Re: [PATCH 4/4] mm/slub: free KFENCE objects in slab_free_hook()

[Marco Elver]

On Wed, 6 Dec 2023 at 14:02, Chengming Zhou <chengming.zhou@xxxxxxxxx> wrote:
>
> On 2023/12/6 17:58, Vlastimil Babka wrote:
> > On 12/5/23 14:27, Chengming Zhou wrote:
> >> On 2023/12/5 03:34, Vlastimil Babka wrote:
> >>> When freeing an object that was allocated from KFENCE, we do that in the
> >>> slowpath __slab_free(), relying on the fact that KFENCE "slab" cannot be
> >>> the cpu slab, so the fastpath has to fallback to the slowpath.
> >>>
> >>> This optimization doesn't help much though, because is_kfence_address()
> >>> is checked earlier anyway during the free hook processing or detached
> >>> freelist building. Thus we can simplify the code by making the
> >>> slab_free_hook() free the KFENCE object immediately, similarly to KASAN
> >>> quarantine.
> >>>
> >>> In slab_free_hook() we can place kfence_free() above init processing, as
> >>> callers have been making sure to set init to false for KFENCE objects.
> >>> This simplifies slab_free(). This places it also above kasan_slab_free()
> >>> which is ok as that skips KFENCE objects anyway.
> >>>
> >>> While at it also determine the init value in slab_free_freelist_hook()
> >>> outside of the loop.
> >>>
> >>> This change will also make introducing per cpu array caches easier.
> >>>
> >>> Tested-by: Marco Elver <elver@xxxxxxxxxx>
> >>> Signed-off-by: Vlastimil Babka <vbabka@xxxxxxx>
> >>> ---
> >>>  mm/slub.c | 22 ++++++++++------------
> >>>  1 file changed, 10 insertions(+), 12 deletions(-)
> >>>
> >>> diff --git a/mm/slub.c b/mm/slub.c
> >>> index ed2fa92e914c..e38c2b712f6c 100644
> >>> --- a/mm/slub.c
> >>> +++ b/mm/slub.c
> >>> @@ -2039,7 +2039,7 @@ static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
> >>>   * production configuration these hooks all should produce no code at all.
> >>>   *
> >>>   * Returns true if freeing of the object can proceed, false if its reuse
> >>> - * was delayed by KASAN quarantine.
> >>> + * was delayed by KASAN quarantine, or it was returned to KFENCE.
> >>>   */
> >>>  static __always_inline
> >>>  bool slab_free_hook(struct kmem_cache *s, void *x, bool init)
> >>> @@ -2057,6 +2057,9 @@ bool slab_free_hook(struct kmem_cache *s, void *x, bool init)
> >>>             __kcsan_check_access(x, s->object_size,
> >>>                                  KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
> >>>
> >>> +   if (kfence_free(kasan_reset_tag(x)))
> >>
> >> I'm wondering if "kasan_reset_tag()" is needed here?
> >
> > I think so, because AFAICS the is_kfence_address() check in kfence_free()
> > could be a false negative otherwise. In fact now I even question some of the
>
> Ok.
>
> > other is_kfence_address() checks in mm/slub.c, mainly
> > build_detached_freelist() which starts from pointers coming directly from
> > slab users. Insight from KASAN/KFENCE folks appreciated :)
> >
> I know very little about KASAN/KFENCE, looking forward to their insight. :)
>
> Just saw a check in __kasan_slab_alloc():
>
>         if (is_kfence_address(object))
>                 return (void *)object;
>
> So thought it seems that a kfence object would be skipped by KASAN.

The is_kfence_address() implementation tolerates tagged addresses,
i.e. if it receives a tagged non-kfence address, it will never return
true.

The KASAN_HW_TAGS patches and KFENCE patches were in development
concurrently, and at the time there was some conflict resolution that
happened when both were merged. The
is_kfence_address(kasan_reset_tag(..)) initially came from [1] but was
squashed into 2b8305260fb.

[1] https://lore.kernel.org/all/9dc196006921b191d25d10f6e611316db7da2efc.1611946152.git.andreyknvl@xxxxxxxxxx/

Andrey, do you recall what issue you encountered that needed kasan_reset_tag()?