I filed a bug about this at: https://bugzilla.redhat.com/show_bug.cgi?id=1252780 Dropping back to libvirt-1.2.2 fixed it for me, but there was no answer to the bug report. Sent from my Sony Xperia™ smartphone ---- chenhongzhen(a)zhicloud.com wrote ----

On 08/24/2015 11:26 PM, Michal Privoznik wrote: Yep, I had come to the same conclusion earlier today, but was going to confirm it before replying and haven't yet had time (chasing other ugly regressions caused by modifying perfectly good working code - someday maybe I will learn!). The horrible part is I'm almost certain that I had either tested it on RHEL6, or was given verification by someone else in an apparent position of knowledge that it would work properly (but I see there's no indication of that in the bugzilla record or commit log, so either it happened in IRC or I'm just on crack). Definitely if it doesn't work, we need to work around it. Even though the official downstream version of libvirt on RHEL6/CentOS6 doesn't have that patch, we so far are supporting upstream builds on RHEL6/CentOS6. Of course if creating a bridge with netlink doesn't work, then deleting the bridge with netlink also likely won't work, and changing back to the old ioctl method will potentially reintroduce a bug that had been stamped out by the switch :-(