Network Working Group R. Callon
Request for Comments: 2185 Cascade Communications Co.
Category: Informational D. Haskin
Bay Networks Inc.
September 1997
Routing Aspects Of IPv6 Transition
Status of this memo
This memo provides information for the Internet community. This memo
does not specify an Internet standard of any kind. Distribution of
this memo is unlimited.
Abstract
This document gives an overview of the routing aspects of the IPv6
transition. It is based on the protocols defined in the document
"Transition Mechanisms for IPv6 Hosts and Routers" [1]. Readers
should be familiar with the transition mechanisms before reading this
document.
The proposals contained in this document are based on the work of the
Ngtrans working group.
This paper uses the following terminology:
node - a protocol module that implements IPv4 or IPv6.
router - a node that forwards packets not explicitly
addressed to itself.
host - any node that is not a router.
border router - a router that forwards packets across
routing domain boundaries.
link - a communication facility or medium over which
nodes can communicate at the link layer, i.e., the layer
immediately below internet layer.
interface - a node's attachment to a link.
address - an network layer identifier for an interface or
a group of interfaces.
Callon & Haskin Informational [Page 1]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
neighbors - nodes attached to the same link.
routing domain - a collection of routers which coordinate
routing knowledge using a single routing protocol.
routing region (or just "region") - a collection of routers
interconnected by a single internet protocol (e.g. IPv6)
and coordinating their routing knowledge using routing
protocols from a single internet protocol stack. A
routing region may be a superset of a routing domain.
tunneling - encapsulation of protocol A within protocol B,
such that A treats B as though it were a datalink layer.
reachability information - information describing the set of
reachable destinations that can be used for packet
forwarding decisions.
routing information - same as reachability information.
address prefix - the high-order bits in an address.
routing prefix - address prefix that expresses destinations
which have addresses with the matching address prefixes.
It is used by routers to advertise what systems they are
capable of reaching.
route leaking - advertisement of network layer reachability
information across routing region boundaries.
This document gives an overview of the routing aspects of IPv4 to
IPv6 transition. The approach outlined here is designed to be
compatible with the existing mechanisms for IPv6 transition [1].
During an extended IPv4-to-IPv6 transition period, IPv6-based systems
must coexist with the installed base of IPv4 systems. In such a dual
internetworking protocol environment, both IPv4 and IPv6 routing
infrastructure will be present. Initially, deployed IPv6-capable
domains might not be globally interconnected via IPv6-capable
internet infrastructure and therefore may need to communicate across
IPv4-only routing regions. In order to achieve dynamic routing in
such a mixed environment, there need to be mechanisms to globally
distribute IPv6 network layer reachability information between
dispersed IPv6 routing regions. The same techniques can be used in
later stages of IPv4-to-IPv6 transition to route IPv4 packets between
isolated IPv4-only routing region over IPv6 infrastructure.
Callon & Haskin Informational [Page 2]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
The IPng transition provides a dual-IP-layer transition, augmented by
use of encapsulation where necessary and appropriate. Routing issues
related to this transition include:
(1) Routing for IPv4 packets
(2) Routing for IPv6 packets
(2a) IPv6 packets with IPv6-native addresses
(2b) IPv6 packets with IPv4-compatible addresses
(3) Operation of manually configured static tunnels
(4) Operation of automatic encapsulation
(4a) Locating encapsulators
(4b) Ensuring that routing is consist with
encapsulation
Basic mechanisms required to accomplish these goals include: (i)
Dual-IP-layer Route Computation; (ii) Manual configuration of point-
to-point tunnels; and (iii) Route leaking to support automatic
encapsulation.
The basic mechanism for routing of IPv4 and IPv6 involves dual-IP-
layer routing. This implies that routes are separately calculated for
IPv4 addresses and for IPv6 addressing. This is discussed in more
detail in section 3.1.
Tunnels (either IPv4 over IPv6, or IPv6 over IPv4) may be manually
configured. For example, in the early stages of transition this may
be used to allow two IPv6 domains to interact over an IPv4
infrastructure. Manually configured static tunnels are treated as if
they were a normal data link. This is discussed in more detail in
section 3.2.
Use of automatic encapsulation, where the IPv4 tunnel endpoint
address is determined from the IPv4 address embedded in the IPv4-
compatible destination address of IPv6 packet, requires consistency
of routes between IPv4 and IPv6 routing domains for destinations
using IPv4-compatible addresses. For example, consider a packet which
starts off as an IPv6 packet, but then is encapsulated in an IPv4
packet in the middle of its path from source to destination. This
packet must locate an encapsulator at the correct part of its path.
