that are both an HBA and a CGA, i.e., that the interface identifier contains cryptographic information of CGA and the prefix-set information of an HBA. The CGA specification already considers the possibility of including additional information into the CGA generation process through the usage of Extension Fields in the CGA Parameter Data Structure. It is then possible to define a Multi-Prefix Extension for CGA so that the prefix set information is included in the interface identifier generation process. In this example, only the HBA capabilities of the Host1 addresses were used. In other words, neither the public key included in the CGA Parameter Data Structure nor its correspondent private key was used in the protocol. In the following section, we will consider a case where its usage is required.

In the previous section, we have presented the mechanisms that allow a host to use different addresses of a predetermined set to exchange packets of a communication. The set of addresses involved was predetermined and known when the communication was initiated. To achieve such functionality, only HBA functionalities of the addresses were needed. In this section, we will explore the case where the goal is to exchange packets using additional addresses that were not known when the communication was established. An example of such a situation is when a new prefix is available in a site after a renumbering event. In this case, the hosts that have the new address available may want to use it in communications that were established before the renumbering event. In this case, HBA functionalities of the addresses are not enough and CGA capabilities are to be used. Consider then the previous case of the communication between Host1 and Host2. Suppose that the communication is up and running, as described earlier. Host1 is using PA:LA:iidA and Host2 is using IPHost2 to exchange packets. Now suppose that a new address, PC:LC:addC is available in Host1. Note that this address is just a regular IPv6 address, and it is neither an HBA nor a CGA. Host1 wants to use this new address in the existent communication with Host2. It should be noted that the HBA mechanism described in the previous section cannot be used to verify this new address, since this address does not belong to the HBA set (since the prefix was not available at the moment of the generation of the HBA set). This means that alternative verification mechanisms will be needed. In order to verify this new address, CGA capabilities of PA:LA:iidA are used. Note that the same address is used, only that the verification mechanism is different. So, if Host1 wants to use PC:LC:addC to exchange packets in the established communication, it will use the UPDATE message defined in the Shim6 protocol , conveying the new address, PC:LC:addC, and this message will be signed using the private key corresponding to the public key contained in CGA_PDS_A. When Host2 receives the message, it will verify the signature using the public key contained in the CGA Parameter Data Structure associated with the address used for establishing the communication, i.e., CGA_PDS_A and PA:LA:iidA, respectively. Once that the signature is verified, the new address (PC:LC:addC) can be used in the communication. In any case, a renumbering event has an impact on a site that is using the HBA technique. In particular, the new prefix added will not be included in the existing HBA set, so it is only possible to use the new prefix with the existing HBA set if CGA capabilities are used. While this is acceptable for the short term, in the long run, the site will need to renumber its HBA addresses. In order to do that, it will need to re-generate the HBA sets assigned to hosts including the new prefix in the prefix set, which will result in different addresses, not only because we need to add a new address with the new prefix, but also because the addresses with the existing prefixes will also change because of the inclusion of a new prefix in the prefix set. Moreover, since HBA addresses need to be generated locally, once these are generated after the renumbering event, the new address information needs to be conveyed to the DNS manager in case that such address information is to be published in the DNS (see DNS considerations section for more details).

HBA sets can be generated using any prefix set. Actually, the only particularity of the HBA is that they contain information about the prefix set in the interface identifier part of the address in the form of a hash, but no assumption about the properties of prefixes used for the HBA generation is made. This basically means that depending on the prefixes used for the HBA set generation, it may or may not be recommended to publish the resulting (HBA) addresses in the DNS. For instance, when Unique Local Address (ULA) prefixes are included in the HBA generation process, specific DNS considerations related to the local nature of the ULA should be taken into account and proper recommendations related to publishing such prefixes in the DNS should followed. Moreover, among its addresses, a given host can have some HBAs and some other IPv6 addresses. The consequence from this is that only HBA addresses will be bound together by the HBA technique, while other addresses would not be bound to the HBA set. This would basically mean that if one of the other addresses is used for initiating a Shim6 communication, it won't be possible to use the HBA technique to bind the address used with the HBA set. Furthermore, since HBA addresses are indistinguishable from other IPv6 addresses in their format, an initiator will not be able to distinguish, by merely looking at the different addresses, which ones belong to the HBA set and which ones do not, so alternative means would be required the initiator is supposed to use only HBA for establishing communications in the presence of non-HBA addresses in the DNS. In addition, it should be noted that the actual HBA values are a result of the HBA generation procedure, meaning that they cannot be arbitrarily chosen. This has an implication with respect to DNS management, because the party that generates the HBA address set needs to convey the address information to the DNS manager, so that the addresses are published and not the other way around. The situation is similar to regular CGA addresses and even to the case where stateless address autoconfiguration is used. In order to do that, it is possible to use Dynamic DNS updates or other proprietary tools. A similar consideration applies when the host wants to publish reverse-DNS entries. Since the host needs to generate its HBA addresses, it will need to convey the address information to the DNS manager so the proper reverse-DNS entry is populated in case it is needed. It should be noted that neither the Shim6 protocol nor the HBA technique rely on the reverse DNS for its proper functioning and the general reasons for requiring reverse-DNS population apply as for any other regular IPv6 address.

