[PATCH 0/4] Random fixes for xfstest

[PATCH 0/4] Random fixes for xfstest

[Anju T Sudhakar]

Below are some fixes for xfstest- mostly centered around 64k
blocksize(ppc64) and with xfs filesystem.


Anju T Sudhakar (4):
  xfs/432: Skip this test for dax config and pagesize not equal to the
    specified block size
  generic/594: Fix test failure for xfs on older quota-tools
  generic/600: Fix this test on xfs with older quota-tools
  generic/219: Fix test failure on 64k block size

 tests/generic/219 | 5 +++++
 tests/generic/594 | 6 ++++++
 tests/generic/600 | 8 ++++++++
 tests/xfs/432     | 4 ++++
 4 files changed, 23 insertions(+)

-- 
2.24.1

[PATCH 1/4] xfs/432: Skip this test for dax config and pagesize not equal to the specified block size

[Anju T Sudhakar]

This test case passes 1k as the block size(to keep that as minimum),
and hence it fails to mount with "-o dax" option on platforms where
pageszie not equal to the specified block size. This leads to test
case reported as "failed" with dax config.

So skip this test when dax is enabled.

Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
---
 tests/xfs/432 | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/tests/xfs/432 b/tests/xfs/432
index 40cb2474..9022af69 100755
--- a/tests/xfs/432
+++ b/tests/xfs/432
@@ -32,6 +32,10 @@ _require_scratch
 
 rm -f "$seqres.full"
 
+# Skip this test, if pagesize !=1k and dax is enabled.
+if [[ $(get_page_size) -ne 1024 ]]; then
+	_exclude_scratch_mount_option dax
+fi
 echo "Format and mount"
 # We need to create a directory with a huuuge extent record.  Normally
 # a rapidly expanding directory gets its blocks allocated in lockstep --
-- 
2.24.1

[PATCH 2/4] generic/594: Fix test failure for xfs on older quota-tools

[Anju T Sudhakar]

For xfs we need the quota-tools containing commit fdd774bf08e568
("quota-tools: pass quota type to QCMD for Q_XFS_GETQSTAT")to run this
test. Otherwise this test will fail on xfs, as the grace time is not set
correctly. So check for the quota-tool with the required commit before
running this test on xfs.

Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
---
 tests/generic/594 | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/tests/generic/594 b/tests/generic/594
index 88f9a0dc..668c1fae 100755
--- a/tests/generic/594
+++ b/tests/generic/594
@@ -6,6 +6,8 @@
 #
 # Test per-type(user, group and project) filesystem quota timers, make sure
 # each of grace time can be set/get properly.
+# For xfs, you need a quota-tools containing commit fdd774bf08e568 for this
+# test to run properly- v4.06 should do.
 #
 . ./common/preamble
 _begin_fstest auto quick quota
@@ -18,9 +20,13 @@ _begin_fstest auto quick quota
 _supported_fs generic
 _require_scratch
 _require_setquota_project
+version=$(setquota -V | head -n1 | cut -d" " -f4)
 # V4 XFS doesn't support to mount project and group quota together
 if [ "$FSTYP" = "xfs" ];then
 	_require_scratch_xfs_crc
+	if [ "$version" \< "4.06" ];then
+		_notrun "This test requires quota-tools 4.06 or greater"
+	fi
 fi
 _require_quota
 
-- 
2.24.1

[PATCH 3/4] generic/600: Fix this test on xfs with older quota-tools

[Anju T Sudhakar]

For xfs, we need quota-tools containing commit be96da2353669d ("quota-tools:
Set FS_DQ_TIMER_MASK for individual xfs grace times"), to run this test.
Otherwise the test fails on xfs. Check the quota-tools before running
the test.

Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
---
 tests/generic/600 | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/tests/generic/600 b/tests/generic/600
index a4233ecc..c8dc1bd5 100755
--- a/tests/generic/600
+++ b/tests/generic/600
@@ -10,6 +10,9 @@
 # This test only exercises user quota because it's not known whether the
 # filesystem can set individual grace timers for each quota type
 #
+# For xfs, you need a quota-tools containing commit be96da2353669d to
+# run this test properly- 4.06 should do.
+
 . ./common/preamble
 _begin_fstest auto quick quota
 
@@ -24,6 +27,11 @@ _require_quota
 _require_user
 _require_setquota_project
 
+version=$(setquota -V | head -n1 | cut -d" " -f4)
+if [ "$FSTYP" = "xfs" ] && [ "$version" \< "4.06" ];then
+	_notrun "This test requires quota-tools 4.06 or greater"
+fi
+
 _scratch_mkfs >$seqres.full 2>&1
 _qmount_option "usrquota"
 _qmount
-- 
2.24.1

[PATCH 4/4] generic/219: Fix test failure on 64k block size

[Anju T Sudhakar]

Skip this test on 64k, as 4k as taken as the default block size.

Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
---
But I have a Query:
This test fails on ext4&xfs with 64k block size. Do we need to
rephrase this test to incorporate the 64k bs scenario as well?
If not we can skip this test.

 tests/generic/219 | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/tests/generic/219 b/tests/generic/219
index d3d257f1..d38fd033 100755
--- a/tests/generic/219
+++ b/tests/generic/219
@@ -79,6 +79,11 @@ test_accounting()
 
 # real QA test starts here
 
+# Skip this test on 64k block size scenario
+bsize=$(_get_block_size $SCRATCH_MNT)
+if [ $bsize -ge 4096 ];then
+	_notrun "Test assumes 4k as the default block size"
+fi
 _scratch_unmount 2>/dev/null
 _scratch_mkfs >> $seqres.full 2>&1
 _scratch_mount "-o usrquota,grpquota"
-- 
2.24.1

Re: [PATCH 1/4] xfs/432: Skip this test for dax config and pagesize not equal to the specified block size

[Dave Chinner]

On Mon, Oct 11, 2021 at 01:25:49PM +0530, Anju T Sudhakar wrote:
> This test case passes 1k as the block size(to keep that as minimum),
> and hence it fails to mount with "-o dax" option on platforms where
> pageszie not equal to the specified block size. This leads to test
> case reported as "failed" with dax config.
> 
> So skip this test when dax is enabled.
> 
> Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
> ---
>  tests/xfs/432 | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/tests/xfs/432 b/tests/xfs/432
> index 40cb2474..9022af69 100755
> --- a/tests/xfs/432
> +++ b/tests/xfs/432
> @@ -32,6 +32,10 @@ _require_scratch
>  
>  rm -f "$seqres.full"
>  
> +# Skip this test, if pagesize !=1k and dax is enabled.
> +if [[ $(get_page_size) -ne 1024 ]]; then
> +	_exclude_scratch_mount_option dax
> +fi

Page size will never be 1kB so there's no point even checking the
page size here.

The problem here is that exclusing the dax mount option is
completely wrong when "-o dax=never" is used - it means that even
though the storage hardware is DAX capable, don't use dax at all,
and so tests that use 1kB block sizes will actaully work.

IOWs, this test should run if dax=never or dax=inode is set, but not
if dax=always or just plain "dax" is set...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 2/4] generic/594: Fix test failure for xfs on older quota-tools

[Dave Chinner]

On Mon, Oct 11, 2021 at 01:25:50PM +0530, Anju T Sudhakar wrote:
> For xfs we need the quota-tools containing commit fdd774bf08e568
> ("quota-tools: pass quota type to QCMD for Q_XFS_GETQSTAT")to run this
> test. Otherwise this test will fail on xfs, as the grace time is not set
> correctly. So check for the quota-tool with the required commit before
> running this test on xfs.
> 
> Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
> ---
>  tests/generic/594 | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/tests/generic/594 b/tests/generic/594
> index 88f9a0dc..668c1fae 100755
> --- a/tests/generic/594
> +++ b/tests/generic/594
> @@ -6,6 +6,8 @@
>  #
>  # Test per-type(user, group and project) filesystem quota timers, make sure
>  # each of grace time can be set/get properly.
> +# For xfs, you need a quota-tools containing commit fdd774bf08e568 for this
> +# test to run properly- v4.06 should do.
>  #
>  . ./common/preamble
>  _begin_fstest auto quick quota
> @@ -18,9 +20,13 @@ _begin_fstest auto quick quota
>  _supported_fs generic
>  _require_scratch
>  _require_setquota_project
> +version=$(setquota -V | head -n1 | cut -d" " -f4)
>  # V4 XFS doesn't support to mount project and group quota together
>  if [ "$FSTYP" = "xfs" ];then
>  	_require_scratch_xfs_crc
> +	if [ "$version" \< "4.06" ];then
> +		_notrun "This test requires quota-tools 4.06 or greater"
> +	fi

We don't test for version numbers, because feature and bug fix back
ports make testing version numbers unreliable. What needs to be done
here is to test whether the feature is available of not via a
_require_setquota_xfs_getqstat() rule that encapsulates the feature
test and _notrun call if the feature is not supported.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 4/4] generic/219: Fix test failure on 64k block size

[Dave Chinner]

On Mon, Oct 11, 2021 at 01:25:52PM +0530, Anju T Sudhakar wrote:
> Skip this test on 64k, as 4k as taken as the default block size.
> 
> Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
> ---
> But I have a Query:
> This test fails on ext4&xfs with 64k block size. Do we need to
> rephrase this test to incorporate the 64k bs scenario as well?
> If not we can skip this test.

