Skip to contentSkip to conversation

slub: choose the right freelist pointer location when creating small caches

PATCH NAK Lin, Zhenpeng 2 participants 9 messages Jun 5, 2021 – Jun 11, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 slub: choose the right freelist pointer location when creating small caches

When enabling CONFIG_SLUB_DEBUG and booting with "slub_debug=Z", the kernel crashes at creating caches if the object size is smaller than 2*sizeof(void*). The problem is due to the wrong calculation of freepointer_area. The freelist pointer can be stored in the middle of object only if the object size is not smaller than 2*sizeof(void*). Otherwise, the freelist pointer will be corrupted by SLUB_RED_ZONE.

Fixes 3202fa62fb43 ("slub: relocate freelist pointer to middle of object")
Fixes 89b83f282d8b ("slub: avoid redzone when choosing freepointer location")

---
mm/slub.c | 4 ++--

1 file changed, 2 insertions(+), 2 deletions(-)
mm/slub.c +0 -0
index 3f96e099817a..cb23233ee683 100644
@@ -3704,7 +3704,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)

* can't use that portion for writing the freepointer, so
* s->offset must be limited within this for the general case. */
- freepointer_area = size;
+ freepointer_area = s->object_size;

#ifdef CONFIG_SLUB_DEBUG
	/*
@@ -3751,7 +3751,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
		 */
		s->offset = size;
		size += sizeof(void *);

- } else if (freepointer_area > sizeof(void *)) {
+ } else if (freepointer_area >= 2 * sizeof(void *)) {
/*
* Store freelist pointer near middle of object to keep
* it away from the edges of the object to avoid small
--
2.17.1

#1 lore raw
Quoted 38 lines · Zhenpeng

When enabling CONFIG_SLUB_DEBUG and booting with "slub_debug=Z", the kernel crashes at creating caches if the object size is smaller than 2*sizeof(void*). The problem is due to the wrong calculation of freepointer_area. The freelist pointer can be stored in the middle of object only if the object size is not smaller than 2*sizeof(void*). Otherwise, the freelist pointer will be corrupted by SLUB_RED_ZONE.

Fixes 3202fa62fb43 ("slub: relocate freelist pointer to middle of object")
Fixes 89b83f282d8b ("slub: avoid redzone when choosing freepointer location")

---
mm/slub.c | 4 ++--

1 file changed, 2 insertions(+), 2 deletions(-)
mm/slub.c +2 -3
index 3f96e099817a..cb23233ee683 100644
@@ -3704,7 +3704,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
 	 * can't use that portion for writing the freepointer, so
 	 * s->offset must be limited within this for the general case.
 	 */
-	freepointer_area = size;
+	freepointer_area = s->object_size;
 
 #ifdef CONFIG_SLUB_DEBUG
 	/*
@@ -3751,7 +3751,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
 		 */
 		s->offset = size;
 		size += sizeof(void *);
-	} else if (freepointer_area > sizeof(void *)) {
+	} else if (freepointer_area >= 2 * sizeof(void *)) {
 		/*
 		 * Store freelist pointer near middle of object to keep
 		 * it away from the edges of the object to avoid small
--
Quoted 1 line

2.17.1

NAK, I'd prefer this get cleaned up more completely, especially since there are no objects of that size in the kernel currently:

https://lore.kernel.org/lkml/[email protected]/

Signature · 1 line
Kees Cook
#2 lore raw

There do exist objects whose size is smaller than 2*sizeof(void*). E.g. struct ccid in DCCP module.

On 6/8/21, 2:26 PM, "Kees Cook" <[email protected]> wrote:

    On Sat, Jun 05, 2021 at 01:58:13AM +0000, Lin, Zhenpeng wrote:
    > When enabling CONFIG_SLUB_DEBUG and booting with "slub_debug=Z", the
    > kernel crashes at creating caches if the object size is smaller
    > than 2*sizeof(void*). The problem is due to the wrong calculation
    > of freepointer_area. The freelist pointer can be stored in the
    > middle of object only if the object size is not smaller than
    > 2*sizeof(void*). Otherwise, the freelist pointer will be corrupted by
    > SLUB_RED_ZONE.
    > 
    > Fixes: 3202fa62fb43 ("slub: relocate freelist pointer to middle of object")
    > Fixes: 89b83f282d8b ("slub: avoid redzone when choosing freepointer location")
    > Signed-off-by: Zhenpeng Lin <[email protected]>
    > ---
    > mm/slub.c | 4 ++--
    > 1 file changed, 2 insertions(+), 2 deletions(-)
    > 
    > diff --git a/mm/slub.c b/mm/slub.c
    > index 3f96e099817a..cb23233ee683 100644
    > --- a/mm/slub.c
    > +++ b/mm/slub.c
    > @@ -3704,7 +3704,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
    > 	 * can't use that portion for writing the freepointer, so
    > 	 * s->offset must be limited within this for the general case.
    > 	 */
    > -	freepointer_area = size;
    > +	freepointer_area = s->object_size;
    > 
    > #ifdef CONFIG_SLUB_DEBUG
    > 	/*
    > @@ -3751,7 +3751,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
    > 		 */
    > 		s->offset = size;
    > 		size += sizeof(void *);
    > -	} else if (freepointer_area > sizeof(void *)) {
    > +	} else if (freepointer_area >= 2 * sizeof(void *)) {
    > 		/*
    > 		 * Store freelist pointer near middle of object to keep
    > 		 * it away from the edges of the object to avoid small
    > --
    > 2.17.1

