[PVFS-developers] chown bug, part 2

[Rob Ross]

Don,

Are you guys testing with just the kernel interface, or kernel and MPI-IO 
(or native)?

Thanks,

Rob

On Mon, 27 Oct 2003, Porter Don wrote:

> 
> The more I looked at this, the problem is caused by the dmeta->sd_path entry
> getting corrupted, most notably when one is calling get_dmeta on the root of
> the filesystem. 
> 
> It seems, however, that the sd_path is used _nowhere_ in the code except
> when reading/writing dmeta and at one point in do_mkdir as a buffer to
> append a '/' onto the rd_path.  
> 
> So, unless I am missing something, there is really no reason to store the
> subdirectory path (or filename for that matter) in all of the .pvfsdir
> files.  I can see why the rd_path (root directory) is useful, but a simple
> stat would be the best way to get the actual sd_path.
> 
> So, as an experiment, I stripped out all of the places in the pvfs code
> where dmeta->sd_path and dmeta->fname are used, and everything seems to work
> just fine.  Also, the problems with chroot go away.  Further, because these
> are the last two entries in the .pvfsdir file, they can be easily ignored if
> they are already there.
> 
> Please let me know if I am missing something, but these seem to be vestigial
> structures that we can do without.  Attached is a patch that basically
> prunes these out, works around the buffer thing in do_mkdir (just in case!),
> and fixes the chown problem.
> 
> Thanks,
> Don Porter
> 
>