Re: [Syslog] Self-signed certs - was: Re: -transport-tls-12, section 4.2.3 (fingerprints)

Hi,

I think the real difference here is not "CA-issued certs" vs.
"self-signed certs", but "accepting any cert" vs. "accepting certs 
you can verify (as trusted peers according to your local policy)".

We definitely want to discourage blindly accepting any certificate
(CA-issued or self-signed); but when properly verified, self-signed
certificates are not any less secure than CA-issued ones.

Best regards,
Pasi

> -----Original Message-----
> From: Chris Lonvick
> Sent: 09 May, 2008 16:20
> To: Rainer Gerhards
> Cc: syslog@ietf.org
> Subject: [Syslog] Self-signed certs - was: Re: 
> -transport-tls-12, section 4.2.3 (fingerprints)
> 
> Hi,
> 
> On Thu, 8 May 2008, Rainer Gerhards wrote:
> <some elided for brevity>
> > However, I wonder why it would be useful to auto-generate certs.
> > Probably I am overlooking somehting obvious. But: isn't cert
> > auto-generation equal to no authentication? After all, if a
> > *self-signed* cert is generated by the remote peer AND we accept
> > it, doesn't that essentially mean we accept any peer because the
> > peer can put whatever it likes into the cert? I do not see why
> > this is any better than having no cert at all...
> 
> It minimally protects against masquerade and disclosure, two of the
> threats we agreed upon.  It will also provide a TCP-based transport
> for anyone who wishes/needs to have a mechanism to throttle the flow
> of packets for congestion control - something that you cannot do
> with the UDP transport.
> 
> Those are the reasons I can think of.  You do raise a good point by
> questioning this and I'd like to see some discussion from the WG.
> Are these reasons sufficient to keep self-signed certs in the
> specification?  If so, should specific comments be made about their
> use?
> 
> WG Chair Hat sort'a on, sort'a off: I'm thinking that a self-signed
> cert is the method of least effort to provide congestion control for
> syslog and it should be included in the document just for that
> reason.  This was the objection raised by the Transport ADs when
> they saw that syslog-transport-udp was the only REQUIRED transport.
> I agree that self-signed certs don't really provide good protection
> and that should be noted in the Security Considerations Section.  If
> you don't agree with this, please object now.
> 
> If you do agree with this, does the following text work:
> ===
> (Perhaps as a third paragraph in Section 4.2.4)
> 
> Self-signed certificates will provide minimal protection against
> modification and disclosure.  Their use will not provide effective
> protection against masqeurade unless they are used with certificate
> fingerprint authorization lists.  The use of self-signed
> certificates without certificate fingerprint authorization lists is
> NOT RECOMMENDED.  However since tls is a tcp-based protocol,
> enabling tls, even with self-signed certificates, will effectively
> enable congestion control in the network.  See Section 8.6 of
> [syslog-protocol].
> 
> And perhaps merge the first three sentences of the above with 
> the second paragraph in Sec Considerations section 5.1.  Current:
>     The use of self-signed certificates with certificate fingerprint
>     authorization lists provides more protection from 
>     masquerade and man-in-the-middle attacks than forgoing certificate 
>     validation and authorization.
> ===
> 
> Thanks,
> Chris
_______________________________________________
Syslog mailing list
Syslog@ietf.org
https://www.ietf.org/mailman/listinfo/syslog