Internet-Draft | IPv6 over OMNI Interfaces | June 2024 |
Templin | Expires 27 December 2024 | [Page] |
Air/land/sea/space mobile nodes (e.g., aircraft of various configurations, terrestrial vehicles, seagoing vessels, space systems, enterprise wireless devices, pedestrians with cell phones, etc.) communicate with networked correspondents over wireless and/or wired-line data links and configure mobile routers to connect end user networks. This document presents a multilink virtual interface specification that enables mobile nodes to coordinate with a network-based mobility service, fixed node correspondents and/or other mobile node peers. The virtual interface provides an adaptation layer service suited for both mobile and more static environments such as enterprise and home networks. Both Provider-Aggregated (PA) and Provider-Independent (PI) addressing services are supported. This document specifies the transmission of IP packets over Overlay Multilink Network (OMNI) Interfaces.¶
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 27 December 2024.¶
Copyright (c) 2024 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.¶
Air/land/sea/space mobile nodes (e.g., aircraft of various configurations, terrestrial vehicles, seagoing vessels, space systems, enterprise wireless devices, pedestrians with cellphones, etc.) configure mobile routers with multiple interface connections to wireless and/or wired-line data links. These data links often have diverse performance, cost and availability properties that can change dynamically according to mobility patterns, flight phases, proximity to infrastructure, etc. The mobile router acts as a Client of a network-based Mobility Service (MS) by configuring a virtual interface over its underlay interface data link connections.¶
Each Client configures a virtual network interface (termed the "Overlay Multilink Network Interface (OMNI)") as a thin layer over its underlay interfaces (which may themselves connect to virtual or physical links). The OMNI interface is therefore the only interface abstraction exposed to the IP layer and behaves according to the Non-Broadcast, Multiple Access (NBMA) interface principle, while each underlay interface appears as a link layer communication channel in the architecture. The OMNI interface internally employs the "OMNI Adaptation Layer (OAL)" to ensure that original IP packets or parcels [I-D.templin-6man-parcels2], [I-D.templin-intarea-parcels2] are adapted to diverse underlay interfaces with heterogeneous properties.¶
The OMNI interface connects to a virtual overlay known as the "OMNI link". The OMNI link spans one or more Internetworks that may include private-use infrastructures (e.g., enterprise networks, operator networks, etc.) and/or the global public Internet itself. Together, OMNI and the OAL provide the foundational elements required to support the "6 M's of Modern Internetworking", including:¶
Multilink - a Client's ability to coordinate multiple diverse underlay interfaces as a single logical unit (i.e., the OMNI interface) to achieve the required communications performance and reliability objectives.¶
Multinet - the ability to span the OMNI link over a segment routing topology with multiple diverse administrative domain network segments while maintaining seamless end-to-end communications between mobile Clients and correspondents such as air traffic controllers, fleet administrators, etc.¶
Mobility - a Client's ability to change network points of attachment (e.g., moving between wireless base stations) which may result in an underlay interface address change, but without disruptions to ongoing communication sessions with peers over the OMNI link.¶
Multicast - the ability to send a single network transmission that reaches multiple Clients belonging to the same interest group, but without disturbing other Clients not subscribed to the interest group.¶
Multihop - a mobile Client peer-to-peer relaying capability useful when multiple forwarding hops between peers may be necessary to "reach back" to an infrastructure access point connection to the OMNI link.¶
(Performance) Maximization - the ability to exchange large packets/parcels between peers without loss due to a link size restriction, and to adaptively adjust packet/parcel sizes to maintain the best performance profile for each independent traffic flow.¶
Client OMNI interfaces coordinate with the MS and/or OMNI peer nodes through IPv6 Neighbor Discovery (ND) control message exchanges [RFC4861]. The MS consists of a distributed set of service nodes (including Proxy/Servers and other infrastructure elements) that also configure OMNI interfaces. Automatic Extended Route Optimization (AERO) in particular provides a companion MS compatible with the OMNI architecture [I-D.templin-6man-aero3]. AERO discusses details of ND message based multilink forwarding, route optimization, mobility management, and multinet traversal while the fundamental aspects of OMNI link operation are discussed in this document.¶
Each OMNI interface provides a multilink nexus for exchanging inbound and outbound traffic flows via selected underlay interfaces. The IP layer sees the OMNI interface as a point of connection to the OMNI link. Each OMNI link assigns one or more associated Mobility Service Prefixes (MSPs), which are typically IP Global Unicast Address (GUA) prefixes. The MS then delegates Mobile Network Prefixes (MNPs) taken from an MSP to Client end systems as Provider-Independent (PI) address blocks. Clients in local domains also obtain Provider-Aggregated (PA) addresses from internal/external Stable Network Prefixes (SNPs) assigned to Proxy/Servers that connect the local domain to the global topology per [I-D.bctb-6man-rfc6296-bis]. If there are multiple OMNI links, the IP layer will see multiple OMNI interfaces.¶
Clients receive SNP addresses and optionally also MNP prefix delegations through IPv6 ND control message exchanges with Proxy/Servers over MANETs, Access Networks (ANETs) and/or open Internetworks (INETs). Clients sub-delegate MNPs to downstream-attached End-user Networks (ENETs) independently of the underlay interfaces selected for upstream data transport. Each Client acts as a fixed or mobile router on behalf of ENET peers, and uses OMNI interface control messaging to coordinate with Hosts, Proxy/Servers and/or other Clients. The Client iterates its control messaging over each of the OMNI interface's (M)ANET/INET underlay interfaces in order to register each interface with the MS (see Section 13). The Client can also provide multihop forwarding services for a recursively extended chain of other Clients and Hosts connected via downstream-attached ENETs.¶
Clients may connect to multiple distinct OMNI links within the same OMNI domain by configuring multiple OMNI interfaces, e.g., omni0, omni1, omni2, etc. Each OMNI interface is configured over a distinct set of underlay interfaces and provides a nexus for Safety-Based Multilink (SBM) operation. The IP layer applies SBM routing to select a specific OMNI interface, then the selected OMNI interface applies Performance-Based Multilink (PBM) internally to select appropriate underlay interfaces. Applications select SBM topologies based on IP layer Segment Routing [RFC8402], while each OMNI interface orchestrates PBM internally based on OAL Multinet traversal.¶
OMNI provides a link model suitable for a wide range of use cases. For example, the International Civil Aviation Organization (ICAO) Working Group-I Mobility Subgroup is developing a future Aeronautical Telecommunications Network with Internet Protocol Services (ATN/IPS) and has issued a liaison statement requesting IETF adoption [ATN] in support of ICAO Document 9896 [ATN-IPS]. The IETF IP Wireless Access in Vehicular Environments (ipwave) working group has further included problem statement and use case analysis for OMNI in [RFC9365]. Still other communities of interest include AEEC, RTCA Special Committee 228 (SC-228) and NASA programs that examine commercial aviation, Urban Air Mobility (UAM) and Unmanned Air Systems (UAS). Pedestrians with handheld mobile devices, home and small office networks, enterprise networks and many others represent additional large classes of potential OMNI users.¶
This document specifies the transmission of original IP packets/parcels and control messages over OMNI interfaces. The operation of both IP protocol versions (i.e., IPv4 [RFC0791] and IPv6 [RFC8200]) is specified as the network layer data plane, while OMNI interfaces use IPv6 ND messaging in the control plane independently of the data plane protocol(s). OMNI interfaces also provide an adaptation layer based on encapsulation and fragmentation over heterogeneous underlay interfaces as an OAL sublayer between L3 and L2. OMNI and the OAL are specified in detail throughout the remainder of this document.¶
The terminology in the normative references applies; especially, the terms "link" and "interface" are the same as defined in the IPv6 [RFC8200] and IPv6 Neighbor Discovery (ND) [RFC4861] specifications. This document assumes the following IPv6 ND control plane message types: Router Solicitation (RS), Router Advertisement (RA), Neighbor Solicitation (NS), Neighbor Advertisement (NA), unsolicited NA (uNA) and Redirect.¶
The terms "All-Routers multicast", "All-Nodes multicast" and "Subnet-Router anycast" are the same as defined in [RFC4291]. Also, IPv6 ND state names, variables and constants including REACHABLE, ReachableTime and REACHABLE_TIME are the same as defined in [RFC4861].¶
The term "IP" is used to refer collectively to either Internet Protocol version (i.e., IPv4 [RFC0791] or IPv6 [RFC8200]) when a specification at the layer in question applies equally to either version.¶
The terms Host, Client and Proxy/Server are intentionally capitalized to denote an instance of that particular node type that also configures an OMNI interface and engages the OMNI Adaptation Layer.¶
The terms "application layer (L5 and higher)", "transport layer (L4)", "network layer (L3)", "(data) link layer (L2)" and "physical layer (L1)" are used consistently with common Internetworking terminology, with the understanding that reliable delivery protocol users of UDP are considered as transport layer elements. The OMNI specification further defines an "adaptation layer" positioned below the network layer but above the link layer, which may include physical links and Internet- or higher-layer tunnels. A (network) interface is a node's attachment to a link (via L2), and an OMNI interface is therefore a node's attachment to an OMNI link (via the adaptation layer).¶
The terms "IP jumbogram", "advanced jumbo (AJ)" and "IP parcel" refer to special packet formats that enable a new link model for the Internet as discussed in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2].¶
The following terms are defined within the scope of this document:¶
OMNI interfaces limit the size of their IPv6 ND control plane messages (plus any original IP packet/parcel attachments) to the minimum IPv6 link MTU minus overhead for adaptation and link layer encapsulation. If there are sufficient OMNI parameters and/or IP packet/parcel attachments that would exceed this size, the OMNI interface forwards the information as multiple smaller IPv6 ND messages and the recipient accepts the union of all information received. This allows the messages to travel without loss due to a size restriction over secured control plane paths that include IPsec tunnels [RFC4301], secured direct point-to-point links and/or unsecured paths that require an authentication signature.¶
Host, Client and Proxy/Server OMNI interfaces that employ IPv6 ND control plane messaging maintain per-neighbor state in Neighbor Cache Entries (NCEs). Each NCE is indexed by the neighbor's network layer address(es) while the neighbor's OAL encapsulation address provides context for Identification verification. The IPv6 ND Protocol Constants defined in Section 10 of [RFC4861] are used in their same format and meaning in this document.¶
The L3, adaptation and (virtual) L2 layers each include distinct packet Identification numbering spaces. The adaptation layer employs an 8-octet Identification numbering space that is distinct from L3/L2 spaces, with an Identification value appearing in an IPv6 Extended Fragment Header [I-D.templin-6man-ipid-ext2] or an OMNI Compressed Header (OCH) (see: Section 6.5) in each adaptation layer encapsulation.¶
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 [RFC2119][RFC8174] when, and only when, they appear in all capitals, as shown here.¶
An OMNI interface is a virtual interface configured over one or more underlay interfaces, which may be physical (e.g., an aeronautical radio link, a cellular wireless link, etc.) or virtual (e.g., an internet-layer or higher-layer "tunnel"). The OMNI interface architectural layering model is the same as in [RFC5558][RFC7847], and augmented as shown in Figure 1. The network layer therefore sees the OMNI interface as a single L3 interface nexus for multiple underlay interfaces that appear as L2 communication channels in the architecture.¶
Each underlay interface provides an L2/L1 abstraction according to one of the following models:¶
(M)ANET interfaces connect to a (M)ANET that is separated from the open INET by Proxy/Servers. The (M)ANET interface may be either on the same link segment as a Proxy/Server, or separated from a Proxy/Server by multiple adaptation layer and/or L2 hops. (Note that NATs may appear internally within a (M)ANET or on the Proxy/Server itself and may require NAT traversal the same as for the INET case.) The OMNI interface configures an ALVIF over each ANET interface, while MANET interfaces already appear as adaptation layer interfaces and need not configure an ALVIF.¶
INET interfaces connect to an INET either natively or through IP Network Address Translators (NATs). Native INET interfaces have global IP addresses that are reachable from any INET correspondent. NATed INET interfaces typically configure private IP addresses and connect to a private network behind one or more NATs with the outermost NAT providing INET access. The same as for ANET interfaces, the OMNI interface configures an ALVIF over each INET interface.¶
ENET interfaces connect a Client's downstream-attached networks, where the Client provides forwarding services for ENET Host and Client communications to remote peers. An ENET may be as simple as a small IoT sub-network that travels with a mobile Client to as complex as a large private enterprise network that the Client connects to a larger *NET. Downstream-attached Hosts and Clients see the ENET as a *NET and see the (upstream) Client as a Proxy/Server.¶
VPN interfaces use security encapsulations (e.g. IPsec tunnels) over underlay networks to connect Client, Proxy/Server or other critical infrastructure nodes. VPN interfaces provide security services at lower layers of the architecture (L2/L1), with securing properties similar to Direct point-to-point interfaces.¶
Direct point-to-point interfaces securely connect Clients, Proxy/Servers and/or other critical infrastructure nodes over physical or virtual media that does not transit any open Internetwork paths. Examples include a line-of-sight link between a remote pilot and an unmanned aircraft, a fiberoptic link between gateways, etc.¶
The OMNI interface forwards original IP packets/parcels from the network layer using the OMNI Adaptation Layer (OAL) (see: Section 5) as an encapsulation and fragmentation sublayer service. This "OAL source" then further encapsulates the resulting OAL packets/fragments in underlay network headers (e.g., UDP/IP, IP-only, Ethernet-only, etc.) to create L2 encapsulated "carrier packets" for fragmentation and transmission over underlay interfaces. The target OMNI interface then receives the carrier packets from underlay interfaces and performs L2 reassembly/decapsulation.¶
If the resulting OAL packets/fragments are addressed to itself, the OMNI interface performs reassembly/decapsulation as an "OAL destination" and delivers the original IP packet/parcel to the network layer. If the OAL packets/fragments are addressed to another node, the OMNI interface instead re-encapsulates them in new underlay network L2 headers as an "OAL intermediate system" then performs L2 fragmentation and forwards the resulting carrier packets over an underlay interface. The OAL source and OAL destination are seen as "neighbors" on the OMNI link, while OAL intermediate systems provide a virtual bridging service that joins the segments of a (multinet) Segment Routing Topology (SRT).¶
The OMNI interface transports carrier packets over either secured or unsecured underlay interfaces to access the secured/unsecured OMNI link spanning trees as discussed further throughout the document. Carrier packets that carry control plane messages over secured underlay interfaces use secured L2/L1 services such as IPsec, direct encapsulation over secured point-to-point links, etc. Carrier packets that carry data plane messages over unsecured underlay interfaces instead use L2 encapsulations appropriate for public or private Internetworks and are subject for the following sections.¶
The OMNI interface and its OAL can forward original IP packets/parcels over underlay interfaces while including/omitting various lower layer encapsulations including OAL, UDP, IP and (ETH)ernet or other link layer header. The network layer can also engage underlay interfaces directly while bypassing the OMNI interface entirely when necessary. This architectural flexibility may be beneficial for underlay interfaces (e.g., some aviation data links) for which encapsulation overhead is a primary consideration. OMNI interfaces that send original IP packets/parcels directly over underlay interfaces without invoking the OAL can only reach peers located on the same OMNI link segment. Source Clients can instead use the OAL to coordinate with target Clients in the same or different OMNI link segments by sending initial carrier packets to a First-Hop Segment (FHS) Proxy/Server. The FHS Proxy/Sever then sends the carrier packets into the SRT spanning tree, which transports them to a Last-Hop Segment (LHS) Proxy/Server for the target Client.¶
The OMNI interface encapsulation/decapsulation layering possibilities are shown in Figure 2 below. Imaginary vertical lines drawn between the Network Layer at the top of the figure and Underlay Interfaces at the bottom of the figure denote the various encapsulation/decapsulation layering combination possibilities. Common combinations include IP-only (i.e., direct access to underlay interfaces with or without using the OMNI interface), IP/IP, IP/UDP/IP, IP/UDP/IP/ETH, IP/OAL/UDP/IP, IP/OAL/UDP/ETH, etc.¶
The OMNI/OAL model gives rise to a number of opportunities:¶
Clients coordinate with the MS and receive both SNP addresses and MNP delegations through IPv6 ND control plane message exchanges with Proxy/Servers. Since GUA and ULA addresses are managed for uniqueness, no Duplicate Address Detection (DAD) or Multicast Listener Discovery (MLD) messaging is necessary over the OMNI interface.¶
underlay interfaces on the same L2 link segment as a Proxy/Server do not require any L3 addresses (i.e., not even link-local) in environments where communications are coordinated entirely over the OMNI interface.¶
as underlay interface properties change (e.g., link quality, cost, availability, etc.), any active interface can be used to update the profiles of multiple additional interfaces in a single message. This allows for timely adaptation and service continuity under dynamically changing conditions.¶
coordinating underlay interfaces in this way allows them to be represented in a unified MS profile with provisions to support the "6 M's of Modern Internetworking".¶
header compression and path MTU determination is conducted on a per-flow basis, with each flow adapting to the best performance profiles and path selections.¶
exposing a single virtual interface abstraction to the network layer allows for multilink operation (including QoS based link selection, carrier packet replication, load balancing, etc.) at L2 while still permitting L3 traffic shaping based on, e.g., DSCP, flow label, etc.¶
the OMNI interface supports multinet traversal over the SRT when communications across different administrative domain network segments are necessary. This mode of operation would not be possible via direct communications over the underlay interfaces themselves.¶
the OAL supports lossless and adaptive path MTU mitigations not available for communications directly over the underlay interfaces themselves. The OAL supports "packing" of multiple original IP payload packets/parcels within a single OAL "super-packet" and also supports transmission of IP packets/parcels of all sizes up to and including (advanced) jumbograms.¶
the OAL assigns per-packet Identification values that allow for adaptation/link layer reliability and data origin authentication.¶
L3 sees the OMNI interface as a point of connection to the OMNI link; if there are multiple OMNI links, L3 will see multiple OMNI interfaces.¶
Multiple independent OMNI interfaces can be used for increased fault tolerance through Safety-Based Multilink (SBM), with Performance-Based Multilink (PBM) applied within each interface.¶
Multiple independent OMNI links can be joined together into a single link without requiring renumbering of infrastructure elements, since the GUAs/ULAs assigned by Proxy/Servers of the different links will be mutually exclusive.¶
the OMNI/OAL model supports transmission of new forms of IP packets known as "IP parcels and Advanced Jumbos (AJs)" that improve performance and efficiency for both transport layer protocols and networked paths.¶
OMNI provides robust support for both Provider-Aggregated (PA) and Provider-Independent (PI) addressing resulting in a versatile service for all Client use cases.¶
Figure 3 depicts the architectural model for a source Client with an attached ENET connecting to the OMNI link via multiple independent *NETs. The Client's OMNI interface forwards adaptation layer IPv6 ND solicitation messages over available *NET underlay interfaces using any necessary L2 encapsulations. The IPv6 ND messages traverse the *NETs until they reach an FHS Proxy/Server (FHS#1, FHS#2, ..., FHS#n), which returns an IPv6 ND advertisement message and/or forwards a proxyed version of the message over the SRT to an LHS Proxy/Server near the target Client (LHS#1, LHS#2, ..., LHS#m). The Hop Limit in IPv6 ND messages is not decremented due to encapsulation; hence, the source and target Client OMNI interfaces appear to be attached to a common link.¶
After the initial IPv6 ND message exchange, the source Client (as well as any nodes on its attached ENETs) can send carrier packets to the target Client via the OMNI interface. OMNI interface multilink services will send the carrier packets via FHS Proxy/Servers for the correct underlay *NETs. The FHS Proxy/Server then re-encapsulates the carrier packets and forwards them over the SRT which delivers them to an LHS Proxy/Server, and the LHS Proxy/Server in turn re-encapsulates and forwards them to the target Client. (Note that when the source and target Client are on the same SRT segment, the FHS and LHS Proxy/Servers may be one and the same.)¶
Mobile Clients select a MAP Proxy/Server (not shown in the figure), which will often be one of their FHS Proxy/Servers but could also be any Proxy/Server on the OMNI link. Clients then register all of their *NET underlay interfaces with the MAP Proxy/Server via per interface FHS Proxy/Servers in a pure proxy role. The MAP Proxy/Server then provides a designated router that advertises the Client's MNPs into the OMNI link routing system, and the Client can quickly migrate to a new MAP Proxy/Server if the former becomes unresponsive.¶
Clients therefore use Proxy/Servers as gateways into the SRT to reach OMNI link correspondents via a spanning tree established in a manner outside the scope of this document. Proxy/Servers forward critical MS control messages via the secured spanning tree and forward other messages via the unsecured spanning tree (see Security Considerations). When AERO route optimization is applied, Clients can instead forward directly to correspondents in the same SRT segment to reduce Proxy/Server and/or Gateway load.¶
Note: while not shown in the figure, a Client's ENET may connect many additional Hosts and even other Clients in a recursive extension of the OMNI link. This OMNI virtual link extension will be discussed more fully throughout the document.¶
Note: Original IP packets/parcels sent into an OMNI interface will receive consistent consideration according to their size as discussed in the following sections, while those sent directly over underlay interfaces that exceed the underlay network path MTU are dropped with an ordinary ICMP Packet Too Big (PTB) message returned. These PTB messages are subject to loss the same as for any non-OMNI IP interface [RFC2923].¶
The OMNI interface observes the link nature of tunnels, including the Maximum Transmission Unit (MTU), Effective MTU to Send (EMTU_S), Effective MTU to Receive (EMTU_R) and the role of fragmentation and reassembly [I-D.ietf-intarea-tunnels]. The OMNI interface is configured over one or more underlay interfaces as discussed in Section 4, where underlay links and network paths may have diverse MTUs. OMNI interface considerations for accommodating original IP packets/parcels of various sizes are discussed in the following sections.¶
IPv6 underlay interfaces are REQUIRED to configure a minimum MTU of 1280 octets and a minimum EMTU_R of 1500 octets [RFC8200]. Therefore, the minimum IPv6 path MTU is 1280 octets since routers on the path are not permitted to perform network fragmentation even though the destination is required to reassemble more. The network therefore MUST forward original IP packets/parcels as large as 1280 octets without generating an IPv6 Path MTU Discovery (PMTUD) Packet Too Big (PTB) message [RFC8201]. Since each OAL intermediate system must configure an EMTU_R of at least 65535 octets (see: Section 6.3), the source can apply "source fragmentation" for carrier packets as large as that size but this does not affect the minimum IPv6 path MTU.)¶
IPv4 underlay interfaces are REQUIRED to configure a minimum MTU of 68 octets [RFC0791] and a minimum EMTU_R of 576 octets [RFC0791][RFC1122]. Therefore, when the Don't Fragment (DF) bit in the IPv4 header is set to 0 the minimum IPv4 path MTU is 576 octets since routers on the path support network fragmentation and the destination is required to reassemble at least that much. The OMNI interface therefore SHOULD set DF to 0 in the IPv4 encapsulation headers of carrier packets no larger than 576 octets, and SHOULD set DF to 1 in larger carrier packets unless it has a way to determine the EMTU_R of the next OAL hop as discussed in Section 6.15. This limitation is therefore relaxed by the requirement that each OAL intermediate system must configure a minimum EMTU_R of 65535 octets (see: Section 6.3) allowing for IPv4 fragmentation and reassembly for larger carrier packets.¶
The OMNI interface itself sets an "unlimited" MTU of (2**32 - 1) octets. The network layer therefore unconditionally admits all original IP packets/parcels into the OMNI interface, where the adaptation layer accommodates them if possible according to their size. For each parcel that it accommodates, the OAL source within the OMNI interface first performs "parcellation" if necessary to break large parcels into smaller sub-parcels that can transit the OAL path (see: Section 5.1). The OAL source then invokes adaptation layer encapsulation/fragmentation services to transform all original IP packets and (sub-)parcels no larger that 65535 octets into OAL packets/fragments. The OAL source then applies L2 encapsulation and fragmentation if necessary to form carrier packets and finally forwards the carrier packets via underlay interfaces.¶
When the OAL source performs IPv6 encapsulation and fragmentation (see: Section 6), the Payload Length field limits the maximum-sized original IP packet/parcel that the OAL can accommodate while applying IPv6 fragmentation to (2**16 - 1) = 65535 octets (i.e., not including the OAL encapsulation header lengths). The OAL source is also permitted to forward packets/parcels larger than this size as a best-effort delivery service if the L2 path can accommodate them through "jumbo-in-jumbo" encapsulation (see: Section 5.2); otherwise, the OAL source discards the packet and arranges to return a PTB "hard error" to the original source (see: Section 6.9).¶
Each OMNI interface therefore sets a minimum EMTU_R of 65535 octets (plus the length of the OAL encapsulation headers), and each OAL destination must consistently either accept or reject still larger whole packets that arrive over any of its underlay interfaces according to their size. If an underlay interface presents a whole packet larger than the OAL destination is prepared to accept (e.g., due to a buffer size restriction), the OAL destination discards the packet and arranges to return a PTB "hard error" to the OAL source which in turn forwards the PTB to the original source (see: Section 6.9).¶
As specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2], an IP parcel is an IP jumbogram variant for which an IPv6 Parcel Payload Option field encodes a value between 256 and 65535 octets denoting the non-final transport layer protocol segment length while the parcel body includes as many as 64 individual transport layer protocol segments. The Jumbo Payload length field is modified to include a Parcel Index field plus flags followed by a Parcel Payload Length field which together determine the size and number of transport layer segments included in the parcel.¶
IP parcel "parcellation" and "reunification" procedures for OMNI interfaces are specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2], while OAL encapsulation and fragmentation procedures are specified in Section 6.13 of this document. The maximum-sized IP parcel that can be conveyed over an OMNI interface using OAL parcellation and IPv6 fragmentation-based assured delivery is one with 64 segments of 65535 (minus headers) octets in length. (The OAL source can instead forward large parcels as a best-effort service using jumbo-in-jumbo encapsulation if the OAL/L2 path can accommodate them.)¶
IP parcels follow the same link models described for Advanced Jumbos below. IP parcels that accumulate link errors on the path are subject to error detection and correction at the final destination.¶
ENET end systems that implement either the full OMNI interface (i.e., Clients) or enough of the OAL to process parcels (i.e., Hosts) are permitted to exchange parcels with consenting peers. This accommodates nodes that connect to the OMNI link but do not assign OAL addresses.¶
While the maximum-sized original IP packet/parcel that the OAL can accommodate using IPv6 fragmentation-based assured delivery is 65535 octets, OMNI interfaces can forward much larger singleton parcels termed "Advanced Jumbos (AJs)" via jumbo-in-jumbo encapsulation as specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2]. For jumbo-in-jumbo encapsulation of large AJs, the OAL source appends an OAL IPv6 header plus extensions then appends any L2 headers to identify this as an AJ. Since the Jumbo Payload Length is 32 bits, the largest possible AJ is limited to (2**32 - 1) octets minus the lengths of any extension/encapsulation headers, or smaller still for transmission over underlay interfaces that include additional extensions/encapsulations.¶
