Re: required/suggested devfs userland tool functionality

["Simon 'corecode' Schubert" <corecode@xxxxxxxxxxxx>]

Matthew Dillon wrote:
> :2) let the userland tool load a whole set of rules (for each devfs
> :mount point) into the kernel. In turn the kernel applies the set of
> :rules every time a device is attached. This has several advantages:
> :- userland wouldn't have to be asked for every device attach
> :- rules would continue to be applied even if the userland tool isn't running
> :- for that same reason, userland tool wouldn't have to be a daemon.
> :
> :While I prefer the second approach, I would like to hear your thoughts
> :about this before making a final decision on which one to use. I'd
> :also welcome suggestions of other things you think the userland devfs
> :tool should be able to do.
> :
> :Cheers,
> :Alex Hornung
>
>     I like the second approach.   Particularly since you already have a
>     a VOP interface so loading the rules into devd could be as simple as
>     doing a write() to a special node in devd.

But do you really want to perform regexp/glob matching in the kernel?  Or 
do you want to restrict the users to prefix matching?

I think we basically need to deal with multiple things here:

1. no race conditions when creating device nodes
2. give the user enough flexibility
3. allow the user to use chmod/chown?

I don't have an opinion yet what is better, but maybe we should assess 
which kind of rules a user is expected to write (which rules do we want to 
ship per default?), and then we can decide whether it is worthwhile to put 
the rules management in the kernel, or whether it better goes into userland.

cheers
  simon

--
  <3 the future  +++  RENT this banner advert  +++   ASCII Ribbon   /"\
  rock the past  +++  space for low CHF NOW!1  +++     Campaign     \ /
Party Enjoy Relax   |   http://dragonflybsd.org      Against  HTML   \
Dude 2c 2 the max   !   http://golden-apple.biz       Mail + News   / \