[PATCH v3 0/2] two regression tests for ext4

[PATCH v3 0/2] two regression tests for ext4

[Sun Ke]

two regression tests for ext4

Sun Ke (2):
  ext4: resize fs after resize_inode without e2fsck
  ext4: set 256 blocks in a block group then apply io pressure

 tests/ext4/057     | 44 ++++++++++++++++++++++++++++++++++++++++++++
 tests/ext4/057.out |  3 +++
 tests/ext4/058     | 33 +++++++++++++++++++++++++++++++++
 tests/ext4/058.out |  2 ++
 4 files changed, 82 insertions(+)
 create mode 100755 tests/ext4/057
 create mode 100644 tests/ext4/057.out
 create mode 100755 tests/ext4/058
 create mode 100644 tests/ext4/058.out

-- 
2.13.6

[PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Sun Ke]

Forget to run requested e2fsck after resize_inode, then resize fs, it
will trigger off null pointer.

Regression test for commit b55c3cd102a6 ext4: add reserved GDT blocks
check.

Signed-off-by: Sun Ke <sunke32@huawei.com>
---
 tests/ext4/057     | 44 ++++++++++++++++++++++++++++++++++++++++++++
 tests/ext4/057.out |  3 +++
 2 files changed, 47 insertions(+)
 create mode 100755 tests/ext4/057
 create mode 100644 tests/ext4/057.out

diff --git a/tests/ext4/057 b/tests/ext4/057
new file mode 100755
index 00000000..44dae76c
--- /dev/null
+++ b/tests/ext4/057
@@ -0,0 +1,44 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2022 HUAWEI.  All Rights Reserved.
+#
+# FS QA Test 057
+#
+# Forget to run requested e2fsck after resize_inode, then resize fs,
+# it will trigger off null pointer.
+#
+# Regression test for commit
+# b55c3cd102a6 ext4: add reserved GDT blocks check
+#
+. ./common/preamble
+_begin_fstest auto resize quick
+
+# real QA test starts here
+
+# Modify as appropriate.
+_supported_fs ext4
+_fixed_by_kernel_commit b55c3cd102a6 \
+	"ext4: add reserved GDT blocks check"
+
+_require_scratch
+_require_command "$TUNE2FS_PROG" tune2fs
+_require_command "$RESIZE2FS_PROG" resize2fs
+_require_scratch_size $((1024 * 1024)) #kB
+
+# set fs size 512M
+dev_size=$((512 * 1024 * 1024))
+_scratch_mkfs_sized $dev_size >> $seqres.full 2>&1
+
+# forget to run requested e2fsck after resize_inode
+$TUNE2FS_PROG -O ^resize_inode $SCRATCH_DEV | grep -w "e2fsck"
+
+_scratch_mount
+
+# resize fs will trigger NULL pointer in ext4_flex_group_add
+$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1
+
+echo "Silence is golden"
+
+# success, all done
+status=0
+exit
diff --git a/tests/ext4/057.out b/tests/ext4/057.out
new file mode 100644
index 00000000..4784ad7e
--- /dev/null
+++ b/tests/ext4/057.out
@@ -0,0 +1,3 @@
+QA output created by 057
+Please run e2fsck -f on the filesystem.
+Silence is golden
-- 
2.13.6

[PATCH v3 2/2] ext4: set 256 blocks in a block group then apply io pressure

[Sun Ke]

Set 256 blocks in a block group, then inject I/O pressure, it will
trigger off kernel BUG in ext4_mb_mark_diskspace_used.

Regression test for commit a08f789d2ab5 ext4: fix bug_on
ext4_mb_use_inode_pa.

