| Internet-Draft | Automatic Stub Networks | October 2022 |
| Lemon | Expires 10 April 2023 | [Page] |
This document describes a set of practices for connecting stub networks to adjacent infrastructure networks, as well as to larger network fabrics. This is applicable in cases such as constrained (Internet of Things) networks where there is a need to provide functional parity of service discovery and reachability between devices on the stub network and devices on an adjacent infrastructure link (for example, a home network).¶
The stub networks use case is intended to fully address the need to attach a single network link to an infrastructure network, where the attached link provides no through routing and in cases where integration to the infrastructure routing fabric (if any) is not available.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 10 April 2023.¶
Copyright (c) 2022 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
This document describes a set of practices for connecting stub networks to adjacent infrastructure networks, as well as to larger network fabrics. This is applicable in cases such as constrained (Internet of Things) networks where there is a need to provide functional parity of service discovery and reachability between devices on the stub network and devices on an adjacent infrastructure link (for example, a home network).¶
The stub networks use case is intended to fully address the need to attach a single network link to an infrastructure network, where the attached link provides no through routing and in cases where integration to the infrastructure routing fabric (if any) is not available.¶
We assume that adjacent infrastructure link supports Router and Prefix Discovery using router advertisements. Adjacent infrastructure links on networks where this is not supported are out of scope for this document.¶
In order to provide IPv6 routing to the stub network, IPv6 addressing must be available on the adjacent infrastructure link. In the ideal case, such addressing is already present on the link, and need not be provided. In this case, the stub router SHOULD NOT provide addressability on the adjacent infrastructure link.¶
IPv6 addressing is considered to be present on the link if a usable on-link prefix is advertised on the adjacent infrastructure link. A usable on-link prefix could be a prefix advertised on the link that is on-link and allows autonomous configuration. A prefix is also a usable on-link prefix if it is advertised on the link as on-link, and if the 'm' bit is set in the Router Advertisement message header ([RFC4861], Section 4.2) that contains the Prefix option. This indicates that node addressibility is being managed using DHCPv6.¶
A prefix is advertised on the link if, when a Router Solicit message ([RFC4861], Section 4.1) is sent, a Router Advertisement message is received in response which contains a prefix information option ([RFC4861], Section 4.6.2) for that prefix.¶
After such an RA message has been received, it can be assumed for some period of time thereafter that the prefix is still valid on the link. However, prefix lifetimes and router lifetimes are often quite long. The mere fact that a prefix that has been advertised is still within its valid lifetime does not mean that that prefix is still being advertised on the link.¶
This is important because when a new host appears on the adjacent infrastructure link and sends an initial router solicit, if it does not receive a usable on-link prefix, it will not be able to communicate. Consequently, the stub router MUST monitor router solicits and advertisements on the link in order to determine whether a prefix that has been advertised on the link is still being advertised.¶
There are several methods that can be used to accomplish this:¶
The stub router MUST listen for router advertisements on the adjacent infrastructure link, and record the time at which each router advertisement was received. A router advertisement that is more than STALE_RA_TIME seconds old MUST be assumed to no longer be advertised on the link. When the last non-stale router advertisement containing a usable prefixes on the link is marked stale, the stub router should begin Router Discovery ([RFC4861], Section 6.3).¶
The stub router MUST listen for router solicits on the adjacent infrastructure link. When a router solicit is received, the router SHOULD set a timer for VICARIOUS_SOLICIT_TIME seconds. If, after that amount of time, no router advertisements are received that contain a usable on-link prefix, the stub router MUST begin router discovery. This is necessary in case the response to the router solicit was unicast, since in this case the stub router would not see that response. When the stub router first connects to the adjacent infrastructure link, it MUST begin router discovery.¶
When router discovery completes, the stub router evaluates whether or not a usable on-link prefix has been seen in a non-stale router advertisement during router discovery. If no usable on-link prefix has been seen, then the stub router MUST begin to provide a usable on-link prefix.¶
As an alternative to the vicarious router discovery process described here, the stub router could monitor the presence of the router advertising the on-link prefix in the neighbor cache. If the neighbor cache entry becomes stale, this could be an indication that the prefix is also stale. If the neighbor cache entry goes stale, the router would need to try to refresh it, and if that fails, then the stub router must begin advertising its own on-link prefix on the stub network.¶
When there is no usable on-link prefix on the adjacent infrastructure network, the stub router provides its own on-link prefix. This prefix has a valid and preferred lifetime of STUB_PROVIDED_PREFIX_LIFETIME seconds. This prefix MUST allow for autonomous configuration (SLAAC).¶
The stub router must advertise this prefix every BEACON_INTERVAL seconds. When the stub router is advertising reachability to the stub network, the on-link prefix advertisement and the route information advertisement must be contained in the same router advertisement.¶
