Internet Draft T. Herbert
<draft-herbert-gue-00.txt> Google
Category: Experimental
Expires June 2014 December 20, 2013
Generic UDP Encapsulation
<draft-herbert-gue-00.txt>
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
This Internet-Draft will expire on June 24, 2014.
Copyright Notice
Copyright (c) 2013 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document.
Herbert Expires June 2014 [Page 1]
Internet Draft herbert-gue.txt December 20, 2013
Abstract
This specification describes Generic UDP Encapsulation (GUE), which
is a scheme for using UDP to encapsulate packets of arbitrary IP
protocols for transport across layer 3 networks. By encapsulating
packets in UDP, specialized capabilities in networking hardware for
efficient handling of UDP packets can be leveraged. GUE specifies
basic encapsulation methods upon which higher level constructs, such
tunnels and overlay networks, can be constructed.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Packet formats . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1. GUE header preamble . . . . . . . . . . . . . . . . . . . . 3
2.2. GUE encapsulation header . . . . . . . . . . . . . . . . . 4
3. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1. Encapsulator operation . . . . . . . . . . . . . . . . . . 6
3.2. Decapsulator operation . . . . . . . . . . . . . . . . . . 7
3.3. Router and switch operation . . . . . . . . . . . . . . . . 7
3.4. Middlebox and NAT interactions . . . . . . . . . . . . . . 7
3.5. UDP checksum . . . . . . . . . . . . . . . . . . . . . . . 8
3.6. MTU and fragmentation issues . . . . . . . . . . . . . . . 8
4. Inner flow identifier properties . . . . . . . . . . . . . . . 8
4.1. Flow classification . . . . . . . . . . . . . . . . . . . . 8
4.2. Inner flow identifier properties . . . . . . . . . . . . . 9
5. Motivation for GUE . . . . . . . . . . . . . . . . . . . . . . 10
6. Security Considerations . . . . . . . . . . . . . . . . . . . . 11
7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 11
8. References . . . . . . . . . . . . . . . . . . . . . . . . . . 12
8.1. Normative References . . . . . . . . . . . . . . . . . . . 12
8.2. Informative References . . . . . . . . . . . . . . . . . . 12
Appendix A: NIC processing for GUE . . . . . . . . . . . . . . . . 13
A.1. Receive multi-queue . . . . . . . . . . . . . . . . . . . . 13
A.2. Checksum offload . . . . . . . . . . . . . . . . . . . . . 14
A.2.1. Transmit checksum offload . . . . . . . . . . . . . . . 14
A.2.2. Receive checksum offload . . . . . . . . . . . . . . . 14
A.3. Transmit Segmentation Offload . . . . . . . . . . . . . . . 14
A.4. Large Receive Offload . . . . . . . . . . . . . . . . . . . 15
Appendix B: Privileged ports . . . . . . . . . . . . . . . . . . . 16
Appendix C: Inner flow identifier as a route selector . . . . . . 16
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 16
1. Introduction
This specification describes a general method for encapsulating
packets of arbitrary IP protocols within User Datagram Protocol (UDP)
[RFC0768] packets. Encapsulating packets in UDP facilitates efficient
Herbert Expires June 2014 [Page 2]
Internet Draft herbert-gue.txt December 20, 2013
transport across networks. Networking devices widely provide protocol
specific processing and optimizations for UDP (as well as TCP)
packets. Packets for atypical IP protocols (those not usually parsed
by networking hardware) can be encapsulated in UDP packets to
maximize deliverability and to leverage flow specific mechanisms for
routing and packet steering.
Hardware devices commonly perform hash computations on packet headers
to classify packets into flows or flow buckets. Flow classification
is done to support load balancing (statistical multiplexing) of flows
across a set of networking resources. Examples of such load balancing
techniques are Equal Cost Multipath routing (ECMP), port selection in
Link Aggregation, and NIC device Receive Side Scaling (RSS). Hashes
are usually either a three-tuple hash of IP protocol, source address,
and destination address; or a five-tuple hash consisting of IP
protocol, source address, destination address, source port, and
destination port. Typically, networking hardware will compute five-
tuple hashes for TCP and UDP, but only three-tuple hashes for other
IP protocols. Since the five-tuple hash provides more granularity,
load balancing can be finer grained with better distribution. When a
packet is encapsulated with GUE, the source port in the outer UDP
packet is set to reflect the flow of the inner packet. When a device
computes a five-tuple hash on the outer UDP/IP header of a GUE
packet, the resultant value classifies the packet per its inner flow.
