On Fri, Oct 28, 2005 at 07:10:24AM -0700, Jason Thorpe wrote:
> 
> On Oct 28, 2005, at 12:39 AM, Jonathan A. Kollasch wrote:
> 
> > Anyway, isn't having
> >extra code for adding entries to devfs, etc. just extra bloat?
> 
> No.  In fact, it's the only reasonable way to handle highly-dynamic  
> device discovery as found on modern systems.

And what are we then supposed to do on another class of "modern systems",
embedded systems which are required to have certain elements of their
configurations static for security reasons (or in order to obtain certain
security certifications)?  I commonly mount all read-write filesystems
nodev, and mount all filesystems containing devices read-only, so that I
can be *guaranteed* that no new device nodes will be available to user
processes no matter what else changes.

It seems like every time we discuss devfs, somewhere near the end of the
discussion some handwaving is done about how this or that hack will be
provided to support requirements like mine -- but at the beginning of the
next discussion, it's entirely plain that the people advocating devfs
couldn't really care less, because they're forgotten all about it.  This
makes me skeptical that if devfs goes into our tree and static device
nodes go out, I will not, in fact, lose the useful functionality of which
I now take advantage.

-- 
 Thor Lancelot Simon	                                      tls@rek.tjls.com

"The inconsistency is startling, though admittedly, if consistency is to be
 abandoned or transcended, there is no problem."		- Noam Chomsky