Coda File System

Re: Version skew is fatal, right?

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Thu, 15 May 2003 01:37:43 -0400
On Thu, May 15, 2003 at 10:31:40AM +0900, Stephen J. Turnbull wrote:
> This
> 
> 10:28:18 WorkerInit: Version Skew with kernel! Get a newer kernel!
> 10:28:18 WorkerInit: Kernel version is 2
> .
> 
> doesn't make that obvious, although I guess you can check that venus
> exited with ps.  Also, it would be nice if venus told you what kernel
> version it wants.

Ahh, it expects version 3. There is a kernel patch for 2.4 and 2.5 in
coda/kernel-src/linux-2.x-coda.patch.

The big difference is that the kernel module now uses opaque 128-bit
identifiers for objects. The kernel doesn't need to know anything about
realms, volumes or other details. But the existing 96-bit structure
couldn't handle the new object space.

When you have a patched kernel and run things for the first time, venus
will come up really quickly and /coda will be empty. Only when you
actually access a 'realm' we will try to resolve the servers. So doing
'ls /coda/testserver.coda.cs.cmu.edu' should find our testserver and
create the directory entry. You can also do 'ls /coda/<my_server_name>'
and your own server root should become visible.

It is really neat, and solves some hairy issues. We cannot get those
unrepairable conflicts on /coda anymore because that volume is 'faked'
locally on the client. We don't really need anymore client side
administration aside from installing the client, as rootservers are
resolved as they are accessed.

But I'm still having some problems taming FreeBSD and WinXP to behave
well with the new code, although the remaining windows problems seem to
be mostly pre-realms bugs.

Jan
Received on 2003-05-15 01:43:11