[PATCH] Fix for bug #9706 - Parameter is incorrect on Android

[Andrew Bartlett]

On Wed, 2013-03-13 at 15:11 -0700, Jeremy Allison wrote:
> On Thu, Mar 14, 2013 at 08:48:32AM +1100, Andrew Bartlett wrote:
> > On Wed, 2013-03-13 at 10:46 -0700, Jeremy Allison wrote:
> > > On Wed, Mar 13, 2013 at 08:48:44AM -0700, Jeremy Allison wrote:
> > > > On Wed, Mar 13, 2013 at 01:00:30PM +1100, Andrew Bartlett wrote:
> > > > > 
> > > > > Why do we need to look for RA_SAMBA, if a client has sent
> > > > > CAP_LARGE_READX?
> > > > 
> > > > Because due to changes in our client library,
> > > > currently the 4.0.x client code doesn't send CAP_LARGE_READX
> > > > or CAP_LARGE_WRITEX to the server in the sessionsetupX
> > > > capabilities field. The 3.6.x client code (and also
> > > > the MacOXC client) do, and I think we should too
> > > > (even though a Windows server ignores it).
> > > 
> > > Wow, this has gotten really interesting. Found a bug in
> > > our client libraries also whilst writing all the tests
> > > for this :-).
> > > 
> > > I'll attach everything to the bug report once I'm done.
> > 
> > If it is only 4.0 clients that are missing this, then I think we should
> > just treat that as a 4.0 bug, and not work around it on the server.  
> 
> I have a new test suite, and a shiney new patchset
> that goes along with it to fix things. I'll attach
> to the bug so we can get it into 4.0.next and 3.6.next.
> 
> Remember we already have this check, so it's nothing
> new and doesn't break anything for other clients, and
> taking it out will just arbitrarily break Samba 4.0.x clients,
> so let's leave it alone.

The reason I would like it removed it that then the client side, not
needing to trigger/not trigger the magic 'samba' detection, totally
controls if the server thinks large reads/writes are acceptable.

Otherwise, how does your test show we behave correctly for clients that
don't advertise this capability, but do (wrongly) ask for a large
read?  

That is, given the number of different combinations we have tripped over
here so far, why not both make it easy to test and remove 'detected OS'
from the matrix we must prove and test?

Andrew Bartlett

-- 
Andrew Bartlett                                http://samba.org/~abartlet/
Authentication Developer, Samba Team           http://samba.org