When the stub router is advertising an on-link prefix on the AIL, it may receive a router advertisement containing a usable on-link prefix for the AIL with a non-zero preferred lifetime. In this case, the stub router should begin to deprecate the on-link prefix it is advertising on the AIL. The preferred lifetime for this prefix should be set to zero in subsequent advertisements.¶
The valid lifetime (VALID) is computed based on three values: the current time when a router advertisement is being generated (NOW), the time at which the new usable on-link prefix advertisement was received (DEPRECATE_TIME), and STUB_PROVIDED_PREFIX_LIFETIME. All of these values are in seconds. VALID is computed as follows:¶
VALID = STUB_PROVIDED_PREFIX_LIFETIME - (NOW - DEPRECATE_TIME)¶
If VALID is less than BEACON_INTERVAL, the stub router does not include the deprecated prefix in the router advertisement. Note that VALID could be less than zero. Otherwise, the prefix is provided in the advertisement, but with a valid lifetime of VALID.¶
It is also possible that all routers on the link that are capable of advertising prefixes might be following the same protocol of deprecating their own prefix when a valid prefix shows up. To prevent a situation where all routers deprecate their prefix and wait until there are no valid prefixes being advertised before advertising a prefix, each stub router must detect the situation where, having deprecated its own prefix, all of the other prefixes being advertised on the link have also been deprecated.¶
When this situation occurs, each router on the link MUST compare the valid lifetimes of all the prefixes that have been seen. If the router's own prefix expires last, then that router should immediately resume publishing its prefix as a preferred prefix.¶
If a router observes this situation and its prefix is not the one that expires last, it MUST set a timer for UNDEPRECATE_WAIT seconds, while continuing to observe prefix advertisements on the link. If, when the timer expires, the prefix that expires last has not been re-published as a preferred prefix, then that prefix is marked as 'really deprecated', and no longer considered a candidate for de-deprecation.¶
Using the remaining list of prefixes, the router should then apply the same algorithm. It should continue to apply this algorithm until either its prefix becomes the one to re-publish as preferred, or some other router has re-published its prefix as preferred.¶
When multiple stub routers are connected to the same AIL, and no usable on-link prefix is being provided on that link by the infrastructure, there will be a competition between routers to provide a usable on-link prefix. In order to avoid duplication, stub routers MUST include a random offset in the time interval across which router discovery is performed. This ensures that after a power failure, not all stub routers will exit router discovery at the exact same time, and so one stub router should advertise a usable on-link prefix before the others. This should prevent the other stub routers from advertising additional on-link prefixes.¶
There is no particular harm caused by advertising multiple on-link prefixes, but it is preferable to minimize this, because each on-link prefix consumes space in every on-link host's routing table, and consumes time when making source address selection and routing decisions.¶
How addressability is managed on stub networks depends on the nature of the stub network. For some stub networks, the stub router can be sure that it is the only router. For example, a stub router that is providing a Wi-Fi network for tethering will advertise its own SSID and use its own joining credentials; in this case, it can assume that it is the only router for that network, and advertise a default route and on-link prefix just like any other router.¶
However, some stub networks are more cooperative in nature, for example IP mesh networks. On such networks, multiple stub routers may be present and be providing addressability and reachability.¶
In either case, some stub router connected to the stub network MUST provide a usable on-link prefix (the OSNR prefix) for the stub network. If the stub network is a multicast-capable medium where Router Advertisements are used for router discovery, the same mechanism described in section [Support for adjacent infrastructure links] is used.¶
Stub networks that do not support the use of Router Advertisements for router discovery must use some similar mechanism that is compatible with that type of network. Describing the process of establishing a common OSNR prefix on such networks is out of scope for this document.¶
Stub routers may restart from time to time; when a restart occurs, the stub router may have been advertising state to the network which, following the restart, is no longer required.¶
For example, suppose there are two stub routers connected to the same infrastructure link. When the first stub router is restarted, the second takes over providing an on-link prefix. Now the first router rejoins the link. It sees that the second stub router's prefix is advertised on the infrastructure link, and therefore does not advertise its own.¶
This behavior can cause problems because the first stub router no longer sees the on-link prefix it had been advertising on infrastructure as on-link. Consequently, if it receives a packet to forward to such an address, it will forward that packet directly to a default router, if one is present; otherwise, it will have no route to the destination, and will drop the packet.¶
To address this problem, stub routers SHOULD remember the last time a prefix was advertised across restarts. On restart, the router can immediately begin deprecating the prefix, and can stop after the prefix valid lifetime goes to zero, based on the recorded time that the last advertisement was sent.¶
When a stub router has only flash memory with limited write lifetime, it may be inappropriate to do a write to flash every time a prefix beacon happens. In this case, the router SHOULD record the set of prefixes that have been advertised on infrastructure and the maximum valid lifetime that was advertised. On restart, the router should assume that hosts on the infrastructure link have received advertisements for any such prefixes, and should immediately deprecate them, and continue to do so until the maximum valid lifetime has elapsed after restart.¶
