]> 464XLAT Customer-side Translator (CLAT): Node Recommendations Google
1 Darling Island Rd Pyrmont NSW 2009 AU furry13@gmail.com furry@google.com
Microsoft
tojens@microsoft.com
Ops IPv6 operations ipv6 slaac 464xlat clat nat64 pref64 ipv6-only ipv6-mostly 464XLAT [RFC6877] defines an architecture for providing IPv4 connectivity across an IPv6-only network. The solution contains two key elements: provider-side translator (PLAT) and customer-side translator (CLAT). This document complements [RFC6877] and updates [RFC8585] by providing recommendations for the node developers on enabling and disabling CLAT functions.
Introduction 464XLAT is widely deployed in 3GPP networks (as described in Section 4.2 of ) where User Equipment (UE) devices (such as mobile phones and CE routers) perform the CLAT function, providing a private IPv4 address and default route for the applications and tethered devices. Enabling 464XLAT allowed mobile operators to transition UE devices to IPv6-only mode, where those devices are only assigned IPv6 addresses on their WAN interfaces. Until recently, IPv6-only hosts were rather uncommon outside of mobile networks and datacenters. Even if the network provides PLAT in the form of NAT64 (), hosts (desktops, laptops, etc.) still needed the network to provide IPv4 addresses, as otherwise IPv4-only applications fail. However, as more and more operating systems outside of the 3GPP world support CLAT, it becomes possible to migrate those devices to IPv6-only mode, while still providing IPv4 as a service via 464XLAT. Networks such as public Wi-Fi, enterprise networks, or even home networks can deploy 464XLAT as described in Section 4.2 of :
  • PLAT functions are performed by NAT64 network devices.
  • CLAT is performed by the UA itself.
Another 464XLAT deployment model is a Wireline one (section 4.1 of ), when a CPE router is connected to an IPv6-only network and provides CLAT functions for IPv4-enabled downstream devices. specifies 464XLAT support requirements for such devices. For both scenarios to work, the node performing CLAT (such as a host or a CPE router) need to enable the CLAT when connecting to an IPv6-only network. This document complements and updates by providing recommendations for the node developers on enabling and disabling CLAT functions.
Requirements Language The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.
Terminology This document reuses most of Terminology section from .
  • CLAT Node: a node (a host or a router) which performs CLAT functions by running one or multiple CLAT instances (e.g. one CLAT instance per interface).
  • IPv6-only network: A network that does not assign IPv4 addresses to hosts and facilitates connectivity to IPv4-only destinations using NAT64 (). In this document, the term "IPv6-only network" specifically refers to networks that provide NAT64 as it's required by the 464XLAT architecture ().
  • Native IPv4 (such as in 'native IPv4 connectivity' or 'native IPv4 default gateway'): IPv4 connectivity or default gateway provided by the network without using any form of IPv4-as-a-service or translation mechanisms (such as 464XLAT).
  • ULA: Unique Local Addresses, .
Currently the only avalable form of PLAT for 464XLAT deployments is NAT64. Therefore this document uses those terms interchangeably.
Multiple Interfaces Considerations A node may have multiple IPv6-only interfaces (for example, a mobile phone can be connected to an IPv6-only Wi-Fi network and to an IPv6-only mobile network). In that case the node SHOULD run an independent, dedicated CLAT instance on each interface connected to a network equipped with PLAT. Consequently, each CLAT instance SHOULD install a separate default IPv4 route on each CLAT-enabled interface. The metrics of IPv4 routes SHOULD be consistent with the metrics of IPv6 default routes.
