Re: sysfs -> /sbin/hotplug -> udev

[Joerg Sonnenberger <joerg@xxxxxxxxxxxxxxxxx>]

On Fri, Apr 23, 2004 at 12:18:11AM -0400, esmith wrote:
> sysctls are not tied to any object. While there are groups of sysctls for
> various subsystems, they are no rigid semantics for binding them together.

Well, you can discuss about this.

> sysctls, at least in Linux, are a bit of a pain to add, and from a
> maintainence standpoint, don't scale (There are large static lists in a
> couple of files somewhere). You also can't add them dynamically (like when
> a device is inserted or a module is loaded). There is easy way to add a

That's a Linux specific problem which is not true for *BSD. While adding
and removing sysctls is currently somewhat expensive, it can be done and is
done by modules.

> per-object sysctl, for a dynamic number of objects. And, there is no
> built-in protection mechanism for the objects when reading/writing
> sysctls.

Sure, so what? It would be possible to add that, but I'm not convienced
that it is necessary. E.g. our sysctl approach for ps & co. is IMO much
better than procfs, though we have to add a compatibility layer one day
to solve the dreaded proc size changed problem.

> kobjects and sysfs provide all of those things, so not only is it handy,
> it's far superior in many ways. :)

Sure, but I still can't find a real use which can't be solved by improving
sysctl. Heck, you could even write a sysctlfs if you want to use the
normal fd API.

Joerg

> 
> 	Pat