[v3] mm: slub: avoid deref of free pointer in sanity checks if object is invalid

[PATCH v3] mm: slub: avoid deref of free pointer in sanity checks if object is invalid

Posted by Li Qiong 2 months, 1 week ago

For debugging, object_err() prints free pointer of the object.
However, if check_valid_pointer() returns false for a object,
dereferncing `object + s->offset` can lead to a crash. Therefore,
print the object's address in such cases.

Fixes: bb192ed9aa71 ("mm/slub: Convert most struct page to struct slab by spatch")
Cc: <stable@vger.kernel.org>
Signed-off-by: Li Qiong <liqiong@nfschina.com>
---
v2:
- rephrase the commit message, add comment for object_err().
v3:
- check object pointer in object_err().
---
 mm/slub.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index 31e11ef256f9..d3abae5a2193 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1104,7 +1104,11 @@ static void object_err(struct kmem_cache *s, struct slab *slab,
 		return;
 
 	slab_bug(s, reason);
-	print_trailer(s, slab, object);
+	if (!check_valid_pointer(s, slab, object)) {
+		print_slab_info(slab);
+		pr_err("invalid object 0x%p\n", object);
+	} else
+		print_trailer(s, slab, object);
 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 
 	WARN_ON(1);
@@ -1587,7 +1591,7 @@ static inline int alloc_consistency_checks(struct kmem_cache *s,
 		return 0;
 
 	if (!check_valid_pointer(s, slab, object)) {
-		object_err(s, slab, object, "Freelist Pointer check fails");
+		slab_err(s, slab, "Freelist Pointer(0x%p) check fails", object);
 		return 0;
 	}
 
-- 
2.30.2

Re: [PATCH v3] mm: slub: avoid deref of free pointer in sanity checks if object is invalid

Posted by Harry Yoo 2 months, 1 week ago

On Tue, Jul 29, 2025 at 04:14:55PM +0800, Li Qiong wrote:
> For debugging, object_err() prints free pointer of the object.
> However, if check_valid_pointer() returns false for a object,
> dereferncing `object + s->offset` can lead to a crash. Therefore,
> print the object's address in such cases.

As the code changed a bit, I think the commit message could better reflect
what this patch actually does.

> Fixes: bb192ed9aa71 ("mm/slub: Convert most struct page to struct slab by spatch")

As Vlastimil mentioned in previous version, this is not the first commit
that introduced this problem.

> Cc: <stable@vger.kernel.org>
> Signed-off-by: Li Qiong <liqiong@nfschina.com>
> ---
> v2:
> - rephrase the commit message, add comment for object_err().
> v3:
> - check object pointer in object_err().
> ---
>  mm/slub.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 31e11ef256f9..d3abae5a2193 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -1104,7 +1104,11 @@ static void object_err(struct kmem_cache *s, struct slab *slab,
>  		return;
>  
>  	slab_bug(s, reason);
> -	print_trailer(s, slab, object);
> +	if (!check_valid_pointer(s, slab, object)) {
> +		print_slab_info(slab);
> +		pr_err("invalid object 0x%p\n", object);

Can we just handle this inside print_trailer() because that's the function
that prints the object's free pointer, metadata, etc.?

Also, the message should start with a capital letter.

and "invalid object" sounds misleading because it's the pointer
that is invalid. Perhaps simply "Invalid pointer 0x%p\n"?
(What would be the most comprehensive message here? :P)

> +	} else
> +		print_trailer(s, slab, object);

>  	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
>  
>  	WARN_ON(1);
> @@ -1587,7 +1591,7 @@ static inline int alloc_consistency_checks(struct kmem_cache *s,
>  		return 0;
>  
>  	if (!check_valid_pointer(s, slab, object)) {
> -		object_err(s, slab, object, "Freelist Pointer check fails");
> +		slab_err(s, slab, "Freelist Pointer(0x%p) check fails", object);
>  		return 0;

Do we really need this hunk after making object_err() resiliant
against wild pointers?

