Re: [PATCH net-next] scm: fix MSG_CTRUNC setting condition for SO_PASSSEC

[Aleksandr Mikhalitsyn]

On Mon, Feb 27, 2023 at 7:32 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
>
> On Mon, Feb 27, 2023 at 10:55:04AM +0100, Aleksandr Mikhalitsyn wrote:
> > On Mon, Feb 27, 2023 at 10:47 AM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
> > >
> > > On Sun, Feb 26, 2023 at 09:17:30PM +0100, Alexander Mikhalitsyn wrote:
> > > > Currently, we set MSG_CTRUNC flag is we have no
> > > > msg_control buffer provided and SO_PASSCRED is set
> > > > or if we have pending SCM_RIGHTS.
> > > >
> > > > For some reason we have no corresponding check for
> > > > SO_PASSSEC.
> > > >
> > > > Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>
> > > > Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> > > > Cc: Jakub Kicinski <kuba@xxxxxxxxxx>
> > > > Cc: Paolo Abeni <pabeni@xxxxxxxxxx>
> > > > Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@xxxxxxxxxxxxx>
> > > > ---
> > > >  include/net/scm.h | 13 ++++++++++++-
> > > >  1 file changed, 12 insertions(+), 1 deletion(-)
> > >
> > > Is it a bugfix? If yes, it needs Fixes line.
> >
> > It's from 1da177e4c3 ("Linux-2.6.12-rc2") times :)
> > I wasn't sure that it's correct to put the "Fixes" tag on such an old
> > and big commit. Will do. Thanks!
> >
> > >
> > > >
> > > > diff --git a/include/net/scm.h b/include/net/scm.h
> > > > index 1ce365f4c256..585adc1346bd 100644
> > > > --- a/include/net/scm.h
> > > > +++ b/include/net/scm.h
> > > > @@ -105,16 +105,27 @@ static inline void scm_passec(struct socket *sock, struct msghdr *msg, struct sc
> > > >               }
> > > >       }
> > > >  }
> > > > +
> > > > +static inline bool scm_has_secdata(struct socket *sock)
> > > > +{
> > > > +     return test_bit(SOCK_PASSSEC, &sock->flags);
> > > > +}
> > > >  #else
> > > >  static inline void scm_passec(struct socket *sock, struct msghdr *msg, struct scm_cookie *scm)
> > > >  { }
> > > > +
> > > > +static inline bool scm_has_secdata(struct socket *sock)
> > > > +{
> > > > +     return false;
> > > > +}
> > > >  #endif /* CONFIG_SECURITY_NETWORK */
> > >
> > > There is no need in this ifdef, just test bit directly.
> >
> > The problem is that even if the kernel is compiled without
> > CONFIG_SECURITY_NETWORK
> > userspace can still set the SO_PASSSEC option. IMHO it's better not to
> > set MSG_CTRUNC
> > if CONFIG_SECURITY_NETWORK is disabled, msg_control is not set but
> > SO_PASSSEC is enabled.
> > Because in this case SCM_SECURITY will never be sent. Please correct
> > me if I'm wrong.
>
> I don't know enough in this area to say if it is wrong or not.
> My remark was due to the situation where user sets some bit which is
> going to be ignored silently. It will be much cleaner do not set it
> if CONFIG_SECURITY_NETWORK is disabled instead of masking its usage.

Hi Leon,

I agree with you, but IMHO then it looks more correct to return -EOPNOTSUPP on
setsockopt(fd, SO_PASSSEC, ...) if CONFIG_SECURITY_NETWORK is disabled.
But such a change may break things.

Okay, anyway I'll wait until net-next will be opened and present a
patch with a more
detailed description and Fixes tag. Speaking about this problem with
CONFIG_SECURITY_NETWORK
if you insist that it will be more correct then I'm ready to fix it too.

Thanks,
Alex

>
> Thanks
>
> >
> > Kind regards,
> > Alex
> >
> > >
> > > >
> > > >  static __inline__ void scm_recv(struct socket *sock, struct msghdr *msg,
> > > >                               struct scm_cookie *scm, int flags)
> > > >  {
> > > >       if (!msg->msg_control) {
> > > > -             if (test_bit(SOCK_PASSCRED, &sock->flags) || scm->fp)
> > > > +             if (test_bit(SOCK_PASSCRED, &sock->flags) || scm->fp ||
> > > > +                 scm_has_secdata(sock))
> > > >                       msg->msg_flags |= MSG_CTRUNC;
> > > >               scm_destroy(scm);
> > > >               return;
> > > > --
> > > > 2.34.1
> > > >