Signed-off-by: Sun Ke <sunke32@huawei.com>
---
 tests/ext4/058     | 33 +++++++++++++++++++++++++++++++++
 tests/ext4/058.out |  2 ++
 2 files changed, 35 insertions(+)
 create mode 100755 tests/ext4/058
 create mode 100644 tests/ext4/058.out

diff --git a/tests/ext4/058 b/tests/ext4/058
new file mode 100755
index 00000000..ddc96f30
--- /dev/null
+++ b/tests/ext4/058
@@ -0,0 +1,33 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2022 HUAWEI.  All Rights Reserved.
+#
+# FS QA Test 058
+#
+# Set 256 blocks in a block group, then inject I/O pressure,
+# it will trigger off kernel BUG in ext4_mb_mark_diskspace_used
+#
+# Regression test for commit
+# a08f789d2ab5 ext4: fix bug_on ext4_mb_use_inode_pa 
+#
+. ./common/preamble
+_begin_fstest auto quick
+
+# real QA test starts here
+
+_supported_fs ext4
+_fixed_by_kernel_commit a08f789d2ab5 \
+	"ext4: fix bug_on ext4_mb_use_inode_pa"
+_require_scratch
+
+# set 256 blocks in a block group
+_scratch_mkfs -g 256 >> $seqres.full 2>&1 || _fail "mkfs failed"
+_scratch_mount
+
+$FSSTRESS_PROG -d $SCRATCH_MNT/stress -n 1000 >> $seqres.full 2>&1
+
+echo "Silence is golden"
+
+# success, all done
+status=0
+exit
diff --git a/tests/ext4/058.out b/tests/ext4/058.out
new file mode 100644
index 00000000..fb5ca60b
--- /dev/null
+++ b/tests/ext4/058.out
@@ -0,0 +1,2 @@
+QA output created by 058
+Silence is golden
-- 
2.13.6

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Zorro Lang]

On Wed, Jul 13, 2022 at 05:28:58PM +0800, Sun Ke wrote:
> Forget to run requested e2fsck after resize_inode, then resize fs, it
> will trigger off null pointer.
> 
> Regression test for commit b55c3cd102a6 ext4: add reserved GDT blocks
> check.
> 
> Signed-off-by: Sun Ke <sunke32@huawei.com>
> ---
>  tests/ext4/057     | 44 ++++++++++++++++++++++++++++++++++++++++++++
>  tests/ext4/057.out |  3 +++
>  2 files changed, 47 insertions(+)
>  create mode 100755 tests/ext4/057
>  create mode 100644 tests/ext4/057.out
> 
> diff --git a/tests/ext4/057 b/tests/ext4/057
> new file mode 100755
> index 00000000..44dae76c
> --- /dev/null
> +++ b/tests/ext4/057
> @@ -0,0 +1,44 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2022 HUAWEI.  All Rights Reserved.
> +#
> +# FS QA Test 057
> +#
> +# Forget to run requested e2fsck after resize_inode, then resize fs,
> +# it will trigger off null pointer.
> +#
> +# Regression test for commit
> +# b55c3cd102a6 ext4: add reserved GDT blocks check
> +#
> +. ./common/preamble
> +_begin_fstest auto resize quick
> +
> +# real QA test starts here
> +
> +# Modify as appropriate.
> +_supported_fs ext4
> +_fixed_by_kernel_commit b55c3cd102a6 \
> +	"ext4: add reserved GDT blocks check"
> +
> +_require_scratch
> +_require_command "$TUNE2FS_PROG" tune2fs
> +_require_command "$RESIZE2FS_PROG" resize2fs
> +_require_scratch_size $((1024 * 1024)) #kB
> +
> +# set fs size 512M
> +dev_size=$((512 * 1024 * 1024))
> +_scratch_mkfs_sized $dev_size >> $seqres.full 2>&1
> +
> +# forget to run requested e2fsck after resize_inode
> +$TUNE2FS_PROG -O ^resize_inode $SCRATCH_DEV | grep -w "e2fsck"
> +
> +_scratch_mount
> +
> +# resize fs will trigger NULL pointer in ext4_flex_group_add
> +$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1
> +
> +echo "Silence is golden"
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/ext4/057.out b/tests/ext4/057.out
> new file mode 100644
> index 00000000..4784ad7e
> --- /dev/null
> +++ b/tests/ext4/057.out
> @@ -0,0 +1,3 @@
> +QA output created by 057
> +Please run e2fsck -f on the filesystem.

