On Fri, Sep 18, 2020 at 02:10:42PM +0200, Nicolas Rybowski wrote: > + > +BPF_CALL_1(bpf_mptcp_sock, struct sock *, sk) > +{ > + if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && > sk_is_mptcp(sk)) { > + struct mptcp_subflow_context *mptcp_sfc = mptcp_subflow_ctx(sk);
Could you add !sk check here as well? See commit 8c33dadc3e0e ("bpf: Bpf_skc_to_* casting helpers require a NULL check on sk") It's not strictly necessary yet, but see below. Also this new helper is not exercised from C test. Only from asm. Could you update patch 4 with such additional logic? > + > + return (unsigned long)mptcp_sfc->conn; I think we shouldn't extend the verifier with PTR_TO_MPTCP_SOCK and similar concept anymore. This approach doesn't scale and we have better way to handle such field access with BTF. > + } > + return (unsigned long)NULL; > +} > + > +const struct bpf_func_proto bpf_mptcp_sock_proto = { > + .func = bpf_mptcp_sock, > + .gpl_only = false, > + .ret_type = RET_PTR_TO_MPTCP_SOCK_OR_NULL, In this particular case you can do: + .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, Then bpf_mptcp_sock_convert_ctx_access() will no longer be necessary and bpf prog will be able to access all mptcp_sock fields right away. Will that work for your use case?