Network Working Group R. Gilligan
Request for Comments: 2893 FreeGate Corp.
Obsoletes: 1933 E. Nordmark
Category: Standards Track Sun Microsystems, Inc.
August 2000

Transition Mechanisms for IPv6 Hosts and Routers

Status of this Memo

This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.

Copyright Notice

Copyright (C) The Internet Society (2000). All Rights Reserved.

Abstract

This document specifies IPv4 compatibility mechanisms that can be
implemented by IPv6 hosts and routers. These mechanisms include
providing complete implementations of both versions of the Internet
Protocol (IPv4 and IPv6), and tunneling IPv6 packets over IPv4
routing infrastrUCtures. They are designed to allow IPv6 nodes to
maintain complete compatibility with IPv4, which should greatly
simplify the deployment of IPv6 in the Internet, and facilitate the
eventual transition of the entire Internet to IPv6. This document
obsoletes RFC1933.

Table of Contents

1. Introduction............................................. 2
1.1. Terminology......................................... 3
1.2. Structure of this Document.......................... 5
2. Dual IP Layer Operation.................................. 6
2.1. Address Configuration............................... 7
2.2. DNS................................................. 7
2.3. Advertising Addresses in the DNS.................... 8
3. Common Tunneling Mechanisms.............................. 9
3.1. Encapsulation....................................... 11
3.2. Tunnel MTU and Fragmentation........................ 11
3.3. Hop Limit........................................... 13
3.4. Handling IPv4 ICMP errors........................... 13
3.5. IPv4 Header Construction............................ 15
3.6. Decapsulation....................................... 16
3.7. Link-Local Addresses................................ 17
3.8. Neighbor Discovery over Tunnels..................... 18
4. Configured Tunneling..................................... 18
4.1. Default Configured Tunnel........................... 19
4.2. Default Configured Tunnel using IPv4 "Anycast Address" 19
4.3. Ingress Filtering................................... 20
5. Automatic Tunneling...................................... 20
5.1. IPv4-Compatible Address Format...................... 20
5.2. IPv4-Compatible Address Configuration............... 21
5.3. Automatic Tunneling Operation....................... 22
5.4. Use With Default Configured Tunnels................. 22
5.5. Source Address Selection............................ 23
5.6. Ingress Filtering................................... 23
6. Acknowledgments.......................................... 24
7. Security Considerations.................................. 24
8. Authors' Addresses....................................... 24
9. References............................................... 25
10. Changes from RFC1933................................... 26
11. Full Copyright Statement................................ 29

1. Introduction

The key to a successful IPv6 transition is compatibility with the
large installed base of IPv4 hosts and routers. Maintaining
compatibility with IPv4 while deploying IPv6 will streamline the task
of transitioning the Internet to IPv6. This specification defines a
set of mechanisms that IPv6 hosts and routers may implement in order
to be compatible with IPv4 hosts and routers.

The mechanisms in this document are designed to be employed by IPv6
hosts and routers that need to interoperate with IPv4 hosts and
utilize IPv4 routing infrastructures. We eXPect that most nodes in

the Internet will need such compatibility for a long time to come,
and perhaps even indefinitely.

However, IPv6 may be used in some environments where interoperability
with IPv4 is not required. IPv6 nodes that are designed to be used
in such environments need not use or even implement these mechanisms.

The mechanisms specified here include:

- Dual IP layer (also known as Dual Stack): A technique for
providing complete support for both Internet protocols -- IPv4 and
IPv6 -- in hosts and routers.

- Configured tunneling of IPv6 over IPv4: Point-to-point tunnels
made by encapsulating IPv6 packets within IPv4 headers to carry
them over IPv4 routing infrastructures.

- IPv4-compatible IPv6 addresses: An IPv6 address format that
employs embedded IPv4 addresses.

- Automatic tunneling of IPv6 over IPv4: A mechanism for using
IPv4-compatible addresses to automatically tunnel IPv6 packets
over IPv4 networks.

The mechanisms defined here are intended to be part of a "transition
toolbox" -- a growing collection of techniques which implementations
and users may employ to ease the transition. The tools may be used
as needed. Implementations and sites decide which techniques are
appropriate to their specific needs. This document defines the
initial core set of transition mechanisms, but these are not expected
to be the only tools available. Additional transition and
compatibility mechanisms are expected to be developed in the future,
with new documents being written to specify them.

1.1. Terminology

The following terms are used in this document:

Types of Nodes

IPv4-only node:

A host or router that implements only IPv4. An IPv4-only node
does not understand IPv6. The installed base of IPv4 hosts and
routers existing before the transition begins are IPv4-only
nodes.

IPv6/IPv4 node:

A host or router that implements both IPv4 and IPv6.

IPv6-only node:

A host or router that implements IPv6, and does not implement
IPv4. The operation of IPv6-only nodes is not addressed here.

IPv6 node:

Any host or router that implements IPv6. IPv6/IPv4 and IPv6-
only nodes are both IPv6 nodes.

IPv4 node:

Any host or router that implements IPv4. IPv6/IPv4 and IPv4-
only nodes are both IPv4 nodes.

Types of IPv6 Addresses

IPv4-compatible IPv6 address:

An IPv6 address bearing the high-order 96-bit prefix
0:0:0:0:0:0, and an IPv4 address in the low-order 32-bits.
IPv4-compatible addresses are used by IPv6/IPv4 nodes which
perform automatic tunneling,

IPv6-native address:

The remainder of the IPv6 address space. An IPv6 address that
bears a prefix other than 0:0:0:0:0:0.

Techniques Used in the Transition

IPv6-over-IPv4 tunneling:

The technique of encapsulating IPv6 packets within IPv4 so that
they can be carried across IPv4 routing infrastructures.

Configured tunneling:

IPv6-over-IPv4 tunneling where the IPv4 tunnel endpoint address
is determined by configuration information on the encapsulating
node. The tunnels can be either unidirectional or
bidirectional. Bidirectional configured tunnels behave as
virtual point-to-point links.