If you hope to match this line, means this case isn't "Silence is golden".

I don't know why you'd to have this line, it looks not suit to be golden
image. If you'd like to make sure current ext4 supports "resize_inode"
feature, you can use:
  _require_scratch_ext4_feature resize_inode

Thanks,
Zorro

> +Silence is golden
> -- 
> 2.13.6
> 

Re: [PATCH v3 2/2] ext4: set 256 blocks in a block group then apply io pressure

[Zorro Lang]

On Wed, Jul 13, 2022 at 05:28:59PM +0800, Sun Ke wrote:
> Set 256 blocks in a block group, then inject I/O pressure, it will
> trigger off kernel BUG in ext4_mb_mark_diskspace_used.
> 
> Regression test for commit a08f789d2ab5 ext4: fix bug_on
> ext4_mb_use_inode_pa.
> 
> Signed-off-by: Sun Ke <sunke32@huawei.com>
> ---

This version looks good to me. I'll merge it if there's not objection
from ext4 folks.

Reviewed-by: Zorro Lang <zlang@kernel.org>

>  tests/ext4/058     | 33 +++++++++++++++++++++++++++++++++
>  tests/ext4/058.out |  2 ++
>  2 files changed, 35 insertions(+)
>  create mode 100755 tests/ext4/058
>  create mode 100644 tests/ext4/058.out
> 
> diff --git a/tests/ext4/058 b/tests/ext4/058
> new file mode 100755
> index 00000000..ddc96f30
> --- /dev/null
> +++ b/tests/ext4/058
> @@ -0,0 +1,33 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2022 HUAWEI.  All Rights Reserved.
> +#
> +# FS QA Test 058
> +#
> +# Set 256 blocks in a block group, then inject I/O pressure,
> +# it will trigger off kernel BUG in ext4_mb_mark_diskspace_used
> +#
> +# Regression test for commit
> +# a08f789d2ab5 ext4: fix bug_on ext4_mb_use_inode_pa 
> +#
> +. ./common/preamble
> +_begin_fstest auto quick
> +
> +# real QA test starts here
> +
> +_supported_fs ext4
> +_fixed_by_kernel_commit a08f789d2ab5 \
> +	"ext4: fix bug_on ext4_mb_use_inode_pa"
> +_require_scratch
> +
> +# set 256 blocks in a block group
> +_scratch_mkfs -g 256 >> $seqres.full 2>&1 || _fail "mkfs failed"
> +_scratch_mount
> +
> +$FSSTRESS_PROG -d $SCRATCH_MNT/stress -n 1000 >> $seqres.full 2>&1
> +
> +echo "Silence is golden"
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/ext4/058.out b/tests/ext4/058.out
> new file mode 100644
> index 00000000..fb5ca60b
> --- /dev/null
> +++ b/tests/ext4/058.out
> @@ -0,0 +1,2 @@
> +QA output created by 058
> +Silence is golden
> -- 
> 2.13.6
> 

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Theodore Ts'o]

On Thu, Jul 14, 2022 at 11:46:07PM +0800, Zorro Lang wrote:
> On Wed, Jul 13, 2022 at 05:28:58PM +0800, Sun Ke wrote:
> > +
> > +# forget to run requested e2fsck after resize_inode
> > +$TUNE2FS_PROG -O ^resize_inode $SCRATCH_DEV | grep -w "e2fsck"
> > +
> > +_scratch_mount
> > +
> > +# resize fs will trigger NULL pointer in ext4_flex_group_add
> > +$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1
> > +
> > +echo "Silence is golden"
  ...
> > diff --git a/tests/ext4/057.out b/tests/ext4/057.out
> > new file mode 100644
> > index 00000000..4784ad7e
> > --- /dev/null
> > +++ b/tests/ext4/057.out
> > @@ -0,0 +1,3 @@
> > +QA output created by 057
> > +Please run e2fsck -f on the filesystem.
> 
> If you hope to match this line, means this case isn't "Silence is golden".
> 
> I don't know why you'd to have this line, it looks not suit to be golden
> image. If you'd like to make sure current ext4 supports "resize_inode"
> feature, you can use:
>   _require_scratch_ext4_feature resize_inode

That's not the problem.

