Re: [PATCH] string: Disable read_word_at_a_time() optimizations if kernel MTE is enabled

[Kees Cook <kees@kernel.org>]

On Mon, Mar 10, 2025 at 05:37:50PM +0000, Catalin Marinas wrote:
> On Fri, Mar 07, 2025 at 07:36:31PM -0800, Kees Cook wrote:
> > On Fri, Mar 07, 2025 at 06:33:13PM -0800, Peter Collingbourne wrote:
> > > The optimized strscpy() and dentry_string_cmp() routines will read 8
> > > unaligned bytes at a time via the function read_word_at_a_time(), but
> > > this is incompatible with MTE which will fault on a partially invalid
> > > read. The attributes on read_word_at_a_time() that disable KASAN are
> > > invisible to the CPU so they have no effect on MTE. Let's fix the
> > > bug for now by disabling the optimizations if the kernel is built
> > > with HW tag-based KASAN and consider improvements for followup changes.
> > 
> > Why is faulting on a partially invalid read a problem? It's still
> > invalid, so ... it should fault, yes? What am I missing?
> 
> read_word_at_a_time() is used to read 8 bytes, potentially unaligned and
> beyond the end of string. The has_zero() function is then used to check
> where the string ends. For this uses, I think we can go with
> load_unaligned_zeropad() which handles a potential fault and pads the
> rest with zeroes.

Agh, right, I keep forgetting that this can read past the end of the
actual allocation. I'd agree, load_unaligned_zeropad() makes sense
there.

> 
> > > Signed-off-by: Peter Collingbourne <pcc@google.com>
> > > Link: https://linux-review.googlesource.com/id/If4b22e43b5a4ca49726b4bf98ada827fdf755548
> > > Fixes: 94ab5b61ee16 ("kasan, arm64: enable CONFIG_KASAN_HW_TAGS")
> > > Cc: stable@vger.kernel.org
> > > ---
> > >  fs/dcache.c  | 2 +-
> > >  lib/string.c | 3 ++-
> > >  2 files changed, 3 insertions(+), 2 deletions(-)
> > 
> > Why are DCACHE_WORD_ACCESS and HAVE_EFFICIENT_UNALIGNED_ACCESS separate
> > things? I can see at least one place where it's directly tied:
> > 
> > arch/arm/Kconfig:58:    select DCACHE_WORD_ACCESS if HAVE_EFFICIENT_UNALIGNED_ACCESS
> 
> DCACHE_WORD_ACCESS requires load_unaligned_zeropad() which handles the
> faults. For some reason, read_word_at_a_time() doesn't expect to fault
> and it is only used with HAVE_EFFICIENT_UNALIGNED_ACCESS. I guess arm32
> only enabled load_unaligned_zeropad() on hardware that supports
> efficient unaligned accesses (v6 onwards), hence the dependency.
> 
> > Would it make sense to sort this out so that KASAN_HW_TAGS can be taken
> > into account at the Kconfig level instead?
> 
> I don't think we should play with config options but rather sort out the
> fault path (load_unaligned_zeropad) or disable MTE temporarily. I'd go
> with the former as long as read_word_at_a_time() is only used for
> strings in conjunction with has_zero(). I haven't checked.

Okay, sounds good. (And with a mild thread-merge: yes, folks want to use
KASAN_HW_TAGS=y in production.)

-- 
Kees Cook