Hi Anuj!

I've been mulling over this for a few days...

The block layer currently infers this by looking at the csum_type
(e.g., in blk_integrity_generate). I assumed userspace could do the
same, so I didn't expose a separate pi_tuple_size field. Do you see
this differently?

When the block layer data integrity code was originally designed, the
concept of non-PI metadata didn't exist.

Then NVMe came along and we added support for opaque metadata in
addition to the PI.

As a result, the block layer considers the opaque metadata part of the
PI but it technically isn't. It really should be the other way around:
The PI is a subset of the metadata.

It would require quite a bit of rototilling to metadata-ize the block
layer plumbing at this point. But for a new user API, I do think we
should try to align with the architecture outlined in the standards.

-- 
Martin K. Petersen