[OpenAFS] openafs 1.2.10 on fedora core 1.0

[Troy Dawson]

> Date: Wed, 12 Nov 2003 16:13:49 -0500
> From: David Botsch <dwb7@ccmr.cornell.edu>
> To: Derek Atkins <warlord@MIT.EDU>
> Cc: David Botsch <dwb7@ccmr.cornell.edu>,
> 	grant williamson <traxtopel@fastmail.fm>,
> 	OPENAFS <openafs-info@openafs.org>
> Subject: Re: [OpenAFS] openafs 1.2.10 on fedora core 1.0
> 
> See below...
> 
> On 2003.11.12 14:19 Derek Atkins wrote:
> 
>>David Botsch <dwb7@ccmr.cornell.edu> writes:
>>
>>
>>>I've discovered one problem so far with the newly built rpms.
>>>
>>>On a SP machine, the name is:
>>>2.4.22-1.2115.nptl
>>>
>>>fine.
>>
>>By "the name" you mean "output from uname -r"?
>>
> 
> 
> Yes. That exactly what I mean. The release.
> 
> 
>>>On a MP machine, the name is:
>>>2.4.22-1.2115.nptlsmp
>>>
>>>kernel modules built are:
>>>
>>>libafs-2.4.22-1.2115.nptl-athlon/x86.o
>>>and
>>>libafs-2.4.22-1.2115.nptl-athlon/x86.mp.o
>>
>>Uhh, the slash is DEFINITELY going to be causing problems.  There's
>>really a slash in the name?  If so the build script is going to HAVE
>>to deal with it.
>>
> 
> 
> No, no slashes. I was just trying to be all-inclusive.
> 
> Modules built were named, for example:
> libafs-2.4.22-1.2115.nptl-athlon.o
> libafs-2.4.22.1.2115.nptl-athlon.mp.o
> libafs-2.4.22.1.2115.nptl-i586.o
> libafs-2.4.22.1.2115.nptl-i586.mp.o
> 
> and so on for i386 and i686
> 
> 
>>>AFS is trying to load a module named:
>>>libafs-2.4.22-1.2115.nptlsmp.mp.o
>>
>>I suspect this is due to the above issue and either afsmodname not
>>dealing properly or something else causing problems.  I'd need to see
>>the contents of <builddir>/libafs/, the contents of one of
>><builddir>/libafs/MODLOAD*/, the contents of
>><instdir>/usr/vice/etc/modload and the contents of
>><instdir>/usr/vice/etc/modload/SymTable to have a better idea of
>>what's going on.
>>
> 
> 
> No problem with the name of the built modules. Just with the name of 
> the module afsd thinks it needs to load. It tries, on an smp machine, 
> to load a module named libafs-2.4.22-1.2115.nptlsmp.mp.o
> 
> so, unlike in previous versions of redhat, that "smp" string is getting 
> in there, somehow.
> Taking a closer look, looks like the init.d script for starting afs 
> needs to be fixed.
> 
> 
>>>Ugh!
>>
>>Yea.  This is what happens when you run out front ;)
>>
>>-derek
>>
>>PS: Thanks for the diff.  I'll take a look sometime soon.  I'm at the
>>IETF this week so I wont be pulling down SP1 until at least this
>>weekend if not next week.
>>
>>--
>>       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
>>       Member, MIT Student Information Processing Board  (SIPB)
>>       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
>>       warlord@MIT.EDU                        PGP key available
>>
I'm not trying to do this on Fedora, but I am doing it on RedHat Enterprise 3 
(as) and ran into the same problem.  This is similar to the Mandrake problem I 
was having a while back.  Thanks to whoever fixed up the 1.2.10 spec file 
(Derek?), but now there is a line that you can change so you can pick 
whichever variable you want it to search on for your kernel.
So in the spec file I changed

# Mandrake users can define symtab to be "-S iget4_locked -x"
%define symtab -x

to

# Mandrake users can define symtab to be "-S iget4_locked -x"
%define symtab -S page_follow_link -x

I then recompiled the rpm's (I needed to do that anyway) and now it finds the 
right kernel module.  I really just pulled the page_follow_link out of the 
list of variables I got when I did a

/usr/vice/etc/afsmodname -c /usr/vice/etc/modload/libafs-2.4.21-4.EL-i386.o

But it worked, so I'll keep that variable in there till it doesn't work 
anymore, then find a better one.

I hope this helps and isn't too disjointed.
Troy


-- 
__________________________________________________
Troy Dawson  dawson@fnal.gov  (630)840-6468
Fermilab  ComputingDivision/CSS  CSI Group
__________________________________________________