On Fri, Feb 05, 2016 at 03:53:08AM -0500, Aaron Conole wrote: > IMHO, this is all just guessing games - testing this way is non-atomic. There > must be a way to have DPDK (or even if there is some non-dpdk vhost-user > library, really) know that this socket is orphaned. I guess, I'm saying > it's ****** that OVS is left holding the bag of deciding that it should > clean up.
It's inherent in the way Unix domain sockets work. Every program that creates a Unix domain socket, in the file system namespace, has the same issue. The only way I can think to sidestep this issue is to make OVS not create the socket at all. Instead, some other helper program (presumably one with fewer privileges) would unlink and bind it and then pass it to OVS with fd-passing over the OVS ctl socket. That's not a terrible idea. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
