On Sat, Oct 18, 2003 at 09:14:30AM -0400, gianni sissa wrote:
> i'm working with coda to build a distribuited storage system for my company. 
> i've a question about the supported fs in coda: which partition need to be
> ext2? i mean: for /vice and /vicepx can I use a nfs mounted filesystem? (a
> little crazy but necessary for me...)

As far as the server is concerned, anything goes. It is a normal
userspace process.

For the Coda client things are a bit different and is also a bit
platform/kernel version specific.

On NetBSD/FreeBSD and Linux-2.2 kernels we pass down a device/inode
number pair and we access the underlying container file directly.
This doesn't work well for several journalled file systems on Linux
because they associate commit actions with the closing of a filehandle,
it also breaks on file systems that do not have a unique device or inode
number (many network file systems, ramfs, tmpfs, jffs2, reiserfs, vfat...)

In Linux-2.4.9 or .10 this was changed, most distributions started to
use ext3 or reiserfs by default and this broke many of the newly
installed Coda clients. So on newer Linux systems we pass down the file
handle and always access the file though that handle. I haven't yet had
a venus.cache file system that doesn't work with this new method. Well
one exception, you clearly can't store venus.cache in /coda.

Jan