Basic IPv6 jumbograms per [RFC2675] use the Jumbo Payload Option and set the IPv6 Payload Length field to 0. IP parcels and AJs instead use an adaptation of the IPv6 Minimum Path MTU option [RFC9268] known as the Parcel Payload Option. The OAL/L2 source forwards basic jumbograms and AJs as giant carrier packets using jumbo-in-jumbo encapsulation, noting that traditional 32-bit link CRCs do not provide adequate integrity protection for such large sizes [CRC]. If a basic jumbogram is dropped along the path to the OAL destination, the OAL source arranges to return an ICMP PTB "hard error" to the original source. If a parcel/AJ is dropped, the OAL source instead arranges to return ICMP PTB "soft errors" (see: Section 6.9).¶
AJs range in size from the largest possible unit as discussed above to the smallest unit that includes only the headers and a small or possibly even null payload. Intermediate hops forward AJs that follow a new DTN link model for the Internet (instead of dropping) even if link errors were incurred along the path. The AJ will then arrive at the destination along with any cumulative link errors collected on the path. The final destination then applies end-to-end integrity checks and/or error correction while requesting retransmission only as a last resort. This link model may be more appropriate for delay/disruption-tolerant environments such as anticipated for air/land/sea/space mobile Internetworking.¶
Advanced jumbo services for both IPv6 and IPv4 (including jumbo path probing and jumbo-in-jumbo encapsulation) are specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2].¶
The above sections primarily concern data plane aspects of the OMNI interface MTU and describe the data plane service model offered to the network layer. OMNI interfaces also internally employ a control plane service based on IPv6 Neighbor Discovery (ND) messaging. These control plane messages must be sent over secured underlay interfaces (e.g., IPsec tunnels, secured direct point-to-point links, etc.) or over unsecured paths but with an authentication signature included. In all control plane path cases, the IPv6 minimum MTU of 1280 octets must be assumed.¶
OMNI interfaces therefore offer an unlimited data plane MTU to the network layer but set a more conservative MTU for the internal control plane operation. OMNI interfaces assume a fixed control plane path MTU of 1280 octets (minus OAL encapsulation overhead) for transmission of IPv6 ND messages. OMNI interfaces should send multiple smaller IPv6 ND messages instead of singleton larger messages whenever possible to minimize fragmentation.¶
The OMNI interface forwards original IP packets/parcels from the network layer for transmission over one or more underlay interfaces. The OMNI Adaptation Layer (OAL) acting as the OAL source then applies IPv6 encapsulation to form OAL packets subject to OAL fragmentation producing fragments suitable for L2 encapsulation and transmission as carrier packets. These carrier packets may in turn be subject to IP fragmentation over underlay interface paths as described in Section 6.1. The carrier packets/fragments then travel over one or more underlay networks spanned by OAL intermediate systems in the SRT. Each successive OAL intermediate system performs L2 reassembly (if necessary) then re-encapsulates by removing the L2 headers of the first underlay network and appending L2 headers appropriate for the next underlay network while re-fragmenting if necessary. (This process supports the multinet concatenation capability needed for joining multiple diverse networks.) Following any forwarding by OAL intermediate systems, the carrier packets arrive at the OAL destination.¶
When the OAL destination receives the carrier packets, it performs L2 reassembly (if necessary) then discards the L2 headers and reassembles the resulting OAL fragments into an OAL packet as described in Section 6.3. The OAL destination next decapsulates the OAL packet to obtain the original IP packet/parcel which it then delivers to the network layer. The OAL source may be either the source Client or its FHS Proxy/Server, while the OAL destination may be either the LHS Proxy/Server or the target Client. Proxy/Servers (and SRT Gateways as discussed in [I-D.templin-6man-aero3]) may also serve as OAL intermediate systems.¶
The OAL presents an OMNI sublayer abstraction similar to ATM Adaptation Layer 5 (AAL5). Unlike AAL5 which performs segmentation and reassembly with fixed-length 53-octet cells over ATM networks, however, the OAL uses IPv6 encapsulation, fragmentation and reassembly with larger variable-length cells over heterogeneous networks. Detailed operations of the OAL are specified in the following sections.¶
When the network layer forwards an original IP packet/parcel into the OMNI interface, it either sets the TTL/Hop Limit for locally-generated packets or decrements the TTL/Hop Limit according to standard IP forwarding rules. The OAL source next creates an "OAL packet" by prepending an IPv6 encapsulation header in the spirit of [RFC2473]. The OAL source includes the IPv6 encapsulation header then sets Version to "OMNI-OFH" (see: Section 6.2) and Next Header to TBD1 (see: IANA Considerations).¶
When the OAL source performs IPv6 encapsulation, it next copies the "Type of Service/Traffic Class" [RFC2983] and "Explicit Congestion Notification (ECN)" [RFC3168] values in the original packet/parcel's IP header into the corresponding fields in the OAL IPv6 header then sets the IPv6 header "Flow Label" as specified in [RFC6438]. The OAL source next sets the IPv6 header Payload Length to the length of the original IP packet/parcel and sets Hop Limit to a value that is sufficiently large to support loop-free forwarding over multiple concatenated OAL intermediate hops. The OAL source next selects OAL IPv6 source and destination addresses associated with its own adaptation layer interface and the adaptation layer interface of the target.¶
The OAL source next inserts any necessary extension headers following the IPv6 header as specified in Section 6.4. For OAL data plane packets, the source first inserts any per-fragment extension headers (e.g., Hop-by-Hop, Routing, etc.) then inserts an IPv6 Extended Fragment Header (see: [I-D.templin-6man-ipid-ext2]) with an 8-octet (64-bit) OAL packet Identification. Note that the extension header insertions could cause the IPv6 Payload Length to exceed 65535 octets by a small amount when the original IP packet is (nearly) the maximum length. The OAL source then fragments the OAL packet if necessary according to an OAL Fragment Size (OFS) maintained in AERO Forwarding Vectors (AVFs) for each OAL destination. (The OAL source processes OAL packets with payloads that are no larger than the OFS and original IP packets/parcels larger than 65535 octets as "atomic fragments".) OAL fragments prepared by the source must not be fragmented further by OAL intermediate systems on the path to the OAL destination.¶
OAL packets that contain original IP parcels no larger than (64*65535) octets may be first subject to OMNI interface parcellation, after which the (sub-)parcels (as well as OAL packets that contain original IP packets no larger than 65535 octets) are subject to OAL fragmentation-based assured delivery. Advanced Jumbos (AJs) larger than 65535 octets (see: [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2]) are not eligible for OAL fragmentation but instead engage a best effort jumbo-in-jumbo encapsulation service as discussed in Section 5.2. (Note: the original source can optionally elect this best-effort jumbo-in-jumbo delivery service for any parcel/AJ regardless of its size.)¶
OAL fragmentation is conducted according to the IPv6 Extended Fragment Header (EFH) fragmentation specification in [I-D.templin-6man-ipid-ext2] with the exception that the IPv6 Payload Length may exceed 65535 by at most the length of the extension headers. The OAL source MUST set a "maximum OFS" to a size no smaller than 1024 octets and thereafter reduce or increase the "effective OFS" according to dynamic network control message feedback. The OAL source SHOULD limit the maximum OFS to a size no larger than 65279 octets unless it has assurance that a larger size can be accommodated. (Note that these sizes allow for up to 256 octets of L2 encapsulation relative to the IPv6 minimum MTU and maximum fragmented packet size.) If an OAL intermediate system or the OAL destination advertises a reduced size, the OAL source SHOULD reduce the effective OFS accordingly (to a size no smaller than 1024 octets) and can later increase the effective OFS as network conditions improve. When the OAL source performs fragmentation, it SHOULD produce the minimum number of fragments under the effective OFS constraints. The fragments produced MUST be non-overlapping and the portion of each non-final fragment following the IPv6 Extended Fragment Header MUST be equal in length while that of the final fragment MAY be smaller and MUST NOT be larger.¶
The OAL source discovers the maximum OFS by including an IPv6 Minimum Path MTU Hop-by-Hop Option [RFC9268] in the OAL encapsulation header of its Neighbor Solicitation (NS) / Neighbor Advertisement (NA) exchanges over the secured spanning tree used to establish multilink forwarding state (see: [I-D.templin-6man-aero3]). Each OAL intermediate system on the path sets the minimum path MTU in the NS message OAL extension header to the maximum OFS capable of traversing the next segment. (Note that segments traversed by L2 encapsulations such as IP tunnels can normally regard the MTU for their unsecured overlay network segments as 65535 octets while those traversed by direct point-to-point links and multihop MANET links must regard the link MTU as a restricting size; therefore, each OAL intermediate system MUST correctly recognize and honor the IPv6 Minimum Path MTU Hop-by-Hop Option. Note also that OAL intermediate systems forward the NS/NA messages in the control plane, but the returned MTU reflects the maximum OFS for the data plane.) When the OAL destination returns an NA message with an OAL header containing an IPv6 Minimum Path MTU Hop-by-Hop Option, the OAL source can then set the maximum OFS for this AFV by subtracting 256 from the returned MTU. The OAL source can later adaptively increase or decrease the effective OFS if it receives dynamic path MTU feedback from an OAL intermediate node or destination with the understanding that larger OFS sizes may provide better performance but also increase the retransmission unit in case of loss.¶
For each first fragment, the OAL source replaces the IPv6 Extended Fragment Header 1-octet "Reserved" field with the encoding shown in Figure 4:¶
For the first fragment, the OAL source then sets "Parcel ID", "(P)arcel" and "More (S)egments" as specified in Section 6.13.¶
For each consecutive fragment beginning with the first, the OAL source then writes a monotonically-increasing "ordinal" value between 0 and 63 in the Extended Fragment Header Index field. Specifically, the OAL source writes the ordinal value '0' for the first fragment, '1' for the first non-first fragment, '2' for the next, '3' for the next, etc. up to the final fragment. The final fragment may assign an ordinal as large as '63' such that at most 64 fragments are possible. During a network path change, an OAL intermediate system may apply further OAL fragmentation to produce minimum-length (sub-)fragments. The OAL destination will then reassemble these (sub-)fragments then combine each reassembled fragment with all other fragments of the same OAL packet and return rate-limited indications to inform the OAL source that the path has changed.¶
The OAL source finally encapsulates the fragments in L2 headers to form carrier packets for transmission over underlay interfaces, while retaining the fragments and their ordinal numbers (i.e., #0, #1, #2, etc.) for a brief period to support adaptation layer retransmissions (see: Section 6.8). OAL fragment and carrier packet formats are shown in Figure 5 (note that IPv4 carrier packets with DF=0 may include trailing checksums ("Csum") as discussed in Section 6.2).¶
The OAL source or intermediate system next encapsulates each OAL fragment (with either full or compressed headers) in L2 encapsulation headers to create a carrier packet. The OAL source or intermediate system (i.e., the L2 source) includes a UDP header as the innermost sublayer if NATs and/or filtering middleboxes might occur on the path. Otherwise, the L2 source includes a full/compressed IP header and/or an actual link layer header (e.g., such as for Ethernet-compatible links) as the innermost sublayer. The L2 source also appends any additional encapsulation sublayer headers necessary (e.g., IPsec AH/ESP, jumbo-in-jumbo encapsulation, etc.).¶
The L2 source encapsulates the OAL information immediately following the innermost L2 sublayer header. The L2 source next interprets the first 4 bits following the L2 headers as a Type field that determines the type of OAL header that follows. The OAL source sets Type to (OMNI-OFH) for an uncompressed IPv6 OMNI Full Header (OFH) or (OMNI-OCH1/2) for an OMNI Compressed Header, Type 1 (OCH1) or 2 (OCH2) as specified in Section 6.5. For IP packets/parcels that do not include an OAL IPv6 encapsulation header, the L2 source instead interprets the first 4 bits as a Version field that encodes '4' (OMNI-IP4) for an ordinary IPv4 packet/parcel or '6' (OMNI-IP6) for an ordinary IPv6 packet/parcel. Other Type values (including a Type for a Hop-by-Hop Options header that includes a Parcel Payload Option) may also appear as specified in Section 6.5.¶
The OAL node prepares the L2 encapsulation headers for OAL packets/fragments as follows:¶
For UDP/IP encapsulation, the L2 source sets the UDP source port to 8060 (i.e., the port number reserved for AERO/OMNI). When the L2 destination is a Proxy/Server or Gateway, the L2 source sets the UDP destination port to 8060; otherwise, the L2 source sets the UDP destination port to its cached port number value for the peer. The L2 source next sets the UDP Length the same as specified in [I-D.ietf-tsvwg-udp-options]. (If the OAL packet is submitted for jumbo-in-jumbo encapsulation, the L2 source instead includes a Hop-by-Hop Options header with a Parcel Payload Option with Advanced Jumbo Type 0 following the L2 UDP/IP header with the length of the L2 UDP header included in the Jumbo Payload Length.) The L2 source then sets the IP {Protocol, Next Header} to '17' (the UDP protocol number) and sets the {Total, Payload} Length the same as specified in the base IP protocol specifications for IP parcels and Advanced Jumbos (see: [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2]) or for ordinary IP packets (see: [RFC0791], [RFC8200] and [I-D.ietf-tsvwg-udp-options]). The L2 source then continues to set the remaining IP header fields as discussed below.¶
For raw IP encapsulation, the L2 source sets the IP {Protocol, Next Header} to TBD1 (see: IANA Considerations) and sets the {Total, Payload} Length the same as specified in [RFC0791] or [RFC8200]. (If the OAL header includes a Parcel Payload Option with an Advanced Jumbo Type, the L2 source includes an Parcel Payload Option with AJ Type 0 in the L2 IP header.) The L2 source then continues to set the remaining IP header fields as discussed below.¶
For IPsec AH/ESP encapsulation, the L2 source sets the appropriate IP or UDP header to indicate AH/ESP then sets the AH/ESP Next Header field to TBD1 the same as for raw IP encapsulation.¶
For direct encapsulations over Ethernet-compatible links, the L2 source prepares an Ethernet Header with EtherType set to TBD2 (see: Section 21.2) (see: Section 7).¶
For OAL packet/fragment encapsulations over secured underlay interface connections to the secured spanning tree, the L2 source applies any L2 security encapsulations according to the protocol (e.g., IPsec). These secured carrier packets are then subject to lower layer security services including fragmentation and reassembly.¶
When an L2 source includes a UDP header, it SHOULD calculate and include a UDP checksum in carrier packets with full OAL headers to prevent mis-delivery and/or detect IPv4 reassembly corruption; the L2 source MAY set UDP checksum to 0 (disabled) in carrier packets with compressed OAL headers (see: Section 6.5) or when reassembly corruption is not a concern. If the L2 source discovers that a path is dropping carrier packets with UDP checksums disabled, it should supply UDP checksums in future carrier packets sent to the same L2 destination. If the L2 source discovers that a path is dropping carrier packets that do not include a UDP header, it should include a UDP header in future carrier packets.¶
When an L2 source sends carrier packets with compressed OAL headers and with UDP checksums disabled, mis-delivery due to corruption of the AERO Forwarding Vector Index (AFVI) is possible but unlikely since the corrupted index would somehow have to match valid state in the (sparsely-populated) AERO Forwarding Information Base (AFIB). In the unlikely event that a match occurs, an OAL destination may receive carrier packets that contain a mis-delivered OAL fragment but can immediately reject any with incorrect Identifications. If the Identification value is somehow accepted, the OAL destination may submit the mis-delivered OAL fragment to the reassembly cache where it will most likely be rejected due to incorrect reassembly parameters. If a reassembly that includes the mis-delivered OAL fragment somehow succeeds (or, for atomic fragments) the OAL destination will verify any included checksums to detect corruption. Finally, any spurious data that somehow eludes all prior checks will be detected and rejected by end-to-end upper layer integrity checks. See: [RFC6935] [RFC6936] for further discussion.¶
For UDP/IP or IP-only L2 encapsulations, when the L2 source is also the OAL source it next copies the "Type of Service/Traffic Class" [RFC2983] and "Explicit Congestion Notification (ECN)" [RFC3168] values in the OAL header into the corresponding fields in the L2 IP header, then (for IPv6) set the L2 IPv6 header "Flow Label" as specified in [RFC6438]. The L2 source then sets the L2 IP TTL/Hop Limit the same as for any host (i.e., it does not copy the Hop Limit value from the OAL header) and finally sets the source and destination IP addresses to direct the carrier packet to the next OAL hop. For carrier packets subject to re-encapsulation, the OAL intermediate system as the L2 source reassembles if necessary then removes the L2 header(s). The L2 source then decrements the OAL header Hop Limit and discards the OAL packet/fragment if the value reaches 0. The L2 source then copies the Type of Service/Traffic Class and ECN values from the previous segment L2 encapsulation header into the next segment L2 encapsulation header while setting the next segment L2 source and destination IP addresses the same as above. (The L2 source also writes the ECN value into the OAL full/compressed header.)¶
The L2 source then applies source fragmentation if necessary by inserting an IPv6 Fragment Header between the L2 headers and the (compressed) OAL header then applying IP fragmentation per [RFC8200] or [I-D.herbert-ipv4-eh] to produce carrier packet fragments no larger than the current Carrier Fragment Size (CFS). (Note that the OMNI protocol L2 headers appear in each fragment and the Fragment Header Next Header field is adjusted as described in Section 6.4 following fragmentation.) The L2 source should prepare carrier packet fragments no larger than 1280 octets (i.e., the IPv6 minimum MTU) until it can determine whether a larger CFS is possible, e.g., through dynamic path probing to the L2 destination. For IPv4, until a probed CFS is determined the L2 source must set DF to 0 and include ancillary integrity checks (see below); these IPv4 carrier packet fragments may be (further) fragmented by intermediate systems in the L2 network.¶
For UDP/IPv4 carrier packets/fragments that set DF to 0, the L2 source calculates the UDP checksum and also includes a trailing 2-octet IPv4 reassembly checksum as specified in Appendix A. The L2 source calculates the checksums simultaneously in a single pass over the UDP pseudo-header plus the remainder of the packet following the header, then writes the UDP result in the UDP header and the IPv4 fragmentation result as the final 2 octets of the packet while incrementing the IPv4 length by 2. For raw IPv4 carrier packet (re-)encapsulation with DF set to 0, the source instead includes a trailing 2-octet IPv4 payload checksum followed by a 2-octet IPv4 reassembly checksum (calculated as above) while incrementing the IPv4 length by 4. The source calculates the IPv4 payload checksum the same as specified for UDP checksums [RFC0768], except that instead of the UDP length the pseudo header includes the length of the IPv4 payload only without including the IPv4 header or trailing checksum lengths. The source calculates the IPv4 payload and reassembly checksums simultaneously in a single pass over the pseudo header plus IPv4 payload the same as for the UDP case without extending to cover the trailing checksum fields themselves. (In both the UDP/IPv4 and raw IPv4 cases, the trailing checksum lengths will not cause the carrier packet to exceed 65535 octets since each OAL fragment reserves space for up to 256 L2 encapsulation octets.)¶
The L2 source then sends the resulting carrier packet fragments over one or more underlay interfaces. Underlay interfaces often connect directly to physical media on the local platform (e.g., an aircraft with a radio frequency link, a laptop computer with WiFi, etc.), but in some configurations the physical media may be hosted on a separate Local Area Network (LAN) node. In that case, the OMNI interface can establish a Layer-2 VLAN or a point-to-point tunnel (at a layer below the underlay interface) to the node hosting the physical media. The OMNI interface may also apply encapsulation at the underlay interface layer (e.g., as for a tunnel virtual interface) such that carrier packets would appear "double-encapsulated" on the LAN; the node hosting the physical media in turn removes the LAN encapsulation prior to transmission or inserts it following reception. Finally, the underlay interface must monitor the node hosting the physical media (e.g., through periodic keepalives) so that it can convey up-to-date Interface Attribute information to the OMNI interface.¶
Note: UDP/IPv4 and IPv4 L2 encapsulations that use IPsec AH/ESP do not include payload or reassembly integrity checks since the security encapsulations already include strong integrity checks.¶
Note: the L2 source must include a suitable Identification value in the IPv6 Fragment Header when it performs source fragmentation and must also include a suitable Identification value in the IPv4 header when it sets DF=0.¶
For paths that cannot rely on network fragmentation to deliver carrier packets that exceed the path MTU, the L2 source should actively probe the path to determine the largest possible Carrier Fragment Size (CFS) for the L2 destination under current path conditions. The L2 source conducts probing in the spirit of "Packetization Layer Path MTU Discovery for Datagram Transports" [RFC8899] using a probe packet such as an NS message that includes Nonce and Timestamp options [RFC3971] plus a discard trailing packet attachment as specified in Section 6.10. The L2 source then encapsulates the message in L2 headers as a whole carrier packet and sends the message over the unsecured underlay interface (for IPv4, the L2 source also sets the probe packet DF flag to 1.)¶
Prior to any probing, the L2 source assumes a nominal CFS of 1280 octets (the IPv6 minimum MTU) for both IPv6 and IPv4. Since this size is greater than the IPv4 minimum MTU, the L2 source must set the DF bit to 0 in each carrier packet to increase the likelihood that it will reach the L2 destination. When the L2 source sets DF to 0, it must include IPv4 payload/reassembly checksum(s) as discussed above.¶
When the L2 source engages probing, it will receive NA responses from the L2 destination to confirm delivery of its OAL and L2 encapsulated padded NS messages. When the L2 source receives an NA with a matching Nonce, it can then advance CFS to the size of the NS probe. The L2 source must then continuously probe to confirm the current CFS or advance to even larger CFS values using the probing strategies specified in [RFC8899].¶
After the L2 source confirms a CFS through probing, it can send carrier packet fragments up to CFS octets in length and with DF set to 1 for IPv4. If the path changes, the L2 source may receive a PTB message from a router on the path and should then reduce and/or re-probe the CFS accordingly.¶
All OAL intermediate systems and destinations MUST configure an L2 EMTU_R of 65535 octets on all unsecured underlay interfaces to enable successful reassembly of fragmented carrier packets no larger than that size (conversely, secured underlay interfaces use an EMTU_R specific to the L2 security service such as IPsec). OAL nodes are permitted to accept still larger unfragmented parcels/AJs as a best-effort service. OAL nodes must further recognize and honor the extended Identifications included in the IPv6 Extended Fragment Header [I-D.templin-6man-ipid-ext2].¶
When an OAL node reassembles an IPv4 or IPv6 carrier packet, it accepts the reassembled packet following L2 checksum verification if necessary. When an OAL node reassembles an IPv4 carrier packet with DF set to 0, it must verify both the UDP or IPv4 payload checksum and the IPv4 reassembly checksum. The OAL node then accepts the reassembled packet only if the included checksums are correct, then trims the trailing payload/reassembly checksum(s) by decrementing the IPv4 length before processing the packet further. When an OAL node detects a checksum error or failed reassembly for either IPv4 or IPv6 carrier packets, and the IP first fragment includes enough of the OAL packet header, the OAL node returns a uNA message with an OMNI Fragmentation Report (FRAGREP) option to the OAL source as specified in Section 6.8.¶
If the carrier packet encodes OMNI L2 extension headers per Section 6.4, the OAL node instead removes the UDP header if necessary and submits the packet for IPv6 extension header processing per [RFC8200] (while converting IPv4/Ethernet headers to IPv6 and converting IPv4/EUI addresses to IPv6 compatible addresses if necessary as specified above). The OAL node first sets the IPv6 Next Header field to the 8 bit protocol value for the first extension. When an (Extended) Fragment Header is included, the OAL node performs L2 reassembly per the IPv6 extension header parameters.¶
When an OMNI interface processes a (reassembled) carrier packet from an underlay interface, it copies the ECN value from the L2 encapsulation headers into the OAL header if the carrier packet contains an OAL first-fragment. The OMNI interface next discards the L2 encapsulation headers and examines the OAL header of the enclosed OAL fragment according to the value in the Type field as discussed in Section 6.2. If the OAL fragment is addressed to a different node, the OMNI interface (acting as an OAL intermediate system) performs L2 encapsulation and fragmentation if necessary then forwards while decrementing the OAL Hop Limit as discussed in Section 6.2. If the OAL fragment is addressed to itself, the OMNI interface (acting as an OAL destination) accepts or drops the fragment based on the (Source, Destination, Identification)-tuple.¶
The OAL destination next drops all ordinal OAL non-first fragments that would overlap or leave "holes" with respect to other ordinal fragments already received. The OAL destination updates a checklist of accepted ordinal fragments of the same OAL packet but admits all accepted fragments into the reassembly cache.¶
During reassembly at the OAL destination, the reassembled OAL packet may exceed 65535 by a small amount equal to the size of the OAL encapsulation extension headers. The OAL destination does not write this (too-large) value into the OAL header Payload Length field, but rather remembers the value during reassembly. When reassembly is complete, the OAL destination finally removes the OAL headers. The OAL destination then delivers the original IP packet/parcel to the network layer. The original IP packet/parcel may therefore be as large as 65535 octets, or larger still for large parcels/AJs delivered through jumbo-in-jumbo encapsulation without invoking fragmentation.¶
When an OAL path traverses an IPv6 network with routers that perform adaptation layer forwarding based on full IPv6 headers with OAL addresses, the OAL intermediate system at the head of the IPv6 path forwards the OAL packet/fragment the same as an ordinary IPv6 packet without decapsulating and delivering to the network layer. Once within the IPv6 network, these OAL packets/fragments may traverse arbitrarily-many IPv6 hops before arriving at an OAL intermediate system which may again encapsulate the OAL packets/fragments as carrier packets for transmission over underlay interfaces.¶
Note: carrier packets often traverse paths with underlying links that use integrity checks such as CRC-32 which provide adequate hop-by-hop integrity assurance for payloads up to ~9K octets [CRC]. However, other paths may traverse links (such as fragmenting tunnels over IPv4 - see: [RFC4963]) that do not include adequate checks. The end-to-end integrity checks in IP parcels and AJs therefore allow the final destination to detect any link errors that may have accumulated along the path even if the links themselves do not provide adequate error checking.¶
The IPv6 specification [RFC8200] defines extension headers that follow the base IPv6 header, while Upper Layer Protocols (ULPs) are specified in other documents. Each extension header present is identified by a "Next Header" octet in the previous (extension) header and encodes a "Next Header" field in the first octet that identifies the next extension header or ULP instance. The OMNI specification supports encoding of IPv6 extension header chains immediately following the OMNI L2 UDP, IP or Ethernet header even if the L2 IP protocol version is IPv4. In all cases, the length of the IPv6 extension header chain is limited by [I-D.ietf-6man-eh-limits].¶
The OAL source prepares an OMNI extension header chain by setting the first 4 bits of the first IPv6 extension header in the chain to a Type value for the extension header itself immediately following the OMNI L2 protocol header. The source then sets the next 4 bits to a Next value that identifies either a terminating ULP or the next extension header in the chain. The source then sets the first 8 bits of each subsequent IPv6 extension header in the chain to the standard Next Header encoding as shown in Figure 6:¶
The following Type/Next values are currently defined:¶
0 (OMNI-RES) - Reserved for experimentation.¶
1 (OMNI-OCH1) - OMNI Compressed Header, Type 1 per Section 6.5.¶
2 (OMNI-OCH2) - OMNI Compressed Header, Type 2 per Section 6.5.¶
3 (OMNI-OFH) - OMNI Full Header, per Section 6.5.¶
5 (OMNI-HBH) - Hop-by-Hop Options per Section 4.3 of [RFC8200].¶
8 (OMNI-FH) - Fragment Header per Section 4.5 of [RFC8200].¶
9 (OMNI-DO) - Destination Options per Section 4.6 of [RFC8200].¶
11 (OMNI-ESP) - Encapsulating Security Payload per [RFC4303].¶
12 (OMNI-NNH) - No Next Header per Section 4.7 of [RFC8200].¶
15 (OMNI-ULP) - Upper Layer Protocol shim (see below).¶
Entries OMNI-OCH1 through OMNI-AH in the above list follow the convention that the OMNI Type/Version appears in the first 4 bits of the extension header (or IP header) itself. Conversely, entries OMNI-ESP through OMNI-UDP represent commonly-used ULPs which do not encode a Type/Version in the first 4 bits.¶
