On Tue, Aug 31, 2021 at 08:47:43PM +0300, Michael Tokarev wrote: > 31.08.2021 20:38, Daniel P. Berrangé wrote: > ... > > > > So it looks like libvirt provides an unnamed socket there, -- > > > > maybe from a socketpair(2)? > > > > > > > > > > Yes > > > > No, libvirt binds to a named socket path and passes in a pre-opened > > FD for the listener socket. There shouldn't be any socketpair involved. > > Here's some more info from the original bugreport: > > 31.08.2021 00:20, dann frazier wrote: > > Aha! It seems that the important difference is whether or not the > > virt-manager GUI window for the VM is active. If it is active, the VM > > crashes regardless of how it is started (virsh console start/clicking > > "play" button). If the GUI is not active, the VM always works. > > > > With this knowledge I am able to confidently say that reverting > > 4cfd970ec1 *does* reliably avoid the problem. > > We'll try to figure out the calltrace, where this socket is coming from..
Oh, it is probably from the graphical console connection to SPICE or VNC. For those virt-manager will pass in a socket creted with socketpair() via libvirt, in order to bypass the need for authentication when running locally. Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