Right, we should fix the test to use multiples of block size rather
than fixed byte sizes as we do for all the other tests that need to
work across different block sizes. e.g. probe the block size after
mkfs and then convert the "48k" numbers to "$((12 * $bs))".

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 2/4] generic/594: Fix test failure for xfs on older quota-tools

[Darrick J. Wong]

On Mon, Oct 11, 2021 at 01:25:50PM +0530, Anju T Sudhakar wrote:
> For xfs we need the quota-tools containing commit fdd774bf08e568
> ("quota-tools: pass quota type to QCMD for Q_XFS_GETQSTAT")to run this
> test. Otherwise this test will fail on xfs, as the grace time is not set
> correctly. So check for the quota-tool with the required commit before
> running this test on xfs.

This is a regression test for a kernel bug where Q_XGETQSTAT[V] ignores
the quota type that userspace specifies in the system call and instead
returns information about the first enabled quotatype.  It just so
happens that quota-tools /also/ had a bug where it would never send the
user's quota type to the kernel, so nobody noticed the nested herp and
derp.

The test should probably have mentioned kernel commit 555b2c3da1fc
("quota: honor quota type in Q_XGETQSTAT[V] calls") directly to make
this more obvious.

--D

> Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
> ---
>  tests/generic/594 | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/tests/generic/594 b/tests/generic/594
> index 88f9a0dc..668c1fae 100755
> --- a/tests/generic/594
> +++ b/tests/generic/594
> @@ -6,6 +6,8 @@
>  #
>  # Test per-type(user, group and project) filesystem quota timers, make sure
>  # each of grace time can be set/get properly.
> +# For xfs, you need a quota-tools containing commit fdd774bf08e568 for this
> +# test to run properly- v4.06 should do.
>  #
>  . ./common/preamble
>  _begin_fstest auto quick quota
> @@ -18,9 +20,13 @@ _begin_fstest auto quick quota
>  _supported_fs generic
>  _require_scratch
>  _require_setquota_project
> +version=$(setquota -V | head -n1 | cut -d" " -f4)
>  # V4 XFS doesn't support to mount project and group quota together
>  if [ "$FSTYP" = "xfs" ];then
>  	_require_scratch_xfs_crc
> +	if [ "$version" \< "4.06" ];then
> +		_notrun "This test requires quota-tools 4.06 or greater"
> +	fi
>  fi
>  _require_quota
>  
> -- 
> 2.24.1
> 

Re: [PATCH 2/4] generic/594: Fix test failure for xfs on older quota-tools

[Eryu Guan]

On Mon, Oct 11, 2021 at 09:46:52AM -0700, Darrick J. Wong wrote:
> On Mon, Oct 11, 2021 at 01:25:50PM +0530, Anju T Sudhakar wrote:
> > For xfs we need the quota-tools containing commit fdd774bf08e568
> > ("quota-tools: pass quota type to QCMD for Q_XFS_GETQSTAT")to run this
> > test. Otherwise this test will fail on xfs, as the grace time is not set
> > correctly. So check for the quota-tool with the required commit before
> > running this test on xfs.
> 
> This is a regression test for a kernel bug where Q_XGETQSTAT[V] ignores
> the quota type that userspace specifies in the system call and instead
> returns information about the first enabled quotatype.  It just so
> happens that quota-tools /also/ had a bug where it would never send the
> user's quota type to the kernel, so nobody noticed the nested herp and
> derp.
> 
> The test should probably have mentioned kernel commit 555b2c3da1fc
> ("quota: honor quota type in Q_XGETQSTAT[V] calls") directly to make
> this more obvious.

Agreed, and maybe mention the quota-tools commit as well.

Thanks,
Eryu

> 
> --D
> 
> > Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
> > ---
> >  tests/generic/594 | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/tests/generic/594 b/tests/generic/594
> > index 88f9a0dc..668c1fae 100755
> > --- a/tests/generic/594
> > +++ b/tests/generic/594
> > @@ -6,6 +6,8 @@
> >  #
> >  # Test per-type(user, group and project) filesystem quota timers, make sure
> >  # each of grace time can be set/get properly.
> > +# For xfs, you need a quota-tools containing commit fdd774bf08e568 for this
> > +# test to run properly- v4.06 should do.
> >  #
> >  . ./common/preamble
> >  _begin_fstest auto quick quota
> > @@ -18,9 +20,13 @@ _begin_fstest auto quick quota
> >  _supported_fs generic
> >  _require_scratch
> >  _require_setquota_project
> > +version=$(setquota -V | head -n1 | cut -d" " -f4)
> >  # V4 XFS doesn't support to mount project and group quota together
> >  if [ "$FSTYP" = "xfs" ];then
> >  	_require_scratch_xfs_crc
> > +	if [ "$version" \< "4.06" ];then
> > +		_notrun "This test requires quota-tools 4.06 or greater"
> > +	fi
> >  fi
> >  _require_quota
> >  
> > -- 
> > 2.24.1
> > 

