Re: [PATCH mm] slub, kasan: improve interaction of KASAN and slub_debug poisoning

[Feng Tang]

On Thu, Nov 23, 2023 at 02:26:13PM +0800, Tang, Feng wrote:
[...]
> > -#ifdef CONFIG_KASAN_GENERIC
> >  	/*
> > -	 * KASAN could save its free meta data in object's data area at
> > -	 * offset 0, if the size is larger than 'orig_size', it will
> > -	 * overlap the data redzone in [orig_size+1, object_size], and
> > -	 * the check should be skipped.
> > +	 * KASAN can save its free meta data inside of the object at offset 0.
> > +	 * If this meta data size is larger than 'orig_size', it will overlap
> > +	 * the data redzone in [orig_size+1, object_size]. Thus, we adjust
> > +	 * 'orig_size' to be as at least as big as KASAN's meta data.
> >  	 */
> > -	if (kasan_metadata_size(s, true) > orig_size)
> > -		orig_size = s->object_size;
> > -#endif
> > +	kasan_meta_size = kasan_metadata_size(s, true);
> > +	if (kasan_meta_size > orig_size)
> > +		orig_size = kasan_meta_size;
> 
> 'orig_size' is to save the orignal request size for kmalloc object,
> and its main purpose is to detect the memory wastage of kmalloc
> objects, see commit 6edf2576a6cc "mm/slub: enable debugging memory
> wasting of kmalloc"
> 
> Setting "orig_size = s->object_size" was to skip the wastage check
> and the redzone sanity check for this 'wasted space'.
> 
> So it's better not to set 'kasan_meta_size' to orig_size.
> 
> And from the below code, IIUC, the orig_size is not used in fixing
> the boot problem found by Hyeonggon?

I just tried Hyeonggon's reproducing method [1], and confirmed the
below change of check_object() itself can fix the problem.

[1]. https://lore.kernel.org/lkml/CAB=+i9RnOz0jDockOfw3oNageCUF5gmF+nzOzPpoTxtr7eqn7g@xxxxxxxxxxxxxx/

Thanks,
Feng

> 
> Thanks,
> Feng
> 
> >  
> >  	p += get_info_end(s);
> >  	p += sizeof(struct track) * 2;
> > @@ -1192,7 +1192,7 @@ static int check_object(struct kmem_cache *s, struct slab *slab,
> >  {
> >  	u8 *p = object;
> >  	u8 *endobject = object + s->object_size;
> > -	unsigned int orig_size;
> > +	unsigned int orig_size, kasan_meta_size;
> >  
> >  	if (s->flags & SLAB_RED_ZONE) {
> >  		if (!check_bytes_and_report(s, slab, object, "Left Redzone",
> > @@ -1222,12 +1222,23 @@ static int check_object(struct kmem_cache *s, struct slab *slab,
> >  	}
> >  
> >  	if (s->flags & SLAB_POISON) {
> > -		if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
> > -			(!check_bytes_and_report(s, slab, p, "Poison", p,
> > -					POISON_FREE, s->object_size - 1) ||
> > -			 !check_bytes_and_report(s, slab, p, "End Poison",
> > -				p + s->object_size - 1, POISON_END, 1)))
> > -			return 0;
> > +		if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON)) {
> > +			/*
> > +			 * KASAN can save its free meta data inside of the
> > +			 * object at offset 0. Thus, skip checking the part of
> > +			 * the redzone that overlaps with the meta data.
> > +			 */
> > +			kasan_meta_size = kasan_metadata_size(s, true);
> > +			if (kasan_meta_size < s->object_size - 1 &&
> > +			    !check_bytes_and_report(s, slab, p, "Poison",
> > +					p + kasan_meta_size, POISON_FREE,
> > +					s->object_size - kasan_meta_size - 1))
> > +				return 0;
> > +			if (kasan_meta_size < s->object_size &&
> > +			    !check_bytes_and_report(s, slab, p, "End Poison",
> > +					p + s->object_size - 1, POISON_END, 1))
> > +				return 0;
> > +		}
> >  		/*
> >  		 * check_pad_bytes cleans up on its own.
> >  		 */
> > -- 
> > 2.25.1
> >