Re: [PATCH 2/3] kill MMF_DUMPABLE and MMF_DUMP_SECURELY

[Oleg Nesterov <oleg@redhat.com>]

On 11/18, Kees Cook wrote:
>
> On Sat, Nov 16, 2013 at 11:01 AM, Oleg Nesterov <oleg@redhat.com> wrote:
> > @@ -1629,24 +1628,13 @@ void set_dumpable(struct mm_struct *mm, int value)
> >
> >         do {
> >                 old = ACCESS_ONCE(mm->flags);
> > -               new = old & ~MMF_DUMPABLE_MASK;
> > -
> > -               switch (value) {
> > -               case SUID_DUMP_ROOT:
> > -                       new |= (1 << MMF_DUMP_SECURELY);
> > -               case SUID_DUMP_USER:
> > -                       new |= (1<< MMF_DUMPABLE);
> > -               }
> > -
> > +               new = (old & ~MMF_DUMPABLE_MASK) | value;
>
> Just to make this safe against insane callers, perhaps mask the value as well?

Well yes, before this patch set_dumpable() silently ignored the wrong
value, perhaps you are right but see below.

>     new = (old & ~MMF_DUMPABLE_MASK) | (value & MMF_DUMPABLE_MASK);
                                          ^^^^^^^^^^^^^^^^^^^^^^^^^^

this doesn't really help, with this patch "mm->flags & MMF_DUMPABLE_MASK"
has a room for yet another SUID_DUMP == 4 we do not have yet.

And I don't really like the "silently ignore" logic, so perhaps

		if (WARN_ON(value > SUID_DUMP_ROOT))
			return;

at the start makes more sense?

Or perhaps we do not really need the additional check? suid_dumpable
is always sane, other callers can't use the wrong value.

But I am fine either way, please tell me what do you prefer.

Oleg.