>  	}

-- 
Cheers,
Harry / Hyeonggon

Re: [PATCH v3] mm: slub: avoid deref of free pointer in sanity checks if object is invalid

Posted by liqiong 2 months, 1 week ago


在 2025/7/29 21:41, Harry Yoo 写道:
> On Tue, Jul 29, 2025 at 04:14:55PM +0800, Li Qiong wrote:
>> For debugging, object_err() prints free pointer of the object.
>> However, if check_valid_pointer() returns false for a object,
>> dereferncing `object + s->offset` can lead to a crash. Therefore,
>> print the object's address in such cases.
> As the code changed a bit, I think the commit message could better reflect
> what this patch actually does.
Yes.

>
>> Fixes: bb192ed9aa71 ("mm/slub: Convert most struct page to struct slab by spatch")
> As Vlastimil mentioned in previous version, this is not the first commit
> that introduced this problem.
>
>> Cc: <stable@vger.kernel.org>
>> Signed-off-by: Li Qiong <liqiong@nfschina.com>
>> ---
>> v2:
>> - rephrase the commit message, add comment for object_err().
>> v3:
>> - check object pointer in object_err().
>> ---
>>  mm/slub.c | 8 ++++++--
>>  1 file changed, 6 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/slub.c b/mm/slub.c
>> index 31e11ef256f9..d3abae5a2193 100644
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -1104,7 +1104,11 @@ static void object_err(struct kmem_cache *s, struct slab *slab,
>>  		return;
>>  
>>  	slab_bug(s, reason);
>> -	print_trailer(s, slab, object);
>> +	if (!check_valid_pointer(s, slab, object)) {
>> +		print_slab_info(slab);
>> +		pr_err("invalid object 0x%p\n", object);
> Can we just handle this inside print_trailer() because that's the function
> that prints the object's free pointer, metadata, etc.?
Maybe it's clearer ,  if  object pointer being invalid, don't enter print_trailer()，
print_trailer() prints  valid object.

>
> Also, the message should start with a capital letter.
>
> and "invalid object" sounds misleading because it's the pointer
> that is invalid. Perhaps simply "Invalid pointer 0x%p\n"?
> (What would be the most comprehensive message here? :P)

Make sense,   will change it.
>
>> +	} else
>> +		print_trailer(s, slab, object);
>>  	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
>>  
>>  	WARN_ON(1);
>> @@ -1587,7 +1591,7 @@ static inline int alloc_consistency_checks(struct kmem_cache *s,
>>  		return 0;
>>  
>>  	if (!check_valid_pointer(s, slab, object)) {
>> -		object_err(s, slab, object, "Freelist Pointer check fails");
>> +		slab_err(s, slab, "Freelist Pointer(0x%p) check fails", object);
>>  		return 0;
> Do we really need this hunk after making object_err() resiliant
> against wild pointers?

That's the origin issue,   it may be  inappropriate to use object_err(), if check_valid_pointer being false.

>
>>  	}

Re: [PATCH v3] mm: slub: avoid deref of free pointer in sanity checks if object is invalid

Posted by Harry Yoo 2 months, 1 week ago

On Wed, Jul 30, 2025 at 09:46:09AM +0800, liqiong wrote:
> 在 2025/7/29 21:41, Harry Yoo 写道:
> > On Tue, Jul 29, 2025 at 04:14:55PM +0800, Li Qiong wrote:
> >> Fixes: bb192ed9aa71 ("mm/slub: Convert most struct page to struct slab by spatch")
> > As Vlastimil mentioned in previous version, this is not the first commit
> > that introduced this problem.

Please don't forget to update Fixes: tag :)

> >> Cc: <stable@vger.kernel.org>
> >> Signed-off-by: Li Qiong <liqiong@nfschina.com>
> >> ---
> >> v2:
> >> - rephrase the commit message, add comment for object_err().
> >> v3:
> >> - check object pointer in object_err().
> >> ---
> >>  mm/slub.c | 8 ++++++--
> >>  1 file changed, 6 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/mm/slub.c b/mm/slub.c
> >> index 31e11ef256f9..d3abae5a2193 100644
> >> --- a/mm/slub.c
> >> +++ b/mm/slub.c
> >> @@ -1104,7 +1104,11 @@ static void object_err(struct kmem_cache *s, struct slab *slab,
> >>  		return;
> >>  
> >>  	slab_bug(s, reason);
> >> -	print_trailer(s, slab, object);
> >> +	if (!check_valid_pointer(s, slab, object)) {
> >> +		print_slab_info(slab);
> >> +		pr_err("invalid object 0x%p\n", object);
> > Can we just handle this inside print_trailer() because that's the function
> > that prints the object's free pointer, metadata, etc.?

