Re: Inquiry: disk format change in the future

[Eric Sandeen <sandeen-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

Ryusuke Konishi wrote:
> Hi!
> On Wed, 19 Aug 2009 22:55:39 -0500, Eric Sandeen wrote:
>> Ryusuke Konishi wrote:
>>> (reply added to "Yann E. MORIN")
>>>
>>> Hi,
>>> On Wed, 19 Aug 2009 11:31:56 +0200, "Yann E. MORIN" wrote:
>>>> Hello All!
>>>>
>>>> I've searched the archives, but could not find the answer to the
>>>> following questions:
>>>>
>>>>  - will there be, or what is the probability of, a disk format change
>>>>    in the near future (few months)?
>>> We still have some possibilities for the format change at this time.
>>>
>>> Basically I intend to keep the disk format, but some new features like
>>>
>>> - Extended attributes, ACLs
>>> - Rollback
>>> - Checkpoint based remote replication
>>>
>>> may require it.  So far, we haven't seen any specific requisition.
>>>  
>> Do you expect that any of the changes may impact the ability of
>> utilities such as blkid or file to recognize a nilfs filesystem?
> 
> I don't think these affect blkid or similar programs even if
> additional fields may be required in the superblock.
> 
> One my concern about the automated recognition is too short magic
> field of nilfs.  I think those tools will need to confirm the checksum
> in addition to the magic field.

That's possible, at least for blkid.  I did that for recognizing lvm
volumes.  I'll take a look.

> If I have a chance to reorganize the superblock, I would like to make
> it longer and adjust some other fields.  But I will not do so for the
> above features.
> 
>> I'd happily write patches for those utilities to add nilfs recognition
>> if you think there is a stable fingerprint that can be recognized.
>>
>> Thanks,
>> -Eric
> 
> Thank you for the proposal! I'm grateful if you could do so.  I think
> the time when we drop the experimental flag from Kconfig is the time.

Ok, I can wait 'til then to submit it.  Once I get time to see what it
takes I can run the patch past you, see if you think it's a stable
fingerprint, and either send it upstream or wait, based on what you
think is best at that point.

Thanks,
-Eric