BPF struct_ops - current status and the last developments
summary
BPF struct_ops lets user BPF programs implement kernel-visible method tables so subsystems can call user-defined functions; TCP congestion control is already an upstream consumer. Recent work creates a real BPF link for struct_ops maps so objects automatically unregister when a program dies and supports atomic link update to swap implementations without a registration gap. Ongoing patch adds runtime registration from kernel modules. Early adopters—fuse-bpf and an skex scheduler—report measurable performance gains, up to ~25% RPS and lower P99s.
AI-written summary · about