]> Cisco Systems

Philip Pedersens vei 1 Lysaker 1366 Norway ot@cisco.com

Cisco Systems

Haarlerbergpark Haarlerbergweg 13-19 Amsterdam, NOORD-HOLLAND 1101 CH The Netherlands wdec@cisco.com

CERNET Center/Tsinghua University

Room 225, Main Building, Tsinghua University Beijing 100084 China xing@cernet.edu.cn

CERNET Center/Tsinghua University

Room 225, Main Building, Tsinghua University Beijing 100084 China congxiao@cernet.edu.cn

SoftBank Telecom

1-9-1 Higashi-Shinbashi, Munato-ku Tokyo Japan satoru.matsushima@g.softbank.co.jp

IP Infusion

1188 East Arques Avenue Sunnyvale CA 94085 United States tetsuya@ipinfusion.com

Huawei Technologies

Ottawa Canada tom.taylor.stds@gmail.com

This document describes a mechanism for transporting IPv4 packets across an IPv6 network using IP encapsulation. It also describes a generic mechanism for mapping between IPv6 addresses and IPv4 addresses as well as transport-layer ports.

Mapping of IPv4 addresses in IPv6 addresses has been described in numerous mechanisms dating back to the mid-1990s . The "automatic tunneling" mechanism as first described in assigned a globally unique IPv6 address to a host by combining the host's IPv4 address with a well-known IPv6 prefix. Given an IPv6 packet with a destination address with an embedded IPv4 address, a node could automatically tunnel this packet by extracting the IPv4 tunnel endpoint address from the IPv6 destination address. There are numerous variations of this idea, as described in 6over4 , 6to4 , the Intra-Site Automatic Tunnel Addressing Protocol (ISATAP) , and IPv6 Rapid Deployment on IPv4 Infrastructures (6rd) . The commonalities of all of these IPv6-over-IPv4 mechanisms are as follows: Automatic provisioning of an IPv6 address for a host or an IPv6 prefix for a site. Algorithmic or implicit address resolution of tunnel endpoint addresses. Given an IPv6 destination address, an IPv4 tunnel endpoint address can be calculated. Embedding of an IPv4 address or part thereof into an IPv6 address. In later phases of IPv4-to-IPv6 migration, it is expected that IPv6-only networks will be common, while there will still be a need for residual IPv4 deployment. This document describes a generic mapping of IPv4 to IPv6 and a mechanism for encapsulating IPv4 over IPv6. Just as for the IPv6-over-IPv4 mechanisms referred to above, the residual IPv4-over-IPv6 mechanism must be capable of: Provisioning an IPv4 prefix, an IPv4 address, or a shared IPv4 address. Algorithmically mapping between an IPv4 prefix, an IPv4 address, or a shared IPv4 address and an IPv6 address. The mapping scheme described here supports encapsulation of IPv4 packets in IPv6 in both mesh and hub-and-spoke topologies, including address mappings with full independence between IPv6 and IPv4 addresses. This document describes the delivery of IPv4 unicast service across an IPv6 infrastructure. IPv4 multicast is not considered in this document. The Address plus Port (A+P) architecture of sharing an IPv4 address by distributing the port space is described in . Specifically, Section 4 of covers stateless mapping. The corresponding stateful solution, Dual-Stack Lite (DS-Lite), is described in . The motivations for this work are described in . defines DHCPv6 options for the provisioning of MAP. Other means of provisioning are possible. Deployment considerations are described in . MAP relies on IPv6 and is designed to deliver dual-stack service while allowing IPv4 to be phased out within the service provider's (SP's) network. The phasing out of IPv4 within the SP network is independent of whether the end user disables IPv4 service or not. Further, "greenfield" IPv6-only networks may use MAP in order to deliver IPv4 to sites via the IPv6 network.

