[v1] mm: check the function kmalloc_slab return value

[PATCH] mm: check the function kmalloc_slab return value

Posted by Ren Yu 3 years, 10 months ago

As the possible failure of the kmalloc_slab,
it should be better to check it.

Signed-off-by: Ren Yu <renyu@nfschina.com>
Reported-by: kernel test robot <lkp@intel.com>
---
v2:
- fix build waring integer from pointer without a cast
---
---
 mm/slab.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/mm/slab.c b/mm/slab.c
index f8cd00f4ba13..72135e555827 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -2064,6 +2064,8 @@ int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
 	if (OFF_SLAB(cachep)) {
 		cachep->freelist_cache =
 			kmalloc_slab(cachep->freelist_size, 0u);
+		if (unlikely(ZERO_OR_NULL_PTR(cachep->freelist_cache)))
+			return cachep->freelist_cache;
 	}
 
 	err = setup_cpu_cache(cachep, gfp);
-- 
2.11.0

Re: [PATCH] mm: check the function kmalloc_slab return value

Posted by Vlastimil Babka 3 years, 10 months ago

On 6/14/22 10:39, Ren Yu wrote:
> As the possible failure of the kmalloc_slab,
> it should be better to check it.

AFAIK failure is not possible, kmalloc_slab() is not an allocation function,
it just returns a member of kmalloc_caches array, which is initialized
elsewhere and shouldn't contain NULLs. So the patch seems unnecessary to me.

> Signed-off-by: Ren Yu <renyu@nfschina.com>
> Reported-by: kernel test robot <lkp@intel.com>
> ---
> v2:
> - fix build waring integer from pointer without a cast
> ---
> ---
>  mm/slab.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/mm/slab.c b/mm/slab.c
> index f8cd00f4ba13..72135e555827 100644
> --- a/mm/slab.c
> +++ b/mm/slab.c
> @@ -2064,6 +2064,8 @@ int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
>  	if (OFF_SLAB(cachep)) {
>  		cachep->freelist_cache =
>  			kmalloc_slab(cachep->freelist_size, 0u);
> +		if (unlikely(ZERO_OR_NULL_PTR(cachep->freelist_cache)))

The usual way is "if (!cachep->freelist_cache)". Not sure why check for ZERO.

> +			return cachep->freelist_cache;

So in case of NULL this would return NULL, thus 0, but __kmem_cache_create()
return 0 on success, so it's wrong. You would have to return e.g. -ENOMEM.

>  	}
>  
>  	err = setup_cpu_cache(cachep, gfp);

Re: [PATCH] mm: check the function kmalloc_slab return value

Posted by tury 3 years, 10 months ago


在 2022年06月14日 16:48, Vlastimil Babka 写道:
> On 6/14/22 10:39, Ren Yu wrote:
>> As the possible failure of the kmalloc_slab,
>> it should be better to check it.
> AFAIK failure is not possible, kmalloc_slab() is not an allocation function,
> it just returns a member of kmalloc_caches array, which is initialized
> elsewhere and shouldn't contain NULLs. So the patch seems unnecessary to me.
>
>> Signed-off-by: Ren Yu <renyu@nfschina.com>
>> Reported-by: kernel test robot <lkp@intel.com>
>> ---
>> v2:
>> - fix build waring integer from pointer without a cast
>> ---
>> ---
>>   mm/slab.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/mm/slab.c b/mm/slab.c
>> index f8cd00f4ba13..72135e555827 100644
>> --- a/mm/slab.c
>> +++ b/mm/slab.c
>> @@ -2064,6 +2064,8 @@ int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
>>   	if (OFF_SLAB(cachep)) {
>>   		cachep->freelist_cache =
>>   			kmalloc_slab(cachep->freelist_size, 0u);
>> +		if (unlikely(ZERO_OR_NULL_PTR(cachep->freelist_cache)))
> The usual way is "if (!cachep->freelist_cache)". Not sure why check for ZERO.
>
>> +			return cachep->freelist_cache;
> So in case of NULL this would return NULL, thus 0, but __kmem_cache_create()
> return 0 on success, so it's wrong. You would have to return e.g. -ENOMEM.
Thanks for the advice ,I'll be re-patching
>
>>   	}
>>   
>>   	err = setup_cpu_cache(cachep, gfp);
>

Re: [PATCH] mm: check the function kmalloc_slab return value

Posted by Vlastimil Babka 3 years, 10 months ago

On 6/14/22 11:26, tury wrote:
> 
> 
> 在 2022年06月14日 16:48, Vlastimil Babka 写道:
>> On 6/14/22 10:39, Ren Yu wrote:
>>> As the possible failure of the kmalloc_slab,
>>> it should be better to check it.
>> AFAIK failure is not possible, kmalloc_slab() is not an allocation function,
>> it just returns a member of kmalloc_caches array, which is initialized
>> elsewhere and shouldn't contain NULLs. So the patch seems unnecessary to me.
>>
>>> Signed-off-by: Ren Yu <renyu@nfschina.com>
>>> Reported-by: kernel test robot <lkp@intel.com>
>>> ---
>>> v2:
>>> - fix build waring integer from pointer without a cast
>>> ---
>>> ---
>>>   mm/slab.c | 2 ++
>>>   1 file changed, 2 insertions(+)
>>>
>>> diff --git a/mm/slab.c b/mm/slab.c
>>> index f8cd00f4ba13..72135e555827 100644
>>> --- a/mm/slab.c
>>> +++ b/mm/slab.c
>>> @@ -2064,6 +2064,8 @@ int __kmem_cache_create(struct kmem_cache *cachep,
>>> slab_flags_t flags)
>>>       if (OFF_SLAB(cachep)) {
>>>           cachep->freelist_cache =
>>>               kmalloc_slab(cachep->freelist_size, 0u);
>>> +        if (unlikely(ZERO_OR_NULL_PTR(cachep->freelist_cache)))
>> The usual way is "if (!cachep->freelist_cache)". Not sure why check for ZERO.
>>
>>> +            return cachep->freelist_cache;
>> So in case of NULL this would return NULL, thus 0, but __kmem_cache_create()
>> return 0 on success, so it's wrong. You would have to return e.g. -ENOMEM.
> Thanks for the advice ,I'll be re-patching

However that was meant just for your information/learning, the patch is
still unecessary as I wrote above, so I will not merge it so we don't
complicate the code needlessly.

>>
>>>       }
>>>         err = setup_cpu_cache(cachep, gfp);
>>
>