[Devel] [RFC PATCH vz9 v6 42/62] dm-ploop: properly end pios attached to piwb pios

[Alexander Atanasov]

On 17.01.25 10:23, Pavel Tikhomirov wrote:
> Seems like an excess patch (leads to double ploop_pio_endio), as the 
> only caller of ploop_data_pio_end already does:


Agree, dropped

> 
>          if (pio->is_data_alloc) {
>                  completed = ploop_data_pio_end(pio);
>                  if (!completed)
>                          return;
>          }
> 
>          ploop_pio_endio(pio);
> 
> On 12/6/24 05:56, Alexander Atanasov wrote:
>> Fix a bug in ploop_data_pio_end.
>>
>> If piwb is already completed, which can happen if we attach
>> a flush pio to it, pio is not added to the list and ploop_pio_endio
>> is never called on it. This results in a hang since pio is never 
>> completed.
>>
>> Avoid this by calling ploop_pio_endio directly in case piwb is
>> already completed.
>>
>> https://virtuozzo.atlassian.net/browse/VSTOR-91821
>> Signed-off-by: Alexander Atanasov <alexander.atanasov at virtuozzo.com>
>> ---
>>   drivers/md/dm-ploop-map.c | 3 +++
>>   1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/md/dm-ploop-map.c b/drivers/md/dm-ploop-map.c
>> index 6192ddca3693..276e8cf3b178 100644
>> --- a/drivers/md/dm-ploop-map.c
>> +++ b/drivers/md/dm-ploop-map.c
>> @@ -1204,6 +1204,9 @@ static bool ploop_data_pio_end(struct pio *pio)
>>       else if (!pio->bi_status)
>>           pio->bi_status = piwb->bi_status;
>>       spin_unlock_irqrestore(&piwb->lock, flags);
>> +    /* If pio is late then end it here. this can happen with flushes */
>> +    if (completed)
>> +        ploop_pio_endio(pio);
>>       ploop_put_piwb(piwb);
> 

-- 
Regards,
Alexander Atanasov