kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

[PATCH] kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

Posted by Joey Jiao 2 years, 2 months ago

Fow low memory device, full enabled kasan just not work.
Set KASAN_SANITIZE to n when CONFIG_KASAN_WHITELIST_ONLY=y.
So we can enable kasan for single file or module.

Signed-off-by: Joey Jiao <quic_jiangenj@quicinc.com>
---
 lib/Kconfig.kasan    | 8 ++++++++
 scripts/Makefile.lib | 3 +++
 2 files changed, 11 insertions(+)

diff --git a/lib/Kconfig.kasan b/lib/Kconfig.kasan
index fdca89c05745..1cec4e204831 100644
--- a/lib/Kconfig.kasan
+++ b/lib/Kconfig.kasan
@@ -153,6 +153,14 @@ config KASAN_INLINE
 
 endchoice
 
+config KASAN_WHITELIST_ONLY
+	bool "Whitelist only KASAN"
+	depends on KASAN && !KASAN_HW_TAGS
+	default n
+	help
+	  Say Y here to only enable KASAN for module or files which has explicitly
+	  set KASAN_SANITIZE:=y which is helpful especially for memory limited devices.
+
 config KASAN_STACK
 	bool "Stack instrumentation (unsafe)" if CC_IS_CLANG && !COMPILE_TEST
 	depends on KASAN_GENERIC || KASAN_SW_TAGS
diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 68d0134bdbf9..e8d608ea369c 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -158,6 +158,9 @@ endif
 #
 ifeq ($(CONFIG_KASAN),y)
 ifneq ($(CONFIG_KASAN_HW_TAGS),y)
+ifeq ($(CONFIG_KASAN_WHITELIST_ONLY),y)
+KASAN_SANITIZE ?= n
+endif
 _c_flags += $(if $(patsubst n%,, \
 		$(KASAN_SANITIZE_$(basetarget).o)$(KASAN_SANITIZE)y), \
 		$(CFLAGS_KASAN), $(CFLAGS_KASAN_NOSANITIZE))
-- 
2.38.1

Re: [PATCH] kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

Posted by Alexander Potapenko 2 years, 2 months ago

(CC Masahiro Yamada)

On Thu, Sep 28, 2023 at 6:16 AM Joey Jiao <quic_jiangenj@quicinc.com> wrote:
>
> Fow low memory device, full enabled kasan just not work.
> Set KASAN_SANITIZE to n when CONFIG_KASAN_WHITELIST_ONLY=y.
> So we can enable kasan for single file or module.

I don't have technical objections here, but it bothers me a bit that
we are adding support for KASAN_SANITIZE:=y, although nobody will be
adding KASAN_SANITIZE:=y to upstream Makefiles - only development
kernels when debugging on low-end devices.

Masahiro, is this something worth having in upstream Kconfig code?

> Signed-off-by: Joey Jiao <quic_jiangenj@quicinc.com>
Reviewed-by: Alexander Potapenko <glider@google.com>

Re: [PATCH] kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

Posted by Masahiro Yamada 2 years, 2 months ago

On Fri, Sep 29, 2023 at 11:06 PM Alexander Potapenko <glider@google.com> wrote:
>
> (CC Masahiro Yamada)
>
> On Thu, Sep 28, 2023 at 6:16 AM Joey Jiao <quic_jiangenj@quicinc.com> wrote:
> >
> > Fow low memory device, full enabled kasan just not work.
> > Set KASAN_SANITIZE to n when CONFIG_KASAN_WHITELIST_ONLY=y.
> > So we can enable kasan for single file or module.
>
> I don't have technical objections here, but it bothers me a bit that
> we are adding support for KASAN_SANITIZE:=y, although nobody will be
> adding KASAN_SANITIZE:=y to upstream Makefiles - only development
> kernels when debugging on low-end devices.
>
> Masahiro, is this something worth having in upstream Kconfig code?


Even if we apply this patch to the upstream,
you will end up with adding 'KASAN_SANITIZE :=y'
to the single file/Makefile.

I am not convinced with this patch
since this nod is not so useful standalone.