Automatic tunneling:

IPv6-over-IPv4 tunneling where the IPv4 tunnel endpoint address
is determined from the IPv4 address embedded in the IPv4-
compatible destination address of the IPv6 packet being
tunneled.

IPv4 multicast tunneling:

IPv6-over-IPv4 tunneling where the IPv4 tunnel endpoint address
is determined using Neighbor Discovery [7]. Unlike configured
tunneling this does not require any address configuration and
unlike automatic tunneling it does not require the use of
IPv4-compatible addresses. However, the mechanism assumes that
the IPv4 infrastructure supports IPv4 multicast. Specified in
[3] and not further discussed in this document.

Other transition mechanisms, including other tunneling mechanisms,
are outside the scope of this document.

Modes of operation of IPv6/IPv4 nodes

IPv6-only operation:

An IPv6/IPv4 node with its IPv6 stack enabled and its IPv4
stack disabled.

IPv4-only operation:

An IPv6/IPv4 node with its IPv4 stack enabled and its IPv6
stack disabled.

IPv6/IPv4 operation:

An IPv6/IPv4 node with both stacks enabled.

The keyWords MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD,
SHOULD NOT, RECOMMENDED, MAY, and OPTIONAL, when they appear in this
document, are to be interpreted as described in [16].

1.2. Structure of this Document

The remainder of this document is organized as follows:

- Section 2 discusses the operation of nodes with a dual IP layer,
IPv6/IPv4 nodes.

- Section 3 discusses the common mechanisms used in both of the
IPv6-over-IPv4 tunneling techniques.

- Section 4 discusses configured tunneling.

- Section 5 discusses automatic tunneling and the IPv4-compatible
IPv6 address format.

2. Dual IP Layer Operation

The most straightforward way for IPv6 nodes to remain compatible with
IPv4-only nodes is by providing a complete IPv4 implementation. IPv6
nodes that provide a complete IPv4 and IPv6 implementations are
called "IPv6/IPv4 nodes." IPv6/IPv4 nodes have the ability to send
and receive both IPv4 and IPv6 packets. They can directly
interoperate with IPv4 nodes using IPv4 packets, and also directly
interoperate with IPv6 nodes using IPv6 packets.

Even though a node may be equipped to support both protocols, one or
the other stack may be disabled for operational reasons. Thus
IPv6/IPv4 nodes may be operated in one of three modes:

- With their IPv4 stack enabled and their IPv6 stack disabled.

- With their IPv6 stack enabled and their IPv4 stack disabled.

- With both stacks enabled.

IPv6/IPv4 nodes with their IPv6 stack disabled will operate like
IPv4-only nodes. Similarly, IPv6/IPv4 nodes with their IPv4 stacks
disabled will operate like IPv6-only nodes. IPv6/IPv4 nodes MAY
provide a configuration switch to disable either their IPv4 or IPv6
stack.

The dual IP layer technique may or may not be used in conjunction
with the IPv6-over-IPv4 tunneling techniques, which are described in
sections 3, 4 and 5. An IPv6/IPv4 node that supports tunneling MAY
support only configured tunneling, or both configured and automatic
tunneling. Thus three modes of tunneling support are possible:

- IPv6/IPv4 node that does not perform tunneling.

- IPv6/IPv4 node that performs configured tunneling only.

- IPv6/IPv4 node that performs configured tunneling and automatic
tunneling.

2.1. Address Configuration

Because they support both protocols, IPv6/IPv4 nodes may be
configured with both IPv4 and IPv6 addresses. IPv6/IPv4 nodes use
IPv4 mechanisms (e.g. DHCP) to acquire their IPv4 addresses, and IPv6
protocol mechanisms (e.g. stateless address autoconfiguration) to
acquire their IPv6-native addresses. Section 5.2 describes a
mechanism by which IPv6/IPv4 nodes that support automatic tunneling
MAY use IPv4 protocol mechanisms to acquire their IPv4-compatible
IPv6 address.

2.2. DNS

The Domain Naming System (DNS) is used in both IPv4 and IPv6 to map
between hostnames and IP addresses. A new resource record type named
"A6" has been defined for IPv6 addresses [6] with support for an
earlier record named "AAAA". Since IPv6/IPv4 nodes must be able to
interoperate directly with both IPv4 and IPv6 nodes, they must
provide resolver libraries capable of dealing with IPv4 "A" records
as well as IPv6 "A6" and "AAAA" records.

DNS resolver libraries on IPv6/IPv4 nodes MUST be capable of handling
both A6/AAAA and A records. However, when a query locates an A6/AAAA
record holding an IPv6 address, and an A record holding an IPv4
address, the resolver library MAY filter or order the results
returned to the application in order to influence the version of IP
packets used to communicate with that node. In terms of filtering,
the resolver library has three alternatives:

- Return only the IPv6 address to the application.

- Return only the IPv4 address to the application.

- Return both addresses to the application.

If it returns only the IPv6 address, the application will communicate
with the node using IPv6. If it returns only the IPv4 address, the
application will communicate with the node using IPv4. If it returns
both addresses, the application will have the choice which address to
use, and thus which IP protocol to employ.

If it returns both, the resolver MAY elect to order the addresses --
IPv6 first, or IPv4 first. Since most applications try the addresses
in the order they are returned by the resolver, this can affect the
IP version "preference" of applications.

The decision to filter or order DNS results is implementation
specific. IPv6/IPv4 nodes MAY provide policy configuration to
control filtering or ordering of addresses returned by the resolver,
or leave the decision entirely up to the application.

An implementation MUST allow the application to control whether or
not such filtering takes place.

2.3. Advertising Addresses in the DNS

There are some constraint placed on the use of the DNS during
transition. Most of these are obvious but are stated here for
completeness.

The recommendation is that A6/AAAA records for a node should not be
added to the DNS until all of these are true:

1) The address is assigned to the interface on the node.