In order to be able to provide addressability either on the stub network or on an adjacent infrastructure network, a stub router must allocate its own ULA prefix. ULA prefixes, described in Unique Local IPv6 Unicast Addresses ([RFC4193]) are randomly allocated prefixes. A stub router MUST allocate a single ULA prefix for use in providing on-link prefixes to the stub network and the infrastructure network, as needed.¶
The ULA prefix allocated by a stub router SHOULD be maintained across reboots, and SHOULD remain stable over time. For privacy reasons, a stub router that roams from network to network may wish to allocate a different ULA prefix each time it connects to a different infrastructure network.¶
If IPv6 prefix delegation is available, which implies that IPv6 service is also available on the infrastructure link, then the stub router MAY use IPv6 prefix delegation to acquire a prefix to advertise on the stub network, rather than allocating one out of its ULA prefix.¶
Stub routers MUST advertise reachability to stub network OSNR prefixes on any AIL to which they are connected.¶
Each stub network will have some set of prefixes that are advertised as on-link for that network. A stub router connected to that network SHOULD advertise reachability to all such prefixes on any AIL to which it is attached using router advertisements¶
The stub router MAY advertise itself as a default router on the stub network, if it itself has a default route on the AIL. In some cases it may not be desirable to advertise reachability to the Internet as a whole; in this case the stub router need not advertise itself as a default router.¶
If the stub router is not advertising itself as a default on the stub network, it MUST advertise reachability to any prefixes that are being advertised as on-link on AILs to which it is attached. This is true for prefixes it is advertising, and for other prefixes being advertised on that link.¶
Note that in some stub network configurations, it is possible for more than one stub router to be connected to the stub network, and each stub router may be connected to a different AIL. In this case, a stub router advertising a default route may receive a packet destined for a link that is not an AIL for that router, but is an AIL for a different router. In such a case, if the infrastructure is not capable of routing between these two AILs, a packet which could have been delivered by another stub router will be lost by the stub router that received it.¶
Consequently, stub routers SHOULD be configurable to not advertise themselves as default routers on the stub network. Stub routers SHOULD be configurable to explicitly advertise AIL prefixes on the stub network even if they are advertising as a default router. Stub routers SHOULD be configurable to advertise NAIL prefixes on the stub network; such configuration would include a list of NAIL prefixes to advertise. This list may be configured in a management interface or as a result of these routes being delivered in a routing protocol or through router discovery. The mechanisms by which such configuration can be accomplished are out of scope for this document.¶
In some cases it will be necessary for hosts on the adjacent infrastructure link to be able to discover devices on the stub network. In other cases, this will be unnecessary or even undesirable. For example, it may be undesirable for devices on an adjacent infrastructure link to be able to discover devices on a Wi-Fi tether, for example provided by a mobile phone.¶
One example of a use case for stub networks where such discovery is desirable is the constrained network use case. In this case a low-power, low-cost stub network provides connectivity for devices that provide services to the infrastructure. For such networks, it is necessary that devices on the infrastructure be able to discover devices on the stub network.¶
The most basic use case for this is to provide feature parity with existing solutions like multicast DNS (mDNS). For example, a light bulb with built-in Wi-Fi connectivity might be discoverable on the infrastructure link to which it is connected, using mDNS, but likely is not discoverable on other links. To provide equivalent functionality for an equivalent device on a constrained network that is a stub network, the stub network device must be discoverable on the infrastructure link (which is an AIL from the perspective of the stub network).¶
If services are to be advertised using DNS Service Discovery [RFC6763], there are in principle two ways to accomplish this. One is to present services on the stub network as a DNS zone which can then be configured as a browsing domain in the DNS ([RFC6763], Section 11). The second is to advertise stub network services on the AIL using multicast DNS (mDNS) [RFC6762].¶
Stub network routers cannot be assumed to be able to integrate into the DNS naming hierarchy of the infrastructure network. Therefore, stub networks must be able to rely on ad-hoc service advertisement protocols. Since mDNS is in wide use, this is a suitable protocol for this use case. This is not to say that mDNS is the only such protocol that could be used, but it is the one that we suggest implementing.¶
In order to provide mDNS discovery for devices on the stub network, one of two solutions is likely to be applicable, depending on the operational practicalities of the stub network. For a constrained stub network, on which battery operated devices may be attached, mass multicast traffic for service discovery is impractical, since every device needs to wake up for every service discovery, even if they don't offer that service, and since many such devices may be operating on battery power. For such a network, multicast DNS is not a good choice.¶