> > Signed-off-by: Joey Jiao <quic_jiangenj@quicinc.com>
> Reviewed-by: Alexander Potapenko <glider@google.com>



--
Best Regards
Masahiro Yamada

RE: [PATCH] kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

Posted by Joey Jiao (QUIC) 2 years, 2 months ago

Right, it will be only useful for low memory kernel where 'KASAN_SANITIZE :=y' has to be added explicitly in local as hotfix.

-----Original Message-----
From: Masahiro Yamada <masahiroy@kernel.org> 
Sent: Saturday, September 30, 2023 6:12 PM
To: Alexander Potapenko <glider@google.com>
Cc: Joey Jiao (QUIC) <quic_jiangenj@quicinc.com>; kasan-dev@googlegroups.com; Kevin Ding (QUIC) <quic_likaid@quicinc.com>; Andrey Ryabinin <ryabinin.a.a@gmail.com>; Andrey Konovalov <andreyknvl@gmail.com>; Dmitry Vyukov <dvyukov@google.com>; Vincenzo Frascino <vincenzo.frascino@arm.com>; Nathan Chancellor <nathan@kernel.org>; Nick Desaulniers <ndesaulniers@google.com>; Nicolas Schier <nicolas@fjasle.eu>; linux-kernel@vger.kernel.org; linux-kbuild@vger.kernel.org
Subject: Re: [PATCH] kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

On Fri, Sep 29, 2023 at 11:06 PM Alexander Potapenko <glider@google.com> wrote:
>
> (CC Masahiro Yamada)
>
> On Thu, Sep 28, 2023 at 6:16 AM Joey Jiao <quic_jiangenj@quicinc.com> wrote:
> >
> > Fow low memory device, full enabled kasan just not work.
> > Set KASAN_SANITIZE to n when CONFIG_KASAN_WHITELIST_ONLY=y.
> > So we can enable kasan for single file or module.
>
> I don't have technical objections here, but it bothers me a bit that 
> we are adding support for KASAN_SANITIZE:=y, although nobody will be 
> adding KASAN_SANITIZE:=y to upstream Makefiles - only development 
> kernels when debugging on low-end devices.
>
> Masahiro, is this something worth having in upstream Kconfig code?


Even if we apply this patch to the upstream, you will end up with adding 'KASAN_SANITIZE :=y'
to the single file/Makefile.

I am not convinced with this patch
since this nod is not so useful standalone.



> > Signed-off-by: Joey Jiao <quic_jiangenj@quicinc.com>
> Reviewed-by: Alexander Potapenko <glider@google.com>



--
Best Regards
Masahiro Yamada

Re: [PATCH] kasan: Add CONFIG_KASAN_WHITELIST_ONLY mode

Posted by Andrey Konovalov 2 years, 2 months ago

On Sat, Sep 30, 2023 at 12:13 PM Masahiro Yamada <masahiroy@kernel.org> wrote:
>
> On Fri, Sep 29, 2023 at 11:06 PM Alexander Potapenko <glider@google.com> wrote:
> >
> > (CC Masahiro Yamada)
> >
> > On Thu, Sep 28, 2023 at 6:16 AM Joey Jiao <quic_jiangenj@quicinc.com> wrote:
> > >
> > > Fow low memory device, full enabled kasan just not work.
> > > Set KASAN_SANITIZE to n when CONFIG_KASAN_WHITELIST_ONLY=y.
> > > So we can enable kasan for single file or module.
> >
> > I don't have technical objections here, but it bothers me a bit that
> > we are adding support for KASAN_SANITIZE:=y, although nobody will be
> > adding KASAN_SANITIZE:=y to upstream Makefiles - only development
> > kernels when debugging on low-end devices.
> >
> > Masahiro, is this something worth having in upstream Kconfig code?
>
>
> Even if we apply this patch to the upstream,
> you will end up with adding 'KASAN_SANITIZE :=y'
> to the single file/Makefile.
>
> I am not convinced with this patch
> since this nod is not so useful standalone.

Yeah, I agree here, I don't think this change belongs as is in the
upstream KASAN code.