2) The address is configured on the interface.

3) The interface is on a link which is connected to the IPv6
infrastructure.

If an IPv6 node is isolated from an IPv6 perspective (e.g. it is not
connected to the 6bone to take a concrete example) constraint #3
would mean that it should not have an address in the DNS.

This works great when other dual stack nodes tries to contact the
isolated dual stack node. There is no IPv6 address in the DNS thus
the peer doesn't even try communicating using IPv6 but goes directly
to IPv4 (we are assuming both nodes have A records in the DNS.)

However, this does not work well when the isolated node is trying to
establish communication. Even though it does not have an IPv6
address in the DNS it will find A6/AAAA records in the DNS for the
peer. Since the isolated node has IPv6 addresses assigned to at
least one interface it will try to communicate using IPv6. If it has
no IPv6 route to the 6bone (e.g. because the local router was
upgraded to advertise IPv6 addresses using Neighbor Discovery but
that router doesn't have any IPv6 routes) this communication will
fail. Typically this means a few minutes of delay as TCP times out.
The TCP specification says that ICMP unreachable messages could be
due to routing transients thus they should not immediately terminate
the TCP connection. This means that the normal TCP timeout of a few
minutes apply. Once TCP times out the application will hopefully try
the IPv4 addresses based on the A records in the DNS, but this will
be painfully slow.

A possible implication of the recommendations above is that, if one
enables IPv6 on a node on a link without IPv6 infrastructure, and
choose to add A6/AAAA records to the DNS for that node, then external
IPv6 nodes that might see these A6/AAAA records will possibly try to
reach that node using IPv6 and suffer delays or communication failure
due to unreachability. (A delay is incurred if the application
correctly falls back to using IPv4 if it can not establish
communication using IPv6 addresses. If this fallback is not done the
application would fail to communicate in this case.) Thus it is
suggested that either the recommendations be followed, or care be
taken to only do so with nodes that will not be impacted by external
Accessing delays and/or communication failure.

In the future when a site or node removes the support for IPv4 the
above recommendations apply to when the A records for the node(s)
should be removed from the DNS.

3. Common Tunneling Mechanisms

In most deployment scenarios, the IPv6 routing infrastructure will be
built up over time. While the IPv6 infrastructure is being deployed,
the existing IPv4 routing infrastructure can remain functional, and
can be used to carry IPv6 traffic. Tunneling provides a way to
utilize an existing IPv4 routing infrastructure to carry IPv6
traffic.

IPv6/IPv4 hosts and routers can tunnel IPv6 datagrams over regions of
IPv4 routing topology by encapsulating them within IPv4 packets.
Tunneling can be used in a variety of ways:

- Router-to-Router. IPv6/IPv4 routers interconnected by an IPv4
infrastructure can tunnel IPv6 packets between themselves. In
this case, the tunnel spans one segment of the end-to-end path
that the IPv6 packet takes.

- Host-to-Router. IPv6/IPv4 hosts can tunnel IPv6 packets to an
intermediary IPv6/IPv4 router that is reachable via an IPv4
infrastructure. This type of tunnel spans the first segment of
the packet's end-to-end path.

- Host-to-Host. IPv6/IPv4 hosts that are interconnected by an IPv4
infrastructure can tunnel IPv6 packets between themselves. In
this case, the tunnel spans the entire end-to-end path that the
packet takes.

- Router-to-Host. IPv6/IPv4 routers can tunnel IPv6 packets to
their final destination IPv6/IPv4 host. This tunnel spans only
the last segment of the end-to-end path.

Tunneling techniques are usually classified according to the
mechanism by which the encapsulating node determines the address of
the node at the end of the tunnel. In the first two tunneling
methods listed above -- router-to-router and host-to-router -- the
IPv6 packet is being tunneled to a router. The endpoint of this type
of tunnel is an intermediary router which must decapsulate the IPv6
packet and forward it on to its final destination. When tunneling to
a router, the endpoint of the tunnel is different from the
destination of the packet being tunneled. So the addresses in the
IPv6 packet being tunneled can not provide the IPv4 address of the
tunnel endpoint. Instead, the tunnel endpoint address must be
determined from configuration information on the node performing the
tunneling. We use the term "configured tunneling" to describe the
type of tunneling where the endpoint is explicitly configured.

In the last two tunneling methods -- host-to-host and router-to-host
-- the IPv6 packet is tunneled all the way to its final destination.
In this case, the destination address of both the IPv6 packet and the
encapsulating IPv4 header identify the same node! This fact can be
exploited by encoding information in the IPv6 destination address
that will allow the encapsulating node to determine tunnel endpoint
IPv4 address automatically. Automatic tunneling employs this
technique, using an special IPv6 address format with an embedded IPv4
address to allow tunneling nodes to automatically derive the tunnel
endpoint IPv4 address. This eliminates the need to explicitly
configure the tunnel endpoint address, greatly simplifying
configuration.

The two tunneling techniques -- automatic and configured -- differ
primarily in how they determine the tunnel endpoint address. Most of
the underlying mechanisms are the same:

- The entry node of the tunnel (the encapsulating node) creates an
encapsulating IPv4 header and transmits the encapsulated packet.

- The exit node of the tunnel (the decapsulating node) receives the
encapsulated packet, reassembles the packet if needed, removes the
IPv4 header, updates the IPv6 header, and processes the received
IPv6 packet.

- The encapsulating node MAY need to maintain soft state information
for each tunnel recording such parameters as the MTU of the tunnel
in order to process IPv6 packets forwarded into the tunnel. Since
the number of tunnels that any one host or router may be using may
grow to be quite large, this state information can be cached and
discarded when not in use.

The remainder of this section discusses the common mechanisms that
apply to both types of tunneling. Subsequent sections discuss how
the tunnel endpoint address is determined for automatic and
configured tunneling.

