On Sun, Oct 25, 2020 at 04:41:26PM +0000, Jason McIntyre wrote:
> On Sun, Oct 25, 2020 at 10:16:54AM -0600, Theo de Raadt wrote:
> > Jason McIntyre <j...@kerhand.co.uk> wrote:
> > 
> > > whereas /etc/netstart is actually doing:
> > > 
> > > - configure non-physical:                           (1)
> > > aggr trunk svlan vlan carp pppoe
> > > - routing                                           (2)
> > > - rest of non-physical:                             (3)
> > > tun tap gif etherip gre egre mobileip pflow wg
> > > 
> > > we could try to keep this list up to date, but it may be easier to just
> > > generally describe what netstart is doing.
> > 
> > I think we goes wrong by trying to maintain these as lists, and part of
> > where this goes wrong is weak definition of the reasons for the
> > ordering.  (Meaning, the developers who tweak netstart to handle the
> > concerns I'm about to describe, don't tend to think about the manual
> > page).
> > 
> > The (1) list of non-physical can probably be called "link-layer control
> > interfaces".  Or let's find a name for this.  These devices mutate the
> > presentation of other devices.  That's why their configuration needs to
> > be done before the physical device.
> > 
> > (2) The physical device is then brought up, including IP addressing. The
> > things in (1) need to be done beforehands, or the physical device is
> > participating in the wrong layer of network.
> > 
> > the (3) list of non-physical devices are layer-2 or layer-3 and operate
> > on devices which are already configured with some some sort of
> > "addressing" configured.
> > 
> > It would be nice to have our networking people come up with nice names
> > for group (1) and (2); words which succinctly describe the
> > classification like I've done above.  We need to increase understanding
> > of this order, rather than just abstractly listing names of devices with
> > complicated behaviours.
> > 
> > Once that is done, I still think it is problematic for us to list all
> > devices in each catagory:
> > 
> > a) new subsystems will be forgotten
> > b) the order of instantiation will sometimes be listed wrong -- for some
> >    of these the order is highly significant.
> > 
> > We can try to list as many as possible, but people who want the precise
> > list (and order) should look in the netstart code.  The lists will get
> > long and wrong.  If we find we cannot maintain the lists correctly
> > because it is duplicated information, man page wording like "such as"
> > could be used, also something which leads people to consider the script
> > source as authoritative, ie. have them go read the script 
> > 
> 
> ok, here is a start.
> 
> i have left the description as "non-physical", because i think that is
> clear. we could easily amend it. ifconfig.8 create talks about "network
> pseudo-devices" - that could be a possibility.
> 
> jmc
> 

an updated diff for this just got committed.
jmc

Reply via email to