On 9/15/20 7:33 AM, Nix wrote:
> On 9 Sep 2020, Mariusz Tkaczyk uttered the following:
>
>> + /*
>> + * If mdmonitor is awaken by event, set small delay once
>> + * to deal with udev and mdadm.
>> + */
>> + if (wait_result != 0) {
>> + if (c->delay > 5)
>> + delay_for_event = 5;
>> + } else
>> + delay_for_event = c->delay;
>
> This is racy: if any delay is needed, any finite delay value will
> now and then be too short.
>
> I think this should be fixed by arranging for mdmonitor to be signalled
> when udev or whatever has finished whatever it's doing. (udev has lots
> of ways it could be asked to do that.)
>
Hi
I have applied this patch for now, since it is better than what we had.
However, I agree it would be better to do this using udev or other
signalling, so please have a look at that.
Cheers,
Jes
