From owner-l2tp@ipsec.org  Wed Jan  5 22:45:37 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA22029
	for <l2tpext-archive@odin.ietf.org>; Wed, 5 Jan 2000 22:45:33 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id RAA23337; Wed, 5 Jan 2000 17:01:30 -0800 (PST)
Received: from ans2-pub.dttus.com (ans2-pub.dttus.com [204.151.11.3]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id RAA23332 for <l2tp@ipsec.org>; Wed, 5 Jan 2000 17:01:01 -0800 (PST)
Received: by ans2-pub.dttus.com id TAA29984
  (InterLock SMTP Gateway 3.0 for l2tp@ipsec.org);
  Wed, 5 Jan 2000 19:28:16 -0600
Received: by ans2-pub.dttus.com (Internal Mail Agent-4);
  Wed, 5 Jan 2000 19:28:16 -0600
Received: by ans2-pub.dttus.com (Internal Mail Agent-3);
  Wed, 5 Jan 2000 19:28:16 -0600
X400-Received: by ans2-pub.dttus.com (Internal Mail Agent-2);
  Wed, 5 Jan 2000 19:28:16 -0600
X400-Received: by ans2-pub.dttus.com (Internal Mail Agent-1);
  Wed, 5 Jan 2000 19:28:16 -0600
X400-Mts-Identifier: [/c=US/admd=TeleMail/prmd=Deloitte/; 10C543873CD511F9-MTAUSCNT1DT]
Content-Identifier: 10C543873CD511F9
Content-Return: Allowed
X400-Content-Type: P2-1988 ( 22 )
Conversion: Allowed
Original-Encoded-Information-Types: IA5-Text
Disclose-Recipients: Prohibited
Alternate-Recipient: Allowed
X400-Originator: rverma@deloitte.com
X400-Recipients: non-disclosure;
Message-Id: <10C543873CD511F9*/c=US/admd=TeleMail/prmd=Deloitte/o=ccMailGW/s=Verma/g=Rohit/@MHS>
Date: 05 Jan 2000 17:01:37 -0600
From: Rohit Verma <rverma@dttus.com>
To: l2tp@ipsec.org
Subject: l2TP Diconnect Cause Codes
Mime-Version: 1.0
Content-Type: text/plain
Sender: owner-l2tp@sailpix.com
Precedence: bulk


     Does any one know of RADIUS extensions for L2TP Disconnect Cause Codes
     
     Regards
     -rv


From owner-l2tp@ipsec.org  Fri Jan 14 17:55:10 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA23431
	for <l2tpext-archive@odin.ietf.org>; Fri, 14 Jan 2000 17:55:09 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id NAA00505; Fri, 14 Jan 2000 13:41:24 -0800 (PST)
Received: from showboat.teradyne.com (showboat.teradyne.com [198.51.251.10]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id NAA00501 for <l2tp@ipsec.org>; Fri, 14 Jan 2000 13:41:12 -0800 (PST)
Received: from chorus.teradyne.com (chorus.teradyne.com [131.101.1.195])
	by showboat.teradyne.com (8.8.8+Sun/8.8.8) with ESMTP id QAA03345
	for <l2tp@ipsec.org>; Fri, 14 Jan 2000 16:40:52 -0500 (EST)
Received: from notes.teradyne.com (jaypeak.corp.teradyne.com [131.101.17.23]) by chorus.teradyne.com (8.8.8+Sun/8.7.1) with SMTP id QAA18327 for <l2tp@ipsec.org>; Fri, 14 Jan 2000 16:40:54 -0500 (EST)
Received: by notes.teradyne.com(Lotus SMTP MTA v4.6.3  (733.2 10-16-1998))  id 85256866.007757DB ; Fri, 14 Jan 2000 16:43:32 -0500
X-Lotus-FromDomain: TERADYNE
From: "David Wong" <David_Wong@notes.teradyne.com>
To: l2tp@ipsec.org
Message-ID: <85256866.00774B2E.00@notes.teradyne.com>
Date: Fri, 14 Jan 2000 16:40:14 -0500
Subject: L2TP over IPSec
Mime-Version: 1.0
Content-type: text/plain; charset=us-ascii
Content-Disposition: inline
Sender: owner-l2tp@sailpix.com
Precedence: bulk

From what I've read, it sounds like IPSec transport mode is preferred over
tunnel mode when one wants to protect L2TP traffic. Which IPSec mode is L2TP
"supposed" to operate over, or is it really dependent on the situation? Does
this mean that there will be quite a few implementations of both flavors out
there or will transport mode be encouraged?




