Re: arch/x86/kernel/vsyscall_64.c: overeager NOP of syscalls

[Andi Kleen <ak@muc.de>]

On Thu, Feb 21, 2008 at 09:59:24PM +0100, Thomas Gleixner wrote:
> On Thu, 21 Feb 2008, Thomas Gleixner wrote:
> > > That or we need to do the NOP/un-NOP part in the update_vsyscall code
> > > dependent on if the current clocksource supports vread, instead of on
> > > the /proc entry access.
> > 
> > That won't fly. We need to sychronize the CPUs when we patch the code,
> > which is not possible from update_wall_time with interrupts disabled.
> 
> Also this is utterly stupid as we keep the syscall in cases where we
> do not have vread anyway, so we keep the attack point open for a lot
> of existing machines due to TSC wreckage and HPET unavailability.

Yes that is true, but I didn't find a cheap way around it without
breaking binary compatibility.

I found an expensive way (essentially just putting
a trampoline to a variable mapped vsyscall page on the old static
address), but since it would have added quite a lot of complexity
(vsyscall is inside the kernel mapping and would need to be rewritten
at context switch) and memory overhead in page tables and one more
page per process I didn't do that.

I also considered doing boot time randomization only (which
would avoid a lot of these problems), but it didn't seem worth it.

Also one must say I don't consider it a big security improvement on most
systems.  That is because most programs are not PIC, but linked at a fixed
address and usually if you grep the larger binary for the few bytes
needed for a system call you'll find it at some known offset as 
part of another instruction sequence.

-Andi