Coda File System

Re: UNIX-domain sockets on a Coda FS?

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Mon, 18 Oct 1999 14:15:39 -0400
On Mon, Oct 18, 1999 at 12:24:39PM -0500, Bill Gribble wrote:
> I'm having a problem running a new window manager and I can't explain
> my problems any other way than by blaming coda :) 
> 
> The WM is called "sawmill", and it uses a UNIX-domain socket in the 
> ~/.sawmill/ directory to allow runtime configuration and communication. 
> When my home directory is a Coda-mounted volume (5.3.1, linux-glibc2.1, 
> kernel 2.2.10), launching sawmill leaves this error message in 
> my .xsession-errors :
> 
> bind: Input/output error
> 
> ... and no socket gets created, so I am unhappy. 
> 
> The same config without Coda works fine (except for kernel version, 
> 2.2.12 instead of 2.2.10).  
> 
> Any thoughts? 
> 
> Thanks,
> Bill Gribble

The concept of pipes, sockets and device nodes was probably considered
useless in a distributed filesystem. And they should be stored on the
local disk.

What kind of behaviour would you expect when starting sawmill on another
client from the same homedirectory? Should it fail to start because
someone else created that socket already?

Also security considerations are involved. Imagine someone creating a
world writeable `/dev/kmem' using a client on which he has root access,
and then uses this to obtain root privileges on any other machine. 

The kernel module is ready to handle sockets, pipes and devices, but the
Coda client and server are not, and I am not so sure whether they
should. Although I can easily see many cases where they would be very
useful.

Jan
Received on 1999-10-18 14:16:56