[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: udev device naming policy concerns

On Fri, Mar 05, 2004 at 05:45:55PM +0100, Tore Anderson wrote:
> * Tollef Fog Heen
>  > I think you are overestimating the problems.  I used to run without
>  > any compat symlinks, just devfs, and apart from the fact that you have
>  > to fix inittab, it mostly Just Worked.
>   When I experimented with devfs and attempted to run without the symlinks
>  my system broke down to the point of not booting fully, due to the fact
>  that the devices where my system partitions was supposed to be located
>  (/dev/hda1 and such) wasn't available any longer.  After having fixed
>  that I just found that many other things was broken, such as for instance
>  XMMS (due to the missing /dev/dsp and /dev/mixer), my floppy drive
>  was inaccessible (no /dev/fd0 either), and so on for almost every piece
>  of software using device nodes directly.  It surprises me to hear that
>  you didn't run into such problems, but then again, I did my testing for
>  quite some time ago - things may have changed since then.

Things certainly have. For me, the big changes I had to make were fstab,
lilo.conf, inittab. Large chunks of Debian is already devfs-aware, (I
dunno about xmms but) mpg123 for example will try both /dev/dsp and
/dev/sound/dsp. And of course, alsa's modules appear in the same place
under both devfs and normal operation. Apart from things like rng-tools,
psaux-using devices and microcode.ctl (for which I've locally moved the
devices out of /dev/misc/ since that's a meaningless distinction to me,
and now I can set those packages back to default) I haven't had to hunt
for device nodes in years.... (Well, cdrecord complains that open by
devicename is unsupported, but it also dislikes ide devices so even if I
did supply the nodes it wanted, it wouldn't be happy)

>   Anyway, my complaint isn't that it will be technically impossible to
>  build a system exclusively using non-standard device names, be it
>  devfs names or something else entirely, but rather that this attempt to
>  do so right now is undesireable and serves no good purpose.

>   The current naming scheme is, after all, a universally accepted and
>  ubiquitous standard, one which has been formalized by the Free Standards
>  Group, and one which no major player in the GNU/Linux arena seem intent
>  on changing in the foreseeable future.  I believe we should have a very
>  compelling rationale at the table before deciding to stray from it.

I think the devfs naming scheme is better personally, as it gives you a
structured view of what's in the system, and a structured way of dealing
with devices... (IE v4l stuff lives in /dev/video, easy.)

On the other hand, I can understand your point about not migrating away
from the documented, standardised approach, (ugly as it may be :-)...

The current default files create nodes in the devfs places, and symlink
to the kernel names for the ide/scsi stuff. Maybe it ought to create
nodes with the kernel names, and symlink to the devfs-style names? After
all, local configuration is quite esay to change. (In my case, I just
whacked the %k from the scsi/ide strings since I'm migrating from

And/or provide examples of different schemes. I suspect the old-style
naming config file would probably be a one line udev.rules:
KERNEL="*" NAME="%k"

Paul "TBBle" Hampson, MCSE
6th year CompSci/Asian Studies student, ANU
The Boss, Bubblesworth Pty Ltd (ABN: 51 095 284 361)

"No survivors? Then where do the stories come from I wonder?"
-- Capt. Jack Sparrow, "Pirates of the Caribbean"

This email is licensed to the recipient for non-commercial
use, duplication and distribution.

Attachment: signature.asc
Description: Digital signature

Reply to: