Re: [PATCH v3 19/19] selftests: ublk: add end-to-end integrity test

[Ming Lei <ming.lei@redhat.com>]

On Tue, Jan 06, 2026 at 05:32:29PM -0800, Caleb Sander Mateos wrote:
> On Tue, Jan 6, 2026 at 4:21 PM Ming Lei <ming.lei@redhat.com> wrote:
> >
> > On Tue, Jan 06, 2026 at 09:15:44AM -0800, Caleb Sander Mateos wrote:
> > > On Tue, Jan 6, 2026 at 6:10 AM Ming Lei <ming.lei@redhat.com> wrote:
> > > >
> > > > On Mon, Jan 05, 2026 at 05:57:51PM -0700, Caleb Sander Mateos wrote:
> > > > > Add test case loop_08 to verify the ublk integrity data flow. It uses
> > > > > the kublk loop target to create a ublk device with integrity on top of
> > > > > backing data and integrity files. It then writes to the whole device
> > > > > with fio configured to generate integrity data. Then it reads back the
> > > > > whole device with fio configured to verify the integrity data.
> > > > > It also verifies that injected guard, reftag, and apptag corruptions are
> > > > > correctly detected.
> > > > >
> > > > > Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
> > > > > ---
> > > > >  tools/testing/selftests/ublk/Makefile        |   1 +
> > > > >  tools/testing/selftests/ublk/test_loop_08.sh | 111 +++++++++++++++++++
> > > > >  2 files changed, 112 insertions(+)
> > > > >  create mode 100755 tools/testing/selftests/ublk/test_loop_08.sh
> > > > >
> > > > > diff --git a/tools/testing/selftests/ublk/Makefile b/tools/testing/selftests/ublk/Makefile
> > > > > index bfd68ae64142..ab745443fd58 100644
> > > > > --- a/tools/testing/selftests/ublk/Makefile
> > > > > +++ b/tools/testing/selftests/ublk/Makefile
> > > > > @@ -33,10 +33,11 @@ TEST_PROGS += test_loop_02.sh
> > > > >  TEST_PROGS += test_loop_03.sh
> > > > >  TEST_PROGS += test_loop_04.sh
> > > > >  TEST_PROGS += test_loop_05.sh
> > > > >  TEST_PROGS += test_loop_06.sh
> > > > >  TEST_PROGS += test_loop_07.sh
> > > > > +TEST_PROGS += test_loop_08.sh
> > > > >  TEST_PROGS += test_stripe_01.sh
> > > > >  TEST_PROGS += test_stripe_02.sh
> > > > >  TEST_PROGS += test_stripe_03.sh
> > > > >  TEST_PROGS += test_stripe_04.sh
> > > > >  TEST_PROGS += test_stripe_05.sh
> > > > > diff --git a/tools/testing/selftests/ublk/test_loop_08.sh b/tools/testing/selftests/ublk/test_loop_08.sh
> > > > > new file mode 100755
> > > > > index 000000000000..ca289cfb2ad4
> > > > > --- /dev/null
> > > > > +++ b/tools/testing/selftests/ublk/test_loop_08.sh
> > > > > @@ -0,0 +1,111 @@
> > > > > +#!/bin/bash
> > > > > +# SPDX-License-Identifier: GPL-2.0
> > > > > +
> > > > > +. "$(cd "$(dirname "$0")" && pwd)"/test_common.sh
> > > > > +
> > > > > +if ! _have_program fio; then
> > > > > +     exit $UBLK_SKIP_CODE
> > > > > +fi
> > > > > +
> > > > > +fio_version=$(fio --version)
> > > > > +if [[ "$fio_version" =~ fio-[0-9]+\.[0-9]+$ ]]; then
> > > > > +     echo "Requires development fio version with https://github.com/axboe/fio/pull/1992"
> > > > > +     exit $UBLK_SKIP_CODE
> > > > > +fi
> > > > > +
> > > > > +TID=loop_08
> > > > > +
> > > > > +_prep_test "loop" "end-to-end integrity"
> > > > > +
> > > > > +_create_backfile 0 256M
> > > > > +_create_backfile 1 32M # 256M * (64 integrity bytes / 512 data bytes)
> > > > > +integrity_params="--integrity_capable --integrity_reftag
> > > > > +                  --metadata_size 64 --pi_offset 56 --csum_type t10dif"
> > > > > +dev_id=$(_add_ublk_dev -t loop -u $integrity_params "${UBLK_BACKFILES[@]}")
> > > >
> > > > I tried above setting:
> > > >
> > > > ./kublk add -t loop --integrity_capable --integrity_reftag --metadata_size 64 --pi_offset 56 --csum_type t10dif --foreground -u /dev/sdb /dev/sdc
> > > > dev id 1: nr_hw_queues 2 queue_depth 128 block size 512 dev_capacity 8388608
> > > >         max rq size 1048576 daemon pid 38295 flags 0x160c2 state LIVE
> > > >         queue 0: affinity(0 )
> > > >         queue 1: affinity(8 )
> > > >
> > > > However, IO error is always triggered:
> > > >
> > > > [ 9202.316382] ublkb1: ref tag error at location 0 (rcvd 128)
> > > > [ 9202.317171] Buffer I/O error on dev ublkb1, logical block 0, async page read
> > >
> > > Hmm, what are the initial contents of /dev/sdc? It looks like they are
> > > nonzero, as the reftag being read for logical block 0 is 128 rather
> > > than the expected 0 (the reftag would be read from bytes 60 to 63 of
> > > /dev/sdc). In general, though, the partition scan may be expected to
> > > fail the bio-integrity-auto checks if the integrity data hasn't been
> > > initialized. I don't think this is an issue, since the partition scan
> > > is looking for a partition table but there's no guarantee that one
> > > exists.
> > > You can disable the kernel integrity checks if you want by writing 0
> > > to /sys/block/ublkb1/integrity/read_verify. However, I'm not sure it's
> > > possible to do this soon enough to take effect before the partition
> > > scan.
> > > We could also use the UBLK_F_NO_AUTO_PART_SCAN feature, once it lands,
> > > to suppress the partition scan and these error messages.
> >
> > UBLK_F_NO_AUTO_PART_SCAN can't avoid the following read failure.
> 
> Are you saying the reads aren't coming from the partition scan? What
> else would be issuing reads to the ublk device before fio is run on
> it?

I mean people can run fio read after disk is setup, or udev may read disk
in background.

> 
> >
> > I guess the issue can be avoided by adding small superblock to the meta
> > data file, then format it in the 1st time when superblock doesn't exist.
> 
> Not sure the complexity of a superblock is necessary. Could just
> initialize all the protection information with valid reftags and guard
> tags.

For verification purpose, the superblock can include only the info if PI meta
file is formatted or not. But it isn't a big deal if the meta backfile can
be formatted with tags in the specified setting.

> 
> >
> > This way will make it usable from test/verify purpose.
> 
> I'm still not following why it's not "usable" currently. Only blocks
> that have been written to (with protection information generated by
> fio or the kernel) can be read back, but that's how block devices are
> expected to be used. It seems fine to me for test purposes.
> test_loop_08.sh writes to the full ublk device before verifying it, so
> it shouldn't encounter any invalid reftags.

It triggers READ error easily, and will cause noisy report, just like
what I did...

Also you can't run FS test workload...


Thanks,
Ming