[PATCH v4 08/13] btrfs:zoned: make sb for npo2 zone devices align with sb log offsets

[Johannes Thumshirn]

On 20/05/2022 11:07, Pankaj Raghav wrote:
> On 5/19/22 09:57, Johannes Thumshirn wrote:
>>> Unfortunately it is not possible to just move the WP in zoned devices.
>>> The only alternative that I could use is to do write zeroes which are
>>> natively supported by some devices such as ZNS. It would be nice to know
>>> if someone had a better solution to this instead of doing write zeroes
>>> in zoned devices.
>>>
>>
>> I have another question. In case we need to pad the sb zone with a write
>> zeros and have a power fail between the write-zeros and the regular 
>> super-block write, what happens? I know this padding is only done for the
>> backup super blocks, never the less it can happen and it can happen when
>> the primary super block is also corrupted.
>>
>> AFAIU we're then trying to reach out for a backup super block, look at the
>> write pointer and it only contains zeros but no super block, as only the 
>> write-zeros has reached the device and not the super block write.
>>
>> How is this situation handled?
>>
> That is a very good point. I did think about this situation while adding
> padding to the mirror superblock with write zeroes. If the drive is
> **less than 4TB** and with the **primary superblock corrupted**, then it
> will be an issue with the situation you have described for npo2 drives.
> That situation is not handled here. Ofc this is not an issue when we
> have the second mirror at 4TB for bigger drives. Do you have some ideas
> in mind for this failure mode?

The only idea I have for this is creating a bounce buffer, write the padding
and the super-block into the buffer and then submit it. But that's too ugly
to live.

And it would involve changing non-zoned super-block writing code, which I think
is way to risky.