From owner-l2tp@ipsec.org  Fri Jan 14 17:55:30 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA23442
	for <l2tpext-archive@odin.ietf.org>; Fri, 14 Jan 2000 17:55:29 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id LAA25912; Thu, 6 Jan 2000 11:44:44 -0800 (PST)
Received: from jindo.cisco.com (jindo.cisco.com [171.69.43.22]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA25904 for <l2tp@ipsec.org>; Thu, 6 Jan 2000 11:37:04 -0800 (PST)
Received: from uzura.cisco.com (uzura.cisco.com [161.44.2.77])
	by jindo.cisco.com (8.8.8/2.5.1/Cisco List Logging/8.8.8) with ESMTP id LAA09821;
	Thu, 6 Jan 2000 12:00:00 -0800 (PST)
Date: Thu, 6 Jan 2000 14:59:57 -0500 (EST)
From: Skip Booth <ebooth@cisco.com>
To: Bernard Aboba <aboba@internaut.com>
cc: "'William Dixon'" <wdixon@MICROSOFT.com>, l2tp@ipsec.org
Subject: RE: securing L2TP using IPSec
In-Reply-To: <002001bf3778$ef074ee0$298939cc@internaut.com>
Message-ID: <Pine.GSO.4.10.10001061454560.20565-100000@uzura.cisco.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-l2tp@sailpix.com
Precedence: bulk


I've been spending some time on this issue as of late and wanted to share my
thoughts.  I would like to spend some time on this next week at the bakeoff if
any cares to join me.  Certainly any comments on list are welcomed as well.

-Skip Booth

L2TP-Security draft, comments on IKE negotiations and filtering
----------

Since L2TP allows for either the LAC or the LNS to choose their UDP port a
combination of static and dynamic filters must be implemented.  The static
and dynamic filters are described below followed by how they should be
implemented.

Cases:

1) LAC uses fixed port (1701) and the LNS uses fixed port (1701)
2) LAC uses dynamic port (ANY) and the LNS uses fixed port (1701)
3) LAC uses fixed port (1701) and the LNS uses dynamic port (ANY)
4) LAC uses dynamic port (ANY) and the LNS uses dynamic port (ANY)

The other part that may be dynamic is the LNS's IP Address.  There have 
been discussions recently of the L2TP mailing list about the ambiguity of
in the L2TP RFC (2661) about whether the LNS is allowed to change their IP
Address as well as their port number.  It has appears to be the consensus
that most implementations of L2TP would allow for this behaviour.  It has
been noted that this would be a useful feature for load-balancing at the
LNS.  As a result I have depicted the most general case for L2TP tunnel
establishment below:

SCCRQ (LAC-IPAddr,  LACPort   -> LNS-IPAddr,  1701)
SCCRP (LNS-IPAddr1, LNSPort1, -> LAC-IPAddr,  LACPort)
SCCCN (LAC-IPAddr,  LACPort,  -> LNS-IPAddr1, LNSPort1)

To this point, there have been two proposals about how to setup static
filters to protect L2TP traffic.  For a history lesson I will show these
below and talk about why they are not satisfactory:

Note: When ANY is mentioned in the Port then a 0 should be negotiated for
the quick mode port.

1st Proposal (The one found in the l2tp-security draft):
------------
	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  ANY Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | ANY Port  |  1701     |    |   1701     |  ANY Port  |
	 +-----------------------+----+-------------------------+

Comments:
This proposal covers two of the (IMO most common) cases.  The first
case is when the LAC and LNS both stay of ports 1701 and the second case
when the LAC choses it's own port but the LNS stays on 1701.  It does not
satisfy any of the cases where the LNS chooses a new port or a new address.

2nd Proposal (Proposed by William Dixon):
------------
	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  Any Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     |    1701   |  ANY Port |    |  ANY Port  |   1701     |
	 +-----------------------+----+-------------------------+

Comments:
This proposal covers two cases.  The first case is when the LAC and LNS
both stay on port 1701 and the second case is when the LNS chooses a new
port and the LAC stays on 1701.  I believe this set of filters exposes the
security issues to the LNS.  Since the LNS filter definition must open his
ports up to ANY, then non-L2TP traffic can be sent to the LNS and IPSec
would not protect against this.  The only thing that this proposal insures
is that the LAC must originate the traffic on a source port of 1701, but
this does not necessary mean that it is L2TP traffic.  Clients are free to
choose any port they want.  It opens the LNS up to receiving any UDP
traffic such as TFTP, SNMP, Routing updates via RIP, etc.  Although you can
argue that for someone to do this, they have to be a) authenticated via IKE
which implies a level of trust, b) savy and malicious enough to know how to
send non L2TP traffic to the LNS with their source port set to 1701.
Although this is unlikely if there are better solutions present I don't
think this proposal should be our first choice.

Summary:
Neither Proposal 1 or 2 cover all of the cases.  IMO Proposal 1 covers the
more common cases and has no security holes while proposal 2 does.  Thus if
we want to define the simple, static filter case we should go with proposal
1.  I would also like to consider the following proposal which covers all
the cases at the expense of being somewhat more complex.

Proposal 3 (New and proposed by Me, although this follows closely what Bernard
has proposed)
----------

Definitions:
LACPort == Port LAC chooses to originate/receive L2TP traffic on.  This can
be a static port such as 1701 or a dynamic one assigned by the socket.  In
either case only L2TP is registered on this port.

LNSPort ==  Port LNS chooses to originate/receive L2TP traffic on.  This
can be the IANA assigned L2TP port number 1701 or can move to a new port
number on the SCCRP.

LNSAddr1 == New IP Address the LNS chooses before sending the SCCRP.  This
address should now be used by the LAC for all future traffic to the LNS.

ANY Port == Accept any port or a specific port during IKE negotiations.

Step 1:
-------
Protect the SCCRQ from the LAC to the LNS.  

LAC Implementation notes:
It is up to the implementation on whether the LAC filter needs to
be dynamically injected or defined statically.  Regardless the LAC MUST 
negotiate the specific port it has chosen for it's source port for L2TP.

