[Ocfs2-devel] [PATCH 1/2 v3] ocfs2/dlm: fix race between convert and recovery

[Joseph Qi <joseph.qi@huawei.com>]

Hi Junxiao,
Do you have any other questions about this solution?

On 2015/9/24 14:13, Joseph Qi wrote:
> On 2015/9/24 12:21, Junxiao Bi wrote:
>> On 09/24/2015 12:11 PM, Joseph Qi wrote:
>>> Hi Junxiao,
>>>
>>> On 2015/9/24 11:58, Junxiao Bi wrote:
>>>> On 09/24/2015 11:14 AM, Joseph Qi wrote:
>>>>> There is a race window between dlmconvert_remote and
>>>>> dlm_move_lockres_to_recovery_list, which will cause a lock with
>>>>> OCFS2_LOCK_BUSY in grant list, thus system hangs.
>>>>>
>>>>> dlmconvert_remote
>>>>> {
>>>>>         spin_lock(&res->spinlock);
>>>>>         list_move_tail(&lock->list, &res->converting);
>>>>>         lock->convert_pending = 1;
>>>>>         spin_unlock(&res->spinlock);
>>>>>
>>>>>         status = dlm_send_remote_convert_request();
>>>>>         >>>>>> race window, master has queued ast and return DLM_NORMAL,
>>>>>                and then down before sending ast.
>>>>>                this node detects master down and calls
>>>>>                dlm_move_lockres_to_recovery_list, which will revert the
>>>>>                lock to grant list.
>>>>>                Then OCFS2_LOCK_BUSY won't be cleared as new master won't
>>>>>                send ast any more because it thinks already be authorized.
>>>> How this race windowed fixed?
>>>> the process have sent convert request to master node successfully(return
>>>> value DLM_NORMAL) then wait on LOCK_BUSY, then when master node panic
>>>> before send out ast, dlm_move_lockres_to_recovery_list() move the lock
>>>> to grant list. Ast never come.
>>> res->state is now having DLM_LOCK_RES_RECOVERING set. 
>> But what happened if master node panic after
>> dlm_send_remote_convert_request() return DLM_NORMAL and wait on LOCK_BUSY?
>>
> Only two cases that it will call dlm_revert_pending_convert:
> 1) request returns other than DLM_NORMAL, current code will retry;
> 2) request returns DLM_NORMAL and recovery does it, this is the case we
> discuss. Consider the lockres spinlock has to be taken when doing this,
> only dlm_move_lockres_to_recovery_list runs first before request return
> code handling can we have the race window. If convert thread runs first,
> it's still in convert list, and recovery won't move it to grant list
> because convert_pending is cleared. And the new master will send ast
> after recovery.
> 
> Thanks,
> Joseph
> 
>> Thanks,
>> Junxiao.
>>
>> This patch will
>>> reset the status to DLM_RECOVERING and then retry convert request. Then
>>> new master will handle it and send ast.
>>>
>>>>
>>>> Thanks,
>>>> Junxiao.
>>>>
>>>>>
>>>>>         spin_lock(&res->spinlock);
>>>>>         lock->convert_pending = 0;
>>>>>         if (status != DLM_NORMAL)
>>>>>                 dlm_revert_pending_convert(res, lock);
>>>>>         spin_unlock(&res->spinlock);
>>>>> }
>>>>>
>>>>> In this case, check if res->state has DLM_LOCK_RES_RECOVERING bit set
>>>>> (res is still in recovering) or res master changed (new master has
>>>>> finished recovery), reset the status to DLM_RECOVERING, then it will
>>>>> retry convert.
>>>>>
>>>>> Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
>>>>> Reported-by: Yiwen Jiang <jiangyiwen@huawei.com>
>>>>> Cc: <stable@vger.kernel.org>
>>>>> ---
>>>>>  fs/ocfs2/dlm/dlmconvert.c | 11 ++++++++++-
>>>>>  1 file changed, 10 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
>>>>> index e36d63f..9e6116e 100644
>>>>> --- a/fs/ocfs2/dlm/dlmconvert.c
>>>>> +++ b/fs/ocfs2/dlm/dlmconvert.c
>>>>> @@ -262,6 +262,7 @@ enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
>>>>>  				  struct dlm_lock *lock, int flags, int type)
>>>>>  {
>>>>>  	enum dlm_status status;
>>>>> +	u8 old_owner = res->owner;
>>>>>
>>>>>  	mlog(0, "type=%d, convert_type=%d, busy=%d\n", lock->ml.type,
>>>>>  	     lock->ml.convert_type, res->state & DLM_LOCK_RES_IN_PROGRESS);
>>>>> @@ -316,11 +317,19 @@ enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
>>>>>  	spin_lock(&res->spinlock);
>>>>>  	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
>>>>>  	lock->convert_pending = 0;
>>>>> -	/* if it failed, move it back to granted queue */
>>>>> +	/* if it failed, move it back to granted queue.
>>>>> +	 * if master returns DLM_NORMAL and then down before sending ast,
>>>>> +	 * it may have already been moved to granted queue, reset to
>>>>> +	 * DLM_RECOVERING and retry convert */
>>>>>  	if (status != DLM_NORMAL) {
>>>>>  		if (status != DLM_NOTQUEUED)
>>>>>  			dlm_error(status);
>>>>>  		dlm_revert_pending_convert(res, lock);
>>>>> +	} else if ((res->state & DLM_LOCK_RES_RECOVERING) ||
>>>>> +			(old_owner != res->owner)) {
>>>>> +		mlog(0, "res %.*s is in recovering or has been recovered.\n",
>>>>> +				res->lockname.len, res->lockname.name);
>>>>> +		status = DLM_RECOVERING;
>>>>>  	}
>>>>>  bail:
>>>>>  	spin_unlock(&res->spinlock);
>>>>>
>>>>
>>>>
>>>> .
>>>>
>>>
>>>
>>
>>
>> .
>>
> 
> 
> 
> _______________________________________________
> Ocfs2-devel mailing list
> Ocfs2-devel at oss.oracle.com
> https://oss.oracle.com/mailman/listinfo/ocfs2-devel
> 
>