The "tune2fs -O ^resize_inode" command is printing that message as a
reminder that it would be a Really Good idea to run e2fsck on the file
system, because tune2fs doesn't completely remove the resize inode
after turning off that feature.

The commit which this test is trying to verify is that the kernel
won't oops if the system adminsitrator ignores the rather explicit
request:

Please run e2fsck -f on the filesystem.

... and blithely mounts the file system without running fsck -f on the
file system first.  While it could be argued that a system
administrator which fails to follow instructions deserves everything
they get, we decided the as a quality of implementation issue, it
would be better if the kernel didn't dereference a NULL pointer in
that case.  :-)

The one thing I'll note is that it is possible that at some point in
the future, tune2fs could be improved so that it cleanly removes the
resize_inode when the resize inode feature is removed, so that running
"fsck.ext4 -f" is no longer necessary.  So if you want to future-proof
the test so it doesn't fail once tune2fs is made more idiot-proof, it
might be better if the test did something like this:

mke2fs -t ext4 -O ^resize_inode /dev/vdc 512m
debugfs -w -R "set_super_value s_reserved_gdt_blocks 100" /dev/vdc
mount -t ext4 /dev/vdc /vdc
resize2fs /dev/vdc 1G

Translating the above from commands suitable for manual trial using
"kvm-xfstests shell" to a proper xfstests script is left as an
exercise for the reader.  :-)

					- Ted

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Zorro Lang]

On Thu, Jul 14, 2022 at 06:00:34PM -0400, Theodore Ts'o wrote:
> On Thu, Jul 14, 2022 at 11:46:07PM +0800, Zorro Lang wrote:
> > On Wed, Jul 13, 2022 at 05:28:58PM +0800, Sun Ke wrote:
> > > +
> > > +# forget to run requested e2fsck after resize_inode
> > > +$TUNE2FS_PROG -O ^resize_inode $SCRATCH_DEV | grep -w "e2fsck"
> > > +
> > > +_scratch_mount
> > > +
> > > +# resize fs will trigger NULL pointer in ext4_flex_group_add
> > > +$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1
> > > +
> > > +echo "Silence is golden"
>   ...
> > > diff --git a/tests/ext4/057.out b/tests/ext4/057.out
> > > new file mode 100644
> > > index 00000000..4784ad7e
> > > --- /dev/null
> > > +++ b/tests/ext4/057.out
> > > @@ -0,0 +1,3 @@
> > > +QA output created by 057
> > > +Please run e2fsck -f on the filesystem.
> > 
> > If you hope to match this line, means this case isn't "Silence is golden".
> > 
> > I don't know why you'd to have this line, it looks not suit to be golden
> > image. If you'd like to make sure current ext4 supports "resize_inode"
> > feature, you can use:
> >   _require_scratch_ext4_feature resize_inode
> 
> That's not the problem.
> 
> The "tune2fs -O ^resize_inode" command is printing that message as a
> reminder that it would be a Really Good idea to run e2fsck on the file
> system, because tune2fs doesn't completely remove the resize inode
> after turning off that feature.
> 
> The commit which this test is trying to verify is that the kernel
> won't oops if the system adminsitrator ignores the rather explicit
> request:
> 
> Please run e2fsck -f on the filesystem.
> 
> ... and blithely mounts the file system without running fsck -f on the
> file system first.  While it could be argued that a system
> administrator which fails to follow instructions deserves everything
> they get, we decided the as a quality of implementation issue, it
> would be better if the kernel didn't dereference a NULL pointer in
> that case.  :-)
> 
> The one thing I'll note is that it is possible that at some point in
> the future, tune2fs could be improved so that it cleanly removes the
> resize_inode when the resize inode feature is removed, so that running
> "fsck.ext4 -f" is no longer necessary.  So if you want to future-proof

Good to know :)

> the test so it doesn't fail once tune2fs is made more idiot-proof, it
> might be better if the test did something like this:
> 
> mke2fs -t ext4 -O ^resize_inode /dev/vdc 512m
> debugfs -w -R "set_super_value s_reserved_gdt_blocks 100" /dev/vdc

So make sure there're reserved GDT blocks, even if disable resize_inode
feature.