2. Packet formats
The payload of a UDP packet destined to a GUE port starts with a GUE
header. If a packet is being encapsulated it immediately follows the
GUE header.
2.1. GUE header preamble
The first byte of the GUE packet header contains a packet type and
header length.
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
| Type | Hlen |
+-+-+-+-+-+-+-+-+
Contents are:
o Type: type of header. The rest of the fields in the header are
defined based the type.
o Hlen: Length in 32-bit words of the GUE header, including
Herbert Expires June 2014 [Page 3]
Internet Draft herbert-gue.txt December 20, 2013
optional fields but not the first four bytes of the header.
Computed as (header_len - 4) / 4. All GUE headers are a multiple
of four bytes in length.
2.2. GUE encapsulation header
The GUE encapsulation header is used to encapsulate packets for
various IP protocols. Encapsulation with a GUE header has the general
format:
+-------------------------------+
| |
| UDP/IP header |
| |
|-------------------------------|
| |
| GUE Header |
| |
|-------------------------------|
| |
| Encapsulated packet |
| |
+-------------------------------+
The GUE encapsulation header is variable length as determined by the
presence of optional fields.
The UDP and GUE encapsulation header format is:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Source port | Destination port |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Length | Checksum |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| 0x0 | Hlen | Protocol |V|R|R|R|R|R|R|R|R|R|R|R|R|R|P|P|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Virtual network ID (optional) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
~ Private fields (optional) ~
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The contents of the UDP header are:
Herbert Expires June 2014 [Page 4]
Internet Draft herbert-gue.txt December 20, 2013
o Source port (inner flow identifier): This should be set to a
value that represents the encapsulated flow. The properties of
the inner flow identifier are described below.
o Destination port: The GUE assigned port number, XXXX.
o Length: Canonical length of the UDP packet (payload length).
o Checksum: Either the standard UDP checksum or zero indicating no
checksum calculated. Zero checksum is recommended.
The GUE header consists of:
o Type: Set to 0x0 to indicate GUE encapsulation header.
o Hlen: Length in 32-bit words of the GUE header, including
optional fields but not the first four bytes of the header.
Computed as (header_len - 4) / 4. The length of the encapsulated
packet is determined from the UDP length and the Hlen:
encapsulated_packet_length = UDP_Length - 8 - GUE_Hlen.
o Protocol: IP protocol number for the next header. The next
header begins at the offset provided by Hlen.
o 'R' Reserved flag. Must be set to zero for sending.
o 'V' Virtualization flag. Indicates presence of the Virtual
Network Identifier (VNID) field. The VNID is used to tunnel
layer 2 or layer 3 packets for network virtualization. Use and
semantics of this field should be defined in separate documents.
o 'P' Private flag. Indicates flags reserved for private use (as
per private use policy specified in [RFC2434]). These flags may
indicate the presence of private fields. These flags can only be
used between a sender and a receiver that have agreement as to
their meaning.
o Virtual network ID (4 octets): Used in network virtualization to
identify the virtual network that packet was sent on. Only
present if virtualization bit is set.
o Private fields: An implementation may define private fields that
are present when a corresponding private bit is set. A private
field must have a length which is a multiple of four bytes, and
must be correctly accounted for in the GUE header length.
3. Operation
Herbert Expires June 2014 [Page 5]
Internet Draft herbert-gue.txt December 20, 2013
The figure below illustrates the use of GUE encapsulation between two
servers. Sever 1 is sending packets to server 2. An encapsulator
performs encapsulation of packets from server 1. These encapsulated
packets traverse the network as UDP packets. At the decapsulator,
packets are decapsulated and sent on to server 2. Packet flow in the
reverse direction need not be symmetric; GUE encapsulation is not
required in the reverse path.
