[PATCH 5/7] x86: Split syscall_trace_enter into two phases

[Andy Lutomirski]

On Wed, Jul 16, 2014 at 1:21 PM, Kees Cook <keescook at chromium.org> wrote:
> On Tue, Jul 15, 2014 at 12:32 PM, Andy Lutomirski <luto at amacapital.net> wrote:
>>
>> +
>> +               ret = seccomp_phase1(&sd);
>> +               if (ret == SECCOMP_PHASE1_SKIP) {
>> +                       regs->orig_ax = -ENOSYS;
>
> Before, seccomp didn't touch orig_ax on a skip. I don't see any
> problem with this, and it's probably more clear this way, but are you
> sure there aren't unexpected side-effects from this?

It's necessary to cause the syscall to be skipped -- see syscall_trace_enter.

That being said, setting it to -ENOSYS is nonsense and probably
confused you at least as much as it confused me.  It should be
regs->orig_ax = -1.

--Andy