Thinking a little more about it, I am not sure if we the requirement to sign 
the same signature group with multiple algorithms by the same originator.  If 
we allow for multiple originators on the same host as I think we are saying we 
would, they each can have their own definitions of the same signature group and 
use a different algorithm.   Likewise, a single originator can define two 
signature groups with separate SPRIs that include the same messages, and sign 
each of them with a different algorithm.  With those capabilities, the space 
should be well covered, and in practice should be able to accomplish everything 
that we could by allowing multiple algorithms for the same signature group 
using the same identifier from the same originator on the same host, without 
requiring the associated completiy.  With this, I think the remainder of the 
issues you bring up are moot.  

On the last item, agree, let's make it "MUST NOT be signed".  

--- Alex

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Martin Schütte
Sent: Monday, August 04, 2008 6:50 AM
To: [email protected]
Subject: Re: [Syslog] Syslog-sign: Minor clarifications, part 2

Alexander Clemm (alex) schrieb:
> This leaves the question whether it should be possible to sign the
> *same* signature group with multiple algorithms.  There are currently 
> no provisions for that.  The spec also states that a payload block 
> (containing the certificate information) is valid for the entire 
> reboot session - we currently do not allow to change it during a 
> reboot session.  If this is an issue, it will have some ramifications.  
> One way to deal with it would be to change the way a "reboot session" 
> is defined.
> 
> Thoughts? </ALEX>

To allow multiple algorithms inside one signature group then one could use 
multiple SIGN params with prefixes to identify their algorithm. E.g. 
[ssign ... SIGN="1;<base64 signature DSA/SHA-1>" SIGN="2;<base64 signature 
DSA/SHA-256>"].

But I do not like the concept.
Just using two Signature Groups in parallel allows the same thing with greater 
flexibility and simpler implementations. (Maybe with more messages, but that is 
not negative for a protocol that has to be redundant by design,)

> This leads me to a problem in 4.2.3.: "In all cases, no more than 192 
> distinct Signature Groups (0-191) are permitted."
> 
> <ALEX> I am not sure I understand the issue?

I think the problem is we have different concepts of a "Signature Group". A 
definition would be useful.

The Introduction states "A Signature Group identifies a group of messages that 
are all kept together for signing purposes by the originator."
I would add the list of attributes that I consider the actual
identifiers: (HOSTNAME, APP-NAME, PROCID, MSGID, VER, RSID, SG, SPRI).

It follows: Two syslog-sign messages with the same values for these fields 
belong to the same Signature Group. Two syslog-sign messages which values 
differ in any of these fields do not belong to the same Signature Group.

>  The fact that the SPRI
> parameter may take only values from 0-191 inclusive already implies 
> there can be no more than 192 values?

Yes, indeed.
But that makes the sencence only more confusing, because what do the given 
numbers 0-191 mean if they do not refer to SPRI values?

   I think the statement is
> correct as is - the signature group refers to a set of syslog messages 
> while the SPRI value defines what messages that set contains
> - we do however want to call out that there can be no more than 192 
> such sets. </ALEX>

I am sorry if this appears like nit-picking, but do you say
- There can be no more (=follows logically from the definitions/construction)? 
or
- There should be no more (=there could be, but we set an arbitrary upper limit 
for interoperability)?

Maybe we can use an example to test the limits:
Say I want to use VER="0111" and VER="0121" in parallel so I have every message 
signed twice. And I also want to use SG="1". Then I have 384 concurrent 
Signature Groups.  Ennumerating with the identifies as above (HOSTNAME, 
APP-NAME, PROCID, MSGID are constant) gives the list:
   1. (..., VER="0111", RSID="123", SG="1", SPRI="0")
   2. (..., VER="0111", RSID="123", SG="1", SPRI="1") ...
192. (..., VER="0111", RSID="123", SG="1", SPRI="191") 193. (..., VER="0121", 
RSID="123", SG="1", SPRI="0") 194. (..., VER="0121", RSID="123", SG="1", 
SPRI="1") ...
384. (..., VER="0121", RSID="123", SG="1", SPRI="191")

Should this be a valid protocol use or not?

 > <ALEX> Section 4.1 states that syslog-sign messages are not counted
> and not signed, so I think the current position is clear.  Martin, I 
> agree with your reasoning and argumentation.  Essentially, the purpose 
> is to sign a stream of messages, not alter that stream of messages. 
> </ALEX>

Could we replace "do not need to be signed" by "MUST not be signed" then?

--
Martin
_______________________________________________
Syslog mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/syslog
_______________________________________________
Syslog mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/syslog

Reply via email to