LNS Implementation notes:
The LNS filter MAY be statically configured and MUST only be used to
respond to phase II quick mode negotiations.  Since the remote address is
unknown this filter MUST NOT be used to initiate a phase II.

LAC: Initiator, LNS: Responder

	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  Any Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | LACPort   |   1701    |    |    1701    |  ANY Port  |
	 +-----------------------+----+-------------------------+

A set of phase II SAs will exist once the quick mode negotiation finishes
and the resulting dynamic filter would be present at the LNS:

Prot 17, LACAddr, LACPort --> LNS Addr, 1701
Prot 17, LNSAddr, 1701    --> LAC Addr, LACPort

Step 2: 
-------
Protect the SCCRP sent from the LNS to the LAC.  In this case a new
phase II might need to be negotiated (port change or LNS addr changes) and
potentially a new phase 1 (if LNSAddr1 != LNSAddr).  If renegotiation occurs
the LNS now becomes the initiator.  The original Phase II and Phase I
SHOULD NOT be deleted since they may still be reused for new L2TP tunnels
between these peers.

LAC Implementation NOTES:

This should be a phase II responder only policy.  No phase II's should be
initiated from via a match on this filter.  It also MUST have a lower
priority than the filter defined in step 1 for the LAC since it is more
generic.  If a LAC chooses a source port of 1701 (LACPort == 1701), then
this filter MAY be statically defined.

LNS Implementation NOTES:
If the LNS decides to move to a new LNSAddr1 or LNSPort then L2TP MUST
inject a new filter with the new information.  This would result in a new
phase II being negotiated.  If LNSAddr1 != LNSAddr then a new phase 1 MUST
be negotiated with the LAC. If everything remains the same then the phase
I and phase II established in step 1 will be used to protect the SCCRP.  If
a new filter is injected this filter must have a higher priority than the
filter defined in step 1 since it is more specific.

LAC: Responder, LNS: Initiator

	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | ANY Addr  |    |  LNSAddr1  |  LAC Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | LACPort   |   ANY     |    |   LNSPort  |  LACPort   |
	 +-----------------------+----+-------------------------+

If the LNS changes either their source address or their source port, a new
set of phase II SAs will exist once the quick mode negotiation finishes
and the resulting dynamic filter would be present at the LNS:

Prot 17, LACAddr,  LACPort --> LNSAddr1, LNSPort
Prot 17, LNSAddr1, LNSPort --> LACAddr,  LACPort

NOTE: Any dynamically injected filters MUST be present until after L2TP
tears down the tunnel by sending a STOPCCN.  This is must be so to handle
the occasional phase 1/phase 2 renegotiation IKE performs.

Examples using proposal 3 (Both LAC and LNS chose ports):

SCCRQ (10.0.0.1,  5000 -> 10.0.0.2, 1701)
SCCRP (10.0.0.2,  6000 -> 10.0.0.1, 5000)
SCCCN (10.0.0.1,  5000 -> 10.0.0.2, 6000)

1) LAC send SCCRQ to LNS.  Matches outbound filter on LAC. Packet queued
or dropped by IPSec Engine and Phase 1 and 2 request are started.  L2TP on
the LAC installs a new filter into the IPSec filter database.  The filter
will allow IPSec to respond to quick mode negotiations to the LAC IP
address, with UDP packets sent to port 5000.

2) Phase 1 establishes between 10.0.0.1 and 10.0.0.2
3) Phase 2 is initiated from LAC to LNS with the proxy as follows:

	Local:	Addr = 10.0.0.1
		Prot = 17
		Port = 5000
	Remote: Addr = 10.0.0.2
		Prot = 17
		Port = 1701

This will be matched by the LNS and quick mode completes.

4) The SCCRQ is retransmitted from the LAC, protected by the SA established
in step 3 and received by the LNS.  L2TP chooses a new port number of 6000
and injects a filter into the IPSec filter database.  

5) The SCCRP is sent out to the LAC.  This packet matches the new filter
injected in step 4 and the LNS initiates a new phase 2 to the LAC.  The
proxy would be as follows:

	Local:	Addr = 10.0.0.2
		Prot = 17
		Port = 6000

	Remote: Addr = 10.0.0.1
		Prot = 17
		Port = 5000

This phase 2 exchange would match the newly injected filter which the LAC
install in step 1.  The phase 2 would complete successfully and all UDP traffic
between 10.0.0.1/5000 and 10.0.0.2/6000 would be protected by ipsec.