For such networks, a unicast service registration protocol such as DNS-SD Service Registration Protocol (SRP) [I-D.ietf-dnssd-srp] is a good solution. The stub router can act as an SRP server on the stub network, accepting service advertisements from stub network devices. On the adjacent infrastructure network, it can advertise those services as multicast DNS Advertising Proxy [I-D.sctl-advertising-proxy].¶
For other stub networks, for example a Wi-Fi-based Personal Area Network provided as part of a tethering function on a mobile device, multicast DNS may be the only option. For Wi-Fi stub networks, there is such a large installed base of devices supporting mDNS that requiring some other service advertisement solution would be problematic simply because it would require new software for that entire installed base. For other networks, particularly constrained networks, where devices do not currently support mDNS, no such obstacle exists.¶
Because the primary use case for discovery of devices on a stub network is the use case where the stub network is joining a constrained network to an existing infrastructure link, we currently only describe a solution (DNS-SD SRP) for that use case. A solution for the use case where the stub router must provide discoverability for a stub network where mDNS advertising is preferred is out of scope for this document.¶
Hosts on the stub network may need to discover hosts on the adjacent infrastructure network. In the IoT network example we've been using, there might be a light switch on the stub network which needs to be able to actuate a light bulb connected to the adjacent infrastructure network. In order to know where to send the actuation messages, the light switch will need to be able to discover the light bulb's address somehow.¶
In the case of a Wi-Fi stub network, devices on the stub network will need to be able to access the Internet, and may also need to be able to access local services on the adjacent infrastructure link.¶
In order to address these use cases, the stub network router SHOULD provide a DNS-SD Discovery Proxy [RFC8766] and a DNS resolver. Since these two functions are combined, if the stub router provides them, it MUST offer both services on the standard DNS UDP and TCP ports.¶
Stub networks are defined to be IPv6-only because it would be difficult to implement a stub network using IPv4 technology. However, stub network devices may need to be able to communicate with IPv4-only services either on the adjacent infrastructure, or on the global internet. Ideally, the infrastructure network fully supports IPv6, and all services on the infrastructure network are IPv6-capable. In this case, perhaps the infrastructure network provides NAT64 service to IPv4-only hosts on the internet. In this ideal setting, the stub router need do nothing-the infrastructure network is doing it all.¶
In this situation, if there are multiple stub routers, each connected to the same adjacent infrastructure link, there is no need for special behavior-each stub router can advertise a default route, and any stub router will do to route NAT64 traffic. If some stub routers are connected to different adjacent infrastructure links than others, some of which support NAT64 and some of which do not, then the default route may not carry traffic to the correct link for NAT64 service. In this case, a more specific address to the infrastructure NAT64 prefix(es) MUST be advertised by those stub routers that are able to discover it.¶
Most infrastructure networks at present do not provide NAT64 service. It is therefore necessary for stub routers to be able to provide NAT64 service if IPv4 hosts are to be reachable from the stub network.¶
To provide NAT64 service, a stub router must allocate a NAT64 prefix. For convenience, the stub network allocates a single prefix out of the /48 ULA prefix that it maintains. Out of the 2^16 possible subnets of the /48, the stub router SHOULD use the numerically highest /64 prefix.¶
If there are multiple stub routers providing connectivity between the stub network and infrastructure, each stub network uses its own NAT64 prefix-there is no common NAT64 prefix. The reason for this is that NAT64 translation is not stateless, and is tied to the stub router's IPv4 address. Therefore each NAT64 egress is not equivalent.¶
A stub network that services a Wi-Fi stub network SHOULD provide DNS64 translation: hosts on the stub network cannot be assumed to be able to do DNS64 synthesis in the stub resolver. In this case the DNS resolver on the stub router MUST honor the CD and DO bits if received in a request, since this indicates that the stub resolver on the requestor intends to do DNSSEC validation. In this case, the resolver on the stub router MUST NOT perform DNS64 synthesis.¶
On specific stub networks it may be desirable to require the stub network device to perform DNS64 synthesis. Stub network routers for such networks do not need to provide DNS64 synthesis. Instead, they MUST provide an ipv4only.arpa answer that advertises the NAT64 prefix for that stub router, and MUST provide an explicit route to that NAT64 prefix on the stub network using RA or whatever technology is specific to that stub network type.¶
In constrained networks it can be very useful if stub network resolvers provide the information required to do DNS64 translation in the answer to the AAAA query. If the answer to an AAAA query comes back with "no data" (not NXDOMAIN), this suggests that there may be an A record. In this case, the stub network's resolver SHOULD attempt to look up an A record on the same name. If such a record exists, the resolver SHOULD return no data in the Answer section of the DNS response, and SHOULD provide any CNAME records that were involved in returning the "no data" answer to the AAAA query, and SHOULD provide any A records that were ultimately returned, in the Additional section. The resolver should also include an ipv4only.arpa record in the Additional section.¶
