[OpenAFS-devel] Re: [OpenAFS] 2.6 kernel support anytime soon? Workarounds?

[Jeffrey Hutzelman]

On Monday, May 10, 2004 14:12:52 -0400 Derek Atkins <warlord@MIT.EDU> wrote:

> There was a lot of history with Linus and Alan about how the AFS
> module would interface with the kernel..  And the whole question of
> "stable" interfaces back around the 2.0 days (like when Linus broke
> binary compatibility by rearranging the members of a structure).  Back
> then AFS _couldn't_ go into the kernel; it wasn't open source, and the
> other Linux developers knew it.  They just haven't relearned that
> times have changed.

It still can't go into the kernel.  It's open source, but not distributed 
under the GPL, and the license is beyond our power to change.

> As if GPL is the only open source license??   Feh!

Ah, but that's exactly what the Linux folks seem to think.  The broken 
"tainting" model the Linux folks have adopted makes no provision for 
non-GPL open-source licenses.  They assume that either your license is the 
GPL, or you suck.

The sad thing is, the original intent of the tainting stuff was to make it 
easy to tell people to go away when their machine was broken by some 
binary-only module, instead of spending lots of time tracking down a 
problem only to discover that it's buried inside something you don't have 
source for.  Fine; I was willing to debug problems with AFS when it was 
binary-only, instead of sending people to linux-kernel.  Really, I still 
am, especially when it's my users involved.

But that doesn't excuse the "your license is not exactly the same as our 
license, and your architecture is not based on whatever bizarre ideas's 
we've had in the last few years, so we hate your forever and ever and will 
do everything in our power to try to screw you".

I'm sorry, but I'm sick and tired of the holier-than-thou attitude of the 
Linux folks.  I've run AFS on Sun 2's and IBM RT's and damn near everything 
else even vaguely UNIX-like made since then, and most of those "closed" OS 
vendors have been more helpful and provided more stable interfaces than 
Linux.  SGI added hooks to login and xdm.  HP went to some trouble to 
publish previously-unavailable headers needed by OpenAFS.  NeXT made 
changes to their directory browser so it wouldn't choke on /afs.  The Linux 
maintainers, on the other hand, tell us we can't have our syscall any more, 
demand that we completly rearchitect the user-kernel interface that has 
worked well for us on many platforms over a couple of decades, and then 
WON'T EVEN GIVE US THE TOOLS TO DO SO.  That's just idiotic.



-- Jeffrey T. Hutzelman (N3NHS) <jhutz+@cmu.edu>
   Sr. Research Systems Programmer
   School of Computer Science - Research Computing Facility
   Carnegie Mellon University - Pittsburgh, PA