On Thu, 25 Nov 1999, Bernard Aboba wrote:

> a. The LNS IKE MM policy will look like: "accept MM from any IP addr".
> b. The LNS IKE QM policy will look like:
> 
> LNS Filters (static):
> 
> From <Any IP> to <LNS IP>, protocol UDP, source port Any, dest port 1701
> 
> Then, when the LNS completes the IKE MM negotiation, receives the first
> packet from the LAC, and chooses the port to reply on, but before sending
> the reply, it can plumb the following dynamic policy:
> 
> LNS Filters (dynamic):
> 
> From <LNS IP> to <LAC IP>, protocol UDP, source port <chosen>, dest port
> <chosen>
> 
> To avoid a race condition, the dynamic policy MUST be effective before the
> LNS sends its reply. Otherwise, the reply might not be IPSEC protected. It
> is possible that this could take so long that the LAC could time out and
> retransmit.
> 
> 
> 
> -----Original Message-----
> From: William Dixon [mailto:wdixon@microsoft.com]
> Sent: Wednesday, November 24, 1999 5:29 PM
> To: 'aboba@internaut.com'; adagio_v@pacbell.net; l2tp@ipsec.org
> Subject: RE: securing L2TP using IPSec
> 
> 
> Yes in the dynamic case the LAC can plumb filters based on it's chosen port.
> But the LAC then initiates IKE to the LNS (before sending any L2TP traffic).
> How will the LNS know to agree to this chosen source port?  Normally it
> can't.  The LNS policy would have to say:
>    From LAC IP, to LNS IP, Any source, Any dest port
> 
> You could configure each LAC with a source port range, so LNS policy is
> specific to the LAC IP for that range, or just specific to a source port
> range.
> 
> When many LACs are involved, inbound policy on the LNS just agrees to secure
> all UDP to/from any IP address.  Now for dedicated boxes, this might be ok.
> But the IPSec policy on the LNS would interfere with other IKE requests to
> secure different UDP traffic.
> 
> I should have also said that the solution presented dealt only with using
> IKE MM/QM.  Because the vendors at that discusion had all implemented MM/QM.
> Having multiple uses of IPSec to secure different UDP port applications
> brings up the issue that there is already a conflict with this approach -
> namely that IKE main mode policy can at best be IP address specific, not
> port specific.  The configuration presented has one main mode policy used to
> reply to all LACs.  There's nothing we can do about this in the RFC2401
> series documents with Main Mode & Quick Mode, except not to use those modes,
> instead use Aggressive Mode.
> 
> I'll have to go back to check, but I didn't think Aggressive Mode allowed
> the use of Certificate Request Payloads, just allows a cert to be presented.
> So this may work, but none of us had experience with this configuration.
> 
> Has anyone on the list used an Aggressive Mode configuration to secure L2TP?
> For the purpose of allowing other UDP services to be protected differently ?
> 
> Wm
> 
> ---------------
> >From <LAC IP> to <LNS IP>, protocol UDP, source port <chosen>, dest port
> Any
> >From <LNS IP> to <LAC IP>, protocol UDP, source port Any, dest port
> <chosen>
> 
> This will cover the case where the LNS and LAC *both* do dynamic port
> assignment.
> 
> -----Original Message-----
> From: Bernard Aboba [mailto:aboba@internaut.com]
> Sent: Wednesday, November 24, 1999 12:16 AM
> To: adagio_v@pacbell.net; l2tp@ipsec.org
> Subject: RE: securing L2TP using IPSec
> 
> 
> >How is it done?
> 
> Here's are some rough thoughts on how this might work:
> 
> LAC Filters (static):
> 
> >From <LAC IP> to <LNS IP>, protocol UDP, source port Any, dest port 1701
> >From <LNS IP> to <LAC IP>, protocol UDP, source port 1701, dest port Any
> 
> These filters will cause IKE to be kicked off when the LAC initiates
> L2TP to the LNS, and will also protect the return traffic if the LNS
> is also using static port assignment.
> 
> Prior to sending the first packet, the LAC can also plumb dynamic
> filter policy once it knows what port it will be sending on:
> 
> LAC Filters (dynamic):
> 
> >From <LAC IP> to <LNS IP>, protocol UDP, source port <chosen>, dest port
> Any
> >From <LNS IP> to <LAC IP>, protocol UDP, source port Any, dest port
> <chosen>
> 
> This will cover the case where the LNS and LAC *both* do dynamic port
> assignment.
> 
> Note that to avoid race conditions it is safest if the dynamic filter policy
> takes effect before the initial L2TP packet is sent by the LAC.
> This will ensure that after the initial IKE negotiation completes setting
> up the first IPSEC SA and the LNS responds, thereby kicking of a QM
> negotiation for the additional (dynamic) SA, that the LAC will be
> prepared for it. Otherwise, the second QM negotiation could fail.
> 
> 
> 



From owner-l2tp@ipsec.org  Fri Jan 14 18:16:36 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA23679
	for <l2tpext-archive@odin.ietf.org>; Fri, 14 Jan 2000 18:16:35 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id FAA02555; Fri, 14 Jan 2000 05:15:09 -0800 (PST)
Received: from ctron-dnm.ctron.com (firewall-user@ctron-dnm.cabletron.com [12.25.1.120]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id FAA02534 for <l2tp@ipsec.org>; Fri, 14 Jan 2000 05:14:02 -0800 (PST)
Received: (from uucp@localhost)
	by ctron-dnm.ctron.com (8.8.7/8.8.7) id IAA15685
	for <l2tp@ipsec.org>; Fri, 14 Jan 2000 08:17:21 -0500 (EST)
Received: from roc-mail2.cabletron.com(134.141.72.230) by ctron-dnm.ctron.com via smap (4.1)
	id xma015522; Fri, 14 Jan 00 08:16:06 -0500
Received: from new-exc1.ctron.com (NEW-EXC1 [134.141.200.123])
	by roc-mail2.ctron.com (8.8.7/8.8.7) with ESMTP id IAA04675
	for <l2tp@ipsec.org>; Fri, 14 Jan 2000 08:12:39 -0500 (EST)
Received: by new-exc1.ctron.com with Internet Mail Service (5.5.2448.0)
	id <C5FLPAH2>; Fri, 14 Jan 2000 13:12:52 -0000
Message-ID: <29752A74B6C5D211A4920090273CA3DCCDE8B2@new-exc1.ctron.com>
From: "Waters, Stephen" <Stephen.Waters@cabletron.com>
To: l2tp@ipsec.org
Subject: fragmentation
Date: Fri, 14 Jan 2000 13:12:49 -0000
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2448.0)
Content-Type: text/plain;
	charset="ISO-8859-1"
Sender: owner-l2tp@sailpix.com
Precedence: bulk