3.1. Encapsulation

The encapsulation of an IPv6 datagram in IPv4 is shown below:

+-------------+
IPv4
Header
+-------------+ +-------------+
IPv6 IPv6
Header Header
+-------------+ +-------------+
Transport Transport
Layer ===> Layer
Header Header
+-------------+ +-------------+

~ Data ~ ~ Data ~

+-------------+ +-------------+

Encapsulating IPv6 in IPv4

In addition to adding an IPv4 header, the encapsulating node also has
to handle some more complex issues:

- Determine when to fragment and when to report an ICMP "packet too
big" error back to the source.

- How to reflect IPv4 ICMP errors from routers along the tunnel path
back to the source as IPv6 ICMP errors.

Those issues are discussed in the following sections.

3.2. Tunnel MTU and Fragmentation

The encapsulating node could view encapsulation as IPv6 using IPv4 as
a link layer with a very large MTU (65535-20 bytes to be exact; 20
bytes "extra" are needed for the encapsulating IPv4 header). The
encapsulating node would need only to report IPv6 ICMP "packet too
big" errors back to the source for packets that exceed this MTU.
However, such a scheme would be inefficient for two reasons:

1) It would result in more fragmentation than needed. IPv4 layer
fragmentation SHOULD be avoided due to the performance problems
caused by the loss unit being smaller than the retransmission unit
[11].

2) Any IPv4 fragmentation occurring inside the tunnel would have to
be reassembled at the tunnel endpoint. For tunnels that terminate
at a router, this would require additional memory to reassemble
the IPv4 fragments into a complete IPv6 packet before that packet
could be forwarded onward.

The fragmentation inside the tunnel can be reduced to a minimum by
having the encapsulating node track the IPv4 Path MTU across the
tunnel, using the IPv4 Path MTU Discovery Protocol [8] and recording
the resulting path MTU. The IPv6 layer in the encapsulating node can
then view a tunnel as a link layer with an MTU equal to the IPv4 path
MTU, minus the size of the encapsulating IPv4 header.

Note that this does not completely eliminate IPv4 fragmentation in
the case when the IPv4 path MTU would result in an IPv6 MTU less than
1280 bytes. (Any link layer used by IPv6 has to have an MTU of at
least 1280 bytes [4].) In this case the IPv6 layer has to "see" a
link layer with an MTU of 1280 bytes and the encapsulating node has
to use IPv4 fragmentation in order to forward the 1280 byte IPv6
packets.

The encapsulating node can employ the following algorithm to
determine when to forward an IPv6 packet that is larger than the
tunnel's path MTU using IPv4 fragmentation, and when to return an
IPv6 ICMP "packet too big" message:

if (IPv4 path MTU - 20) is less than or equal to 1280
if packet is larger than 1280 bytes
Send IPv6 ICMP "packet too big" with MTU = 1280.
Drop packet.
else
Encapsulate but do not set the Don't Fragment
flag in the IPv4 header. The resulting IPv4
packet might be fragmented by the IPv4 layer on
the encapsulating node or by some router along
the IPv4 path.
endif
else
if packet is larger than (IPv4 path MTU - 20)
Send IPv6 ICMP "packet too big" with
MTU = (IPv4 path MTU - 20).
Drop packet.
else

Encapsulate and set the Don't Fragment flag
in the IPv4 header.
endif
endif

Encapsulating nodes that have a large number of tunnels might not be
able to store the IPv4 Path MTU for all tunnels. Such nodes can, at
the expense of additional fragmentation in the network, avoid using
the IPv4 Path MTU algorithm across the tunnel and instead use the MTU
of the link layer (under IPv4) in the above algorithm instead of the
IPv4 path MTU.

In this case the Don't Fragment bit MUST NOT be set in the
encapsulating IPv4 header.

3.3. Hop Limit

IPv6-over-IPv4 tunnels are modeled as "single-hop". That is, the
IPv6 hop limit is decremented by 1 when an IPv6 packet traverses the
tunnel. The single-hop model serves to hide the existence of a
tunnel. The tunnel is opaque to users of the network, and is not
detectable by network diagnostic tools such as traceroute.

The single-hop model is implemented by having the encapsulating and
decapsulating nodes process the IPv6 hop limit field as they would if
they were forwarding a packet on to any other datalink. That is,
they decrement the hop limit by 1 when forwarding an IPv6 packet.
(The originating node and final destination do not decrement the hop
limit.)

The TTL of the encapsulating IPv4 header is selected in an
implementation dependent manner. The current suggested value is
published in the "Assigned Numbers RFC. Implementations MAY provide
a mechanism to allow the administrator to configure the IPv4 TTL such
as the one specified in the IP Tunnel MIB [17].

3.4. Handling IPv4 ICMP errors

In response to encapsulated packets it has sent into the tunnel, the
encapsulating node might receive IPv4 ICMP error messages from IPv4
routers inside the tunnel. These packets are addressed to the
encapsulating node because it is the IPv4 source of the encapsulated
packet.

The ICMP "packet too big" error messages are handled according to
IPv4 Path MTU Discovery [8] and the resulting path MTU is recorded in
the IPv4 layer. The recorded path MTU is used by IPv6 to determine
if an IPv6 ICMP "packet too big" error has to be generated as
described in section 3.2.

The handling of other types of ICMP error messages depends on how
much information is included in the "packet in error" field, which
holds the encapsulated packet that caused the error.

Many older IPv4 routers return only 8 bytes of data beyond the IPv4
header of the packet in error, which is not enough to include the
address fields of the IPv6 header. More modern IPv4 routers are
likely to return enough data beyond the IPv4 header to include the
entire IPv6 header and possibly even the data beyond that.

If the offending packet includes enough data, the encapsulating node
MAY extract the encapsulated IPv6 packet and use it to generate an
IPv6 ICMP message directed back to the originating IPv6 node, as
shown below:

+--------------+
IPv4 Header
dst = encaps
node
+--------------+
ICMP
Header
- - +--------------+
IPv4 Header
src = encaps
IPv4 node
+--------------+ - -
Packet IPv6
Header Original IPv6
in +--------------+ Packet -
Transport Can be used to
Error Header generate an
+--------------+ IPv6 ICMP
error message
~ Data ~ back to the source.

- - +--------------+ - -

IPv4 ICMP Error Message Returned to Encapsulating Node

3.5. IPv4 Header Construction

When encapsulating an IPv6 packet in an IPv4 datagram, the IPv4
header fields are set as follows:

Version:

4

IP Header Length in 32-bit words:

5 (There are no IPv4 options in the encapsulating header.)

Type of Service:

0. [Note that work underway in the IETF is redefining the Type
of Service byte and as a result future RFCs might define a
different behavior for the ToS byte when tunneling.]

Total Length:

Payload length from IPv6 header plus length of IPv6 and IPv4
headers (i.e. a constant 60 bytes).

Identification:

Generated uniquely as for any IPv4 packet transmitted by the
system.

Flags:

Set the Don't Fragment (DF) flag as specified in section 3.2.
Set the More Fragments (MF) bit as necessary if fragmenting.

Fragment offset:

Set as necessary if fragmenting.

Time to Live:

Set in implementation-specific manner.

Protocol:

41 (Assigned payload type number for IPv6)

Header Checksum:

Calculate the checksum of the IPv4 header.

Source Address:

IPv4 address of outgoing interface of the encapsulating node.

Destination Address:

IPv4 address of tunnel endpoint.

Any IPv6 options are preserved in the packet (after the IPv6 header).

3.6. Decapsulation

When an IPv6/IPv4 host or a router receives an IPv4 datagram that is
addressed to one of its own IPv4 address, and the value of the
protocol field is 41, it reassembles if the packet if it is
fragmented at the IPv4 level, then it removes the IPv4 header and
submits the IPv6 datagram to its IPv6 layer code.

The decapsulating node MUST be capable of reassembling an IPv4 packet
that is 1300 bytes (1280 bytes plus IPv4 header).

The decapsulation is shown below:

+-------------+
IPv4
Header
+-------------+ +-------------+
IPv6 IPv6
Header Header
+-------------+ +-------------+
Transport Transport
Layer ===> Layer
Header Header
+-------------+ +-------------+

~ Data ~ ~ Data ~

+-------------+ +-------------+

Decapsulating IPv6 from IPv4

When decapsulating the packet, the IPv6 header is not modified.
[Note that work underway in the IETF is redefining the Type of
Service byte and as a result future RFCs might define a different
behavior for the ToS byte when decapsulating a tunneled packet.] If
the packet is subsequently forwarded, its hop limit is decremented by
one.

As part of the decapsulation the node SHOULD silently discard a
packet with an invalid IPv4 source address such as a multicast
address, a broadcast address, 0.0.0.0, and 127.0.0.1. In general it
SHOULD apply the rules for martian filtering in [18] and ingress
filtering [13] on the IPv4 source address.

The encapsulating IPv4 header is discarded.

After the decapsulation the node SHOULD silently discard a packet
with an invalid IPv6 source address. This includes IPv6 multicast
addresses, the unspecified address, and the loopback address but also
IPv4-compatible IPv6 source addresses where the IPv4 part of the
address is an (IPv4) multicast address, broadcast address, 0.0.0.0,
or 127.0.0.1. In general it SHOULD apply the rules for martian
filtering in [18] and ingress filtering [13] on the IPv4-compatible
source address.

The decapsulating node performs IPv4 reassembly before decapsulating
the IPv6 packet. All IPv6 options are preserved even if the
encapsulating IPv4 packet is fragmented.

After the IPv6 packet is decapsulated, it is processed almost the
same as any received IPv6 packet. The only difference being that a
decapsulated packet MUST NOT be forwarded unless the node has been
explicitly configured to forward such packets for the given IPv4
source address. This configuration can be implicit in e.g., having a
configured tunnel which matches the IPv4 source address. This
restriction is needed to prevent tunneling to be used as a tool to
circumvent ingress filtering [13].

3.7. Link-Local Addresses

Both the configured and automatic tunnels are IPv6 interfaces (over
the IPv4 "link layer") thus MUST have link-local addresses. The
link-local addresses are used by routing protocols operating over the
tunnels.

The Interface Identifier [14] for such an Interface SHOULD be the
32-bit IPv4 address of that interface, with the bytes in the same
order in which they would appear in the header of an IPv4 packet,
padded at the left with zeros to a total of 64 bits. Note that the

"Universal/Local" bit is zero, indicating that the Interface
Identifier is not globally unique. When the host has more than one
IPv4 address in use on the physical interface concerned, an
administrative choice of one of these IPv4 addresses is made.

The IPv6 Link-local address [14] for an IPv4 virtual interface is
formed by appending the Interface Identifier, as defined above, to
the prefix FE80::/64.

+-------+-------+-------+-------+-------+-------+------+------+
FE 80 00 00 00 00 00 00
+-------+-------+-------+-------+-------+-------+------+------+
00 00 00 00 IPv4 Address
+-------+-------+-------+-------+-------+-------+------+------+

3.8. Neighbor Discovery over Tunnels

Automatic tunnels and unidirectional configured tunnels are
considered to be unidirectional. Thus the only ASPects of Neighbor
Discovery [7] and Stateless Address Autoconfiguration [5] that apply
to these tunnels is the formation of the link-local address.

If an implementation provides bidirectional configured tunnels it
MUST at least accept and respond to the probe packets used by
Neighbor Unreachability Detection [7]. Such implementations SHOULD
also send NUD probe packets to detect when the configured tunnel
fails at which point the implementation can use an alternate path to
reach the destination. Note that Neighbor Discovery allows that the
sending of NUD probes be omitted for router to router links if the
routing protocol tracks bidirectional reachability.

