On Wed, 2017-11-22 at 12:12 +0800, Ed Greshko wrote:
On 11/15/17 20:20, Patrick O'Callaghan wrote:
> On Tue, 2017-11-14 at 14:50 -0800, Rick Stevens wrote:
> > On 11/14/2017 02:31 PM, Patrick O'Callaghan wrote:
> > > On trying to fire up a VM using virt-manager, I get "unable to map
> > > backing store for guest RAM: Permission denied".
> > >
> > > I use hugepages to lock down memory for the VM, which may be relevant
> > > for the error. This worked correctly on F26 just before the upgrade to
> > > F27. I have changed nothing in my VM configuration, nor in the config
> > > file for QEMU. However, temporarily turning off SElinux allows the
> > > startup to proceed, after which I can re-enable SElinux with no ill
> > > effects, i.e. the VM runs correctly.
> > >
> > > A complete relabel of my system (touch /.autolabel and reboot) has made
> > > no difference.
> > Have you checked the AVC reports from selinux? It smells like a bad
> > policy re: hugepages for KVMs. It's difficult to say without the actual
> > AVC denials.
> Not seeing anything. The only AVC reports in journalctl are about some
> dbus stuff unrelated to libvirtd. Interestingly the libvirtd startup
> lines in journalctl show:
> ... error : virAuditOpen:62 : Unable to initialise audit layer: Protocol not
> but it's not clear whether that has any relevance. As I say, toggling
> setenforce lets libvirtd start and from then on everything works.
I added a comment to your BZ but thought it would be worth mentioning here as well.
If you find your problem going away by disabling selinux but you're not getting an
AVC then it may be due to "donotaudit" in the policy. So, you should have
enabled and run "semodule -D" to turn off donotaudit and then run your test and
if you get an AVC.
I think you mean "semodule -DB" as that's what the manpage gives as an
example (BTW the page is phenomenally hard to understand, not least
because at one point it uses MODE, at another it says "module" and at a
third it says KIND, and it isn't clear if these are supposed to be the
Anyway, it had no effect. The error persists of course, but there is
still no AVC report.
Thanks all the same.