Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 21 Jun 2017 10:24:56 +0100
From: Mark Rutland <mark.rutland@....com>
To: Kees Cook <keescook@...omium.org>
Cc: Laura Abbott <labbott@...hat.com>,
	Alexander Popov <alex.popov@...ux.com>,
	"kernel-hardening@...ts.openwall.com" <kernel-hardening@...ts.openwall.com>,
	PaX Team <pageexec@...email.hu>,
	Brad Spengler <spender@...ecurity.net>,
	Tycho Andersen <tycho@...ker.com>,
	Ard Biesheuvel <ard.biesheuvel@...aro.org>
Subject: Re: Re: [PATCH RFC v2 1/1] gcc-plugins: Add
 stackleak feature erasing the kernel stack at the end of syscalls

On Tue, Jun 20, 2017 at 12:11:56PM -0700, Kees Cook wrote:
> On Tue, Jun 20, 2017 at 4:20 AM, Mark Rutland <mark.rutland@....com> wrote:
> > On Tue, Jun 13, 2017 at 02:51:59PM -0700, Laura Abbott wrote:
> >> On 06/09/2017 10:28 AM, Kees Cook wrote:
> >> > It seems like it shouldn't be too hard to add on-user-return erasure
> >> > code to other architectures too.
> >>
> >> I played around getting this to compile for arm64 with a dummy
> >> stack clearing function. arm64 is doing something special with the
> >> efistub so it fails to link with
> >>
> >> drivers/firmware/efi/libstub/arm-stub.c:45:(.init.text+0x54): relocation
> >>       truncated to fit: R_AARCH64_CALL26 against undefined symbol `__efistub_track_stack'
> >>
> >> The relocation to the .init.text section and appending __efistub happens after
> >> compilation so the checks in the plugin itself don't work. I haven't come up
> >> with a solution to not have the plugin run on the stub yet.
> >
> > Can we do something like what we do for KCOV, and (only) place the
> > plugin-invoking flags to into CFLAGS_STACKLEAK, which we can filter out
> > in scripts/Makefile.lib?
> 
> That seems fine to me, though plugins have tended to provide a
> DISABLE_foo_bar_PLUGIN export (like used in ppc for turning off the
> latent entropy plugin):
> 
> scripts/Makefile.gcc-plugins:    DISABLE_LATENT_ENTROPY_PLUGIN +=
> -fplugin-arg-latent_entropy_plugin-disable
> 
> arch/powerpc/kernel/Makefile:CFLAGS_cputable.o +=
> $(DISABLE_LATENT_ENTROPY_PLUGIN)
> arch/powerpc/kernel/Makefile:CFLAGS_prom_init.o +=
> $(DISABLE_LATENT_ENTROPY_PLUGIN)
> ...

Interesting. I hadn't encountered that style before.

I'd seen the per-makefile config disables:

drivers/firmware/efi/libstub/Makefile:GCOV_PROFILE                      := n
drivers/firmware/efi/libstub/Makefile:KASAN_SANITIZE                    := n
drivers/firmware/efi/libstub/Makefile:UBSAN_SANITIZE                    := n
drivers/firmware/efi/libstub/Makefile:KCOV_INSTRUMENT                   := n

... and per-object config disables:

arch/arm64/mm/Makefile:KASAN_SANITIZE_physaddr.o        += n
arch/arm64/mm/Makefile:KASAN_SANITIZE_kasan_init.o      := n

... which AFAICT are both handled by the standard scripts/Makefile.lib idiom:

ifeq ($(CONFIG_KASAN),y)
_c_flags += $(if $(patsubst n%,, \
                $(KASAN_SANITIZE_$(basetarget).o)$(KASAN_SANITIZE)y), \
                $(CFLAGS_KASAN))
endif

... it seems a shame for plugins to follow a different pattern.

Thanks,
Mark.

Powered by blists - more mailing lists

Your e-mail address:

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.