Re: devfs - why not ?

[Johannes Erdfelt]

On Thu, Apr 13, 2000, Stephen Frost <[e-mail]> wrote:
> On Thu, 13 Apr 2000, Johannes Erdfelt wrote:
> 
> > On Thu, Apr 13, 2000, Stephen Frost <[e-mail]> wrote:
> > > 	Heh, of course, there's the question of: can you *get* the
> > > information.  I have no clue if you can or not, but I'm curious, *can*
> > > you get the serial number of a mouse?  Does that work for any ps/2
> > > mouse?  Just curious.  If you can't, well, it don't make no difference
> > > if they're unique. :)
> > 
> > PS/2 mice don't have that information. USB devices have the capability
> > to, but all of the mice I've seen, don't have serial numbers.
> 
> 	Okay, well, just kind of FYI, I don't want a USB mouse plugged in
> on some random machine on my network to control the pointer on another
> machine on my network. :)  The same goes for ps/2, and in general 
> filesystems as well.

Heh, I wouldn't want that either.

> 	Now, if you want data on a partition on another machine, it would
> be nice to get that information.  That, however, is in general a userspace
> issue.  Sure, the kernel can provide information that makes sense to 
> provide to user space about a given device, but the kernel shouldn't be
> out hunting down filesystems to mount locally.
> 	I would have thought the kernel would provide information to
> userspace about a device (physical attributes, where the filesystem is
> physically), and that userspace can then figure out what it wants to do
> w/ it.  If it wants to use UUID, physical location, MAC address, what have
> you, it can, it's userspace.  The kernel shouldn't be defining these things,
> the kernel should just be telling you where it is *physically*.  devfs,
> mostly I think, tells you where it is physically.  That's fine, userspace
> can then use that info to figure out what to do w/ it.  Userspace can
> also go figure out other things about whatever it is to determine where
> to mount it, or symlink it, or change permissions, etc, etc, etc.
> 	My thought is simply this, the kernel shouldn't be looking up in
> tables and whatnot where to mount something, that isn't its problem.  It
> has some useful info about a device that just showed up, it needs to pass
> that info to userspace so userspace can go do something about that device
> showing up.
> 	The interface between the kernel and userspace is where things are
> getting hung-up, I think.  It used to be major,minor numbers.  Now we have
> the option of devfs.  Well, devfs is nice because it gives us a larger
> address space, but at the same time it's introduced some confusion as to
> what the kernel should be doing and what userspace should be handling.
> 	So, let's get back to figuring out how the kernel should be talking
> to userspace about devices and quit w/ the grand ideas for the userspace
> half, if someone finds it useful, they'll implement it. :)

What's wrong with the current devfsd protocol?

Most of the objections I've seen center around devfs as a VFS, but not
necessarily devfsd.

JE