Enabling CLAT For performance and security reasons CLAT MUST NOT be enabled if the node has IPv4 native connectivity over the given interface. Therefore recommendations provided in this section are only applicable to an IPv6-only node (a node which does not have a native IPv4 default route configured). To enable CLAT, an IPv6-only node needs to discover the PLAT-side translation IPv6 prefix, also known as the NAT64 prefix (see Section 6.3 of ). The PREF64 Router Advertisement (RA) option () provides that information and can be used as a strong indication that the network supports PLAT (NAT64) functionality. Therefore an IPv6-only node SHOULD enable CLAT as soon as a Router Advertisement containing PREF64 option is received. If RAs received by the node do not contain a PREF64 option, the node MAY use other mechanisms to detect the PLAT presence and obtain the NAT64 prefix (such as ). When discovering the NAT64 prefix using the mechanism defined in , the node MUST follow recommendations provided in . Specifically, the node MUST send the query to the DNS servers for the specific network interface per Section 7.1 of . In particular, queries for AAAA resource records of "ipv4only.arpa." MUST be sent to the recursive resolvers provided by the network, not to the resolvers configured manually, local recursive resolvers etc. Any delay in enabling CLAT on an IPv6-only node would be impactful for IPv4-only applications, as such applications cannot benefit from 464XLAT until CLAT is operational. Therefore it's desirable that the node enables CLAT as soon as network support for PLAT is detected while IPv4 connectivity is not yet detected. The node SHOULD enable CLAT after discovering the NAT64 prefix, unless by that time the node has already obtained a non-link-local IPv4 address. The node SHOULD NOT wait for an explicit (DHCP Option 108) or an implicit (DHCP timeouts) indication that native IPv4 connectivity is not available. However, to mitigate attacks described in Section 7 of , the node MAY delay enabling CLAT if the NAT64 prefix was discovered via DNS () only. The delay is implementation specific. If IPv4 connectivity becomes available later, the node MUST disable CLAT (unless explicitly configured to keep it running) as discussed in the following section. If the node supports multiple IPv4 continuity solutions, the node MUST follow recommendations from Section 4 of to avoid IPv4 address space conflicts.
Disabling CLAT It is possible that after the CLAT instance has started, native IPv4 becomes available (e.g. an IPv4 address received via DHCP). Unless explicitly configured otherwise, the node MUST disable CLAT immediately upon obtaining a native IPv4 default gateway. While disabling CLAT is impactful for all applications and traffic flows already utilizing CLAT, it is recommended not only from a performance perspective, but also from a security point of view. Because IPv4-only networks are inherently IPv6-ignorant, they might lack IPv6 layer2 security features, such as RA Guard, that would prevent spoofed RAs. An attacker can send an RA containing the PREF64 option, while the network doesn't provide any PLAT functionality. If the node keeps CLAT enabled and uses it for IPv4-only destinations, that traffic could be dropped (availability attack) or routed through an attacker-controlled route (active attack on insecure traffic or hoarding secure traffic for "Harvest Now, Decrypt Later" attacks for non-PQC secure traffic, see Section 8 of ). Rogue RAs can also disturb traffic to destinations that support both IPv4 and IPv6 by causing IPv6 through an attacker's PLAT to be used instead of the legitimate network owner's IPv4 path. There are some corner cases when the administrator might prefer the node to use CLAT even if the native IPv4 connectivity is available (e.g. for performance reasons, if IPv4 as a Service performs better than native IPv4). However for the reasons described above such behaviour MUST be explicitly enabled by the administrator via a configuration knob and MUST NOT be a default behaviour, especially for unmanaged nodes.
CLAT Addresses Considerations
CLAT IPv4 Addresses There are two different models in 464XLAT deployments:
  • A dedicated prefix model, which Section 4.1 of calls "wireline network architecture". In that case, the node performing CLAT functions also extends the network downstream and provides network connectivity services to other connected systems. Those systems can be physical (e.g. various clients connected to a CPE router), or logical (e.g. virtual systems running on a node, while the host system acts as a router and performs CLAT). In all those cases, systems behind the CLAT node usually use addresses. Despite the word "wireline", this architecture is applicable for wireless or 3GPP routers: such router can provide IPv4 connectivity to connected devices, performing CLAT functions for traffic sent/received via the IPv6-only uplink interface.
  • A single-address model, which section 4.2 of calls "wireless network architecture". In that case, the CLAT instance provides an IPv4 address and the default route to the local node's network stack only. It should be noted that implies that the second deployment scenario is limited to 3GPP cases, while currently it is also deployed in other types of networks, such as enterprise networks and Wi-Fi hotspots, where hosts (as mobile phones, laptops and desktops) use CLAT to provide connectivity to IPv4-only local applications. Despite the word "wireless" in the name, this architecture is applicable for wired networks as well (e.g. desktops or servers using wired connections).