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

One or more MAP Customer Edge (CE) devices and Border Relays (BRs) connected to the same virtual link. A service provider may deploy a single MAP domain or may utilize multiple MAP domains. A set of parameters describing the mapping between an IPv4 prefix, IPv4 address, or shared IPv4 address and an IPv6 prefix or address. Each domain uses a different mapping rule set. A device that implements MAP. A MAP-enabled router managed by the service provider at the edge of a MAP domain. A BR has at least an IPv6-enabled interface and an IPv4 interface connected to the native IPv4 network. A MAP BR may also be referred to as simply a "BR" within the context of MAP. A device functioning as a Customer Edge router in a MAP deployment. A typical MAP CE adopting MAP Rules will serve a residential site with one WAN-side interface and one or more LAN-side interfaces. A MAP CE may also be referred to as simply a "CE" within the context of MAP. Each node has a separate part of the transport-layer port space; this is denoted as a port set. Algorithmically identifies a set of ports exclusively assigned to a CE. An IPv4 address that is shared among multiple CEs. Only ports that belong to the assigned port set can be used for communication. Also known as a port-restricted IPv4 address. The IPv6 prefix assigned to an End-user CE by means other than MAP itself, e.g., provisioned using DHCPv6 Prefix Delegation (PD) , assigned via Stateless Address Autoconfiguration (SLAAC) , or configured manually. It is unique for each CE. The IPv6 address used to reach the MAP function of a CE from other CEs and from BRs. An IPv6 prefix assigned by a service provider for a mapping rule. An IPv4 prefix assigned by a service provider for a mapping rule. The IPv4 EA-bits in the IPv6 address identify an IPv4 prefix/address (or part thereof) or a shared IPv4 address (or part thereof) and a Port Set Identifier.

In accordance with the requirements stated above, the MAP mechanism can operate with shared IPv4 addresses, full IPv4 addresses, or IPv4 prefixes. Operation with shared IPv4 addresses is described here, and the differences for full IPv4 addresses and prefixes are described below. The MAP mechanism uses existing standard building blocks. The existing Network Address and Port Translator (NAPT) on the CE is used with additional support for restricting transport-protocol ports, ICMP identifiers, and fragment identifiers to the configured port set. For packets outbound from the private IPv4 network, the CE NAPT MUST translate transport identifiers (e.g., TCP and UDP port numbers) so that they fall within the CE's assigned port range. The NAPT MUST in turn be connected to a MAP-aware forwarding function that does encapsulation/decapsulation of IPv4 packets in IPv6. MAP supports the encapsulation mode specified in . In addition, MAP specifies an algorithm to do "address resolution" from an IPv4 address and port to an IPv6 address. This algorithmic mapping is specified in . The MAP architecture described here restricts the use of the shared IPv4 address to only be used as the global address (outside) of the NAPT running on the CE. A shared IPv4 address MUST NOT be used to identify an interface. While it is theoretically possible to make host stacks and applications port-aware, it would be a drastic change to the IP model . For full IPv4 addresses and IPv4 prefixes, the architecture just described applies, with two differences: first, a full IPv4 address or IPv4 prefix can be used as it is today, e.g., for identifying an interface or as a DHCP pool, respectively. Second, the NAPT is not required to restrict the ports used on outgoing packets. This architecture is illustrated in .

The MAP BR connects one or more MAP domains to external IPv4 networks.

