devfs
[Top] [All Lists]

RE: devfs and raw devices again

To: "'Richard Gooch'" <rgooch@xxxxxxxxxxxxxxx>, jurgen@xxxxxxxx
Subject: RE: devfs and raw devices again
From: Borsenkow Andrej <Andrej.Borsenkow@xxxxxxxxxxxxxx>
Date: Tue, 28 May 2002 09:15:48 +0400
Cc: devfs@xxxxxxxxxxx
Importance: Normal
In-reply-to: <200205272210.g4RMAkJ22002@xxxxxxxxxxxxxxxxxxxxxxxx>
Sender: owner-devfs@xxxxxxxxxxx
> > I took a look at a couple of the patches that are floating around
> > to make raw devices work with devfs, and I found them all a bit
> > unsatisfying.  The problem is that they pre-generate a fixed
> > number of raw device files in devfs which don't actually do anything
> > until they are bound.  It seems to me that it would be much more
> > devfs-ish to just create the control device initially and then
> > generate raw device files as needed.  This is possible, but would
> > require some minor changes to the 'raw' utility.
> >
> > Here's how I think it should work... initially we just have...
> >
> >   /dev/raw/ctl
> >
> > To bind a raw device  for i.e. /dev/scsi/.../part2 you just do...
> >
> >   $ raw /dev/scsi/.../part2
> >   /dev/raw/1
> >
> >  ..the raw utility prints the name of the raw device it has bound,
> > which now appears in devfs.  In addition devfs now has symlinks...
> >
> >   /dev/scsi/.../rpart2 -> /dev/raw/1
> >   /dev/discs/disc0/rpart2 -> /dev/raw/1
> >
> >  ..etc.  Comments?
> 
> Sounds reasonable.
> 

I do not think so. Most applications for raw devices need fixed
predictable names.  Oracle admin hardly will be pleased with the fact
that his tablespaces change name on every reboot. 

Ability to add (or possibly free) raw devices online is useful though.

-andrej

<Prev in Thread] Current Thread [Next in Thread>