Date: Fri, 22 Dec 2000 09:49:25 -0800 (PST) From: Archie Cobbs <archie@dellroad.org> To: John Smith <tiburon359@usa.net> Cc: freebsd-net@freebsd.org, freebsd-hackers@freebsd.org Subject: Re: [Re: New netgraph features?] Message-ID: <200012221749.eBMHnPG81236@curve.dellroad.org> In-Reply-To: <20001222074211.20124.qmail@nw174.netaddress.usa.net> "from John Smith at Dec 22, 2000 00:42:11 am"
next in thread | previous in thread | raw e-mail | index | archive | help
John Smith writes:
> >- We'd need to enhace the definition of a netgraph address
> > to include, say, an IP address, eg.:
> >
> > $ ngctl msg 192.168.1.12:foo: blah blah
>
> Well, I was thinking about this. I would like to share
> this pre-idea and to receive your opinions.
> I have one question here. Why should it be limited
> to UDP... (or whatever protocol)? Are we going to
> loose something, if we, say, create special node
> for 'netgraph tunneling' (so that it may ot may not
> be included into a running kernel) then connect
> this node to another one, which will be used for
> 'transport' layer. Such a node could possibly be
> used to encode/decode the inter-netgraph messages.
> Other nodes' names then should include
> the transport layer address. This way I think we
> won't get limited to one protcol...
Your idea of being protocol agnostic is more general.. however,
there is a slightly larger problem I didn't mention before.
Address syntax is parsed by the base code (ng_base.c). The
syntax of an address is not something an individual node gets
to decide.. so there are two possibilities..
You could do multi-host netgraph using a "private" addressing
scheme, simply by defining a control message that contained inside
it an IP address (or whatever), a netgraph address on the remote
machine, and a payload control message. Then write a node that
knows how to (de)encapsulate these control messages and send/recv
them over the network. But then you have something like this:
$ ngctl msg relaynode: { ip=192.168.1.12 addr="foo:" ... }
instead of this:
$ ngctl msg 192.168.1.12:foo: blah blah
To get the "cooler" case #2 behavior, there would need to be
"global" knowledge of the addressing scheme, which is of course
tied into the delivery protocol because the generalized netgraph
address must contain a protocol address (IP address or whatever).
Now.. we do have a syntax for describing a struct sockaddr of any
type (eg, ng_ksocket(4) node). So the protocol address could just
be an ASCII struct sockaddr and that would be as general as it gets.
-Archie
__________________________________________________________________________
Archie Cobbs * Packet Design * http://www.packetdesign.com
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200012221749.eBMHnPG81236>