Re: xfs (was: reiserfs on all my SRM-alpha drives...)

["Martin K. Petersen" <mkp mkp net>]

>>>>> "Michal" == Michal Jaegermann <michal@ellpspace.math.ualberta.ca> writes:

#include <disclaimers/i-am-not-speaking-for-sgi.h>

>> Uhm, uhm, uhm.  A rewrite?  Why would we rewrite it?

Michal> To better fit with the rest of the kernel.  I recall quite
Michal> distincly suggestions of that sort from some people who will
Michal> have some say in matters of a final acceptance.  

We have a slim layer mapping between Linux VFS and what the core XFS
code thinks is the IRIX vnode layer.  That's what Al dislikes.

And while I mostly share his opinion, ripping linvfs out and teaching
XFS to talk directly to Linux' VFS would not only be a big task.  It
would also mean throwing away hundreds of man-years of development,
testing, and QA that SGI has put into the filesystem.

And that is simply not worth the effort.  A stable filesystem is more
important.  Plus SGI is obviously interested in sharing code base
between IRIX and Linux.


Michal> I hear (mind you, I hear!) that right now, for example, it is
Michal> non-trivial to have at once both ext3 and xfs support.  

Not true.  There's an export symbol conflict and that's it.

I have both on my machine right here.


Michal> An issue of xfs going "too deep" into other structures was
Michal> mentioned on linux-kernel list quite a few times.  

A few points I'd like to make:

Chances are that people making statements like that haven't looked at
the patches recently.

XFS used to have a bigger footprint, yes.  When we started the
project we kept joking about porting Linux to XFS.  And to some
extent that turned out to be the case.  XFS is by far the most
advanced Linux filesystem out there.  And it has feature requirements
beyond what currently Linux offers.

SGI has been actively participating in the kernel development process
in order to add the required functionality to the generic kernel.
Most of the things XFS needs are now part of 2.4 as a result of this.
And the process is ongoing.  We're helping define a common interface
for access control lists and extended attributes, for instance.

The single big XFS feature Linux is missing right now is support for
delayed allocations in the VM.  And that's where the bulk of our
changes are.  None of the other filesystems on Linux support delayed
allocations yet, so there's no infrastructure to support it.  But I
know that at least the reiser folks are interested in adding support
to their filesystem.

So when you see comments about ``XFS going too deep into other
structures'' it usually means ``Linux doesn't have feature XYZ yet''.


Michal> If you suggest that xfs is likely to become a part of a
Michal> standard kernel in 2.4 series then you know something which I
Michal> did not hear yet.  

Well, who knows?  Right now a stable VM is the primary target...

-- 
Martin K. Petersen, Principal Linux Consultant, Linuxcare, Inc.
mkp@linuxcare.com, http://www.linuxcare.com/
SGI XFS for Linux Developer, http://oss.sgi.com/projects/xfs/