A MAP node is provisioned with one or more mapping rules. Mapping rules are used differently, depending on their function. Every MAP node must be provisioned with a Basic Mapping Rule. This is used by the node to configure its IPv4 address, IPv4 prefix, or shared IPv4 address. This same basic rule can also be used for forwarding, where an IPv4 destination address and, optionally, a destination port are mapped into an IPv6 address. Additional mapping rules are specified to allow for multiple different IPv4 subnets to exist within the domain and optimize forwarding between them. Traffic outside of the domain (i.e., when the destination IPv4 address does not match (using longest matching prefix) any Rule IPv4 prefix in the Rules database) is forwarded to the BR. There are two types of mapping rules: Basic Mapping Rule (BMR) - mandatory. A CE can be provisioned with multiple End-user IPv6 prefixes. There can only be one Basic Mapping Rule per End-user IPv6 prefix. However, all CEs having End-user IPv6 prefixes within (aggregated by) the same Rule IPv6 prefix may share the same Basic Mapping Rule. In combination with the End-user IPv6 prefix, the Basic Mapping Rule is used to derive the IPv4 prefix, address, or shared address and the PSID assigned to the CE. Forwarding Mapping Rule (FMR) - optional; used for forwarding. The Basic Mapping Rule may also be a Forwarding Mapping Rule. Each Forwarding Mapping Rule will result in an entry in the rule table for the Rule IPv4 prefix. Given a destination IPv4 address and port within the MAP domain, a MAP node can use the matching FMR to derive the End-user IPv6 address of the interface through which that IPv4 destination address and port combination can be reached. In hub-and-spoke mode, there are no FMRs. Both mapping rules share the same parameters: Rule IPv6 prefix (including prefix length) Rule IPv4 prefix (including prefix length) Rule EA-bit length (in bits) A MAP node finds its BMR by doing a longest match between the End-user IPv6 prefix and the Rule IPv6 prefix in the Mapping Rules table. The rule is then used for IPv4 prefix, address, or shared address assignment. A MAP IPv6 address is formed from the BMR Rule IPv6 prefix. This address MUST be assigned to an interface of the MAP node and is used to terminate all MAP traffic being sent or received to the node. Port-restricted IPv4 routes are installed in the rule table for all the Forwarding Mapping Rules, and a default route is installed to the MAP BR (see ). Forwarding Mapping Rules are used to allow direct communication between MAP CEs; this is known as "Mesh mode". In hub-and-spoke mode, there are no Forwarding Mapping Rules; all traffic MUST be forwarded directly to the BR. While an FMR is optional in the sense that a MAP CE MAY be configured with zero or more FMRs -- depending on the deployment -- all MAP CEs MUST implement support for both rule types.

The port-mapping algorithm is used in domains whose rules allow IPv4 address sharing. The simplest way to represent a port range is using a notation similar to Classless Inter-Domain Routing (CIDR) . For example, the first 256 ports are represented as port prefix 0.0/8 and the last 256 ports as 255.0/8. In hexadecimal, these would be 0x0000/8 (PSID = 0) and 0xFF00/8 (PSID = 0xFF), respectively. Using this technique but wishing to avoid allocating the system ports to the user, one would have to exclude the use of one or more PSIDs (e.g., PSIDs 0 to 3 in the example just given). When the PSID is embedded in the End-user IPv6 prefix, it is desirable to minimize the restrictions of possible PSID values in order to minimize dependencies between the End-user IPv6 prefix and the assigned port set. This is achieved by using an infix representation of the port value. Using such a representation, the well-known ports are excluded by restrictions on the value of the high-order bit field (A) rather than the PSID. The infix algorithm allocates ports to a given CE as a series of contiguous ranges spaced at regular intervals throughout the complete range of possible port-set values.

0 | | | +-----------+-----------+-------+ | a bits | k bits |m bits | ]]>

The number of offset bits -- 6 by default, as this excludes the system ports (0-1023). To guarantee non-overlapping port sets, the offset 'a' MUST be the same for every MAP CE sharing the same address. Selects the range of the port number. For 'a' > 0, A MUST be larger than 0. This ensures that the algorithm excludes the system ports. For the default value of 'a' (6), the system ports are excluded by requiring that A be greater than 0. Smaller values of 'a' exclude a larger initial range, e.g., 'a' = 4 will exclude ports 0-4095. The interval between initial port numbers of successive contiguous ranges assigned to the same user is 2^(16 - a). The length in bits of the PSID field. To guarantee non-overlapping port sets, the length 'k' MUST be the same for every MAP CE sharing the same address. The sharing ratio is 2^k.  The number of ports assigned to the user is 2^(16 - k) - 2^m (excluded ports). The Port Set Identifier (PSID). Different PSID values guarantee non-overlapping port sets, thanks to the restrictions on 'a' and 'k' stated above, because the PSID always occupies the same bit positions in the port number. The number of contiguous ports is given by 2^m. Selects the specific port within a particular range specified by the concatenation of A and the PSID.

