Re: [PATCH v3 3/3] x86/asm: Pin sensitive CR0 bits

From: Kees Cook
Date: Tue Jun 18 2019 - 13:18:06 EST


On Tue, Jun 18, 2019 at 02:24:30PM +0200, Peter Zijlstra wrote:
> On Tue, Jun 18, 2019 at 11:38:02AM +0200, Jann Horn wrote:
> > On Tue, Jun 18, 2019 at 6:55 AM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> > > With sensitive CR4 bits pinned now, it's possible that the WP bit for
> > > CR0 might become a target as well. Following the same reasoning for
> > > the CR4 pinning, this pins CR0's WP bit (but this can be done with a
> > > static value).
> > >
> > > Suggested-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > > ---
> > > arch/x86/include/asm/special_insns.h | 15 ++++++++++++++-
> > > 1 file changed, 14 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/arch/x86/include/asm/special_insns.h b/arch/x86/include/asm/special_insns.h
> > > index c8c8143ab27b..b2e84d113f2a 100644
> > > --- a/arch/x86/include/asm/special_insns.h
> > > +++ b/arch/x86/include/asm/special_insns.h
> > > @@ -31,7 +31,20 @@ static inline unsigned long native_read_cr0(void)
> > >
> > > static inline void native_write_cr0(unsigned long val)
> > > {
> >
> > So, assuming a legitimate call to native_write_cr0(), we come in here...
> >
> > > - asm volatile("mov %0,%%cr0": : "r" (val), "m" (__force_order));
> > > + unsigned long bits_missing = 0;
>
> ^^^
>
> > > +
> > > +set_register:
> > > + asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
> >
> > ... here we've updated CR0...
> >
> > > + if (static_branch_likely(&cr_pinning)) {
> >
> > ... this branch is taken, since cr_pinning is set to true after boot...
> >
> > > + if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
> >
> > ... this branch isn't taken, because a legitimate update preserves the WP bit...
> >
> > > + bits_missing = X86_CR0_WP;
>
> ^^^
>
> > > + val |= bits_missing;
> > > + goto set_register;
> > > + }
> > > + /* Warn after we've set the missing bits. */
> > > + WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
> >
> > ... and we reach this WARN_ONCE()? Am I missing something, or does
> > every legitimate CR0 write after early boot now trigger a warning?
>
> bits_missing will be 0 and WARN will not be issued.
>
> > > + }
> > > }

Yup, as Peter points out, bits_missing is only non-zero when bits went
missing. The normal case will skip the WARN_ONCE() (which is also
internally wrapped in unlikely()). And I would have noticed the very
loud WARN at every boot if this wasn't true. ;)

--
Kees Cook