Hi Bazsi,

Pardon me for not putting comments in-line but I'd like to summarize and
see where this discussion is going.  As I see it, various people have
said:

- 3195 is too heavy
- a minimal implementation (just say 'no') of 3195 may be light enough
- there are many tcp-based implementations to transport syslog but none
  of them interoperate
- a simple tcp-based implementation would need all of the aspects of
  BEEP to optionally call the security features that are desired
  (reliable transport, confidentiality, device authentication)
- BEEP channels can be a good thing

(Did I miss any other thoughts?)

I'm still not seeing what features/functions are missing from 3195, or
what features/functions a simple tcp-based syslog would fill.  I don't see
that documenting any/all of the old methods would serve a purpose either.

I'll keep this discussion open to see if we can get a clear reason for a
simple reliable tcp protocol.

Thanks,
Chris

On Fri, 20 Dec 2002, Balazs Scheidler wrote:

> Ccing to syslog-sec.
>
> On Fri, Dec 20, 2002 at 09:40:57AM -0800, Darren New wrote:
> > Balazs Scheidler wrote:
> >> I know I can do that, but there's no need. I wanted to explain that
> >> currently deployed syslog implementations don't do negotiation.
> >
> >Yes, I know that.
> >
> >The point is that there *is* a need, yes. What you meant to say is *you*
> >don't have that need. Which is fine, which is why it can be hardcoded in
> >*your* implementation. Others, making more generic implementations, need
> >to know whether the other side wants encryption, whether the other side
> >is using better timestamps, whether the other side is a creator of
> >messages or a relay or both, etc.
> >
> >> Look, there are implementations of syslog-over-TCP. There was a couple
> >> listed in another mail.
> >
> >So use them.
>
> I do. I want to get router logs on TCP, PIX does that fine (probably because
> of its CC audit), but other routers don't. Probably because there's no
> document on the protocol.
>
> >> You tell us that everything can be done in BEEP as
> >> well, that's true. But what are the _benefits_ of adding BEEP over the
> >> already implemented TCP transport?
> >
> >The ability to negotiate TLS, SASL, raw vs cooked, and the ability to
> >tunnel through firewalls to get your messages out. Did you attend any of
> >the meetings where the requirements were discussed? Did you read the
> >mailing list archives?
>
> Yes, I followed the mailing list, I even contributed to the RFC3164 effort.
> And no, I didn't attend any meetings because the meetings were held on a
> different continent.
>
> >> The problem is that those TCP implementations does not necessarily
> >> interoperate,
> >
> >Right. In part because they lack any negotiation.
>
> No, because there's no RFC on syslog over TCP, and everybody did it on its
> own.
>
> > > I'd argue that BEEP is viable if interoperation is the only benefit.
> >
> > It's not the only benefit. How would you, for example, implement a
> > syslog-over-TCP that could log into a firewall, authenticate itself,
> > then have the firewall pass that connection through to the collector on
> > the other side of the firewall, then have the device and collector
> > negotiate TLS so the firewall (if broken into) cannot see the messages
> > going thru it, in the current implementations of
> > syslog-over-TCP-without-negotiation?
>
> So BEEP can use TLS on one of its subchannels?  Last I heard it was only
> possible to wrap the whole BEEP stream into TLS, thus the whole argument is
> void as you'd need to use two BEEP streams (one ended on the
> firewall, the other on the collector), the same can be done using two
> independent TCP streams.
>
> Firewall authentication (e.g. authentication to cross the firewall) can be
> implemented in two different ways:
> * inband or inline auth where the authentication is hacked into the protocol
>   (as implemented by fwtk for instance where you have to auth twice in FTP:
>   first to the firewall, then to the server)
> * outband auth: when the authentication is based on a completely separate
>   channel
>
> The latter is something that you have in SASL defined as "external" method.
> This can mean many things: IPSec VPN, SSL wrapping, special outband protocol
> (this is also called session authentication in CheckPoint, or satyr
> authentication in Zorp)
>
> If you want automatic authentication on the firewall you either wrap the
> stream twice (first wrap is decrypted on the firewall, the second at the
> collector)
>
> > > > > For this it uses its own windowing, has its own channel control messages
> > > > > etc. This does not mean too much to syslog right now either. We can use
> > > > > several TCP streams if we need to.
> > > >
> > > > The overhead of handling these messages when there's only one window is
> > > > trivial.
> > >
> > > It is trivial, but is overhead (in both run-time and development time) which
> > > gives us nothing.
> >
> > Except that it doesn't give you nothing. It gives you the ability to do
> > more in the future. I'm sure when folks did syslog-over-UDP, they said
> > things like "TCP gives us overead we don't need."
>
> ... and we are left in the cold while BEEP based products are developed.
> I'm not arguing that BEEP is cool and is not necessary. I'd like to extend
> RFC3164 (or syslog-sign) with TCP transport.
>
> I might propose some text but I'd first like to be sure that it would be
> accepted.
>
> Chris, would proposed updates to syslog/sign be accepted?
>
> > > Certainly there are implementations of syslog-over-TCP.
> >
> > Right. But they don't do what you need. If they do what you need, great,
> > use em. If they're not interoperable enough for you, propose a new RFC.
> > Write it up and send it to the right WG. You know how the process works.
> > :-)
>
> Is this the right WG?
>
> > > I have implemented a syslog implementation, and also deployed it in
> > > demanding environments. Those implementations work *right* now.
> >
> > So explain again what the problem is?
>
> That there are some other plain TCP implementations, and as there's no RFC
> on this matter I'd doubt they interoperate. If they do without an RFC, it
> would still be wise to document it, just like it was the first
> accomplishment of this WG to release RFC3164.
>
> > > We *have* implementations, or implementations that are trivial to make
> > > RFC conformant.
> >
> > Um, trivially conformant to what RFC? Did I do all this syslog-over-TCP
> > RFC work for nothing? Is there another syslog-over-TCP RFC already
> > approved?
>
> I don't know about any. I meant that if there was an RFC which defined plain
> TCP transport (messages going one way, '\n' terminated lines which should be
> in the format that was described in rfc3164), _then_ many of the current
> implementations could be trivially modified to match that RFC.
>
> --
> Bazsi
> PGP info: KeyID 9AF8D0A9 Fingerprint CD27 CFB0 802C 0944 9CFD 804E C82C 8EB1
>
>


Reply via email to