On Thu, Sep 10, 2020 at 8:51 AM Murphy Zhou <jencce.kernel@xxxxxxxxx> wrote:
>
> When nfs_export is enabled, the link count of upper dir
> objects are more then the expected number in this testcase.
> Because extra index entries are linked to upper inodes.
>
> QA output created by 073
> +Expected link count is 12 but real count is 23, file name is dir
> +Expected link count is 12 but real count is 23, file name is 1
> ...
> +Expected link count is 12 but real count is 23, file name is 10
> Silence is golden
>
> Signed-off-by: Murphy Zhou <jencce.kernel@xxxxxxxxx>
> ---
> Hi folks,
>
> Please help review that with nfs_export enabled, this is expected.
> I think so but I'm not 100% sure about it. Maybe it's a bug in
> the kernel.
>
Indeed this is expected.
With nfs_export, for every unlinked lower file/dir there is a whiteout
"tombstone" in the index directory, whose name is the file handle
of the unlinked inode.
So this test creates extra 11 tombstones with nfs_export enabled.
> Thanks.
>
> tests/overlay/073 | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/tests/overlay/073 b/tests/overlay/073
> index 37860c92..b78551eb 100755
> --- a/tests/overlay/073
> +++ b/tests/overlay/073
> @@ -99,7 +99,7 @@ run_test_case()
> {
> _scratch_mkfs
> make_lower_files ${1}
> - _scratch_mount -o "index=on"
> + _scratch_mount -o "index=on,nfs_export=off"
So the fix looks fine, but let's document why nfs_export=off is needed.
Thanks,
Amir.