For the purposes of Neighbor Discovery the automatic and configured
tunnels specified in this document as assumed to NOT have a link-
layer address, even though the link-layer (IPv4) does have address.
This means that a sender of Neighbor Discovery packets

- SHOULD NOT include Source Link Layer Address options or Target
Link Layer Address options on the tunnel link.

- MUST silently ignore any received SLLA or TLLA options on the
tunnel link.

4. Configured Tunneling

In configured tunneling, the tunnel endpoint address is determined
from configuration information in the encapsulating node. For each
tunnel, the encapsulating node must store the tunnel endpoint
address. When an IPv6 packet is transmitted over a tunnel, the

tunnel endpoint address configured for that tunnel is used as the
destination address for the encapsulating IPv4 header.

The determination of which packets to tunnel is usually made by
routing information on the encapsulating node. This is usually done
via a routing table, which directs packets based on their destination
address using the prefix mask and match technique.

4.1. Default Configured Tunnel

IPv6/IPv4 hosts that are connected to datalinks with no IPv6 routers
MAY use a configured tunnel to reach an IPv6 router. This tunnel
allows the host to communicate with the rest of the IPv6 Internet
(i.e. nodes with IPv6-native addresses). If the IPv4 address of an
IPv6/IPv4 router bordering the IPv6 backbone is known, this can be
used as the tunnel endpoint address. This tunnel can be configured
into the routing table as an IPv6 "default route". That is, all IPv6
destination addresses will match the route and could potentially
traverse the tunnel. Since the "mask length" of such a default route
is zero, it will be used only if there are no other routes with a
longer mask that match the destination. The default configured
tunnel can be used in conjunction with automatic tunneling, as
described in section 5.4.

4.2. Default Configured Tunnel using IPv4 "Anycast Address"

The tunnel endpoint address of such a default tunnel could be the
IPv4 address of one IPv6/IPv4 router at the border of the IPv6
backbone. Alternatively, the tunnel endpoint could be an IPv4
"anycast address". With this approach, multiple IPv6/IPv4 routers at
the border advertise IPv4 reachability to the same IPv4 address. All
of these routers accept packets to this address as their own, and
will decapsulate IPv6 packets tunneled to this address. When an
IPv6/IPv4 node sends an encapsulated packet to this address, it will
be delivered to only one of the border routers, but the sending node
will not know which one. The IPv4 routing system will generally
carry the traffic to the closest router.

Using a default tunnel to an IPv4 "anycast address" provides a high
degree of robustness since multiple border router can be provided,
and, using the normal fallback mechanisms of IPv4 routing, traffic
will automatically switch to another router when one goes down.
However, care must be taking when using such a default tunnel to
prevent different IPv4 fragments from arriving at different routers
for reassembly. This can be prevented by either avoiding
fragmentation of the encapsulated packets (by ensuring an IPv4 MTU of
at least 1300 bytes) or by preventing frequent changes to IPv4
routing.

4.3. Ingress Filtering

The decapsulating node MUST verify that the tunnel source address is
acceptable before forwarding decapsulated packets to avoid
circumventing ingress filtering [13]. Note that packets which are
delivered to transport protocols on the decapsulating node SHOULD NOT
be subject to these checks. For bidirectional configured tunnels
this is done by verifying that the source address is the IPv4 address
of the other end of the tunnel. For unidirectional configured
tunnels the decapsulating node MUST be configured with a list of
source IPv4 address prefixes that are acceptable. Such a list MUST
default to not having any entries i.e. the node has to be explicitly
configured to forward decapsulated packets received over
unidirectional configured tunnels.

5. Automatic Tunneling

In automatic tunneling, the tunnel endpoint address is determined by
the IPv4-compatible destination address of the IPv6 packet being
tunneled. Automatic tunneling allows IPv6/IPv4 nodes to communicate
over IPv4 routing infrastructures without pre-configuring tunnels.

5.1. IPv4-Compatible Address Format

IPv6/IPv4 nodes that perform automatic tunneling are assigned IPv4-
compatible address. An IPv4-compatible address is identified by an
all-zeros 96-bit prefix, and holds an IPv4 address in the low-order
32-bits. IPv4-compatible addresses are structured as follows:

96-bits 32-bits
+--------------------------------------+--------------+
0:0:0:0:0:0 IPv4 Address
+--------------------------------------+--------------+
IPv4-Compatible IPv6 Address Format

IPv4-compatible addresses are assigned exclusively to nodes that
support automatic tunneling. A node SHOULD be configured with an
IPv4-compatible address only if it is prepared to accept IPv6 packets
destined to that address encapsulated in IPv4 packets destined to the
embedded IPv4 address.

An IPv4-compatible address is globally unique as long as the IPv4
address is not from the private IPv4 address space [15]. An
implementation SHOULD behave as if its IPv4-compatible address(es)
are assigned to the node's automatic tunneling interface, even if the
implementation does not implement automatic tunneling using a concept
of interfaces. Thus the IPv4-compatible address SHOULD NOT be viewed
as being attached to e.g. an Ethernet interface i.e. implications

should not use the Neighbor Discovery mechanisms like NUD [7] at the
Ethernet. Any such interactions should be done using the
encapsulated packets i.e. over the automatic tunneling (conceptual)
interface.

5.2. IPv4-Compatible Address Configuration

An IPv6/IPv4 node with an IPv4-compatible address uses that address
as one of its IPv6 addresses, while the IPv4 address embedded in the
low-order 32-bits serves as the IPv4 address for one of its
interfaces.

An IPv6/IPv4 node MAY acquire its IPv4-compatible IPv6 addresses via
IPv4 address configuration protocols. It MAY use any IPv4 address
configuration mechanism to acquire its IPv4 address, then "map" that
address into an IPv4-compatible IPv6 address by pre-pending it with
the 96-bit prefix 0:0:0:0:0:0. This mode of configuration allows
IPv6/IPv4 nodes to "leverage" the installed base of IPv4 address
configuration servers.