The Basic Mapping Rule is mandatory and is used by the CE to provision itself with an IPv4 prefix, IPv4 address, or shared IPv4 address. Recall from that the BMR consists of the following parameters: Rule IPv6 prefix (including prefix length) Rule IPv4 prefix (including prefix length) Rule EA-bit length (in bits) shows the structure of the complete MAP IPv6 address as specified in this document.

| ]]>

The Rule IPv6 prefix is common among all CEs using the same Basic Mapping Rule within the MAP domain. The EA bit field encodes the CE-specific IPv4 address and port information. The EA bit field, which is unique for a given Rule IPv6 prefix, can contain a full or partial IPv4 address and, in the shared IPv4 address case, a PSID. An EA bit field length of 0 signifies that all relevant MAP IPv4 addressing information is passed directly in the BMR and is not derived from the EA bit field in the End-user IPv6 prefix. The MAP IPv6 address is created by concatenating the End-user IPv6 prefix with the MAP subnet identifier (if the End-user IPv6 prefix is shorter than 64 bits) and the interface identifier as specified in . The MAP subnet identifier is defined to be the first subnet (s bits set to zero). Define: r = length of the IPv4 prefix given by the BMR; o = length of the EA bit field as given by the BMR; p = length of the IPv4 suffix contained in the EA bit field. The length r MAY be zero, in which case the complete IPv4 address or prefix is encoded in the EA bits. If only a part of the IPv4 address / prefix is encoded in the EA bits, the Rule IPv4 prefix is provisioned to the CE by other means (e.g., a DHCPv6 option). To create a complete IPv4 address (or prefix), the IPv4 address suffix (p) from the EA bits is concatenated with the Rule IPv4 prefix (r bits). The offset of the EA bit field in the IPv6 address is equal to the BMR Rule IPv6 prefix length. The length of the EA bit field (o) is given by the BMR Rule EA-bit length and can be between 0 and 48. A length of 48 means that the complete IPv4 address and port are embedded in the End-user IPv6 prefix (a single port is assigned). A length of 0 means that no part of the IPv4 address or port is embedded in the address. The sum of the Rule IPv6 Prefix length and the Rule EA-bit length MUST be less than or equal to the End-user IPv6 prefix length. If o + r < 32 (length of the IPv4 address in bits), then an IPv4 prefix is assigned. This case is shown in .

If o + r is equal to 32, then a full IPv4 address is to be assigned. The address is created by concatenating the Rule IPv4 prefix and the EA-bits. This case is shown in .

If o + r is > 32, then a shared IPv4 address is to be assigned. The number of IPv4 address suffix bits (p) in the EA bits is given by 32 - r bits. The PSID bits are used to create a port set. The length of the PSID bit field within the EA bits is q = o - p.

The length of r MAY be 32, with no part of the IPv4 address embedded in the EA bits. This results in a mapping with no dependence between the IPv4 address and the IPv6 address. In addition, the length of o MAY be zero (no EA bits embedded in the End-user IPv6 prefix), meaning that the PSID is also provisioned using, for example, DHCP. See for an example of the Basic Mapping Rule.

The Forwarding Mapping Rule is optional and is used in Mesh mode to enable direct CE-to-CE connectivity. On adding an FMR rule, an IPv4 route is installed in the rule table for the Rule IPv4 prefix (Figures , , and ).

| ]]>

See for an example of the Forwarding Mapping Rule.

