Re: mdadm --detail works, mdadm --stop says "does not appear to be an md device"

[Mariusz Tkaczyk <mariusz.tkaczyk@linux.intel.com>]

On Mon, 20 Mar 2023 19:01:01 -0700
Marc MERLIN <marc@merlins.org> wrote:

> On Mon, Mar 20, 2023 at 04:16:59PM +0100, Mariusz Tkaczyk wrote:
> > On Mon, 20 Mar 2023 07:50:35 -0700
> > Marc MERLIN <marc@merlins.org> wrote:
> >   
> > > On Mon, Mar 20, 2023 at 03:36:39PM +0100, Mariusz Tkaczyk wrote:  
> > > > Hi,
> > > > mdadm is unable to complete this task because it cannot ensure that it
> > > > is safe to stop the array. It cannot open the array with O_EXCL.
> > > > If it is mounted then it may hang if filesystem needs to flush some
> > > > data.    
> > >  
> > > Thanks for the reply. The array was not mounted, that said, given that
> > > it was fully down, there wasn't a way to flush the data if it had been
> > > (cable problem to an enclosure, all the drives disappeared at once)
> > >   
> > > > Please, try umount the array if it mounted somewhere and then try:
> > > > 
> > > > # echo inactive > /sys/block/md6/md/array_state
> > > > # echo clear > /sys/block/md6/md/array_state    
> > > 
> > > I can try this next time (already had to reboot), thanks.
> > > 
> > > That said, mdadm should output a better message in this case  
> > > > > mdadm: /dev/md6 does not appear to be an md device    
> > > is clearly wrong 
> > > 
> > > Is that something easy to fix/improve?  
> > 
> > Oh, sorry my bad, please see the code:
> > https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/tree/mdopen.c#n472
> > 
> > We are failing to "understand" the array:
> > https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/tree/util.c#n229
> > It has nothing to do with open and O_EXCL. I didn't dig into to determine
> > why.
> > 
> > Anyway, now error seems to be reasonable but maybe we should be able to
> > tract this array as valid? I requires more work and analysis so it is not
> > simple fix.  
> 
> You are definitely more knowledgeable about this than I am.
> All I can say is that the array was down, not mounted, and I couldn't
> stop it without a reboot, and that's a problem.
> 
> Any way to force stop in a case like this, would be quite welcome :)
> 
> Thanks,
> Marc

Jes, how you see this?

Thanks,
Mariusz