Re: [PATCH] drivers:of:property.c: fix a potential double put (release) bug

[Saravana Kannan]

On Wed, Aug 18, 2021 at 6:11 PM Saravana Kannan <saravanak@xxxxxxxxxx> wrote:
>
> On Wed, Aug 18, 2021 at 7:07 AM Rob Herring <robh+dt@xxxxxxxxxx> wrote:
> >
> > +Saravana
> >
> > On Wed, Aug 18, 2021 at 8:26 AM Wentao_Liang <Wentao_Liang_g@xxxxxxx> wrote:
> > >
> > > In line 1423 (#1), of_link_to_phandle() is called. In the function
> > > (line 1140, #2), "of_node_put(sup_np);" drops the reference to phandle
> > > and may cause phandle to be released. However, after the function
> > > returns, the phandle is subsequently dropped again (line 1424, #3) by
> > > the same put function. Double putting the phandle can lead to an
> > > incorrect reference count.
> > >
> > > We believe that the first put of the phandle is unnecessary (#3). We
> > > can fix the above bug by removing the redundant "of_node_put()" in line
> > > 1423.
> > >
> > > 1401 static int of_link_property(struct device_node *con_np,
> > >                                 const char *prop_name)
> > > 1402 {
> > > ...
> > > 1409     while (!matched && s->parse_prop) {
> > > ...
> > > 1414
> > > 1415         while ((phandle = s->parse_prop(con_np, prop_name, i))) {
> > > ...
> > >                  //#1 phandle is dropped in this function
> > > 1423             of_link_to_phandle(con_dev_np, phandle);
> > >
> > > 1424             //#3 the second drop to phandle
> > >                  of_node_put(phandle);
> > >
> > > 1425             of_node_put(con_dev_np);
> > > 1426         }
> > > ...
> > > 1428     }
> > > 1429     return 0;
> > > 1430 }
> > >
> > > 1095 static int of_link_to_phandle(struct device_node *con_np,
> > > 1096                   struct device_node *sup_np)
> > > 1097 {
> > > 1098     struct device *sup_dev;
> > > 1099     struct device_node *tmp_np = sup_np;
> > > ...
> > > 1140     of_node_put(sup_np);  //#2 the first drop to phandle
> > >                                //   (unnecessary)
> > > 1141
> > > 1142     return 0;
> > > 1143 }
> > >
> > > Signed-off-by: Wentao_Liang <Wentao_Liang_g@xxxxxxx>
> > > ---
> > >  drivers/of/property.c | 1 -
> > >  1 file changed, 1 deletion(-)
> > >
> > > diff --git a/drivers/of/property.c b/drivers/of/property.c
> > > index 6c028632f425..408fdde1a20c 100644
> > > --- a/drivers/of/property.c
> > > +++ b/drivers/of/property.c
> > > @@ -1137,7 +1137,6 @@ static int of_link_to_phandle(struct device_node *con_np,
> > >         put_device(sup_dev);
> > >
> > >         fwnode_link_add(of_fwnode_handle(con_np), of_fwnode_handle(sup_np));
> > > -       of_node_put(sup_np);
>
> Hi Wentao,
>
> Thanks for noticing and reporting the bug! Your analysis is correct,
> but the fix is definitely wrong. For one, the reference to the node
> phandle is pointing to can be dropped in of_link_to_phandle() when it
> calls of_get_compat_node(). It could also be dropped in one of the
> error paths. So, now you'll be incorrectly dropping the reference for
> the wrong node. Let me send out a fix and  mention you as the
> reporter.
>

I spoke too soon. I think there is no refcount problem because
of_link_to_phandle() makes sure it doesn't change the ref count of any
of the DT nodes passed in as input. If you see of_get_compat_node(),
you'll notice that it does a of_node_get() first. So it returns a node
pointer (that could be the same as the input) and it makes sure it
increments that refcount for the node it's returning. And since we are
doing:

sup_np = of_get_compat_node(sup_np);

We are ensuring that by the time of_link_phandle() returns, we haven't
changed the refcount of any of the nodes.

So, I don't think there's any bug here.

-Saravana