> mount -t ext4 /dev/vdc /vdc
> resize2fs /dev/vdc 1G

Thanks Ted! That's really helpful to get review points from ext4 expert.

Hi Ke, would you mind re-sending this case refer to above review points?
You can refer to below code, but I didn't test it, so please test and make
sure it works and can reproduce the bug. Feel free to improve it if something
wrong.

_require_command "$DEBUGFS_PROG" debugfs
...

MKFS_OPTIONS="-O ^resize_inode $MKFS_OPTIONS" _scratch_mkfs_sized $dev_size \
	>>$seqres.full 2>&1 || _fail "mkfs failed"
$DEBUGFS_PROG -w -R "set_super_value s_reserved_gdt_blocks 100" $SCRATCH_DEV \
	>>$seqres.full 2>&1
$DEBUGFS_PROG -R "show_super_stats -h" $SCRATCH_DEV 2>/dev/null | \
	grep "Reserved GDT blocks"
_scratch_mount
$RESIZE2FS_PROG $SCRATCH_DEV 1g >> $seqres.full 2>&1


Thanks,
Zorro


> 
> Translating the above from commands suitable for manual trial using
> "kvm-xfstests shell" to a proper xfstests script is left as an
> exercise for the reader.  :-)
> 
> 					- Ted
> 

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Sun Ke]

在 2022/7/16 2:08, Zorro Lang 写道:
> On Thu, Jul 14, 2022 at 06:00:34PM -0400, Theodore Ts'o wrote:
>> On Thu, Jul 14, 2022 at 11:46:07PM +0800, Zorro Lang wrote:
>>> On Wed, Jul 13, 2022 at 05:28:58PM +0800, Sun Ke wrote:
>>>> +
>>>> +# forget to run requested e2fsck after resize_inode
>>>> +$TUNE2FS_PROG -O ^resize_inode $SCRATCH_DEV | grep -w "e2fsck"
>>>> +
>>>> +_scratch_mount
>>>> +
>>>> +# resize fs will trigger NULL pointer in ext4_flex_group_add
>>>> +$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1
>>>> +
>>>> +echo "Silence is golden"
>>    ...
>>>> diff --git a/tests/ext4/057.out b/tests/ext4/057.out
>>>> new file mode 100644
>>>> index 00000000..4784ad7e
>>>> --- /dev/null
>>>> +++ b/tests/ext4/057.out
>>>> @@ -0,0 +1,3 @@
>>>> +QA output created by 057
>>>> +Please run e2fsck -f on the filesystem.
>>>
>>> If you hope to match this line, means this case isn't "Silence is golden".
>>>
>>> I don't know why you'd to have this line, it looks not suit to be golden
>>> image. If you'd like to make sure current ext4 supports "resize_inode"
>>> feature, you can use:
>>>    _require_scratch_ext4_feature resize_inode
>>
>> That's not the problem.
>>
>> The "tune2fs -O ^resize_inode" command is printing that message as a
>> reminder that it would be a Really Good idea to run e2fsck on the file
>> system, because tune2fs doesn't completely remove the resize inode
>> after turning off that feature.
>>
>> The commit which this test is trying to verify is that the kernel
>> won't oops if the system adminsitrator ignores the rather explicit
>> request:
>>
>> Please run e2fsck -f on the filesystem.
>>
>> ... and blithely mounts the file system without running fsck -f on the
>> file system first.  While it could be argued that a system
>> administrator which fails to follow instructions deserves everything
>> they get, we decided the as a quality of implementation issue, it
>> would be better if the kernel didn't dereference a NULL pointer in
>> that case.  :-)
>>
>> The one thing I'll note is that it is possible that at some point in
>> the future, tune2fs could be improved so that it cleanly removes the
>> resize_inode when the resize inode feature is removed, so that running
>> "fsck.ext4 -f" is no longer necessary.  So if you want to future-proof
> 
> Good to know :)
> 
>> the test so it doesn't fail once tune2fs is made more idiot-proof, it
>> might be better if the test did something like this:
>>
>> mke2fs -t ext4 -O ^resize_inode /dev/vdc 512m
>> debugfs -w -R "set_super_value s_reserved_gdt_blocks 100" /dev/vdc
> 
> So make sure there're reserved GDT blocks, even if disable resize_inode
> feature.
> 
>> mount -t ext4 /dev/vdc /vdc
>> resize2fs /dev/vdc 1G
> 
> Thanks Ted! That's really helpful to get review points from ext4 expert.
> 
> Hi Ke, would you mind re-sending this case refer to above review points?
> You can refer to below code, but I didn't test it, so please test and make
> sure it works and can reproduce the bug. Feel free to improve it if something
> wrong.