Re: [PATCH 2/4] generic/594: Fix test failure for xfs on older quota-tools

[xuyang2018.jy]

on 2021/10/12 10:57, Eryu Guan wrote:
> On Mon, Oct 11, 2021 at 09:46:52AM -0700, Darrick J. Wong wrote:
>> On Mon, Oct 11, 2021 at 01:25:50PM +0530, Anju T Sudhakar wrote:
>>> For xfs we need the quota-tools containing commit fdd774bf08e568
>>> ("quota-tools: pass quota type to QCMD for Q_XFS_GETQSTAT")to run this
>>> test. Otherwise this test will fail on xfs, as the grace time is not set
>>> correctly. So check for the quota-tool with the required commit before
>>> running this test on xfs.
>>
>> This is a regression test for a kernel bug where Q_XGETQSTAT[V] ignores
>> the quota type that userspace specifies in the system call and instead
>> returns information about the first enabled quotatype.  It just so
>> happens that quota-tools /also/ had a bug where it would never send the
>> user's quota type to the kernel, so nobody noticed the nested herp and
>> derp.
>>
>> The test should probably have mentioned kernel commit 555b2c3da1fc
>> ("quota: honor quota type in Q_XGETQSTAT[V] calls") directly to make
>> this more obvious.
>
> Agreed, and maybe mention the quota-tools commit as well.
I think recording kernel commit and package commit(not only in commit 
message but also in comment at the begin of test case) should become a 
habit when it is a regression test.  It is more friendly to user.

Best Regards
Yang Xu
>
> Thanks,
> Eryu
>
>>
>> --D
>>
>>> Signed-off-by: Anju T Sudhakar<anju@linux.vnet.ibm.com>
>>> ---
>>>   tests/generic/594 | 6 ++++++
>>>   1 file changed, 6 insertions(+)
>>>
>>> diff --git a/tests/generic/594 b/tests/generic/594
>>> index 88f9a0dc..668c1fae 100755
>>> --- a/tests/generic/594
>>> +++ b/tests/generic/594
>>> @@ -6,6 +6,8 @@
>>>   #
>>>   # Test per-type(user, group and project) filesystem quota timers, make sure
>>>   # each of grace time can be set/get properly.
>>> +# For xfs, you need a quota-tools containing commit fdd774bf08e568 for this
>>> +# test to run properly- v4.06 should do.
>>>   #
>>>   . ./common/preamble
>>>   _begin_fstest auto quick quota
>>> @@ -18,9 +20,13 @@ _begin_fstest auto quick quota
>>>   _supported_fs generic
>>>   _require_scratch
>>>   _require_setquota_project
>>> +version=$(setquota -V | head -n1 | cut -d" " -f4)
>>>   # V4 XFS doesn't support to mount project and group quota together
>>>   if [ "$FSTYP" = "xfs" ];then
>>>   	_require_scratch_xfs_crc
>>> +	if [ "$version" \<  "4.06" ];then
>>> +		_notrun "This test requires quota-tools 4.06 or greater"
>>> +	fi
>>>   fi
>>>   _require_quota
>>>
>>> --
>>> 2.24.1
>>>
>
>

Re: [PATCH 4/4] generic/219: Fix test failure on 64k block size

[Theodore Ts'o]

On Mon, Oct 11, 2021 at 07:36:39PM +1100, Dave Chinner wrote:
> On Mon, Oct 11, 2021 at 01:25:52PM +0530, Anju T Sudhakar wrote:
> > Skip this test on 64k, as 4k as taken as the default block size.
> > 
> > Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
> > ---
> > But I have a Query:
> > This test fails on ext4&xfs with 64k block size. Do we need to
> > rephrase this test to incorporate the 64k bs scenario as well?
> > If not we can skip this test.
> 
> Right, we should fix the test to use multiples of block size rather
> than fixed byte sizes as we do for all the other tests that need to
> work across different block sizes. e.g. probe the block size after
> mkfs and then convert the "48k" numbers to "$((12 * $bs))".

... and then make sure that you understand what the test is trying to
do, and with those changes, audit the test to make sure it is correct
and valid when the block size is as small as 1k, and as large as 64k.
It might require adjusting other parameters, including the size of the
scratch file system, etc.

Cheers,

					- Ted