Re: [PATCH v2 1/4] mm/migration: reduce the rcu lock duration

[Miaohe Lin <linmiaohe@huawei.com>]

On 2022/5/6 11:23, ying.huang@intel.com wrote:
> On Mon, 2022-04-25 at 21:27 +0800, Miaohe Lin wrote:
>> rcu_read_lock is required by grabbing the task refcount but it's not
>> needed for ptrace_may_access. So we could release the rcu lock after
>> task refcount is successfully grabbed to reduce the rcu holding time.
>>
>> Reviewed-by: Muchun Song <songmuchun@bytedance.com>
>> Reviewed-by: Christoph Hellwig <hch@lst.de>
>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>> Cc: Huang Ying <ying.huang@intel.com>
>> Cc: David Howells <dhowells@redhat.com>
>> Cc: Christoph Lameter <cl@linux.com>
>> ---
>>  mm/migrate.c | 3 +--
>>  1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/mm/migrate.c b/mm/migrate.c
>> index b2678279eb43..b779646665fe 100644
>> --- a/mm/migrate.c
>> +++ b/mm/migrate.c
>> @@ -1902,17 +1902,16 @@ static struct mm_struct *find_mm_struct(pid_t pid, nodemask_t *mem_nodes)
>>  		return ERR_PTR(-ESRCH);
>>  	}
>>  	get_task_struct(task);
>> +	rcu_read_unlock();
>>  
>>
>>  	/*
>>  	 * Check if this process has the right to modify the specified
>>  	 * process. Use the regular "ptrace_may_access()" checks.
>>  	 */
>>  	if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
>> -		rcu_read_unlock();
>>  		mm = ERR_PTR(-EPERM);
>>  		goto out;
>>  	}
>> -	rcu_read_unlock();
>>  
>>
>>  	mm = ERR_PTR(security_task_movememory(task));
>>  	if (IS_ERR(mm))
> 
> Hi, Miaohe,
> 
> Please check the previous discussion and verify whether the original
> reported race condition is stll valid by yourself before resending this
> patch again.  If you find that the original race condition isn't
> possible now, please add the analysis in your change log.
> 

Sorry for late respond. It's a pity that this change is still not verified by the relevant
experts. I will try to give my analysis in my change log instead if the original race condition
is invalid now.

Many thanks!

> Best Regards,
> Huang, Ying
> 
> .
>