[OpenAFS-devel] automatically spawning afsd from kernelspace?

Niklas Edmundsson Niklas.Edmundsson@hpc2n.umu.se
Wed, 18 May 2005 12:55:15 +0200 (MEST)


On Wed, 18 May 2005, Chris Huebsch wrote:

>> I think one point being missed is that the current way of doing things is 
>> pretty similar for all architectures.
>
> Is it?

I was mostly refering to the afsd-situation where afsd is the way to 
provide startup parameters, and the argument that you should do away 
with afsd and let the module handle all parameters. Changing this 
would add quite a lot to the mess.

<snip>

Regarding location/loading of modules, each architecture has it's way 
of doing things. IMO, the current framework provides a way of having 
each platform do its own fiddling to get the module into the kernel, 
so the openafs can have pretty generic code for parameter 
handling/passing and whatnot.

> The current practice breaks a lot of common Linux administration
> techniques, building up obstacles installing AFS.

That must be a packaging question, because the Debian packages are 
easily installed, including the kernel module (install 
module-source-package, wave your arms appropriately to build the 
kernel module package for your kernel, install module package, done).

Building openafs by hand and installing on a Linux system might be 
another matter, but my opinion here is that openafs should provide a 
reasonably clean framework for building and installing in a --prefix, 
and full-fleshed-distribution-specific-install with modules in the 
right place, startups, configs, whatnot should be done by those who 
package it for your distribution. I would go so far to say that this 
includes selecting the sysname too, since it makes absolutely no sense 
for openafs to do this by itself (maintaining an array of all 
supported architechtures on all different versions of distributions 
gets just plain silly, and no, "linux on $arch" is not good enough).

Having seen a number of different thingies producing linux kernel 
modules that needs to be loaded, the ones with the least amount of 
"clever solutions" have been the ones that are easiest to package and 
keep up to date. OpenAFS might be full of crud and could do good with 
a large pickaxe applied to the more annoying historical baggage, but 
it's a dream linux-kernel-module-wise compared to certain 
kernel-modules for HPC cluster networks, hardware monitoring (if 
anyone has a Dell bashing session, I'm on) and so on. They might work 
good for those having the exact RedHat release the thing was made for, 
but for the rest of us it's a plain nightmare.


/Nikke - hoping for Ubuntu to achieve world domination so he never has
          to touch an RPM again ;)
-- 
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
  Niklas Edmundsson, Admin @ {acc,hpc2n}.umu.se     |    nikke@hpc2n.umu.se
---------------------------------------------------------------------------
  So, you want to pilot a starship?  Better eat your Wheaties.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=