The specific algorithm for acquiring an IPv4-compatible address using
IPv4-based address configuration protocols is as follows:

1) The IPv6/IPv4 node uses standard IPv4 mechanisms or protocols to
acquire the IPv4 address for one of its interfaces. These
include:

- The Dynamic Host Configuration Protocol (DHCP) [2]

- The Bootstrap Protocol (BOOTP) [1]

- The Reverse Address Resolution Protocol (RARP) [9]

- Manual configuration

- Any other mechanism which accurately yields the node's own IPv4
address

2) The node uses this address as the IPv4 address for this interface.

3) The node prepends the 96-bit prefix 0:0:0:0:0:0 to the 32-bit IPv4
address that it acquired in step (1). The result is an IPv4-
compatible IPv6 address with one of the node's IPv4-addresses
embedded in the low-order 32-bits. The node uses this address as
one of its IPv6 addresses.

5.3. Automatic Tunneling Operation

In automatic tunneling, the tunnel endpoint address is determined
from the packet being tunneled. If the destination IPv6 address is
IPv4-compatible, then the packet can be sent via automatic tunneling.
If the destination is IPv6-native, the packet can not be sent via
automatic tunneling.

A routing table entry can be used to direct automatic tunneling. An
implementation can have a special static routing table entry for the
prefix 0:0:0:0:0:0/96. (That is, a route to the all-zeros prefix
with a 96-bit mask.) Packets that match this prefix are sent to a
pseudo-interface driver which performs automatic tunneling. Since
all IPv4-compatible IPv6 addresses will match this prefix, all
packets to those destinations will be auto-tunneled.

Once it is delivered to the automatic tunneling module, the IPv6
packet is encapsulated within an IPv4 header according to the rules
described in section 3. The source and destination addresses of the
encapsulating IPv4 header are assigned as follows:

Destination IPv4 address:

Low-order 32-bits of IPv6 destination address

Source IPv4 address:

IPv4 address of interface the packet is sent via

The automatic tunneling module always sends packets in this
encapsulated form, even if the destination is on an attached
datalink.

The automatic tunneling module MUST NOT send to IPv4 broadcast or
multicast destinations. It MUST drop all IPv6 packets destined to
IPv4-compatible destinations when the embedded IPv4 address is
broadcast, multicast, the unspecified (0.0.0.0) address, or the
loopback address (127.0.0.1). Note that the sender can only tell if
an address is a network or subnet broadcast for broadcast addresses
assigned to directly attached links.

5.4. Use With Default Configured Tunnels

Automatic tunneling is often used in conjunction with the default
configured tunnel technique. "Isolated" IPv6/IPv4 hosts -- those
with no on-link IPv6 routers -- are configured to use automatic
tunneling and IPv4-compatible IPv6 addresses, and have at least one
default configured tunnel to an IPv6 router. That IPv6 router is

configured to perform automatic tunneling as well. These isolated
hosts send packets to IPv4-compatible destinations via automatic
tunneling and packets for IPv6-native destinations via the default
configured tunnel. IPv4-compatible destinations will match the 96-
bit all-zeros prefix route discussed in the previous section, while
IPv6-native destinations will match the default route via the
configured tunnel. Reply packets from IPv6-native destinations are
routed back to the an IPv6/IPv4 router which delivers them to the
original host via automatic tunneling. Further examples of the
combination of tunneling techniques are discussed in [12].

5.5. Source Address Selection

When an IPv6/IPv4 node originates an IPv6 packet, it must select the
source IPv6 address to use. IPv6/IPv4 nodes that are configured to
perform automatic tunneling may be configured with global IPv6-native
addresses as well as IPv4-compatible addresses. The selection of
which source address to use will determine what form the return
traffic is sent via. If the IPv4-compatible address is used, the
return traffic will have to be delivered via automatic tunneling, but
if the IPv6-native address is used, the return traffic will not be
automatic-tunneled. In order to make traffic as symmetric as
possible, the following source address selection preference is
RECOMMENDED:

Destination is IPv4-compatible:

Use IPv4-compatible source address associated with IPv4 address
of outgoing interface

Destination is IPv6-native:

Use IPv6-native address of outgoing interface

If an IPv6/IPv4 node has no global IPv6-native address, but is
originating a packet to an IPv6-native destination, it MAY use its
IPv4-compatible address as its source address.

5.6. Ingress Filtering

The decapsulating node MUST verify that the encapsulated packets are
acceptable before forwarding decapsulated packets to avoid
circumventing ingress filtering [13]. Note that packets which are
delivered to transport protocols on the decapsulating node SHOULD NOT
be subject to these checks. Since automatic tunnels always
encapsulate to the destination (i.e. the IPv4 destination will be
the destination) any packet received over an automatic tunnel SHOULD
NOT be forwarded.

6. Acknowledgments

We would like to thank the members of the IPng working group and the
Next Generation Transition (ngtrans) working group for their many
contributions and extensive review of this document. Special thanks
are due to Jim Bound, Ross Callon, and Bob Hinden for many helpful
suggestions and to John Moy for suggesting the IPv4 "anycast address"
default tunnel technique.

7. Security Considerations

Tunneling is not known to introduce any security holes except for the
possibility to circumvent ingress filtering [13]. This is prevented
by requiring that decapsulating routers only forward packets if they
have been configured to accept encapsulated packets from the IPv4
source address in the receive packet. Additionally, in the case of
automatic tunneling, nodes are required by not forwarding the
decapsulated packets since automatic tunneling ends the tunnel and
the destination.

8. Authors' Addresses

Robert E. Gilligan
FreeGate Corp
1208 E. Arques Ave
Sunnyvale, CA 94086
USA

Phone: +1-408-617-1004
Fax: +1-408-617-1010
EMail: gilligan@freegate.com

Erik Nordmark
Sun Microsystems, Inc.
901 San Antonio Rd.
Palo Alto, CA 94303
USA

Phone: +1-650-786-5166
Fax: +1-650-786-5896
EMail: nordmark@eng.sun.com

9. References

[1] Croft, W. and J. Gilmore, "Bootstrap Protocol", RFC951,
September 1985.