Entries OMNI-HBH, OMNI-RH, OMNI-FH, OMNI-DO and OMNI-AH represent true IPv6 extension headers encoded for OMNI, which may be chained. Source and destination processing of OMNI extension headers follows exactly per their definitions in the normative references, with the exception of the special (Type, Next) coding in the first 8 bits of the first extension header.¶
When a ULP not found in the above table immediately follows the OMNI L2 UDP, IP or Ethernet header, the source includes a 2-octet "Type 1 ULP Shim" before the ULP where both the first 4 bit (Type) and next 4 bit (Next) fields encode the special value 15 (OMNI-ULP). The source then includes a Next Header field that encodes the IP protocol number of the ULP. The source then includes the ULP data immediately after the shim as shown in Figure 7.¶
When a ULP "OMNI-(N)" found in the above table immediately follows the OMNI L2 UDP, IP or Ethernet header, the source includes a 1-octet "Type 2 ULP Shim" before the ULP where the first 4 bits encode the special Type value 15 (OMNI-ULP) and the next 4 bits encode the Next ULP type "N" taken from the table above. The source then includes the ULP data immediately after the shim as shown in Figure 8.¶
When a ULP not found in the above table follows a first OMNI extension header, the source sets the extension header Next field to OMNI-ULP (15) and includes a 1-octet "Type 3 ULP Shim" that encodes the IP protocol number for the Next Header of the ULP data that follows as shown in Figure 9.¶
When a ULP "OMNI-(N)" found in the above table follows a first OMNI extension header, the source sets the extension header Next field to the ULP Type "N" and does not include a shim. The ULP then begins immediately after the first OMNI extension header.¶
When a ULP of any kind follows a non-first OMNI extension header, the source sets the extension header Next Header field to the IP protocol number for the ULP and does not include a shim. The ULP then begins immediately after the non-first OMNI extension header.¶
Note: The L2 UDP header (when present) is logically considered as the first L2 extension header in the chain. If an Advanced Jumbo extension header is also present, its Jumbo Payload length includes the length of the L2 UDP header.¶
Note: After a node parses the extension header chain, it changes the "Type/Next" field in the first extension header back to the correct "Next Header" value before processing the first extension header.¶
OAL sources that send OAL packets with OMNI Full Headers (OFH) include a Compressed Routing Header (CRH) [I-D.ietf-6man-comp-rtg-hdr] and IPv6 Extended Fragment Header extensions for segment-by-segment forwarding based on an AERO Forwarding Information Base (AFIB) in each OAL intermediate system. OAL sources, intermediate systems and destinations establish header compression state in the AFIB through IPv6 ND NS/NA message exchanges. After an initial NS/NA exchange, OAL nodes can apply OMNI Header Compression to significantly reduce header overhead.¶
OAL nodes apply header compression in order to avoid transmission of redundant data found in the original IP packet and OAL encapsulation headers; the resulting compressed headers are often significantly smaller than the original IP packet header itself even when OAL encapsulation is applied. Header compression is limited to the OAL IPv6 encapsulation header plus extensions along with the base original IP packet header; it does not extend to include any extension headers of the original IP packet which appear as upper layer payload immediately following the compressed headers.¶
Each OAL node establishes AFIB soft state entries known as AERO Forwarding Vectors (AFVs) which support both OAL packet/fragment forwarding and OAL/IPv6 header compression/decompression. The FHS OAL sources references each AFV by an AERO Forwarding Vector Index (AFVI) which in conjunction with the previous hop L2ADDR provides compression/decompression and next hop forwarding context.¶
When an OAL node sends carrier packets that contain OAL packets/fragments to a next hop, it includes an OFH with a CRH containing AFVI forwarding information followed by an Extended Fragment Header. If the OAL source applied OAL encapsulation, the first 4 bits following the L2 headers must encode the Type OMNI-OFH to signify that an uncompressed OFH (plus extensions) is present; otherwise, the first 4 bits must encode the value OMNI-IP6 as a Type/Version value for IPv6. The CRH include a single 32-bit AFVI (as CRH-32) and with Segments Left set to 1.¶
When an OAL intermediate system forwards an OAL packet, it determines the AFVI for the next OAL hop by using the AFVI included in the CRH to search for a matching AFV. The OAL intermediate system then writes the next hop AFVI into the CRH and forwards the OAL packet to the next hop without decrementing Segments Left. This same AFVI re-writing progression begins with the OAL source then continues over all OAL intermediate nodes and finally ends at the OAL destination.¶
When AFV state is available, the OAL source should omit significant portions of the OAL header (plus extensions) and the entire original IP packet header by applying OMNI header compression. For OAL first fragments (including atomic fragments), the OAL source uses OMNI Compressed Header, Type 1 (OCH1) Format (a) as shown in Figure 10:¶
The format begins with a 4-bit Type followed by the 8-bit Traffic Class (copied into the OAL header from the original IP packet header) followed by an 8-bit (OAL) Hop Limit followed by followed by a 6-bit Parcel ID with 2 P/S control bits followed by 4 flag bits. The header next includes the 4 least significant octets of the OAL Identification followed by a 2/4-octet AFVI according to whether the A flag is set to 0/1, respectively. The format then includes a 2-octet Payload Length only if the L2 header does not include a length field. The format finally includes the Next Header and Hop Limit values from the original (L3) IP packet header, plus a 2-octet Header Checksum only for IPv4 original packets. (Note that these values represent compression of the original IP packet header plus the OFH header along with its CRH-32 and Extended Fragment Header in a unified concatenation.)¶
The OAL node sets Type to OMNI-OCH1, sets Hop Limit to the uncompressed OAL header Hop Limit and sets the ECN bits in the Traffic Class field the same as for an uncompressed IP header. The OAL node next sets (F)irst to 1 as a first fragment then sets (M)ore Fragments, Parcel ID, (P)arcel, and More (S)egments the same as for an uncompressed Extended Fragment Header. The OAL node finally sets the L3 Next Header and Hop Limit fields to the values that would appear in the uncompressed original IP header; the OAL node also includes a 2-octet Header Checksum for IPv4 original packets, or omits the Header Checksum for IPv6 original packets.¶
The payload of the OAL first fragment (i.e., beginning after the original IP header) is then included immediately following the OCH1 header, and the L2 header length field (if present) is reduced by the difference in length between the compressed and full-length headers. If the L2 header includes a length field, the OAL destination can determine the payload length by examining the L2 header; otherwise, the OCH1 header itself includes a 2-octet Payload Length field that encodes the length of the packet payload (or first fragment) that follows the OCH1. Note that first fragments (and atomic packets) are logically considered ordinal fragment 0 even though no ordinal value is transmitted.¶
When the OAL source has multiple original atomic IP packets enqueued that would include identical original IP headers (except for the Payload Length), it can set the (Q)ueued flag and perform "compressed packing" (see: Section 6.10). When the Q flag is set, the M flag MUST be 0, meaning that the payload MUST NOT extend beyond the first fragment. The Payload Length field MUST be included, but encodes the length of the first queued packet payload only. The OCH1 header is then followed by the payload of the first queued packet (i.e., with the IP header removed) which is followed by a second Payload Length field that encodes the length of the second queued packet payload. The second Payload Length is then followed by the payload of the second queued packet which is followed by a third Payload Length (and possibly also a third packet payload), etc., until a final Payload Length field that encodes the value 0 appears. When the OAL destination receives an OCH1 OAL packet with the Q flag set, it extracts each packet payload (while appending the original IP header with only the Payload Length values differing) by following the chain of Payload Length fields present.¶
For OAL non-first fragments (i.e., those with non-zero Index), the OAL uses OMNI Compressed Header, Type 1 (OCH1) Format (b) as shown in Figure 11:¶
The format begins with a 4-bit Type followed by an 8-bit Traffic Class followed by an 8-bit OAL Hop Limit the same as for first fragments. The format next includes a 6-bit ordinal fragment Index followed by a (F)irst flag, an (A)FVI extension flag and finally a (M)ore Fragments flag. The format next includes the least-significant 4 octets of the OAL Identification followed by a 2/4-octet AFVI according to the A flag followed by a 0/2-octet Payload Length field the same as for an OCH1 first fragment.¶
The OAL node sets Type to OMNI-OCH1, sets Hop Limit to the uncompressed OAL header Hop Limit value, and sets (Index, (F)irst, (M)ore Fragments, Identification) to their appropriate values as a non-first fragment. In particular, the OAL Node sets Index to a monotonically increasing ordinal value beginning with 1 for the first non-first fragment, 2 for the second non-first fragment, 3 for the third non-first fragment, etc., up to at most 63 for the final fragment.¶
The OAL non-first fragment body is then included immediately following the OCH1 header, and the L2 header length field (if present) is reduced by the difference in length between the compressed headers and full-length original IP header with OFH plus extensions. The OAL destination will then be able to determine the Payload Length by examining the L2 header length field if present; otherwise by examining the 2-octet OCH1 Payload Length the same as for first fragments.¶
The OCH1 Format (a) is used for all original IPv6 packets that do not include a Fragment Header as well as for original IPv4 packets that set IHL to 5, DF to 1 and (MF; Fragment Offset) to 0 (the OCH1 Format (b) is used for all non-first fragments regardless of the original IP version). For other "non-atomic" original IP packets and first fragments, the OAL uses the "Type 2" OMNI Compressed Header (OCH2) formats shown in Figure 12 and Figure 13:¶
In both of the above OCH2 formats, the leading octets include the same information that would appear in a corresponding OCH1 header with the exception that the (Q, F) flags are replaced by a 2-bit Reserved field. The remainder of the OCH2 format (a) includes fields that would appear in an uncompressed IPv6 header plus Fragment Header extension per [RFC8200], while the remainder of format (b) includes fields that would appear in an uncompressed IPv4 header per [RFC0791] with the Options and Padding lengths calculated based on IHL. In both cases, the Source and Destination addresses are not transmitted. (Note that packing is not supported with the OCH2 format since each non-atomic IP packet header will often include different values.)¶
When an OAL destination or intermediate system receives a carrier packet, it determines the length of the encapsulated OAL information and verifies that the innermost L2 next header field indicates OMNI (see: Section 6.2), then processes any included OMNI L2 extension headers as specified in Section 6.4. The OAL destination then examines the Next Header field of the final L2 extension header. If the Next Header field contains the value TBD1, and the 4-bit Type that follows encodes a value OMNI-IP6, OMNI-OFH, OMNI-OCH1 or OMNI-OCH2 the OAL node processes the remainder of the OAL header as a full or compressed header as specified above.¶
The OAL node then uses the AFVI to locate the cached AFV which determines the next hop. During forwarding for compressed headers, the OAL node changes the OCH AFVI to the cached value for the AFV next hop. If the OAL node is the destination, it instead reconstructs the OFH and original IP headers based on the information cached in the AFV combined with the received information in the OCH1/2. For non-atomic fragments, the OAL node then adds the resulting OAL fragment to the reassembly cache if the Identification is acceptable. Following OAL reassembly if necessary, the OAL node delivers the original IP packet to the network layer.¶
For all OCH1/2 types, the source node sets all Reserved fields and bits to 0 on transmission and the destination node ignores the values on reception. For both OCH1/2, ECN information is compiled for first fragments, and not for non-first fragments.¶
Finally, if an IPv6 Hop-by-Hop (HBH) and/or Routing Header extension header is required to appear as per-fragment extensions with each OAL fragment that uses OCH1 format (b) or OCH2 compression the OAL node inserts an OMNI-HBH and/or OMNI-RH header as the first extension(s) following the L2 header and before the OMNI-OCH1/2 as discussed in Section 6.4.¶
When the OAL node is unable to determine whether the next OAL hop is connected to the same underlay link, it should perform carrier packet L2 encapsulation for initial packets sent via the next hop over a specific underlay interface by including full UDP/IP headers and with the UDP port numbers set as discussed in Section 6.2. The node can thereafter attempt to send an NS to the next OAL hop in carrier packet(s) that omit the UDP header and set the IP protocol number to TBD1. If the OAL node receives an NA reply, it can omit the UDP header in subsequent packets. The node can further attempt to send an NS in carrier packet(s) that omit both the UDP and IP headers and set EtherType to TBD2. If the source receives an NA reply, it can begin omitting both the UDP and IP headers in subsequent packets.¶
Note: in the above, "next OAL hop" refers to the first OAL node encountered on the optimized path to the destination over a specific underlay interface as determined through route optimization (e.g., see: [I-D.templin-6man-aero3]). The next OAL hop could be a Proxy/Server, Gateway or the OAL destination itself.¶
The OAL encapsulates each original IP packet/parcel as an OAL packet then performs fragmentation to produce one or more carrier packets with the same 8-octet Identification value. In environments where spoofing is not considered a threat, OMNI interfaces send OAL packets with Identifications beginning with an unpredictable Initial Send Sequence (ISS) value [RFC7739] monotonically incremented (modulo 2**64) for each successive OAL packet sent to either a specific neighbor or to any neighbor. (The OMNI interface may later change to a new unpredictable ISS value as long as the Identifications are assured unique within a timeframe that would prevent the fragments of a first OAL packet from becoming associated with the reassembly of a second OAL packet.) In other environments, OMNI interfaces should maintain explicit per-flow send and receive windows to detect and exclude spurious carrier packets that might clutter the reassembly cache as discussed below.¶
OMNI interface neighbors use a window synchronization service similar to TCP [RFC9293] to maintain unpredictable ISS values incremented (modulo 2**64) for each successive OAL packet and re-negotiate windows often enough to maintain an unpredictable profile. OMNI interface neighbors exchange IPv6 ND messages that include OMNI Multilink Vector sub-options (see: Section 10.2.8) that include TCP-like information fields and flags to manage streams of OAL packets instead of streams of octets. As a link layer service, the OAL provides low-persistence best-effort retransmission with no mitigations for duplication, reordering or deterministic delivery. Since the service model is best-effort and only control message sequence numbers are acknowledged, OAL nodes can select unpredictable new initial sequence numbers outside of the current window without delaying for the Maximum Segment Lifetime (MSL).¶
OMNI interface end neighbors and intermediate systems maintain current and previous per-flow window state in IPv6 ND NCEs and/or AFVs to support dynamic rollover to a new window while still sending OAL packets and accepting carrier packets from the previous windows. OMNI interface neighbors synchronize windows through asymmetric and/or symmetric IPv6 ND message exchanges. When OMNI end and intermediate systems receive an IPv6 ND message with new per-flow window information, it resets the previous window state based on the current window then resets the current window based on new and/or pending information.¶
The IPv6 ND message OMNI option Multilink Vector sub-option includes TCP-like information fields including Sequence Number, Acknowledgement Number, Window and flags (see: Section 10). OMNI interface neighbors and intermediate systems maintain the following TCP-like state variables on a per-interface-pair basis (i.e., through a combination of NCE and/or AFV state):¶
Send Sequence Variables (current, previous and pending) SND.NXT - send next SND.WND - send window ISS - initial send sequence number Receive Sequence Variables (current and previous) RCV.NXT - receive next RCV.WND - receive window IRS - initial receive sequence number¶
OMNI interface neighbors "OAL A" and "OAL B" exchange IPv6 ND messages per [RFC4861] with OMNI options that include TCP-like information fields in a Multilink Vector. When OAL A synchronizes with OAL B, it maintains both a current and previous SND.WND beginning with a new unpredictable ISS and monotonically increments SND.NXT for each successive OAL packet transmission. OAL A initiates synchronization by including the new ISS in the Sequence Number of an authentic IPv6 ND message with the SYN flag set and with Window set to M (up to 2**24) as its advertised send window size while creating a NCE in the INCOMPLETE state if necessary. OAL A caches the new ISS as pending, uses the new ISS as the Identification for OAL encapsulation, then sends the resulting OAL packet to OAL B and waits up to RetransTimer milliseconds to receive an IPv6 ND message response with the ACK flag set (retransmitting up to MAX_UNICAST_SOLICIT times if necessary).¶
When OAL B receives the SYN, it creates a NCE in the STALE state and also an AFV if necessary, resets its RCV variables and caches the source's send window size M as its receive window size. OAL B then prepares an IPv6 ND message with the ACK flag set, with the Acknowledgement Number set to OAL A's next sequence number, and with Window set to M. Since OAL B does not assert an ISS of its own, it uses the IRS it has cached for OAL A as the Identification for OAL encapsulation then sends the ACK to OAL A.¶
When OAL A receives the ACK, it notes that the Identification in the OAL header matches its pending ISS. OAL A then sets the NCE state to REACHABLE and resets its SND variables based on the Window size and Acknowledgement Number (which must include the sequence number following the pending ISS). OAL A can then begin sending OAL packets to OAL B with Identification values within the (new) current SND.WND for this interface pair for up to ReachableTime milliseconds or until the NCE is updated by a new IPv6 ND message exchange. This implies that OAL A must send a new SYN before sending more than N OAL packets within the current SND.WND, i.e., even if ReachableTime is not nearing expiration. After OAL B returns the ACK, it accepts carrier packets received from OAL A via this interface pair within either the current or previous RCV.WND as well as any new authentic NS/RS SYN messages received from OAL A even if outside the windows.¶
OMNI interface neighbors can employ asymmetric window synchronization as described above using 2 independent (SYN -> ACK) exchanges (i.e., a 4-message exchange), or they can employ symmetric window synchronization using a modified version of the TCP "3-way handshake" as follows:¶
OAL A prepares a SYN with an unpredictable ISS not within the current SND.WND and with Window set to M as its advertised send window size. OAL A caches the new ISS and Window size as pending information, uses the pending ISS as the Identification for OAL encapsulation, then sends the resulting OAL packet to OAL B and waits up to RetransTimer milliseconds to receive an ACK response (retransmitting up to MAX_UNICAST_SOLICIT times if necessary).¶
OAL B receives the SYN, then resets its RCV variables based on the Sequence Number while caching OAL A's send window size M as its receive window size. OAL B then selects a new unpredictable ISS outside of its current window, then prepares a response with Sequence Number set to the pending ISS and Acknowledgement Number set to OAL A's next sequence number. OAL B then sets both the SYN and ACK flags, sets Window to a chosen send window size N and sets the OPT flag according to whether an explicit concluding ACK is optional or mandatory. OAL B then uses the pending ISS as the Identification for OAL encapsulation, sends the resulting OAL packet to OAL A and waits up to RetransTimer milliseconds to receive an acknowledgement (retransmitting up to MAX_UNICAST_SOLICIT times if necessary).¶
OAL A receives the SYN/ACK, then resets its SND variables based on the Acknowledgement Number (which must include the sequence number following the pending ISS). OAL A then resets its RCV variables based on the Sequence Number and OAL B's advertised send Window N and marks the NCE as REACHABLE. If the OPT flag is clear, OAL A next prepares an immediate unsolicited NA message with the ACK flag set, the Acknowledgement Number set to OAL B's next sequence number, with Window set to N, and with the OAL encapsulation Identification to SND.NXT, then sends the resulting OAL packet to OAL B. If the OPT flag is set and OAL A has OAL packets queued to send to OAL B, it can optionally begin sending their carrier packets under the current SND.WND as implicit acknowledgements instead of returning an explicit ACK.¶
OAL B receives the implicit/explicit acknowledgement(s) then resets its SND state based on the pending/advertised values and marks the NCE as REACHABLE. Note that OAL B sets the OPT flag in the SYN/ACK to assert that it will interpret timely receipt of carrier packets within the (new) current window as an implicit acknowledgement. Potential benefits include reduced delays and control message overhead, but use case analysis is outside the scope of this specification.)¶
Following synchronization, OAL A and OAL B hold updated NCEs and AFVs, and can exchange OAL packets with Identifications set to SND.NXT for each flow while the state remains REACHABLE and there is available window capacity. (Intermediate systems that establish AFVs for the per-flow window synchronization exchanges can also use the Identification window for source validation.) Either neighbor may at any time send a new SYN to assert a new ISS. For example, if OAL A's current SND.WND for OAL B is nearing exhaustion and/or ReachableTime is nearing expiration, OAL A can continue sending OAL packets under the current SND.WND while also sending a SYN with a new unpredictable ISS. When OAL B receives the SYN, it resets its RCV variables and may optionally return either an asymmetric ACK or a symmetric SYN/ACK to also assert a new ISS. While sending SYNs, both neighbors continue to send OAL packets with Identifications set to the current SND.NXT for each interface pair then reset the SND variables after an acknowledgement is received.¶
While the optimal symmetric exchange is efficient, anomalous conditions such as receipt of old duplicate SYNs can cause confusion for the algorithm as discussed in Section 3.5 of [RFC9293]. For this reason, the OMNI Multilink Vector sub-option includes an RST flag which OAL nodes set in solicited NA responses to ACKs received with incorrect acknowledgement numbers. The RST procedures (and subsequent synchronization recovery) are conducted exactly as specified in [RFC9293].¶
OMNI interfaces that employ the window synchronization procedures described above observe the following requirements:¶
OMNI interfaces MUST select new unpredictable ISS values that are at least a full window outside of the current SND.WND.¶
OMNI interfaces MUST set the Window field in SYN messages as a non-negotiable advertised send window size.¶
OMNI interfaces MUST send IPv6 ND messages used for window synchronization securely while using unpredictable initial Identification values until synchronization is complete.¶
It is essential to understand that the above window synchronization operations between nodes OAL(A) and OAL(B) are conducted in IPv6 ND message exchanges over multihop paths with potentially many OAL(i) intermediate hops in the forward and reverse paths (which may be disjoint). Each such forward path OAL(i) caches the sequence number and window size advertised from OAL(A) to OAL(B) in its AFV entry indexed by the previous hop L2ADDR and AFVI, while each such reverse path OAL(i) caches the sequence number, window size and AFVI advertised from OAL(B) to OAL(A). (The forward/reverse path OAL(i) nodes then select new unique next-hop AFVIs before forwarding.)¶
Note: Although OMNI interfaces employ TCP-like window synchronization and support uNA ACK responses to SYNs, all other aspects of the IPv6 ND protocol (e.g., control message exchanges, NCE state management, timers, retransmission limits, etc.) are honored exactly per [RFC4861]. OMNI interfaces further manage per-interface-pair window synchronization parameters in one or more AFVs for each neighbor pair.¶
Note: Recipients of OAL-encapsulated IPv6 ND messages index the NCE based on the message source address, which also determines the carrier packet Identification window. However, IPv6 ND messages may contain a message source address that does not match the OMNI encapsulation source address when the recipient acts as a proxy.¶
Note: OMNI interface neighbors apply separate send and receive windows for all of their (multilink) underlay interface pairs that exchange carrier packets. Each interface pair represents a distinct underlay network path, and the set of paths traversed may be highly diverse when multiple interface pairs are used. OMNI intermediate systems therefore become aware of each distinct set of interface pair window synchronization parameters based on periodic IPv6 ND message updates to their respective AFVs.¶
When the round-trip delay from the original source to the final destination is long while the route-trip time from the OAL source the OAL destination is significantly shorter, the OAL source can maintain a short-term cache of the OAL fragments it sends to OAL destinations in case timely best-effort selective retransmission is requested. The OAL destination in turn maintains a checklist for (Source, Destination, Identification)-tuples of recently received OAL fragments and notes the ordinal numbers of OAL fragments already received (i.e., as ordinals #0, #1, #2, #3, etc.). The timeframe for maintaining the OAL source and destination caches determines the link persistence (see: [RFC3366]).¶
If the OAL destination notices some fragments missing after most other fragments within the same link persistence timeframe have already arrived, it may issue an Automatic Repeat Request (ARQ) with Selective Repeat (SR) by sending a uNA message to the OAL source. The OAL destination creates a uNA message with an OMNI option with one or more Fragmentation Report (FRAGREP) sub-options that include (Identification, Bitmap)-tuples for fragments received and missing from this OAL source (see: Section 10). The OAL destination includes an authentication signature if necessary, performs OAL encapsulation (with its own address as the OAL source and the source address of the message that prompted the uNA as the OAL destination) and sends the message to the OAL source.¶
If an OAL intermediate system or OAL destination processes an OAL fragment for which corruption is detected, it may similarly issue an immediate ARQ/SR the same as described above. The FRAGREP provides an immediate (rather than time-bounded) indication to the OAL source that a fragment has been lost.¶
When the OAL source receives the uNA message, it authenticates the message then examines any enclosed FRAGREPs. For each (Source, Destination, Identification)-tuple, the OAL source determines whether it still holds the corresponding OAL fragments in its cache and retransmits any for which the Bitmap indicates a loss event. For example, if the Bitmap indicates that ordinal fragments #3, #7, #10 and #13 from the OAL packet with Identification 0x0123456789abcdef are missing the OAL source only retransmits those fragments. When the OAL destination receives the retransmitted OAL fragments, it admits them into the reassembly cache and updates its checklist. If some fragments are still missing, the OAL destination may send a small number of additional uNA ARQ/SRs within the link persistence timeframe.¶
The OAL therefore provides a link layer low-to-medium persistence ARQ/SR service consistent with [RFC3366] and Section 8.1 of [RFC3819]. The service provides the benefit of timely best-effort link layer retransmissions which may reduce OAL fragment loss and avoid some unnecessary end-to-end delays. This best-effort network-based service therefore compliments transport and higher layer end-to-end protocols responsible for true reliability.¶
When the OMNI interface forwards original IP packets/parcels from the network layer, it invokes the OAL and returns internally-generated Path MTU Discovery (PMTUD) ICMPv4 "Fragmentation Needed and Don't Fragment Set" [RFC1191] or ICMPv6 "Packet Too Big (PTB)" [RFC8201] messages as necessary. This document refers to both message types as "PTBs" and introduces a distinction between PTB "hard" and "soft" errors as discussed below.¶
Ordinary PTB messages are hard errors that always indicate loss due to a real MTU restriction has occurred. However, the OMNI interface can also forward original IP packets/parcels via OAL encapsulation and fragmentation while at the same time returning PTB soft error messages (subject to rate limiting) to the original source to suggest smaller sizes due to factors such as link performance characteristics, excessive number of fragments needed, reassembly congestion, etc.¶
This ensures that the path MTU is adaptive and reflects the current path used for a given data flow. The OMNI interface can therefore continuously forward original IP packets/parcels without loss while returning PTB soft error messages that recommend smaller sizes. Original sources that receive the soft errors in turn reduce the size of the original IP packets/parcels they send the same as for hard errors, but not necessarily due to a loss event. The original source can then resume sending larger packets/parcels if the soft errors subside.¶
OAL intermediate systems that experience fragment loss and OAL end systems that experience reassembly cache congestion can return uNA messages that include OMNI encapsulated PTB soft error messages to OAL sources that originate fragments (subject to rate limiting). The OAL node creates a secured uNA message with an OMNI option containing an ICMPv6 Error sub-option. The OAL node encodes a PTB message in the sub-option with MTU set to a reduced value and with the leading portion an OAL first fragment containing the header of an original IP packet/parcel for which the source must be notified (see: Section 10).¶
