Keyed IPv6 TunnelCisco Systemsmaciek@cisco.comCisco Systemsgiheron@cisco.comDeutsche Telekom AGrainer.schatzmayr@telekom.deAlcatel-Lucent, Inc.wim.henderickx@alcatel-lucent.com
Internet
L2TPEXT Working GroupThis document describes a tunnel
encapsulation for Ethernet over IPv6 with a mandatory 64-bit cookie for
connecting Layer 2 (L2)
Ethernet attachment circuits identified by IPv6 addresses. The
encapsulation is based on the Layer 2 Tunneling Protocol Version 3
(L2TPv3) over IP and does not use the
L2TPv3 control plane.L2TPv3, as defined in ,
provides a mechanism for tunneling Layer 2 (L2) "circuits" across a
packet-oriented data network (e.g., over IP), with multiple attachment
circuits multiplexed over a single pair of IP address endpoints (i.e., a
tunnel) using the L2TPv3 Session ID as a circuit discriminator.Implementing L2TPv3 over IPv6 provides the opportunity to utilize
unique IPv6 addresses to identify Ethernet attachment circuits directly,
leveraging the key property that IPv6 offers -- a vast number of unique IP
addresses. In this case, processing of the L2TPv3 Session ID may be
bypassed upon receipt, as each tunnel has one and only one associated
session. This local optimization does not hinder the ability to continue
supporting the multiplexing of circuits via the Session ID on the same
router for other L2TPv3 tunnels.There are various advantages to this approach when compared to the
"traditional" L2TPv3 approach of using a loopback address to terminate
the tunnel and then carrying multiple sessions over the tunnel.
These include better ECMP load balancing (since each tunnel has a
unique source/destination IPv6 address pair) and finer-grained
control when advertising tunnel endpoints using a routing protocol.The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in RFC 2119
.The L2TPv3 control plane defined in is not used for this encapsulation.
The management plane is used to create and maintain matching configurations
at either end of each tunnel. Local configuration by the management plane creates
a one-to-one mapping between the access-side L2 attachment circuit and the IP
address used in the network-side IPv6 encapsulation.The IPv6 L2TPv3 tunnel encapsulating device uniquely identifies each
Ethernet L2 attachment connection by a port ID or a combination of a port
ID and VLAN ID(s) on the access side and by a local IPv6 address on the
network side. The local IPv6 address also identifies the tunnel
endpoint. The local IPv6 addresses identifying L2TPv3 tunnels
SHOULD NOT be assigned from connected IPv6 subnets facing
towards remote tunnel endpoints, since that approach would
result in an IPv6 Neighbor Discovery cache entry per tunnel on
the next-hop router towards the remote tunnel endpoint. It is
RECOMMENDED that local IPv6 addresses identifying L2TPv3 tunnels
are assigned from dedicated subnets used only for such tunnel endpoints.Certain deployment scenarios may require using a single IPv6 address
(such as a unicast or anycast address assigned to a specific service instance,
for example, a virtual switch) to identify a tunnel endpoint for multiple IPv6
L2TPv3 tunnels. For such cases, the tunnel decapsulating device uses the local
IPv6 address to identify the service instance and the remote IPv6 address to
identify the individual tunnel within that service instance.As mentioned above, Session ID processing is not required, as each
keyed IPv6 tunnel has one and only one associated session. However, for
compatibility with existing implementations, the packets
need to be sent with the Session ID. Routers implementing L2TPv3
according to can be configured with multiple L2TPv3
tunnels, with one session per tunnel, to interoperate with routers
implementing the keyed IPv6 tunnel as specified by this
document. Note that as Session ID processing is not enabled
for keyed IPv6 tunnels, there can only be a single keyed
IPv6 tunnel between two IPv6 addresses.In line with , the 64-bit cookie is used for an
additional tunnel endpoint context check. This is the largest
cookie size permitted in . All packets MUST carry the 64-bit
L2TPv3 cookie field. The cookie MUST be 64 bits long in order to provide
sufficient protection against spoofing and brute-force blind insertion
attacks. The cookie values SHOULD be randomly selected.In the absence of the L2TPv3 control plane, the L2TPv3 encapsulating
router MUST be provided with a local configuration of the 64-bit cookie
for each local and remote IPv6 endpoint. Note that cookies are
asymmetric, so local and remote endpoints may send different cookie
values and, in fact, SHOULD do so. The value of the cookie MUST
be able to be changed at any time in a manner that does not drop