There used to be a lot of text somewhere in the L2TP drafts on fragmentation
and avoiding it by setting the virtual PPP MRU size appropriately.  Is this
being progressed?

Ideally, L2TP should detect the MTU of the underlying IP link (or ATM,
or...).  

In the case where L2TP is protected with IPSEC, the IPSEC architecture calls
for the support of DF propagation and PMTU discovery support. This is not
possible if L2TP has hidden the DF bit in side L2TP and PPP.

In our implementation, L2TP calculates the MTU on the outbound network, and
causes the virtual PPP instances to use an appropriate MRU size - it also
'asks' IPSEC how much overhead will be applied by the request IPSEC
proposal. 

IPSEC can set the DF independently of what L2TP has done, I guess, but
processing 'Too Big' ICMP messages will be hard without the participation of
L2TP.

Is this issue being worked in new drafts?

Cheers, Steve. 


From owner-l2tp@ipsec.org  Fri Jan 14 18:20:57 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA23736
	for <l2tpext-archive@odin.ietf.org>; Fri, 14 Jan 2000 18:20:56 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id GAA28157; Fri, 7 Jan 2000 06:23:45 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id GAA28150 for <l2tp@ipsec.org>; Fri, 7 Jan 2000 06:15:45 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id GAA01616;
	Fri, 7 Jan 2000 06:44:52 -0800 (PST)
Received: from cisco.com (townsley-home-1.cisco.com [161.44.99.51]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id JAA08642; Fri, 7 Jan 2000 09:42:32 -0500 (EST)
Message-ID: <3875F8A8.7998215E@cisco.com>
Date: Fri, 07 Jan 2000 09:31:04 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: l2tp@ipsec.org
CC: ebooth@cisco.com
Subject: [Fwd: RE: securing L2TP using IPSec (fwd)]
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


Posting for Skip Booth, who might be having some majordomo troubles with
ipsec.org. This is in response to the thread that Bernard and William
began before the holidays.

- Mark

---------- Forwarded message ----------
Date: Thu, 6 Jan 2000 22:57:47 -0500 (EST)
From: Skip Booth <ebooth@cisco.com>
To: l2tp@ipsec.org
Subject: RE: securing L2TP using IPSec (fwd)


I've been spending some time on this issue as of late and wanted to
share my
thoughts.  I would like to spend some time on this next week at the
bakeoff if
any cares to join me.  Certainly any comments on list are welcomed as
well.

-Skip Booth

L2TP-Security draft, comments on IKE negotiations and filtering
----------

Since L2TP allows for either the LAC or the LNS to choose their UDP port
a
combination of static and dynamic filters must be implemented.  The
static
and dynamic filters are described below followed by how they should be
implemented.

Cases:

1) LAC uses fixed port (1701) and the LNS uses fixed port (1701)
2) LAC uses dynamic port (ANY) and the LNS uses fixed port (1701)
3) LAC uses fixed port (1701) and the LNS uses dynamic port (ANY)
4) LAC uses dynamic port (ANY) and the LNS uses dynamic port (ANY)

The other part that may be dynamic is the LNS's IP Address.  There have 
been discussions recently of the L2TP mailing list about the ambiguity
of
in the L2TP RFC (2661) about whether the LNS is allowed to change their
IP
Address as well as their port number.  It has appears to be the
consensus
that most implementations of L2TP would allow for this behaviour.  It
has
been noted that this would be a useful feature for load-balancing at the
LNS.  As a result I have depicted the most general case for L2TP tunnel
establishment below:

SCCRQ (LAC-IPAddr,  LACPort   -> LNS-IPAddr,  1701)
SCCRP (LNS-IPAddr1, LNSPort1, -> LAC-IPAddr,  LACPort)
SCCCN (LAC-IPAddr,  LACPort,  -> LNS-IPAddr1, LNSPort1)

To this point, there have been two proposals about how to setup static
filters to protect L2TP traffic.  For a history lesson I will show these
below and talk about why they are not satisfactory:

Note: When ANY is mentioned in the Port then a 0 should be negotiated
for
the quick mode port.

1st Proposal (The one found in the l2tp-security draft):
------------
	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  ANY Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | ANY Port  |  1701     |    |   1701     |  ANY Port  |
	 +-----------------------+----+-------------------------+

Comments:
This proposal covers two of the (IMO most common) cases.  The first
case is when the LAC and LNS both stay of ports 1701 and the second case
when the LAC choses it's own port but the LNS stays on 1701.  It does
not
satisfy any of the cases where the LNS chooses a new port or a new
address.

