Re: [PATCH v2] btrfs: Add new test for qgroup assign functionality

[Sidong Yang <realwakka@gmail.com>]

On Wed, Sep 23, 2020 at 09:55:02AM +0800, Qu Wenruo wrote:
> 
> 
> On 2020/9/22 下午11:36, Sidong Yang wrote:
> > This new test will test btrfs's qgroup assign functionality. The
> > test has 3 cases.
> > 
> >  - assign, no shared extents
> >  - assign, shared extents
> >  - snapshot -i, shared extents
> > 
> > Each cases create subvolumes and assign qgroup in their own way
> > and check with the command "btrfs check".
> > 
> > Cc: Qu Wenruo <wqu@suse.com>
> > Cc: Eryu Guan <guan@eryu.me>
> > 
> > Signed-off-by: Sidong Yang <realwakka@gmail.com>
> > ---
> > v2: create new test and use the cases
> > ---
> >  tests/btrfs/221     | 121 ++++++++++++++++++++++++++++++++++++++++++++
> >  tests/btrfs/221.out |   2 +
> >  tests/btrfs/group   |   1 +
> >  3 files changed, 124 insertions(+)
> >  create mode 100755 tests/btrfs/221
> >  create mode 100644 tests/btrfs/221.out
> > 
> > diff --git a/tests/btrfs/221 b/tests/btrfs/221
> > new file mode 100755
> > index 00000000..7fe5d78f
> > --- /dev/null
> > +++ b/tests/btrfs/221
> > @@ -0,0 +1,121 @@
> > +#! /bin/bash
> > +# SPDX-License-Identifier: GPL-2.0
> > +# Copyright (c) 2020 YOUR NAME HERE.  All Rights Reserved.
> 
> So, "YOUR NAME HERE" is your name? :)

Oops! I missed it.

> 
> > +#
> > +# FS QA Test 221
> > +#
> > +# Test the assign functionality of qgroups
> > +#
> > +seq=`basename $0`
> > +seqres=$RESULT_DIR/$seq
> > +echo "QA output created by $seq"
> > +
> > +here=`pwd`
> > +tmp=/tmp/$$
> > +status=1	# failure is the default!
> > +trap "_cleanup; exit \$status" 0 1 2 3 15
> > +
> > +_cleanup()
> > +{
> > +	cd /
> > +	rm -f $tmp.*
> > +}
> > +
> > +# get standard environment, filters and checks
> > +. ./common/rc
> > +. ./common/filter
> > +. ./common/reflink
> > +
> > +# remove previous $seqres.full before test
> > +rm -f $seqres.full
> > +
> > +# real QA test starts here
> > +
> > +# Modify as appropriate.
> > +_supported_fs generic
> 
> It's a btrfs specific test.
Thanks!
> 
> > +_supported_os Linux
> > +
> > +_require_test
> 
> You don't need test_dev at all.
Yeah, I just realized that I used only scratch dev noy test dev.
Thanks!
> 
> > +_require_scratch
> > +_require_btrfs_qgroup_report
> > +_require_cp_reflink
> > +
> > +# Test assign qgroup for submodule with shared extents by reflink
> > +assign_shared_test()
> > +{
> > +	echo "=== qgroup assign shared test ===" >> $seqres.full
> > +	_run_btrfs_util_prog quota enable $SCRATCH_MNT
> 
> I'm not sure if _run_btrfs_util_prog is still recommended.
> IIRC nowadays we recommend to call $BTRFS_UTIL_PROG directly.
> 
> Test case btrfs/193 would be the example.
It's good to replace it! Thanks.
> 
> > +	_run_btrfs_util_prog qgroup create 1/100 $SCRATCH_MNT
> > +
> > +	_run_btrfs_util_prog subvolume create $SCRATCH_MNT/a
> > +	subvolid=$(_btrfs_get_subvolid $SCRATCH_MNT a)
> > +	_run_btrfs_util_prog qgroup assign 0/$subvolid 1/100 $SCRATCH_MNT
> 
> "btrfs qgroup assign" can take path directly. This would save your some
> lines. E.g:
> 
>   # btrfs qgroup create 1/100 /mnt/btrfs/
>   # btrfs qgroup assign /mnt/btrfs/ 1/100 /mnt/btrfs/
>   # btrfs qgroup  show -pc /mnt/btrfs/
>   qgroupid         rfer         excl parent  child
>   --------         ----         ---- ------  -----
>   0/5          16.00KiB     16.00KiB 1/100   ---
>   1/100        16.00KiB     16.00KiB ---     0/5
Thanks for good tip!

> 
> > +
> > +	_run_btrfs_util_prog subvolume create $SCRATCH_MNT/b
> > +	subvolid=$(_btrfs_get_subvolid $SCRATCH_MNT b)
> > +	_run_btrfs_util_prog qgroup assign 0/$subvolid 1/100 $SCRATCH_MNT
> 
> Shouldn't this assign happens when we have shared extents between the
> two subvolumes?
> 
> Now you're just testing the basic qgroup functionality of accounting,
> not assign.
> 
> For real assign tests, what we want is either:
> - After assign, qgroup accounting is still correct
>   We don't need even to rescan.
>   And "btrfs check" will verify the numbers are correct.
> 
> - After assign, qgroup accounting is inconsistent
>   At least we should either have qgroup inconsistent bit set, or qgroup
>   rescan kicked in automatically.
>   And "btrfs check" will skip the qgroup numbers.
> 
> But in your case, we're assigning two empty subovlumes into the same
> qgroup, then do some operations.
> This only covers the "assign, no shared extents" case.