> Maybe it's clearer ,  if  object pointer being invalid, don't enter print_trailer()，
> print_trailer() prints  valid object.

You're probably right. No strong opinion.
object_err() is the only user anyway.

> >> +	} else
> >> +		print_trailer(s, slab, object);
> >>  	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
> >>  
> >>  	WARN_ON(1);
> >> @@ -1587,7 +1591,7 @@ static inline int alloc_consistency_checks(struct kmem_cache *s,
> >>  		return 0;
> >>  
> >>  	if (!check_valid_pointer(s, slab, object)) {
> >> -		object_err(s, slab, object, "Freelist Pointer check fails");
> >> +		slab_err(s, slab, "Freelist Pointer(0x%p) check fails", object);
> >>  		return 0;
> > Do we really need this hunk after making object_err() resiliant
> > against wild pointers?
> 
> That's the origin issue,   it may be  inappropriate to use object_err(), if check_valid_pointer being false.

That was the original issue, but you're making it not crash even if
with bad pointers are passed?

> >>  	}
> 

-- 
Cheers,
Harry / Hyeonggon

Re: [PATCH v3] mm: slub: avoid deref of free pointer in sanity checks if object is invalid

Posted by liqiong 2 months, 1 week ago


在 2025/7/30 13:04, Harry Yoo 写道:
> On Wed, Jul 30, 2025 at 09:46:09AM +0800, liqiong wrote:
>> 在 2025/7/29 21:41, Harry Yoo 写道:
>>> On Tue, Jul 29, 2025 at 04:14:55PM +0800, Li Qiong wrote:
>>>> Fixes: bb192ed9aa71 ("mm/slub: Convert most struct page to struct slab by spatch")
>>> As Vlastimil mentioned in previous version, this is not the first commit
>>> that introduced this problem.
> Please don't forget to update Fixes: tag :)

It seems that it's the first commit:    Fixes: 81819f0fc828 ("SLUB core"  )


>
>>>> Cc: <stable@vger.kernel.org>
>>>> Signed-off-by: Li Qiong <liqiong@nfschina.com>
>>>> ---
>>>> v2:
>>>> - rephrase the commit message, add comment for object_err().
>>>> v3:
>>>> - check object pointer in object_err().
>>>> ---
>>>>  mm/slub.c | 8 ++++++--
>>>>  1 file changed, 6 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/mm/slub.c b/mm/slub.c
>>>> index 31e11ef256f9..d3abae5a2193 100644
>>>> --- a/mm/slub.c
>>>> +++ b/mm/slub.c
>>>> @@ -1104,7 +1104,11 @@ static void object_err(struct kmem_cache *s, struct slab *slab,
>>>>  		return;
>>>>  
>>>>  	slab_bug(s, reason);
>>>> -	print_trailer(s, slab, object);
>>>> +	if (!check_valid_pointer(s, slab, object)) {
>>>> +		print_slab_info(slab);
>>>> +		pr_err("invalid object 0x%p\n", object);
>>> Can we just handle this inside print_trailer() because that's the function
>>> that prints the object's free pointer, metadata, etc.?
>> Maybe it's clearer ,  if  object pointer being invalid, don't enter print_trailer()，
>> print_trailer() prints  valid object.
> You're probably right. No strong opinion.
> object_err() is the only user anyway.
>
>>>> +	} else
>>>> +		print_trailer(s, slab, object);
>>>>  	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
>>>>  
>>>>  	WARN_ON(1);
>>>> @@ -1587,7 +1591,7 @@ static inline int alloc_consistency_checks(struct kmem_cache *s,
>>>>  		return 0;
>>>>  
>>>>  	if (!check_valid_pointer(s, slab, object)) {
>>>> -		object_err(s, slab, object, "Freelist Pointer check fails");
>>>> +		slab_err(s, slab, "Freelist Pointer(0x%p) check fails", object);
>>>>  		return 0;
>>> Do we really need this hunk after making object_err() resiliant
>>> against wild pointers?
>> That's the origin issue,   it may be  inappropriate to use object_err(), if check_valid_pointer being false.
> That was the original issue, but you're making it not crash even if
> with bad pointers are passed?

Make sense, fix in object_err(),  it wouldn't  crash and print the message.

>
>>>>  	}