In the single-address model, the CLAT instance needs a single IPv4 CLAT address and a single CLAT-only IPv6 address (which is distinct from the one or more IPv6 addresses used by the node running CLAT for its own native IPv6 connectivity, see ). The node providing CLAT functions to local applications SHOULD use IPv4 addresses from the dedicated 192.0.0.0/29 range (), reserved for IPv4 continuity solutions including but not limited to 464XLAT. If the node runs multiple CLAT instances (see ), the node SHOULD use different local IPv4 addresses for each CLAT instance. This approach limits the number of CLAT instances per node to 8, which seems to be more that sufficient at the time of writing. If in the future some deployment scenarios require more that 8 CLAT instances per node, a new larger IPv4 range will be requested from IANA. The node MUST NOT send packets on wire from the local CLAT address. The host SHOULD use 255.255.255.255 as a netmask for the CLAT address. That allows all 8 addresses from 192.0.0.0/29 to be used, if needed. It should be noted that 192.0.0.0/29 is shared between multiple IPv4 continuity solutions such as 464XLAT and DS-Lite (see . For example, Section 10 of reserves 192.0.0.1 for the Dual-Stack Lite default router. However, as per Section 4 of , the host MUST NOT enable two active IPv4 continuity solutions simultaneously in a way that would cause a node to have overlapping 192.0.0.0/29 address space. Therefore, as long as the host is not using DS-Lite, it MAY use 192.0.0.1 as a CLAT address.
CLAT IPv6 Addresses Section 6.3 of recommends that the CLAT instance acquires a dedicated /64 for translating between IPv4 and IPv6, and only uses a single interface IPv6 address if a dedicated prefix is not available via DHCPv6-PD. However, deployments where each node can obtain a dedicated /64 just for CLAT are rather uncommon, especially in environments like enterprise networks, Wi-Fi hotspots, etc. Quite often the CLAT instance uses a single IPv6 address as a source for all IPv4 traffic translated by CLAT. In particular, in a single address model (see ) the CLAT instance only need a single CLAT IPv6 address. Even in a dedicated prefix model, the CLAT instance can first perform stateful NAT44 to translate all IPv4 addresses from the dedicated prefix to a single IPv4 address, and then perform stateless CLAT. The CLAT instance SHOULD obtain a dedicated IPv6 address used exclusively for CLAT functions. This is required as when the node receives a packet from a NAT64 source, the node needs to differentiate between native IPv6 traffic and traffic which needs to be passed to the CLAT instance. For example, an ICMPv6 Echo Reply packet from 64:ff9b::203.0.113.1 can be a response to either an IPv6 ping to 64:ff9b::203.0.113.1, or an IPv4 ping to 203.0.113.1, translated by CLAT. Using a dedicated IPv6 source address for CLAT traffic allows the node to make that distinction without keeping state and operate in the stateless mode (see Section 1.3 of . In a dedicated prefix model, the instance MAY obtain a dedicated IPv6 address for each CLAT IPv4 address. The node MUST treat the CLAT addresses as any other address and comply with and . In particular:
  • Performing Duplicate Address Detection for each dedicated CLAT address (Section 5.4 of );
  • Processing received Neighbor Solicitations sent to the solicited-node multicast address () for the node CLAT addresses.
  • Sending Gratuitous Neighbor Advertisements () for the CLAT addresses.
  • Registering the CLAT addresses using DHCPv6 (), if the network supports the registration.
If the dedicated CLAT address is obtained via Stateless Address Autoconfiguration (SLAAC, ), the CLAT instance SHOULD ensure that the address is checksum-neutral for the given local IPv4 CLAT address and the NAT64 prefix (this means that the local IPv4 address needs to be assigned/known before the IPv6 address is configured). Using a checksum-neutral CLAT address provides the following benefits:
  • Better performance as CLAT doesn't need to recalculate the checksum.