If a stub network is constructed using mesh technology, it may become partitioned. In such a situation, it may be one stub router is connected to one partition, and another stub router is connected to the other partition. In this situation, in order for all nodes to be reachable, it is necessary that each partition of the stub network have its own prefix. When such a partition occurs, the stub routers must detect that it has occurred. If a stub router is currently providing a prefix on the stub network, it need take no action. If a stub router had not been providing a prefix on the stub network, and now discovers that there is no stub router providing a prefix on the network, it MUST begin to provide its own prefix on the stub network. It MUST also advertise reachability to that new prefix on its adjacent infrastructure link(s).¶
When partitions of this type occur, they may also heal. When a partition heals in a situation where two stub routers have both been advertising a prefix, it will now appear that there are two prefixes on the stub network. Since partition events may represent a recurring situation, stub routers SHOULD wait for at least PARTITION_HEAL_WAIT_TIME before deprecating one of these prefixes.¶
When the time comes to deprecate one or more prefixes as a result of a network partition healing, only one prefix should remain. If there are any GUA prefixes, and if there is no specific configuration contradicting this, the GUA prefix that is numerically lowest should be kept, and all others deprecated. If there are no GUA prefixes, then the ULA prefix that is numerically lowest should be kept, and the others deprecated. By using this approach, it is not necessary for the routers to coordinate in advance.¶
There are two ways that connectivity to non-adjacent links can be established. The first is that if the infrastructure network as a whole has a working IPv4 routing fabric, NAT64 can be used to enable hosts on the stub network to establish communications with hosts on non-adjacent links, including the Internet. In some cases, this is all that is needed.¶
However, if it will be necessary for nodes on non-adjacent networks to establish communications with nodes on the stub network, this will require a working IPv6 routing fabric connecting the stub network to any non-adjacent links from which communications will need to be established.¶
In order for such routing to work, the stub network will also need to acquire a prefix that the infrastructure network is aware of and can route to. The ULA prefix that can work for communicating to adjacent infrastructure links will not work for communicating to non-adjacent links.¶
A prefix may be acquired by using DHCPv6 Prefix Delegation ([RFC8415], Section 6.3). The stub router then advertises this prefix as the on-link prefix for the stub network, as before. It also advertises reachability to this prefix using router advertisements, as before.¶
In the case where there is more than one stub router, it would be best if only one stub router requested a delegated prefix. This can be managed through the mechanism described earlier: the stub router only acquires a prefix to advertise when it has decided that it needs to advertise a prefix, and so in most cases only one stub router at a time will request a delegated prefix.¶
In order to avoid excessive consumption of delegated prefixes, stub routers connected to stub networks that support multiple stub routers SHOULD request short lifetimes for delegated prefixes and renew frequently. Stub routers SHOULD request a lifetime of PREFIX_DELEGATION_INTERVAL. Stub routers SHOULD record the time that a prefix was acquired in stable storage, and SHOULD release the prefix using a "DHCP Release" transaction when shutting down, or when it determines that a prefix is no longer needed (See "graceful shutdown" in Figure 9 of [RFC8415] for details). Stub routers SHOULD release any remembered still-valid prefix after reboot, if after rebooting it is discovered that another prefix is being advertised on the stub network.¶
We can assume that a side effect of the prefix delegation process will be to establish routing to the stub router that requested the prefix. This should mean that any node that wishes to establish communication with a node on the stub network will be able to do so through the delegating router that provides the prefix or, if it is attached to an infrastructure link that is adjacent to the stub router, through the stub router itself by means of the router advertisement it is providing.¶
The case of multiple stub routers is more complicated however. Any routing that comes as a side-effect of DHCPv6 Prefix Delegation will only route through the stub router that acquired the prefix. Other stub routers can provide reachability on their respective adjacent infrastructure links, but reachability across the full routing fabric of the infrastructure network will only be possible if there is some routing protocol present on the infrastructure network. Addressing this problem is out of scope for this document.¶
In order for service advertisements to be available on non-adjacent infrastructure, the infrastructure must provide SRP service for constrained stub networks, and must advertise the availability of such service so that stub routers can forward SRP updates to that SRP service, rather than providing SRP as a local service. This SRP service can be discovered using DNS-SD, using the _dnssd-srp-tls service type. If the stub network requires UDP-based SRP rather than tls-based SRP, the stub router MUST act as a proxy to deliver SRP updates over the tcp+tls transport.¶
For stub networks that use multicast DNS, stub routers must provide a discovery proxy service, and most advertise that service to the infrastructure. In turn, the infrastructure must configure that service to be discoverable by devices on the infrastructure, as described in [RFC8766], Section 6.¶
The mechanism described previously for making service advertisements available to non-adjacent infrastructure also scales to the internet, since it uses DNS. Indeed, the question an operator should ask before enabling such discovery is, do they want their stub network devices to be discoverable on the internet. If it becomes possible to configure service advertising automatically, behavior similar to that specified in [RFC6092], Section 3.2 and 3.3, would be advised: do not automatically advertise stub network devices on the Internet.¶
