rsync - Oct 2008 - Patch for GPFS support

List,

We, IBM, have developed a patch that enhances RSYNC.
This patch provides extensions to RSYNC to map to and use filesystem
extensions present in the IBM GPFS cluster filesystem.
(GPFS is a cluster filesystem for Linux and AIX that provides additional
attributes such as NFSv4 ACLs, extended attributes and windows attributes
in addition to the common posix permissions)

This will greatly enhance the RSYNC functionality for users of the GPFS
filesystem.


We would like to contribute this patch to RSYNC and hope it will be useful
for all users of GPFS.
This patch may also provide ideas/startingpoint for others with filesystems
that would need to transfer additional fs-specific metadata using rsync.


Please review and apply if there are no issues with the patch.



best regards
Ronnie Sahlberg



(See attached file: README.patch)(See attached file:
rsync_gpfs_attrs-git.diff)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: README.patch
Type: application/octet-stream
Size: 2282 bytes
Desc: not available
Url :
http://lists.samba.org/archive/rsync/attachments/20081021/015484b1/README.obj
-------------- next part --------------
A non-text attachment was scrubbed...
Name: rsync_gpfs_attrs-git.diff
Type: application/octet-stream
Size: 32976 bytes
Desc: not available
Url :
http://lists.samba.org/archive/rsync/attachments/20081021/015484b1/rsync_gpfs_attrs-git.obj

On Tue, Oct 21, 2008 at 10:02:42AM +1100, Ronnie Sahlberg
wrote:
> This patch provides extensions to RSYNC to map to and use filesystem
> extensions present in the IBM GPFS cluster filesystem.
Thanks for the patch!  I have looked over your changes, and I have some
preliminary feedback:

 - When a protocol change requires an new option to be used, there's not
   usually a need to bump the protocol version.

 - It is rsync's job to pass options to the remote server, not the user.
   When a new option is added, you should change the server_options()
   function in options.c rather than requiring the user to use
   --rsync-path (or --remote-options).

 - That said, I'd prefer a change such as this to not need a new option.
   Ideally, rsync would notice when a filesystem supported GPFS
   attributes and include the data in the regular xattr data using a
   special xattr name, such as "user.rsync.GPFS".  That would require
no
   protocol change, and would even allow the backing up of GPFS info to
   a non-GPFS disk.  Note that the xattr code might need some speed
   improvements to handle the case where every file has a chunk of
   differing data, as it currently tries to search for matching data
   between files, and its search algorithm is not very good.  You appear
   to have supplied a better search algorithm in your patch, which might
   be good to use in the xattr and ACL code.

 - I would be interested to know if the GPFS data could be read in
   smaller pieces than one big chunk, as it might be better to fit the
   data into rsync's current data structures in separate bits.  E.g. if
   there is ACL info in the data, can the ACL code be extended to handle
   the data and give the user the option of choosing what parts of the
   GPFS data they want to copy?

 - Are there application interface docs on-line?  I found some man pages
   in a tar file, but on-line docs would be nice too.

..wayne..