The OAL node that sends the uNA encapsulates the leading portion of the OAL first fragment (beginning with the OAL header) in the PTB "packet in error" field and signs the message if an authentication signature is included. The OAL node then performs OAL encapsulation (with its own address as the source and the source address of the message that prompted the uNA as the destination) and sends the message to the OAL source. (Note that OAL intermediate systems forward uNAs via the secured spanning tree while OAL destination end systems include an authentication signature when necessary.)¶
When the OAL source receives a uNA message from an OAL intermediate system, it can reduce its OFS estimate and begin sending smaller OAL fragments and/or reduce its CFS estimate and begin sending smaller carrier packet fragments. When the OAL source receives a uNA message from the OAL destination, it sends a corresponding network layer PTB soft error to the original source.¶
The OAL source prepares the PTB soft error by first setting the Type field to 2 for IPv6 [RFC4443] or TBD6 for IPv4 (see: IANA considerations). The OAL source then sets the Code field to "PTB Soft Error (no loss)" if the OAL destination forwarded the original IP packet/parcel successfully or "PTB Soft Error (loss)" if it was dropped (see: IANA considerations). The OAL source next sets the PTB destination address to the original IP packet/parcel source, and sets the source address to one of its OMNI interface addresses that is reachable from the perspective of the original source.¶
The OAL source then sets the MTU field to a value smaller than the original IP packet/parcel size but no smaller than 1280, writes as much of the original IP packet/parcel first fragment as possible into the "packet in error" field such that the entire PTB including the IP header is no larger than 1280 octets for IPv6 or 576 octets for IPv4. The OAL source then calculates and sets the ICMP Checksum and returns the PTB to the original source.¶
An original sources that receives these PTB soft errors first verifies that the ICMP Checksum is correct and the packet-in-error contains the leading portion of one of its recent packet/parcel transmissions. The original source can then adaptively tune the size of the original IP packets/parcels it sends to produce the best possible throughput and latency, with the understanding that these parameters may fluctuate over time due to factors such as congestion, mobility, network path changes, etc. Original sources should therefore consider receipt or absence of soft errors as hints of when decreasing or increasing packet/parcel sizes may provide better performance.¶
The OMNI interface supports continuous transmission and reception of packets/parcels of various sizes in the face of dynamically changing network conditions. Moreover, since PTB soft errors do not indicate a hard limit, original sources that receive soft errors can resume sending larger packets/parcels without waiting for the recommended 10 minutes specified for PTB hard errors [RFC1191][RFC8201]. The OMNI interface therefore provides an adaptive service that accommodates MTU diversity especially well-suited for air/land/sea/space mobile Internetworking.¶
The OMNI interface may also return PTB messages with Parcel Report and/or Jumbo Report Codes in response to parcels and/or AJs delivered by the network layer and forwarded through jumbo-in-jumbo encapsulation. These Parcel/Jumbo Report messages are prepared the same as for PTB soft errors discussed above. IP parcels and AJs are discussed in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2].¶
Note: when the OAL source receives persistent Fragmentation Reports for a given flow (see: Section 6.8), it should return PTB soft errors to the original source (subject to rate limiting) the same as if it had received PTB soft errors from the OAL destination. When the original source is likely to retransmit an entire original IP packet on its own behalf in case of loss, the OAL destination can elect to return only PTB soft errors and refrain from returning Fragmentation Reports.¶
Note: the OAL source may receive uNA messages that include both a PTB soft error and Fragmentation Report(s). If so, the OAL source both returns PTB soft errors to the original source (subject to rate limiting) and retransmits any missing fragments if it is configured to do so.¶
Note: OAL intermediate nodes that reassemble fragmented carrier packets should return PTB soft errors subject to rate limiting during periods of fragment loss and/or L2 reassembly cache congestion. The OAL previous hop should regard these PTB soft errors as an indication to reduce the current CFS for this L2 destination.¶
The OAL source ordinarily includes a 40-octet IPv6 encapsulation header for each original IP packet/parcel during OAL encapsulation. The OAL source then performs fragmentation such that a copy of the 40-octet IPv6 header plus a 16-octet IPv6 Extended Fragment Header is included in each OAL fragment (when a Routing Header is added, the OAL encapsulation headers become larger still). However, these encapsulations may represent excessive overhead in some environments.¶
OAL header compression as discussed in Section 6.5 can dramatically reduce encapsulation overhead, however a complimentary technique known as "packing" (see: [I-D.ietf-intarea-tunnels]) supports encapsulation of multiple original IP packets/parcels and/or control messages within a single OAL "super-packet".¶
When the OAL source has multiple original IP packets/parcels to send to the same OAL destination with total length no larger than the OAL destination EMTU_R, it can concatenate them into a super-packet encapsulated in a single OAL header. Within the OAL super-packet, the IP header of the first original IP packet/parcel (iHa) followed by its data (iDa) is concatenated immediately following the OAL header. The IP header of the next original packet/parcel (iHb) followed by its data (iDb) is then concatenated immediately following the first, with each remaining original IP packet/parcel concatenated in succession. The OAL super-packet format is transposed from [I-D.ietf-intarea-tunnels] and shown in Figure 14:¶
When the OAL source prepares a super-packet, it applies OAL fragmentation then applies L2 encapsulation/fragmentation and sends the resulting carrier packets to the OAL destination. When the OAL destination receives the super-packet it first reassembles if necessary. The OAL destination then selectively extracts each original IP packet/parcel (e.g., by setting pointers into the super-packet buffer and maintaining a reference count, by copying each packet into a separate buffer, etc.) and forwards each one to the network layer. During extraction, the OAL determines the IP protocol version of each successive original IP packet/parcel 'j' by examining the 4 most-significant bits of iH(j), and determines the length of each one by examining the rest of iH(j) according to the IP protocol version.¶
When an OAL source prepares a super-packet that includes an IPv6 ND message with an authentication signature as the first original IP packet/parcel (i.e., iHa/iDa), it calculates the authentication signature over the remainder of super-packet. Authentication and integrity for forwarding initial data messages in conjunction with IPv6 ND messages used to establish NCE state are therefore supported. (A second common use case entails a path MTU probe beginning with an unsigned IPv6 ND message followed by a suitably large NULL packet (e.g., an IP packet with padding octets added beyond the IP header and with {Protocol, Next Header} set to 59 ("No Next Header"), a UDP/IP packet with port number set to 9 ("discard") [RFC0863], etc.)¶
The OAL source can also apply this super-packet packing technique at the same time it performs OCH1 header compression as discussed in Section 6.5. Note that this technique can only be applied when all original IP packets are atomic packets with IP headers that differ only in Payload Length, such as for a stream of packets for a single flow that are queued for transmission service at roughly the same time.¶
The OAL header of a super packet may also include a Parcel Payload Option with AJ Type 0 if the total length of all payload packets/parcels exceeds 65535 octets. In that case, the super-packet must be forwarded as an atomic fragment over OAL paths that support such large sizes.¶
OAL sources may send NULL OAL packets known as "bubbles" for the purpose of establishing Network Address Translator (NAT) state on the path to the OAL destination. The OAL source prepares a bubble by crafting an OAL header with appropriate IPv6 source and destination ULAs, with the IPv6 Next Header field set to the value 59 ("No Next Header" - see [RFC8200]) and with 0 or more octets of NULL protocol data immediately following the IPv6 header.¶
The OAL source includes a random Identification value then encapsulates the OAL packet in L2 headers destined to either the mapped address of the OAL destination's first-hop ingress NAT or the L2 address of the OAL destination itself. When the OAL source sends the resulting carrier packet, any egress NATs in the path toward the L2 destination will establish state based on the activity. At the same time, the bubble themselves will be harmlessly discarded by either an ingress NAT on the path to the OAL destination or by the OAL destination itself.¶
The bubble concept for establishing NAT state originated in [RFC4380] and was later updated by [RFC6081]. OAL bubbles may be employed by mobility services such as AERO.¶
OMNI Hosts are end systems that connect to the OMNI link over ENET underlay interfaces (i.e., either via an OMNI interface or as a sublayer of the ENET interface itself). Each ENET connects to the rest of the OMNI link via a Client that distributes an MNP delegation. Clients delegate MNP addresses and/or sub-prefixes to ENET nodes (i.e., Hosts, other Clients, routers and non-OMNI hosts) using standard mechanisms such as DHCP [I-D.ietf-dhc-rfc8415bis][RFC2131] and IPv6 Stateless Address AutoConfiguration (SLAAC) [RFC4862]. Clients forward original IP packets/parcels between their ENET Hosts and peers on external networks acting as routers and/or OAL intermediate systems.¶
OMNI Hosts coordinate with Clients and/or other Hosts connected to the same ENET using OMNI L2 encapsulation of OMNI IPv6 ND messages. The L2 encapsulation headers and ND messages both use the MNP-based addresses assigned to ENET underlay interfaces as source and destination addresses (i.e., instead of ULAs). For IPv4 MNPs, the ND messages use IPv4-Compatible IPv6 addresses [RFC4291] in place of the IPv4 addresses.¶
Hosts discover Clients by sending encapsulated RS messages using an OMNI link IP anycast address (or the unicast address of the Client) as the RS L2 encapsulation destination as specified in Section 13. The Client configures the IPv4 and/or IPv6 anycast addresses for the OMNI link on its ENET interface and advertises the address(es) into the ENET routing system. The Client then responds to the encapsulated RS messages by sending an encapsulated RA message that uses its ENET unicast address as the source. (To differentiate itself from an INET border Proxy/Server, the Client sets the RA message OMNI Interface Attributes sub-option LHS field to 0 for the Host's interface index. When the RS message includes an L2 anycast destination address, the Client also includes an Interface Attributes sub-option for interface index 0 to inform the Host of its L2 unicast address - see: Section 13 for full details on the RS and RA message contents.)¶
Hosts coordinate with peer Hosts on the same ENET by sending encapsulated NS messages to receive an NA reply. (Hosts determine whether a peer is on the same ENET by matching the peer's IP address with the MNP (sub)-prefix for the ENET advertised in the Client's RA message [RFC8028].) Each ENET peer then creates a NCE and synchronizes Identification windows the same as for OMNI link neighbors, and the Host can then engage in OMNI link transactions with the Client and/or other ENET Hosts. The Host therefore regards the Client as if it were an ANET Proxy/Server, and the Client provides the same services that a Proxy/Server would provide. By coordinating with other Hosts, the peers can exchange large IP packets/parcels over the ENET using encapsulation and fragmentation if necessary.¶
When a Host prepares an original IP packet/parcel, it uses the IP address of its OMNI interface (which is the same as the IP address of the underlying native ENET interface) as the source and the IP address of the (remote) peer as the destination. The Host next performs parcellation if necessary (see: Section 6.13) then encapsulates the packet(s)/(sub-)parcel(s) in OMNI L2 headers while setting the L2 source to the L3 source address and L2 destination to either the L3 destination address if the peer is on the local ENET, or to the IP address of the Client otherwise. The Host can then proceed to exchange packets/parcels with the destination, either directly or via the Client as an intermediate system.¶
The encapsulation procedures are coordinated per Section 6.1, except that the OMNI L2 encapsulation header is followed by an IPv6 (Extended) Fragment Header. When the L2 encapsulation is based on an EUI or IPv4 address, the Host next translates the encapsulation header into an IPv6 header with IPv6 compatible addresses per Appendix B. Next, for IPv4 ENETs the Host sets the {IPv6 Traffic Class, Payload Length, Next Header, Hop Limit} fields according to the IPv4 {Type of Service, Total Length, Protocol, TTL} fields, respectively and also sets Flow Label as specified in [RFC6438]. The Host then applies IPv6 fragmentation to produce IPv6 fragments no smaller than the effective OFS described in Section 6.1. The Host next translates the IPv6 encapsulation headers back to OMNI L2 headers for the native ENET address format and with Type set to indicate the presence of the L2 IPv6 (Extended) Fragment Header. The Host finally sends the resultant carrier packets to the ENET peer.¶
When the ENET peer receives the carrier packets, it first translates the OMNI L2 headers back to IPv6 headers with compatible addresses. The peer then reassembles then removes the encapsulation headers and applies parcel reunification if necessary. The peer then either delivers the original IP packet/parcel to the transport layers if it is also the final destination or forwards the packet/parcel via the next hop if it is a Client acting as an intermediate system.¶
Hosts and Clients that initiate OMNI-based original IP packet/parcel transactions should first test the path toward the final destination using the parcel path qualification procedure specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2]. An OMNI Host that sends and receives parcels need not implement the full OMNI interface abstraction but MUST implement enough of the OAL to be capable of fragmenting and reassembling maximum-length encapsulated IP packets/parcels and sub-parcels as discussed above and in the following section.¶
Note: Hosts and their peer Clients/Hosts on the same ANET/ENET can improve efficiency by forwarding original IP packets/parcels that do not require fragmentation as direct encapsulations within the OMNI L2 header and without including a L2 IPv6 (Extended) Fragment Header. In that case, the first 4 bits immediately following the OMNI L2 encapsulation header encode the value '4' for IPv4 or '6' for IPv6. Note that this savings comes at the expense of omitting a well-behaved Identification, but this may be an acceptable tradeoff in many secured ANET/ENET instances.¶
IP parcels are formed by an OMNI Host or Client transport layer protocol entity identified by the "5-tuple" (source address, destination address, source port, destination port, protocol number) when it produces a {TCP,UDP} protocol data unit containing the concatenation of multiple transport layer protocol segments. The transport layer protocol entity then presents the buffer and non-final segment size to the network layer which appends a single {TCP,UDP}/IP header (plus any extension headers) before presenting the parcel to the OMNI Interface. Transport and network protocol formatting and processing rules as well as parcellation and reunification procedures for IP parcels are specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2], while detailed OAL encapsulation and fragmentation procedures are specified here.¶
When the network layer forwards a parcel, the OMNI interface invokes the OAL which forwards it to either an intermediate system or the final destination itself. The OAL source first invokes parcellation by subdividing the parcel into sub-parcels if necessary with each sub-parcel no larger than 65535 (minus headers). The OAL source also maintains a Parcel ID for each sub-parcel of the same original parcel that along with the Identification value for this OAL packet supports reassembly; the OAL source increments Parcel ID (modulo 64) for each successive parcel.¶
The OAL source next performs encapsulation on each sub-parcel with destination set to the next hop address. If the next hop is reached via a (M)ANET/INET interface, the OAL source inserts an OAL header the same as discussed in Section 6.1 and sets the destination to the ULA of the target Client. If the next hop is reached via an ENET interface, the OAL source instead inserts an IP header of the appropriate protocol version for the underlay ENET (i.e., even if the encapsulation header is IPv4) and sets the destination to the ENET IP address of the next hop. The OAL source inserts the encapsulation header even if no actual fragmentation is needed and/or even if the Parcel Payload Option is present.¶
The OAL source next assigns an appropriate Identification number that is monotonically-incremented for each consecutive sub-parcel, then performs IPv6 fragmentation over the sub-parcel if necessary to create fragments small enough to traverse the path to the next hop. (If the encapsulation header is IPv4, the OAL source first translates the encapsulation header into an IPv6 header with IPv4-Compatible IPv6 addresses during fragmentation/reassembly while inserting the IPv6 Extended Fragment Header.) The OAL source then writes the "Parcel ID" and sets/clears the "(P)arcel" and "More (S)egments" bits in the Reserved field of the IPv6 Extended Fragment Header of the first fragment (see: Figure 4). (The OAL source sets P to 1 for a parcel or to 0 for a non-parcel. When P is 1, the OAL next sets S to 1 for non-final sub-parcels or to 0 if the sub-parcel contains the final segment.) The OAL source then sends each resulting carrier packet to the next hop, i.e., after first translating the IPv6 encapsulation header back to IPv4 if necessary.¶
When the OAL destination receives the carrier packets, it reassembles if necessary (i.e., after first translating the IPv4 encapsulation header to IPv6 if necessary). If the P flag in the first fragment is 0, the OAL destination then processes the reassembled entity as an ordinary IP packet; otherwise it continues processing as a sub-parcel. If the OAL destination is not the final destination, it can optionally retain the sub-parcels along with their Parcel ID and Identification values for a brief time for opportunistic reunification with peer sub-parcels of the same original parcel identified by the 4-tuple consisting of the adaptation layer (OAL source, OAL destination, Parcel ID, Identification). (Note that the OAL destination must not consult the parcel's network layer "5-tuple" at the adaptation layer, since it is possible that multiple sub-parcels of the same parcel may be forwarded over different network paths).¶
The OAL destination performs adaptation layer reunification by concatenating the segments included in sub-parcels with the same Parcel ID and Identification values within 64 of one another to create a larger sub-parcel possibly even as large as the entire original (sub)parcel. Order of concatenation is determined by increasing Identification values, noting that a sub-parcel that sets any TCP control flags must occur as a first concatenation, and the final sub-parcel (i.e., the one with S set to 0) must occur as a final concatenation and not as an intermediate. The OAL destination then appends common {TCP,UDP}/IP headers plus extensions to each reunified sub-parcel as specified in [I-D.templin-6man-parcels2] and [I-D.templin-intarea-parcels2].¶
When the OAL destination is not the final destination, it next forwards the reunified (sub-)parcel(s) to the next hop toward the final destination while ensuring that the S flag remains set to 0 in the sub-parcel that contains the final segment. When the parcel or sub-parcels arrive at the final destination, it performs network layer reunification to form the largest possible (sub)-parcels (while honoring the S flag) then delivers them to the transport layer entity which acts on the enclosed 5-tuple information supplied by the original source.¶
Note: IP parcels may also originate from a non-OMNI original source and travel over multiple parcel-capable IP links before reaching an OMNI link ingress node (i.e., either a Client or Proxy/Server acting as a "relay"). The ingress node then forwards the parcel into the OMNI link according to the rules established above for locally-generated parcels, with the exception that the parcel IP TTL/Hop Limit is decremented. Similarly, when the IP parcel arrives at the OMNI link egress node (i.e., either a Client or Proxy/Server acting as a "relay"), the parcel may travel over multiple parcel-capable IP links before reaching the final destination.¶
Note: The OAL destination process of reunifying parcels at the adaptation layer is optional, and should be avoided in cases where performance could be negatively impacted. It is always acceptable (albeit sometimes sub-optimal) for the OAL destination to forward sub-parcels on toward the final destination without performing adaptation layer reunification, since each sub-parcel will contain a well-formed header and an integral number of transport layer protocol segments and with the Parcel ID field and P, S flag set appropriately. The final destination can then optionally perform network layer reunification independently of any adaptation layer reunification that may have been applied by the OAL.¶
Note: The "Parcel ID" that appears in the OAL Extended Fragment Header and OCH1/2 headers is an adaptation layer value that encodes the same value for all sub-parcels of the original parcel at the adaptation layer. This is different than the "(Parcel) Index" that appears in the Parcel Payload Option header as well as L2/L3 IPv6 Extended Fragment Headers, which is a network layer value that encodes a transport layer segment index.¶
Note: Parcel Path Qualification procedures require 2 additional ICMP PTB message Code values to identify a Parcel Report and Jumbo Report. These Code values are specified in [I-D.templin-6man-parcels2] for IPv6 and [I-D.templin-intarea-parcels2] for IPv4.¶
In light of the above, OAL sources, destinations and intermediate systems observe the following normative requirements:¶
OAL sources MUST forward original IP packets/parcels either larger than the OMNI interface minimum EMTU_R or smaller than the minimum OFS as atomic fragments (i.e., and not as multiple fragments).¶
OAL sources MUST perform OAL fragmentation such that all non-final fragments are equal in length while the final fragment may be a different length.¶
OAL sources MUST produce non-final fragments with payloads no smaller than the minimum OFS during fragmentation.¶
OAL intermediate systems SHOULD and OAL destinations MUST unconditionally drop any non-final OAL fragments with payloads smaller than the minimum OFS.¶
OAL destinations MUST drop any new OAL fragments that would overlap with other fragments and/or leave holes smaller than the minimum OFS between fragments that have already been received.¶
Note: Under the minimum OFS, an ordinary 1500-octet original IP packet/parcel would require at most 2 OAL fragments, with the first fragment containing 1024 payload octets and the final fragment containing the remainder. For all packet/parcel sizes, the likelihood of successful reassembly may improve when the OMNI interface sends all fragments of the same fragmented OAL packet consecutively over the same underlay interface pair instead of distributed across multiple underlay interface pairs. Finally, an assured minimum OFS allows continuous operation over all paths including those that traverse bridged L2 media with dissimilar MTUs.¶
Note: Certain legacy network hardware of the past millennium was unable to accept IP fragment "bursts" resulting from a fragmentation event - even to the point that the hardware would reset itself when presented with a burst. This does not seem to be a common problem in the modern era, where fragmentation and reassembly can be readily demonstrated at line rate (e.g., using tools such as 'iperf3') even over fast links on ordinary hardware platforms. Even so, while the OAL destination is reporting reassembly congestion (see: Section 6.9) the OAL source could impose "pacing" by inserting an inter-fragment delay and increasing or decreasing the delay according to congestion indications.¶
As discussed in Section 3.7 of [RFC8900], there are 4 basic threats concerning IPv6 fragmentation; each of which is addressed by effective mitigations as follows:¶
Overlapping fragment attacks - reassembly of overlapping fragments is forbidden by [RFC8200]; therefore, this threat does not apply to the OAL.¶
Resource exhaustion attacks - this threat is mitigated by providing a sufficiently large OAL reassembly cache and instituting "fast discard" of incomplete reassemblies that may be part of a buffer exhaustion attack. The reassembly cache should be sufficiently large so that a sustained attack does not cause excessive loss of good reassemblies but not so large that (timer-based) data structure management becomes computationally expensive. The cache should also be indexed based on the arrival underlay interface such that congestion experienced over a first underlay interface does not cause discard of incomplete reassemblies for uncongested underlay interfaces.¶
Attacks based on predictable fragment Identification values - in environments where spoofing is possible, this threat is mitigated through the use of Identification windows beginning with unpredictable values per Section 6.7. By maintaining windows of acceptable Identifications, OAL neighbors can quickly discard spurious carrier packets that might otherwise clutter the reassembly cache.¶
Evasion of Network Intrusion Detection Systems (NIDS) - since the OAL source employs a robust OFS, network-based firewalls can inspect and drop OAL fragments containing malicious data thereby disabling reassembly by the OAL destination. However, since OAL fragments may take different paths through the network (some of which may not employ a firewall) each OAL destination must also employ a firewall.¶
IPv4 includes a 2-octet (16-bit) Identification (IP ID) field with only 65535 unique values such that even at moderate data rates the field could wrap and apply to new carrier packets while the fragments of old carrier packets using the same IP ID are still alive in the network [RFC4963]. Carrier packets sent via an IPv4 path with DF set to 0 and with trailing payload/reassembly checksum(s) therefore ensure sufficient integrity to detect and discard reassembly errors. Since IPv6 provides a 4-octet (32-bit) Identification value, IP ID wraparound for IPv6 fragmentation may only be a concern at extreme data rates (e.g., 1Tbps or more). Note that these limitations are fully addressed through the Extended Identification format supported by [I-D.templin-6man-ipid-ext2].¶
Unless the path is secured at the network layer or below (i.e., in environments where spoofing is possible), OMNI interfaces MUST NOT send OAL packets/fragments with Identification values outside the current window and MUST secure IPv6 ND messages used for address resolution or window state synchronization. OAL destinations SHOULD therefore discard without reassembling any out-of-window OAL fragments received over an unsecured path.¶
The above sections primarily concern data plane aspects of the OMNI interface service and describe the data plane service model offered to the network layer. OMNI interfaces also internally employ a control plane service based on IPv6 Neighbor Discovery (ND) messaging. These control plane messages are forwarded over secured underlay interfaces (e.g., IPsec tunnels, secured direct point-to-point links, etc.) or over unsecured underlay interfaces and with an authentication signature included. In both cases, the IPv6 minimum MTU of 1280 octets must be assumed.¶
OMNI interfaces therefore send all control plane messages as "atomic OAL packets" that are no larger than 1280 octets and do not include an IPv6 Extended Fragment Header nor Compressed Routing Header (CRH) in contrast to the data plane. This means that these messages must not be subject to OAL fragmentation and reassembly, although they may be subject to L2 fragmentation and reassembly along some paths. Fragmentation security concerns for large IPv6 ND messages are documented in [RFC6980].¶
When the OMNI interface forwards original IP packets/parcels from the network layer it first invokes OAL encapsulation and fragmentation, then wraps each resulting OAL packet/fragment in any necessary L2 headers to produce carrier packets according to the native frame format of the underlay interface. For example, for Ethernet-compatible interfaces the frame format is specified in [RFC2464], for aeronautical radio interfaces the frame format is specified in standards such as ICAO Doc 9776 (VDL Mode 2 Technical Manual), for various forms of tunnels the frame format is found in the appropriate tunneling specification, etc.¶
When the OMNI interface encapsulates an OAL packet/fragment directly over an Ethernet-compatible link layer, the over-the-wire transmission format is shown in Figure 15:¶
The format includes a standard Ethernet Header ("eth-hdr") with EtherType TBD2 (see: Section 21.2) followed by an Ethernet Payload that includes zero or more OMNI Extension Headers followed by an OAL (or native IPv6/IPv4) Packet/Fragment. The Ethernet Payload is then followed by a standard Ethernet Trailer ("eth-trail").¶
The first OMNI extension header and the OAL Packet/Fragment both begin with a 4-bit "Type/Version" as discussed in Section 6.2. When "Type/Version" encodes an OMNI extension header type, the length of the extension headers is limited by [I-D.ietf-6man-eh-limits] and the length of the OAL Packet/Fragment is determined by the IP header fields that follow the extension headers.¶
When "Type/Version" encodes OMNI-OFH, OMNI-OCH1/2, OMNI-IP4 or OMNI-IP6 the length of the OAL Packet/Fragment is determined by the {Total, Payload} Length field found in the full/compressed header according to the specific protocol rules.¶
See Figure 2 for a map of the various L2 layering combinations possible. For any layering combination, the final layer (e.g., UDP, IP, Ethernet, etc.) must have an assigned number and frame format representation that is compatible with the selected underlay interface.¶