Stub routers may be mobile, or fixed. That is, they may move from location to location along with some or all of their connected devices, attaching to whatever infrastructure is available. Or they may be fixed devices that are only ever expected to exist in one particular location.¶
For devices that are intended to be in a fixed location, the distinction between infrastructure links and the internet as a whole is meaningful; for mobile nodes it most likely is not, unless such a node is only going to ever attach to trusted infrastructure as it moves from location to location-not a common scenario.¶
For fixed links, the infrastructure may be trusted, in which case the distinction between infrastructure and internet can be expected to be managed by the infrastructure, and therefore only visible to the stub router in the sense that some non-adjacent destinations may be reachable (infrastructure destinations, for example) while others are not.¶
The reason for mentioning this here is to point out that the stub router can't be expected to manage this interface: it is up to the infrastructure network to do so, either implicitly or explicitly. [RFC7084] provides a set of default behaviors for home routers that may be adequate for automatically managing this interface, but further work in this area may be warranted.¶
These appendices describe the problem of linking stub networks to existing networks automatically, in the same way that hosts, when connected to an existing network, are able to discover network addressing parameters, information about routing, and services that are advertised on the network.¶
There are several use cases for stub networks. Motivating factors include:¶
What makes stub networks a distinct type of network is simply that a stub network never provides transit between networks to which it is connected. The term "stub" refers to the way the network is seen by the link to which it is connected: there is reachability through a stub network router to the stub network from that link, but there is no reachability to any link beyond that one.¶
Stub networks may be globally reachable, or may be only locally reachable. A host on a globally reachable stub network can interoperate with other hosts anywhere on the Internet. A host on a locally reachable stub network can only interoperate with hosts on the network link(s) to which it is connected.¶
The goal of this document is to describe the minimal set of changes or behaviors required to use existing IETF specifications to support the stub network use case. The result should be a small set of protocol enhancements (ideally no changes at all to protocols) and should be deployable on existing networks without requiring changes to those networks. Both the locally-reachable and globally-reachable use case should be able to be made to work, and ideally the globally-reachable use case should build on what is used to make the locally-reachable use case work, rather than requiring two separate solutions.¶
What we mean by "interoperate" is that a host on a stub network:¶
Discoverability here means "discoverable using DNS, or DNS Service Discovery". As an example, when one host connected to a specific WiFi network wishes to discover services on hosts connected to that same WiFi network, it can do so using multicast DNS (RFC6762), which is an example of DNS Service Discovery. Similarly, when a host on some other network wishes to discover the same service, it must use DNS-based DNS Service Discovery [RFC6763]. In both cases, "discoverable using DNS" means that the host has an entry in the DNS.¶
NOTE: it may be tempting to ask, why do we lump discoverability in with reachability and addressability, both of which are essentially Layer 3 issues? The answer is that it does us no good to automatically set up connectivity between stub network hosts and infrastructure hosts if the infrastructure hosts have no mechanism for learning about the availability of services provided by stub network hosts. For stub networks that only consume cloud services this will not be an issue, but for stub networks that provide services, e.g. the incompatible media use case mentioned earlier, discoverability is necessary in order for stub network connectivity to be useful.¶
Ability to acquire an IP address that can be used to communicate means that the IP address a host on the stub network acquires can be used to communicate with it by hosts on neighbor networks, for locally reachable stub networks, or by hosts on any network, for globally reachable networks. Various means of providing such addresses are discussed later.¶
Reachability to networks on which applicable hosts are attached means that when a host on the stub network has the IP address of an applicable host with which it intends to communicate, that host knows of a next-hop router to which it can send datagrams, so that they will ultimately reach the host with that IP address.¶
Reachability from networks on which applicable hosts are attached means that when such a host has a datagram destined for an IP address on the stub network, a next-hop router is known by that host which, when the datagram is sent to that router, will ultimately result in the datagram reaching the intended stub network host.¶
In addition to the interoperability goals we've described above, the additional goal for stub networks is that they be able to be connected automatically, with no user intervention. The experience of connecting a stub network to an infrastructure should be as straightforward as connecting a new host to the same infrastructure network.¶
Currently there is one known way to accomplish what we are describing here [[Michael, does ANIMA have a second way?]]. The Homenet working group produced a protocol, HomeNet Configuration Protocol (HNCP), the purpose of which is to allow a collection of routers to self-configure. HNCP is not technically constrained to home environments; in principle, it can work in any environment.¶
