[PATCH 1/2] mm/slub: remove redundant kasan_reset_tag() from freelist_ptr calculations

[PATCH 1/2] mm/slub: remove redundant kasan_reset_tag() from freelist_ptr calculations

[Vlastimil Babka]

Commit d36a63a943e3 ("kasan, slub: fix more conflicts with
CONFIG_SLAB_FREELIST_HARDENED") has introduced kasan_reset_tags() to
freelist_ptr() encoding/decoding when CONFIG_SLAB_FREELIST_HARDENED is
enabled to resolve issues when passing tagged or untagged pointers
inconsistently would lead to incorrect calculations.

Later, commit aa1ef4d7b3f6 ("kasan, mm: reset tags when accessing
metadata") made sure all pointers have tags reset regardless of
CONFIG_SLAB_FREELIST_HARDENED, because there was no other way to access
the freepointer metadata safely with hw tag-based KASAN.

Therefore the kasan_reset_tag() usage in freelist_ptr_encode()/decode()
is now redundant, as all callers use kasan_reset_tag() unconditionally
when constructing ptr_addr. Remove the redundant calls and simplify the
code and remove obsolete comments.

Also in freelist_ptr_encode() introduce an 'encoded' variable to make
the lines shorter and make it similar to the _decode() one.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
---
These 2 patches build on top of:
https://lore.kernel.org/all/20230704135834.3884421-1-matteorizzo@google.com/

 mm/slub.c | 22 ++++++----------------
 1 file changed, 6 insertions(+), 16 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index f8cc47eff742..07edad305512 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -374,22 +374,14 @@ typedef struct { unsigned long v; } freeptr_t;
 static inline freeptr_t freelist_ptr_encode(const struct kmem_cache *s,
 					    void *ptr, unsigned long ptr_addr)
 {
+	unsigned long encoded;
+
 #ifdef CONFIG_SLAB_FREELIST_HARDENED
-	/*
-	 * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
-	 * Normally, this doesn't cause any issues, as both set_freepointer()
-	 * and get_freepointer() are called with a pointer with the same tag.
-	 * However, there are some issues with CONFIG_SLUB_DEBUG code. For
-	 * example, when __free_slub() iterates over objects in a cache, it
-	 * passes untagged pointers to check_object(). check_object() in turns
-	 * calls get_freepointer() with an untagged pointer, which causes the
-	 * freepointer to be restored incorrectly.
-	 */
-	return (freeptr_t){.v = (unsigned long)ptr ^ s->random ^
-			swab((unsigned long)kasan_reset_tag((void *)ptr_addr))};
+	encoded = (unsigned long)ptr ^ s->random ^ swab(ptr_addr);
 #else
-	return (freeptr_t){.v = (unsigned long)ptr};
+	encoded = (unsigned long)ptr;
 #endif
+	return (freeptr_t){.v = encoded};
 }
 
 static inline void *freelist_ptr_decode(const struct kmem_cache *s,
@@ -398,9 +390,7 @@ static inline void *freelist_ptr_decode(const struct kmem_cache *s,
 	void *decoded;
 
 #ifdef CONFIG_SLAB_FREELIST_HARDENED
-	/* See the comment in freelist_ptr_encode */
-	decoded = (void *)(ptr.v ^ s->random ^
-		swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
+	decoded = (void *)(ptr.v ^ s->random ^ swab(ptr_addr));
 #else
 	decoded = (void *)ptr.v;
 #endif
-- 
2.41.0

[PATCH 2/2] mm/slub: remove freelist_dereference()

[Vlastimil Babka]

freelist_dereference() is a one-liner only used from get_freepointer().
Remove it and make get_freepointer() call freelist_ptr_decode()
directly to make the code easier to follow.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
---
 mm/slub.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index 07edad305512..c4556a5dab4b 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -397,18 +397,14 @@ static inline void *freelist_ptr_decode(const struct kmem_cache *s,
 	return decoded;
 }
 
-/* Returns the freelist pointer recorded at location ptr_addr. */
-static inline void *freelist_dereference(const struct kmem_cache *s,
-					 void *ptr_addr)
-{
-	return freelist_ptr_decode(s, *(freeptr_t *)(ptr_addr),
-			    (unsigned long)ptr_addr);
-}
-
 static inline void *get_freepointer(struct kmem_cache *s, void *object)
 {
-	object = kasan_reset_tag(object);
-	return freelist_dereference(s, (freeptr_t *)(object + s->offset));
+	unsigned long ptr_addr;
+	freeptr_t p;
+
+	ptr_addr = ((unsigned long)kasan_reset_tag(object)) + s->offset;
+	p = *(freeptr_t *)(ptr_addr);
+	return freelist_ptr_decode(s, p, ptr_addr);
 }
 
