Re: [Xen-devel] [PATCH] Prevent xl save from segfaulting when control/shutdown key is removed

[Paul Durrant <Paul.Durrant@xxxxxxxxxx>]

True. I was conflating the key rm issue with the fact that the XenServer 
Windows PV guest agent gets unhappy because it can't advertise its features 
(which I hadn't mentioned before).

I suppose that libxl will just have to continue to speculatively write 
control/shutdown and timeout if it's not NUL-ed or rm-ed as it does now.

  Paul

> -----Original Message-----
> From: Ian Campbell
> Sent: 23 November 2011 13:05
> To: Paul Durrant
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-devel] [PATCH] Prevent xl save from segfaulting
> when control/shutdown key is removed
> 
> On Wed, 2011-11-23 at 12:59 +0000, Paul Durrant wrote:
> > > -----Original Message-----
> > > From: Ian Campbell
> > > Sent: 23 November 2011 11:24
> > > To: Paul Durrant
> > > Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
> > > Subject: RE: [Xen-devel] [PATCH] Prevent xl save from
> segfaulting
> > > when control/shutdown key is removed
> > >
> > > On Wed, 2011-11-23 at 11:19 +0000, Paul Durrant wrote:
> > > > what is the reason for creating control ro to the guest?
> > >
> > > In general libxl prefers to whitelist paths which the guest can
> > > write too, just to prevent a complete free for all, keep things
> > > somewhat under control and to help avoid situations where tools
> > > might inadvertently rely on a guest-writeable key in an unsafe
> way..
> > >
> > > >  In XenServer we allow the guest to write the control key to
> > > advertise
> > > > feature-shutdown, feature-suspend etc. so that the tools know
> what
> > > > values of control/shutdown the guest will respond to.
> > >
> > > The libxl way would be to create these at build time (perhaps
> empty)
> > > with the appropriate permissions.
> > >
> > > It's not clear how that functionality can be added in a way
> which is
> > > compatible with existing guests though, e.g. no Linux guest
> writes
> > > those but many can be suspended etc.
> > >
> >
> > So the simple solution, for compatibility's sake, is to make
> control rw isn't it?
> 
> The problem I'm thinking of exists even if control is rw.
> 
> Given an empty control directory how do you know if a guest supports
> suspend or not, given that most existing guests which do support
> suspend do not write any key there?
> 
> Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel