Stream Control Transmission Protocol over User Datagram Protocol (SCTP over UDP, also known as UDP encapsulation of SCTP) is a feature defined in RFC6951 and implemented in the Linux kernel space since 5.11.0. It is planned to be supported by Red Hat Enterprise Linux (RHEL) 8.5.0 and 9.0.
This article is a quick introduction to SCTP over UDP in the Linux kernel.
Why we need SCTP over UDP
As described in the Internet Engineering Task Force Request for Comments (RFC), there are two main reasons that we need SCTP over UDP:
To allow SCTP traffic to pass through legacy NATs, which do not
provide native SCTP support as specified in [BEHAVE] and
To allow SCTP to be implemented on hosts that do not provide
direct access to the IP layer. In particular, applications can
use their own SCTP implementation if the operating system does not
The first reason will solve the middlebox issues that have brought many troubles to users and prevented SCTP’s wide use. The second reason is to allow user space applications to develop their own SCTP implementation based on the UDP protocol.
How SCTP over UDP works
With this feature enabled, all SCTP packets are encapsulated into UDP packets. SCTP over UDP is implemented with kernel UDP tunnel APIs that have previously been used by the VXLAN, GENEVE, and TIPC protocols.
For receiving encapsulated packets, the kernel listens on one specific UDP port on all local interfaces. The default port is 9899. This port also acts as the UDP packet
src port for this host as a sender. As you might anticipate, the
dest port should be the listening port of the peer, which also defaults to 9899. The
dest addresses bound to by SCTP would still be used in IP headers:
| IP(v6) Header (addresses bound by SCTP) |
| UDP Header (src: 9899, dest: 9899) |
| SCTP Common Header (SCTP ports) |
| SCTP Chunk #1 |
| ... |
| SCTP Chunk #n |
Note: SCTP also considers the UDP header while calculating the fragmentation point.
How to use SCTP over UDP
When programming, you don't need to do anything different: All the standard SCTP features still apply, and all the APIs are available to use as before. Old applications will work well without any changes or recompilation. The only adjustment is to set up a UDP port (a local listening port or
src port) and an encapsulation port (a remote listening or
dest port), which could be done globally for the network namespace by
# sysctl -w net.sctp.encap_port=9899
# sysctl -w net.sctp.udp_port=9899
Alternatively, you could set the encapsulation port per socket, association, or transport, using
On the server side, the
encapsulation port normally doesn’t need to be set explicitly, as detailed in the next section.
The UDP encapsulation port
The UDP encapsulation port allows for very flexible usage. On the sender side, the global encapsulation port only provides a default value:
- The per-socket encapsulation port can be used when another socket on one host connects to a different host on which a different UDP port is used.
- The per-association encapsulation port can be used when the same socket connects to a different host on which a different UDP port is used.
- The per-transport encapsulation port can be used when the same association wants to send UDP-encapsulated SCTP packets on one transport.
On the receiver side, the encapsulation port normally doesn’t need to be set:
- The encapsulation port of one association would be learned from the first
INITs with different UDP
srcports would then be discarded.
- The encapsulation port of each transport would be learned from the incoming packets on the corresponding path, and can be updated anytime.
- Plain SCTP packets can still be processed even if the encapsulation ports of the association and its transports are set.
If you’re using SCTP and enjoying its features, like multi-homing, multi-streaming, and partial-reliability, but having issues with middleboxes, the Linux kernel now provides an easier way to get around them.Last updated: August 24, 2022