Re: [PATCH v4 0/6] Micro-optimize vclock_gettime

[Andrew Lutomirski <luto@mit.edu>]

On Tue, May 17, 2011 at 4:00 AM, Ingo Molnar <mingo@elte.hu> wrote:
>
> * Thomas Gleixner <tglx@linutronix.de> wrote:
>
>> > see if I can persuade Uli to take accept a glibc patch to stop calling it
>> > in future static glibc versions.
>>
>> How wide spread is this in reality on 64bit systems ?
>>
>> IOW, what's the damage if we take a trap and emulate it in the most painful
>> way we can come up with ?

I dunno.  I'll measure it.

>
> Well, how does that differ from having the real syscall instruction there? How
> are we going to filter real (old-)glibc calls from exploits?

Because there are only four vsyscalls: vgettimeofday, vtime, vgetcpu,
and venosys.  None of them have side-effects, so they only allow an
attacker to write something to user memory somewhere.  The
implementation of vgettimeofday needs a syscall instruction internally
for its fallback, which means that an attack could jump there instead
of to the start of the vsyscall implementation.

>
> If it can be filtered in a meaningful way then we should just do that and
> perhaps offer a (default enabled) .config COMPAT_VDSO_EMU=y switch to turn the
> emulation off.
>
> That way we keep the ABI and also have a way out for users who *really* need
> this to work in a performant way.

Yeah, that probably makes more sense.  It'll make for an uglier
diffstat, though -- there's a lot of ugly duplicate code around to
make vgettimeofday and vgetcpu work.

--Andy

>
> Thanks,
>
>        Ingo
>