From: Sven Conrad (SvenConrad@gmx.de)
Date: Sat Sep 02 2000 - 19:30:19 CEST
Hi Me again
Hmmmpf. I'am not sure. But I got some Ideas in this disskusion that
sound strange to me. This can allso be, because I missunderstand things,
so please don't worry if I critizise something which is totaly correct.
I want to help as good as possible and never blame someone.
> > Because, in this case, network.fun have to scan any configured
> > interface and ask the user for netmask/gateway/broadcast
No good Idea, because netmask/interface is setup stuff!
Netmask/gateway is routing. This is totaly different! Don't
mix this up. The easy going one interface/network stuff,
netmask is in both cases identical, but only in this special
case.
>
> I don't think so. It seems to me that a gateway is global to all the interfaces. It is set
> up with
Nope. Gateway routes can only address ONE interface per route. You can
have as
manny routes per inferface as you like.
> > Anycase, this requires (as you told) some duplication in help
> > messages about this parameters and theyr default values.
>
Yes, could be. This is why the idea of logical dependent fun is so
glamoures.
It is independent of the physics. E.g. a point-to-point can be done on a
eth,
no problem. But only Bus-interfaces can have Netmask/Broadcast things.
(BTW: I count tokenring as a bus, even if it is not exact, what it is)
> I think not, finally. I think that as this stuff will only be usefull for experimented
> users, it won't need more help messages.
>
Ok, what do we want? Much advaced netwok setup, but only understandable
by experts?
Don't like the idea, that mulinx become a pain for newbys. What do you
mean?
I think I don't got you.
> > Yes. I summarized with the keyword (peer?) any special parameters
> > for ppp/plip.
>
ppp is allmost a point-to-point but there is one problem. It is a
protocoll to handle
all ip stuff by it's own. So it call itself ipconfig. This is a big
difference between
ppp and plip. I nerver tryed out, if you can setup a serial line without
pppd and
more like plip.
> Yes, but I also think that it is not usefull to setup a network for plip in the interface
> setup, as it is quite unusual.
>
I gues the oposite. plip is a usual interface. You can do everything.
E.g. I have
done a connect from a eth-subnet to a big LAN with the WAN behind over
just one
plip. Impressing, I guaratee. It's fast enought for runnung X over it.
> > Please, discuss this topics:
> >
> > 1. I have to ask for netmask and broadcast for ANY interface?
>
Only for bus(eth) interfaces. Neither plip nor ppp have a netmask or
broadcast.
>
> > 2. Any interfaces live in own subnetwork, or no?
>
Yes. One interface is limited to one subnet because you can only assign
one
IP number to a interface. For eth this means, arp assigns the adapater
MAC to
this IP. This is allso vallid, when someone setups two subnets on one
collision
domain, (i.e. the same bus/cable/hub...) what is nonsens.
A interface without a subnet is just not usable.
> Not necessarily. I think you can have 2 interfaces sharing a network, and also interfaces
> without explicit network, in case of point to point communication.
>
I see a point-to-point connect like a subnet. This makes it easyer to
understand this for
me. But this subnet kowns no broadcast and there is only one box
reachable. But never the
less, you need your route to the partner.
> > 3. the 'route' command istruct the kernel to move packet toward
> > a specific interface, based on netmask. Now, is the ipfwadm
> > command enough to fix rules moving packets from an
> > interface to another, and what is the common policy?
> >
Try to change your standpoint. There is nothing like a link between
interfaces.
The kernel get one packet and does not care about what the source is.
(firewall
roules do, but not routing)
If the destinaion-IP is not one IP of this box and routing is enabled,
the kernel
ask the routing table, where this packet should be delivered to. This is
done
with masking out using netmasks and compare to netaddress. On maches, it
puts the
packed as is on the net or in case of a gateway route, send it to the
gateway.
Thats (nearby) all. Default gateways have netmask 0.0.0.0 and address
0.0.0.0
which allways match.
ipfwadm sets roules for routing. This can be done by sepecific source or
destination
address ranges. In case of NAT(==masqurading) the source IP of the
packed will be
replaced by the interface-IP.
Is this understandable? This viewpoint has been my major breakthroug in
understandig
networking.
---------------------------------------------------------------------
To unsubscribe, e-mail: mulinux-unsubscribe@sunsite.auc.dk
For additional commands, e-mail: mulinux-help@sunsite.auc.dk
This archive was generated by hypermail 2.1.6 : Sat Feb 08 2003 - 15:27:15 CET