Re: [PATCH v4 07/10] x86: narrow out of bounds syscalls to sys_read under speculation

[Andy Lutomirski]

On Tue, Feb 6, 2018 at 8:42 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Tue, Feb 6, 2018 at 12:37 PM, Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
>>
>> Are there any compilers that would miscompile:
>>
>>     mask = 0 - (index < size);
>>
>> That might be a way to improve the assembly.
>
> Sadly, that is *very* easy to miscompile. In fact, I'd be very
> surprised indeed if any compiler worth its name wouldn't combine the
> comparison with the conditional branch it accompanies, and just turn
> that into a constant. IOW, you'd get
>
>         mask = 0 - (index < size);
>         if (index <= size) {
>                  ... use mask ..
>
> and the compiler would just turn that into
>
>         if (index <= size) {
>                 mask = -1;
>
> and be done with it.
>
>                Linus

Can you use @cc to make an asm statement that outputs both the masked
array index and the "if" condition?  I can never remember the syntax,
but something like:

asm ("cmp %[limit], %[index]\n\tcmovae %[zero], %[index]" : [index]
"+" (index), "@ccb" (result));

Then you shove this into a statement expression macro so you can do:

if (index_mask_nospec(&nr, NR_syscalls)) {
  ... sys_call_table[nr] ..;
}

(Caveat emptor: I can also *ever* remember which way the $*!& AT&T
syntax cmp instruction goes.)

A down side is that nr actually ends up containing zero outside the
if.  *That* could be avoided with jump labels.

--Andy