Re: [PATCH] cgroup: fix reverse unlock sequence in cgroup_get_sb

[Danny Feng <dfeng@redhat.com>]

On 07/22/2009 08:53 AM, Li Zefan wrote:
>>>> Seems reasonable to me. You might also want to mention that elsewhere
>>>> the sequence is unlock cgroup_mutex followed by inode->i_mutex.
>>>>
>>>> Acked-by: Balbir Singh balbir@linux.vnet.ibm.com
>>>>          
>>> No, the unlock order is irrelevant. It's the lock order that matters. So
>>> this patch
>>> fixes nothing.
>>>
>>> Xiaotian, you didn't run into deadlock, did you?
>>>
>>>        
>> Li, Consider the following
>>
>>
>> lock(A)
>> lock(B)
>> unlock(A)
>> unlock(B)
>>
>> Tomorrow if a unsuspecting programmer does this
>>
>> lock(A)
>> lock(B)
>> unlock(A)
>>
>> code block
>>
>> unlock(B)
>>
>>
>> What protects code block? lock B? Is that the intention?
>>
>>      
>
> I won't worry about that. If unlock order is a concern,
> we should have taught lockdep to detect it.
>
> Here's a reply from Linus on this issue:
>
> 	http://lkml.org/lkml/2008/10/8/150
>    
OK, this patch is trivial. Just for consistency with previous unlock 
sequence:-)