The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2002-Aug> msg00108



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

LDP TCP/MD5 Option implementation understanding!

  • From: Morvan Daniel Müller <morvan@softplan.com.br>
  • Date: Thu, 22 Aug 2002 00:50:48 -0300
  • Cc: mpls@UU.NET
  • X-MIME-Autoconverted: from quoted-printable to 8bit by cell.onecall.net id g7M3ki313117
  • X-OriginalArrivalTime: 22 Aug 2002 03:48:19.0589 (UTC) FILETIME=[C13BA350:01C2498E]

Hi Eric!

>	If you can configure a different "secret" for
>each interface, then you probably could configure
>no secret for some interfaces, thus disabling the
>TCP MD5 Signature option for those interfaces.
>
>	Does this help?

Ok, it clarified!

Thank you very much!


At 18:53 21/08/2002 -0400, you wrote:
>Morvan,
>
>	The TCP MD5 Signature option works at a level
>below TCP.  In many cases, the option is enabled by 
>providing a "secret" to the TCP stack (possibly via
>ioctl, for example).  The extent to which this will
>apply - i.e. to interfaces or to the entire LSR - 
>depends on your implementation (in particular how
>TCP is bound to specific interfaces).  It is not a 
>good idea to have a single "secret" apply to every 
>interface in an LSR.  If this were done generally, 
>or even often, the secret would be quite out in the 
>open since many LSRs would end up using the same 
>"secret" for all of their interfaces - forcing the 
>same "secret" to be shared by a lot of LSRs.
>
>	If you can configure a different "secret" for
>each interface, then you probably could configure
>no secret for some interfaces, thus disabling the
>TCP MD5 Signature option for those interfaces.
>
>	Does this help?
>
>Eric W. Gray
>Systems Architect
>Celox Networks, Inc.
>egray@celoxnetworks.com
>508 305 7214
>
>
>-----Original Message-----
>From: Morvan Daniel Müller [mailto:morvan@softplan.com.br] 
>Sent: Wednesday, August 21, 2002 6:35 PM
>To: mpls@UU.NET
>Subject: LDP TCP/MD5 Option implementation understanding!
>
>Hello!
>
>Excuse me about erros, my english is poor :)!.
>
>At 17:18 13/08/2002 +0200, you wrote:
>>LDP Implementation Survey Form [V 1.0]
>>...
>>================+=============================+========================
>>TCP MD5 Option  | 2.9                         |
>>================+=============================+========================
>...
>>
>
>About "TCP MD5 Option" into LDP I have some doubts about inform locally, in
>each LSR, all possible peers and respective keys, like:
>LSR A local confuguration:
>LDP Authentication Option = enabled
>lsr-peer-idC : md5-keyC
>lsr-peer-idB : md5-keyB
>...
>
>Considering that "TCP MD5 Option" is a global option on the LSR, i.e., if 
>the authentication=ON this LSR only talk with peers who could authenticate
>(know the md5-key), OK?
>
>Consider a mpls domain:
>C
>|
>|
>A--- B ------- E
>|              |
>|              |
>D              F
>
>If you configure A with auth=ON. You need configure B, C and D with auth=ON
>and inform the md5-key, to permit talking between they.
>If B have auth=ON it only could speack with E if it have too auth=ON and
>know the Md5-key, on so on ...
>
>How is the MD5 Auth OPtion implemented in the LSR police:
>
>1) you could inform that any peer can speak without authenticate and only
>especific peers need authenticate. I think No, is no meaning!
>
>2) you could identify local domain sessions and permit it, and supply
>authentication for external domain sessions, or indirect (non adjacent)
>sessions? if so, how its possible identify such session? (I think in
>stack-bit, but local domains LSRs can too use label stacking for especial
>situations like traffic eng., etc..)
>
>3) Ldp authentication is based on the LSR physical interface? So you can
>suplly authentication for specific LSR interfaces (like LERs ingress
>interfaces) and for internal domain interfaces don't supply authentication.
>
>4) Its necessary configure manually all LSRs in the mpls domain to talk
>with each other, by inform all possible ldp_peers and respective md5-key.
>It will be a problem for scale?
>
>I may miss-understood something in this environment, please coul'd someone
>clarify me?
>
>Thanks, 
>
>Morvan.
>