IPv4 traffic between MAP nodes that are all within one MAP domain is encapsulated in IPv6, with the sender's MAP IPv6 address as the IPv6 source address and the receiving MAP node's MAP IPv6 address as the IPv6 destination address. To reach IPv4 destinations outside of the MAP domain, traffic is also encapsulated in IPv6, but the destination IPv6 address is set to the configured IPv6 address of the MAP BR. On the CE, the path to the BR can be represented as a point-to-point IPv4-over-IPv6 tunnel with the source address of the tunnel being the CE's MAP IPv6 address and the BR IPv6 address as the remote tunnel address. When MAP is enabled, a typical CE router will install a default IPv4 route to the BR. The BR forwards traffic received from the outside to CEs using the normal MAP forwarding rules.

The interface identifier format of a MAP node is described below.

In the case of an IPv4 prefix, the IPv4 address field is right-padded with zeros up to 32 bits. The PSID field is left-padded with zeros to create a 16-bit field. For an IPv4 prefix or a complete IPv4 address, the PSID field is zero. If the End-user IPv6 prefix length is larger than 64, the most significant parts of the interface identifier are overwritten by the prefix.

For a given MAP domain, the BR and CE MUST be configured with the following MAP elements. The configured values for these elements are identical for all CEs and BRs within a given MAP domain. The Basic Mapping Rule and, optionally, the Forwarding Mapping Rules, including the Rule IPv6 prefix, Rule IPv4 prefix, and Length of EA bits. Hub-and-spoke mode or Mesh mode (if all traffic should be sent to the BR, or if direct CE-to-CE traffic should be supported). In addition, the MAP CE MUST be configured with the IPv6 address(es) of the MAP BR ().

The MAP elements are set to values that are the same across all CEs within a MAP domain. The values may be configured in a variety of ways, including provisioning methods such as the Broadband Forum's "TR-69" Residential Gateway management interface , an XML&nbhy;based object retrieved after IPv6 connectivity is established, or manual configuration by an administrator. IPv6 DHCP options for MAP configuration are defined in . Other configuration and management methods may use the formats described by these options for consistency and convenience of implementation on CEs that support multiple configuration methods. The only remaining provisioning information the CE requires in order to calculate the MAP IPv4 address and enable IPv4 connectivity is the IPv6 prefix for the CE. The End-user IPv6 prefix is configured as part of obtaining IPv6 Internet access. The MAP provisioning parameters, and hence the IPv4 service itself, are tied to the associated End-user IPv6 prefix lifetime; thus, the MAP service is also tied to this in terms of authorization, accounting, etc. A single MAP CE MAY be connected to more than one MAP domain, just as any router may have more than one IPv4-enabled service-provider-facing interface and more than one set of associated addresses assigned by DHCP. Each domain within which a given CE operates would require its own set of MAP configuration elements and would generate its own IPv4 address. Each MAP domain requires a distinct End-user IPv6 prefix. MAP DHCP options are specified in .

The MAP BR MUST be configured with corresponding mapping rules for each MAP domain for which it is acting as a BR. For increased reliability and load balancing, the BR IPv6 address MAY be an anycast address shared across a given MAP domain. As MAP is stateless, any BR may be used at any time. If the BR IPv6 address is anycast, the relay MUST use this anycast IPv6 address as the source address in packets relayed to CEs. Since MAP uses provider address space, no specific routes need to be advertised externally for MAP to operate in IPv6 or IPv4 BGP. However, if anycast is used for the MAP IPv6 relays, the anycast addresses must be advertised in the service provider's IGP.

depicts the overall MAP architecture with IPv4 users connected to a routed IPv6 network. MAP uses encapsulation mode as specified in . For a shared IPv4 address, a MAP CE forwarding IPv4 packets from the LAN performs NAT44 functions first and creates appropriate NAT44 bindings. The resulting IPv4 packets MUST contain the source IPv4 address and source transport identifiers specified by the MAP provisioning parameters. The IPv4 packet is forwarded using the CE's MAP forwarding function. The IPv6 source and destination addresses MUST then be derived as per of this document.

