[lsb-discuss] xfs fileystem fails tests passed by ext3 and reiserfs?

[Stew Benedict]

On Tue, 4 Mar 2008 jrc at skylon.demon.co.uk wrote:

> I have run a series of comparison 3.1.1 tests using the reiserfs,
> ext3, and xfs filesystems hosted on a SuSE SLES 10 SP1 system.  There
> are two test purposes that pass for ext3 and reiserfs but fail for xfs.
> 
> Here is an excerpt from the journal of the xfs run:
> 
> 10|77 /tset/POSIX.os/ioprim/write/T.write 13:52:33|TC Start, scenario ref 80-0
> 15|77 3.6-lite 28|TCM Start
> [...]
> 400|77 15 1 13:53:34|IC Start
> 200|77 15 13:53:34|TP Start
> 520|77 15 00018119 1 1|time stamps incorrect: mtime, ctime were not updated
> 220|77 15 1 13:53:40|FAIL
> 410|77 15 1 13:53:40|IC End
> [...]
> 10|113 /tset/LSB.os/ioprim/writev_L/T.writev_L 14:03:41|TC Start, scenario ref 119-0
> 15|113 3.6-lite 30|TCM Start
> [...]
> 400|113 15 1 14:04:41|IC Start
> 200|113 15 14:04:41|TP Start
> 520|113 15 00018514 1 1|time stamps incorrect: mtime, ctime were not updated
> 220|113 15 1 14:04:47|FAIL
> 410|113 15 1 14:04:47|IC End
> [...]
> 
> Are these failures the result of a legitimate optimisation or is this
> a bug?  The xfs filesystem was mounted with the options set to
> "defaults".

These types of failures do tend to show up with different filesystems or 
mount settings (noatime or nodiratime on ext3 will cause a number of failures).

I'm not seeing anything in "man mount" about xfs defaults that implies 
that mtime, ctime would not be updated, but I also don't recall having run 
many tests on xfs systems.  It could well be that the test method needs to 
be reviewed to handle the various contemporary filesystems better.

Google does show old (2003) references to an xfs bug mentioning "XFS 
does not update directory mtime when changes made in directory", but these 
tests look to be creating/altering/checking only a file, not a directory. 


-- 
Stew Benedict
The Linux Foundation