 #ifndef CONFIG_SLUB_TINY
-- 
2.41.0

Re: [PATCH 1/2] mm/slub: remove redundant kasan_reset_tag() from freelist_ptr calculations

[Kees Cook]

On Tue, Jul 11, 2023 at 03:46:24PM +0200, Vlastimil Babka wrote:
> Commit d36a63a943e3 ("kasan, slub: fix more conflicts with
> CONFIG_SLAB_FREELIST_HARDENED") has introduced kasan_reset_tags() to
> freelist_ptr() encoding/decoding when CONFIG_SLAB_FREELIST_HARDENED is
> enabled to resolve issues when passing tagged or untagged pointers
> inconsistently would lead to incorrect calculations.
> 
> Later, commit aa1ef4d7b3f6 ("kasan, mm: reset tags when accessing
> metadata") made sure all pointers have tags reset regardless of
> CONFIG_SLAB_FREELIST_HARDENED, because there was no other way to access
> the freepointer metadata safely with hw tag-based KASAN.
> 
> Therefore the kasan_reset_tag() usage in freelist_ptr_encode()/decode()
> is now redundant, as all callers use kasan_reset_tag() unconditionally
> when constructing ptr_addr. Remove the redundant calls and simplify the
> code and remove obsolete comments.
> 
> Also in freelist_ptr_encode() introduce an 'encoded' variable to make
> the lines shorter and make it similar to the _decode() one.
> 
> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>

Thanks, this is much more readable!

Acked-by: Kees Cook <keescook@chromium.org>

-- 
Kees Cook

Re: [PATCH 2/2] mm/slub: remove freelist_dereference()

[Kees Cook]

On Tue, Jul 11, 2023 at 03:46:25PM +0200, Vlastimil Babka wrote:
> freelist_dereference() is a one-liner only used from get_freepointer().
> Remove it and make get_freepointer() call freelist_ptr_decode()
> directly to make the code easier to follow.
> 
> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
> ---
>  mm/slub.c | 16 ++++++----------
>  1 file changed, 6 insertions(+), 10 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 07edad305512..c4556a5dab4b 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -397,18 +397,14 @@ static inline void *freelist_ptr_decode(const struct kmem_cache *s,
>  	return decoded;
>  }
>  
> -/* Returns the freelist pointer recorded at location ptr_addr. */
> -static inline void *freelist_dereference(const struct kmem_cache *s,
> -					 void *ptr_addr)
> -{
> -	return freelist_ptr_decode(s, *(freeptr_t *)(ptr_addr),
> -			    (unsigned long)ptr_addr);
> -}
> -
>  static inline void *get_freepointer(struct kmem_cache *s, void *object)
>  {
> -	object = kasan_reset_tag(object);
> -	return freelist_dereference(s, (freeptr_t *)(object + s->offset));
> +	unsigned long ptr_addr;
> +	freeptr_t p;
> +
> +	ptr_addr = ((unsigned long)kasan_reset_tag(object)) + s->offset;
> +	p = *(freeptr_t *)(ptr_addr);
> +	return freelist_ptr_decode(s, p, ptr_addr);
>  }
>  
>  #ifndef CONFIG_SLUB_TINY
> -- 
> 2.41.0
> 

I like reducing the complexity here, but I find dropping the "object"
reassignment makes this a bit harder to read. What about:

	object = kasan_reset_tag(object);
	unsigned long ptr_addr = (unsigned long)object + s->offset;
	freeptr_t p = *(freeptr_t *)(ptr_addr);

	return freelist_ptr_decode(s, p, ptr_addr);

?

They're the same result, so either way:

Acked-by: Kees Cook <keescook@chromium.org>

-- 
Kees Cook

Re: [PATCH 1/2] mm/slub: remove redundant kasan_reset_tag() from freelist_ptr calculations

[Andrey Konovalov]