This document defines a new CGA Extension, the Multi-Prefix Extension. This extension has been assigned the CGA Extension Type value 0x0012.

The goal of HBAs is to create a group of addresses that are securely bound, so that they can be used interchangeably when communicating with a node. If there is no secure binding between the different addresses of a node, a number of attacks are enabled, as described in . In particular, it would be possible for an attacker to redirect the communications of a victim to an address selected by the attacker, hijacking the communication. When using HBAs, only the addresses belonging to an HBA set can be used interchangeably, limiting the addresses that can be used to redirect the communication to a predetermined set that belongs to the original node involved in the communication. So, when using HBAs, a node that is communicating using address A can redirect the communication to a new address B if and only if B belongs to the same HBA set as A. This means that if an attacker wants to redirect communications addressed to address HBA1 to an alternative address IPX, the attacker will need to create a CGA Parameter Data Structure that generates an HBA set that contains both HBA1 and IPX. In order to generate the required HBA set, the attacker needs to find a CGA Parameter Data Structure that fulfills the following conditions: the prefix of HBA1 and the prefix of IPX are included in the Multi-Prefix Extension. HBA1 is included in the HBA set generated. Note: this assumes that it is acceptable for the attacker to redirect HBA1 to any address of the prefix of IPX. The remaining fields that can be changed at will by the attacker in order to meet the above conditions are: the Modifier, other prefixes in the Multi-Prefix Extension, and other extensions. In any case, in order to obtain the desired HBA set, the attacker will have to use a brute-force attack, which implies the generation of multiple HBA sets with different parameters (for instance with a different Modifier) until the desired conditions are meet. The expected number of times that the generation process will have to be repeated until the desired HBA set is found is exponentially related with the number of bits containing hash information included in the interface identifier of the HBA. Since 59 of the 64 bits of the interface identifier contain hash bits, then the expected number of generations that will have to be performed by the attacker are O(2^59). Note: We assume brute force is the best attack against HBA/CGAs. Also, note that the assumption that the Sec tool defined in multiplies the attack factor holds for brute-force attacks but may not hold for other attack classes. The protection against brute-force attacks can be improved by increasing the Sec parameter. A non-zero Sec parameter implies that steps 3-4 of the generation process will be repeated O(2^(16*Sec)) times (expected number of times). If we assimilate the cost of repeating the steps 3-4 to the cost of generating the HBA address, we can estimate the number of times that the generation is to be repeated in O(2^(59+16*Sec)), in the case of Sec values of 1 and 2. For other Sec values, Sec protection mechanisms will be defined by the specifications pointed by the CGA SEC registry defined in RFC 4982 .

In this section, we will analyze the security provided by HBAs in the context of a Shim6 protocol as described in of this memo. First of all, it must be noted that HBAs cannot prevent man&nbhy;in&nbhy;the&nbhy;middle (hereafter MITM) attacks. This means that in the scenario described in Section 8, if an attacker is located along the path between Host1 and Host2 during the lifetime of the communication, the attacker will be able to change the addresses used for the communication. This means that he will be able to change the addresses used in the communication, adding or removing prefixes at his will. However, the attacker must make sure that the CGA Parameter Data Structure and the HBA set is changed accordingly. This essentially means that the attacker will have to change the interface identifier part of the addresses involved, since a change in the prefix set will result in different interface identifiers of the addresses of the HBA set, unless the appropriate Modifier value is used (which would require O(2(59+16*Sec)) attempts). So, HBA doesn't provide MITM attacks protection, but a MITM attacker will have to change the address used in the communication in order to change the prefix set valid for the communication. HBAs provide protection against time shifting attacks , . In the multihoming context, an attacker would perform a time shifted attack in the following way: an attacker placed along the path of the communication will modify the packets to include an additional address as a valid address for the communication. Then the attacker would leave the on-path location, but the effects of the attack would remain (i.e., the address would still be considered as a valid address for that communication). Next we will present how HBAs can be used to prevent such attacks. If the attacker is not on-path when the initial CGA Parameter Data Structure is exchanged, his only possibility to launch a redirection attack is to fake the signature of the message for adding new addresses using CGA capabilities of the addresses. This implies discovering the public key used in the CGA Parameter Data Structure and then cracking the key pair, which doesn't seem feasible. So in order to launch a redirection attack, the attacker needs to be on&nbhy;path when the CGA Parameter Data Structure is exchanged, so he can modify it. Now, in order to launch the redirection attack, the attacker needs to add his own prefix in the prefix set of the CGA Parameter Data Structure. We have seen in the previous section that there are two possible approaches for this: Find the right Modifier value, so that the address initially used in the communication is contained in the new HBA set. The cost of this attack is O(2(59+16*Sec)) iterations of the generation process, so it is deemed unfeasible. Use any Modifier value, so that the address initially used in the communication is probably not included in the HBA set. In this case, the attacker must remain on-path, since he needs to rewrite the address carried in the packets (if not, the endpoints will notice a change in the address used in the communication). This essentially means that the attacker cannot launch a time shifted attack, but he must be a full-time man-in-the-middle. So, the conclusion is that HBAs provide protection against time shifted attacks HBAs do not provide complete protection against flooding attacks, and, as a result, the SHIM6 protocol has other means to deal with them. However, HBAs make it very difficult to launch a flooding attack towards a specific address. It is possible though, to launch a flooding attack against a prefix. And of course, the protection that HBA offers applies only to nodes that employ it; HBA provides no solution for general-purpose flooding-attack protection for other nodes. Suppose that an attacker has easy access to a prefix PX::/nX and that he wants to launch a flooding attack on a host located in the address P:iid. The attack would consist of establishing communication with a server S and requesting a heavy flow from it. Then simply redirecting the flow to P:iid, flooding the target. In order to perform this attack, the attacker needs to generate an HBA set including P and PX in the prefix set, and that the resulting HBA set contains P:iid. In order to do this, the attacker needs to find the appropriate Modifier value. The expected number of attempts required to find such Modifier value is O(2(59+16*Sec)), as presented earlier. So, we can conclude that such attack is not feasible. However, the target of a flooding attack is not limited to specific hosts, but it can also be launched against other elements of the infrastructure, such as router or access links. In order to do that, the attacker can establish a communication with a server S and request a download of a heavy flow. Then, the attacker redirects the communication to any address of the target network. Even if the target address is not assigned to any host, the flow will flood the access link of the target site, and the site access router will also suffer the overload. Such attack cannot be prevented using HBAs, since the attacker can easily generate an HBA set using his own prefix and the target network prefix. In order to prevent such attacks, additional mechanisms are required, such as reachability tests.

