On Thu, Sep 30, 2004 at 11:18:29AM -0400, Jan Harkes wrote:
>     pioctl opens a magic file /coda/.CONTROLPIPE
>     venus returns an open socket or pipe.
 ...

Hello Jan,

glad you find it interesting.

If I am not totally off, you considering the communication between
other processes and Venus. I thought also about the kernel-Venus
communication. Now it is ioctl()s which is quite natural,
but can break when ioctls are remapped by an unaware abi-compatibility
layer.

We would be rather safe if we could communicate via write()/read()
as the semantics of such syscalls is well defined,
in contrast to "non-standard" ioctl()s.

I suppose it is going to stay a theoretical discussion, as we have much
more important things to change :) but still it is good to choose
the most flexible approach next time we happen to make changes.

If we'd have a simple and generic interface between the kernel and userland,
we could run the cache manager in an isolated box like a virtual machine
- or even on another host :)
[if Venus can access cache files on the "coda client" in any way,
and talk to the kernel module by some means, that is enough]

Just generating wild ideas :)
--
Ivan