Glad to do that, it can reproduce the bug.

Thanks,
Sun Ke
> 
> _require_command "$DEBUGFS_PROG" debugfs
> ...
> 
> MKFS_OPTIONS="-O ^resize_inode $MKFS_OPTIONS" _scratch_mkfs_sized $dev_size \
> 	>>$seqres.full 2>&1 || _fail "mkfs failed"
> $DEBUGFS_PROG -w -R "set_super_value s_reserved_gdt_blocks 100" $SCRATCH_DEV \
> 	>>$seqres.full 2>&1
> $DEBUGFS_PROG -R "show_super_stats -h" $SCRATCH_DEV 2>/dev/null | \
> 	grep "Reserved GDT blocks"
> _scratch_mount
> $RESIZE2FS_PROG $SCRATCH_DEV 1g >> $seqres.full 2>&1
> 
> 
> Thanks,
> Zorro
> 
> 
>>
>> Translating the above from commands suitable for manual trial using
>> "kvm-xfstests shell" to a proper xfstests script is left as an
>> exercise for the reader.  :-)
>>
>> 					- Ted
>>
> 
> .
> 

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Sun Ke]

在 2022/7/16 2:08, Zorro Lang 写道:
> On Thu, Jul 14, 2022 at 06:00:34PM -0400, Theodore Ts'o wrote:
>> On Thu, Jul 14, 2022 at 11:46:07PM +0800, Zorro Lang wrote:
>>> On Wed, Jul 13, 2022 at 05:28:58PM +0800, Sun Ke wrote:
>>>> +
>>>> +# forget to run requested e2fsck after resize_inode
>>>> +$TUNE2FS_PROG -O ^resize_inode $SCRATCH_DEV | grep -w "e2fsck"
>>>> +
>>>> +_scratch_mount
>>>> +
>>>> +# resize fs will trigger NULL pointer in ext4_flex_group_add
>>>> +$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1
>>>> +
>>>> +echo "Silence is golden"
>>    ...
>>>> diff --git a/tests/ext4/057.out b/tests/ext4/057.out
>>>> new file mode 100644
>>>> index 00000000..4784ad7e
>>>> --- /dev/null
>>>> +++ b/tests/ext4/057.out
>>>> @@ -0,0 +1,3 @@
>>>> +QA output created by 057
>>>> +Please run e2fsck -f on the filesystem.
>>>
>>> If you hope to match this line, means this case isn't "Silence is golden".
>>>
>>> I don't know why you'd to have this line, it looks not suit to be golden
>>> image. If you'd like to make sure current ext4 supports "resize_inode"
>>> feature, you can use:
>>>    _require_scratch_ext4_feature resize_inode
>>
>> That's not the problem.
>>
>> The "tune2fs -O ^resize_inode" command is printing that message as a
>> reminder that it would be a Really Good idea to run e2fsck on the file
>> system, because tune2fs doesn't completely remove the resize inode
>> after turning off that feature.
>>
>> The commit which this test is trying to verify is that the kernel
>> won't oops if the system adminsitrator ignores the rather explicit
>> request:
>>
>> Please run e2fsck -f on the filesystem.
>>
>> ... and blithely mounts the file system without running fsck -f on the
>> file system first.  While it could be argued that a system
>> administrator which fails to follow instructions deserves everything
>> they get, we decided the as a quality of implementation issue, it
>> would be better if the kernel didn't dereference a NULL pointer in