Also, this packet has to follow a consistent route for the entire
path from source to destination. This is discussed in more detail in
section 3.3.
The mechanisms for tunneling IPv6 over IPv4 are defined in the
transition mechanisms specification [1].
Callon & Haskin Informational [Page 3]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
In the basic dual-IP-layer transition scheme, routers may
independently support IPv4 and IPv6 routing. Other parts of the
transition, such as DNS support, and selection by the source host of
which packet format to transmit (IPv4 or IPv6) are discussed in [1].
Forwarding of IPv4 packets is based on routes learned through running
IPv4-specific routing protocols. Similarly, forwarding of IPv6
packets (including IPv6-packets with IPv4-compatible addresses) is
based on routes learned through running IPv6-specific routing
protocols. This implies that separate instances of routing protocols
are used for IPv4 and for IPv6 (although note that this could consist
of two instances of OSPF and/or two instances of RIP, since both OSPF
and RIP are capable of supporting both IPv4 and IPv6 routing).
A minor enhancement would be to use an single instance of an
integrated routing protocol to support routing for both IPv4 and
IPv6. At the time that this is written there is no protocol which
has yet been enhanced to support this. This minor enhancement does
not change the basic dual-IP-layer nature of the transition.
For initial testing of IPv6 with IPv4-compatible addresses, it may be
useful to allow forwarding of IPv6 packets without running any IPv6-
compatible routing protocol. In this case, a dual (IPv4 and IPv6)
router could run routing protocols for IPv4 only. It then forwards
IPv4 packets based on routes learned from IPv4 routing protocols.
Also, it forwards IPv6 packets with an IPv4-compatible destination
address based on the route for the associated IPv4 address. There are
a couple of drawbacks with this approach: (i) It does not
specifically allow for routing of IPv6 packets via IPv6-capable
routers while avoiding and routing around IPv4-only routers; (ii) It
does not produce routes for "non-compatible" IPv6 addresses. With
this method the routing protocol does not tell the router whether
neighboring routers are IPv6-compatible. However, neighbor discovery
may be used to determine this. Then if an IPv6 packet needs to be
forwarded to an IPv4-only router it can be encapsulated to the
destination host.
Tunneling techniques are already widely deployed for bridging non-IP
network layer protocols (e.g. AppleTalk, CLNP, IPX) over IPv4 routed
infrastructure. IPv4 tunneling is an encapsulation of arbitrary
packets inside IPv4 datagrams that are forwarded over IPv4
infrastructure between tunnel endpoints. For a tunneled protocol, a
tunnel appears as a single-hop link (i.e. routers that establish a
Callon & Haskin Informational [Page 4]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
tunnel over a network layer infrastructure can inter-operate over the
tunnel as if it were a one-hop, point-to-point link). Once a tunnel
is established, routers at the tunnel endpoints can establish routing
adjacencies and exchange routing information. Describing the
protocols for performing encapsulation is outside the scope of this
paper (see [1]). Static point-to-point tunnels may also be
established between a host and a router, or between two hosts. Again,
each manually configured point-to-point tunnel is treated as if it
was a simple point-to-point link.
Automatic tunneling may be used when both the sending and destination
nodes are connected by IPv4 routing. In order for automatic
tunneling to work, both nodes must be assigned IPv4-compatible IPv6
addresses. Automatic tunneling can be especially useful where either
source or destination hosts (or both) do not have any adjacent IPv6-
capable router. Note that by "adjacent router", this includes
routers which are logically adjacent by virtue of a manually
configured point-to-point tunnel (which is treated as if it is a
simple point-to-point link).
With automatic tunneling, the resulting IPv4 packet is forwarded by
IPv4 routers as a normal IPv4 packet, using IPv4 routes learned from
routing protocols. There are therefore no special issues related to
IPv4 routing in this case. There are however routing issues relating
to how IPv6 routing works in a manner which is compatible with
automatic tunneling, and how tunnel endpoint addresses are selected
during the encapsulation process. Automatic tunneling is useful from
a source host to the destination host, from a source host to a
router, and from a router to the destination host. Mechanisms for
automatic tunneling from a router to another router are not currently
defined.
If both source and destination hosts make use of IPv4-compatible IPv6
addresses, then it is possible for automatic tunneling to be used for
the entire path from the source host to the destination host. In this
case, the IPv6 packet is encapsulated in an IPv4 packet by the source
host, and is forwarded by routers as an IPv4 packet all the way to
the destination host. This allows initial deployment of IPv6-capable
hosts to be done prior to the update of any routers.