A MAP CE receiving an IPv6 packet to its MAP IPv6 address sends this packet to the CE's MAP function, where it is decapsulated. The resulting IPv4 packet is then forwarded to the CE's NAT44 function, where it is handled according to the NAT's translation table. A MAP BR receiving IPv6 packets selects a best matching MAP domain rule (Rule IPv6 prefix) based on a longest address match of the packet's IPv6 source address, as well as a match of the packet destination address against the configured BR IPv6 address(es). The selected MAP Rule allows the BR to determine the EA-bits from the source IPv6 address. To prevent spoofing of IPv4 addresses, any MAP node (CE and BR) MUST perform the following validation upon reception of a packet. First, the embedded IPv4 address or prefix, as well as the PSID (if any), are extracted from the source IPv6 address using the matching MAP Rule. These represent the range of what is acceptable as source IPv4 address and port. Second, the node extracts the source IPv4 address and port from the IPv4 packet encapsulated inside the IPv6 packet. If they are found to be outside the acceptable range, the packet MUST be silently discarded and a counter incremented to indicate that a potential spoofing attack may be underway. The source validation checks just described are not done for packets whose source IPv6 address is that of the BR (BR IPv6 address). By default, the CE router MUST drop packets received on the MAP virtual interface (i.e., after decapsulation of IPv6) for IPv4 destinations not for its own IPv4 shared address, full IPv4 address, or IPv4 prefix.

ICMP messages should be supported in MAP domains. Hence, the NAT44 in the MAP CE MUST implement the behavior for ICMP messages conforming to the best current practice documented in . If a MAP CE receives an ICMP message having the ICMP Identifier field in the ICMP header, the NAT44 in the MAP CE MUST rewrite this field to a specific value assigned from the port set. BRs and other CEs must handle this field in a way similar to the handling of a port number in the TCP/UDP header upon receiving the ICMP message with the ICMP Identifier field. If a MAP node receives an ICMP error message without the ICMP Identifier field for errors that are detected inside an IPv6 tunnel, a node should relay the ICMP error message to the original source. This behavior SHOULD be implemented in accordance with Section 8 of .

Due to the different sizes of the IPv4 and IPv6 headers, handling the maximum packet size is relevant for the operation of any system connecting the two address families. There are three mechanisms to handle this issue: Path MTU Discovery (PMTUD), fragmentation, and transport-layer negotiation such as the TCP Maximum Segment Size (MSS) option . MAP uses all three mechanisms to deal with different cases.

Encapsulating an IPv4 packet to carry it across the MAP domain will increase its size (typically by 40 bytes). It is strongly recommended that the MTU in the MAP domain be well managed and that the IPv6 MTU on the CE WAN-side interface be set so that no fragmentation occurs within the boundary of the MAP domain. For an IPv4 packet entering a MAP domain, fragmentation is performed as described in Section 7.2 of . The use of an anycast source address could lead to an ICMP error message generated on the path being sent to a different BR. Therefore, using a dynamically set tunnel MTU (Section 6.7 of ) is subject to IPv6 Path MTU black holes. A MAP BR using an anycast source address SHOULD NOT by default use Path MTU Discovery across the MAP domain. Multiple BRs using the same anycast source address could send fragmented packets to the same CE at the same time. If the fragmented packets from different BRs happen to use the same fragment ID, incorrect reassembly might occur. See for an analysis of the problem; Section 3.4 of suggests solving the problem by fragmenting the inner packet.

The forwarding of an IPv4 packet received from outside of the MAP domain requires the IPv4 destination address and the transport-protocol destination port. The transport-protocol information is only available in the first fragment received. As described in Section 5.3.3 of , a MAP node receiving an IPv4 fragmented packet from outside has to reassemble the packet before sending the packet onto the MAP link. If the first packet received contains the transport-protocol information, it is possible to optimize this behavior by using a cache and forwarding the fragments unchanged. Implementers of MAP should be aware that there are a number of well-known attacks against IP fragmentation; see and . Implementers should also be aware of additional issues with reassembling packets at high rates, as described in .