On Tue, Jul 11, 2023 at 3:46 PM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> Commit d36a63a943e3 ("kasan, slub: fix more conflicts with
> CONFIG_SLAB_FREELIST_HARDENED") has introduced kasan_reset_tags() to
> freelist_ptr() encoding/decoding when CONFIG_SLAB_FREELIST_HARDENED is
> enabled to resolve issues when passing tagged or untagged pointers
> inconsistently would lead to incorrect calculations.
>
> Later, commit aa1ef4d7b3f6 ("kasan, mm: reset tags when accessing
> metadata") made sure all pointers have tags reset regardless of
> CONFIG_SLAB_FREELIST_HARDENED, because there was no other way to access
> the freepointer metadata safely with hw tag-based KASAN.
>
> Therefore the kasan_reset_tag() usage in freelist_ptr_encode()/decode()
> is now redundant, as all callers use kasan_reset_tag() unconditionally
> when constructing ptr_addr. Remove the redundant calls and simplify the
> code and remove obsolete comments.
>
> Also in freelist_ptr_encode() introduce an 'encoded' variable to make
> the lines shorter and make it similar to the _decode() one.
>
> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
> ---
> These 2 patches build on top of:
> https://lore.kernel.org/all/20230704135834.3884421-1-matteorizzo@google.com/
>
>  mm/slub.c | 22 ++++++----------------
>  1 file changed, 6 insertions(+), 16 deletions(-)
>
> diff --git a/mm/slub.c b/mm/slub.c
> index f8cc47eff742..07edad305512 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -374,22 +374,14 @@ typedef struct { unsigned long v; } freeptr_t;
>  static inline freeptr_t freelist_ptr_encode(const struct kmem_cache *s,
>                                             void *ptr, unsigned long ptr_addr)
>  {
> +       unsigned long encoded;
> +
>  #ifdef CONFIG_SLAB_FREELIST_HARDENED
> -       /*
> -        * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
> -        * Normally, this doesn't cause any issues, as both set_freepointer()
> -        * and get_freepointer() are called with a pointer with the same tag.
> -        * However, there are some issues with CONFIG_SLUB_DEBUG code. For
> -        * example, when __free_slub() iterates over objects in a cache, it
> -        * passes untagged pointers to check_object(). check_object() in turns
> -        * calls get_freepointer() with an untagged pointer, which causes the
> -        * freepointer to be restored incorrectly.
> -        */
> -       return (freeptr_t){.v = (unsigned long)ptr ^ s->random ^
> -                       swab((unsigned long)kasan_reset_tag((void *)ptr_addr))};
> +       encoded = (unsigned long)ptr ^ s->random ^ swab(ptr_addr);
>  #else
> -       return (freeptr_t){.v = (unsigned long)ptr};
> +       encoded = (unsigned long)ptr;
>  #endif
> +       return (freeptr_t){.v = encoded};
>  }
>
>  static inline void *freelist_ptr_decode(const struct kmem_cache *s,
> @@ -398,9 +390,7 @@ static inline void *freelist_ptr_decode(const struct kmem_cache *s,
>         void *decoded;
>
>  #ifdef CONFIG_SLAB_FREELIST_HARDENED
> -       /* See the comment in freelist_ptr_encode */
> -       decoded = (void *)(ptr.v ^ s->random ^
> -               swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
> +       decoded = (void *)(ptr.v ^ s->random ^ swab(ptr_addr));
>  #else
>         decoded = (void *)ptr.v;
>  #endif
> --
> 2.41.0

Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>

Thanks!

Re: [PATCH 2/2] mm/slub: remove freelist_dereference()

[Vlastimil Babka]

On 7/11/23 18:21, Kees Cook wrote:
> On Tue, Jul 11, 2023 at 03:46:25PM +0200, Vlastimil Babka wrote:
>>  
>>  #ifndef CONFIG_SLUB_TINY
>> -- 
>> 2.41.0
>> 
> 
> I like reducing the complexity here, but I find dropping the "object"
> reassignment makes this a bit harder to read. What about:

Alright.

> 	object = kasan_reset_tag(object);
> 	unsigned long ptr_addr = (unsigned long)object + s->offset;
> 	freeptr_t p = *(freeptr_t *)(ptr_addr);

Are we really so benevolent with declaration-after-statement now? :)

> 	return freelist_ptr_decode(s, p, ptr_addr);
> 
> ?
> 
> They're the same result, so either way:
> 
> Acked-by: Kees Cook <keescook@chromium.org>
> 

Re: [PATCH 2/2] mm/slub: remove freelist_dereference()

[Vlastimil Babka]

On 7/13/23 09:44, Vlastimil Babka wrote:
> On 7/11/23 18:21, Kees Cook wrote:
>> On Tue, Jul 11, 2023 at 03:46:25PM +0200, Vlastimil Babka wrote:
>>>  
>>>  #ifndef CONFIG_SLUB_TINY
>>> -- 
>>> 2.41.0
>>> 
>> 
>> I like reducing the complexity here, but I find dropping the "object"
>> reassignment makes this a bit harder to read. What about:
> 
> Alright.
> 
>> 	object = kasan_reset_tag(object);
>> 	unsigned long ptr_addr = (unsigned long)object + s->offset;
>> 	freeptr_t p = *(freeptr_t *)(ptr_addr);
> 
> Are we really so benevolent with declaration-after-statement now? :)

I've left the declarations separate for now so it's similar to
get_freepointer_safe(). Pushed the result to slab/for-6.6/cleanup and
for-next. Thanks for the reviews!

>> 	return freelist_ptr_decode(s, p, ptr_addr);
>> 
>> ?
>> 
>> They're the same result, so either way:
>> 
>> Acked-by: Kees Cook <keescook@chromium.org>
>> 
>