Beyond socket options: Towards fully extensible Linux transport stacks

Abstract

The Transmission Control Protocol (TCP) is one of the most important protocols in today’s Internet. It was designed to be extensible for various use cases. A client can propose to use an extension over a given TCP connection by sending a TCP option that identifies this extension. In practice, deploying a TCP extension is difficult as the maintainers of client stacks often wait until servers implement a given extension and server maintainers look at clients in the same manner. It often takes several years if not a decade to actually deploy a TCP option widely. Our goal is to support experimenting and deploying new TCP options in a quick, simple, and efficient way. This includes inserting new TCP options at the sender side and parsing them at the receiver side. The implementation and the interface should be simple, generic, and introduce as few changes to the kernel code as possible. In this paper, we focus on the Linux TCP stack since it is one of the most widely used TCP stacks, given its utilization on many servers and Android devices. For this purpose, we leverage the extended Berkeley Packet Filter (eBPF), which is a recently developed in-kernel infrastructure to enable high performance and safe programmability to the Linux kernel space.Multipath TCP (MPTCP) is a major TCP extension that enables more capabilities and has richer semantics than regular TCP. We implemented a similar methodology in the Linux MPTCP stack to support new use-cases through custom MPTCP options. Moreover, an eBPF-based framework for user-defined path managers is also proposed, given that subflow management is an important task in Multipath TCP.