+---------------+ +---------------+
| | | |
| Server 1 | | Server 2 |
| | | |
+---------------+ +---------------+
| ^
V |
+---------------+ +---------------+ +---------------+
| | | | | |
| Encapsulator |-->| Layer 3 |-->| Decapsulator |
| | | Network | | |
+---------------+ +---------------+ +---------------+
The encapsulator and decapsulator may be co-resident with the
corresponding servers, or may be on separate nodes in the network.
Network tunneling can be achieved by encapsulating layer 2 or layer 3
packets. In this case the encapsulator and decapsulator nodes are the
tunnel endpoints. These could be routers that provide network tunnels
on behalf of communicating servers.
When encapsulating layer 4 packets, the encapsulator and decapsulator
should be co-resident with the servers. In this case, the
encapsulation headers are inserted between the IP header and the
transport packet. The addresses in the IP header refer to both the
endpoints of the encapsulation and the endpoints for terminating the
the transport protocol.
3.1. Encapsulator operation
Encapsulators create encapsulation headers, set the source port to
the inner flow identifier, set flags and optional fields in the GUE
header, and forward packets to a decapsulator.
An encapsulator may be an end host originating the packets of a flow,
or may be a network device performing encapsulation on behalf of
servers (routers implementing tunnels for instance). In either case,
the intended target (decapsulator) is indicated by the outer
destination IP address.
Herbert Expires June 2014 [Page 6]
Internet Draft herbert-gue.txt December 20, 2013
If an encapsulator is tunneling packets, that is encapsulating
packets of layer 2 or layer 3 protocols (e.g. EtherIP, IPIP, ESP
tunnel mode), it should follow standard conventions for tunneling of
one IP protocol over another. Diffserv interaction with tunnels is
described in [RFC2983], ECN propagation for tunnels is described in
[RFC6040].
3.2. Decapsulator operation
A decapsulator performs decapsulation of GUE packets. A decapsulator
is addressed by the outer destination IP address of a GUE packet.
The decapsulator validates packets, including fields of the GUE
header. If a packet is acceptable, the UDP and GUE headers are
removed and the packet is resubmitted for IP protocol processing.
If a decapsulator receives a GUE packet with an unknown flag, bad
header length (too small for included optional fields), or an
otherwise malformed header, it must drop the packet and may log the
event. No error message is returned back to the encapsulator.
3.3. Router and switch operation
Routers and switches should forward GUE packets as standard UDP/IP
packets. The outer five-tuple should contain sufficient information
to perform flow classification corresponding to the flow of the inner
packet. A switch should not need to parse a GUE header, and none of
the flags or optional fields in the GUE header should affect routing.
A router should not modify a GUE header when forwarding a packet. It
may encapsulate a GUE packet in another GUE packet, for instance to
implement a network tunnel. In this case the router takes the role of
an encapsulator, and the corresponding decapsulator is the logical
endpoint of the tunnel.
3.4. Middlebox and NAT interactions
A middle box may interpret some flags and optional fields of the GUE
header for classification purposes, but is not required to understand
all flags and fields in GUE packets. A middle box should not drop a
GUE packet because there are flags unknown to it. The header length
in the GUE header allows a middlebox to inspect the payload packet
without needing to parse the flags or optional fields.
In certain instances a middlebox may infer bidirectional connection
semantics to a UDP flow. For instance a stateful firewall may create
a five-tuple rule to match flows on egress, and a corresponding five-
tuple rule for matching ingress packets where the roles of source and
destination are reversed for the IP addresses and UDP port numbers.
Herbert Expires June 2014 [Page 7]
Internet Draft herbert-gue.txt December 20, 2013
NAT for UDP assumes bidirectional connection semantics.
GUE primarily assumes unidirectional flow properties, there is no
necessary correspondence between the UDP ports of GUE packet for
encapsulated flows in different directions. GUE could be extended to
provide bidirectional semantics, however that is outside the scope of
this document.