>> that case.  :-)
>>
>> The one thing I'll note is that it is possible that at some point in
>> the future, tune2fs could be improved so that it cleanly removes the
>> resize_inode when the resize inode feature is removed, so that running
>> "fsck.ext4 -f" is no longer necessary.  So if you want to future-proof
> 
> Good to know :)
> 
>> the test so it doesn't fail once tune2fs is made more idiot-proof, it
>> might be better if the test did something like this:
>>
>> mke2fs -t ext4 -O ^resize_inode /dev/vdc 512m
>> debugfs -w -R "set_super_value s_reserved_gdt_blocks 100" /dev/vdc
> 
> So make sure there're reserved GDT blocks, even if disable resize_inode
> feature.
> 
>> mount -t ext4 /dev/vdc /vdc
>> resize2fs /dev/vdc 1G
> 
> Thanks Ted! That's really helpful to get review points from ext4 expert.
> 
> Hi Ke, would you mind re-sending this case refer to above review points?
> You can refer to below code, but I didn't test it, so please test and make
> sure it works and can reproduce the bug. Feel free to improve it if something
> wrong.
> 
> _require_command "$DEBUGFS_PROG" debugfs
> ...
> 
> MKFS_OPTIONS="-O ^resize_inode $MKFS_OPTIONS" _scratch_mkfs_sized $dev_size \
> 	>>$seqres.full 2>&1 || _fail "mkfs failed"
> $DEBUGFS_PROG -w -R "set_super_value s_reserved_gdt_blocks 100" $SCRATCH_DEV \
> 	>>$seqres.full 2>&1
> $DEBUGFS_PROG -R "show_super_stats -h" $SCRATCH_DEV 2>/dev/null | \
> 	grep "Reserved GDT blocks"
> _scratch_mount
> $RESIZE2FS_PROG $SCRATCH_DEV 1g >> $seqres.full 2>&1
> 
> 
> Thanks,
> Zorro

two questions.

1. The test run fsck automatically, and complain inconsistent，I think 
it need not run fsck.
2. It warn missing kernel fix, but the commit had merged.

##test log##
[root@testvm xfstests-dev]# ./check tests/ext4/057
FSTYP         -- ext4
PLATFORM      -- Linux/x86_64 testvm 5.10.0-00009-g08870cf6d88a #35 SMP 
Fri Jul 22 15:50:00 CST 2022
MKFS_OPTIONS  -- /dev/sdb
MOUNT_OPTIONS -- -o acl,user_xattr /dev/sdb /mnt/scratch

ext4/057 5s ... _check_generic_filesystem: filesystem on /dev/sdb is 
inconsistent
(see /root/xfstests-dev/results//ext4/057.full for details)


HINT: You _MAY_ be missing kernel fix:
       b55c3cd102a6 ext4: add reserved GDT blocks check

Ran: ext4/057
Failures: ext4/057
Failed 1 of 1 tests



##test##
# Modify as appropriate.
_supported_fs ext4
_fixed_by_kernel_commit b55c3cd102a6 \
     "ext4: add reserved GDT blocks check"

_require_scratch
_require_command "$TUNE2FS_PROG" tune2fs
_require_command "$RESIZE2FS_PROG" resize2fs
_require_command "$DEBUGFS_PROG" debugfs
_require_scratch_size $((1024 * 1024)) #kB

# set fs size 512M
dev_size=$((512 * 1024 * 1024))
MKFS_OPTIONS="-O ^resize_inode $MKFS_OPTIONS" _scratch_mkfs_sized 
$dev_size \
     >>$seqres.full 2>&1 || _fail "mkfs failed"

