Oplock logic bugs in Samba 2.2.2

[acherry at pobox.com]

 > Hmmm. This looks like the problem that was confirmed fixed by
 > several large Solaris sites with the Samba 2.2.2 CVS code.
 > 
 > Getting more backtraces would be very useful here.

I'm still working on this one -- the problem has so far proven
impossible for us to reproduce in a test environment, and there are
obvious issues with collecting the debugging data in our production
environment (which has to be done carefully at off-peak hours).  Our
production server is pretty complex (Sun Cluster, HSM, Veritas
volumes and vxfs), so it's difficult to build an equivalent test
system, not to mention a test load of >100 users.

Our main reason for wanting to move forward with the Samba 2.2.2
upgrade was the improved wildcard handling code, which fixes a problem
we're having with a particular piece of Windows software.  As a
possible interim solution, I've tried pulling the ms_fnmap() code out
of 2.2.2 and integrating it into the Samba 2.0.6 sources.  The
modified 2.0.6 seems to be working so far in testing.  I was hoping
that someone more intimate with the Samba code base would be able to
point out if there are any major problems with doing this.  The
modifications I have made are:

     - Copy ms_fnmatch.c into the 2.0.6 lib and modify Makefile.in
       and proto.h to include it in the build.

     - Replace mask_match() in util.c with the 2.2.2 version.

     - Remove the final argument (BOOL trans2) from the prototype
       for mask_match() and all calls to mask_match(), since it was
       removed in the 2.2.2 version (not needed anymore??).

     - Change the call to fnmatch() in interpret_interface() to use
       ms_fnmatch() (probably not necessary, but I did it to match
       the 2.2.2 sources).

Are there any obvious problems with doing this (it's only a stopgap
solution until we can figure out why we are having problems 2.2.2).

Thanks...

-Andrew Cherry
 UNIX System Admin
 Cummins, Inc