[2] Droms, R., "Dynamic Host Configuration Protocol", RFC1541,
October 1993.

[3] Carpenter, B. and C. Jung, "Transmission of IPv6 over IPv4
Domains without Explicit Tunnels", RFC2529, March 1999.

[4] Deering, S. and R. Hinden, "Internet Protocol, Version 6 (IPv6)
Specification", RFC2460, December 1998.

[5] Thomson, S. and T. Narten, "IPv6 Stateless Address
Autoconfiguration," RFC2462, December 1998.

[6] Crawford, M., Thomson, S., and C. Huitema. "DNS Extensions to
Support IPv6 Address Allocation and Renumbering", RFC2874, July
2000.

[7] Narten, T., Nordmark, E. and W. Simpson, "Neighbor Discovery for
IP Version 6 (IPv6)", RFC2461, December 1998.

[8] Mogul, J. and S. Deering, "Path MTU Discovery", RFC1191,
November 1990.

[9] Finlayson, R., Mann, T., Mogul, J. and M. Theimer, "Reverse
Address Resolution Protocol", STD 38, RFC903, June 1984.

[10] Braden, R., "Requirements for Internet Hosts - Communication
Layers", STD 3, RFC1122, October 1989.

[11] Kent, C. and J. Mogul, "Fragmentation Considered Harmful". In
Proc. SIGCOMM '87 Workshop on Frontiers in Computer
Communications Technology. August 1987.

[12] Callon, R. and D. HaSKIN, "Routing Aspects of IPv6 Transition",
RFC2185, September 1997.

[13] Ferguson, P. and D. Senie, "Network Ingress Filtering: Defeating
Denial of Service Attacks which employ IP Source Address
Spoofing", RFC2267, January 1998.

[14] Hinden, R. and S. Deering, "IP Version 6 Addressing
Architecture", RFC2373, July 1998.

[15] Rechter, Y., Moskowitz, B., Karrenberg, D., de Groot, G.J. and
E. Lear, "Address Allocation for Private Internets", BCP 5, RFC
1918, February 1996.

[16] Bradner, S., "Key words for use in RFCs to Indicate Requirement
Levels", BCP 14, RFC2119, March 1997.

[17] Thaler, D., "IP Tunnel MIB", RFC2667, August 1999.

[18] Baker, F., "Requirements for IP Version 4 Routers", RFC1812,
June 1995.

10. Changes from RFC1933

- Deleted section 3.1.1 (IPv4 loopback address) in order to prevent
it from being mis-construed as requiring routers to filter the
address ::127.0.0.1, which would put another test in the
forwarding path for IPv6 routers.

- Deleted section 4.4 (Default Sending Algorithm). This section
allowed nodes to send packets in "raw form" to IPv4-compatible
destinations on the same datalink. Implementation experience has
shown that this adds complexity which is not justified by the
minimal savings in header overhead.

- Added definitions for operating modes for IPv6/IPv4 nodes.

- Revised DNS section to clarify resolver filtering and ordering
options.

- Re-wrote the discussion of IPv4-compatible addresses to clarify
that they are used exclusively in conjunction with the automatic
tunneling mechanism. Re-organized document to place definition of
IPv4-compatible address format with description of automatic
tunneling.

- Changed the term "IPv6-only address" to "IPv6-native address" per
current usage.

- Updated to algorithm for determining tunnel MTU to reflect the
change in the IPv6 minimum MTU from 576 to 1280 bytes [4].

- Deleted the definition for the term "IPv6-in-IPv4 encapsulation."
It has not been widely used.

- Revised IPv4-compatible address configuration section (5.2) to
recognize multiple interfaces.

- Added discussion of source address selection when using IPv4-
compatible addresses.

- Added section on the combination of the default configured
tunneling technique with hosts using automatic tunneling.

- Added prohibition against automatic tunneling to IPv4 broadcast or
multicast destinations.

- Clarified that configured tunnels can be unidirectional or
bidirectional.

- Added description of bidirectional virtual links as another type
of tunnels. Nodes MUST respond to NUD probes on such links and
SHOULD send NUD probes.

- Added reference to [16] specification as an alternative for
tunneling over a multicast capable IPv4 cloud.

- Clarified that IPv4-compatible addresses are assigned exclusively
to nodes that support automatic tunnels i.e. nodes that can
receive such packets.

- Added text about formation of link-local addresses and use of
Neighbor Discovery on tunnels.

- Added restriction that decapsulated packets not be forwarded
unless the source address is acceptable to the decapsulating
router.

- Clarified that decapsulating nodes MUST be capable of reassembling
an IPv4 packet that is 1300 bytes (1280 bytes plus IPv4 header).

- Clarified that when using a default tunnel to an IPv4 "anycast
address" the network must either have an IPv4 MTU of least 1300
bytes (to avoid fragmentation of minimum size IPv6 packets) or be
configured to avoid frequent changes to IPv4 routing to the
"anycast address" (to avoid different IPv4 fragments arriving at
different tunnel endpoints).

- Using A6/AAAA instead of AAAA to reference IPv6 address records in
the DNS.

- Specified when to put IPv6 addresses in the DNS.

- Added reference to the tunnel mib for TTL specification for the
tunnels.

- Added a table of contents.

- Added recommendations for use of source and target link layer
address options for the tunnel links.

- Added checks in the decapsulation checking both an IPv4-compatible
IPv6 source address and the outer IPv4 source addresses for
multicast, broadcast, all-zeros etc.

11. Full Copyright Statement

Copyright (C) The Internet Society (2000). All Rights Reserved.

This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.

The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.

This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Acknowledgement

Funding for the RFCEditor function is currently provided by the
Internet Society.