The problem with HNCP is twofold. First, it only works if it is deployed on all routers within the network infrastructure for a site. Secondly, it attempts to do too much, and invents too much that is new. Let's look at these in order.¶
First, HNCP only works when deployed on all routers within the network infrastructure. To be clear, this does not mean that it is impossible to use HNCP on a network where, for instance, the edge router(s) do not support HNCP. What it does mean is that if this configuration works, the reason it works is that the network supports prefix delegation to routers inside the network. So a router doing HNCP can get a prefix using prefix delegation from, for example, an edge router, and this will work.¶
Unfortunately, the way that such an HNCP server should behave is not documented, and it's not actually clear how it should behave. What if the DHCP server allocates it a /64? HNCP is designed to get a larger prefix and subdivide it-there is no provision for requesting multiple delegations. So if we wanted to use HNCP to solve this problem, we would need to do additional work.¶
Secondly, HNCP tries to do too much, and invents too much that is new. HNCP is a complicated protocol for propagating network configuration information in a mesh. It does not assume that any network is a stub network, and because of that, using it to support stub networks is needlessly complicated.¶
Despite having been an IETF proposed standard since 2016, and having been worked on for quite some time before that, it is not possible to purchase a router that implements HNCP. There exists a prototype implementation in OpenWRT, but getting it to actually work is problematic, and many problems have been left unsolved, and would be quite difficult to solve without additional standards work.¶
Because of the first point-the utter lack of commercial implementations of HNCP-any stub network solution that is intended to be deployed to arbitrary networks can't rely on the availability of HNCP. This may come in the future, but is not available now, and may never be. Therefore, whatever approach is taken MAY use HNCP if available, but MUST work without HNCP. Therefore, using HNCP represents additional implementation complexity; whether this is worth doing is something that should be considered, but because using HNCP is necessarily optional, it probably makes the most sense to assume that any functionality provided by HNCP will be external to the stub network router, and that the stub network router itself need not participate in the HNCP mesh.¶
Proxy Neighbor Discovery provides reachability to hosts on the stub network by simply pretending that they are on the infrastructure network. This reachability can be local or global depending on what IPv6 service (if any) is available on the infrastructure link. The use of Proxy ND for providing connectivity to stub networks is described in [RFC8929].¶
If IPv6 service is available on the infrastructure link, this service can be used to provide addressability on the stub network, and also provides addressability on the infrastructure link.¶
If IPv6 service is not available on the infrastructure link, addressability for proxy ND can be provided by advertising an on-link autoconfigurable prefix in a Router Advertisement offered by the stub router.¶
Discoverability for stub network hosts can be provided using DNS-SD service registration protocol on the stub network, in combination with an Advertising Proxy on the stub router which would advertise registered services to the infrastructure link.¶
Discoverability of infrastructure link hosts by stub network hosts can be provided using a DNS-SD discovery proxy and/or regular DNS. As long as the stub network requires that each stub router provide a DNS-SD Discovery Proxy and also provide name resolution, this will work even in the multiple stub router case.¶
Can definitely work in specific circumstances, but probably doesn't lend itself to full automation.¶
Reachability to the stub network is provided using the Route Information Option [RFC4191] in a router advertisement [RFC4861] issued by the stub router. Since the stub router does not provide IPv6 connectivity, it must not advertise itself as a default router. Each stub router can provide a default route to the stub network.¶
Addressability on the stub network is provided using a ULA prefix generated by the stub router. Addressibility on the infrastructure link is either provided by the infrastructure, or else must be provided by the stub router.¶
Discoverability for this approach is the same as for the Proxy ND approach.¶
This option has the advantage of relying primarily on ordinary IPv6 routing, as opposed to workarounds like proxy neighbor discovery or NAT64. The cooperation that is required between stub routers is minimal: they need simply minimize the advertising of redundant information. When redundant information is advertised, this is an aesthetic issue rather than an operational issue, and can be allowed to heal gradually.¶
Additionally, this option does not require any new behavior on the part of existing hosts or routers. It does assume that infrastructure hosts actually implement [RFC4191], but it is not unreasonable to expect that this either is already the case, or can easily be accomplished. It also assumes that the infrastructure does not enforce RA Guard [RFC6105]. This is compatible with the recommendations in RFC6105, which indicates that RA guard needs to be configured before it is enabled.¶
The approach described in this section only makes it possible for stub network hosts to interoperate with hosts on the link to which the stub router is directly attached. The "Global Reachability" approach talks about how to establish interoperability between stub network hosts and hosts on links to which the stub network is not directly attached.¶
Global reachability for stub networks requires either the use of NAT64, or else the presence of global IPv6 service on the link. As such it is more of an add-on approach than a different approach. This section talks about a specific example of global reachability: how to make global reachability work for the "Stub Reachability using RA" approach mentioned earlier.¶
