Re: [PATCH] generic/563: use a loop device to avoid partition incompatibility

[Zorro Lang <zlang@redhat.com>]

On Fri, Dec 11, 2020 at 10:21:40AM -0500, Brian Foster wrote:
> On Fri, Dec 11, 2020 at 04:45:08PM +0800, Zorro Lang wrote:
> > On Thu, Dec 10, 2020 at 11:14:26AM -0500, Brian Foster wrote:
> > > cgroup writeback accounting does not track partition level
> > > statistics. Instead, I/O is accounted against the parent device. As
> > > a result, the test fails if the scratch device happens to be a
> > > device partition. Since parent level stats are potentially polluted
> > > by factors external to the test, wrap the scratch device in a
> > > loopback device to guarantee the test always runs on a top-level
> > > block device.
> > > 
> > > Reported-by: Boyang Xue <bxue@redhat.com>
> > > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > > ---
> > >  tests/generic/563 | 21 ++++++++++++++-------
> > >  1 file changed, 14 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/tests/generic/563 b/tests/generic/563
> > > index 51deaa2f..9292dece 100755
> > > --- a/tests/generic/563
> > > +++ b/tests/generic/563
> > > @@ -2,7 +2,7 @@
> > >  # SPDX-License-Identifier: GPL-2.0
> > >  # Copyright (c) 2019 Red Hat, Inc.  All Rights Reserved.
> > >  #
> > > -# FS QA Test No. 011
> > > +# FS QA Test No. 563
> > >  #
> > >  # This test verifies that cgroup aware writeback properly accounts I/Os in
> > >  # various scenarios. We perform reads/writes from different combinations of
> > > @@ -26,6 +26,8 @@ _cleanup()
> > >  
> > >  	echo $$ > $cgdir/cgroup.procs
> > >  	rmdir $cgdir/$seq-cg* > /dev/null 2>&1
> > > +	umount $SCRATCH_MNT > /dev/null 2>&1
> > > +	_destroy_loop_device $LOOP_DEV > /dev/null 2>&1
> > >  }
> > >  
> > >  # get standard environment, filters and checks
> > > @@ -42,14 +44,12 @@ rm -f $seqres.full
> > >  _supported_fs generic
> > >  _require_scratch
> > >  _require_cgroup2 io
> > > +_require_loop
> > >  
> > >  # cgroup v2 writeback is only support on block devices so far
> > >  _require_block_device $SCRATCH_DEV
> > >  
> > > -smajor=$((0x`stat -L -c %t $SCRATCH_DEV`))
> > > -sminor=$((0x`stat -L -c %T $SCRATCH_DEV`))
> > >  cgdir=$CGROUP2_PATH
> > > -
> > >  iosize=$((1024 * 1024 * 8))
> > >  
> > >  # Check cgroup read/write charges against expected values. Allow for some
> > > @@ -89,12 +89,19 @@ reset()
> > >  	rmdir $cgdir/$seq-cg* > /dev/null 2>&1
> > >  	$XFS_IO_PROG -fc "pwrite 0 $iosize" $SCRATCH_MNT/file \
> > >  		>> $seqres.full 2>&1
> > > -	_scratch_cycle_mount || _fail "mount failed"
> > > +	umount $SCRATCH_MNT || _fail "umount failed"
> > > +	_mount $LOOP_DEV $SCRATCH_MNT || _fail "mount failed"
> > >  	stat $SCRATCH_MNT/file > /dev/null
> > >  }
> > >  
> > > -_scratch_mkfs >> $seqres.full 2>&1
> > > -_scratch_mount
> > > +# cgroup I/O accounting doesn't work on partitions. Use a loop device to rule
> > > +# that out.
> > > +LOOP_DEV=$(_create_loop_device $SCRATCH_DEV)
> > 
> > I recommend using a file to create loop device. If you'd like to use SCRATCH_DEV
> > to create loop device directly, you'd better to change the "_require_scratch"
> > to "_require_scratch_nocheck". Or I think it might be failed, e.g. if SCRATCH_DEV
> > is a 4k sector size device.
> > 
> 
> What's the error that occurs with a 4k device, out of curiosity? I
> suppose if it's just a repair thing then using _nocheck probably makes
> sense (or technically might make sense regardless since we're not
> formatting the scratch device directly). I don't mind creating a file
> and using loop on that, but would like to make sure I understand if/why
> it's necessary.

The XFS on underlying device will cause fsck fail, likes this:

# modprobe scsi_debug sector_size=4096 physblk_exp=0 dev_size_mb=1024
# losetup -f --show /dev/sdc
/dev/loop0
# # mkfs.xfs -f /dev/loop0
meta-data=/dev/loop0             isize=512    agcount=4, agsize=65536 blks
         =                       sectsz=512   attr=2, projid32bit=1
         =                       crc=1        finobt=1, sparse=1, rmapbt=0
         =                       reflink=1
data     =                       bsize=4096   blocks=262144, imaxpct=25
         =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0, ftype=1
log      =internal log           bsize=4096   blocks=2560, version=2
         =                       sectsz=512   sunit=0 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0
Discarding blocks...Done.
# xfs_repair -n /dev/loop0
[passed]
# losetup -d /dev/loop0
# xfs_repair -n /dev/sdc
Phase 1 - find and verify superblock...
xfs_repair: read failed: Invalid argument
xfs_repair: data size check failed
xfs_repair: cannot repair this filesystem.  Sorry.

The xfstests always do fsck on SCRATCH_DEV except you use _require_scratch_nocheck
at the beginning of a sub-case, to skip the fsck.

Thanks,
Zorro

> 
> > Others look good to me.
> > 
> 
> Thanks for the feedback.
> 
> Brian
> 
> > Thanks,
> > Zorro
> > 
> > > +smajor=$((0x`stat -L -c %t $LOOP_DEV`))
> > > +sminor=$((0x`stat -L -c %T $LOOP_DEV`))
> > > +
> > > +_mkfs_dev $LOOP_DEV >> $seqres.full 2>&1
> > > +_mount $LOOP_DEV $SCRATCH_MNT || _fail "mount failed"
> > >  
> > >  echo "+io" > $cgdir/cgroup.subtree_control || _fail "subtree control"
> > >  
> > > -- 
> > > 2.26.2
> > > 
> > 
>