Hello, I am in the process of refactoring/refreshing an internal system that heavily uses libbpf/libxdp - and taking the time to refresh API usage. Previously we've used the bpf_xdp_attach() API to attach an XDP program to a device, and in doing so we can specify XDP mode flags etc. I've noticed a general shift towards use of bpf_link infrastructure, and behold there is an API to link XDP programs to a NIC: bpf_program__attach_xdp(), which looks to just proxy to the internal bpf_program_attach_fd() routine. While the internal implementation looks to support passing of link options via flags, the higher level XDP API simply passes NULL into the opts parameter, so the link creation flags will default to zero. But then, when chasing through the syscall into the kernel, it does look like link mode flags are supported within the kernel and it would do the right thing. I'm wondering if the lack of flags/options in bpf_program__attach_xdp() API is just an oversight, or if I'm missing something - would someone be able to confirm my thoughts? If so I'll stick to bpf_xdp_attach() for now! Thanks Alasdair
|