Re: [PATCH v2] slab: Fix obj_ext is mistakenly considered NULL due to race condition

[Hao Ge <hao.ge@linux.dev>]

Hi Harry


On 2025/10/24 16:54, Harry Yoo wrote:
> On Thu, Oct 23, 2025 at 10:33:13PM +0800, Hao Ge wrote:
>> From: Hao Ge <gehao@kylinos.cn>
>>
>> If two competing threads enter alloc_slab_obj_exts(), if the process
>> that allocates the vector wins cmpxchg(), and the other thread mistakenly
>> assume slab->obj_ext is still empty due to its own allocation failure.
> Massaging this a little bit:
>
>    "If two competing threads enter alloc_slab_obj_exts(), and the one that
>     allocates the vector wins the cmpxchg(), the other thread that failed
>     allocation mistakenly assumes that slab->obj_exts is still empty due to
>     its own allocation failure."
>
>> This
>> will then trigger warnings enforced by CONFIG_MEM_ALLOC_PROFILING_DEBUG
>> checks in the subsequent free path.
>>
>> Therefore, let's add an additional check when the process that allocates
>> the vector loses the cmpxchg()
> You mean "when the process that failed to allocate the vector loses the
> cmpxchg()"?

Yes, I apologize for not being clear enough in my description here.

>> Suggested-by: Harry Yoo <harry.yoo@oracle.com>
>> Signed-off-by: Hao Ge <gehao@kylinos.cn>
>> ---
>> v2: Revise the solution according to Harry's suggestion.
>>      Add Suggested-by: Harry Yoo <harry.yoo@oracle.com>
>> ---
>>
>>   mm/slub.c | 16 +++++++++++-----
>>   1 file changed, 11 insertions(+), 5 deletions(-)
>>
>> diff --git a/mm/slub.c b/mm/slub.c
>> index d4403341c9df..d7bfec6c0171 100644
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -2052,9 +2052,9 @@ static inline void mark_objexts_empty(struct slabobj_ext *obj_exts)
>>   	}
>>   }
>>   
>> -static inline void mark_failed_objexts_alloc(struct slab *slab)
>> +static inline bool mark_failed_objexts_alloc(struct slab *slab)
>>   {
>> -	cmpxchg(&slab->obj_exts, 0, OBJEXTS_ALLOC_FAIL);
>> +	return cmpxchg(&slab->obj_exts, 0, OBJEXTS_ALLOC_FAIL) == 0;
>>   }
>>   
>>   static inline void handle_failed_objexts_alloc(unsigned long obj_exts,
>> @@ -2076,7 +2076,7 @@ static inline void handle_failed_objexts_alloc(unsigned long obj_exts,
>>   #else /* CONFIG_MEM_ALLOC_PROFILING_DEBUG */
>>   
>>   static inline void mark_objexts_empty(struct slabobj_ext *obj_exts) {}
>> -static inline void mark_failed_objexts_alloc(struct slab *slab) {}
>> +static inline bool mark_failed_objexts_alloc(struct slab *slab) { return false; }
>>   static inline void handle_failed_objexts_alloc(unsigned long obj_exts,
>>   			struct slabobj_ext *vec, unsigned int objects) {}
>>   
>> @@ -2124,8 +2124,14 @@ int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
>>   				   slab_nid(slab));
>>   	}
>>   	if (!vec) {
>> -		/* Mark vectors which failed to allocate */
>> -		mark_failed_objexts_alloc(slab);
>> +		/*
>> +		 * Try to mark vectors which failed to allocate
> nit:
>                                                                 ^ missing
> 							       period
> 							       (.) here
>
> With the comments resolved,
> Reviewed-by: Harry Yoo <harry.yoo@oracle.com>


A period is indeed missing here.


Hi Vlastimil


Thank you for adding V2 to your tree. Now, should I resubmit V3,

or can you assist with making these modifications in your tree?


>
>> +		 * If this operation fails, there may be a racing process
>> +		 * that has already completed the allocation.
>> +		 */
>> +		if (!mark_failed_objexts_alloc(slab) &&
>> +		    slab_obj_exts(slab))
>> +			return 0;
>>   
>>   		return -ENOMEM;
>>   	}
>> -- 
>> 2.25.1