3.5. UDP checksum
GUE packets should be sent with a zero checksum if the encapsulated
packet contains its own checksum or can be checked with some
alternate means. Applicability Statement for the Use of IPv6 UDP
Datagrams with Zero Checksums [RFC6936] provides analysis and
motivation of sending zero checksums when using UDP as an
encapsulation protocol.
If a receiver receives a GUE packet with a non-zero checksum, it must
perform normal UDP checksum verification.
3.6. MTU and fragmentation issues
Standard conventions for handling of MTU (Maximum Transmission Unit)
and fragmentation in conjunction with networking tunnels
(encapsulation of layer 2 or layer 3 packets) should be followed.
Details are described in MTU and Fragmentation Issues with In-the-
Network Tunneling [RFC4459]
If a packet is fragmented before encapsulation in GUE, all the
related fragments must be encapsulated using the same source port
(inner flow identifier). An operator may set MTU to account for
encapsulation overhead and reduce the likelihood of fragmentation.
4. Inner flow identifier properties
4.1. Flow classification
A major objective of using GUE is that a network device can perform
flow classification corresponding to the flow of the inner
encapsulated packet based on the contents in the outer headers.
To support flow classification, the source port of the UDP header in
GUE is set to a value that maps to the inner flow. This is referred
to as the inner flow identifier. The inner flow identifier is set by
the encapsulator; it can be computed on the fly based on packet
contents or retrieved from a state maintained for the inner flow.
Examples of deriving an inner flow identifier are:
Herbert Expires June 2014 [Page 8]
Internet Draft herbert-gue.txt December 20, 2013
o If the encapsulated packet is a layer 4 packet, TCP/IPv4 for
instance, the inner flow identifier could be based on the
canonical five-tuple hash of the inner packet.
o If the encapsulated packet is an AH transport mode packet with
TCP as next header, the inner flow identifier could be a hash
over a three-tuple: TCP protocol and TCP ports of the
encapsulated packet.
o If a node is encrypting a packet using ESP tunnel mode and GUE
encapsulation, the inner flow identifier could be based on the
contents of clear-text packet. For instance, a canonical five-
tuple hash for a TCP/IP packet could be used.
The five-tuple hash commonly used to identify a flow in UDP will
cover the outer source address, destination address, source port
(inner flow identifier), and destination port. These values should be
mostly persistent for the lifetime of an encapsulated flow, only
changing infrequently (at most once every thirty seconds).
4.2. Inner flow identifier properties
The inner flow identifier is the value set in the UDP source port of
a GUE packet. The inner flow identifier should adhere to the
following properties:
o The value set in the source port should be within the ephemeral
port range. IANA suggest this range to be 49152 to 65535, where
the high order two bits of the port are set to one. This
provides fourteen bits for the inner flow identifier value.
o The inner flow identifier should have a uniform distribution
across encapsulated flows.
o An encapsulator may occasionally change the inner flow
identifier used for an inner flow per its discretion (for
security, route selection, etc). Changing the value should
happen no more than once every thirty seconds.
o Decapsulators, or any networking devices, should not attempt any
interpretation of the inner flow identifier, nor should they
attempt to reproduce any hash calculation. They may use the
value to match further receive packets for steering decisions,
but cannot assume that the hash uniquely or permanently
identifies a flow.
o Input to the inner flow identifier is not restricted to ports
and addresses; input could include flow label from an IPv6
Herbert Expires June 2014 [Page 9]
Internet Draft herbert-gue.txt December 20, 2013
packet, SPI from an ESP packet, or other flow related state in
the encapsulator that is not necessarily conveyed in the packet.
o The assignment function for inner flow identifiers should be
randomly seeded to mitigate denial of service attacks. The seed
may be changed periodically.
5. Motivation for GUE
This section presents the motivation for GUE with respect to other
encapsulation methods.
