Re: [PATCH RESEND] block: pin user PI pages for writing when ref tags are remapped
From: Martin K. Petersen
Date: Fri Nov 20 2026 - 22:14:55 EST
Christoph,
> We could fine tune based on if we need remapping due to partitions, if
> we limited remapping to partitions, but I think Martin last time
> voiced an opinion that remapping could also be used to pass in
> arbitrary tags (which always felt a bit odd to me).
The application has no way of knowing what the first ref tag should be
except in cases where it is writing directly to a single block device.
Once the storage configuration involves multiple devices, or when
writing to a file, the concept of the user-visible ref tag being
LBA-based goes out the window. All the application knows is that it is
writing to block N of a given file.
That's why we then allow the application to pass in the seed value to
validate the PI against. Just like DIX and ILBRT/EILBRT, except this
particular translation happens at the application-kernel boundary.
--
Martin K. Petersen