OMNI addressing follows the IPv6 addressing architecture [RFC4291] which states that: "IPv6 addresses of all types are assigned to interfaces, not nodes. An IPv6 unicast address refers to a single interface. Since each interface belongs to a single node, any of that node's interfaces' unicast addresses may be used as an identifier for the node." OMNI addressing further follows the IPv6 address preference policies specified in [RFC6724] as updated by [I-D.ietf-6man-rfc6724-update].¶
Each OMNI interface is configured over a set of MANET and/or ALVIF interfaces, and each ALVIF is in turn configured over a single *NET underlay interface. OMNI nodes assign IP addresses to their *NET interfaces according to the native underlay network autoconfiguration service(s) or through manual configuration. OMNI nodes assign IPv6 addresses to their OMNI, MANET and ALVIF interfaces as specified in this section.¶
[RFC4861] requires that hosts and routers assign Link-Local Addresses (LLAs) to all interfaces (including the OMNI interface), and that routers use their LLAs as the source address for RA and Redirect messages. Since the OMNI "link" comprises the concatenation of potentially many OMNI link segments, however, LLA uniqueness assurance is possible only on a per-segment basis and not across the entire OMNI link. For example, a Proxy/Server and all of the Client's that connect through it via a local *NET all share a common OMNI link segment over which LLA uniqueness applies. However, the LLAs used within a local *NET may overlap with those in other *NETs which represent different segments. Finally, MANET and ALVIF interfaces need not ensure uniqueness for the LLAs they assign since they will not be used as adaptation layer source or destination addresses nor as an identifier for the node.¶
[I-D.templin-6man-mla] specifies the MANET Local Address (MLA), and [RFC9374] specifies a Hierarchical Host Identity Tag (HHIT). OMNI nodes assign a unique MLA or HHIT to each MANET and ALVIF interface as well as to the OMNI interface itself. According to the IPv6 scoped addressing architecture [RFC4007], the node may assign the same MLA/HHIT to multiple ALVIF or MANET interfaces that connect to the same *NET. MLAs and HHITs are considered as adaptation layer addresses in the architecture, but nodes may also use them as the source and destination addresses of original IP packets exchanged between peers in isolated MANETs with no connection to the global Internet. When the node uses the OMNI interface, the original IP packet with MLA or HHIT addresses is subject to OAL encapsulation with an IPv6 header that also uses MLA or HHIT addresses. When the node uses a MANET or ALVIF interface, the original IP packet with MLA or HHIT addresses is presented for L2 encapsulation without including an OAL IPv6 header.¶
OMNI interfaces assign IPv6 Unique-Local Addresses (ULAs) and use them as the source and destination addresses in IPv6 packets forwarded over the OMNI interface within the local *NET. ULAs are routable only within the scope of each individual *NET, and are derived from the IPv6 prefix fd00::/8 (i.e., the ULA prefix fc00::/7 followed by the L bit set to 1). The 56 bits following fd00::/8 encode a 40-bit Global ID followed by a 16-bit Subnet ID followed by a 64-bit Interface Identifier as specified in Section 3 of [RFC4193].¶
When a Proxy/Server configures a ULA prefix for OMNI, it selects a 40-bit Global ID for the OMNI link segment initialized to a candidate pseudo-random value as specified in Section 3 of [RFC4193]. All nodes on the same OMNI link segment use the same Global ID, and statistical uniqueness of the pseudo-random Global ID provides a unique OMNI link segment identifier. This property allows different link segments to join together in the future without requiring renumbering even if the segments come in contact with one another and overlap, e.g., as a result of a mobility event.¶
Proxy/Servers for each OMNI link segment use the DHCPv6 service to delegate 1x1 mapped ULA/GUA SNP addresses for each Client that requests an address delegation. Clients in turn assign the ULA/GUA delegations to their OMNI interfaces which ensures that the addresses are available for use and that no duplicates will be assigned within each subnet. Considerations for 1x1 ULA/GUA address mapping are discussed in [I-D.ietf-v6ops-ula-usage-considerations] and [I-D.bctb-6man-rfc6296-bis].¶
The ULA presents an IPv6 address format that is routable within the local OMNI link segment and can be used to convey link-scoped (i.e., single-hop) IPv6 ND messages across multiple hops through OAL IPv6 encapsulation. The OMNI link extends across one or more underlying Internetworks to include all Proxy/Servers and other service nodes. All Clients are also considered to be connected to the OMNI link, however unnecessary encapsulations are omitted whenever possible to conserve bandwidth (see: Section 12).¶
OMNI domains manage MSPs delegated from the IP GUA prefix space [RFC4291] from which the MS delegates MNPs to support Client PI addressing. OMNI Proxy Servers also configure SNPs paired with a ULA configured as above to delegate PA internal (ULA) and external (GUA) addresses to Clients within their local *NETs.¶
For IPv6, MSPs are assigned to the OMNI link by IANA and/or an associated Regional Internet Registry [IPV6-GUA] such that the link can be interconnected to the global IPv6 Internet without causing inconsistencies in the routing system. Instead of GUAs, an OMNI link could use ULAs with the 'L' bit set to 0 (i.e., from the "ULA-C" prefix fc00::/8) [RFC4193], however this would require IPv6 NAT if the domain were ever connected to the global IPv6 Internet.¶
For IPv4, MSPs are assigned to the OMNI link by IANA and/or an associated RIR [IPV4-GUA] such that the link can be interconnected to the global IPv4 Internet without causing routing inconsistencies. An OMNI *NET could instead use private IPv4 prefixes (e.g., 10.0.0.0/8, etc.) [RFC3330], however this would require IPv4 NAT at the *NET boundary. OMNI interfaces advertise IPv4 MSPs into IPv6 routing systems as "6to4 prefixes" [RFC3056] (e.g., the IPv6 prefix for the IPv4 MSP "V4ADDR/24" is 2002:V4ADDR::/40).¶
IPv4 routers that configure OMNI interfaces advertise the prefix TBD3/N (see: IANA Considerations) into the routing systems of their connected *NETs and assign the IPv4 OMNI anycast address TBD3.1 to their *NET interfaces. IPv6 routers that configure OMNI interfaces advertise the prefix 2002:TBD3::/(N+16) into the routing systems of their connected *NETs and assign the IPv6 OMNI anycast address 2002:TBD3:: to their *NET interfaces.¶
Proxy/Server OMNI interfaces configure ULA/GUA IPv6 SNP SRA addresses per [RFC4291] and accept packets addressed to the SRA the same as for any IPv6 router. Proxy/Servers also configure the global IPv6 SRA address for each MSP managed by this OMNI link and accept packets addressed to the SRA address on their internal interfaces to support Client OMNI link discovery. Client OMNI interfaces configure the IPv6 SRA address corresponding to their MNP delegations.¶
OMNI interfaces use their OMNI IPv6 and IPv4 anycast addresses to support control plane Service Discovery in the spirit of [RFC7094], i.e., the addresses are not intended for use in supporting longer term data plane flows. Specific applications for OMNI IPv6 and IPv4 anycast addresses are discussed throughout the document as well as in [I-D.templin-6man-aero3].¶
OMNI Clients and Proxy/Servers that connect over open Internetworks include a unique node identification value for themselves in the OMNI options of their IPv6 ND messages (see: Section 10.2.3). An example identification value alternative is the (H)HIT per [RFC7401] and [RFC9374]. (Another example is the Universally Unique IDentifier (UUID) [RFC9562] which can be self-generated by a node without supporting infrastructure with very low probability of collision.)¶
When a Client is truly outside the context of any infrastructure, it may have no addressing information at all. In that case, the Client can use an MLA or HHIT as an IPv6 source/destination address for sustained communications in Vehicle-to-Vehicle (V2V) and (multihop) Vehicle-to-Infrastructure (V2I) scenarios. The Client can also propagate the MLA or HHIT into the multihop routing tables of (collective) Mobile/Vehicular Ad-hoc Networks (MANETs/VANETs) using only the vehicles themselves as communications relays.¶
MLAs and HHITs provide an especially useful construct since they appear as properly-formed IPv6 addresses and can therefore be assigned to interfaces. Clients may assign an MLA or HHIT to their MANET or ALVIF interfaces to support peer-to-peer communications with other nodes within the same OMNI link segment without the need for OMNI encapsulation. Clients also assign the same MLA or HHIT to the OMNI interface itself so that the MLA or HHIT can appear as the source address of an original IP packet that becomes subject to OMNI encapsulation. Clients may inject their MLA or HHIT into the local routing system of each OMNI link segment, but Proxy/Servers must not inject MLAs or HHITs into the OMNI link global routing system.¶
OMNI interfaces maintain a network layer conceptual neighbor cache per [RFC1256] or [RFC4861] the same as for any IP interface. The network layer maintains state through static and/or dynamic Neighbor Cache Entry (NCE) configurations.¶
Each OMNI interface also maintains a separate internal adaptation layer conceptual neighbor cache that includes a NCE for each of its active OAL neighbors. For each peer NCE, OAL neighbors also maintain AERO Forwarding Vectors (AFVs) which map per-interface-pair parameters. Throughout this document, the terms "neighbor cache", "NCE" and "AFV" refer to this OAL neighbor information unless otherwise specified.¶
IPv6 Neighbor Discovery (ND) [RFC4861] messages sent over OMNI interfaces without OAL encapsulation observe the native underlay interface Source/Target Link-Layer Address Option (S/TLLAO) format (e.g., for Ethernet the S/TLLAO is specified in [RFC2464]). IPv6 ND messages sent from within the OMNI interface using OAL encapsulation do not include S/TLLAOs, but instead include a new option type that encodes OMNI link-specific information. Hence, this document does not define a new S/TLLAO format but instead defines a new option type termed the "OMNI option" designed for these purposes. (Note that OMNI interface IPv6 ND messages sent without encapsulation may include both OMNI options and S/TLLAOs, but the information conveyed in each is mutually exclusive.)¶
For each IPv6 ND message, the OMNI interface includes one or more OMNI options (and any other ND message options) then completely populates all option information. OMNI options should be padded when necessary to ensure that they end on their natural 64-bit boundaries the same as for any IPv6 ND message option.¶
If the OMNI interface includes an OMNI option with an authentication signature, it first sets the signature field to 0 then calculates the authentication signature beginning after the IPv6 ND message header checksum field. The OMNI interface extends the calculation over the entire length of the ND message (as well as any concatenated extensions in the case of a super-packet) then writes the authentication signature value into the appropriate OMNI authentication sub-option field.¶
The OMNI interface then applies any non-OMNI authentication signatures, calculates the IPv6 ND message checksum per [RFC4443] beginning with a pseudo-header of the IPv6 header and writes the value into the Checksum field. OMNI interfaces verify first integrity then authenticity of each IPv6 ND message or super-packet received, and process the message further only following successful verification.¶
OMNI interface Clients such as aircraft typically have multiple wireless data link types (e.g. satellite-based, cellular, terrestrial, air-to-air directional, etc.) with diverse performance, cost and availability properties. The OMNI interface would therefore appear to have multiple L2 connections, and may include information for multiple underlay interfaces in a single IPv6 ND message exchange. OMNI interfaces manage their dynamically-changing multilink profiles by including OMNI options in IPv6 ND messages as discussed in the following subsections.¶
OMNI options appear in IPv6 ND messages formatted as shown in Figure 16:¶
In this format:¶
Type is set to TBD4 (see: IANA Considerations).¶
Length is set to the number of 8-octet blocks in the option. The value 0 is invalid, while the values 1 through 255 (i.e., 8 through 2040 octets, respectively) indicate the total length of the OMNI option. If multiple OMNI option instances appear in the same IPv6 ND message, the union of the contents of all OMNI options is accepted unless otherwise qualified for specific sub-options below.¶
Sub-Options is a Variable-length field padded with Pad1/N sub-options if necessary (see below) such that the complete OMNI option is an integer multiple of 8 octets long. The Sub-Options field contains zero or more sub-options as specified in Section 10.2.¶
The OMNI option is included in OMNI interface IPv6 ND messages; the option is processed by receiving interfaces that recognize it and otherwise ignored. The OMNI interface processes all OMNI option instances received in the same IPv6 ND message in the consecutive order in which they appear. The OMNI option(s) included in each IPv6 ND message may include full or partial information for the neighbor. The OMNI interface therefore retains the union of the information in the most recently received OMNI options in the corresponding NCE.¶
Each OMNI option includes a Sub-Options block containing zero or more individual sub-options. Each consecutive sub-option is concatenated immediately following its predecessor. All sub-options except Pad1 (see below) are in an OMNI-specific type-length-value (TLV) format encoded as follows:¶
Sub-Type is a 5-bit field that encodes the sub-option type. Sub-option types defined in this document are:¶
Sub-Types 16-29 are available for future assignment for major protocol functions, while Sub-Type 30 supports scalable extension to include other functions. Sub-Type 31 is reserved by IANA.¶
Sub-Length is an 11-bit field that encodes the length of the Sub-Option Data in octets.¶
Sub-Option Data is a block of data with format determined by Sub-Type and length determined by Sub-Length. Note that each sub-option is concatenated consecutively with the previous and may therefore begin and/or end on an arbitrary octet boundary.¶
The OMNI interface codes each sub-option with a 2-octet header that includes Sub-Type in the most significant 5 bits followed by Sub-Length in the next most significant 11 bits. Each sub-option encodes a maximum Sub-Length value of 2038 octets minus the lengths of the OMNI option header and any preceding sub-options. This allows ample Sub-Option Data space for coding large objects (e.g., ASCII strings, domain names, protocol messages, security codes, etc.), while a single OMNI option is limited to 2040 octets the same as for any IPv6 ND option.¶
The OMNI interface codes initial sub-options in a first OMNI option instance and any additional sub-options in additional instances in the same IPv6 ND message in the intended order of processing. If the size of all OMNI options with their sub-options would cause the IPv6 ND message to exceed the OMNI interface MTU, the OMNI interface can code any remaining sub-options in additional IPv6 ND messages.¶
The OMNI interface processes all OMNI options received in an IPv6 ND message while skipping over and ignoring any unrecognized sub-options. The OMNI interface processes the sub-options of all OMNI option instances in the consecutive order in which they appear in the IPv6 ND message, beginning with the first instance and continuing through any additional instances to the end of the message. If an individual sub-option length would cause processing to exceed the OMNI option instance and/or IPv6 ND message lengths, the OMNI interface accepts any sub-options already processed and ignores the remainder of that instance.¶
IPv6 ND messages that require OMNI authentication services include a Node Identification sub-option as the first sub-option of the first OMNI option if necessary. Whether or not a Node Identification is included, the IPv6 ND message includes some form of authentication (e.g., HMAC, HIP, QUIC, etc.) as the immediately next sub-option whether in the same or a different OMNI option. A single IPv6 ND messages includes a single effective OMNI authentication service sub-option; if multiple are included, the first sub-option is processed and all others are ignored. The IPv6 ND message may instead include non-OMNI authentication options such as those specified in [RFC3971] or [RFC8928]. Nodes that receive IPv6 ND messages over unsecured underlying networks first verify the IPv6 ND message checksum then authenticate the message by processing the authentication option/sub-option.¶
Note: large objects that exceed the maximum Sub-Option Data length are not supported under the current specification; if this proves to be limiting in practice, future specifications may define support for fragmenting large sub-options across multiple OMNI options within the same IPv6 ND message (or even across multiple IPv6 ND messages, if necessary).¶
The following sub-option types and formats are defined in this document:¶
Sub-Type is set to 0. If multiple instances appear in OMNI options of the same message all are processed.¶
Sub-Type is followed by 3 'x' bits, set to any value on transmission (typically all-zeros) and ignored on reception. Pad1 therefore consists of a single octet with the most significant 5 bits set to 0, and with no Sub-Length or Sub-Option Data fields following.¶
If more than a single octet of padding is required, the PadN option, described next, should be used, rather than multiple Pad1 options.¶
Sub-Type is set to 1. If multiple instances appear in OMNI options of the same message all are processed.¶
Sub-Length is set to N that encodes the number of padding octets that follow.¶
Sub-Option Data consists of N octets, set to any value on transmission (typically all-zeros) and ignored on receipt.¶
When an intermediate system forwards an IPv6 ND message with OMNI options, it can void any non-Pad1 sub-options that should not be processed by the next hop by simply writing the value '1' (PadN) over the Sub-Type. When the intermediate system alters the IPv6 ND message in this way, the integrity check is invalidated and must be re-calculated. See: Appendix C for a discussion of IPv6 ND message authentication and integrity.¶
The Node Identification sub-option (when present) must appear as the first sub-option of the first OMNI option in each IPv6 ND message. If multiple instances appear in OMNI options of the same IPv6 ND message the first instance of a specific ID-Type is processed and all other instances of the same ID-Type are ignored. (A single IPv6 ND message can therefore convey multiple distinct Node Identifications - each with a different ID-Type.)¶
The format and contents of the sub-option are shown in Figure 21:¶
Sub-Type is set to 2. Multiple instances are processed as discussed above.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow. The ID-Type field is always present, and the maximum Node Identification Value length is limited by the remaining available space in this OMNI option.¶
ID-Type is a 1-octet field that encodes the type of the Node Identification Value. The following ID-Type values are currently defined:¶
0 - Universally Unique IDentifier (UUID) [RFC9562]. Indicates that Node Identification Value contains a 16-octet UUID.¶
1 - Host Identity Tag (HIT) [RFC7401]. Indicates that Node Identification Value contains a 16-octet HIT.¶
2 - Hierarchical HIT (HHIT) [RFC9374]. Indicates that Node Identification Value contains a 16-octet HHIT.¶
3 - Network Access Identifier (NAI) [RFC7542]. Indicates that Node Identification Value contains an (N-1)-octet NAI.¶
4 - Fully-Qualified Domain Name (FQDN) [RFC1035]. Indicates that Node Identification Value contains an (N-1)-octet FQDN.¶
5 - IPv6 Address. Indicates that Node Identification contains a 16-octet IPv6 address that is not a (H)HIT. The IPv6 address type is determined according to the IPv6 addressing architecture [RFC4291].¶
6 - 252 - Unassigned.¶
253 - 254 - reserved for experimentation, as recommended in [RFC3692].¶
255 - reserved by IANA.¶
Node Identification Value is an (N-1)-octet field encoded according to the appropriate the "ID-Type" reference above.¶
OMNI interfaces code Node Identification Values used for DHCPv6 messaging purposes as a DHCP Unique IDentifier (DUID) using the "DUID-EN for OMNI" format with enterprise number 45282 (see: Section 21) as shown in Figure 22:¶
In this format, the OMNI interface codes the ID-Type and Node Identification Value fields from the OMNI sub-option following a 6-octet DUID-EN header, then includes the entire "DUID-EN for OMNI" in a DHCPv6 message per [I-D.ietf-dhc-rfc8415bis].¶
The Authentication sub-option includes a Hashed Message Authentication Code (HMAC) computed according to [RFC2104] and [RFC6234].¶
The Authentication sub-option is formatted as shown in Figure 23:¶
Sub-Type is set to 3. The Authentication sub-option must appear at most once in any IPv6 ND message; if multiple instances appear in OMNI options of the same message the first is processed and all others are ignored.¶
Sub-Length is set to N, i.e., the length of the option in octets beginning immediately following the Sub-Length field and extending to the end of the HMAC. The length of the HMAC is therefore limited by the remaining available space for this sub-option.¶
Type encodes the authentication algorithm type found in the IANA "ICMPv6 Parameters - Trust Anchor Option (Type 15) Name Field" registry, and determines the length of the HMAC. For example, when Type is 3 the authentication algorithm is SHA-1 and the HMAC is 160 bits (20 octets) in length, when Type is 5 the algorithm is SHA-256 and the HMAC is 256 bits (32 octets) in length, etc. A full list of available Types is found in the registry, which cites [RFC6495] for several well-known Types. The Type value TBD7 is reserved for the Edwards-Curve Digital Signature Algorithm (EdDSA) (see IANA Considerations) with the HMAC (i.e., digital signature) including 64 octets for Ed25519 or 114 octets for Ed448 per [RFC8032].¶
HMAC includes a Hashed Message Authentication Code or digital signature for this IPv6 ND message with (N-1)-octet length according to Type.¶
IPv6 ND messages used to manage neighbor relationships between Clients and their Proxy/Servers (and also between Clients and their peer Clients) include a Neighbor Control OMNI sub-option. Each IPv6 ND message includes at most one Neighbor Control sub-option which must be specific to the underlying interface over which the ND message is sent.¶
The Neighbor Control sub-option is formatted as follows:¶
Sub-Type is set to 4. If multiple instances appear in OMNI options of the same message, the first is processed and all others are ignored.¶
Sub-Length is set to N.¶
Sub-Option Data includes an N-octet neighbor control flags field. This specification defines several Control flags in the first octet.¶
Clients set the Neighbor Unreachability Detection (NUD), Address Resolution Responder (ARR) and Report (RPT) flags in RS messages to control the operation of their Proxy/Server neighbors as discussed in Section 13.¶
Nodes set the Synchronous (u)NA Required (SNR) flag in non-solicitation IPv6 ND messages (i.e., solicited/unsolicited NA/RA and Redirects) for which they require a synchronous (but technically "unsolicited") NA reply (see: [I-D.templin-6man-aero3]).¶
OAL intermediate systems set the Path Change (PCH) flag in uNA messages used to report a change in a path established by multilink forwarding.¶
All remaining flags in the first octet plus any additional octets are Reserved and must be set to 0; future specifications may define new flags.¶
The Interface Attributes sub-option provides neighbors with forwarding information for the multilink conceptual sending algorithm discussed in Section 12. Neighbors use the forwarding information to select among candidate underlay interfaces that can be used to forward carrier packets to the neighbor based on factors such as traffic selectors and link metrics. Interface Attributes further include link layer address information to be used for either direct INET encapsulation for targets in the local SRT segment or spanning tree forwarding for targets in remote SRT segments.¶
OMNI nodes include Interface Attributes for some/all of a source or target Client's underlay interfaces in NS/NA and uNA messages used to publish Client information (see: [I-D.templin-6man-aero3]). At most one Interface Attributes sub-option for each distinct ifIndex may be included; if an IPv6 ND message includes multiple Interface Attributes sub-options for the same ifIndex, the first is processed and all others are ignored. OMNI nodes that receive NS/NA messages can use all of the included Interface Attributes and/or Traffic Selectors to formulate a map of the prospective source or target node as well as to seed the information to be populated in future neighbor exchanges.¶
OMNI Clients and Proxy/Servers also include Interface Attributes sub-options in RS/RA messages used to initialize, discover and populate routing and addressing information. Each RS message MUST contain exactly one Interface Attributes sub-option with an ifIndex corresponding to the Client's underlay interface used to transmit the message, and each RA message MUST echo the same Interface Attributes sub-option with any (proxyed) information populated by the FHS Proxy/Server to provide operational context.¶
When an FHS Proxy/Server receives an RS message destined to an anycast L2 address, it MUST include an additional Interface Attributes sub-option with ifIndex '0' that encodes its own unicast L2 address relative to the Client's underlay interface in the solicited RA response. Any additional Interface Attributes sub-options that appear in RS/RA messages (i.e., besides those for the Client's own ifIndex and ifIndex '0') are ignored.¶
The Interface Attributes sub-option is formatted as shown below:¶
Sub-Type is set to 5. Multiple instances are processed as discussed above.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow.¶
Sub-Option Data contains an "Interface Attributes" option encoded as follows:¶
ifIndex is a 4-octet index value corresponding to a specific underlay interface. Client OMNI interfaces MUST number each distinct underlay interface with a non-zero ifIndex value assigned by network management per [RFC2863] and include the value in this field. The ifIndex value '0' denotes "unspecified".¶
ifType is a 4-octet type value corresponding to this underlay interface. The value is coded per the 'IANAifType-MIB' registry [http://www.iana.org].¶
ifProvider is a 4-octet provider identifier corresponding to this underlay interface. This document defines the single provider identifier value '0' (undefined). Future documents may define other values.¶
ifMetric encodes a 4-octet interface metric. Lower values indicate higher priorities, and the highest value indicates an interface that should not be selected. The ifMetric setting provides an instantaneous indication of the interface bandwidth, link quality, signal strength, cost, etc.; hence, its value may change in successive IPv6 ND messages.¶
ifGroup is a 4-octet identifier for a Link Aggregation Group (LAG) [IEEE802.1AX] corresponding to the underlay interface identified by ifIndex. Interface attributes for ifIndex members of the same group will encode the same value in ifGroup. This document defines the single ifGroup value '0' meaning "no group assigned". Future documents will specify the setting of other values.¶
SRT is a 1-octet Segment Routing Topology prefix length that determines the length associated with this sub-tree of a larger topology that may include the concatenation of multiple connected segments. The SRT value ranges from 0 to 128.¶
FMT - a 1-octet "Forward/Mode/Type" code interpreted as follows:¶
The most significant 2 bits (i.e., "FMT-Forward" and "FMT-Mode") are interpreted in conjunction with one another. When FMT-Forward is clear, the LHS Proxy/Server performs OAL reassembly and decapsulation to obtain the original IP packet/parcel before forwarding. If the FMT-Mode bit is clear, the LHS Proxy/Server then forwards the original IP packet/parcel at L3; otherwise, it invokes the OAL to re-encapsulate, re-fragment and sends the resulting carrier packets to the Client via the selected underlay interface. When FMT-Forward is set, the LHS Proxy/Server forwards unmodified OAL fragments to the Client without reassembling. If FMT-Mode is clear, all carrier packets destined to the Client must always be sent via the LHS Proxy/Server; otherwise the Client is eligible for direct forwarding over the open INET where it may be located behind one or more NATs.¶
The next most significant 2 bits are reserved, and the value encoded in the least significant 4 bits (i.e., "FMT-Type") determines the type and length of the L2ADDR field. The following values are currently defined:¶
0 - L2ADDR is 0 octets in length and unused.¶
1 - L2ADDR is 4 octets in length and encodes an IPv4 address.¶
2 - L2ADDR is 6 octets in length and encodes an EUI-48 address [EUI].¶
3 - L2ADDR is 8 octets in length and encodes an EUI-64 address [EUI].¶
4 - L2ADDR is 16 octets in length and encodes an IPv6 address.¶
LHS GUA/L2ADDR - encodes the 16 octet SNP IPv6 GUA of the node relative to the LHS Proxy/Server followed by the L2ADDR field formatted as above. FMT, SRT and LHS together provide guidance for the OMNI interface forwarding algorithm. Specifically, if LHS::/SRT is located in the local OMNI link segment, then the source can address the target Client either through its dependent Proxy/Server or through direct encapsulation following NAT traversal according to FMT. Otherwise, the target Client is located on a different SRT segment and the path from the source must employ a combination of route optimization and spanning tree hop traversals. L2ADDR identifies the LHS Proxy/Server's INET-facing interface not located behind NATs, therefore no UDP port number is included since port number 8060 is used when the L2 encapsulation includes a UDP header. Instead, L2ADDR includes only an L2 address with type and length determined by FMT-Type as described above. When L2ADDR includes an IPv4 or IPv6 address, it is recorded in network byte order in ones-compliment "obfuscated" form per [RFC4380].¶
Traffic Selector Blocks(s) - zero or more Traffic Selector blocks follow, with their total length determined by the number of octets remaining in the Interface Attributes sub-option beyond the end of the LHS Proxy/Server information. Each Traffic Selector block is formatted the same as specified in Section 10.2.7 and processed consecutively, with its length subtracted from the remaining length of the Interface Attributes sub-option.¶
The Traffic Selector sub-option provides forwarding information for the multilink conceptual sending algorithm discussed in Section 12. The sub-option includes traffic selector information per [RFC6088] as ancillary information for an Interface Attributes sub-option with the same ifIndex value, or as discrete information for the included ifIndex when no Interface Attributes sub-option is present.¶
IPv6 ND messages may include multiple Traffic Selectors for some or all of the source/target Client's underlay interfaces (see: [I-D.templin-6man-aero3] for further discussion). Traffic Selectors must be honored by all implementations in the format shown below:¶