A number of different encapsulation techniques have been proposed for
the encapsulation of one protocol over another. EtherIP [RFC3378]
provides layer 2 tunneling of Ethernet frames over IP. GRE [RFC2784],
MPLS [RFC4023], and L2TP [RFC2661] provide methods for tunneling
layer 2 and layer 3 packets over IP. NVGRE [NVGRE] and VXLAN [VXLAN]
are proposals for encapsulation of layer 2 packets for network
virtualization. IPIP [RFC2003] and Generic packet tunneling in IPv6
[RFC2473] provide methods for tunneling IP packets over IP.
Several proposals exist for encapsulating packets over UDP including
ESP over UDP [RFC3948], TCP directly over UDP [TCPUDP], VXLAN, LISP
[RFC6830] which encapsulates layer 3 packets, and Generic UDP
Encapsulation for IP Tunneling (GRE over UDP)[GREUDP]. Generic UDP
tunneling [GUT] is a proposal similar to GUE in that it aims to
tunnel packets of IP protocols over UDP.
GUE has the following discriminating features:
o UDP encapsulation leverages specialized network device
processing for efficient transport. The semantics for using the
UDP source port as an identifier for an inner flow are defined.
o GUE permits encapsulation of arbitrary IP protocols, which
includes layer 2 3, and 4 protocols. This potentially allows
nearly all traffic within a data center to be normalized to be
either TCP or UDP on the wire.
o Multiple protocols can be multiplexed over a single UDP port
number. This is in contrast to techniques to encapsulate
specific protocols over UDP using a protocol specific port
number (such as ESP/UDP, GRE/UDP, SCTP/UDP). GUE provides a
uniform and extensible mechanism for encapsulating all IP
protocols in UDP with minimal overhead (four bytes of additional
header).
o GUE is extensible. New flags and fields can be defined.
Herbert Expires June 2014 [Page 10]
Internet Draft herbert-gue.txt December 20, 2013
o The GUE header includes a header length field. This allows a
network node to inspect an encapsulated packet without needing
to parse the full encapsulation header.
o Private flags and fields allow local customization and
experimentation while being compatible with processing in
network nodes (routers and middleboxes).
o GUE can provide encapsulation for a virtual network that
provides layer 3 connectivity. In contrast, VXLAN and NVGRE are
defined to only provide layer 2 services (encapsulation of
Ethernet).
o GUE defines a 32 bit virtual networking identifier (in contrast
to 24 bit values defined for VXLAN and NVGRE). This facilitates
hierarchical assignment, local flag definitions in the
identifier, and potentially obfuscation of the identifier on the
wire.
6. Security Considerations
Encapsulation of IP protocols within GUE should not increase
security risk, nor provide additional security in itself. As
suggested in section 3 the source port for of UDP packets in GUE
should be randomly seeded to mitigate some possible denial
service attacks.
7. IANA Considerations
A well known UDP port number assignment for GUE will be
requested.
Herbert Expires June 2014 [Page 11]
Internet Draft herbert-gue.txt December 20, 2013
8. References
8.1. Normative References
[RFC0768]Postel, J., "User Datagram Protocol", STD 6, RFC 768, August
1980.
[RFC2434]Narten, T. and H. Alvestrand, "Guidelines for Writing an
IANA Considerations Section in RFCs", RFC 2434, October 1998.
[RFC2983]Black, D., "Differentiated Services and Tunnels", RFC 2983,
October 2000.
[RFC6040]Briscoe, B., "Tunnelling of Explicit Congestion
Notification", RFC 6040, November 2010.
[RFC6936]Fairhurst, G. and M. Westerlund, "Applicability Statement
for the Use of IPv6 UDP Datagrams with Zero Checksums", RFC 6936,
April 2013.
[RFC4459]Savola, P., "MTU and Fragmentation Issues with In-the-
Network Tunneling", RFC 4459, April 2006.
8.2. Informative References
[RFC2003]Perkins, C., "IP Encapsulation within IP", RFC 2003, October
1996.
[RFC3948]Huttunen, A., Swander, B., Volpe, V., DiBurro, L., and M.
Stenberg, "UDP Encapsulation of IPsec ESP Packets", RFC 3948, January
2005.
[RFC6830]Farinacci, D., Fuller, V., Meyer, D., and D. Lewis, "The
Locator/ID Separation Protocol (LISP)", RFC 6830, January 2013.
