Extended Berkeley Packet Filter (eBPF) allows for high-performance introspection of the Linux kernel execution. eBPF is widely available (part of the mainline kernel and enabled by most distributions), flexible (any kernel code path can be probed) and safe (driven from userspace and statically verified). In this talk, I will introduce eBPF, explaining how it can be used to track TCP connections in real time. On the way I will demonstrate it is possible to access eBPF from languages other than C (Golang) and remove undesirable runtime dependencies (LLVM compiler and kernel-headers). At Weaveworks we are using eBPF for the connection-tracker of the Weave Scope visualization tool.