$DEBUGFS_PROG -w -R "set_super_value s_reserved_gdt_blocks 100" 
$SCRATCH_DEV \
     >>$seqres.full 2>&1

$DEBUGFS_PROG -R "show_super_stats -h" $SCRATCH_DEV 2>/dev/null | \
     grep "Reserved GDT blocks"

_scratch_mount

# resize fs will trigger NULL pointer in ext4_flex_group_add
$RESIZE2FS_PROG $SCRATCH_DEV 1G >> $seqres.full 2>&1


> 
> 
>>
>> Translating the above from commands suitable for manual trial using
>> "kvm-xfstests shell" to a proper xfstests script is left as an
>> exercise for the reader.  :-)
>>
>> 					- Ted
>>
> 
> .
> 

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Theodore Ts'o]

On Fri, Jul 22, 2022 at 04:16:26PM +0800, Sun Ke wrote:
> 
> 1. The test run fsck automatically, and complain inconsistent，I think it
> need not run fsck.

The check script always run fscks after the test.  In order to
suppress the fsck complaint, we'll need to add this after the
resize2fs invocation:

_scratch_unmount
$DEBUGFS_PROG -w -R "set_super_value s_reserved_gdt_blocks 0" $SCRATCH_DEV \
        >>$seqres.full 2>&1

This resets the s_reserved_gdt_blocks field back to zero, so the fsck
doesn't fail.  Which is fine, because the point of the test is to see
whether the kernel dereferences a NULL pointer or not.

> 2. It warn missing kernel fix, but the commit had merged.

The way _fixed_by_kernel_commit works is if the test fails (for any
reason), it prints that you MAY be missing the bugfix commit:

> HINT: You _MAY_ be missing kernel fix:
>       b55c3cd102a6 ext4: add reserved GDT blocks check

		     	       		- Ted

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Zorro Lang]

On Fri, Jul 22, 2022 at 07:51:57AM -0400, Theodore Ts'o wrote:
> On Fri, Jul 22, 2022 at 04:16:26PM +0800, Sun Ke wrote:
> > 
> > 1. The test run fsck automatically, and complain inconsistent，I think it
> > need not run fsck.
> 
> The check script always run fscks after the test.  In order to
> suppress the fsck complaint, we'll need to add this after the
> resize2fs invocation:
> 
> _scratch_unmount
> $DEBUGFS_PROG -w -R "set_super_value s_reserved_gdt_blocks 0" $SCRATCH_DEV \
>         >>$seqres.full 2>&1
> 
> This resets the s_reserved_gdt_blocks field back to zero, so the fsck
> doesn't fail.  Which is fine, because the point of the test is to see
> whether the kernel dereferences a NULL pointer or not.

Or maybe just replace _reuqire_scratch with _require_scratch_nocheck, if we
corrupt the fs on SCRATCH_DEV intentionally?

Thanks,
Zorro

> 
> > 2. It warn missing kernel fix, but the commit had merged.
> 
> The way _fixed_by_kernel_commit works is if the test fails (for any
> reason), it prints that you MAY be missing the bugfix commit:
> 
> > HINT: You _MAY_ be missing kernel fix:
> >       b55c3cd102a6 ext4: add reserved GDT blocks check
> 
> 		     	       		- Ted
> 

Re: [PATCH v3 1/2] ext4: resize fs after resize_inode without e2fsck

[Theodore Ts'o]

On Fri, Jul 22, 2022 at 11:11:14PM +0800, Zorro Lang wrote:
> > This resets the s_reserved_gdt_blocks field back to zero, so the fsck
> > doesn't fail.  Which is fine, because the point of the test is to see
> > whether the kernel dereferences a NULL pointer or not.
> 
> Or maybe just replace _reuqire_scratch with _require_scratch_nocheck, if we
> corrupt the fs on SCRATCH_DEV intentionally?

Agreed, that would be a better solution.

					- Ted