Callon & Haskin Informational [Page 5]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
A source host may make use of Host to Host automatic tunneling
provided that the following are both true:
- the source address is an IPv4-compatible IPv6 address.
- the destination address is an IPv4-compatible IPv6 address.
- the source host does know of one or more neighboring IPv4-
capable routers, or the source and destination are on the
same subnet.
If all of these requirements are true, then the source host may
encapsulate the IPv6 packet in an IPv4 packet, using a source IPv4
address which is extracted from the associated source IPv6 address,
and using a destination IPv4 address which is extracted from the
associated destination IPv6 address.
Where host to host automatic tunneling is used, the packet is
forwarded as a normal IPv4 packet for its entire path, and is
decapsulated (i.e., the IPv4 header is removed) only by the
destination host.
In some cases "configured default" tunneling may be used to
encapsulate the IPv6 packet for transmission from the source host to
an IPv6-backbone. However, this requires that the source host be
configured with an IPv4 address to use for tunneling to the backbone.
Configured default tunneling is particularly useful if the source
host does not know of any local IPv6-capable router (implying that
the packet cannot be forwarded as a normal IPv6 packet directly over
the link layer), and when the destination host does not have an
IPv4-compatible IPv6 address (implying that host to host tunneling
cannot be used).
Host to router configured default tunneling may optionally also be
used even when the host does know of a local IPv6 router. In this
case it is a policy decision whether the host prefers to send a
native IPv6 packet to the IPv6-capable router or prefers to send an
encapsulated packet to the configured tunnel endpoint.
Similarly host to router default configured tunneling may be used
even when the destination address is an IPv4-compatible IPv6 address.
In this case for example a policy decision may be made to prefer
tunneling for part of the path and native IPv6 for part of the path,
or alternatively to use tunneling for the entire path from source
host to destination host.
Callon & Haskin Informational [Page 6]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
A source host may make use of host to router configured default
tunneling provided that ALL of the following are true:
- the source address is an IPv4-compatible IPv6 address.
- the source host does know of one or more neighboring IPv4-
capable routers
- the source host has been configured with an IPv4 address of
an dual router which can serve as the tunnel endpoint.
If all of these requirements are true, then the source host may
encapsulate the IPv6 packet in an IPv4 packet, using a source IPv4
address which is extracted from the associated source IPv6 address,
and using a destination IPv4 address which corresponds to the
configured address of the dual router which is serving as the tunnel
endpoint.
When host to router configured default tunneling is used, the packet
is forwarded as a normal IPv4 packet from the source host to the dual
router serving as tunnel endpoint, is decapsulated by the dual
router, and is then forwarded as a normal IPv6 packet by the tunnel
endpoint.
The dual router which is serving as the end point of the host to
router configured default tunnel must advertise reachability into
IPv4 routing sufficient to cause the encapsulated packet to be
forwarded to it.
The simplest approach is for a single IPv4 address to be assigned for
use as a tunnel endpoint. One or more dual routers, which have
connectivity to the IPv6 backbone and which are capable of serving as
tunnel endpoint, advertise a host route to this address into IPv4
routing in the IPv4-only region. Each dual host in the associated
IPv4-only region is configured with the address of this tunnel
endpoint and selects a route to this address for forwarding
encapsulated packet to a tunnel end point (for example, the nearest
tunnel end point, based on whatever metric(s) the local routing
protocol is using).
Finally, in some cases there may be some reason for specific hosts to
prefer one of several tunnel endpoints, while allowing all potential
tunnel endpoints to serve as backups in case the preferred endpoint
is not reachable. In this case, each dual router with IPv6 backbone
connectivity which is serving as potential tunnel endpoint is given a
unique IPv4 address taken from a single IPv4 address block (where the
IPv4 address block is assigned either to the organization
administering the IPv4-only region, or to the organization
Callon & Haskin Informational [Page 7]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
administering the local part of the IPv6 backbone). In the likely
case that there are much less than 250 such dual routers serving as
tunnel endpoints, we suggest using multiple IPv4 addresses selected
from a single 24-bit IPv4 address prefix for this purpose. Each dual
router then advertises two routes into the IPv4 region: A host route
corresponding to the tunnel endpoint address specifically assigned to
it, and also a standard (prefix) route to the associated IPv4 address
block. Each dual host in the IPv4-only region is configured with a
tunnel endpoint address which corresponds to the preferred tunnel
endpoint for it to use. If the associated dual router is operating,
then the packet will be delivered to it based upon the host route
that it is advertising into the IPv4-only region. However, if the
associated dual router is down, but some other dual router serving as
a potential tunnel endpoint is operating, then the packet will be
delivered to the nearest operating tunnel endpoint.
In some cases the source host may have direct connectivity to one or
more IPv6-capable routers, but the destination host might not have
direct connectivity to any IPv6-capable router. In this case,
provided that the destination host has an IPv4-compatible IPv6
address, normal IPv6 forwarding may be used for part of the packet's
path, and router to host tunneling may be used to get the packet from
an encapsulating dual router to the destination host.
In this case, the hard part is the IPv6 routing required to deliver
the IPv6 packet from the source host to the encapsulating router. For
this to happen, the encapsulating router has to advertise
reachability for the appropriate IPv4-compatible IPv6 addresses into
the IPv6 routing region. With this approach, all IPv6 packets
(including those with IPv4-compatible addresses) are routed using
routes calculated from native IPv6 routing. This implies that
encapsulating routers need to advertise into IPv6 routing specific
route entries corresponding to any IPv4-compatible IPv6 addresses
that belong to dual hosts which can be reached in an neighboring
IPv4-only region. This requires manual configuration of the
encapsulating routers to control which routes are to be injected into
IPv6 routing protocols. Nodes in the IPv6 routing region would use
such a route to forward IPv6 packets along the routed path toward the
router that injected (leaked) the route, at which point packets are
encapsulated and forwarded to the destination host using normal IPv4
routing.
Depending upon the extent of the IPv4-only and dual routing regions,
the leaking of routes may be relatively simple or may be more
complex. For example, consider a dual Internet backbone, connected
via one or two dual routers to an IPv4-only stub routing domain. In
Callon & Haskin Informational [Page 8]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
this case, it is likely that there is already one summary address
prefix which is being advertised into the Internet backbone in order
to summarize IPv4 reachability to the stub domain. In such a case,
the border routers would be configured to announce the IPv4 address
prefix into the IPv4 routing within the backbone, and also announce
the corresponding IPv4-compatible IPv6 address prefix into IPv6
routing within the backbone.
A more difficult case involves the border between a major Internet
backbone which is IPv4-only, and a major Internet backbone which
supports both IPv4 and IPv6. In this case, it requires that either
(i) the entire IPv4 routing table be fed into IPv6 routing in the
dual routing domain (implying a doubling of the size of the routing
tables in the dual domain); or (ii) Manual configuration is required
to determine which of the addresses contained in the Internet routing
table include one or more IPv6-capable systems, and only these
addresses be advertised into IPv6 routing in the dual domain.
Clearly tunneling is useful only if communication can be achieved in
both directions. However, different forms of tunneling may be used in
each direction, depending upon the local environment, the form of
address of the two hosts which are exchanging IPv6 packets, and the
policies in use.
Table 1 summarizes the form of tunneling that will result given each
possible combination of host capabilities, and given one possible set
of policy decisions. This table is derived directly from the
requirements for automatic tunneling discussed above.
The example in table 1 uses a specific set of policy decisions: It is
assumed in table 1 that the source host will transmit a native IPv6
where possible in preference over encapsulation. It is also assumed
that where tunneling is needed, host to host tunneling will be
preferred over host to router tunneling. Other combinations are
therefore possible if other policies are used.
Due to a specific policy choice, the default sending rules in [1] may
not be followed.
Note that IPv6-capable hosts which do not have any local IPv6 router
must be given an IPv4-compatible v6 address in order to make use of
their IPv6 capabilities. Thus, there are no entries for IPv6-capable
hosts which have an incompatible IPv6 address and which also do not
have any connectivity to any local IPv6 router. In fact, such hosts
could communicate with other IPv6 hosts on the same local network
without the use of a router. However, since this document focuses on
Callon & Haskin Informational [Page 9]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
routing and router implications of IPv6 transition, direct
communication between two hosts on the same local network without any
intervening router is outside the scope of this document.
Also, table 1 does not consider manually configured point-to-point
tunnels. Such tunnels are treated as if they were normal point-to-
point links. Thus any two IPv6-capable devices which have a manually
configured tunnel between them may be considered to be directly
connected.