2nd Proposal (Proposed by William Dixon):
------------
	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  Any Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     |    1701   |  ANY Port |    |  ANY Port  |   1701     |
	 +-----------------------+----+-------------------------+

Comments:
This proposal covers two cases.  The first case is when the LAC and LNS
both stay on port 1701 and the second case is when the LNS chooses a new
port and the LAC stays on 1701.  I believe this set of filters exposes
the
security issues to the LNS.  Since the LNS filter definition must open
his
ports up to ANY, then non-L2TP traffic can be sent to the LNS and IPSec
would not protect against this.  The only thing that this proposal
insures
is that the LAC must originate the traffic on a source port of 1701, but
this does not necessary mean that it is L2TP traffic.  Clients are free
to
choose any port they want.  It opens the LNS up to receiving any UDP
traffic such as TFTP, SNMP, Routing updates via RIP, etc.  Although you
can
argue that for someone to do this, they have to be a) authenticated via
IKE
which implies a level of trust, b) savy and malicious enough to know how
to
send non L2TP traffic to the LNS with their source port set to 1701.
Although this is unlikely if there are better solutions present I don't
think this proposal should be our first choice.

Summary:
Neither Proposal 1 or 2 cover all of the cases.  IMO Proposal 1 covers
the
more common cases and has no security holes while proposal 2 does.  Thus
if
we want to define the simple, static filter case we should go with
proposal
1.  I would also like to consider the following proposal which covers
all
the cases at the expense of being somewhat more complex.

Proposal 3 (New and proposed by Me, although this follows closely what
Bernard
has proposed)
----------

Definitions:
LACPort == Port LAC chooses to originate/receive L2TP traffic on.  This
can
be a static port such as 1701 or a dynamic one assigned by the socket. 
In
either case only L2TP is registered on this port.

LNSPort ==  Port LNS chooses to originate/receive L2TP traffic on.  This
can be the IANA assigned L2TP port number 1701 or can move to a new port
number on the SCCRP.

LNSAddr1 == New IP Address the LNS chooses before sending the SCCRP. 
This
address should now be used by the LAC for all future traffic to the LNS.

ANY Port == Accept any port or a specific port during IKE negotiations.

Step 1:
-------
Protect the SCCRQ from the LAC to the LNS.  

LAC Implementation notes:
It is up to the implementation on whether the LAC filter needs to
be dynamically injected or defined statically.  Regardless the LAC MUST 
negotiate the specific port it has chosen for it's source port for L2TP.

LNS Implementation notes:
The LNS filter MAY be statically configured and MUST only be used to
respond to phase II quick mode negotiations.  Since the remote address
is
unknown this filter MUST NOT be used to initiate a phase II.

LAC: Initiator, LNS: Responder

	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  Any Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | LACPort   |   1701    |    |    1701    |  ANY Port  |
	 +-----------------------+----+-------------------------+

A set of phase II SAs will exist once the quick mode negotiation
finishes
and the resulting dynamic filter would be present at the LNS:

Prot 17, LACAddr, LACPort --> LNS Addr, 1701
Prot 17, LNSAddr, 1701    --> LAC Addr, LACPort

Step 2: 
-------
Protect the SCCRP sent from the LNS to the LAC.  In this case a new
phase II might need to be negotiated (port change or LNS addr changes)
and
potentially a new phase 1 (if LNSAddr1 != LNSAddr).  If renegotiation
occurs
the LNS now becomes the initiator.  The original Phase II and Phase I
SHOULD NOT be deleted since they may still be reused for new L2TP
tunnels
between these peers.

LAC Implementation NOTES:

This should be a phase II responder only policy.  No phase II's should
be
initiated from via a match on this filter.  It also MUST have a lower
priority than the filter defined in step 1 for the LAC since it is more
generic.  If a LAC chooses a source port of 1701 (LACPort == 1701), then
this filter MAY be statically defined.

LNS Implementation NOTES:
If the LNS decides to move to a new LNSAddr1 or LNSPort then L2TP MUST
inject a new filter with the new information.  This would result in a
new
phase II being negotiated.  If LNSAddr1 != LNSAddr then a new phase 1
MUST
be negotiated with the LAC. If everything remains the same then the
phase
I and phase II established in step 1 will be used to protect the SCCRP. 
If
a new filter is injected this filter must have a higher priority than
the
filter defined in step 1 since it is more specific.

LAC: Responder, LNS: Initiator

	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | ANY Addr  |    |  LNSAddr1  |  LAC Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | LACPort   |   ANY     |    |   LNSPort  |  LACPort   |
	 +-----------------------+----+-------------------------+

If the LNS changes either their source address or their source port, a
new
set of phase II SAs will exist once the quick mode negotiation finishes
and the resulting dynamic filter would be present at the LNS:

Prot 17, LACAddr,  LACPort --> LNSAddr1, LNSPort
Prot 17, LNSAddr1, LNSPort --> LACAddr,  LACPort

NOTE: Any dynamically injected filters MUST be present until after L2TP
tears down the tunnel by sending a STOPCCN.  This is must be so to
handle
the occasional phase 1/phase 2 renegotiation IKE performs.

