[OpenAFS] openafs 1.2.10 on fedora core 1.0

David Botsch dwb7@ccmr.cornell.edu
Wed, 12 Nov 2003 16:07:59 -0500


See points below:

On 2003.11.12 14:31 Derek Atkins wrote:
> Hi,
> 
> David Botsch <dwb7@ccmr.cornell.edu> writes:
> 
> > Sure. See attached.
> 
> Thanks.  I've got a few questions...
> 
> >  # Set 'krb5support' to 1 if you want to build the openafs-krb5
> package
> >  # to distribute aklog and asetkey
> > -%define krb5support 1
> > +%define krb5support 0
> 

Initially, I received an error about not being able to find libkrb524. 
So, my initial fix was to use perl to get rid of the -lkrb524 in the 
linker options (since in the krb docs, those apis have been moved into 
the libkrb.a library). When I received other errors with the afs-krb5 
stuff not building, I just turned it off (and didn't bother removing 
the perl command since it was no longer executed, at any rate).

> Why did you set krb5 support to 0 if you also (later) modify the
> aklog build?
> 
> > -config_opts="--enable-redhat-buildsys \
> > -%if %{enable_bitmap_later}
> > -	--enable-bitmap-later \
> > -%endif
> 
> First, why do you disable the "--enable-redhat-buildsys"?  What

See the email from Grant Williamson concerning a lot of errors about 
CONFIG_NR_CPUS being "undeclared here". Removing the 
--enable-redhat-buildsys fixed that.

> exactly is failing when this remains in the build?  Note that this
> could be why you're failing to get properly-named modules.  It's also
> possible that the SPEC doesn't deal with overriding the /boot/kernel.h
> file, or that Red Hat has removed support for older platforms like
> i386 and/or i586.  However, removing this is the wrong answer, and
> WILL cause the SPEC to fail in a number of other ways, in particular
> when trying to build for a kernel on a different processor-type.
> 

I believe that i586 is still supported. The modules were not named 
improperly (they were named according to the same scheme as has been 
used on redhat for a while). But, when afs attempts to load a module, 
it is not using the same naming scheme. So, one or the other needs to 
be fixed.

I suspect it has to do with there no longer being a number before the 
"smp" in the release name.

Compare
2.4.18-27.7.x.ccmr.1smp

with
2.4.22-1.2115.nptlsmp

There would prolly not be a problem if it was named something like: 
2.4.22-1.2115.nptl.1smp



> It would be much better if you could help me figure out why the
> "redhat-buildsys" is failing.
> 

Yep. What exactly is this supposed to do, at any rate?

> > -%if %{enable_bos_restricted_mode}
> > -	--enable-bos-restricted-mode \
> > -%endif
> > -%if %{enable_fast_restart}
> > -	--enable-fast-restart \
> > -%endif
> > -	--enable-transarc-paths"
> > +#config_opts="--enable-redhat-buildsys \
> > +#%if %{enable_bitmap_later}
> > +#	--enable-bitmap-later \
> > +#%endif
> > +#%if %{enable_bos_restricted_mode}
> > +#	--enable-bos-restricted-mode \
> > +#%endif
> > +#%if %{enable_fast_restart}
> > +#	--enable-fast-restart \
> > +#%endif
> > +config_opts="--enable-transarc-paths"
> 
> Another question is why you disable the ability to enable fast
> restart, bitmap-later, bos-restricted-mode, and fast-restart?
> 

Can't think of a good one other than they got commented out when I made 
the change of getting rid of redhat-buildsys


> > @@ -447,10 +447,11 @@
> >  %if %{krb5support}
> >  # Now build aklog/asetkey
> >  (cd afs-krb5/src &&
> >  	./configure --prefix=/usr --with-krb5=/usr/kerberos \
> >  	    --with-afs=`pwd`/../../${sysname}/dest/ && \
> > +	perl -i -pe 's/-lkrb524//' Makefile && \
> >  	make all PROGS="asetkey aklog ka-forwarder" && \
> >  	touch afs2k5db && touch fakeka && \
> >  	make install PROGS="asetkey aklog ka-forwarder" \
> >  	    DESTDIR=`pwd`/../../${sysname}/dest/ INSTALL_BIN=/bin \
> >  	    INSTALL_SBIN=/etc) || exit 1
> 
> Why do you do this if you've disabled krb5 support?  Actually a better
> way to deal with this is to change the afs-krb5 patch, or to add
> another patch that performs the same operation.
> 

As for the why on the "perl" command, see above.
Yes, the appropriate solution is to fix afs-krb5. The idea for the 
moment was just to get afs to compile, with or without krb5 support.

> -derek
> 
> --
>        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
> 

-- 
********************************
David William Botsch
Consultant/Advisor II
CCMR Computing Facility
dwb7@ccmr.cornell.edu
********************************