If two IPv4 hosts behind two different MAP CEs with the same IPv4 address send fragments to an IPv4 destination host outside the domain, those hosts may use the same IPv4 fragmentation identifier, resulting in incorrect reassembly of the fragments at the destination host. Given that the IPv4 fragmentation identifier is a 16-bit field, it could be used similarly to port ranges. A MAP CE could rewrite the IPv4 fragmentation identifier to be within its allocated port set, if the resulting fragment identifier space was large enough related to the rate at which fragments were sent. However, splitting the identifier space in this fashion would increase the probability of reassembly collisions for all connections through the Customer Premises Equipment (CPE). See also .

The NAT44 implemented in the MAP CE SHOULD conform to the behavior and best current practices documented in , , and . In MAP address-sharing mode (determined by the MAP domain / rule configuration parameters), the operation of the NAT44 MUST be restricted to the available port numbers derived via the Basic Mapping Rule.

With consistency checks between IPv4 and IPv6 sources that are performed on IPv4/IPv6 packets received by MAP nodes, MAP does not introduce any new opportunity for spoofing attacks that would not already exist in IPv6. In MAP domains where IPv4 addresses are shared, the fact that IPv4 datagram reassembly may be necessary introduces an opportunity for DoS attacks. This is inherent in address sharing and is common with other address-sharing approaches such as DS-Lite and NAT64/DNS64. The best protection against such attacks is to accelerate IPv6 deployment so that address sharing is used less and less where MAP is supported. Routing loop attacks may exist in some "automatic tunneling" scenarios and are documented in . They cannot exist with MAP because each BR checks that the IPv6 source address of a received IPv6 packet is a CE address based on the Forwarding Mapping Rule. From hosts that are not subject to ingress filtering , an attacker can inject spoofed packets during ongoing transport connections . The attacks depend on guessing which ports are currently used by target hosts. Using an unrestricted port set is preferable, i.e., using native IPv6 connections that are not subject to MAP port-range restrictions. To minimize these types of attacks when using a restricted port set, the MAP CE's NAT44 filtering behavior SHOULD be "Address-Dependent Filtering" as described in Section 5 of . Furthermore, the MAP CEs SHOULD use a DNS transport proxy function to handle DNS traffic and source such traffic from IPv6 interfaces not assigned to MAP. outlines general issues with IPv4 address sharing.

&rfc2119; &rfc2473; &rfc5625; &rfc6346; &rfc6335; &rfc6333; &rfc1933; &rfc4213; &rfc5969; &rfc3056; &rfc2529; &rfc5214; &rfc3633; &rfc6269; &rfc6250; &rfc2663; &rfc6324; &rfc2827; &rfc4953; &rfc5961; &rfc6056; &rfc4787; &rfc5508; &rfc5382; &rfc4862; &rfc4632; &rfc4459; &rfc1858; &rfc3128; &rfc4963; &rfc6864; Broadband Forum TR-069

Example 1 - Basic Mapping Rule:

Example 2 - BR:

Example 3 - Forwarding Mapping Rule:

Example 4 - Rule with no embedded address bits and no address sharing:

Example 5 - Rule with no embedded address bits and address sharing (sharing ratio of 256):

This appendix describes how the port-mapping algorithm described in was derived. The algorithm is used in domains whose rules allow IPv4 address sharing. The basic requirement for a port-mapping algorithm is that the port sets it assigns to different MAP CEs MUST be non-overlapping. A number of other requirements guided the choice of the algorithm: In keeping with the general MAP algorithm, the port set MUST be derivable from a Port Set identifier (PSID) that can be embedded in the End-user IPv6 prefix. The mapping MUST be reversible such that, given the port number, the PSID of the port set to which it belongs can be quickly derived. The algorithm MUST allow a broad range of address-sharing ratios. It SHOULD be possible to exclude subsets of the complete port numbering space from assignment. Most operators would exclude the system ports (0-1023). A conservative operator might exclude all but the transient ports (49152-65535). The effect of port exclusion on the possible values of the End-user IPv6 prefix (i.e., due to restrictions on the PSID value) SHOULD be minimized. For administrative simplicity, the algorithm SHOULD allocate the same or almost the same number of ports to each CE sharing a given IPv4 address. The two extreme cases that an algorithm satisfying those conditions might support are when (1) the port numbers are not contiguous for each PSID but uniformly distributed across the allowed port range and (2) the port numbers are contiguous in a single range for each PSID. The port-mapping algorithm proposed here is called the Generalized Modulus Algorithm (GMA) and supports both of these cases. For a given IPv4 address-sharing ratio (R) and the maximum number of contiguous ports (M) in a port set, the GMA is defined as follows: The port numbers (P) corresponding to a given PSID are generated by:

