Re: [Libguestfs] [p2v PATCH 10/10] nbd.c: bind listening socket without AI_ADDRCONFIG

Now the question is if we are bothered by this "RST" in the conversion log, as it's expected.

I think wait_for_nbd_server_to_start() does add a little bit of safety. Namely, if the child goes away unexpectedly (because the execlp() fails, or the new image exits for whatever reason very soon), we find out about that (with a timeout in the worst case), and don't start up "ssh -R". With wait_for_nbd_server_to_start() removed, this "child failure" will only be visible by virt-v2v attempting to connect back via "ssh -R", and then (with ssh on the p2v box getting an RST), witnessing a similarly abrubpt EOF / RST from sshd on the conversion server. Ultimately virt-v2v will fail, and this will be visible on virt-p2v, but that's a detour.

I agree that functionally, wait_for_nbd_server_to_start() is not necessary, as long as the child does not vanish. With the child present, there is always at least one file descriptor pointing to the file description pointing to the listening socket, so the p2v side of "ssh -R" will always have a socket to connect to. I'll append a patch anyway to the series, just for discussion's sake. > Does it work OK if you simply remove wait_for_nbd_server_to_start and > the place where that function is called? Yes, the message disappears then.