[RFC3378]Housley, R. and S. Hollenbeck, "EtherIP: Tunneling Ethernet
Frames in IP Datagrams", RFC 3378, September 2002.
[RFC2784]Farinacci, D., Li, T., Hanks, S., Meyer, D., and P. Traina,
"Generic Routing Encapsulation (GRE)", RFC 2784, March 2000.
[RFC4023]Worster, T., Rekhter, Y., and E. Rosen, Ed., "Encapsulating
MPLS in IP or Generic Routing Encapsulation (GRE)", RFC 4023, March
2005.
[RFC2661]Townsley, W., Valencia, A., Rubens, A., Pall, G., Zorn, G.,
Herbert Expires June 2014 [Page 12]
Internet Draft herbert-gue.txt December 20, 2013
and B. Palter, "Layer Two Tunneling Protocol "L2TP"", RFC 2661,
August 1999.
[NVGRE] NVGRE: Network Virtualization using Generic Routing
Encapsulation draft-sridharan-virtualization-nvgre-03
[VXLAN] VXLAN: A Framework for Overlaying Virtualized Layer 2
Networks over Layer 3 Networks draft-mahalingam-dutt-dcops-vxlan-06
[TCPUDP] Encapsulation of TCP and other Transport Protocols over UDP
draft-cheshire-tcp-over-udp-00
[GREUDP] Generic UDP Encapsulation for IP Tunneling draft-yong-tsvwg-
gre-in-udp-encap-02
[GUT] Generic UDP Tunnelling (GUT) draft-manner-tsvwg-gut-02.txt
Appendix A: NIC processing for GUE
This appendix provides some guidelines for Network Interface Cards
(NICs) to implement common offloads and accelerations to support GUE.
Note that most of this discussion is generally applicable to other
methods of encapsulation.
A.1. Receive multi-queue
Contemporary NICs support multiple receive descriptor queues (multi-
queue). Multi-queue enables load balancing of network processing for
a NIC across multiple CPUs. On packet reception, a NIC must select
the appropriate queue for host processing. Receive Side Scaling is a
common method which uses the flow hash for a packet to index an
indirection table where each entry stores a queue number. Flow
Director and Accelerated Receive Flow Steering (aRFS) allow a host to
program the queue that is used for a given flow which is identified
either by an explicit five-tuple or by flow hash.
GUE encapsulation should be compatible with multi-queue NICs that
support five-tuple hash calculation for UDP/IP packets as input to
RSS. The inner flow identifier (source port) ensures classification
of the encapsulated flow even in the case that the outer source and
destination addresses are the same for all flows (e.g. all flows are
going over a single tunnel).
By default, UDP support may be disabled in NICs to avoid out of order
reception that can occur when UDP packets are fragmented. As
discussed above, fragmentation of GUE packets should be mitigated by
fragmenting packets before entering a tunnel, path MTU discovery in
higher layer protocols, or operator adjusting MTUs. Other UDP traffic
Herbert Expires June 2014 [Page 13]
Internet Draft herbert-gue.txt December 20, 2013
may not implement such procedures to avoid fragmentation, so enabling
UDP support in the NIC should be a considered tradeoff during
configuration.
A.2. Checksum offload
Many NICs provide capabilities to calculate standard ones complement
payload checksum for packets in transmit or receive. When using GUE
encapsulation there are two checksums that may be of interest, the
payload checksum of an encapsulated packet, and the UDP checksum of
in the outer header.
A.2.1. Transmit checksum offload
NICs may provide a protocol agnostic method to offload transmit
checksum that can be used with GUE. In this method the host provides
checksum related parameters in a transmit descriptor for a packet.
These parameters include the starting offset of data to checksum, the
length of data to checksum, and the offset in the packet where the
computed checksum is to be written. The host may seed the checksum
with for data not covered by the NIC computation (the checksum of the
pseudo header for instance).
In the case of GUE, the checksum for an encapsulated transport layer
packet, a TCP packet for instance, can be offloaded by setting the
appropriate checksum parameters.