any legitimate tunneled packets, i.e., the receiver MUST be
configurable to accept two discrete cookies for a single tunnel
simultaneously. This enables the receiver to hold both the
'old' and 'new' cookie values during a change of cookie value. Cookie
values SHOULD be changed periodically by the management plane.Note that mandating a 64-bit cookie is a change from the optional
variable-length cookie of and that this requirement constrains
interoperability with existing implementations to those supporting
a 64-bit cookie. The management plane MUST NOT configure a keyed IP tunnel
unless both endpoints support the 64-bit cookie.The ingress router encapsulates the entire Ethernet frame,
without the preamble and Frame Check Sequence (FCS) in L2TPv3 as
per . The L2-specific sublayer
MAY be carried if Virtual Circuit Connectivity Verification (VCCV)
and/or frame sequencing is required,
but it SHOULD NOT be carried otherwise. The L2TPv3 packet is encapsulated
directly over IPv6 (i.e., no UDP header is carried).The ingress router MAY retain the FCS as per . Support for retaining the FCS and for
receiving packets with a retained FCS is OPTIONAL and, if present,
MUST be configurable. In the absence of the L2TPv3 control
plane, such configuration MUST be consistent for the two
endpoints of any given tunnel, i.e., if one router is configured
to retain the FCS, then the other router MUST be configured to
receive packets with the retained FCS. Any router configured to
retain FCS for a tunnel MUST retain FCS for all frames sent over
that tunnel. All routers implementing this specification MUST
support the ability to send frames without retaining the FCS and
to receive such frames.Any service-delimiting IEEE
802.1Q or IEEE 802.1ad
VLAN IDs -- S-tag, C-tag, or the tuple (S-tag, C-tag) -- are treated with
local significance within the Ethernet L2 port and MUST NOT be
forwarded over the IPv6 L2TPv3 tunnel.Note that the same approach may be used to transport protocols
other than Ethernet, though this is outside the scope of this
specification.The full encapsulation is as follows:The combined IPv6 and keyed IP tunnel header contains the following
fields:IPv6 Header. Note that:The traffic class may be set by the ingress router to ensure
correct Per-Hop Behavior (PHB) treatment by transit routers between the ingress and
egress and to correct QoS disposition at the egress router.The flow label, as defined in , may be
set by the ingress router to indicate a flow of packets from the client,
which may not be reordered by the network (if there is a requirement
for finer-grained ECMP load balancing rather than per-circuit load balancing).The next header will be set to 0x73 to indicate that the next header is
L2TPv3.In the "Static 1:1 Mapping" case described in Section 2, the IPv6
source address may correspond to a port
or port/VLAN being transported as an L2 circuit, or it may correspond to a virtual
interface terminating inside the router (e.g., if L2 circuits are being used
within a multipoint VPN or if an anycast address is being terminated on a set of
data-center virtual machines.)As with the source address, the IPv6 destination address may correspond to a
port or port/VLAN being transported as an L2 circuit or to a virtual interface.Session ID.
In the "Static 1:1 Mapping" case described in , the IPv6 address identifies an L2TPv3 session directly;
thus, at endpoints supporting one-stage resolution (IPv6 Address Only), the Session ID
SHOULD be ignored upon receipt. It is RECOMMENDED that the remote endpoint is configured
to set the Session ID to all ones (0xFFFFFFFF) for easy identification in case of
troubleshooting. For compatibility with other tunnel termination platforms
supporting only two-stage resolution (IPv6 Address + Session ID), this
specification recommends supporting explicit configuration of
Session ID to any value other than zero (including all ones). The Session ID of zero
MUST NOT be used, as it is reserved for use by L2TP control messages
as specified in . Note that the Session ID is
unidirectional; the sent and
received Session IDs at an endpoint may be different.Cookie.
The 64-bit cookie, configured and described as in .
All packets for a destined L2 circuit (or L2TPv3 Session) MUST match one of the cookie
values configured for that circuit. Any packets that do not contain a valid cookie
value MUST be discarded (see for more details).L2-Specific Sublayer (Optional).
As noted above, this will be present if VCCV
and/or frame sequencing is required. If VCCV is required, then any frames with
bit 0 (the "V-bit") set are VCCV messages. If frame sequencing is required, then
any frames with bit 1 (the "S-bit") set have a valid frame sequence number in
bits 8-31.Payload (variable).
