Re: [PATCH -tip] kasan: Emit different calls for instrumentable memintrinsics

[Andrey Konovalov]

On Fri, Feb 10, 2023 at 12:35 AM Marco Elver <elver@xxxxxxxxxx> wrote:
>
> On Thu, 9 Feb 2023 at 23:43, Andrey Konovalov <andreyknvl@xxxxxxxxx> wrote:
> >
> > On Wed, Feb 8, 2023 at 7:42 PM Marco Elver <elver@xxxxxxxxxx> wrote:
> > >
> > > Clang 15 will provide an option to prefix calls to memcpy/memset/memmove
> > > with __asan_ in instrumented functions: https://reviews.llvm.org/D122724
> >
> > Hi Marco,
> >
> > Does this option affect all functions or only the ones that are marked
> > with no_sanitize?
>
> Only functions that are instrumented, i.e. wherever
> fsanitize=kernel-address inserts instrumentation.

Ack.

> > Based on the LLVM patch description, should we also change the normal
> > memcpy/memset/memmove to be noninstrumented?
>
> They are no longer instrumented as of 69d4c0d32186 (for
> CONFIG_GENERIC_ENTRY arches).

Ah, sorry, overlooked that.

> > These __asan_mem* functions are not defined in the kernel AFAICS.
> > Should we add them?
>
> Peter introduced them in 69d4c0d32186, and we effectively have no
> mem*() instrumentation on x86 w/o the compiler-enablement patch here.
>
> > Or maybe we should just use "__" as the prefix, as right now __mem*
> > functions are the ones that are not instrumented?
>
> __asan_mem* is for instrumented code, just like ASan userspace does
> (actually ASan userspace has been doing it like this forever, just the
> kernel was somehow special).
>
> [...]
> > > Fixes: 69d4c0d32186 ("entry, kasan, x86: Disallow overriding mem*() functions")
> > > Signed-off-by: Marco Elver <elver@xxxxxxxxxx>
> > > ---
> > >
> > > The Fixes tag is just there to show the dependency, and that people
> > > shouldn't apply this patch without 69d4c0d32186.
>
> ^^^ Depends on this commit, which is only in -tip.

Got it. Missed that patch.

> > > +ifdef CONFIG_GENERIC_ENTRY
>
> It also only affects GENERIC_ENTRY arches.
>
> > > +# Instrument memcpy/memset/memmove calls by using instrumented __asan_mem*()
> > > +# instead. With compilers that don't support this option, compiler-inserted
> > > +# memintrinsics won't be checked by KASAN.
> > > +CFLAGS_KASAN += $(call cc-param,asan-kernel-mem-intrinsic-prefix)
> > > +endif
>
> Probably the same should be done for SW_TAGS, because arm64 will be
> GENERIC_ENTRY at one point or another as well.

Yes, makes sense. I'll file a bug for this once I fully understand the
consequences of these changes.

> KASAN + GCC on x86 will have no mem*() instrumentation after
> 69d4c0d32186, which is sad, so somebody ought to teach it the same
> param as above.

Hm, with that patch we would have no KASAN checking within normal mem*
functions (not the ones embedded by the compiler) on GENERIC_ENTRY
arches even with Clang, right?