You mean that there should be some data with reflink before assigning?
If so, the code below should be executed before assigning qgroups.
Should test process be like this?
make submodules -> make data -> copy with reflink -> assign qgroup

> 
> > +	_run_btrfs_util_prog quota rescan -w $SCRATCH_MNT
> > +
> > +	_ddt of="$SCRATCH_MNT"/a/file1 bs=1M count=1 >> $seqres.full 2>&1
> > +	cp --reflink=always "$SCRATCH_MNT"/a/file1 "$SCRATCH_MNT"/b/file1 >> $seqres.full 2>&1
> > +
> > +	_scratch_unmount
> 
> Since you're unmounting here, why not keep the _scratch_mkfs and
> _scratch_unmount pair in the same function?
> 
> > +	_run_btrfs_util_prog check $SCRATCH_DEV
> > +}
> > +
> > +# Test assign qgroup for submodule without shared extents
> > +assign_no_shared_test()
> > +{
> > +	echo "=== qgroup assign no shared test ===" >> $seqres.full
> > +	_run_btrfs_util_prog quota enable $SCRATCH_MNT
> > +	_run_btrfs_util_prog qgroup create 1/100 $SCRATCH_MNT
> > +
> > +	_run_btrfs_util_prog subvolume create $SCRATCH_MNT/a
> > +	subvolid=$(_btrfs_get_subvolid $SCRATCH_MNT a)
> > +	_run_btrfs_util_prog qgroup assign 0/$subvolid 1/100 $SCRATCH_MNT
> > +
> > +	_run_btrfs_util_prog subvolume create $SCRATCH_MNT/b
> > +	subvolid=$(_btrfs_get_subvolid $SCRATCH_MNT b)
> > +	_run_btrfs_util_prog qgroup assign 0/$subvolid 1/100 $SCRATCH_MNT
> > +
> > +	_run_btrfs_util_prog quota rescan -w $SCRATCH_MNT
> 
> No, we don't want rescan.
> 
> And the timing is still wrong.
Yeah, I'll delete it.
> 
> > +	_scratch_unmount
> > +
> > +	_run_btrfs_util_prog check $SCRATCH_DEV
> > +}
> > +
> > +# Test snapshot with assigning qgroup for submodule
> > +snapshot_test()
> > +{
> > +	echo "=== qgroup snapshot test ===" >> $seqres.full
> > +	_run_btrfs_util_prog quota enable $SCRATCH_MNT
> > +
> > +	_run_btrfs_util_prog subvolume create $SCRATCH_MNT/a
> > +	subvolid=$(_btrfs_get_subvolid $SCRATCH_MNT a)
> > +
> > +	_run_btrfs_util_prog subvolume snapshot -i 0/$subvolid $SCRATCH_MNT/a $SCRATCH_MNT/b
> > +	subvolid=$(_btrfs_get_subvolid $SCRATCH_MNT b)
> 
> Even we're snapshotting on the source subvolume, since it's empty, we
> will only copy the root item, resulting two empty subvolumes without
> sharing anything.
> 
> You need to at least fill the source subvolumes with some data.
> It's better to bump the tree level with some inline extents.

I should write some data before snapshot. is it right?
Thanks for all your comments.

Thanks,
Sidong

> 
> Thanks,
> Qu
> 
> > +
> > +	_run_btrfs_util_prog quota rescan -w $SCRATCH_MNT
> > +	_scratch_unmount
> > +
> > +	_run_btrfs_util_prog check $SCRATCH_DEV
> > +}
> > +
> > +
> > +_scratch_mkfs > /dev/null 2>&1
> > +_scratch_mount
> > +assign_no_shared_test
> > +
> > +_scratch_mkfs > /dev/null 2>&1
> > +_scratch_mount
> > +assign_shared_test
> > +
> > +_scratch_mkfs > /dev/null 2>&1
> > +_scratch_mount
> > +snapshot_test
> > +
> > +# success, all done
> > +echo "Silence is golden"
> > +status=0
> > +exit
> > diff --git a/tests/btrfs/221.out b/tests/btrfs/221.out
> > new file mode 100644
> > index 00000000..aa4351cd
> > --- /dev/null
> > +++ b/tests/btrfs/221.out
> > @@ -0,0 +1,2 @@
> > +QA output created by 221
> > +Silence is golden
> > diff --git a/tests/btrfs/group b/tests/btrfs/group
> > index 1b5fa695..cdda38f3 100644
> > --- a/tests/btrfs/group
> > +++ b/tests/btrfs/group
> > @@ -222,3 +222,4 @@
> >  218 auto quick volume
> >  219 auto quick volume
> >  220 auto quick
> > +221 auto quick qgroup
> > 
>