Alexander Dalloz wrote:
On the Fedora user list the question came up whether dump & restore are extended attributes aware. This is an issue i.e. when using SELinux. Do the tools take care?
Since 2002, this is the #1 reason I've been advocating that Red Hat adopt XFS as its' 2nd, officially supported filesystem. xfsdump for XFS does this, and always has. You don't have to use "getfacl" to save POSIX ACLs (although that was always an option, even back in the early 2.4.x days for XFS).
Everything (short of SGI-integrated hardware API support) came over from Irix in XFS for Linux, and that included POSIX ACLs, Quotas, etc... This is in stark contrast from IBM JFS which came from OS/2 instead of AIX (because of the legal issues with Monterey), and ReiserFS is a non-traditional UNIX design (I won't go there, even if it is very innovative in most respects). Most of these features in kernel 2.6 ere thanx to the SGI XFS team, and the team also worked with the Ext3 and Linux & co. as of 2.5.3+ in ensuring POSIX EAs were uniform across all filesystems c/o the VFS.
I haven't wanted to bother Tweedie and/or Red Hat in getting an explaination why they aren't looking at supporting XFS, but I think it would solve a lot of scalability issues. I always assumed it wasn't merely a "NIH" attitude, and they had real reasons for not wanting to support it until Ext3 proved to be a serious limitation for them.
Rodrigo Barbosa wrote:
That depends very much on what you are calling extended attributes. I have no experience with SELinux, so I really can't make an educated guess on what you are talking about. Care to enlight me ?
I have to be honest, I've never used xfsdump and verified SELinux attributes were preserved. But it should bring _all_ POSIX EAs with it. I've just typically relied on it for ACLs.
-- Bryan J. Smith mailto:b.j.smith@ieee.org
On Tue, 14 Jun 2005 at 2:19pm, Bryan J. Smith b.j.smith@ieee.org wrote
I haven't wanted to bother Tweedie and/or Red Hat in getting an explaination why they aren't looking at supporting XFS, but I think it would solve a lot of scalability issues. I always assumed it wasn't merely a "NIH" attitude, and they had real reasons for not wanting to support it until Ext3 proved to be a serious limitation for them.
The canonical reasons I've heard are 1) they don't want to spend the money/time/resources to acquire enough XFS expertise to support it at the Enterprise level and 2) besides, as of RHEL4 (they claim), XFS doesn't provide anything ext3 already provides, so why bother.
Yes, I've pointed out on official Red Hat mailing lists that 2 is false due (at least) to the issue of backing up ACLs (use star they say -- no thanks, say I), but I got no response to that. And I've got benchmarks showing XFS pretty handily beating ext3 on nice new hardware, but I don't have much faith that would get any response either.