[libvirt] [PATCH 2/2] rpc: Fix segmentation fault when no worker pool is available

[Daniel P. Berrangé]

On Wed, Jun 20, 2018 at 11:14:41AM +0200, Marc Hartmayer wrote:
> On Tue, Jun 19, 2018 at 05:03 PM +0200, "Daniel P. Berrangé" <berrange at redhat.com> wrote:
> > On Tue, Jun 19, 2018 at 03:55:09PM +0100, Daniel P. Berrangé wrote:
> >> On Tue, Jun 19, 2018 at 04:51:13PM +0200, Erik Skultety wrote:
> >> > On Mon, Jun 18, 2018 at 12:38:06PM +0200, Marc Hartmayer wrote:
> >> > > On Mon, Jun 18, 2018 at 09:47 AM +0200, Erik Skultety <eskultet at redhat.com> wrote:
> >> > > > On Fri, Jun 15, 2018 at 03:31:34PM +0200, Marc Hartmayer wrote:
> >> > > >> On Fri, Jun 15, 2018 at 01:39 PM +0200, Marc Hartmayer <mhartmay at linux.ibm.com> wrote:
> >> > > >> > If srv->workers is a NULL pointer, as it is the case if there are no
> >> > > >> > workers, then don't try to dereference it.
> >> > > >> >
> >> > > >> > Signed-off-by: Marc Hartmayer <mhartmay at linux.ibm.com>
> >> > > >> > Reviewed-by: Boris Fiuczynski <fiuczy at linux.ibm.com>
> >> > > >> > ---
> >> > > >> >  src/rpc/virnetserver.c | 22 +++++++++++++++-------
> >> > > >> >  1 file changed, 15 insertions(+), 7 deletions(-)
> >> > > >> >
> >> > > >> > diff --git a/src/rpc/virnetserver.c b/src/rpc/virnetserver.c
> >> > > >> > index 5ae809e372..be6f610880 100644
> >> > > >> > --- a/src/rpc/virnetserver.c
> >> > > >> > +++ b/src/rpc/virnetserver.c
> >> > > >> > @@ -933,13 +933,21 @@ virNetServerGetThreadPoolParameters(virNetServerPtr srv,
> >> > > >> >                                      size_t *jobQueueDepth)
> >> > > >> >  {
> >> > > >> >      virObjectLock(srv);
> >> > > >> > -
> >> > > >> > -    *minWorkers = virThreadPoolGetMinWorkers(srv->workers);
> >> > > >> > -    *maxWorkers = virThreadPoolGetMaxWorkers(srv->workers);
> >> > > >> > -    *freeWorkers = virThreadPoolGetFreeWorkers(srv->workers);
> >> > > >> > -    *nWorkers = virThreadPoolGetCurrentWorkers(srv->workers);
> >> > > >> > -    *nPrioWorkers = virThreadPoolGetPriorityWorkers(srv->workers);
> >> > > >> > -    *jobQueueDepth = virThreadPoolGetJobQueueDepth(srv->workers);
> >> > > >> > +    if (srv->workers) {
> >> > > >> > +        *minWorkers = virThreadPoolGetMinWorkers(srv->workers);
> >> > > >> > +        *maxWorkers = virThreadPoolGetMaxWorkers(srv->workers);
> >> > > >> > +        *freeWorkers = virThreadPoolGetFreeWorkers(srv->workers);
> >> > > >> > +        *nWorkers = virThreadPoolGetCurrentWorkers(srv->workers);
> >> > > >> > +        *nPrioWorkers = virThreadPoolGetPriorityWorkers(srv->workers);
> >> > > >> > +        *jobQueueDepth = virThreadPoolGetJobQueueDepth(srv->workers);
> >> > > >> > +    } else {
> >> > > >> > +        *minWorkers = 0;
> >> > > >> > +        *maxWorkers = 0;
> >> > > >> > +        *freeWorkers = 0;
> >> > > >> > +        *nWorkers = 0;
> >> > > >> > +        *nPrioWorkers = 0;
> >> > > >> > +        *jobQueueDepth = 0;
> >> > > >> > +    }
> >> > > >> >
> >> > > >> >      virObjectUnlock(srv);
> >> > > >> >      return 0;
> >> > > >> > --
> >> > > >> > 2.13.6
> >> > > >>
> >> > > >> After thinking again it probably makes more sense (and the code more
> >> > > >> beautiful) to initialize the worker pool even for maxworker=0 (within
> >> > > >
> >> > > > I don't understand why should we do that.
> >> > >
> >> > > Because right now there are several functionalities broken. Segmentation
> >> > > faults in virNetServerGet/SetThreadPoolParameters, it’s not possible to
> >> > > start with maxworkers=0 and then change it at runtime via
> >> >
> >> > Naturally, since no workers means noone to process the request, that is IMHO
> >> > the expected behaviour.
> >>
> >> Yes, a daemon should either run with no workers, or should run with
> >> 1 or more workers. It is not value to change between these two modes.
> >>
> >> So if there's a codepath that lets you change from 0 -> 1 workers,
> >> or the reverse, we should make sure that reports an error.
> >>
> >> Essentially workers=0 is only intended for things like virtlockd
> >> or virlogd which don't need to be multithreaded, or indeed must
> >> *never* be multithreaded to avoid tickling kernel bugs like
> >> virtlockd did in the past.
> >
> > Also note that workers=0 will cause libvirtd to deadlock, because
> > the QEMU driver (and others too) assume that they run in a seperate
> > thread from the main event loop.
> 
> Shall we then disallow this value for the option "max_workers" in
> libvirtd.conf?

Yes, we should, as it can't work correctly.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|