-----------------+------------------+--------------------------
Host A | Host B | Result
-----------------+------------------+--------------------------
v4-compat. addr. | v4-compat. addr. | host to host tunneling
no local v6 rtr. | no local v6 rtr. | in both directions
-----------------+------------------+--------------------------
v4-compat. addr. | v4-compat. addr. | A->B: host to host tunnel
no local v6 rtr. | local v6 rtr. | B->A: v6 forwarding plus
| | rtr->host tunnel
-----------------+------------------+--------------------------
v4-compat. addr. | incompat. addr. | A->B: host to rtr tunnel
no local v6 rtr. | local v6 rtr. | plus v6 forwarding
| | B->A: v6 forwarding plus
| | rtr to host tunnel
-----------------+------------------+--------------------------
v4-compat. addr. | v4-compat. addr. | end to end native v6
local v6 rtr. | local v6 rtr. | in both directions
-----------------+------------------+--------------------------
v4-compat. addr. | incompat. addr. | end to end native v6
local v6 rtr. | local v6 rtr. | in both directions
-----------------+------------------+--------------------------
incompat. addr. | incompat. addr. | end to end native v6
local v6 rtr. | local v6 rtr. | in both directions
-----------------+------------------+--------------------------
Table 1: Summary of Automatic Tunneling Combinations
Figure 2 illustrates an example network with two regions A and B.
Region A is dual, meaning that the routers within region A are
capable of forwarding both IPv4 and IPv6. Region B is IPv4-only,
implying that the routers within region B are capable of routing only
IPv4. The illustrated routers R1 through R4 are dual. The illustrated
routers r5 through r9 are IPv4-only. Also assume that hosts H3
through H8 are dual. Thus H7 and H8 have been upgraded to be IPv6-
capable, even though they exist in a region in which the routers are
not IPv6-capable. However, host h1 and h2 are IPv4-only.
Callon & Haskin Informational [Page 10]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
......................... .......................
. . . .
. h1 . . |-h2 .
. | . . | .
. H3---R1--------R2---------------r5----r9----+ .
. | | . . | |-H7 .
. | | . . | .
. | | . . | .
. H4---R3--------R4---------------r6----r8-----H8 .
. . . .
......................... .......................
Region A (Dual Routers) Region B (IPv4-only Rtrs)
Figure 2: Example of Automatic Tunneling
Consider a packet from h1 to H8. In this case, since h1 is IPv4-only,
it will send an IPv4 packet. This packet will traverse regions A and
B as a normal IPv4 packet for the entire path. Routing will take
place using normal IPv4 routing methods, with no change from the
operation of the current IPv4 Internet (modulo normal advances in the
operation of IPv4, of course). Similarly, consider a return packet
from H8 to h1. Here again H8 will transmit an IPv4 packet, which will
be forwarded as a normal IPv4 packet for the entire path.
Consider a packet from H3 to H8. In this case, since H8 is in an
IPv4-only routing domain, we can assume that H8 uses an IPv4-
compatible IPv6 address. Since both source and destination are IPv6-
capable, H3 may transmit an IPv6 packet destined to H8. The packet
will be forwarded as far as R2 (or R4) as an IPv6 packet.
Router R2 (or R4) will then encapsulate the full IPv6 packet in an
IPv4 header for delivery to H8. In this case it is necessary for
routing of IPv6 within region A to be capable of delivering this
packet correctly to R2 (or R4). As explained in section 3.3, routers
R2 and R4 may inject routes to IPv4-compatible IPv6 addresses into
the IPv6 routing used within region A corresponding to the routes
which are available via IPv4 routing within region B.
Consider a return packet from H8 to H3. Again, since both source and
destination are IPv6-capable, a IPv6 packet may be transmitted by H8.
However, since H8 does not have any direct connectivity to an IPv6-
capable router, H8 must make use of an automatic tunnel. Which form
of automatic tunnel will be used depends upon the type of address
assigned to H3.
Callon & Haskin Informational [Page 11]
RFC 2185 Routing Aspects Of IPv6 Transition September 1997
If H3 is assigned an IPv4-compatible address, then the requirements
specified in section 3.3.1 will all be satisfied. In this case host
H8 may encapsulate the full IPv6 packet in an IPv4 header using a
source IPv4 address extracted from the IPv6 address of H8, and using
a destination IPv4 address extracted from the IPv6 address of H3.
If H3 has an IPv6-only address, then it is not possible for H8 to
extract an IPv4 address to use as the destination tunnel address from
the IPv6 address of H3. In this case H8 must use host to router
tunneling, as specified in section 3.3.2. In this case one or both of
R2 and R4 must have been configured with a tunnel endpoint IPv4
address (R2 and R4 may use either the same address or different
addresses for this purpose). R2 and/or R4 therefore advertise
reachability to the tunnel endpoint address to r5 and r6
(respectively), which advertise this reachability information into
region B. Also, H8 must have been configured to know which tunnel
endpoint address to use for host to router tunneling. This will
result in the IPv6 packet, encapsulated in an IPv4 header, to be
transmitted as far as the border router R2 or R4. The border router
will then strip off the IPv4 header, and forward the remaining IPv6
packet as a normal IPv6 packet using the normal IPv6 routing used in
region A.