[Users] Can't start a VM - sanlock permission denied

[Mike Burns]

On Mon, 2012-10-15 at 05:55 -0400, Federico Simoncelli wrote:
> ----- Original Message -----
> > From: "Dan Kenigsberg" <danken at redhat.com>
> > To: "Mike Burns" <mburns at redhat.com>
> > Cc: "Federico Simoncelli" <fsimonce at redhat.com>, users at ovirt.org
> > Sent: Monday, October 15, 2012 11:02:45 AM
> > Subject: Re: [Users] Can't start a VM - sanlock permission denied
> > 
> > On Sun, Oct 14, 2012 at 09:53:51PM -0400, Mike Burns wrote:
> > > On Sun, 2012-10-14 at 19:11 -0400, Federico Simoncelli wrote:
> > > > ----- Original Message -----
> > > > > From: "Alexandre Santos" <santosam72 at gmail.com>
> > > > > To: "Dan Kenigsberg" <danken at redhat.com>
> > > > > Cc: "Haim Ateya" <hateya at redhat.com>, users at ovirt.org,
> > > > > "Federico Simoncelli" <fsimonce at redhat.com>
> > > > > Sent: Sunday, October 14, 2012 7:23:36 PM
> > > > > Subject: Re: [Users] Can't start a VM - sanlock permission
> > > > > denied
> > > > > 
> > > > > 2012/10/13 Dan Kenigsberg < danken at redhat.com >
> > > > > 
> > > > > On Sat, Oct 13, 2012 at 11:25:37AM +0100, Alexandre Santos
> > > > > wrote:
> > > > > > Hi,
> > > > > > after getting to the oVirt Node console (F2) I figured out
> > > > > > that
> > > > > > selinux
> > > > > > wasn't allowing the sanlock, so I entered the setsebool
> > > > > > virt_use_sanlock 1
> > > > > > and the problem is fixed.
> > > > > 
> > > > > Which version of vdsm is istalled on your node? and which
> > > > > selinux-policy? sanlock should work out-of-the-box.
> > > > > 
> > > > > 
> > > > > vdsm-4.10.0-10.fc17
> > > > > 
> > > > > on /etc/sysconfig/selinux
> > > > > SELINUX=enforcing
> > > > > SELINUXTYPE=targeted
> > > > 
> > > > As far as I understand the selinux policies for the ovirt-node
> > > > are set
> > > > by recipe/common-post.ks (in the ovirt-node repo):
> > > > 
> > > > semanage  boolean -m -S targeted -F /dev/stdin  << \EOF_semanage
> > > > allow_execstack=0
> > > > virt_use_nfs=1
> > > > EOF_semanage
> > > > 
> > > > We should update it with what vdsm is currently setting:
> > > > 
> > > > virt_use_sanlock=1
> > > > sanlock_use_nfs=1
> > > > 
> > > 
> > > Shouldn't vdsm be setting these if they're needed?
> > 
> > It should - I'd like to know which vdsm version was it, and why this
> > was skipped.
> 
> The version was 4.10.0-10.fc17 and what I thought (but I didn't test yesterday
> night) is that the ovirt-node was overriding what we were setting.
> Anyway this is not the case.
> 
> > > I can certainly set
> > > the values, but IMO, if vdsm needs it, vdsm should set it.
> > 
> > virt_use_nfs=1 made it into the node. Maybe there was a good reason
> > for it that applies to virt_use_sanlock as well. (I really hate to
> > persist the policy files, and dislike the idea of setting virt_use_sanlock
> > every time vdsmd starts - it's slooooow).

Agreed, we shouldn't do this every time vdsm starts.  I posted a patch
to pull these 2 booleans into ovirt-node config.  When we start the
process of making an ovirt/vdsm plugin rpm, it should move into the new
plugin rather than in base ovirt-node.

> 
> We set them when we install vdsm (not when the service starts) so they should
> be good to go in the iso. It might be a glitch during the vdsm package
> installation, it could be something like semanage taking the boolean from the
> host where the iso is built rather than the root where the package is installed.
> 
> Do we have the iso build logs?
> 

No, not at this point.  I'm still building ISOs and RPMs for ovirt-node
on my local machine and don't generally persist the build logs for long.
Eventually when we move to building only in Jenkins for official
releases, we'll have logs persisted long term.

Mike