[CRIU] [PATCH] cg: don't leak properties when cg dir exists

[Andrew Vagin]

On Thu, Oct 02, 2014 at 12:33:56PM -0500, Tycho Andersen wrote:
> On Thu, Oct 02, 2014 at 07:27:51PM +0400, Pavel Emelyanov wrote:
> > On 10/02/2014 07:24 PM, Tycho Andersen wrote:
> > > On Thu, Oct 02, 2014 at 07:21:55PM +0400, Pavel Emelyanov wrote:
> > >>
> > >>>> configure them and execute "criu restore".
> > >>>> And I doesn't expect that criu will change parameters for the initial
> > >>>> set of cgroup. Could you explain why it should?
> > >>>
> > >>> If you're using --cgroup-root, it should certainly not touch the
> > >>> original ones. However, if you're not, it has to re-create them
> > >>> because of the case I described above.
> > >>
> > >> Absolutely. Guys, do I get the issue right -- the tasks from test live
> > >> in root cpuset cgrpup, so on restore criu tries to set "special values"
> > >> to this cgroup too and fails with -EPERM?
> > > 
> > > Yes, although I think it is -EACCES.
> > 
> > Of course :) And why do we get one?
> 
> Actually reading the man page, it might be ENOSPC (Andrey can confirm
> by pasting the whole log), but I think we are probably writing to
> cpuset:/cpuset.mems when cpuset:/tasks has pids in it. In either case,
> though, we shouldn't be trying to write cpuset:/cpuset.mems at all.

In my case it's -EACCES.

> 
> Tycho
> 
> > > Definitely a bug with the patch I
> > > just sent, though. I will send another to fix it.
> > 
> > Thanks for looking into this.
> > 
> >