Re: [PATCH RESEND] block: pin user PI pages for writing when ref tags are remapped
From: Keith Busch
Date: Fri Nov 20 2026 - 13:37:30 EST
On Fri, Nov 20, 2026 at 09:44:13AM -0800, Caleb Sander Mateos wrote:
> On Fri, Nov 20, 2026 at 9:23 AM Keith Busch <kbusch@xxxxxxxxxx> wrote:
> >
> > On Fri, Nov 20, 2026 at 09:13:41AM -0800, Christoph Hellwig wrote:
> > > On Fri, Nov 20, 2026 at 09:58:32AM -0700, Keith Busch wrote:
> > > > > Agreed. I'll keep the PI iterator as ITER_SOURCE. In v2,
> > > > > bio_integrity_map_user() will bounce the user PI buffer for REQ_OP_WRITE
> > > > > when the integrity profile has BLK_INTEGRITY_REF_TAG. That way the
> > > > > block layer can remap the tags in its own copy.
> > > >
> > > > Can avoid bouncing? I think the remapping flags can be tied to the
> > > > user's reftag check request.
> > > >
> > > > If BIP_CHECK_REFTAG is set from the user setting IO_INTEGRITY_CHK_REFTAG
> > > > on a write, we can map the PI buffer for writing. The user explicitly
> > > > asked to have the ref tags checked by the device, and partitions require
> > > > remapping them, so they've told the kernel to write to their buffer.
> > >
> > > No, the user did not ask for violating the mapping direction. The
> > > whole remapping thing very much is an internal implementation detail
> >
> > Okay, fine. But can we skip bouncing and remapping if the flag isn't
> > set?
>
> Currently, the remapping happens regardless of BIP_CHECK_REFTAG. It
> depends only on whether the block device reports ref tag support via
> BLK_INTEGRITY_REF_TAG (i.e. Type 1 or Type 2 PI). I think it would be
> a UAPI behavior change to make the ref tag remapping conditional on
> BIP_CHECK_REFTAG.
Right, today remapping is checked when BLK_INTEGRITY_REF_TAG is set in
the profile flags. My proposal is to change that check to
BIP_CHECK_REFTAG in the bio integrity flags. That is indeed a user
visible change in behavior, so that really hurts the suggestion.
> I don't see a way to tell ahead of time whether remapping will be
> required without walking the metadata buffer an additional time in
> bio_integrity_map_iter() to check the current ref tags.
The flag is set in the bip_flags, and you can't merge bio's that have
different bip_flags. So you only need to check the first one rather than
walk all of them.
But it's a moot point because of the user visible change. I suppose you
just shouldn't use a PI format if you don't want your metadata to have
that interpretation overlayed on it.