you need to call a program to adjust the permissions.
there is no big performance difference IMO between chown() and
mknod() + chown().
and with devfs you need to do that stuff everytime.
with a normal /dev, you only need to care if your kernel+modules
changes.
for example if you swithc between a plip interface and lp, it doesn't
hurt you to have both devices available. but with devfs, between every
switch the old inodes will be removed and new inodes added (and a call
to kerneld to adjust the permissions). no thanks.
> And I certainly have SCSI disc naming in my sights... With devfs you
> can have both the old-style /dev/sd{a,b,c} as well as something like
> the Solaris scheme:
that can also work pretty well without devfs.
(create entries in /proc/devlist, and call makedev to create the
corresponding devices).
andreas