NICs typically can offload only one transmit checksum per packet, so
simultaneously offloading both an inner transport packet's checksum
and the outer UDP checksum is likely not possible. In this case
setting UDP checksum to zero (per above discussion) and offloading
the inner transport packet checksum is desirable.
A.2.2. Receive checksum offload
GUE is compatible with NICs that perform a protocol agnostic receive
checksum. In this technique, a NIC computes a ones complement
checksum over all (or some predefined portion) of a packet. The
computed value is provided to the host stack in the packet's receive
descriptor. The host driver can use this checksum to "patch up" and
validate any inner packet transport checksum, as well as the outer
UDP checksum if it is non-zero.
A.3. Transmit Segmentation Offload
Transmit Segmentation Offload (TSO) is a NIC feature where a host
provides a large (>MTU size) TCP packet to the NIC, which in turn
splits the packets into separate segments and transmits each one.
Herbert Expires June 2014 [Page 14]
Internet Draft herbert-gue.txt December 20, 2013
This is useful to reduce CPU load on host.
The process of TSO could be generalized as:
1. Split the TCP payload into segments which will allow less than
MTU size packets.
2. For each segment, replicate the TCP header and all preceding
headers of the original packet.
3. For each protocol header set any payload length fields to
reflect the length for the segment.
4. Set TCP sequence number to correctly reflect the offset of the
TCP data in the stream.
5. Recompute and set any checksums that either cover the payload
of the packet or cover header which was changed by setting a
payload length.
Following this general process, TSO can be extended to support TCP
encapsulation in GUE. For each segment the Ethernet, outer IP, UDP
header, GUE header, inner IP header if tunneling, and TCP headers are
replicated. Any packet length header fields need to be set properly
(including the length in the outer UDP header), and checksums need to
be set correctly (including the outer UDP checksum if being used).
A.4. Large Receive Offload
Large Receive Offload (LRO) is a NIC feature where packets of a TCP
connection are reassembled, or coalesced, in the NIC and delivered to
the host as one large packet. This feature can reduce CPU utilization
in the host.
LRO requires significant protocol awareness to be implemented
correctly and is difficult to generalize. Packets in the same flow
need to be unambiguously identified. In the presence of tunnels or
network virtualization, this may require more than a five-tuple match
(for instance packets for flows in two different virtual networks may
have identical five-tuples). Additionally, a NIC needs to perform
validation over packets that are being coalesced, and needs to
fabricate a single meaningful header from all the coalesced packets.
The conservative approach to supporting LRO for GUE would be to
assign packets to the same flow only if they have the same five-tuple
and were encapsulated the same way. That is the outer IP addresses,
the outer UDP ports, GUE protocol, GUE flags and fields, and inner
five tuple are all identical.
Herbert Expires June 2014 [Page 15]
Internet Draft herbert-gue.txt December 20, 2013
Appendix B: Privileged ports
Using the source port to contain an inner flow identifier value
disallows the security method of a receiver enforcing that the source
port be a privileged port. Privileged ports are defined by some
operating systems to restrict source port binding. Unix, for
instance, considered port number less than 1024 to be privileged.
Enforcing that packets are sent from a privileged port is widely
considered an inadequate security mechanism and has been mostly
deprecated. To approximate this behavior, an implementation could
restrict a user from sending a packet destined to the GUE port
without proper credentials.
Appendix C: Inner flow identifier as a route selector
A encapsulator generating an inner flow identifier may modulate the
value to perform a type of multipath source routing. Assuming that
networking switches perform ECMP based on the flow hash, a sender can
affect this decision by altering the inner flow identifier. For
instance, a sender may store a flow hash in its PCB for an inner
flow, and may alter the value upon detecting that packets are
traversing a lossy path. Changing the inner flow identifier for a
flow should be subject to hysteresis (at most once every thirty
seconds) to limit the number of out of order packets delivered.
Authors' Addresses
Tom Herbert
Google
1600 Amphitheatre Parkway
Mountain View, CA
EMail: therbert@google.com
Herbert Expires June 2014 [Page 16]