    NAK, I'd prefer this get cleaned up more completely, especially since
    there are no objects of that size in the kernel currently:

    https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Flkml%2F20201015033712.1491731-1-keescook%40chromium.org%2F&amp;data=04%7C01%7Czplin%40psu.edu%7C28b6f3c5a3b149be56e808d92aaafd26%7C7cf48d453ddb4389a9c1c115526eb52e%7C0%7C0%7C637587736155493816%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=%2F8CXVkqlhA7RnfX%2BDP07%2F4t1NIw1CHsUpuuWrsLyU9o%3D&amp;reserved=0

    -- 
    Kees Cook
#3 lore raw
Quoted 1 line · Zhenpeng

There do exist objects whose size is smaller than 2*sizeof(void*). E.g. struct ccid in DCCP module.

Yes, sorry, I meant sizeof(void *). I've sent an updated v4 series and CCed you. Are you able to test that and see if it fixes it for you too?

Thanks for the push to dust this series off again! :)

-Kees

Signature · 1 line
Kees Cook
#4 lore raw

No problem. Just took a look and tested the patch, it looks good to me!

On 6/8/21, 2:41 PM, "Kees Cook" <[email protected]> wrote:

On Tue, Jun 08, 2021 at 06:33:01PM +0000, Lin, Zhenpeng wrote:
> There do exist objects whose size is smaller than 2*sizeof(void*). E.g. struct ccid in DCCP module.

Yes, sorry, I meant sizeof(void *). I've sent an updated v4 series and CCed you. Are you able to test that and see if it fixes it for you too?

Thanks for the push to dust this series off again! :)

-Kees

--
Kees Cook

#5 lore raw
Quoted 1 line · Zhenpeng

No problem. Just took a look and tested the patch, it looks good to me!

Great; thank you! Sorry I dropped the ball on this series. I got distracted. :) It looks like akpm took it into -mm now, so this should be fixed in -next soon.

Signature · 1 line
Kees Cook
#6 lore raw

Sounds good. But I would suggest this to go to -stable as soon as possible. Because this bug is affecting the basic functionality of DCCP. It crashes kernel whenever a new socket in this module is created.

Best,
Zhenpeng

-----Original Message-----
From: Kees Cook <[email protected]>
Date: Tuesday, June 8, 2021 at 7:14 PM
To: "Lin, Zhenpeng" <[email protected]>
Cc: Christoph Lameter <[email protected]>, Pekka Enberg <[email protected]>, David Rientjes <[email protected]>, Joonsoo Kim <[email protected]>, Andrew Morton <[email protected]>, Vlastimil Babka <[email protected]>, "[email protected]" <[email protected]>, "[email protected]" <[email protected]>
Subject: Re: [PATCH] slub: choose the right freelist pointer location when creating small caches

On Tue, Jun 08, 2021 at 07:06:45PM +0000, Lin, Zhenpeng wrote:
> No problem. Just took a look and tested the patch, it looks good to me!

Great; thank you! Sorry I dropped the ball on this series. I got distracted. :) It looks like akpm took it into -mm now, so this should be fixed in -next soon.

--
Kees Cook

#7 lore raw
Quoted 1 line · Zhenpeng

Sounds good. But I would suggest this to go to -stable as soon as possible. Because this bug is affecting the basic functionality of DCCP. It crashes kernel whenever a new socket in this module is created.

But only when redzoning is enabled, yes?

Signature · 1 line
Kees Cook
#8 lore raw

Yes, that should be enabled to trigger.

-----Original Message-----
From: Kees Cook <[email protected]>
Date: Thursday, June 10, 2021 at 7:01 PM
To: "Lin, Zhenpeng" <[email protected]>
Cc: Christoph Lameter <[email protected]>, Pekka Enberg <[email protected]>, David Rientjes <[email protected]>, Joonsoo Kim <[email protected]>, Andrew Morton <[email protected]>, Vlastimil Babka <[email protected]>, "[email protected]" <[email protected]>, "[email protected]" <[email protected]>
Subject: Re: [PATCH] slub: choose the right freelist pointer location when creating small caches

On Thu, Jun 10, 2021 at 08:20:31PM +0000, Lin, Zhenpeng wrote:
> Sounds good. But I would suggest this to go to -stable as soon as possible. Because this bug is affecting the basic functionality of DCCP. It crashes kernel whenever a new socket in this module is created.

But only when redzoning is enabled, yes?

--
Kees Cook