Re: [PATCH v3 bpf-next 8/8] selftests/bpf: introduce XDP compliance test tool

["Toke Høiland-Jørgensen" <toke@redhat.com>]

Stanislav Fomichev <sdf@google.com> writes:

>> > > +
>> > > +   ctrl_sockfd = accept(sockfd, (struct sockaddr *)&ctrl_addr, &len);
>> > > +   if (ctrl_sockfd < 0) {
>> > > +           fprintf(stderr, "Failed to accept connection on DUT socket\n");
>> > > +           close(sockfd);
>> > > +           return -errno;
>> > > +   }
>> > > +
>>
>> [...]
>>
>> >
>> > There is also connect_to_fd, maybe we can use that? It should take
>> > care of the timeouts.. (requires plumbing server_fd, not sure whether
>> > it's a problem or not)
>>
>> please correct me if I am wrong, but in order to have server_fd it is mandatory
>> both tester and DUT are running on the same process, right? Here, I guess 99% of
>> the times DUT and tester will run on two separated devices. Agree?
>
> Yes, it's targeting more the case where you have a server fd and a
> bunch of clients in the same process. But I think it's still usable in
> your case, you're not using fork() anywhere afaict, so even if these
> are separate devices, connect_to_fd should still work. (unless I'm
> missing something, haven't looked too closely)

Just to add a bit of context here, "separate devices" can refer to the
hosts as well as the netdevs. I.e., it should also be possible to run
this in a mode where the client bit runs on a different physical machine
than the server bit (as it will not be feasible in any case to connect
things with loopback cables).

-Toke