Re: [PATCH bpf-next v2 3/6] net: remove duplicate reuseport_lookup functions

[Kuniyuki Iwashima]

From: Lorenz Bauer <lmb@xxxxxxxxxxxxx>
Date: Wed, 14 Jun 2023 16:25:05 +0100
> On Tue, Jun 13, 2023 at 7:57 PM Kuniyuki Iwashima <kuniyu@xxxxxxxxxx> wrote:
> >
> >                         else {
> >                                 result = sk;
> >                         }
> >
> > The assignment to result below is buggy.  Let's say SO_REUSEPROT group
> > have TCP_CLOSE and TCP_ESTABLISHED sockets.
> 
> I'm not very familiar with SO_REUSEPORT, I assumed (incorrectly
> probably) that such a group would only ever have TCP_CLOSE in UDP case
> and TCP_LISTENING in TCP case. Can you explain how I could end up in
> this situation?

When we call conenct() for UDP socket in SO_REUSEPORT group, the state
is changed from TCP_CLOSE to TCP_ESTABLISHED in __ip4_datagram_connect(),
and the socket remains in the group.

That's why we check TCP_ESTABLISHED in reuseport_select_sock_by_hash()
that is always false for TCP but true for UDP in the case above.