Sub-Type is set to 6. Multiple instances with the same or different ifIndex values may appear in the same IPv6 ND message. When multiple instances appear, all are processed and the cumulative information from all is accepted.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow.¶
Sub-Option data begins with a 4-octet ifIndex value corresponding to a specific underlay interface. (Note that when traffic selector blocks appear within an Interface Attributes sub-option, the ifIndex field already appears and is not included multiple times.)¶
The remainder of Sub-Option Data contains one or more "Traffic Selector" blocks for this ifIndex that each begin with 1-octet "TS Length" and "TS Format" fields. TS length encodes the combined lengths of the TS* fields plus the Traffic Selector body that follows (i.e. a value between 2-255 octets). When TS Format encodes the value 1 or 2, the Traffic Selector body encodes an IPv4 or IPv6 traffic selector per [RFC6088] beginning with 16 flag bits ("A-N" plus 2 "Reserved"); when TS Format encodes any other value the Traffic Selector block is skipped and processing resumes beginning with the next Traffic Selector block (if any). The Traffic Selector block elements then appear immediately after the flags (with no 16-bit Reserved field included) and encode the information corresponding to any set flag bit(s) in order the same as specified in [RFC6088]. Each included Traffic Selector block is processed consecutively, with its length subtracted from the remaining sub-option length until all blocks are processed. If the length of any Traffic Selector block would exceed the remaining length for the entire sub-option, the remainder of the sub-option is ignored.¶
Clients exchange IPv6 ND messages with their FHS Proxy/Servers and other Client peers to populate forward and reverse path AFIB state.¶
The Multilink Vector sub-option provides the necessary information allowing OAL intermediate and end systems in the path to establish AFVs to support future packet forwarding.¶
Each IPv6 ND message contain at most one Multilink Vector sub-option; if multiple are present, the first is processed and all others are ignored.¶
The Multilink Vector sub-option is formatted as follows:¶
Sub-Type is set to 7 and Sub-Length is set to 32. If multiple instances appear in OMNI options of the same message, the first is processed and all others are ignored.¶
the first 4 octets of Sub-Option Data include an AFVI generated by the IPv6 ND message source. When the SYN flag is set, each OAL intermediate hop records this (previous hop) AFVI value along with the previous hop L2 address in an AFIB AFV, then also generates and records a new (next hop) AFVI value. When the SYN flag is not set, the intermediate hop instead uses the AFVI value to locate an existing AFV without creating a new one. Each intermediate hop then rewrites the Multilink Vector AFVI field to the next hop value and forwards the message to the next hop. The process continues until the message arrives at the IPv6 ND message destination.¶
the next 8 octets include the 4-octet ifIndex of the FHS (initiator) node followed by the 4-octet ifIndex of the LHS (responder) node.¶
the final 20 octets of Sub-Option Data follows from the Transmission Control Protocol (TCP) header specified in Section 3.1 of [RFC9293]. The field is formatted as an 8-octet Sequence Number, followed by an 8-octet Acknowledgement Number, followed by a 1-octet flags field followed by a 3-octet Window size. The TCP (ACK, RST, SYN) flags are used for TCP-like window synchronization, while the TCP (CWR, ECE, URG, PSH, FIN) flags are unused. The OPT flag (discussed in Section 6.7) is an OMNI-specific replacement for the TCP PSH flag, the TST flag (discussed in [I-D.templin-6man-aero3] is an OMNI-specific replacement for the TCP FIN flag and the 3 remaining unused flags appear as reserved (RES). Together, these fields support the OAL window synchronization services specified in Section 6.7.¶
When an IPv6 ND message source includes a Multilink Vector sub-option, it MUST temporarily reset the AFVI to 0 before calculating an authentication signature over the message since OAL intermediate nodes in the path will rewrite the AFVI to a different value. The source then MUST reset the AFVI to its actual value before calculating the IPv6 ND message checksum and forwarding the message. When an OAL intermediate system or destination receives the message, it first verifies the checksum then MUST temporarily reset the AFVI to 0 before verifying the authentication signature.¶
Sub-Type is set to 8. If multiple instances appear in OMNI options of the same message all are processed.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow.¶
Geo Type is a 1-octet field that encodes a type designator that determines the format and contents of the Geo Coordinates field that follows. The following types are currently defined:¶
0 - NULL, i.e., the Geo Coordinates field is zero-length.¶
Geo Coordinates is a type-specific format field of length up to the remaining available space for this OMNI option. New formats to be specified in future documents and may include attributes such as latitude/longitude, altitude, heading, speed, etc.¶
The Dynamic Host Configuration Protocol for IPv6 (DHCPv6) sub-option may be included in the OMNI options of Client RS messages and Proxy/Server RA messages. The DHCPv6 sub-option is formatted per Section 8 of [I-D.ietf-dhc-rfc8415bis] as shown in Figure 29:¶
Sub-Type is set to 9. If multiple instances appear in OMNI options of the same message the first is processed and all others are ignored.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow. The 'msg-type' and 'transaction-id' fields are always present; hence, the length of the DHCPv6 options is limited by the remaining available space for this OMNI option.¶
'msg-type' and 'transaction-id' are coded according to Section 8 of [I-D.ietf-dhc-rfc8415bis].¶
A set of DHCPv6 options coded according to Section 21 of [I-D.ietf-dhc-rfc8415bis] follows.¶
The Protocol Independent Multicast - Sparse Mode (PIM-SM) Message sub-option may be included in the OMNI options of IPv6 ND messages. The PIM-SM message sub-option is formatted per Section 4.9 of [RFC7761] and as shown in Figure 30:¶
Sub-Type is set to 10. If multiple instances appear in OMNI options of the same message all are processed.¶
Sub-Length is set to N, i.e., the length of the option in octets beginning immediately following the Sub-Length field and extending to the end of the PIM-SM message. The length of the entire PIM-SM message is therefore limited by the remaining available space for this OMNI option.¶
The PIM-SM message is coded exactly as specified in Section 4.9 of [RFC7761], except that the Checksum field is omitted since message integrity is already assured by the IPv6 ND message Checksum. The Reserved field is set to 0 on transmission and ignored on reception. The "PIM Ver" field encodes the value 2, and the "Type" field encodes the PIM message type. (See Section 4.9 of [RFC7761] for a list of PIM-SM message types and formats.)¶
The Host Identity Protocol (HIP) Message sub-option (when present) provides an authentication service alternative for IPv6 ND messages exchanged between Clients and FHS Proxy/Servers (or between Clients and their peers) over an open Internetwork. When the HIP service is used, FHS Proxy/Servers verify the HIP authentication signatures in source Client IPv6 ND messages then remove the HIP message sub-option and securely forward the ND messages to other OMNI nodes. LHS Proxy/Servers that receive secured IPv6 ND messages from other OMNI nodes that do not already include a security sub-option can insert HIP authentication signatures before forwarding them to the target Client.¶
OMNI interfaces that use the HIP service include the HIP message sub-option when they forward IPv6 ND messages that require security over INET underlay interfaces, i.e., where authentication and integrity is not already assured by link/physical layers or other OMNI layer services. The OMNI interface calculates the authentication signature over the entire length of the OAL packet (or super-packet) beginning after the IPv6 ND message header and extending over the remainder of the OAL packet or super-packet. OMNI interfaces that process OAL packets containing secured IPv6 ND messages verify the signature then either process the rest of the message locally or forward a proxyed copy to the next hop.¶
When an FHS Client inserts a HIP message sub-option in an IPv6 ND message destined to a target in a remote spanning tree segment, it must ensure that the insertion does not cause the message to exceed the IPv6 minimum MTU. If the LHS Proxy/Server cannot create sufficient space through any means without causing the IPv6 ND message to exceed the IPv6 minimum MTU, it returns a suitable error (see: Section 10.2.15) and drops the message.¶
The HIP message sub-option is formatted as shown below:¶
Sub-Type is set to 11. If multiple instances appear in OMNI options of the same message the first is processed and all others are ignored.¶
Sub-Length is set to N, i.e., the length of the option in octets beginning immediately following the Sub-Length field and extending to the end of the HIP parameters. The length of the entire HIP message is therefore limited by the remaining available space for this OMNI option.¶
The HIP message is coded per Section 5 of [RFC7401], except that the OMNI "Sub-Type" and "Sub-Length" fields replace the first 2 octets of the HIP message header (i.e., the Next Header and Header Length fields). Also, since the IPv6 ND message is already protected by its own checksum, the 2-octet HIP message Checksum field is omitted.¶
Sub-Type is set to 12. If multiple instances appear in OMNI options of the same IPv6 ND message, the first is processed and all others are ignored.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow.¶
The QUIC-TLS message [RFC9000][RFC9001][RFC9002] encodes the QUIC and TLS message parameters necessary to support QUIC connection establishment.¶
IPv6 ND messages serve as couriers to transport the QUIC and TLS parameters necessary to establish a secured QUIC connection.¶
Fragmentation Report (FRAGREP) sub-options may be included in the OMNI options of uNA messages sent from an OAL destination to an OAL source. The message consists of (N/16)-many (Identification, Bitmap)-tuples which include the Identification values of OAL fragments received plus a Bitmap marking the ordinal positions of individual fragments received and missing.¶
Sub-Type is set to 13. If multiple instances appear in OMNI options of the same message all are processed.¶
Sub-Length is set to N which must be a multiple of 16, i.e., the combined lengths of each (Identification, Bitmap) pair beginning immediately following the Sub-Length field and extending to the end of the sub-option.¶
Identification(i) includes the 8-octet Identification value found in a received OAL fragment.¶
Bitmap(i) includes a 64-bit checklist of up to 64 ordinal fragments for this Identification, with each bit set to 1 for a fragment received or 0 for a fragment corrupted, lost or still in transit. For example, for a 20-fragment OAL packet with ordinal fragments #3, #10, #13 and #17 missing or corrupted and all other fragments received or still in transit, Bitmap(i) encodes the following:¶
Sub-Type is set to 14. If multiple instances appear in OMNI options of the same IPv6 ND message all are processed.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow.¶
Sub-Option Data includes an N-octet ICMPv6 Error Message body encoded per Section 2.1 of [RFC4443], but with the IPv6 header and Checksum fields omitted. OMNI interfaces include as much of the "packet in error" in the ICMPv6 error message body as possible without causing the IPv6 ND message to exceed the IPv6 minimum MTU. While all ICMPv6 error message types are supported, OAL destinations often include ICMPv6 PTB messages in uNA messages to provide MTU feedback information via the OAL source (see: Section 6.9). Note: ICMPv6 informational messages must not be included and must be ignored if received.¶
OMNI Clients include a Proxy/Server Departure sub-option in RS messages when they associate with a new FHS and/or MAP Proxy/Server and need to send a departure indication to an old FHS and/or MAP Proxy/Server. The Proxy/Server Departure sub-option is formatted as shown below:¶
Sub-Type is set to 15. If multiple instances appear in OMNI options of the same message, the first is processed and all others are ignored.¶
Sub-Length is set to 32.¶
Sub-Option Data contains the 16-octet GUA for the "Old FHS Proxy/Server" followed by a 16-octet GUA for an "Old MAP Proxy/Server. If the Old FHS/MAP is a different node, the corresponding GUA includes the address of the (foreign) Proxy/Server. If the Old FHS/MAP is the local node, the corresponding GUA includes the node's own address. If the FHS/MAP is unspecified, the corresponding GUA instead includes the value "::/128".¶
Since the Sub-Type field is only 5 bits in length, future specifications of major protocol functions may exhaust the remaining Sub-Type values available for assignment. This document therefore defines Sub-Type 30 as an "extension", meaning that the actual sub-option type is determined by examining a 1-octet "Extension-Type" field immediately following the Sub-Length field. The Sub-Type Extension is formatted as shown in Figure 37:¶
Sub-Type is set to 30. If multiple instances appear in OMNI options of the same message all are processed, where each individual extension defines its own policy for processing multiple of that type.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow. The Extension-Type field is always present, and the maximum Extension-Type Body length is limited by the remaining available space in this OMNI option.¶
Extension-Type contains a 1-octet Sub-Type Extension value between 0 and 255.¶
Extension-Type Body contains an (N-1)-octet block with format defined by the given extension specification.¶
Initial Extension-Type values are defined in the following subsections, while remaining Extension-Type values are available for assignment by future specifications which must also define the format of the Extension-Type Body and its processing rules. Extension-Type values 253 and 254 are reserved for experimentation, as recommended in [RFC3692], while value 255 is reserved by IANA.¶
Sub-Type is set to 30.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow. The Extension-Type and Header Type fields are always present, and the Header Option Value is limited by the remaining available space in this OMNI option.¶
Extension-Type is set to 0. Each instance encodes exactly one header option per Section 5.1.1 of [RFC4380], with Ext-Type and Header Type representing the first 2 octets of the option. If multiple instances of the same Header Type appear in OMNI options of the same message the first instance is processed and all others are ignored.¶
Header Type and Header Option Value are coded exactly as specified in Section 5.1.1 of [RFC4380]; the following types are currently defined:¶
0 - Origin Indication (IPv4) - value coded as a UDP port number followed by a 4-octet IPv4 address both in "obfuscated" form per Section 5.1.1 of [RFC4380].¶
1 - Authentication Encapsulation - value coded per Section 5.1.1 of [RFC4380].¶
2 - Origin Indication (IPv6) - value coded as a UDP port number followed by an IP address both in "obfuscated" form per Section 5.1.1 of [RFC4380], except that the IP address is a 16-octet IPv6 address instead of a 4-octet IPv4 address.¶
Header Type values 3 through 252 are available for assignment by future specifications, which must also define the format of the Header Option Value and its processing rules. Header Type values 253 and 254 are reserved for experimentation, as recommended in [RFC3692], and value 255 is reserved by IANA.¶
Sub-Type is set to 30.¶
Sub-Length is set to N that encodes the number of Sub-Option Data octets that follow. The Extension-Type and Trailer Type fields are always present, and the maximum-length Trailer Option Value is limited by the remaining available space in this OMNI option.¶
Extension-Type is set to 1. Each instance encodes exactly one trailer option per Section 4 of [RFC6081]. If multiple instances of the same Trailer Type appear in OMNI options of the same message the first instance is processed and all others ignored.¶
Trailer Type and Trailer Option Value are coded exactly as specified in Section 4 of [RFC6081]; the following Trailer Types are currently defined:¶
0 - Unassigned¶
1 - Nonce Trailer - value coded per Section 4.2 of [RFC6081].¶
2 - Unassigned¶
3 - Alternate Address Trailer (IPv4) - value coded per Section 4.3 of [RFC6081].¶
4 - Neighbor Discovery Option Trailer - value coded per Section 4.4 of [RFC6081].¶
5 - Random Port Trailer - value coded per Section 4.5 of [RFC6081].¶
6 - Alternate Address Trailer (IPv6) - value coded per Section 4.3 of [RFC6081], except that each address is a 16-octet IPv6 address instead of a 4-octet IPv4 address.¶
Trailer Type values 7 through 252 are available for assignment by future specifications, which must also define the format of the Trailer Option Value and its processing rules. Trailer Type values 253 and 254 are reserved for experimentation, as recommended in [RFC3692], while value 255 is reserved by IANA.¶
The multicast address mapping of the native underlay interface applies. The Client mobile router also serves as an IGMP/MLD Proxy for its ENETs and/or hosted applications per [RFC4605].¶
The Client uses Multicast Listener Discovery (MLDv2) [RFC3810] to coordinate with Proxy/Servers, and underlay network elements use MLD snooping [RFC4541]. The Client can also employ multicast routing protocols to coordinate with network-based multicast sources as specified in [I-D.templin-6man-aero3].¶
Since the OMNI link model is NBMA, OMNI links support link-scoped multicast through iterative unicast transmissions to individual multicast group members (i.e., unicast/multicast emulation).¶
The Client's network layer selects the outbound OMNI interface according to SBM considerations when forwarding original IP packets/parcels from local or ENET applications to external correspondents. Each OMNI interface maintains an internal OAL neighbor cache maintained the same as discussed in [RFC4861], but also includes additional state for multilink coordination. Each Client OMNI interface maintains default routes via Proxy/Servers discovered as discussed in Section 13, and may configure more-specific routes discovered through means outside the scope of this specification.¶
For each original IP packet/parcel it forwards, the OMNI interface selects one or more source underlay interfaces based on PBM factors (e.g., traffic attributes, cost, performance, message size, etc.) and one or more target underlay interfaces for the neighbor based on Interface Attributes received in IPv6 ND messages (see: Section 10.2.5). Multilink forwarding may also direct carrier packet replication across multiple underlay interface pairs for increased reliability at the expense of duplication. The set of all Interface Attributes and Traffic Selectors received in IPv6 ND messages determines the multilink forwarding profile for selecting target underlay interfaces.¶
When the OMNI interface forwards an original IP packet/parcel over a selected source underlay interface, it first employs OAL encapsulation and fragmentation as discussed in Section 5, then performs L2 encapsulation as directed by the appropriate AFV. The OMNI interface also performs L2 encapsulation (following OAL encapsulation) when the nearest Proxy/Server is located multiple hops away as discussed in Section 13.2.¶
OMNI interface multilink service designers MUST observe the BCP guidance in Section 15 [RFC3819] in terms of implications for reordering when original IP packets/parcels from the same flow may be spread across multiple underlay interfaces having diverse properties.¶
Clients may connect to multiple independent OMNI links within the same or different OMNI domains to support SBM. The Client configures a separate OMNI interface for each link so that multiple interfaces (e.g., omni0, omni1, omni2, etc.) are exposed to the network layer. Each OMNI interface is configured over a separate set of underlying interfaces and configures one or more OMNI link SRA addresses (see: Section 8); the Client injects the corresponding SRA prefixes into the ENET routing system. Multiple distinct OMNI links can therefore be used to support fault tolerance, load balancing, reliability, etc.¶
Applications in ENETs can use Segment Routing to select the desired OMNI interface based on SBM considerations. The application writes an OMNI link SRA address into the original IP packet/parcel's destination address, and writes the actual destination (along with any additional intermediate hops) into the Segment Routing Header. Standard IP routing directs the packet/parcel to the Client's mobile router entity, where the OMNI link SRA address identifies the correct OMNI interface for next hop forwarding. When the Client receives the packet/parcel, it replaces the IP destination address with the next hop found in the Segment Routing Header and forwards the message via the OMNI interface identified by the SRA address.¶
Note: The Client need not configure its OMNI interface indexes in one-to-one correspondence with the global OMNI Link-IDs configured for OMNI domain administration since the Client's indexes (i.e., omni0, omni1, omni2, etc.) are used only for its own local interface management.¶
After a Proxy/Server has registered an MNP for a Client (see: Section 13), the Proxy/Server will forward all original IP packets/parcels (or carrier packets) destined to an address within the MNP to the Client. The Client will under normal circumstances then forward the resulting original IP packet/parcel to the correct destination within its connected (downstream) ENETs.¶
If at some later time the Client loses state (e.g., after a reboot), it may begin returning original IP packets/parcels (or carrier packets) with destinations corresponding to its MNP to the Proxy/Server as its default router. The Proxy/Server therefore drops any original IP packets/parcels received from the Client with a destination address that corresponds to the Client's MNP (i.e., whether ULA or GUA), and drops any carrier packets with both source and destination address corresponding to the same Client's MNP regardless of their origin.¶
Proxy/Servers support "hairpinning" for packets with SNP source and destination addresses that would convey useful data from a source SNP Client to a target SNP Client both located in the same OMNI link segment. Proxy/Servers support this hairpinning according to [I-D.bctb-6man-rfc6296-bis], however ULA-to-ULA addressing between peer nodes within the same OMNI link segment is preferred whenever possible.¶
Clients engage their FHS Proxy/Servers and the MS by sending OAL encapsulated RS messages with OMNI options under the assumption that one or more Proxy/Server will process the message and respond. The RS message is received by a FHS Proxy/Server, which may in turn forward a proxyed copy to a MAP Proxy/Server located in a local or remote SRT segment if the Client requires MNP service. The MAP Proxy/Server then returns an OAL encapsulated RA message either directly to the Client or via the original FHS Proxy/Server acting as a proxy.¶
To support Client to service coordination, OMNI defines flag bits in the OMNI Neighbor Control sub-option discussed in Section 10.2.5. Clients set or clear the NUD, ARR and/or RPT flags in RS messages as directives to the Mobility Service FHS/MAP Proxy/Servers. Proxy/Servers interpret the flags as follows:¶
When an FHS Proxy/Server forwards or processes an RS with the NUD flag set, it responds directly to future NS Neighbor Unreachability Detection (NUD) messages with the Client as the target by returning NA(NUD) replies; otherwise, it forwards NS(NUD) messages to the Client.¶
When the MAP Proxy/Server receives an RS with the ARR flag set, it responds directly to future NS Address Resolution (AR) messages with the Client as the target by returning NA(AR) replies; otherwise, it forwards NS(AR) messages to the Client.¶
When the MAP Proxy/Server receives an RS with the RPT flag set, it maintains a Report List of recent NS(AR) message sources for the source or target Client and sends uNA messages to all list members if any aspects of the Client's underlay interfaces change.¶
Mobility Service Proxy/Servers function according to the NUD, ARR and RPT flag settings received in the most recent RS message to support dynamic Client updates.¶
Clients and FHS Proxy/Servers include an authentication signature as an OMNI sub-option in their RS/RA exchanges when necessary but always include a valid IPv6 ND message checksum as the final step. FHS and MAP Proxy/Server RS/RA message exchanges over the SRT secured spanning tree instead always include the checksum and omit the authentication signature. Clients and Proxy/Servers use the information included in RS/RA messages to establish NCE state and OMNI link autoconfiguration information as discussed in this section.¶
For each underlay interface, the Client sends RS messages with OMNI options to coordinate with a (potentially) different FHS Proxy/Server for each interface but typically with a limited set of MAP Proxy/Servers (normally only one). All Proxy/Servers are identified by their ULA/GUA SRA addresses and accept carrier packets addressed to their anycast/unicast L2ADDRs; the MAP Proxy/Server may be chosen among any of the Client's FHS Proxy/Servers or may be any other Proxy/Server for the OMNI link. Example L2ADDR discovery methods appear in [RFC5214] and include data link login parameters, name service lookups, static configuration, a DHCP option, a static "hosts" file, etc. In the absence of other information, the Client can resolve the DNS Fully-Qualified Domain Name (FQDN) "linkupnetworks.[domainname]" where "linkupnetworks" is a constant text string and "[domainname]" is a DNS suffix for the OMNI link (e.g., "example.com"). The name resolution will return a set of DNS resource records with the addresses of Proxy/Servers for the local OMNI link segment. When the underlay *NET does not support standard unicast server-based name resolution [RFC1035] the Client can engage a multicast service such as mDNS [RFC6762] within the local OMNI link segment.¶
Each FHS Proxy/Server configures a SNP SRA ULA/GUA address pair for the local OMNI link segment and advertises its L2ADDR(s) for discovery as above. The Client can then manage its own SNP ULA/GUA addresses through DHCPv6 address autoconfiguration exchanges with FHS Proxy/Servers. The FHS Proxy/Servers discovered over multiple of the Client's underlay interfaces may configure the same or different SNP SRA ULAs/GUAs, and the Client's ULA for each underlay interface will fall within the ULA OMNI link segment relative to each FHS Proxy/Server.¶
Clients configure OMNI interfaces that observe the properties discussed in previous sections. The OMNI interface and its underlay interfaces are said to be in either the "UP" or "DOWN" state according to administrative actions in conjunction with the interface connectivity status. An OMNI interface transitions to UP/DOWN through administrative action and/or through underlay interface state transitions. When a first underlay interface transitions to UP, the OMNI interface also transitions to UP. When all underlay interfaces transition to DOWN, the OMNI interface also transitions to DOWN.¶
When a Client OMNI interface transitions to UP, it sends RS messages to register an initial set of underlay interfaces that are also UP and to optionally register/request an MNP. The Client sends additional RS messages to refresh lifetimes and to register/deregister underlay interfaces as they transition to UP or DOWN. The Client's OMNI interface sends initial RS messages over an UP underlay interface with source set to an SNP ULA for the local OMNI link segment if it has one (otherwise with source set to the unspecified address ("::/128") per [RFC4861]) and with destination set to either the SRA GUA of a specific (MAP) Proxy/Server or link-scoped All-Routers multicast (ff02::2) [RFC4291]. The Client includes an OMNI option per Section 10 with a Neighbor Control sub-option with the RS NUD, ARR and RPT flags set or cleared as necessary.¶
Clients in MANETs and open INET deployments also include an OMNI Multilink Vector sub-option with FHS ifIndex set to the ifIndex of its own underlay interface and with LHS ifIndex set to 0 (i.e., the default ifIndex configured by all Proxy/Servers). The Client also sets AFVI to 0, sets Sequence Number to a randomly-chosen 8-octet value and sets the Flow Label in the IPv6 header to 0. The resulting exchange will establish symmetric Identification windows for the Client and Proxy/Server for use in authenticating control messages.¶
The Client next includes an Interface Attributes sub-option for the underlay interface, a DHCPv6 Solicit sub-option with IA_NA and (optionally) IA_PD DHCPv6 options, and with any other necessary OMNI sub-options such as authentication, Proxy/Server Departure, etc. The OMNI interface finally sets or clears the Interface Attributes FMT-Forward and FMT-Mode bits according to its desired FHS Proxy/Server service model as described in Section 10.2.5.¶
The Client next prepares to forward the RS over the underlay interface using OAL encapsulation. The OMNI interface first includes a Nonce and/or Timestamp if necessary, then calculates and sets the authentication signature if necessary followed by the RS message checksum. The OMNI interface next sets the OAL source address to the MLA or HHIT for the outgoing MANET or ALVIF interface and sets the OAL destination to site-scoped All-Routers multicast (ff05::2) [RFC4291], a known Proxy/Server SNP SRA ULA or an anycast address. When L2 encapsulation is used, the Client next includes the discovered FHS Proxy/Server L2ADDR or an anycast address as the L2 destination then fragments if necessary and forwards the resulting carrier packet(s) into the underlay network. Note that the Client does not yet create a NCE, but instead caches the Nonce and/or Timestamp values included in its RS message transmissions to match against any received RA messages.¶
When an FHS Proxy/Server receives the carrier packets containing an RS it performs L2 reassembly if necessary, sets aside the L2 and OAL headers, then verifies the RS checksum/authentication signature. The FHS Proxy/Server then creates/updates a NCE indexed by the RS ULA source address unless unspecified (in which case indexed by the OAL source address). The FHS Proxy/Server then caches the OMNI Interface Attributes and any Traffic Selector sub-options while also caching the L2 (UDP/IP) and OAL source and destination address information. The FHS Proxy/Server then searches for DHCPv6 IA_NA options in the OMNI DHCPv6 sub-option. If IA_NA options are present, the FHS Proxy/Server coordinates with the local DHCPv6 server to either allocate new SNP GUA/ULA pairs or extend the lease lifetime for existing SNP GUA/ULA pairs for the Client. The FHS Proxy/Server next caches the SNP GUA/ULA in the (newly-created) NCE, then caches the RS Neighbor Control NUD flag and Multilink Vector parameters if present (see: Section 10.1) and examines the RS destination address.¶
If the destination matches one of its own addresses and the OMNI DHCPv6 sub-option includes DHCPv6 IA_PD options, the FHS Proxy/Server assumes the MAP role as a default router entry point for injecting the Client's MNP(s) into the OMNI link routing system (i.e., after performing any necessary prefix delegation operations). The FHS/MAP Proxy/Server then caches the RS ARR and RPT flags to determine its role in processing NS(AR) messages and generating uNA messages (see: Section 10.1).¶