  • If a protocol uses the standard IP checksum, CLAT doesn't need to recalculate the checksum. That improves the chances of the protocol working via CLAT even if CLAT is not aware of the protocol's semantics.
To protect user privacy and prevent user tracking through CLAT addresses, the node SHOULD generate a different interface id for the CLAT address when connecting to different networks, even if the NAT64 prefix and the local IPv4 CLAT address do not change. In particular, the node SHOULD generate a random CLAT address every time the network attachement changes to another network.
CLAT and Multiple Prefixes per Interface IPv6 multihoming, particularly when multiple routers on the same link advertise different prefixes PIOs, presents a complex and not yet fully resolved challenge. When routers on a given link are managed independently (e.g., by different ISPs), the resulting set of configuration parameters received by a host can be difficult to utilize without creating a complex and fragile state machine. For example, if router_A advertises a PIO with prefix_A and PREF64_A, while router_B advertises a PIO with prefix_B and a PREF64_B, it is crucial that the CLAT bundles the information received from each router. A CLAT instance must use PREF64_A and generate a CLAT address from prefix_A, sending translated packets to router_A. Alternatively, it must use PREF64_B, generate an address from prefix_B, and send translated packets to router_B. Mixing configuration information from different routers (e.g., generating a CLAT address from prefix_A but using PREF64_B for translation) can lead to packet loss. For example, if packets with source addresses from prefix_A are sent to router_B, that router (or the uplink network) might drop the packets according to BCP 38 (). Similarly, if the CLAT instance uses PREF64_A, advertised by router_A, but those packets are sent to router_B, that router might not be configured to translate packets for that prefix. This document does not aim to define CLAT behavior for every possible multi-router/multi-prefix scenario. Instead, this section provides recommendations for common scenarios, leaving numerous corner cases out of scope. A node discovering multiple routers on the same interface advertising the same PIOs and NAT64 prefix, SHOULD only create one CLAT instance using one of the PIOs to form a CLAT address. A node discovering multiple routers on the same interface signalling the different PIOs and NAT64 prefixes, MAY create one CLAT instance for each tuple of PIOs and NAT64 prefix, or only a single CLAT instance using the NAT64 prefix discovered through the selected IPv6 default router and the address formed from a PIO advertised by that router. When a node creates a single CLAT instance and must choose between multiple PIOs, the node SHOULD select a single PIO using the same algorithm as for choosing the source address for a destination within the selected NAT64 prefix (, updated by ). Discussion: This approach, leveraging the default source address selection algorithm (Section 5 of ), typically results in the policy table (rule 6) and longest prefix match (rule 8) being used for prefix selection. This ensures CLAT address selection aligns with default source address selection for native IPv6 flows, offering the following advantages:
  • When using the well-known NAT64 prefix (64:ff9b::/96), non-ULA prefixes are preferred over ULA prefixes by default. This is beneficial as ULA source packets may not reach PLAT devices.
  • For network-specific NAT64 prefixes within the known-local ULA range (), the ULA prefix is preferred. This can be advantageous in home and enterprise environments where administrators intend to perform NAT64 for specific source prefixes only.
  • For network-specific NAT64 prefixes within the operator's global non-ULA range, the longest prefix match selects the PIO, ensuring CLAT uses the operator's source address for traffic to the operator's PLAT in multi-prefix environments.
  • In managed environments, operators can customize CLAT behavior by modifying the policy table if the default prefix selection is unsuitable.
Creating a single CLAT instance significantly simplifies the CLAT state machine. However, this approach may concentrate all traffic from that instance onto the same first-hop router and NAT64 device in some multihomed topologies. As traffic shifts from CLAT to native IPv6, this drawback becomes less significant and does not justify the added complexity of multiple instances.
