Re: fine-grained PI control

["Martin K. Petersen" <martin.petersen@oracle.com>]

Hi Anuj!

> When the drive is formatted with type1 integrity and dix is supported,
> the block layer would generate/verify the guard and reftag. For scsi,
> apptag would also need to be generated/verified as reftag-check can't
> be specified alone. But I am not not able to find (in SCSI code) where
> exactly: 1. this apptag is being generated
> 2. and getting added to the PI buffer and scsi command.

The block layer on its own doesn't have a way to fill out the app tag in
any meaningful way. But if the PI comes from userland, the application
can set the app tag to whatever it pleases.

Originally we had code in Linux allowing filesystems to attach
additional metadata to bios which would then be stored in the app tag
space. This was intended for backpointers but never really took off.
However, if we move the burden of PI generation to filesystems as
Christoph suggested, then we can revisit using the app tag space.

No matter what, though, for target we need to be able to store whichever
app tag the initiator sends.

-- 
Martin K. Petersen	Oracle Linux Engineering