Examples using proposal 3 (Both LAC and LNS chose ports):

SCCRQ (10.0.0.1,  5000 -> 10.0.0.2, 1701)
SCCRP (10.0.0.2,  6000 -> 10.0.0.1, 5000)
SCCCN (10.0.0.1,  5000 -> 10.0.0.2, 6000)

1) LAC send SCCRQ to LNS.  Matches outbound filter on LAC. Packet queued
or dropped by IPSec Engine and Phase 1 and 2 request are started.  L2TP
on
the LAC installs a new filter into the IPSec filter database.  The
filter
will allow IPSec to respond to quick mode negotiations to the LAC IP
address, with UDP packets sent to port 5000.

2) Phase 1 establishes between 10.0.0.1 and 10.0.0.2
3) Phase 2 is initiated from LAC to LNS with the proxy as follows:

	Local:	Addr = 10.0.0.1
		Prot = 17
		Port = 5000
	Remote: Addr = 10.0.0.2
		Prot = 17
		Port = 1701

This will be matched by the LNS and quick mode completes.

4) The SCCRQ is retransmitted from the LAC, protected by the SA
established
in step 3 and received by the LNS.  L2TP chooses a new port number of
6000
and injects a filter into the IPSec filter database.  

5) The SCCRP is sent out to the LAC.  This packet matches the new filter
injected in step 4 and the LNS initiates a new phase 2 to the LAC.  The
proxy would be as follows:

	Local:	Addr = 10.0.0.2
		Prot = 17
		Port = 6000

	Remote: Addr = 10.0.0.1
		Prot = 17
		Port = 5000

This phase 2 exchange would match the newly injected filter which the
LAC
install in step 1.  The phase 2 would complete successfully and all UDP
traffic
between 10.0.0.1/5000 and 10.0.0.2/6000 would be protected by ipsec.

On Thu, 25 Nov 1999, Bernard Aboba wrote:

> a. The LNS IKE MM policy will look like: "accept MM from any IP addr".
> b. The LNS IKE QM policy will look like:
> 
> LNS Filters (static):
> 
> From <Any IP> to <LNS IP>, protocol UDP, source port Any, dest port 1701
> 
> Then, when the LNS completes the IKE MM negotiation, receives the first
> packet from the LAC, and chooses the port to reply on, but before sending
> the reply, it can plumb the following dynamic policy:
> 
> LNS Filters (dynamic):
> 
> From <LNS IP> to <LAC IP>, protocol UDP, source port <chosen>, dest port
> <chosen>
> 
> To avoid a race condition, the dynamic policy MUST be effective before the
> LNS sends its reply. Otherwise, the reply might not be IPSEC protected. It
> is possible that this could take so long that the LAC could time out and
> retransmit.
> 
> 
> 
> -----Original Message-----
> From: William Dixon [mailto:wdixon@microsoft.com]
> Sent: Wednesday, November 24, 1999 5:29 PM
> To: 'aboba@internaut.com'; adagio_v@pacbell.net; l2tp@ipsec.org
> Subject: RE: securing L2TP using IPSec
> 
> 
> Yes in the dynamic case the LAC can plumb filters based on it's chosen port.
> But the LAC then initiates IKE to the LNS (before sending any L2TP traffic).
> How will the LNS know to agree to this chosen source port?  Normally it
> can't.  The LNS policy would have to say:
>    From LAC IP, to LNS IP, Any source, Any dest port
> 
> You could configure each LAC with a source port range, so LNS policy is
> specific to the LAC IP for that range, or just specific to a source port
> range.
> 
> When many LACs are involved, inbound policy on the LNS just agrees to secure
> all UDP to/from any IP address.  Now for dedicated boxes, this might be ok.
> But the IPSec policy on the LNS would interfere with other IKE requests to
> secure different UDP traffic.
> 
> I should have also said that the solution presented dealt only with using
> IKE MM/QM.  Because the vendors at that discusion had all implemented MM/QM.
> Having multiple uses of IPSec to secure different UDP port applications
> brings up the issue that there is already a conflict with this approach -
> namely that IKE main mode policy can at best be IP address specific, not
> port specific.  The configuration presented has one main mode policy used to
> reply to all LACs.  There's nothing we can do about this in the RFC2401
> series documents with Main Mode & Quick Mode, except not to use those modes,
> instead use Aggressive Mode.
> 
> I'll have to go back to check, but I didn't think Aggressive Mode allowed
> the use of Certificate Request Payloads, just allows a cert to be presented.
> So this may work, but none of us had experience with this configuration.
> 
> Has anyone on the list used an Aggressive Mode configuration to secure L2TP?
> For the purpose of allowing other UDP services to be protected differently ?
> 
> Wm
> 
> ---------------
> >From <LAC IP> to <LNS IP>, protocol UDP, source port <chosen>, dest port
> Any
> >From <LNS IP> to <LAC IP>, protocol UDP, source port Any, dest port
> <chosen>
> 
> This will cover the case where the LNS and LAC *both* do dynamic port
> assignment.
> 
> -----Original Message-----
> From: Bernard Aboba [mailto:aboba@internaut.com]
> Sent: Wednesday, November 24, 1999 12:16 AM
> To: adagio_v@pacbell.net; l2tp@ipsec.org
> Subject: RE: securing L2TP using IPSec
> 
> 
> >How is it done?
> 
> Here's are some rough thoughts on how this might work:
> 
> LAC Filters (static):
> 
> >From <LAC IP> to <LNS IP>, protocol UDP, source port Any, dest port 1701
> >From <LNS IP> to <LAC IP>, protocol UDP, source port 1701, dest port Any
> 
> These filters will cause IKE to be kicked off when the LAC initiates
> L2TP to the LNS, and will also protect the return traffic if the LNS
> is also using static port assignment.
> 
> Prior to sending the first packet, the LAC can also plumb dynamic
> filter policy once it knows what port it will be sending on:
> 
> LAC Filters (dynamic):
> 
> >From <LAC IP> to <LNS IP>, protocol UDP, source port <chosen>, dest port
> Any
> >From <LNS IP> to <LAC IP>, protocol UDP, source port Any, dest port
> <chosen>
> 
> This will cover the case where the LNS and LAC *both* do dynamic port
> assignment.
> 
> Note that to avoid race conditions it is safest if the dynamic filter policy
> takes effect before the initial L2TP packet is sent by the LAC.
> This will ensure that after the initial IKE negotiation completes setting
> up the first IPSEC SA and the LNS responds, thereby kicking of a QM
> negotiation for the additional (dynamic) SA, that the LAC will be
> prepared for it. Otherwise, the second QM negotiation could fail.
> 
> 
>