Link Renumbering As discussed above, a single CLAT instance per interface, using a single PIO, is typically sufficient, even if the link has multiple assigned subnets. However, PIO selection can significantly impact user experience during link renumbering. discusses various examples of "flash renumbering," where the IPv6 prefix assigned to the link changes without explicit host notification. and discuss methods to mitigate the impact of flash renumbering. These methods generally rely on hosts with addresses from both old and new prefixes ceasing use of the old prefix and adopting the new prefix. For nodes running CLAT instances, this requires disabling instances using addresses from the old prefix and creating an instance using an address from the new prefix. The CLAT node SHOULD use at least the following signals to detect link renumbering events:
  • A prefix used to form the CLAT address becomes deperecated or invalid ().
  • The router (or routers) advertizing the PIO used to form the CLAT address
    • has changed it state from being reachable or probably reachable to being unknown or suspect (i.e., it's neighbor cache entry moved to the INCOMPLETE state or ceased to exist, see Section 6.3.6 of ).
    • ceased to be a router (see Section 7.3.3 of ).
Upon receiving a signal indicating a possible renumbering event, the node SHOULD disable the CLAT instance(s) affected by the renumbering, and create new instance(s). In case of implicit signals (provided by the Neighbor Unreachability Detection, , rather than by a Router Advertisement deprecating or invalidating a prefix), the node MAY send Router Solicitations to obtain the most up-to-date network configuration information. When sending Router Solicitations the node MUST follow recommendations specified in Section 6.3.7 of . The node MAY react to a potential renumebring event in a "make-before-break" manner, when old instances are still running until all required information to enable new ones becomes available.
MTU Considerations The IPv4 header is 20 bytes long (or longer if IP options are present), while the IPv6 header is 40 bytes. This means that when CLAT translates an IPv4 packet to IPv6, it usually adds 20 bytes to the packet size. However, when CLAT transates a fragmented IPv4 packet, then Fragment Header needs to be added to the resulting IPv6 packet (Section 4.1 of ). The length of IPv6 Fragment Extension header is 8 bytes (Section 4.5 of ). Therefore, to minimize undesirable IP fragmentation (), the CLAT instance in a single-address mode SHOULD present IPv4-only applications with an IPv4 MTU which is 28 bytes smaller than the IPv6 MTU of the interface the instance is running on.
Updates to RFC8585 This document makes the following changes to Section 3.2.1 of : OLD TEXT: === 464XLAT requirements: === NEW TEXT: === 464XLAT requirements: 464XLAT-0: The IPv6 Transition CE Router SHOULD follow recommendations provided in draft-link-v6ops-claton. === OLD TEXT: === 464XLAT-4: The IPv6 Transition CE Router MUST implement ("Discovery of the IPv6 Prefix Used for IPv6 Address Synthesis") in order to discover the provider-side translator (PLAT) translation IPv4 and IPv6 prefix(es)/suffix(es). === NEW TEXT: === 464XLAT-4: The IPv6 Transition CE Router MUST implement ("Discovering PREF64 in Router Advertisements") and SHOULD implement ("Discovery of the IPv6 Prefix Used for IPv6 Address Synthesis") in order to discover the provider-side translator (PLAT) translation IPv4 and IPv6 prefix(es)/suffix(es). === OLD TEXT: === 464XLAT-6: If the network provides several choices for the discovery/learning of the NAT64 prefix, the priority to use one or the other MUST follow this order: 1) and 2) . The NAT64 prefix could be discovered by means of the method defined in only if the service provider uses DNS64 . It may be the case that the service provider does not use or does not trust DNS64 because the DNS configuration at the CE (or hosts behind the CE) can be modified by the customer. In that case, the service provider may opt to configure the NAT64 prefix by means of the option defined in . This can also be used if the service provider uses DNS64 . === NEW TEXT === 464XLAT-6: If the network provides several choices for the discovery/learning of the NAT64 prefix, the priority to use one or the other MUST follow this order: 1) 2) and 3) . 464XLAT-7: If the IPv6 Transition CE Router performs CLAT functions it SHOULD also include the PLAT prefix in Router Advertisements () sent via the LAN interfaces. If the IPv6 Transition CE Router acts as a DHCP server it SHOULD enable DHCP Option 108 () processing. The router SHOULD have a configuration knob to disable DHCP Option 108 processing. allows the service provider to signal NAT64 prefix independently from DNS64 presence. At the same time the NAT64 prefix could be discovered by means of the method defined in only if the service provider uses DNS64 . It may be the case that the service provider does not use or does not trust DNS64 because the DNS configuration at the CE (or hosts behind the CE) can be modified by the customer. In that case, the service provider may opt to configure the NAT64 prefix by means of the option defined in . This can also be used if the service provider uses DNS64 . ===
Security Considerations If a malicious actor spoofs PLAT presence signals (such as an RA with PREF64 option) or DNS responses for DNS-based NAT64 prefix detection (), traffic of IPv4-only applications using CLAT can be affected:
  • if there is no PLAT (NAT64) devices, traffic to NAT64 destinations would be dropped.
