On 2022/9/5 22:31, Christoph Hellwig wrote:

On Mon, Sep 05, 2022 at 02:59:33PM +0800, Qu Wenruo wrote:

quoted

Mostly due to the fact that metadata and data go split ways for
verification.

All the verification for data happens at endio time.

Yes.

quoted

While part of the verification of metadata (bytenr, csum, level,
tree-checker) goes at endio, but transid, checks against parent are all
done at btrfs_read_extent_buffer() time.

This also means, the read-repair happens at different timing.

Yes.  read-repair for metadata currently is very different than that
from data.  But that is something that exists already in is not new
in this series.

quoted

But what about putting all the needed metadata info (first key, level,
transid etc) also into bbio (using a union to take the same space of
data csum), so that all verification and read repair can happen at endio
time, the same timing as data?

I thought about that.  And I suspect it probably is the right thing
to do.  I'm mostly stayed away from it because it doesn't really
help with the goal in this series, and I also don't have good
code coverage to fail comfortable touching the metadata checksum
handling and repair.  I can offer this sneaky deal:  if someone
help creating good metadata repair coverage in xfstests, I will look
into this next.

Then may I take this work since it's mostly independent and you can
continue your existing work without being distracted?

Thanks,
Qu