On Fri, Feb 01, 2002 at 04:21:42PM -0500, Jan Harkes wrote:
> On Fri, Feb 01, 2002 at 10:07:06PM +0100, Ivan Popov wrote:
> > $ while true;do echo -n .;./cat</dev/null|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat|./cat;done
> ...
> > 
> > It looks like the error happens more often while I am typing on another
> > virtual screen. And A LOT more often while I'm running "find /local" that
> > is a lot of disk activity and screen output :)
> > 
> > Hope it helps to reproduce and debug...
> 
> Reproduced it, and from the strace I discovered that it is an execve()
> syscall that returns the ENOENT error. Now I'm going to try to see how
> we handle this in the kernel module.

Ok, our readlink implementation isn't failing, and it's not a venus
problem. There are only 3 possible places in the Coda kernel module that
could be causing this, or else it is a race in the generic Linux VFS
layers.

Jan