Re: [PATCH] mm/slab: ensure all metadata in slab object are word-aligned

[Andrey Konovalov <andreyknvl@gmail.com>]

On Fri, Oct 24, 2025 at 10:35 AM Harry Yoo <harry.yoo@oracle.com> wrote:
>
> > An alternative to unpoisoning or disabling KASAN could be to add
> > helper functions annotated with __no_sanitize_address that do the
> > required accesses. And make them inlined when KASAN is disabled to
> > avoid the performance hit.
>
> This sounds reasonable, let me try this instead of unpoisoning
> metadata. Thanks.

But note that you still need kasan_reset_tag() for HW_TAGS KASAN: this
mode is not based on compiler instrumentation and thus
__no_sanitize_address has no effect on it.

(There's been some discussion on making __no_sanitize_address work for
HW_TAGS: https://bugzilla.kernel.org/show_bug.cgi?id=212513#c2, but
this was never attempted.)

> > On a side note, you might also need to check whether SW_TAGS KASAN and
> > KMSAN would be unhappy with your changes:
> >
> > - When we do kasan_disable_current() or metadata_access_enable(), we
> > also do kasan_reset_tag();
> > - In metadata_access_enable(), we disable KMSAN as well.
>
> Thanks for pointing this out!
>
> Just to clarify, by calling kasan_reset_tag() we clear tag from the address
> so that SW or HW tag based KASAN won't report access violation? (because
> there is no valid tag in the address?)

Yeah, kind of: kasan_reset_tag() sets the pointer tag (the top byte)
to 0xFF. With SW_TAGS KASAN, the compiler knows not to embed validity
checks for accesses through pointers with 0xFF in the top byte. With
HW_TAGS KASAN, the CPU is instructed to behave the same.

(This is slightly different than kasan_disable_current(): with
kasan_reset_tag(), validity checks do not happen at all. With
kasan_disable_current(), the checks happen but the bug reports are
ignored.)

Thank you!