Re: Is xt_owner's owner_mt() racy with sock_orphan()? [worse with new TYPESAFE_BY_RCU file lifetime?]

[Christian Brauner <brauner@kernel.org>]

On Wed, Dec 06, 2023 at 03:38:50PM +0100, Jann Horn wrote:
> On Wed, Dec 6, 2023 at 2:58 PM Christian Brauner <brauner@kernel.org> wrote:
> >
> > On Tue, Dec 05, 2023 at 06:08:29PM +0100, Jann Horn wrote:
> > > On Tue, Dec 5, 2023 at 5:40 PM Jann Horn <jannh@google.com> wrote:
> > > >
> > > > Hi!
> > > >
> > > > I think this code is racy, but testing that seems like a pain...
> > > >
> > > > owner_mt() in xt_owner runs in context of a NF_INET_LOCAL_OUT or
> > > > NF_INET_POST_ROUTING hook. It first checks that sk->sk_socket is
> > > > non-NULL, then checks that sk->sk_socket->file is non-NULL, then
> > > > accesses the ->f_cred of that file.
> > > >
> > > > I don't see anything that protects this against a concurrent
> > > > sock_orphan(), which NULLs out the sk->sk_socket pointer, if we're in
> > >
> > > Ah, and all the other users of ->sk_socket in net/netfilter/ do it
> > > under the sk_callback_lock... so I guess the fix would be to add the
> > > same in owner_mt?
> >
> > In your other mail you wrote:
> >
> > > I also think we have no guarantee here that the socket's ->file won't
> > > go away due to a concurrent __sock_release(), which could cause us to
> > > continue reading file credentials out of a file whose refcount has
> > > already dropped to zero?
> >
> > Is this an independent worry or can the concurrent __sock_release()
> > issue only happen due to a sock_orphan() having happened first? I think
> > that it requires a sock_orphan() having happend, presumably because the
> > socket gets marked SOCK_DEAD and can thus be released via
> > __sock_release() asynchronously?
> >
> > If so then taking sk_callback_lock() in owner_mt() should fix this.
> > (Otherwise we might need an additional get_active_file() on
> > sk->sk_socker->file in owner_mt() in addition to the other fix.)
> 
> My understanding is that it could only happen due to a sock_orphan()
> having happened first, and so just sk_callback_lock() should probably
> be a sufficient fix. (I'm not an expert on net subsystem locking rules
> though.)

Ok, so as I suspected. That's good.