As noted above, the preamble and any service-delimiting
tags MUST be stripped before encapsulation, and the FCS MUST be stripped unless FCS
retention is configured at both ingress and egress routers. Since a new FCS is added
at each hop when the encapsulating IP packet is transmitted, the payload is protected
against bit errors.Using tunnel encapsulation of Ethernet L2 datagrams in IPv6 will reduce the
effective MTU allowed for the encapsulated traffic.The recommended solution to deal with this problem is for the network
operator to increase the MTU size of all the links between the devices
acting as IPv6 L2TPv3 tunnel endpoints to accommodate both the IPv6
L2TPv3 encapsulation header and the Ethernet L2 datagram without requiring
fragmentation of the IPv6 packet.It is RECOMMENDED that routers implementing this
specification implement IPv6 Path MTU (PMTU) discovery as defined in to confirm that the path over
which packets are sent has sufficient MTU to transport a maximum-length
Ethernet frame plus encapsulation overhead.Routers implementing this specification MAY implement L2TPv3
fragmentation (as defined in Section 5 of ). In the absence of the L2TPv3 control plane,
it is RECOMMENDED that fragmentation (if implemented) is locally
configured on a per-tunnel basis. Fragmentation configuration
MUST be consistent between the two ends of a tunnel.It is NOT RECOMMENDED for routers implementing this
specification to enable IPv6 fragmentation (as defined in
Section 4.5 of ) for keyed IP tunnels.Operations, Administration, and Maintenance (OAM) is an important
consideration when providing circuit-oriented
services such as those described in this document; it is all the more important
in the absence of a dedicated tunnel control plane, as OAM becomes the
only way to detect failures in the tunnel overlay.Note that in the context of keyed IP tunnels, failures in the IPv6
underlay network can be detected using the usual methods such as through
the routing protocol, including the use of single-hop Bidirectional
Forwarding Detection (BFD) to rapidly detect link
failures. Multihop BFD MAY also be enabled between tunnel endpoints as per
.Since keyed IP tunnels always carry an Ethernet payload and since
OAM at the tunnel layer is unable to detect failures in the Ethernet
service processing at the ingress or egress router or on the Ethernet
attachment circuit between the router and the Ethernet client,
it is RECOMMENDED that Ethernet OAM as defined in and/or be enabled for keyed IP tunnels.
As defined in those specifications, the following Connectivity Fault Management
(CFM) and/or Ethernet Continuity Check (ETH-CC) configurations are to be used in
conjunction with keyed IPv6 tunnels:Connectivity verification between the tunnel endpoints
across the tunnel:
Use an Up Maintenance End Point (MEP) located at the tunnel endpoint for
transmitting the CFM PDUs towards, and receiving them from, the
direction of the tunnel.Connectivity verification from the tunnel endpoint across
the local attachment circuit:
Use a Down MEP located at the tunnel
endpoint for transmitting the CFM PDUs towards, and receiving them
from, the direction of the local attachment circuit.Intermediate connectivity verification:
Use a Maintenance Intermediate Point (MIP) located at the
tunnel endpoint to relay CFM PDUs.In addition, Pseudowire VCCV
MAY be used. Furthermore, BFD MAY be enabled over the
VCCV channel .Note that since there is no control plane, it is RECOMMENDED that the management
plane take action when attachment circuit failure is detected, for example, by dropping
the remote attachment circuit.This document does not require any IANA actions.Packet spoofing for any type of Virtual Private Network (VPN)
tunneling protocol is of particular concern as insertion of carefully
constructed rogue packets into the VPN transit network could result in a
violation of VPN traffic separation, leaking data into a customer VPN.
This is complicated by the fact that it may be particularly difficult
for the operator of the VPN to even be aware that it has become a point
of transit into or between customer VPNs.Keyed IPv6 encapsulation provides traffic separation for its VPNs via the
use of separate 128-bit IPv6 addresses to identify the endpoints. The
mandatory use of the 64-bit L2TPv3 cookie provides an additional
check to ensure that an arriving packet is intended for the identified
tunnel.In the presence of a blind packet-spoofing attack, the 64-bit L2TPv3
cookie provides security against inadvertent leaking of frames into a
customer VPN, as documented in Section 8.2 of .For protection against brute-force blind insertion attacks, the 64-
bit cookie MUST be used with all tunnels.Note that the cookie provides no protection against a sophisticated
man-in-the-middle attacker who can sniff and correlate captured data
between nodes for use in a coordinated attack.The L2TPv3 64-bit cookie must not be regarded as a substitute for
security such as that provided by IPsec when operating over an open or
untrusted network where packets may be sniffed, decoded, and correlated
for use in a coordinated attack.IEEE Standard for Local and
metropolitan area networks - Bridges and Bridged NetworksIEEEIEEE Standard for Local and
Metropolitan Area Networks - Virtual Bridged Local Area
Networks, Amendment 4: Provider BridgesIEEEIEEE Standard for Local and metropolitan area networks -
Virtual Bridged Local Area Networks, Amendment 5: Connectivity Fault
ManagementIEEEOperation, administration and maintenance (OAM) functions and
mechanisms for Ethernet-based networksITU-TThe authors would like to thank Carlos Pignataro, Stewart
Bryant, Karsten Thomann, Qi Sun, and Ian Farrer for their insightful
suggestions and review.Peter Weinberger
Cisco Systems
Email: peweinbe@cisco.comMichael Lipman
Cisco Systems
Email: mlipman@cisco.comMark Townsley
Cisco Systems
Email: townsley@cisco.com