Hi Linus,
I was curious, after the last week of discussion, what you thought of
the future of Alexander's port of the STACKLEAK plugin[0]. Given that
there is progress being made (by at least 8 people at last count) to
actually remove VLAs (and bogus warnings)[1], and that there appears
to be consensus[2] on the approach for how to deal with uninitialized
stack variables, this still leaves an aspect of STACKLEAK unaddressed,
which is reducing the lifetime of stack content validity.
We have options available for heap memory poison-on-free[3] (which can
serve both as a debugging feature and a security feature), but we
continue not to have this for the stack. I'd still like to be able to
provide coverage here. AIUI, your objections revolved around not
directly addressing the VLA and uninit cases (which are now underway).
Would you reconsider your NACK, and if not, what do you think the
right approach would be for performing stack clearing at the end of
syscalls?
Thanks!
-Kees
[0] http://www.openwall.com/lists/kernel-hardening/2018/03/03/7
[1] https://patchwork.kernel.org/project/LKML/list/?q=VLA
[2] https://marc.info/?l=kernel-hardening&m=152036383124266&w=2
[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/Kconfig.debug?h=v4.15#n42
--
Kees Cook
Pixel Security
On Mon, Mar 12, 2018 at 10:06 AM, Kees Cook <[email protected]> wrote:
>
> I was curious, after the last week of discussion, what you thought of
> the future of Alexander's port of the STACKLEAK plugin[0]. Given that
> there is progress being made (by at least 8 people at last count) to
> actually remove VLAs (and bogus warnings)[1], and that there appears
> to be consensus[2] on the approach for how to deal with uninitialized
> stack variables, this still leaves an aspect of STACKLEAK unaddressed,
> which is reducing the lifetime of stack content validity.
Honestly, I consider that to be one of those crazy patches that people
can apply if they want to, but that there is no point in having
upstream.
Hundreds of extra lines of assembly for something that isn't even a
leak or a theoretical fix, when there is a better model for just
improving the compiler? Yeah, no.
The fact is, people can do their own thing. But for it to make sense
_mainline_, it has to improve kernel development, and I don't see it
doing that.
I just haven't seen an argument for why it makes sense to do the
belt-and-suspenders-and-glue-your-pants-on approach.
Linus