Re: [PATCH v12 8/8] x86: Disallow vsyscall emulation when CET is enabled

[Sean Christopherson]

On Mon, Sep 21, 2020 at 04:48:25PM -0700, Andy Lutomirski wrote:
> On Mon, Sep 21, 2020 at 3:37 PM Yu-cheng Yu <yu-cheng.yu@xxxxxxxxx> wrote:
> > diff --git a/arch/x86/entry/vsyscall/vsyscall_64.c
> > b/arch/x86/entry/vsyscall/vsyscall_64.c
> > index 44c33103a955..0131c9f7f9c5 100644
> > --- a/arch/x86/entry/vsyscall/vsyscall_64.c
> > +++ b/arch/x86/entry/vsyscall/vsyscall_64.c
> > @@ -38,6 +38,9 @@
> >  #include <asm/fixmap.h>
> >  #include <asm/traps.h>
> >  #include <asm/paravirt.h>
> > +#include <asm/fpu/xstate.h>
> > +#include <asm/fpu/types.h>
> > +#include <asm/fpu/internal.h>
> >
> >  #define CREATE_TRACE_POINTS
> >  #include "vsyscall_trace.h"
> > @@ -286,6 +289,32 @@ bool emulate_vsyscall(unsigned long error_code,
> >         /* Emulate a ret instruction. */
> >         regs->ip = caller;
> >         regs->sp += 8;
> > +
> > +       if (current->thread.cet.shstk_size ||
> > +           current->thread.cet.ibt_enabled) {
> > +               u64 r;
> > +
> > +               fpregs_lock();
> > +               if (test_thread_flag(TIF_NEED_FPU_LOAD))
> > +                       __fpregs_load_activate();
> 
> Wouldn't this be nicer if you operated on the memory image, not the registers?
> 
> > +
> > +#ifdef CONFIG_X86_INTEL_BRANCH_TRACKING_USER
> > +               /* Fixup branch tracking */
> > +               if (current->thread.cet.ibt_enabled) {
> > +                       rdmsrl(MSR_IA32_U_CET, r);
> > +                       wrmsrl(MSR_IA32_U_CET, r & ~CET_WAIT_ENDBR);
> > +               }
> > +#endif
> 
> Seems reasonable on first glance.
> 
> > +
> > +#ifdef CONFIG_X86_INTEL_SHADOW_STACK_USER
> > +               /* Unwind shadow stack. */
> > +               if (current->thread.cet.shstk_size) {
> > +                       rdmsrl(MSR_IA32_PL3_SSP, r);
> > +                       wrmsrl(MSR_IA32_PL3_SSP, r + 8);
> > +               }
> > +#endif
> 
> What happens if the result is noncanonical?  A quick skim of the SDM
> didn't find anything.  This latter issue goes away if you operate on
> the memory image, though -- writing a bogus value is just fine, since
> the FP restore will handle it.

#GP, the SSP MSRs do canonical checks.