Does btrfs get nlink on directories wrong? -- was Re: [PATCH 2/4] xfstests: Add first statx test [ver #5]

Does btrfs get nlink on directories wrong? -- was Re: [PATCH 2/4] xfstests: Add first statx test [ver #5]

[David Howells]

I've added a test to xfstests that exercises the new statx syscall.  However,
it fails on btrfs:

     Test statx on a directory
    +[!] stx_nlink differs, 1 != 2
    +Failed
    +stat_test failed

because a new directory it creates has an nlink of 1, not 2.  Is this a case
of my making an incorrect assumption or is it an fs bug?

David

Re: Does btrfs get nlink on directories wrong? -- was Re: [PATCH 2/4] xfstests: Add first statx test [ver #5]

[David Sterba]

On Wed, Apr 05, 2017 at 11:53:41AM +0100, David Howells wrote:
> I've added a test to xfstests that exercises the new statx syscall.  However,
> it fails on btrfs:
> 
>      Test statx on a directory
>     +[!] stx_nlink differs, 1 != 2
>     +Failed
>     +stat_test failed
> 
> because a new directory it creates has an nlink of 1, not 2.  Is this a case
> of my making an incorrect assumption or is it an fs bug?

Afaik nlink == 1 means that there's no accounting of subdirectories, and
it's a valid value. The 'find' utility can use nlink to optimize
directory traversal but otherwise I'm not aware of other usage.

All directories in btrfs have nlink == 1.

Re: Does btrfs get nlink on directories wrong? -- was Re: [PATCH 2/4] xfstests: Add first statx test [ver #5]

[Amir Goldstein]

On Wed, Apr 5, 2017 at 3:30 PM, David Sterba <dsterba@suse.cz> wrote:
> On Wed, Apr 05, 2017 at 11:53:41AM +0100, David Howells wrote:
>> I've added a test to xfstests that exercises the new statx syscall.  However,
>> it fails on btrfs:
>>
>>      Test statx on a directory
>>     +[!] stx_nlink differs, 1 != 2
>>     +Failed
>>     +stat_test failed
>>
>> because a new directory it creates has an nlink of 1, not 2.  Is this a case
>> of my making an incorrect assumption or is it an fs bug?
>
> Afaik nlink == 1 means that there's no accounting of subdirectories, and
> it's a valid value. The 'find' utility can use nlink to optimize
> directory traversal but otherwise I'm not aware of other usage.
>
> All directories in btrfs have nlink == 1.

FYI,

Overlayfs uses nlink = 1 for merge dirs to silence 'find' et al.
Ext4 uses nlink = 1 for directories with more than 32K subdirs
(EXT4_FEATURE_RO_COMPAT_DIR_NLINK).

But in both those fs newly created directories will have nlink = 2.

Re: Does btrfs get nlink on directories wrong? -- was Re: [PATCH 2/4] xfstests: Add first statx test [ver #5]

[Eryu Guan]

On Wed, Apr 05, 2017 at 03:32:30PM +0300, Amir Goldstein wrote:
> On Wed, Apr 5, 2017 at 3:30 PM, David Sterba <dsterba@suse.cz> wrote:
> > On Wed, Apr 05, 2017 at 11:53:41AM +0100, David Howells wrote:
> >> I've added a test to xfstests that exercises the new statx syscall.  However,
> >> it fails on btrfs:
> >>
> >>      Test statx on a directory
> >>     +[!] stx_nlink differs, 1 != 2
> >>     +Failed
> >>     +stat_test failed
> >>
> >> because a new directory it creates has an nlink of 1, not 2.  Is this a case
> >> of my making an incorrect assumption or is it an fs bug?
> >
> > Afaik nlink == 1 means that there's no accounting of subdirectories, and
> > it's a valid value. The 'find' utility can use nlink to optimize
> > directory traversal but otherwise I'm not aware of other usage.
> >
> > All directories in btrfs have nlink == 1.
> 
> FYI,
> 
> Overlayfs uses nlink = 1 for merge dirs to silence 'find' et al.
> Ext4 uses nlink = 1 for directories with more than 32K subdirs
> (EXT4_FEATURE_RO_COMPAT_DIR_NLINK).
> 
> But in both those fs newly created directories will have nlink = 2.

Is there a conclusion on this? Seems the test should be updated
accordingly?

Thanks,
Eryu

Re: Does btrfs get nlink on directories wrong? -- was Re: [PATCH 2/4] xfstests: Add first statx test [ver #5]

[David Howells]

Eryu Guan <eguan@redhat.com> wrote:

> > Overlayfs uses nlink = 1 for merge dirs to silence 'find' et al.
> > Ext4 uses nlink = 1 for directories with more than 32K subdirs
> > (EXT4_FEATURE_RO_COMPAT_DIR_NLINK).
> > 
> > But in both those fs newly created directories will have nlink = 2.
> 
> Is there a conclusion on this? Seems the test should be updated
> accordingly?

I've dropped the nlink check on directories.

David