HBAs can be used as RFC 4941 addresses. If a node wants to use temporary addresses, it will need to periodically generate new HBA sets. The effort required for this operation depends on the Sec parameter value. If Sec=0, then the cost of generating a new HBA set is similar to the cost of generating a random number, i.e., one iteration of the HBA set generation procedure. However, if Sec>0, then the cost of generating an HBA set is significantly increased, since it required O(2(16*Sec)) iterations of the generation process. In this case, depending on the frequency of address change required, the support for RFC 4941 address may be more expensive.

Recent attacks on currently used hash functions have motivated a considerable amount of concern in the Internet community. The recommended approach to deal with this issue is first to analyze the impact of these attacks on the different Internet protocols that use hash functions, and second to make sure that the different Internet protocols that use hash functions are capable of migrating to an alternative (more secure) hash function without a major disruption in the Internet operation. The aforementioned analysis for CGAs and their extensions (including HBAs) is performed in RFC 4982 . The conclusion of the analysis is that the security of the protocols using CGAs and their extensions are not affected by the recently available attacks against hash functions. In spite of that, the CGA specification was updated by RFC 4982 to enable the support of alternative hash functions.

In order to use the HBA technique, the owner of the HBA set must inform its peer about the CGA Parameter Data Structure in order to allow the peer to verify that the different HBAs belong to the same HBA set. Such information must then be stored by the peer to verify alternative addresses in the future. This can be a vector for DoS attacks, since the peer must commit resources (in this particular case memory) to be able to use the HBA technique for address verification. It is then possible for an attacker to launch a DoS attack by conveying HBA information to a victim, imposing on the victim to use memory for storing HBA related state, and eventually running out of memory for other genuine operations. In order to prevent such an attack, protocols that use the HBA technique should implement proper DoS prevention techniques. For instance, the Shim6 protocol includes a 4-way handshake to establish the Shim6 context and, in particular, to establish the HBA-related state. In this 4-way handshake, the receiver remains stateless during the first 2 messages, while the initiator must keep state throughout the exchange of the 4 messages so that the cost of the context establishment is higher in memory terms for the initiator (i.e., the potential attacker) than for the receiver (i.e., the potential victim). In addition to that, the 4-way handshake prevents the usage of spoofed addresses from off-path attacker, since the initiator must be able to receive information through the address it has used as source address, enabling the tracking of the location from which the attack was launched.

This document was originally produced by a MULTI6 design team consisting of (in alphabetical order): Jari Arkko, Marcelo Bagnulo, Iljitsch van Beijnum, Geoff Huston, Erik Nordmark, Margaret Wasserman, and Jukka Ylitalo.

The initial discussion about HBA benefited from contributions from Alberto Garcia-Martinez, Tuomas Aura, and Arturo Azcorra. The HBA-set generation and HBA verification processes described in this document contain several steps extracted from . Jari Arkko, Matthew Ford, Francis Dupont, Mohan Parthasarathy, Pekka Savola, Brian Carpenter, Eric Rescorla, Robin Whittle, Matthijs Mekking, Hannes Tschofenig, Spencer Dawkins, Lars Eggert, Tim Polk, Peter Koch, Niclas Comstedt, David Ward, and Sam Hartman have reviewed this document and provided valuable comments. The text included in was provided by Eric Rescorla. We would also like to thank Francis Dupont for providing the first implementation of HBA.