The FHS/MAP Proxy/Server then prepares to return an RA message directly to the Client by first populating the Cur Hop Limit, Flags, Router Lifetime, Reachable Time and Retrans Timer fields with values appropriate for the OMNI link. The FHS/MAP Proxy/Server next includes as the first RA message option an OMNI option with a Neighbor Control sub-option and a responsive Multilink Vector sub-option with AFVI set to 0 and with responsive window synchronization information. The FHS/MAP Proxy/Server also includes an authentication sub-option if necessary and a (proxyed) copy of the Client's original Interface Attributes sub-option with its INET-facing interface information written in the FMT, SRT and LHS Proxy/Server GUA/L2ADDR fields. The Proxy/Server also includes a DHCPv6 Reply sub-option with any IA_NA/IA_PD options that have been processed/populated by the DHCPv6 exchange(s).¶
The FHS/MAP Proxy/Server next sets or clears the FMT-Forward and FMT-Mode flags if necessary to convey its capabilities to the Client, noting that it should honor the Client's stated preferences for those parameters if possible or override otherwise. The FMT-Forward/Mode flags thereafter remain fixed unless and until a new RS/RA exchange establishes different values (see: Section 10.2.5 for further discussion). If the FHS/MAP Proxy/Server's Client-facing interface is different than its INET-facing interface, the Proxy/Server next includes a second Interface Attributes sub-option with ifIndex set to '0', with a unicast L2 address for its Client-facing interface in the L2ADDR field and with its SRA ULA in the GUA field.¶
The FHS/MAP Proxy/Server next includes an Origin Indication sub-option that includes the RS L2 source L2ADDR information (see: Section 10.2.17.1), then includes any other necessary OMNI sub-options (either within the same OMNI option or in additional OMNI options). Following the OMNI option(s), the FHS/MAP Proxy/Server next includes any other necessary RA options including 2 PIOs with (A=0; L=0) that include the ULA/GUA SNP prefixes for the segment per [RFC8028], RIOs with more-specific routes per [RFC4191], Nonce and Timestamp options, etc. The FHS/MAP Proxy/Server then sets the RA source address to its own SNP SRA GUA and destination address to the (new) SNP ULA for the Client, then calculates the authentication signature/checksum. The FHS/MAP Proxy/Server finally performs OAL encapsulation while setting the source to its own MLA or HHIT and destination to the OAL source that appeared in the RS, performs L2 encapsulation/fragmentation with L2 source and destination address information reversed from the RS L2 information and returns the resulting carrier packets to the Client over the same underlay interface the RS arrived on.¶
When an FHS Proxy/Server receives an RS with a valid checksum and authentication signature with destination set to link-scoped All-Routers multicast (ff02::2), it can either assume the MAP role itself the same as above or act as a proxy and select the SNP SRA GUA of another Proxy/Server to serve as the MAP. When an FHS Proxy/Server assumes the proxy role or receives an RS with destination set to the SNP SRA GUA of another Proxy/Server, it forwards the message as a proxy. The FHS Proxy/Server creates or updates a NCE for the Client (i.e., based on the RS source address) and caches the OAL source, Neighbor Control, Multilink Vector and Interface Attributes addressing information as above. The FHS Proxy/Server then locally processes any DHCPv6 IA_NA options found in the RS OMNI option and assigns the SNP ULA/GUA address pairs to the Client NCE. The FHS Proxy/Server then writes its own INET-facing FMT, SRT and LHS Proxy/Server GUA/L2ADDR information into the appropriate Interface Attributes sub-option fields (while also setting/clearing FMT-Forward and FMT-Type as above) where the GUA is the Client's SNP GUA address. Next, the FHS Proxy/Server caches the Multilink Vector sub-option and removes it from the RS message, sets the RS source address to the Client's SNP GUA and sets the RS destination to the SNP SRA address of the MAP Proxy/Server. The FHS Proxy/Server then calculates and includes the RS message checksum, sets the OAL source to the Client's SNP GUA and destination to the MAP Proxy/Server SNP SRA GUA, performs L2 encapsulation/fragmentation and sends the resulting carrier packets into the SRT secured spanning tree.¶
When the MAP Proxy/Server receives the carrier packets, it performs L2 reassembly/decapsulation and OAL decapsulation to obtain the proxyed RS, verifies the checksum, then performs DHCPv6 Prefix Delegation (PD) to obtain or update any MNPs for the Client. The MAP Proxy/Server then creates/updates a NCE for the Client's MNP(s) and caches any state (including the ARR and RPT flags, IA_NA addresses, OAL addresses, Interface Attributes information and Traffic Selectors), then finally performs routing protocol injection. The MAP Proxy/Server then returns an RA that echoes the Client's (proxyed) Interface Attributes sub-option and with any RA parameters the same as specified for the FHS/MAP Proxy/Server case above. The MAP Proxy/Server sets the RA source address to its own SNP SRA GUA and destination address to the RS source address (i.e., the Client SNP GUA). The MAP Proxy/Server next calculates the RA message checksum then encapsulates the RA as an OAL packet with source set to the RS message destination (i.e., its own SNP SRA GUA) and destination set to the RS message source (i.e., the Client's SNP GUA). The MAP Proxy/Server finally performs L2 encapsulation/fragmentation and sends the resulting carrier packets into the secured spanning tree.¶
When the FHS Proxy/Server receives the carrier packets it performs L2 reassembly/decapsulation followed by OAL decapsulation to obtain the RA message, verifies checksums then updates the OMNI interface NCE for the Client and creates/updates a NCE for the MAP. The FHS Proxy/Server then sets the P flag in the RA flags field [RFC4389] and proxys the RA by changing the OAL source to its SNP SRA ULA and changing the OAL destination to the source address from the Client's original RS message while also recording any DHCPv6 IA_NA SNP GUA/ULA address pairs as alternate indexes into the Client NCE. The FHS Proxy/Server then includes 2 PIOs with (A=0; L=0) with the SNP ULA/GUA prefixes for the segment per [RFC8028]. The FHS Proxy/Server next includes Neighbor Control parameters responsive to those in the Client's RS and a Multilink Vector sub-option with its responses to its cached initiations from the Client. The FHS Proxy/Server also includes an Interface Attributes sub-option with ifIndex '0' and with its Client-facing interface unicast L2 address if necessary (see above), an Origin Indication sub-option with the Client's cached L2ADDR and an authentication sub-option if necessary. The FHS Proxy/Server finally calculates the authentication signature and RA message checksum, performs L2 encapsulation/fragmentation with addresses taken from the Client's NCE and sends the resulting carrier packets via the same underlay interface over which the RS was received.¶
When the Client receives the carrier packets, it performs L2 reassembly/decapsulation followed by OAL decapsulation to obtain the RA message. The Client next verifies the authentication signature/checksum, then matches the RA with its previously-sent RS by comparing the RS Sequence Number with the RA Acknowledgement Number and also comparing the Nonce and/or Timestamp values. If the values match, the Client then creates/updates OMNI interface NCEs for both the MAP and FHS Proxy/Server and caches the information in the RA message. The Client also caches the RA source address as the MAP Proxy/Server SNP SRA GUA and uses the OAL source address to configure the SNP SRA ULA of this FHS Proxy/Server. The Client next discovers its own SNP ULA by examining the RA destination address, discovers its own SNP GUA by examining the IA_NA DHCPv6 delegated addresses, and discovers the SNP ULA/GUA PIO prefixes for the OMNI link segment per [RFC8028]. If the Client has multiple underlay interfaces, it creates additional FHS Proxy/Server NCEs as necessary when it receives RAs over those interfaces (noting that multiple of the Client's underlay interfaces may be serviced by the same or different FHS Proxy/Servers). The Client finally adds the MAP Proxy/Server SRA GUA to the default router list if necessary.¶
For each underlay interface, the Client next caches the (filled-out) Interface Attributes for its own ifIndex and Origin Indication information that it received in an RA message over that interface so that it can include them in future NS/NA messages to provide neighbors with accurate FMT/SRT/LHS information. (If the message includes an Interface Attributes sub-option with ifIndex '0', the Client also caches the L2ADDR as the underlay network-local unicast address of the FHS Proxy/Server via that underlay interface.) The Client then compares the Origin Indication L2ADDR information with its own underlay interface addresses to determine whether there may be NATs on the path to the FHS Proxy/Server; if the L2ADDR information differs, the Client is behind one or more NATs and must supply the Origin information in IPv6 ND message exchanges with prospective neighbors on the same SRT segment. The Client then caches the Multilink Vector responsive window synchronization parameters for use in future IPv6 ND message exchanges via this FHS Proxy/Server. The Client finally configures default routes and assigns the IPv6 SRA address corresponding to the MNP (e.g., 2001:db8:1:2::) to the OMNI interface.¶
Following the initial exchange, the FHS Proxy/Server MAY later send additional periodic and/or event-driven unsolicited RA messages per [RFC4861]. (The unsolicited RAs may be initiated either by the FHS Proxy/Server itself or by the MAP via the FHS as a proxy.) The Client then continuously manages its underlay interfaces according to their states as follows:¶
When an underlay interface transitions to UP, the Client sends an RS over the underlay interface with an OMNI option with sub-options as specified above.¶
When an underlay interface transitions to DOWN, the Client sends unsolicited NA messages over any UP underlay interface with an OMNI option containing Interface Attributes sub-options for the DOWN underlay interface with ifMetric set to 'ffffffff'. The Client sends isolated unsolicited NAs when reliability is not thought to be a concern (e.g., if redundant transmissions are sent on multiple underlay interfaces), or may instead set the SNR flag in an OMNI Neighbor Control sub-option to trigger an unsolicited NA reply (see: [I-D.templin-6man-aero3]).¶
When the Router Lifetime for the MAP Proxy/Server nears expiration, the Client sends an RS over any underlay interface to receive a fresh RA from the MAP. If no RA messages are received over a first underlay interface (i.e., after retrying), the Client marks the underlay interface as DOWN and should attempt to contact the MAP Proxy/Server via a different underlay interface. If the MAP Proxy/Server is unresponsive over additional underlay interfaces, the Client sends an RS message with destination set to the SNP SRA GUA of another Proxy/Server which will then assume the MAP role.¶
When all of a Client's underlay interfaces have transitioned to DOWN (or if a prefix delegation lifetime expires), the MAP Proxy/Server withdraws the MNP the same as if it had received a message with a release indication.¶
The Client is responsible for retrying each RS exchange up to MAX_RTR_SOLICITATIONS times separated by RTR_SOLICITATION_INTERVAL seconds until an RA is received. If no RA is received over an UP underlay interface (i.e., even after attempting to contact alternate Proxy/Servers), the Client can either declare this underlay interface as DOWN or continue to use the interface to support any peer-to-peer local communications with peers located in the same *NET. When changing to a new FHS/MAP Proxy/Server, the Client also includes a Proxy/Server Departure OMNI sub-option in new RS messages; the (new) FHS Proxy/Server will in turn send uNA messages to the old FHS and/or MAP Proxy/Server to announce the Client's departure as discussed in [I-D.templin-6man-aero3].¶
The network layer sees the OMNI interface as an ordinary IPv6 interface. Therefore, when the network layer sends an RS message the OMNI interface returns an internally-generated RA message as though the message originated from an IPv6 router. The internally-generated RA message contains configuration information consistent with the information received from the RAs generated by the MAP Proxy/Server. Whether the OMNI interface IPv6 ND messaging process is initiated from the receipt of an RS message from the network layer or independently of the network layer is an implementation matter. Some implementations may elect to defer the OMNI interface internal RS/RA messaging process until an RS is received from the network layer, while others may elect to initiate the process independently. Still other deployments may elect to administratively disable network layer RS/RA messaging over the OMNI interface, since the messages are not required to drive the OMNI interface internal RS/RA process. (Note that this same logic applies to IPv4 implementations that employ "ICMP Router Discovery" [RFC1256].)¶
Note: The Router Lifetime value in RA messages indicates the time before which the Client must send another RS message over this underlay interface (e.g., 600 seconds), however that timescale may be significantly longer than the lifetime the MS has committed to retain the prefix registration (e.g., REACHABLE_TIME seconds). Proxy/Servers are therefore responsible for keeping MS state alive on a shorter timescale than the Client may be required to do on its own behalf.¶
Note: On certain multicast-capable underlay interfaces, Clients should send periodic unsolicited multicast NA messages and Proxy/Servers should send periodic unsolicited multicast RA messages as "beacons" that can be heard by other nodes on the link. If a node fails to receive a beacon after a timeout value specific to the link, it can initiate Neighbor Unreachability Detection (NUD) exchanges to test reachability.¶
Note: Although the Client's FHS Proxy/Server is a first-hop segment node from its own perspective, the Client stores the Proxy/Server's FMT/SRT/GUA/L2ADDR as last-hop segment (LHS) information to supply to neighbors. This allows both the Client and MAP Proxy/Server to supply the information to neighbors that will perceive it as LHS information on the return path to the Client.¶
Note: The MAP Proxy/Server injects Client MNPs into the OMNI link routing system by simply creating a route-to-interface forwarding table entry for MNP::/N via the OMNI interface. The dynamic routing protocol will notice the new entry and propagate the route to its peers. If the MAP receives additional RS messages, it need not re-create the forwarding table entry (nor disturb the dynamic routing protocol) if an entry is already present. If the MAP ceases to receive RS messages from any of the Client's interfaces, it removes the Client MNP(s) from the forwarding table (i.e., after a short delay) which also results in their removal from the routing system.¶
Note: If the Client's initial RS message includes an anycast L2 destination address, the FHS Proxy/Server returns the solicited RA using the same anycast address as the L2 source while including an Interface Attributes sub-option with ifIndex '0' and its true unicast address in the L2ADDR. When the Client sends additional RS messages, it includes this FHS Proxy/Server unicast address as the L2 destination and the FHS Proxy/Server returns the solicited RA using the same unicast address as the L2 source. This will ensure that RS/RA exchanges are not impeded by any NATs on the path while avoiding long-term exposure of messages that use an anycast address as the source.¶
Note: The Origin Indication sub-option is included only by the FHS Proxy/Server and not by the MAP (unless the MAP is also serving as an FHS).¶
Note: Clients should set the NUD, ARR and RPT flags consistently in successive RS messages and only change those settings when an FHS/MAP Proxy/Server service profile update is necessary.¶
Note: Although the Client adds the MAP Proxy/Server SNP SRA GUA to the default router list, it also caches the ULAs of the FHS Proxy/Servers on the path to the MAP over each underlying interface. When the Client needs to send an original IP packet/parcel to a default router, it engages OAL encapsulation/fragmentation while using a destination ULA corresponding to the selected interface which directs the packet to an FHS Proxy/Server for that interface. The FHS Proxy/Server then performs L2 encapsulation/fragmentation and forwards the resulting carrier packets without disturbing the MAP.¶
The RS/RA exchanges discussed above observe the principles specified in Section 6.7. Window synchronization is conducted between the Client and each FHS Proxy/Server used to contact the MAP Proxy/Server, i.e., and not between the Client and the MAP. This is due to the fact that the MAP Proxy/Server is responsible only for forwarding messages via the secured spanning tree to FHS Proxy/Servers, and is not responsible for forwarding messages directly to the Client.¶
When a Client sends an RS to perform window synchronization via a new FHS Proxy/Server, it includes an OMNI Multilink Vector sub-option with window synchronization parameters with FHS ifIndex set to its own interface index, with LHS ifIndex set to 0, with AFVI set to 0, with the SYN flag set and ACK flag clear, and with an initial Sequence Number. The Client finally includes an Interface Attributes sub-option then performs OAL encapsulation and L2 encapsulation/fragmentation then sends the resulting carrier packets to the FHS Proxy/Server. When the FHS Proxy/Server receives the carrier packets, it performs L2 reassembly/decapsulation, then extracts the RS message and caches the Multilink Vector parameters. In the process, the FHS Proxy/Server removes the Multilink Vector sub-option itself, since the path to the MAP Proxy/Server is not included in window synchronization.¶
The FHS Proxy/Server then performs L2 encapsulation/fragmentation and sends the resulting carrier packets via the secured spanning tree to the MAP Proxy/Server, which updates the Client's Interface Attributes and returns a unicast RA message. The MAP Proxy/Server performs OAL encapsulation followed by L2 encapsulation/fragmentation and sends the carrier packets via the secured spanning tree to the FHS Proxy/Server. The FHS Proxy/Server then proxys the message as discussed in the previous section and includes a Multilink Vector sub-option with responsive window synchronization information. The FHS Proxy/Server then forwards the message to the Client via OAL encapsulation which updates its window synchronization information for the FHS Proxy/Server as necessary.¶
Following the initial RS/RA-driven window synchronization, the Client can re-assert new windows with specific FHS Proxy/Servers by performing RS/RA exchanges between its own ULAs and the ULAs of the FHS Proxy/Servers at any time without having to disturb the MAP. When the Client also needs to refresh MAP state, it can set the RS destination address to the MAP SNP SRA address.¶
This window synchronization is necessary only for MANET and INET Clients that must include authentication signatures with their IPv6 ND messages; Clients in secured ANETs can omit window synchronization. When Client-to-Proxy/Server window synchronization is used, subsequent IPv6 ND NS/NA messages exchanged between peers include IPv6 Extended Fragment Headers in the OAL encapsulations with in-window Identification values to support message authentication. No header compression state is maintained by OAL intermediate systems, which only maintain state for per-flow data plane windows.¶
On some *NETs, a Client may be located multiple intermediate OAL hops away from the nearest OMNI link Proxy/Server. Clients in multihop networks perform route discovery through the application of an adaptation layer routing protocol (e.g., a MANET routing protocol over omnidirectional wireless interfaces, etc.) then apply corresponding forwarding entries to the OMNI interface. Example routing protocols optimized for MANET operations include OSPFv3 [RFC5340] with MANET Designated Router (OSPF-MDR) extensions [RFC5614], OLSRv2 [RFC7181], AODVv2 [I-D.perkins-manet-aodvv2] and others. Clients employ the routing protocol according to the link model found in [RFC5889] and subnet model articulated in [RFC5942]. For unique identification within the MANET, Clients use an MLA or HHIT as a Router ID.¶
A Client located potentially multiple OAL hops away from the nearest Proxy/Server prepares an RS message, sets the source address to its ULA or unspecified ("::/128"), and sets the destination to link-scoped All-Routers multicast (ff02::2) or the SNP SRA ULA of a Proxy/Server the same as discussed above. The OMNI interface then employs OAL encapsulation, sets the OAL source address to its MLA/HHIT and sets the OAL destination to either the site- scoped All-Routers multicast address (ff05::2) or the OMNI IPv6 anycast address.¶
For IPv6-enabled *NETs where the underlay interface observes the MANET properties discussed above, the Client injects the MLA/HHIT into the IPv6 multihop routing system and forwards the message without further encapsulation. Otherwise, the Client encapsulates the message in UDP/IPv6 L2 headers, sets the source to the underlay interface IPv6 address and sets the destination to the discovered L2 unicast or anycast address of a Proxy/Server. The Client then forwards the message into the IPv6 multihop routing system which conveys it to the nearest Proxy/Server. If the nearest Proxy/Server is too busy, it should forward (without Proxying) the OAL-encapsulated RS to another nearby Proxy/Server connected to the same IPv6 (multihop) network.¶
For IPv4-only *NETs, the Client encapsulates the RS message in UDP/IPv4 L2 headers, sets the source to the underlay interface IPv4 address and sets the destination to the discovered L2 unicast address of a Proxy/Server or the OMNI IPv4 anycast address. The Client then forwards the message into the IPv4 multihop routing system which conveys it to the nearest Proxy/Server that advertises the corresponding IPv4 prefix. If the nearest Proxy/Server is too busy, it should forward (without Proxying) the OAL-encapsulated RS to another nearby Proxy/Server connected to the same IPv4 (multihop) network that configures the OMNI IPv6 anycast address. (In environments where reciprocal RS forwarding cannot be supported, the first Proxy/Server should instead return an RA based on its own MSP(s).)¶
When an OAL intermediate node that participates in the routing protocol receives the encapsulated RS, it forwards the message according to its OAL IPv6 forwarding table (note that an OAL intermediate system could be a fixed infrastructure element such as a roadside unit or another MANET/VANET Client). This process repeats iteratively until the RS message is received by a penultimate OAL hop within single-hop communications range of a Proxy/Server, which forwards the message to the Proxy/Server final hop.¶
When a Proxy/Server that configures the OMNI IPv6 anycast destination address receives the message, it decapsulates the RS and assumes either the MAP or FHS role (in which case, it may forward the RS to a candidate MAP). The MAP/FHS Proxy/Server then prepares an RA message using the same addressing disciplines as discussed in Section 13 and forwards the RA either to the FHS Proxy/Server or directly to the Client.¶
When the MAP or FHS Proxy/Server forwards the RA to the Client, it encapsulates the message in L2 encapsulation headers (if necessary) The Proxy/Server then forwards the message to an OAL node within communications range, which forwards the message according to the next OAL hop by consulting its OAL IPv6 forwarding tables. The multihop forwarding process within the *NET continues repetitively until the message arrives at the original Client, which decapsulates the message and performs autoconfiguration the same as if it had received the RA directly from a Proxy/Server on the same physical link. The Client then injects the delegated ULA and any MNP SRA GUAs into the IPv6 multihop routing system.¶
Note: When the RS message includes anycast OAL and/or L2 encapsulation destinations, the FHS Proxy/Server must use the same anycast addresses as the OAL and/or L2 encapsulation sources to support forwarding of the RA message plus any initial data messages. The FHS Proxy/Server then sends the resulting carrier packets over any NATs on the path. When the Client receives the RA, it will discover the FHS Proxy/Server unicast ULAs and/or L2 encapsulation addresses and can send future carrier packets using the unicast (instead of anycast) addresses to populate NAT state in the forward path. (If the Client does not have immediate data to send to the FHS Proxy/Server, it can instead send an OAL "bubble" - see Section 6.11.) After the Client begins using unicast OAL/L2 encapsulation addresses in this way, the FHS Proxy/Server should also begin using the same unicast addresses in the reverse direction.¶
Note: When an OMNI interface configures an MLA/HHIT, any nodes that forward an encapsulated RS message with the MLA/HHIT as the OAL source must not consider the message as being specific to a particular OMNI link segment. MLAs/HHITs can therefore also serve as the source and destination addresses of unencapsulated IPv6 data communications within the local routing region, and if the MLAs/HHITs are injected into the local network routing protocol their prefix length must be set to 128.¶
Note: intermediate forwarding systems often coordinate multi-hop relaying using the same underlay interface in both the inbound and outbound directions, i.e. as opposed to different underlay interfaces. The final forwarding node within range of a Proxy/Server could use the same or a different underlay interface to exchange carrier packets with the Proxy/Server, but may not be well positioned to perform multilink selections over multiple underlay interfaces on behalf of multihop dependent peers.¶
When a Client requires SNP ULA/GUA delegations via a specific Proxy/Server (or, when the Client requires MNP delegations for the OMNI link), it invokes the DHCPv6 service [I-D.ietf-dhc-rfc8415bis] in conjunction with its OMNI RS/RA message exchanges.¶
When a Client requires the MS to delegate PA ULA/GUA pairs or PI MNPs, it sends an RS message to a FHS Proxy/Server. If the Client requires one or more address or MNP delegations, it includes a DHCPv6 Message sub-option containing a Client Identifier, one or more IA_NA/IA_PD options and a Rapid Commit option then sets the 'msg-type' field to "Solicit" and includes a 3-octet 'transaction-id'. The Client then sets the RS destination to link-scoped All-Routers multicast (ff02::2) and sends the message using OAL encapsulation and fragmentation if necessary as discussed above.¶
When the FHS/MAP Proxy/Server receives the RS message, it performs OAL reassembly if necessary. Next, if the OMNI option includes a DHCPv6 message sub-option, the FHS/MAP Proxy/Server acts as a "Proxy DHCPv6 Client" in a message exchange with the locally-resident DHCPv6 server. The FHS/MAP Proxy/Server then sends the DHCPv6 message to the DHCPv6 Server, which delegates SNP ULA/GUA pairs or MNPs and returns a DHCPv6 Reply message with autoconfiguration parameters.¶
When the FHS Proxy/Server receives a DHCPv6 Reply with delegated addresses, it records the delegated SNP ULA/GUA pairs in the NCE for the Client, then forwards the RS message to the MAP Proxy/Server for prefix delegation if necessary; otherwise, it returns an immediate RA message to the Client.¶
When the MAP Proxy/Server receives a DHCPv6 Reply with delegated prefixes, it creates OMNI interface MNP forwarding table entries (i.e., to prompt the dynamic routing protocol). The MAP Proxy/Server then sends an RA back to the FHS Proxy/Server with the DHCPv6 Reply message included in an OMNI DHCPv6 message sub-option, and the FHS Proxy/Server returns the RA to the Client.¶
Clients can provide an OMNI link ingress point for other nodes on their (downstream) ENETs that also act as Clients. When Client A has already coordinated with an (upstream) (M)ANET/INET Proxy/Server, Client B on an ENET serviced by Client A can send OAL-encapsulated RS messages with addresses set the same as specified in Section 13.2. When Client A receives the RS message, it infers from the OAL encapsulation that Client B is seeking to establish itself as a Client instead of just a simple ENET Host.¶
Client A then returns an RA message the same as a Proxy/Server would do as specified in Section 13.2 except that it instead uses its own MNP SRA GUA as the RA and OAL source addresses and performs (recursive) DHCPv6 Prefix Delegation. The MNP delegation in the RA message must be a sub-MNP from the MNP delegated to Client A. For example, if Client A receives the MNP 2001:db8:1000::/48 it can provide a sub-delegation such as 2001:db8:1000:2000::/56 to Client B. Client B can in turn sub-delegate 2001:db8:1000:2000::/56 to its own ENET(s), where there may be a further prospective Client C that would in turn request OMNI link services via Client B.¶
To support this Client-to-Client chaining, Clients send IPv6 ND messages addressed to link-scoped All-Routers multicast (ff02::2) via their *NET (i.e., upstream) interfaces, but respond to IPv6 ND messages addressed to link-scoped All-Routers multicast over their ENET (i.e., downstream) networks where there may be further prospective Clients wishing to join the chain. The ENET of the upstream Client is therefore seen as an ANET by downstream Clients, and the upstream Client is seen as a Proxy/Server by downstream Clients.¶
If the *NET link model is multiple access, the FHS Proxy/Server is responsible for assuring that address duplication cannot corrupt the neighbor caches of other nodes on the link through the use of the DHCPv6 address delegation service. When the Client sends an RS message on a multiple access *NET, the Proxy/Server verifies that the Client is authorized to use the address and responds with an RA (or forwards the RS to the MAP) only if the Client is authorized.¶
After verifying Client authorization and returning an RA, the Proxy/Server MAY return IPv6 ND Redirect messages in response to subsequent data plane packet transmissions to direct Clients located on the same *NET to exchange OAL packets directly without transiting the Proxy/Server. In that case, the Clients can exchange OAL packets according to their unicast L2 addresses discovered from the Redirect message instead of using the dogleg path through the Proxy/Server. In some *NETs, however, such direct communications may be undesirable and continued use of the dogleg path through the Proxy/Server may provide better performance. In that case, the Proxy/Server can refrain from sending Redirects, and/or Clients can ignore them.¶
*NETs SHOULD deploy Proxy/Servers in Virtual Router Redundancy Protocol (VRRP) [RFC5798] configurations so that service continuity is maintained even if one or more Proxy/Servers fail. Using VRRP, the Client is unaware which of the (redundant) FHS Proxy/Servers is currently providing service, and any service discontinuity will be limited to the failover time supported by VRRP. Widely deployed public domain implementations of VRRP are available.¶
