Re: [PATCH 10/11] capsicum: prctl(2) to force use of O_BENEATH

[David Drysdale]

On Fri, Jul 25, 2014 at 5:00 PM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>
> On Jul 25, 2014 7:02 AM, "Paolo Bonzini" <pbonzini@xxxxxxxxxx> wrote:
> >
> > Il 25/07/2014 15:47, David Drysdale ha scritto:
> > > @@ -1996,6 +2013,17 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> > >               if (arg2 || arg3 || arg4 || arg5)
> > >                       return -EINVAL;
> > >               return current->no_new_privs ? 1 : 0;
> > > +     case PR_SET_OPENAT_BENEATH:
> > > +             if (arg2 != 1 || arg4 || arg5)
> > > +                     return -EINVAL;
> > > +             if ((arg3 & ~(PR_SET_OPENAT_BENEATH_TSYNC)) != 0)
> > > +                     return -EINVAL;
> > > +             error = prctl_set_openat_beneath(me, arg3);
> > > +             break;
> > > +     case PR_GET_OPENAT_BENEATH:
> > > +             if (arg2 || arg3 || arg4 || arg5)
> > > +                     return -EINVAL;
> > > +             return me->openat_beneath;
> > >       case PR_GET_THP_DISABLE:
> > >               if (arg2 || arg3 || arg4 || arg5)
> > >                       return -EINVAL;
> > >
> >
> > Why are you always forbidding a change of prctl from 1 to 0?  It should
> > be safe if current->no_new_privs is clear.
>
> I don't immediately see why you're forbidding unsettling it at all.
> If you need it to be sticky, then use seccomp or Capsicum to make it
> sticky.

Good point, that would make the function more generic -- needing to
latch is specific to Capsicum's use of it.

>
> Also, the way implementation is dangerously racy -- if anyone pokes at
> adjacent bitfields without the lock, they can get corrupted.  Try
> basing on Kees' seccomp tree or security-next and using the new atomic
> flags field.

Ah yes, sorry -- I hadn't yet shifted the implementation to line up with
the work you and Kees have put into the seccomp stuff.

>
>
> --Andy
>
> >
> > Do new threads inherit from the parent?
> >
> > Also, I wonder if you need something like this check:
> >
> >         /*
> >          * Installing a seccomp filter requires that the task has
> >          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
> >          * This avoids scenarios where unprivileged tasks can affect the
> >          * behavior of privileged children.
> >          */
> >         if (!current->no_new_privs &&
> >             security_capable_noaudit(current_cred(), current_user_ns(),
> >                                      CAP_SYS_ADMIN) != 0)
> >                 return -EACCES;
> >
> > Paolo

Yes, new threads inherit the flag from the parent so the
NNP||CAP_SYS_ADMIN check is probably needed.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/