Date: Fri, 12 Apr 1996 12:35:56 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@GndRsh.aac.dev.com> To: jgreco@brasil.moneng.mei.com (Joe Greco) Cc: current@FreeBSD.org Subject: Re: feedback sought on proposed change to netstart Message-ID: <199604121935.MAA21749@GndRsh.aac.dev.com> In-Reply-To: <199604121627.LAA17682@brasil.moneng.mei.com> from Joe Greco at "Apr 12, 96 11:27:09 am"
next in thread | previous in thread | raw e-mail | index | archive | help
> > > Rod,
> > >
> > > See my previous mail. What if we _remove_ the ifconfig in netstart in the
> > > case where there is a start_if.<xxx> file?
> > >
> > > Theory being, that script is entirely responsible for starting the
> > > interface. In my book that may include establishing a physical link,
> > > and/or
> > > setting up whatever addresses and aliases are appropriate.
Can't this all be done on one line???
/etc/sysconfig: ifconfig_de0="inet a.b.c.d netmask 0xffffff00 alias v.x.y.z
Will that not work correctly? [I have no idea, since I don't play with
aliased interfaces :-(]
> > > This seems like a very generalized solution to me (at least, it's what I do
> > > here, and it works great for me).
> >
> > That seems quite reasonable... I have done this by setting ifconfig_${INT}
> > to null in /etc/sysconfig. That does have the nasty side effect of of
> > echoing the config parameters out twice, but it does what you want.
> >
> > I have no objection to the removal of if_config ${ifn} ${ifconfig_args]
> > if an /etc/start_if.${ifn} is called.
>
> Hi Rod,
>
> Um, well. That would be an acceptable punt, I suppose, but in my "ideal"
> situation I would like to still set ifconfig_<xxx> in /etc/sysconfig..
> memory aid type thing. :-) (I have numerous boxes with multi interfaces)
My ``memory aid'' to remind me that I am actually setting these values up
some place else is to set the ifconfig_${int} to null in /etc/sysconfig, that
way I know I need to go look else where for the real stuff...
> Thinking out loud:
>
> 1) it does not make a lot of sense to set up a variable that has no effect,
> therefore setting ifconfig_<xxx> to nothing in sysconfig makes sense.
Yep... kinda why I do it.. though some times I forget to do this and then
pound my head against the wall for a while when I change it in /etc/sysconfig
and it seems to have no effect :-) :-).
> 2) it's still nice to see a high level explanation of interfaces in
> sysconfig... so it would be nice to propagate this data into the
> start_if.<xxx> thingie. So then we would keep the ifconfig_<xxx> thingie.
Since the start_if.${ifn} is a ``sourced'' shell script it has full access
to ${ifn} and ifconfig_${ifn}. You can't pass args on a ``.'' command
though. We could change the ``.'' to a ``sh'' and pass the args so it
was clearer that you had access to them in /etc/start_if.${ifn}.
> 3) that only (easily) works if start_if.xxx is a sourceable shell script..
> but the current construction makes that assumption anyways... maybe
> that's not a problem.
The problem is lack of documentation, perhaps a comment or even a sample
/etc/start_if.FOOBAR that showed how you can use ${ifn} and ifconfig_${ifn}
to do some fancier things. [I once had an /etc/start_if.slip that was
symlinked to by 3 /etc/start_if.sl{0,1,2}'s, this script used ${ifn} to figure
out what phone numbers to dial, and what modem ports to use with a case
statement, no reason to duplicate code :-)].
> 4) this still hides alias address declarations, it's also not immediately
> obvious at the sysconfig level that the interface configuration may actually
> be set elsewhere, and could be set to something else entirely.. but that is
> not a new problem, and it's not immediately apparent what the fix is.
See above, can't you just tack the alias stuff onto the end of the ifconfig
command, or must these be done as seperate ifconfig commands?
> I guess my conclusion is that this is all an ugly hack, this change only
> serves to make it somewhat more functional and flexible. Ideas on how to
> elegantly fix it? I don't have any, at least any that retain the same level
> of flexibility..
Here is one proposal... not much cleaner, it moves the ifconfig into the
/etc/start_if.${ifn} and does the eval of ifconfig_${ifn} to create
ifconfig_args before sourceing the script.
Index: netstart
===================================================================
RCS file: /home/ncvs/src/etc/netstart,v
retrieving revision 1.32.4.3
diff -c -r1.32.4.3 netstart
*** netstart 1995/08/25 07:21:46 1.32.4.3
--- netstart 1996/04/12 19:26:19
***************
*** 35,45 ****
# Set up all the network interfaces, calling startup scripts if needed
for ifn in ${network_interfaces}; do
if [ -e /etc/start_if.${ifn} ]; then
! . /etc/start_if.${ifn} ${ifn}
fi
- eval ifconfig_args=\$ifconfig_${ifn}
- ifconfig ${ifn} ${ifconfig_args}
ifconfig ${ifn}
done
--- 35,46 ----
# Set up all the network interfaces, calling startup scripts if needed
for ifn in ${network_interfaces}; do
+ eval ifconfig_args=\$ifconfig_${ifn}
if [ -e /etc/start_if.${ifn} ]; then
! . /etc/start_if.${ifn}
! else
! ifconfig ${ifn} ${ifconfig_args}
fi
ifconfig ${ifn}
done
You could also change the ``. /etc/start_if.${ifn}'' to
``sh /etc/start_if.${ifn} ${ifconfig_args}'' if you wanted to make
it clear that these are being passed in. This would remove the ability
of /etc/start_if.${ifn} to play games with other netstart variables though.
--
Rod Grimes rgrimes@gndrsh.aac.dev.com
Accurate Automation Company Reliable computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199604121935.MAA21749>