Proxy/Servers SHOULD use high availability clustering services so that multiple redundant systems can provide coordinated response to failures. As with VRRP, widely deployed public domain implementations of high availability clustering services are available. Note that special-purpose and expensive dedicated hardware is not necessary, and public domain implementations can be used even between lightweight virtual machines in cloud deployments.¶
In environments where fast recovery from Proxy/Server failure is essential, FHS Proxy/Servers SHOULD use proactive Neighbor Unreachability Detection (NUD) in a manner that parallels Bidirectional Forwarding Detection (BFD) [RFC5880] to track MAP Proxy/Server reachability. FHS Proxy/Servers can then quickly detect and react to failures so that cached information is re-established through alternate paths. Proactive NUD control messaging is carried only over well-connected ground domain networks (i.e., and not low-end links such as aeronautical radios) and can therefore be tuned for rapid response.¶
FHS Proxy/Servers perform proactive NUD for MAP Proxy/Servers for which there are currently active Clients. If a MAP Proxy/Server fails, the FHS Proxy/Server can quickly inform Clients of the outage by sending multicast RA messages. The FHS Proxy/Server sends RA messages to Clients with source set to the ULA of the MAP, with destination address set to link-scoped All-Nodes multicast (ff02::1) [RFC4291] and with Router Lifetime set to 0.¶
The FHS Proxy/Server SHOULD send MAX_FINAL_RTR_ADVERTISEMENTS RA messages separated by small delays [RFC4861]. Any Clients that have been using the (now defunct) MAP Proxy/Server will receive the RA messages.¶
When a Client connects to a *NET link for the first time, it sends an RS message with an OMNI option. If the first hop router recognizes the option, it responds according to the appropriate FHS/MAP Proxy/Server role resulting in an RA message with an OMNI option returned to the Client. The Client then engages this FHS Proxy/Sever according to the OMNI link model specified above. If the first hop router is a legacy IPv6 router, however, it instead returns an RA message with no OMNI option and with an ordinary unicast source LLA as specified in [RFC4861]. In that case, the Client engages the *NET according to the legacy IPv6 link model and without the OMNI extensions specified in this document.¶
If the *NET link model is multiple access, there must be assurance that address duplication cannot corrupt the neighbor caches of other nodes on the link. When the Client sends an RS message on a multiple access *NET link with an OMNI option, first hop routers that recognize the option ensure that the Client is authorized to use the address and return an RA with a non-zero Router Lifetime only if the Client is authorized. First hop routers that do not recognize the OMNI option instead return an RA that makes no statement about the Client's authorization to use the source address. In that case, the Client should perform Duplicate Address Detection to ensure that it does not interfere with other nodes on the link.¶
An alternative approach for multiple access *NET links to ensure isolation for Client-Proxy/Server communications is through link layer address mappings as discussed in Appendix E. This arrangement imparts a (virtual) point-to-point link model over the (physical) multiple access link.¶
Client OMNI interfaces configured over IPv6-enabled underlay interfaces on an open Internetwork without an OMNI-aware first-hop router receive IPv6 RA messages with no OMNI options, while OMNI interfaces configured over IPv4-only underlay interfaces receive no IPv6 RA messages at all (but may receive IPv4 RA messages per [RFC1256]). Client OMNI interfaces that receive RA messages with OMNI options configure addresses, on-link prefixes, etc. on the underlay interface that received the RA according to standard IPv6 ND and address resolution conventions [RFC4861] [RFC4862]. Client OMNI interfaces configured over IPv4-only underlay interfaces configure IPv4 address information on the underlay interfaces using mechanisms such as DHCPv4 [RFC2131].¶
Client OMNI interfaces configured over underlay interfaces connected to open Internetworks can apply lower layer security services such as VPNs (e.g., IPsec tunnels) to connect to a Proxy/Server, or can establish a secured direct point-to-point link to the Proxy/Server through some other means (see Section 4). In environments where lower layer security may be impractical or undesirable, Client OMNI interfaces can instead send IPv6 ND messages with OMNI options that include authentication signatures.¶
OMNI interfaces use UDP/IP as L2 encapsulation headers for transmission over open Internetworks with UDP service port number 8060 for both IPv4 and IPv6 underlay interfaces. The OMNI interface submits original IP packets/parcels for OAL encapsulation, then encapsulates the resulting OAL fragments in UDP/IP L2 headers to form carrier packets. (The first 4 bits following the UDP header determine whether the OAL headers are uncompressed/compressed as discussed in Section 6.5.) The OMNI interface sets the UDP length to the encapsulated OAL fragment length and sets the IP length to an appropriate value at least as large as the UDP datagram.¶
When necessary, sources include an OMNI option with an authentication sub-option in IPv6 ND messages. The source can employ a simple Hashed Message Authentication Code (HMAC) as specified in [RFC2104][RFC6234], EdDSA [RFC8032], or a message-based authentication service such as HIP [RFC7401], QUIC-TLS [RFC9000][RFC9001], etc., by using the IPv6 ND message OMNI option as a "shipping container". Before calculating the authentication signature, the source fully populates any necessary OMNI sub-options as well as any ordinary IPv6 ND options as necessary.¶
The source then sets both the IPv6 ND message Checksum and authentication signature fields to 0 and calculates the authentication signature over the full length of the IPv6 ND message beginning after the IPv6 ND message checksum field and extending over the length of the message. (If the IPv6 ND message is part of an OAL super-packet, the source instead continues to calculate the authentication signature over the entire length of the super-packet.) The source next writes the authentication signature into the appropriate sub-option field, calculates and writes the message checksum, then forwards the message.¶
After establishing a secured underlay link or preparing for UDP/IP encapsulation, OMNI interfaces send RS/RA messages for Client-Proxy/Server coordination (see: Section 13) and NS/NA messages for multilink forwarding, route optimization, and mobility management (see: [I-D.templin-6man-aero3]). These control plane messages must be authenticated while other control and data plane messages are delivered the same as for ordinary best effort traffic with source address and/or Identification window-based data origin verification. Transport and higher layer protocol sessions over OMNI interfaces that connect over open Internetworks without an explicit underlay link security services should therefore employ security at their layers to ensure authentication, integrity and/or confidentiality.¶
Clients should avoid using INET Proxy/Servers as general-purpose routers for steady streams of carrier packets that do not require authentication. Clients should therefore perform route optimization to coordinate with other INET nodes that can provide forwarding services (or preferably coordinate with peer Clients directly) instead of burdening the Proxy/Server. Procedures for coordinating with peer Clients and discovering INET nodes that can provide better forwarding services are discussed in [I-D.templin-6man-aero3].¶
Clients that attempt to contact peers over INET underlay interfaces often encounter NATs in the path. OMNI interfaces accommodate NAT traversal using UDP/IP encapsulation and the mechanisms discussed in [I-D.templin-6man-aero3]. FHS Proxy/Servers include Origin Indications in RA messages to allow Clients to detect the presence of NATs.¶
Note: Following the initial IPv6 ND message exchange, OMNI interfaces configured over INET underlay interfaces maintain neighbor relationships by transmitting periodic IPv6 ND messages with OMNI options that include authentication signatures. Other authentication services that use their own IPv6 ND option types such as [RFC3971] and [RFC8928] can also be used in addition to any OMNI authentication services.¶
Note: OMNI interfaces configured over INET underlay interfaces should employ the Identification window synchronization mechanisms specified in Section 6.7 in order to exclude spurious carrier packets that might otherwise clutter the reassembly cache. This is especially important in environments where carrier packet spoofing and/or corruption is a threat.¶
Note: NATs may be present on the path from a Client to its FHS Proxy/Server, but never on the path from the FHS Proxy/Server to the MAP where only INET and/or spanning tree hops occur. Therefore, the FHS Proxy/Server does not communicate Client origin information to the MAP where it would serve no purpose.¶
In some use cases, it is desirable, beneficial and efficient for the Client to receive a constant MNP that travels with the Client wherever it moves. For example, this would allow air traffic controllers to easily track aircraft, etc. In other cases, however (e.g., intelligent transportation systems), the Client may be willing to sacrifice a modicum of efficiency in order to have time-varying MNPs that can be changed occasionally to defeat adversarial tracking.¶
The prefix delegation services discussed in Section 13.3 allows Clients that desire time-varying MNPs to obtain short-lived prefixes to send RS messages with an OMNI option with DHCPv6 IA_PD sub-options. The Client would then be obligated to renumber its internal networks whenever its MNPs change. This should not present a challenge for Clients with automated network renumbering services, but may disrupt persistent sessions that would prefer to use a constant address.¶
An OAL destination or intermediate system may need to return ICMPv6-like error messages (e.g., Destination Unreachable, Packet Too Big, Time Exceeded, etc.) [RFC4443] to an OAL source. Since ICMPv6 error messages do not themselves include authentication codes, OAL nodes can instead return error messages as an OMNI ICMPv6 Error sub-option in a secured IPv6 ND uNA message.¶
The following IANA actions are requested in accordance with [RFC8126] and [RFC8726]:¶
The IANA is instructed to allocate an Internet Protocol number TBD1 from the 'protocol numbers' registry for the Overlay Multilink Network Interface (OMNI) protocol. Guidance is found in [RFC5237] (registration procedure is IESG Approval or Standards Action).¶
During final publication stages, the IESG will be requested to procure an IEEE EtherType value TBD2 for OMNI according to the statement found at https://www.ietf.org/about/groups/iesg/statements/ethertypes/.¶
Following this procurement, the IANA is instructed to register the value TBD2 in the 'ieee-802-numbers' registry for Overlay Multilink Network Interface (OMNI) encapsulation on Ethernet networks. Guidance is found in [RFC7042] (registration procedure is Expert Review).¶
The IANA is instructed to assign TBD3/N as an "OMNI IPv4 anycast" address/prefix in the "IPv4 Special-Purpose Address" registry in a similar fashion as for [RFC3068]. The assignment also automatically provides the basis for an "OMNI IPv6 anycast" address configured as 2002:TBD3::. The IANA is requested assist the author's efforts to obtain a TBD3/N public IPv4 prefix, whether through an RIR allocation, a delegation from IANA's "IPv4 Recovered Address Space" registry or through an unspecified third party donation.¶
The IANA is instructed to allocate an official Type number TBD4 from the "IPv6 Neighbor Discovery Option Formats" registry for the OMNI option (registration procedure is RFC required).¶
The IANA is instructed to allocate one Ethernet unicast address TBD5 (suggested value '00-52-14') in the 'ethernet-numbers' registry under "IANA Unicast 48-bit MAC Addresses" (registration procedure is Expert Review). The registration should appear as follows:¶
The IANA is instructed to assign new Code values in the "ICMPv6 Code Fields: Type 2 - Packet Too Big" table in the 'icmpv6-parameters' registry (registration procedure is Standards Action or IESG Approval). The registry entries should appear as follows:¶
The IANA is instructed to assign a new Type number TBD6 in the 'icmp-parameters' registry "ICMP Type Numbers" table (registration procedures IESG Approval or Standards Action). The entry should set "Type" to TBD6, "Name" to "Packet Too Big (PTB)" and "Reference" to [RFCXXXX] (i.e., this document).¶
The IANA is further instructed to create a new table titled: "Type TBD6 - Packet Too Big (PTB)" in the 'icmp-parameters' Code tables, with registration procedures IESG Approval or Standards Action. The table should have the following initial format:¶
The OMNI option defines a 5-bit Sub-Type field, for which IANA is instructed to create and maintain a new registry entitled "OMNI Option Sub-Type Values". Initial values are given below (registration procedure is RFC required):¶
The OMNI Node Identification sub-option (see: Section 10.2.3) contains an 8-bit ID-Type field, for which IANA is instructed to create and maintain a new registry entitled "OMNI Node Identification ID-Type Values". Initial values are given below (registration procedure is RFC required):¶
The OMNI Geo Coordinates sub-option (see: Section 10.2.9) contains an 8-bit Type field, for which IANA is instructed to create and maintain a new registry entitled "OMNI Geo Coordinates Type Values". Initial values are given below (registration procedure is RFC required):¶
The OMNI option defines an 8-bit Extension-Type field for Sub-Type 30 (Sub-Type Extension), for which IANA is instructed to create and maintain a new registry entitled "OMNI Option Sub-Type Extension Values". Initial values are given below (registration procedure is RFC required):¶
The OMNI Sub-Type Extension "RFC4380 UDP/IP Header Option" defines an 8-bit Header Type field, for which IANA is instructed to create and maintain a new registry entitled "OMNI RFC4380 UDP/IP Header Option". Initial registry values are given below (registration procedure is RFC required):¶
The OMNI Sub-Type Extension for "RFC6081 UDP/IP Trailer Option" defines an 8-bit Trailer Type field, for which IANA is instructed to create and maintain a new registry entitled "OMNI RFC6081 UDP/IP Trailer Option". Initial registry values are given below (registration procedure is RFC required):¶
The IANA "ICMPv6 Parameters - Trust Anchor Option (Type 15) Name Field" registry includes Type values for common authentication signature values that could be used for SEcure Neighbor Discovery (SEND). IANA is instructed to assign the value TBD7 for "Edwards- Curve Digital Signature Algorithm (EdDSA) [RFC8032] in this registry with reference set to [RFCXXXX] (i.e., this document).¶
The IANA has assigned the UDP port number "8060" for an earlier experimental version of AERO [RFC6706]. This document reclaims the UDP port number "8060" for 'aero' as the service port for UDP/IP encapsulation. (Note that, although [RFC6706] is not widely implemented or deployed, any messages coded to that specification can be easily distinguished and ignored since they include an invalid ICMPv6 message type number '0'.) The IANA is therefore instructed to update the reference for UDP port number "8060" from "RFC6706" to "RFCXXXX" (i.e., this document) while retaining the existing name 'aero'.¶
The IANA has assigned a 4-octet Private Enterprise Number (PEN) code "45282" in the "enterprise-numbers" registry. This document is the normative reference for using this code in DHCP Unique IDentifiers based on Enterprise Numbers ("DUID-EN for OMNI Interfaces") (see: Section 9). The IANA is therefore instructed to change the enterprise designation for PEN code "45282" from "LinkUp Networks" to "Overlay Multilink Network Interface (OMNI)".¶
The IANA has assigned the ifType code "301 - omni - Overlay Multilink Network Interface (OMNI)" in accordance with Section 6 of [RFC8892]. The registration appears under the IANA "Structure of Management Information (SMI) Numbers (MIB Module Registrations) - Interface Types (ifType)" registry.¶
No further IANA actions are required.¶
Security considerations for IPv4 [RFC0791], IPv6 [RFC8200] and IPv6 Neighbor Discovery [RFC4861] apply. OMNI interface IPv6 ND messages SHOULD include Nonce and Timestamp options [RFC3971] when transaction confirmation and/or time synchronization is needed.¶
OMNI interfaces configured over secured ANET/ENET interfaces inherit the physical and/or link layer security properties (i.e., "protected spectrum") of the connected networks. OMNI interfaces configured over open *NET interfaces can use symmetric securing services such as IPsec tunnels [RFC4301] or can by some other means establish a direct point-to-point link secured at lower layers. When lower layer security may be impractical or undesirable, however, control message integrity and authorization services such as those specified in [RFC7401], [RFC4380], [RFC6234], [RFC8032], [RFC9000], etc. must be employed.¶
OMNI link mobility services MUST support strong network layer authentication for control plane messages and forwarding path integrity for data plane messages. In particular, the AERO service [I-D.templin-6man-aero3] constructs a secured spanning tree with Proxy/Servers as leaf nodes and secures the spanning tree links with network layer security services based on IPsec [RFC4301] with IKEv2 [RFC7296]. (Note that direct point-to-point links secured at lower layers can also be used instead of or in addition to network layer security.) These network (and/or lower-layer) services together provide connectionless integrity and data origin authentication with optional protection against replays.¶
Control plane messages that affect the routing system or neighbor state are constrained to travel only over secured spanning tree paths and are therefore protected by network (and/or lower-layer) security. Other control and data plane messages can travel over unsecured route optimized paths that do not strictly follow the spanning tree, therefore end-to-end sessions should employ transport or higher layer security services (e.g., TLS/SSL [RFC8446], DTLS [RFC6347], etc.). Additionally, the OAL Identification value can provide a first level of data origin authentication to mitigate off-path spoofing.¶
Identity-based key verification infrastructure services such as iPSK may be necessary for verifying the identities claimed by Clients. This requirement should be harmonized with the manner in which identifiers such as (H)HITs are attested in a given operational environment.¶
Security considerations for specific access network interface types are covered under the corresponding IP-over-(foo) specification (e.g., [RFC2464], [RFC2492], etc.).¶
Security considerations for IPv6 fragmentation and reassembly are discussed in Section 6.15. In environments where spoofing is considered a threat, OMNI nodes SHOULD employ Identification window synchronization and OAL destinations SHOULD configure an (end-system-based) firewall.¶
AERO/OMNI Release-3.2 was tagged on March 30, 2021, and was subject to internal testing. The implementation is not planned for public release.¶
A new implementation architecture based on a clean-slate has been developed and will incorporate updated aspects of the AERO/OMNI specs, with the goal of producing a reference implementation for future release.¶
This document suggests that the following could be updated through future IETF initiatives:¶
Updates can be through, e.g., standards action, the errata process, etc. as appropriate.¶
The first version of this document was prepared per the consensus decision at the 7th Conference of the International Civil Aviation Organization (ICAO) Working Group-I Mobility Subgroup on March 22, 2019. Consensus to take the document forward to the IETF was reached at the 9th Conference of the Mobility Subgroup on November 22, 2019. Attendees and contributors included: Guray Acar, Danny Bharj, Francois D´Humieres, Pavel Drasil, Nikos Fistas, Giovanni Garofolo, Bernhard Haindl, Vaughn Maiolla, Tom McParland, Victor Moreno, Madhu Niraula, Brent Phillips, Liviu Popescu, Jacky Pouzet, Aloke Roy, Greg Saccone, Robert Segers, Michal Skorepa, Michel Solery, Stephane Tamalet, Fred Templin, Jean-Marc Vacher, Bela Varkonyi, Tony Whyman, Fryderyk Wrobel and Dongsong Zeng.¶
The following individuals are acknowledged for their useful comments: Amanda Baber, Scott Burleigh, Stuart Card, Donald Eastlake, Adrian Farrel, Michael Matyas, Robert Moskowitz, Madhu Niraula, Greg Saccone, Stephane Tamalet, Eliot Lear, Eduard Vasilenko, Eric Vyncke. Pavel Drasil, Zdenek Jaron and Michal Skorepa are especially recognized for their many helpful ideas and suggestions. Akash Agarwal, Madhuri Madhava Badgandi, Sean Dickson, Don Dillenburg, Joe Dudkowski, Vijayasarathy Rajagopalan, Ron Sackman, Bhargava Raman Sai Prakash and Katherine Tran are acknowledged for their hard work on the implementation and technical insights that led to improvements for the spec.¶
Discussions on the IETF 6man and atn mailing lists during the fall of 2020 suggested additional points to consider. The authors gratefully acknowledge the list members who contributed valuable insights through those discussions. Eric Vyncke and Erik Kline were the intarea ADs, while Bob Hinden and Ole Troan were the 6man WG chairs at the time the document was developed; they are all gratefully acknowledged for their many helpful insights. Many of the ideas in this document have further built on IETF experiences beginning in the 1990s, with insights from colleagues including Ron Bonica, Brian Carpenter, Ralph Droms, Tom Herbert, Bob Hinden, Christian Huitema, Thomas Narten, Dave Thaler, Joe Touch, Pascal Thubert, and many others who deserve recognition.¶
Early observations on IP fragmentation performance implications were noted in the 1986 Digital Equipment Corporation (DEC) "qe reset" investigation, where fragment bursts from NFS UDP traffic triggered hardware resets resulting in communication failures. Jeff Chase, Fred Glover and Chet Juzsczak of the Ultrix Engineering Group led the investigation, and determined that setting a smaller NFS mount block size reduced the amount of fragmentation and suppressed the resets. Early observations on L2 media MTU issues were noted in the 1988 DEC FDDI investigation, where Raj Jain, KK Ramakrishnan and Kathy Wilde represented architectural considerations for FDDI networking in general including FDDI/Ethernet bridging. Jeff Mogul (who led the IETF Path MTU Discovery working group) and other DEC colleagues who supported these early investigations are also acknowledged.¶
Throughout the 1990's and into the 2000's, many colleagues supported and encouraged continuation of the work. Beginning with the DEC Project Sequoia effort at the University of California, Berkeley, then moving to the DEC research lab offices in Palo Alto CA, then to Sterling Software at the NASA Ames Research Center, then to SRI in Menlo Park, CA, then to Nokia in Mountain View, CA and finally to the Boeing Company in 2005 the work saw continuous advancement through the encouragement of many. Those who offered their support and encouragement are gratefully acknowledged.¶
This work is aligned with the NASA Safe Autonomous Systems Operation (SASO) program under NASA contract number NNA16BD84C.¶
This work is aligned with the FAA as per the SE2025 contract number DTFAWA-15-D-00030.¶
This work is aligned with the Boeing Information Technology (BIT) Mobility Vision Lab (MVL) program.¶
Honoring life, liberty and the pursuit of happiness.¶
The IPv4 reassembly checksum algorithm adopts the 8-bit Fletcher algorithm specified in Appendix I of [RFC1146] as also analyzed in [CKSUM]. [RFC6247] declared [RFC1146] historic for the reason that the algorithms had never seen widespread use with TCP, however this document adopts the 8-bit Fletcher algorithm for a different purpose. Quoting from Appendix I of [RFC1146], the IPv4 Fragmentation Checksum Algorithm proceeds as follows:¶
"The 8-bit Fletcher Checksum Algorithm is calculated over a sequence of data octets (call them D[1] through D[N]) by maintaining 2 unsigned 1's-complement 8-bit accumulators A and B whose contents are initially zero, and performing the following loop where i ranges from 1 to N:¶
It can be shown that at the end of the loop A will contain the 8-bit 1's complement sum of all octets in the datagram, and that B will contain (N)D[1] + (N-1)D[2] + ... + D[N]."¶
To calculate the IPv4 reassembly checksum, the above algorithm is applied over the N-octets of the L2-encapsulated OAL packet/fragment body beginning immediately after the L2 encapsulation header(s).¶
Section 2.5.5.1 of [RFC4291] defines an "IPv4-Compatible IPv6 Address" with the following structure:¶
Although [RFC4291] deprecates the address format from its former use in IPv6 transition mechanisms, this document now assigns new uses and therefore updates [RFC4291].¶
When an IPv4-Compatible IPv6 address appears in a packet sent over the wire, the most significant 96 bits are 0 and the least significant 32 bits include an IPv4 address as shown above.¶
When the address format is used for temporary local address conversions to IPv6, however, it can also be used to represent EUI-48 and EUI-64 addresses as shown below:¶
The above EUI-48 and EUI-64 compatible IPv6 forms MAY be used for temporary local address conversions, such as when converting EUI addresses to IPv6 to support IPv6 fragmentation/reassembly. The address forms MUST NOT appear in the IPv6 headers of packets sent over the wire, however they MAY appear in the body of a packet if also accompanied by a Type designator.¶
OMNI interface IPv6 ND messages are subject to authentication and integrity checks at multiple levels. When an OMNI interface sends an IPv6 ND message over an INET interface, it includes an authentication sub-option with a valid signature if necessary and always includes an IPv6 ND message checksum. The OMNI interface that receives the message verifies the IPv6 ND message checksum followed by the authentication signature (if present) to ensure IPv6 ND message integrity and authenticity.¶
When an OMNI interface sends an IPv6 ND message over an underlay interface connected to a secured network, it omits authentication (sub-)options but always calculates/includes an IPv6 ND message checksum beginning with a pseudo-header of the IPv6 header and extending to the end of the IPv6 ND message only with the Checksum field itself set to 0. When an OMNI interface sends an IPv6 ND message over an underlay interface connected to an unsecured network, it first includes an authentication (sub-)option and calculates the signature beginning with the first octet following the IPv6 ND message header Checksum field and extending to the end of the entire packet or super-packet with the authentication signature field set to 0. The OMNI interface next writes the signature into the signature field, then calculates the IPv6 ND message checksum as above.¶
The OMNI interface that receives the message applies any link layer authentication and integrity checks, then verifies the IPv6 ND message checksum. If the checks are correct, the OMNI interface next verifies the authentication signature. The OMNI interface then processes the packet further only if all checksums and authentication signatures were correct.¶
OAL destinations also discard carrier packets with unacceptable Identifications and submit the encapsulated fragments in all others for reassembly. The reassembly algorithm rejects any fragments with unacceptable sizes, offsets, etc. and reassembles all others. During reassembly, the extended Identification value provides an integrity assurance vector that compliments any integrity checks already applied by lower layers as well as a first-pass filter for any checks that will be applied later by upper layers.¶
ICAO Doc 9776 is the "Technical Manual for VHF Data Link Mode 2" (VDLM2) that specifies an essential radio frequency data link service for aircraft and ground stations in worldwide civil aviation air traffic management. The VDLM2 link type is "multicast capable" [RFC4861], but with considerable differences from common multicast links such as Ethernet and IEEE 802.11.¶
First, the VDLM2 link data rate is only 31.5Kbps - multiple orders of magnitude less than most modern wireless networking gear. Second, due to the low available link bandwidth only VDLM2 ground stations (i.e., and not aircraft) are permitted to send broadcasts, and even so only as compact link layer "beacons". Third, aircraft employ the services of ground stations by performing unicast RS/RA exchanges upon receipt of beacons instead of listening for multicast RA messages and/or sending multicast RS messages.¶
This beacon-oriented unicast RS/RA approach is necessary to conserve the already-scarce available link bandwidth. Moreover, since the numbers of beaconing ground stations operating within a given spatial range must be kept as sparse as possible, it would not be feasible to have different classes of ground stations within the same region observing different protocols. It is therefore highly desirable that all ground stations observe a common language of RS/RA as specified in this document.¶
Note that links of this nature may benefit from compression techniques that reduce the bandwidth necessary for conveying the same amount of data. The IETF lpwan working group is considering possible alternatives: [https://datatracker.ietf.org/wg/lpwan/documents].¶
Per [RFC4861], IPv6 ND messages may be sent to either a multicast or unicast link-scoped IPv6 destination address. However, IPv6 ND messaging should be coordinated between the Client and Proxy/Server only without invoking other nodes on the underlay network. This implies that Client-Proxy/Server control messaging should be isolated and not overheard by other nodes on the link.¶
To support Client-Proxy/Server isolation on some links, Proxy/Servers can maintain an OMNI-specific unicast link layer address ("MSADDR"). For Ethernet-compatible links, this specification reserves one Ethernet unicast address TBD5 (see: IANA Considerations). For non-Ethernet statically-addressed links MSADDR is reserved per the assigned numbers authority for the link layer addressing space. For still other links, MSADDR may be dynamically discovered through other means, e.g., link layer beacons.¶
Clients map the L3 addresses of all IPv6 ND messages they send (i.e., both multicast and unicast) to MSADDR instead of to an ordinary unicast or multicast link layer address. In this way, all of the Client's IPv6 ND messages will be received by Proxy/Servers that are configured to accept carrier packets destined to MSADDR. Note that multiple Proxy/Servers on the link could be configured to accept carrier packets destined to MSADDR, e.g., as a basis for supporting redundancy.¶
Therefore, Proxy/Servers must accept and process carrier packets destined to MSADDR, while all other devices must not process carrier packets destined to MSADDR. This model has well-established operational experience in Proxy Mobile IPv6 (PMIP) [RFC5213][RFC6543].¶
<< RFC Editor - remove prior to publication >>¶
Differences from earlier versions:¶
Submit for review.¶