Re: [Qemu-devel] [PATCH] ivshmem: fix memory backend leak

[Igor Mammedov]

On Thu, 1 Nov 2018 15:02:04 +0400
Marc-André Lureau <address@hidden> wrote:

> On Thu, Nov 1, 2018 at 2:53 PM Igor Mammedov <address@hidden> wrote:
> >
> > object_new() returns a new backend with refcount == 1 and
> > then later object_property_add_child() increases refcount to 2
> > So when ivshmem is desroyed, the backend it has created isn't
> > destroyed along with it as children cleanup will bring
> > backend's refcount only to 1, which leaks backend including
> > resources it is using.
> >
> > Drop the original reference from object_new() once backend
> > is attached to its parent.
> >
> > Signed-off-by: Igor Mammedov <address@hidden>  
> 
> I would rather have the unref in finalize, but that is ok too.
I followed the pattern we use else where, i.e. drop reference
as soon as we set the parent (virtio-rng/cpus) within the same
scope as object_new().

There is no point in keeping reference until finalize time since
backend is kept alive as child and is destroyed well after all
nonexistent ivshmem::unrealize/finilize() are finished when generic
Object is being destroyed.

>  Reviewed-by: Marc-André Lureau <address@hidden>
> 
> > ---
> >  hw/misc/ivshmem.c | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/hw/misc/ivshmem.c b/hw/misc/ivshmem.c
> > index f88910e..ecfd10a 100644
> > --- a/hw/misc/ivshmem.c
> > +++ b/hw/misc/ivshmem.c
> > @@ -1279,6 +1279,7 @@ static void desugar_shm(IVShmemState *s)
> >      object_property_set_bool(obj, true, "share", &error_abort);
> >      object_property_add_child(OBJECT(s), "internal-shm-backend", obj,
> >                                &error_abort);
> > +    object_unref(obj);
> >      user_creatable_complete(obj, &error_abort);
> >      s->hostmem = MEMORY_BACKEND(obj);
> >  }
> > --
> > 2.7.4
> >
> >  
> 
>