  • If the attacker intercepts traffic for the NAT64 prefix (e.g. by providing the victim with a bogus NAT64 prefix and steering traffic for those destinations towards themselves), the attacker might be able to perform man-in-the-middle attacks.
Using the PREF64 RA option to detect PLAT presence and the NAT64 prefix is less prone to such attacks than DNS-based detection (), as the attacker needs to be on-link and be able to bypass layer-2 security features such as RA Guard. Therefore recommends the PREF64 RA option as a preferred way to detect PLAT presence. In networks lacking explicit IPv6 deployment and consequently IPv6 security features, administrators may inadvertently expose link-local IPv6 connectivity when endpoints have IPv6 enabled. This unintended exposure can facilitate PLAT presence signal falsification, enabling malicious actors to conduct Man-in-the-Middle (MitM) or Denial-of-Service (DoS) attacks. This document mitigates this risk by requiring endpoints to disable CLAT when the network provides non-link-local IPv4 connectivity, as outlined in .
Privacy Considerations If the instance utilizes the same CLAT address for an extensive period of time or, much worse, uses the same CLAT address when connecting to different networks, eavesdroppers and information collectors can correlate various network activity to the same node. To mitigate that risk and make address-based network-activity correlation more difficult, the node SHOULD generate a different interface id for the CLAT address when connecting to different networks (see ). It should be noted that the node's CLAT IPv6 address is only used (and visible to observers) when the traffic is carried from the CLAT node to the PLAT device. In the vast majority of the cases it means that address is never visible outside of the network Internet edge, so to perform address-based network-activity correlation the observer needs to be located in the same network as the CLAT node. To summarize, this document does not introduce any new privacy considerations.
IANA Considerations This memo does not introduce any requests to IANA.
References Normative References Informative References
Enabling and Disabling CLAT: Flowchart
Enabling and Disabling CLAT Instance YES Has native IPv4 default route? CLAT Disabled NO NO NO PLAT prefix Start DHCP Received PREF64 in RA? discovered via RFC7050? YES YES Wait for DHCP Offer Enable CLAT YES Option 108 Present? NO | CLAT Disabled | | | | | +------------------+---------------+ +------------------------------+ | NO ^ ^ +---------+-------------+ | NO | | | | | +--------v--------+ +-----------v------------+ +--+----------+ | | | | | NO | PLAT prefix | | | Start DHCP | | Received PREF64 in RA? +------>| discovered | | | | | | | via RFC7050?| | +---------+-------+ +-------------------+----+ +-------+-----+ | | ^ | | | | | | YES | | v | v | YES | +---------------------+ | +---------------+ | | | | | | | | | | Wait for DHCP Offer |<----+------+ Enable CLAT |<---------+ | | | | | | | +----------+----------+ | +---------------+ | | | | | | | v | YES | +---------------------+ | | | | | | | +-----+ | | Option 108 Present? | | | | NO | | +--------------------------------------------------+ +---------------------+ ]]>
Acknowledgements Thanks to Ondrej Caletka, Stuart Cheshire, Lorenzo Colitti, Jeremy Duncan, Jason Healy, Ed Horley, KAWASHIMA Masanobu, Ted Lemon, George Michaelson, Jordi Palet, Dieter Siegmund, Philipp S. Tiesel, Eric Vyncke for the discussions, the input, and all contribution.