On Dec 18, 2008 18:35 -0800, Tim Kientzle wrote: > Are these compatible with star and bsdtar ACL support? > Do they use the star or bsdtar xattr extensions?
The xattrs are stored using the same extensions as star (SCHILY.xattr.*), not totally sure about ACLs because I haven't looked at those. > Ondřej Vašík wrote: >> Bdale Garbee wrote: >>> These look like generally useful stuff. However, I've been trying hard >>> to keep Debian's tar package as close to upstream bits as I can. Are >>> these patches things you'd be interested in merging into a new upstream >>> release? >> >> As Fedora maintainer, I would be happy to have those patches accepted by >> upstream, anyway, I know they have several issues to be acceted >> generally... If you have any complaint (e.g. accepting xattrs as defaut, >> feel free to add this there, I'd amend fedora patch and propose >> upstream :) ) I've been trying to get some feedback on some changes to the xattr patches, but haven't been able to contact any of the people who worked on these in Red Hat. Is it better to just email you directly (and/or this list), or file a bug in Bugzilla? In particular, we'd like to: - Always backup all xattrs on a file instead of current practise of filtering the xattrs at backup time. This ensures they are all available later on if needed, even if they are not all restored by default. The filtering would only be done at restore time, and that can be fixed if it is wrong. - Change the restoration of xattrs to be before any file data is written. This allows the xattrs to contain layout hints, which is important for distributed filesystems (Lustre, pNFS) and possibly also ext4, btrfs, XFS, that have smarter allocators that can take advantage of this in the future once they realize what is possible. Currently we backup/restore the Lustre layout xattr in our filesystem-internal binary xattr format. It has been proposed in a few places (pNFS meetings, Posix HECIOWG) to have a "standard" way of saving/restoring common filesystem layout information (e.g. stripe count, stripe width, RAID level) so that this can be at least somewhat portable between high-end filesystems. Cheers, Andreas -- Andreas Dilger Sr. Staff Engineer, Lustre Group Sun Microsystems of Canada, Inc.