The "global reachability" approach has applicability both in the literal sense, and also in the sense of "reachability beyond the link to which the stub router is directly attached." The behavior of the stub router is the same in both cases: it is up to the network infrastructure what prefix is delegated to the stub router, and what reachability is provided.¶
Reachability in this case requires integration into the routing infrastructure. This is most easily accomplished by having the DHCPv6 prefix delegation server add an entry in the routing table pointing to the stub router to which the prefix has been delegated. Stub routers can also advertise reachability to the stub network using router advertisements, but these will only work on the local link.¶
Addressability in this case for hosts on the infrastructure link is assumed to be provided by the infrastructure, since we are relying on the infrastructure to provide DHCPv6 prefix delegation. Addressibility on the stub network is provided using the prefix acquired with prefix delegation.¶
Discoverability for devices on the link to which the stub network is attached can be done as described earlier under the "Proxy ND" approach.¶
This approach should be a proper superset of the "Stub Reachability using RA" approach. The primary technical challenge here is specifying how multiple stub routers cooperate in doing prefix delegation.¶
This document generally assumes that stub networks only support IPv6. Bidirectional reachability for IPv4 can be provided using a combination of NAT44 and Port Control Protocol [RFC6887]. The use of NAT44 and PCP in this way has already been solved and need not be discussed here.¶
Reachability is complicated for NAT64. Typical NAT64 deployments provide reachability from the stub network to the rest of the Internet, but do not provide reachability from the rest of the internet to the stub network. As with NAT44 and PCP, this type of reachability is a solved problem and need not be discussed here. To provide complete reachability to the IPv4 internet, a stub router must not only provide reachability to the cloud, but also reachability from the cloud. That additional work is discussed here.¶
To provide reachability from the cloud to devices on the network, devices on the network will need to obtain static mappings from the external IPv4 address and a port to the internal IPv6 address and a port. There are three ways to do this:¶
The first option has the advantage that the stub host is in complete control over what is advertised. However, it places an additional burden on the stub host which may not be desirable: the host has to implement PCP and link the PCP port allocation to the SRP registration.¶
For a constrained network device, it is most likely preferable to combine the two transactions: the SRP server can receive the registration from the stub host and acquire a PCP mapping for it, and then register an AAAA and A record for the host along with an SRV record for the IPv4 and IPv6 mappings. The hostname mapping would need to be different for the A record and the AAAA record in order to avoid spurious connections to the IPv4 port on the IPv6 address and vice versa.¶
Addressability on the stub network can be provided using a ULA prefix specific to the stub network or, if NAT64 is being used in addition to one of the other solutions discussed here, the prefix allocated on the stub network for that purpose can also be used for NAT64.¶
IPv4 addressability on the infrastructure network is provided by the infrastructure network. It is also possible that the infrastructure network is an IPv6 network. In that case, the NAT64 edge router may be provided by the infrastructure as well.¶
The discoverability described for the "ND Proxy" approach should work here as well, except for the caveat mentioned above under "reachability".¶
Support for NAT64 may be required for some deployments. NAT64 support requires either close cooperation between stub routers, or else requires that the NAT64 translation be done externally. The latter choice is likely quite a bit easier; solutions that provide load balancing and high availability are already available on the market, and hence do not require that the stub routers perform this function. This is expected to be the best approach to serve the needs of consumers of this capability.¶
We can divide the set of hosts needing to be discovered and the set of hosts needing to discover them into four categories:¶
To enable stub hosts to discover direct hosts, a Discovery Proxy [RFC8766] can be used. This must be resident on any stub network router that is seen by the stub host as a resolver.¶
To enable stub hosts to discover infrastructure hosts using DNS-SD [RFC6763], the infrastructure must provide support for RFC6763 service discover using DNS.¶
To enable stub hosts to discover infrastructure hosts and cloud hosts using DNS, DNS resolution must be provided by the stub router, and the infrastructure must additionally provide the stub router with the ability to resolve names.¶
To enable direct hosts to discover stub hosts, stub routers must implement a DNS-SD Advertising Proxy. Stub hosts must register with the advertising proxy using SRP.¶
To enable infrastructure hosts to discover stub hosts, stub routers must provide authoritative DNS service for the stub network link so that it can be integrated into the infrastructure DNS-SD service. To do this automatically will require additional protocol work.¶
To enable cloud hosts to discover stub hosts, stub hosts would need to register with the DNS, and the infrastructure would need to make those registrations available globally, perhaps with whitelisting. This is probably not a very widely applicable use case, and we do not consider specifying how this works to be part of the work of this document.¶
In the case of a stub network that has multiple stub routers, it is possible that, either when the stub network is initially set up, or subsequently, one or more stub routers might be connected to a different infrastructure link than one or more other stub routers. There are two viable approaches to this problem:¶
Explain further.¶