On Tue, Nov 26, 2024 at 11:34:25PM -0800, John Fastabend wrote: > Zijian Zhang wrote: > > > > LGTM! > > > > I think it will be better if the test could also cover the case you > > indicated in the first patch, where skb_transport_offset is a negative > > value. > > > > Thanks, > > Zijian > > > > Hi Cong, > > I agree it would be great to see the skb_transport_offset is > negative pattern. Could we add it? Hmm? It is already negative for sockmap, as I already mentioned in patch 1/1: "skb_transport_offset() and skb_transport_offset() can be negative when they are called after we pull the transport header, for example, when we use eBPF sockmap (aka at the point of ->sk_data_ready())." My test case uses skb verdict, which is one of the sockmap hooks. Or I guess you mean positive? In that case, we would need hook different locations, like TC. I can certainly add it, but once again, it would make backporting this patchset even harder. Thanks!
