[OpenAFS-devel] Re: kabi-tracking kmods

[Stephan Wiesand]

On Sep 26, 2012, at 21:03 , Troy Benjegerdes wrote:

> Based on my understanding, this is a business strategy of Red Hat
> to get big customers, and third-party non-GPL software vendors to
> get them to sign big money service and support contracts for RHEL.
> 
> OpenAFS falls into the 'third party non-GPL software' category.
> 
> It's completely clear to me why it doesn't work... that's because
> subtle breakage of the 'stable' interface by anything linking into
> the linux kernel is in the best interest of everyone who is employed
> by Red Hat writing and patching the kernel. Red Hat employees will
> say they are 'committed to providing stable interfaces' publicly,
> and what you'll see in practice is something else entirely.  
> 
> Kind of like the commitments I keep hearing about people make on 
> this mailing list.

I disagree with, well, everything said above.

> We can either keep recompiling, pay big money, or use the Red Hat 
> developed kAFS.

Another option would be to make the OpenAFS FUSE client fully functional.

Any serious obstacles to that? What would it take?

- Stephan

> 
> On Wed, Sep 26, 2012 at 01:45:44PM -0500, Andrew Deason wrote:
>> On Wed, 26 Sep 2012 11:38:31 -0500
>> Troy Benjegerdes <hozer@hozed.org> wrote:
>> 
>>> 2) RHEL defines which kabi functions are 'stable'
>> 
>> I'm not sure if you understand what Ken is talking about. As far as I
>> can tell, Red Hat's intent is that this is supposed to still work even
>> for things that do not stick to any "stable" whitelist of functions (for
>> RHEL6 and beyond). So if something not "stable" changes, the kernel will
>> recognize that the relevant kernel module is not compatible, and you
>> need to recompile the thing. So we're not supposed to need to stick to a
>> stable ABI.
>> 
>> But as the thread that Ken linked shows, it doesn't seem to work.  We're
>> not sure why it doesn't work for that particular case, and Red Hat's new
>> kernel patching policies make it difficult to figure out. I think it's
>> hard to say how often that will reoccur without knowing what actually
>> broke.
>> 
>> It's of course safer to just recompile for any version change, but if
>> that were acceptable to everyone I expect this kabi stuff wouldn't exist
>> in the first place. I personally don't find it "worth it" to try and
>> figure this out at the moment, but I've treated Linux's lack of
>> interface stability/design as something to just 'live with' for awhile,
>> so it doesn't bother me so much.

-- 
Stephan Wiesand
DESY -DV-
Platanenenallee 6
15738 Zeuthen, Germany