[PATCH] drivers/base/memory: Fix a reference counting issue in __add_memory_block()

[PATCH] drivers/base/memory: Fix a reference counting issue in __add_memory_block()

[Christophe JAILLET]

There is no point in doing put_device()/device_unregister() on a
device that has just been registered a few lines above. This will lead to
a double reference decrement.

I guess that this put_device()/device_unregister() is a cut'n'paste from
remove_memory_block() (i.e. unregister_memory() at the time being) which
does need it.

Fixes: 4fb6eabf1037 ("drivers/base/memory.c: cache memory blocks in xarray to accelerate lookup")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
---
 drivers/base/memory.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/base/memory.c b/drivers/base/memory.c
index 7222ff9b5e05..084d67fd55cc 100644
--- a/drivers/base/memory.c
+++ b/drivers/base/memory.c
@@ -636,10 +636,9 @@ static int __add_memory_block(struct memory_block *memory)
 	}
 	ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
 			      GFP_KERNEL));
-	if (ret) {
-		put_device(&memory->dev);
+	if (ret)
 		device_unregister(&memory->dev);
-	}
+
 	return ret;
 }
 
-- 
2.32.0

Re: [PATCH] drivers/base/memory: Fix a reference counting issue in __add_memory_block()

[Michal Hocko]

On Fri 22-04-22 09:15:21, Christophe JAILLET wrote:
> There is no point in doing put_device()/device_unregister() on a
> device that has just been registered a few lines above. This will lead to
> a double reference decrement.

This is a bit confusing. I would rephrase.
"
__add_memory_block calls both put_device and device_unregister when
storing the memory block into the xarray. This is incorrect because
xarray doesn't take an additional reference and device_unregister
already calls put_device.
"

Btw. I do not think this failure path can be triggered, or is there a
way to hit it?
 
> I guess that this put_device()/device_unregister() is a cut'n'paste from
> remove_memory_block() (i.e. unregister_memory() at the time being) which
> does need it.
> 
> Fixes: 4fb6eabf1037 ("drivers/base/memory.c: cache memory blocks in xarray to accelerate lookup")
> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>

Other than that looks good to me. With the changelog clarified,
especially the part that evaluates whether this is a real or a
theoretical problem, feel free to add
Acked-by: Michal Hocko <mhocko@suse.com>

Thanks!

> ---
>  drivers/base/memory.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
> index 7222ff9b5e05..084d67fd55cc 100644
> --- a/drivers/base/memory.c
> +++ b/drivers/base/memory.c
> @@ -636,10 +636,9 @@ static int __add_memory_block(struct memory_block *memory)
>  	}
>  	ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
>  			      GFP_KERNEL));
> -	if (ret) {
> -		put_device(&memory->dev);
> +	if (ret)
>  		device_unregister(&memory->dev);
> -	}
> +
>  	return ret;
>  }
>  
> -- 
> 2.32.0

-- 
Michal Hocko
SUSE Labs

Re: [PATCH] drivers/base/memory: Fix a reference counting issue in __add_memory_block()

[Christophe JAILLET]

Le 22/04/2022 à 09:34, Michal Hocko a écrit :
> On Fri 22-04-22 09:15:21, Christophe JAILLET wrote:
>> There is no point in doing put_device()/device_unregister() on a
>> device that has just been registered a few lines above. This will lead to
>> a double reference decrement.
> 
> This is a bit confusing. I would rephrase.
> "
> __add_memory_block calls both put_device and device_unregister when
> storing the memory block into the xarray. This is incorrect because
> xarray doesn't take an additional reference and device_unregister
> already calls put_device.
> "
> 
> Btw. I do not think this failure path can be triggered, or is there a
> way to hit it?
>   
>> I guess that this put_device()/device_unregister() is a cut'n'paste from
>> remove_memory_block() (i.e. unregister_memory() at the time being) which
>> does need it.
>>
>> Fixes: 4fb6eabf1037 ("drivers/base/memory.c: cache memory blocks in xarray to accelerate lookup")
>> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> 
> Other than that looks good to me. With the changelog clarified,
> especially the part that evaluates whether this is a real or a
> theoretical problem, feel free to add
> Acked-by: Michal Hocko <mhocko@suse.com>

I'll send a v2 with the updated changelog.

I do agree that this is certainly a theoretical issue.
Moreover, should it be triggered, I think that it would only print a 
warning message about an erroneous decrement on a ref counted resource 
that is already 0.

Well, it also saves 2 LoC and will look more logical to other readers 
(and static checkers :) )

Thanks for the quick review for both of you.

CJ
> 
> Thanks!
> 
>> ---
>>   drivers/base/memory.c | 5 ++---
>>   1 file changed, 2 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
>> index 7222ff9b5e05..084d67fd55cc 100644
>> --- a/drivers/base/memory.c
>> +++ b/drivers/base/memory.c
>> @@ -636,10 +636,9 @@ static int __add_memory_block(struct memory_block *memory)
>>   	}
>>   	ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
>>   			      GFP_KERNEL));
>> -	if (ret) {
>> -		put_device(&memory->dev);
>> +	if (ret)
>>   		device_unregister(&memory->dev);
>> -	}
>> +
>>   	return ret;
>>   }
>>   
>> -- 
>> 2.32.0
> 

Re: [PATCH] drivers/base/memory: Fix a reference counting issue in __add_memory_block()

[David Hildenbrand]

On 22.04.22 09:15, Christophe JAILLET wrote:
> There is no point in doing put_device()/device_unregister() on a
> device that has just been registered a few lines above. This will lead to
> a double reference decrement.
> 
> I guess that this put_device()/device_unregister() is a cut'n'paste from
> remove_memory_block() (i.e. unregister_memory() at the time being) which
> does need it.
> 
> Fixes: 4fb6eabf1037 ("drivers/base/memory.c: cache memory blocks in xarray to accelerate lookup")
> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> ---
>  drivers/base/memory.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
> index 7222ff9b5e05..084d67fd55cc 100644
> --- a/drivers/base/memory.c
> +++ b/drivers/base/memory.c
> @@ -636,10 +636,9 @@ static int __add_memory_block(struct memory_block *memory)
>  	}
>  	ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
>  			      GFP_KERNEL));
> -	if (ret) {
> -		put_device(&memory->dev);
> +	if (ret)
>  		device_unregister(&memory->dev);
> -	}
> +
>  	return ret;
>  }
>  

Looks correct to me. I assume this will only happen once the stars
align, but it's most certainly wrong I think.

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 
Thanks,

David / dhildenb