[PATCH] ext4/271: Fix test failure with data=<foo> mount option

[PATCH] ext4/271: Fix test failure with data=<foo> mount option

[Jan Kara]

The test is failing when MOUNT_OPTIONS contain 'data=<foo>' mount option
as it does not combine with 'noload' mount option the test uses. Fix it
by removing the data=<foo> mount option. Arguably we could also fix the
problem by just skipping the test when such option is set but e.g. using
'data=writeback' is common to excercise somewhat different behavior and
it is undesirable to skip the test just because of this.

Signed-off-by: Jan Kara <jack@suse.cz>
---
 tests/ext4/271 | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/tests/ext4/271 b/tests/ext4/271
index d68c2719024e..417a796ffac9 100755
--- a/tests/ext4/271
+++ b/tests/ext4/271
@@ -45,6 +45,10 @@ _scratch_mkfs_sized $((128 * 1024 * 1024)) >> $seqres.full 2>&1
 
 # -onoload and EXT4_SYNC_FL on file is important becase result in
 # metadata sync writes inside ext4_handle_dirty_metadata()
+#
+# We have to be careful and remove data=<foo> mount options as they don't
+# combine with 'noload'.
+MOUNT_OPTIONS=$(echo $MOUNT_OPTIONS | sed -e 's/\([ ,]\)data=[a-zA-Z]*[,$]/\1/')
 _scratch_mount -onoload
 touch $SCRATCH_MNT/file
 $CHATTR_PROG +S $SCRATCH_MNT/file
-- 
2.6.6

Re: [PATCH] ext4/271: Fix test failure with data=<foo> mount option

[Eryu Guan]

Hi Jan,

On Thu, Jun 16, 2016 at 12:05:11PM +0200, Jan Kara wrote:
> The test is failing when MOUNT_OPTIONS contain 'data=<foo>' mount option
> as it does not combine with 'noload' mount option the test uses. Fix it
> by removing the data=<foo> mount option. Arguably we could also fix the
> problem by just skipping the test when such option is set but e.g. using
> 'data=writeback' is common to excercise somewhat different behavior and

When MOUNT_OPTIONS contains 'data=writeback' ext4/271 also fails

[275528.681541] EXT4-fs (sda6): can't mount with data=, fs mounted w/o journal

Seems kernel commit 1e381f60dad9 ("ext4: do not allow journal_opts for
fs w/o journal") refuses all mounts with "data=" mount option.

> it is undesirable to skip the test just because of this.

So skipping ext4/271 in this case makes more sense to me.

> 
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---
>  tests/ext4/271 | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/tests/ext4/271 b/tests/ext4/271
> index d68c2719024e..417a796ffac9 100755
> --- a/tests/ext4/271
> +++ b/tests/ext4/271
> @@ -45,6 +45,10 @@ _scratch_mkfs_sized $((128 * 1024 * 1024)) >> $seqres.full 2>&1
>  
>  # -onoload and EXT4_SYNC_FL on file is important becase result in
>  # metadata sync writes inside ext4_handle_dirty_metadata()
> +#
> +# We have to be careful and remove data=<foo> mount options as they don't
> +# combine with 'noload'.
> +MOUNT_OPTIONS=$(echo $MOUNT_OPTIONS | sed -e 's/\([ ,]\)data=[a-zA-Z]*[,$]/\1/')

This doesn't remove 'data=<foo>' mount option for me

  $ MKFS_OPTIONS="-o data=journal"
  $ echo $MKFS_OPTIONS | sed -e 's/\([ ,]\)data=[a-zA-Z]*[,$]/\1/'
  -o data=journal

Even with that works, the 'sed' filter only removes the 'data=<foo>'
part and leaves the '-o' part, which results in MOUNT_OPTIONS="-o" and
still causes mount to fail.

And perhaps other journal related mount options should be removed too?
e.g. journal_checksum, journal_async_commit, commit=<n>

I sent two patches back in Jan. to fix ext4/271 (skip ext4/271 when
MOUNT_OPTIONS contain journal related options), what do you think?

https://patchwork.kernel.org/patch/8052251/
https://patchwork.kernel.org/patch/8052261/

Thanks,
Eryu

Re: [PATCH] ext4/271: Fix test failure with data=<foo> mount option

[Jan Kara]

Hi Eryu,

On Mon 20-06-16 17:25:41, Eryu Guan wrote:
> On Thu, Jun 16, 2016 at 12:05:11PM +0200, Jan Kara wrote:
> > The test is failing when MOUNT_OPTIONS contain 'data=<foo>' mount option
> > as it does not combine with 'noload' mount option the test uses. Fix it
> > by removing the data=<foo> mount option. Arguably we could also fix the
> > problem by just skipping the test when such option is set but e.g. using
> > 'data=writeback' is common to excercise somewhat different behavior and
> 
> When MOUNT_OPTIONS contains 'data=writeback' ext4/271 also fails
> 
> [275528.681541] EXT4-fs (sda6): can't mount with data=, fs mounted w/o journal
> 
> Seems kernel commit 1e381f60dad9 ("ext4: do not allow journal_opts for
> fs w/o journal") refuses all mounts with "data=" mount option.
> 
> > it is undesirable to skip the test just because of this.
> 
> So skipping ext4/271 in this case makes more sense to me.

OK, I didn't think of other journal options. Yes, so skipping the test
probably makes more sense then.

> > Signed-off-by: Jan Kara <jack@suse.cz>
> > ---
> >  tests/ext4/271 | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/tests/ext4/271 b/tests/ext4/271
> > index d68c2719024e..417a796ffac9 100755
> > --- a/tests/ext4/271
> > +++ b/tests/ext4/271
> > @@ -45,6 +45,10 @@ _scratch_mkfs_sized $((128 * 1024 * 1024)) >> $seqres.full 2>&1
> >  
> >  # -onoload and EXT4_SYNC_FL on file is important becase result in
> >  # metadata sync writes inside ext4_handle_dirty_metadata()
> > +#
> > +# We have to be careful and remove data=<foo> mount options as they don't
> > +# combine with 'noload'.
> > +MOUNT_OPTIONS=$(echo $MOUNT_OPTIONS | sed -e 's/\([ ,]\)data=[a-zA-Z]*[,$]/\1/')
> 
> This doesn't remove 'data=<foo>' mount option for me
> 
>   $ MKFS_OPTIONS="-o data=journal"
>   $ echo $MKFS_OPTIONS | sed -e 's/\([ ,]\)data=[a-zA-Z]*[,$]/\1/'
>   -o data=journal
> 
> Even with that works, the 'sed' filter only removes the 'data=<foo>'
> part and leaves the '-o' part, which results in MOUNT_OPTIONS="-o" and
> still causes mount to fail.
> 
> And perhaps other journal related mount options should be removed too?
> e.g. journal_checksum, journal_async_commit, commit=<n>
> 
> I sent two patches back in Jan. to fix ext4/271 (skip ext4/271 when
> MOUNT_OPTIONS contain journal related options), what do you think?
> 
> https://patchwork.kernel.org/patch/8052251/
> https://patchwork.kernel.org/patch/8052261/

OK, those two patches are fine with me. Why didn't they get merged? Care to
resend?

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR