Re: [PATCH] md: forbid a RAID5 from having both a bitmap and a journal.

[Wols Lists <antlists@youngman.org.uk>]

On 18/10/17 15:48, John Stoffel wrote:
>>>>>> "NeilBrown" == NeilBrown  <neilb@suse.com> writes:
> 
> NeilBrown> On Tue, Oct 17 2017, John Stoffel wrote:
>>>>>>>> "NeilBrown" == NeilBrown  <neilb@suse.com> writes:
>>>
> NeilBrown> Having both a bitmap and a journal is pointless.
> NeilBrown> Attempting to do so can corrupt the bitmap if the journal
> NeilBrown> replay happens before the bitmap is initialized.
> NeilBrown> Rather than try to avoid this corruption, simply
> NeilBrown> refuse to allow arrays with both a bitmap and a journal.
> NeilBrown> So:
> NeilBrown> - if raid5_run sees both are present, fail.
>>>
>>> So what happens if there's someone out there with an array setup with
>>> both already?  Should the journal or the bitmap be removed at this
>>> time?  
> 
> NeilBrown> If someone has an array like that they can assemble it with
> NeilBrown>   --update=no-bitmap
> 
> NeilBrown> I'd rather not automatically disable things.
> 
> That works for me, assuming there's a useful message in the logs
> explaining what's gone wrong here.  Either mdadm emits the message, or
> the kernel does.
> 
Or are we better off, initially, just adding the code that prevents
adding a bitmap or journal? Allow existing arrays to run, emitting
messages to the console and/or log that this is not a good idea?

Then maybe a few versions down the line actually refuse to start an
array in this state.

I don't know how many arrays are in this state now, but doing this
should reduce the number of people getting a nasty shock when their
system fails to boot ...

Cheers,
Wol