where i and j are indices and the ranges of i, j, and the PSID are discussed below. For any given port number P, the PSID is calculated as:

where trunc() is the operation of rounding down to the nearest integer. Formula (1) can be interpreted as follows. First, the available port space is divided into blocks of size R * M.  Each block is divided into R individual ranges of length M.  The index i in formula (1) selects a block, PSID selects a range within that block, and the index j selects a specific port value within the range. On the basis of this interpretation: i ranges from ceil(N / (R * M)) to trunc(65536/(R * M)) - 1, where ceil is the operation of rounding up to the nearest integer and N is the number of ports (e.g., 1024) excluded from the lower end of the range. That is, any block containing excluded values is discarded at the lower end, and if the final block has fewer than R * M values it is discarded. This ensures that the same number of ports is assigned to every PSID. PSID ranges from 0 to R - 1. j ranges from 0 to M - 1.

If R and M are powers of 2 (R = 2^k, M = 2^m), formula (1) translates to a computationally convenient structure for any port number represented as a 16-bit binary number. This structure is shown in .

|<-----k bits---->|<------m bits----->| ]]>

As shown in the figure, the index value i of formula (1) is given by the first a = 16 - k - m bits of the port number. The PSID value is given by the next k bits, and the index value j is given by the last m bits. Because the PSID is always in the same position in the port number and always the same length, different PSID values are guaranteed to generate different sets of port numbers. In the reverse direction, the generating PSID can be extracted from any port number by a bitmask operation. Note that when M and R are powers of 2, 65536 divides evenly by R * M.  Hence, the final block is complete, and the upper bound on i is exactly 65536/(R * M) - 1. The lower bound on i is still the minimum required to ensure that the required set of ports is excluded. No port numbers are wasted through the discarding of blocks at the lower end if block size R * M is a factor of N, the number of ports to be excluded. As a final note, the number of blocks into which the range 0-65535 is being divided in the above representation is given by 2^a. Hence, the case where a = 0 can be interpreted as one where the complete range has been divided into a single block, and individual port sets are contained in contiguous ranges in that block. We cannot throw away the whole block in that case, so port exclusion has to be achieved by putting a lower bound equal to ceil(N / M) on the allowed set of PSID values instead.

For example, for R = 256, PSID = 0, offset: a = 6 and PSID length: k = 8 bits:

For example, for R = 64, PSID = 0, a = 0 (PSID offset = 0 and PSID length = 6 bits), no port exclusion:

This document is based on the ideas of many, including Masakazu Asama, Mohamed Boucadair, Gang Chen, Maoke Chen, Wojciech Dec, Xiaohong Deng, Jouni Korhonen, Tomek Mrugalski, Jacni Qin, Chunfa Sun, Qiong Sun, and Leaf Yeh.  The authors want in particular to recognize Remi Despres, who has tirelessly worked on generalized mechanisms for stateless address mapping. The authors would like to thank Lichun Bao, Guillaume Gottard, Dan Wing, Jan Zorz, Necj Scoberne, Tina Tsou, Kristian Poscic, and especially Tom Taylor and Simon Perreault for the thorough review and comments of this document. Useful IETF Last Call comments were received from Brian Weis and Lei Yan.

This document is the result of the IETF Softwire MAP design team effort and numerous previous individual contributions in this area: