Re: [PATCH net-next] net: ethtool: Fix suspicious rcu_dereference usage

[Eric Dumazet]

On Tue, Dec 17, 2024 at 6:17 PM Kory Maincent <kory.maincent@xxxxxxxxxxx> wrote:
>
> On Tue, 17 Dec 2024 16:47:07 +0100
> Eric Dumazet <edumazet@xxxxxxxxxx> wrote:
>
> > On Tue, Dec 17, 2024 at 3:03 PM Kory Maincent <kory.maincent@xxxxxxxxxxx>
> > wrote:
> > >
> > > The __ethtool_get_ts_info function can be called with or without the
> > > rtnl lock held. When the rtnl lock is not held, using rtnl_dereference()
> > > triggers a warning due to improper lock context.
> > >
> > > Replace rtnl_dereference() with rcu_dereference_rtnl() to safely
> > > dereference the RCU pointer in both scenarios, ensuring proper handling
> > > regardless of the rtnl lock state.
> > >
> > > Reported-by: syzbot+a344326c05c98ba19682@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > Closes:
> > > https://lore.kernel.org/netdev/676147f8.050a0220.37aaf.0154.GAE@xxxxxxxxxx/
> > > Fixes: b9e3f7dc9ed9 ("net: ethtool: tsinfo: Enhance tsinfo to support
> > > several hwtstamp by net topology") Signed-off-by: Kory Maincent
> > > <kory.maincent@xxxxxxxxxxx> --- net/ethtool/common.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/net/ethtool/common.c b/net/ethtool/common.c
> > > index 02f941f667dd..ec6f2e2caaf9 100644
> > > --- a/net/ethtool/common.c
> > > +++ b/net/ethtool/common.c
> > > @@ -870,7 +870,7 @@ int __ethtool_get_ts_info(struct net_device *dev,
> > >  {
> > >         struct hwtstamp_provider *hwprov;
> > >
> > > -       hwprov = rtnl_dereference(dev->hwprov);
> > > +       hwprov = rcu_dereference_rtnl(dev->hwprov);
> > >         /* No provider specified, use default behavior */
> > >         if (!hwprov) {
> > >                 const struct ethtool_ops *ops = dev->ethtool_ops;
> >
> > I have to ask : Can you tell how this patch has been tested ?
>
> Oh, it was not at all sufficiently tested. Sorry!
> I thought I had spotted the issue but I hadn't.
>
> > rcu is not held according to syzbot report.
> >
> > If rtnl and rcu are not held, lockdep will still complain.
>
> You are totally right.
> I may be able to see it with the timestamping kselftest. I will try.

syzbot has a repro that you can compile and run.

Make sure to build and use a kernel with

CONFIG_PROVE_LOCKING=y