Re: [PATCHv3 bpf-next 3/5] bpf: Don't refcount LISTEN sockets in sk_assign()

On Fri, Mar 27, 2020 at 7:26 AM Jamal Hadi Salim [off-list ref] wrote:

On 2020-03-27 12:25 a.m., Joe Stringer wrote:

quoted

BPF_CALL_1(bpf_sk_release, struct sock *, sk)
  {
-     /* Only full sockets have sk->sk_flags. */
-     if (!sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE))
+     if (sk_is_refcounted(sk))
              sock_gen_put(sk);
      return 0;
  }

Would it make sense to have both the bpf_sk_release and bpf_sk_assign()
centralized so we dont replicate the functionality in tc? Reduces
maintainance overhead.

I think sock_pfree() steps in that direction, we would just need the
corresponding refactoring for sk_assign bits. Sounds like a good idea.

This shouldn't functionally affect this series, I'm happy to either
spin this into next revision of this series (if there's other
feedback), or send a followup refactor for this, or defer this to your
TC follow-up series that would consume the refactored functions.