> > We'd ignore those IP interfaces beyond the fourth one. Although I've > > written code to handle this before, I'm not convinced it matters, since > > it'd mean the addition of new IP interfaces was racing with our > > SIOCGLIFCONF and I don't think it'd be observable to know whether a given > > IP interface was created before or after the SIOCGLIFCONF returned. > > (In the old days when SIOCGLIFCONF returned EINVAL when the buffer was > > too small, it was necessary.) > > A bad outcome would have been to fail the entire operation and ignored > all interfaces (instead of just the new ones), but if that isn't the > case, then that's fine.
Looking at this more, I think there's a risk that will miss an arbitrary set of interfaces. That is, I don't think there's any API guarantee that the new IP interface will be the one that gets chopped off. So I guess it's still an issue, but I'm not sure it's worth spending more time here. -- meem