From owner-l2tp@ipsec.org  Fri Jan 14 18:22:48 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA23753
	for <l2tpext-archive@odin.ietf.org>; Fri, 14 Jan 2000 18:22:47 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id TAA26380; Thu, 6 Jan 2000 19:38:53 -0800 (PST)
Received: from jindo.cisco.com (jindo.cisco.com [171.69.43.22]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id TAA26376 for <l2tp@ipsec.org>; Thu, 6 Jan 2000 19:31:00 -0800 (PST)
Received: from uzura.cisco.com (uzura.cisco.com [161.44.2.77])
	by jindo.cisco.com (8.8.8/2.5.1/Cisco List Logging/8.8.8) with ESMTP id TAA24225
	for <l2tp@ipsec.org>; Thu, 6 Jan 2000 19:57:48 -0800 (PST)
Date: Thu, 6 Jan 2000 22:57:47 -0500 (EST)
From: Skip Booth <ebooth@cisco.com>
To: l2tp@ipsec.org
Subject: RE: securing L2TP using IPSec (fwd)
Message-ID: <Pine.GSO.4.10.10001062256130.20618-100000@uzura.cisco.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-l2tp@sailpix.com
Precedence: bulk


I've been spending some time on this issue as of late and wanted to share my
thoughts.  I would like to spend some time on this next week at the bakeoff if
any cares to join me.  Certainly any comments on list are welcomed as well.

-Skip Booth

L2TP-Security draft, comments on IKE negotiations and filtering
----------

Since L2TP allows for either the LAC or the LNS to choose their UDP port a
combination of static and dynamic filters must be implemented.  The static
and dynamic filters are described below followed by how they should be
implemented.

Cases:

1) LAC uses fixed port (1701) and the LNS uses fixed port (1701)
2) LAC uses dynamic port (ANY) and the LNS uses fixed port (1701)
3) LAC uses fixed port (1701) and the LNS uses dynamic port (ANY)
4) LAC uses dynamic port (ANY) and the LNS uses dynamic port (ANY)

The other part that may be dynamic is the LNS's IP Address.  There have 
been discussions recently of the L2TP mailing list about the ambiguity of
in the L2TP RFC (2661) about whether the LNS is allowed to change their IP
Address as well as their port number.  It has appears to be the consensus
that most implementations of L2TP would allow for this behaviour.  It has
been noted that this would be a useful feature for load-balancing at the
LNS.  As a result I have depicted the most general case for L2TP tunnel
establishment below:

SCCRQ (LAC-IPAddr,  LACPort   -> LNS-IPAddr,  1701)
SCCRP (LNS-IPAddr1, LNSPort1, -> LAC-IPAddr,  LACPort)
SCCCN (LAC-IPAddr,  LACPort,  -> LNS-IPAddr1, LNSPort1)

To this point, there have been two proposals about how to setup static
filters to protect L2TP traffic.  For a history lesson I will show these
below and talk about why they are not satisfactory:

Note: When ANY is mentioned in the Port then a 0 should be negotiated for
the quick mode port.

1st Proposal (The one found in the l2tp-security draft):
------------
	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  ANY Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | ANY Port  |  1701     |    |   1701     |  ANY Port  |
	 +-----------------------+----+-------------------------+

Comments:
This proposal covers two of the (IMO most common) cases.  The first
case is when the LAC and LNS both stay of ports 1701 and the second case
when the LAC choses it's own port but the LNS stays on 1701.  It does not
satisfy any of the cases where the LNS chooses a new port or a new address.

2nd Proposal (Proposed by William Dixon):
------------
	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  Any Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     |    1701   |  ANY Port |    |  ANY Port  |   1701     |
	 +-----------------------+----+-------------------------+

Comments:
This proposal covers two cases.  The first case is when the LAC and LNS
both stay on port 1701 and the second case is when the LNS chooses a new
port and the LAC stays on 1701.  I believe this set of filters exposes the
security issues to the LNS.  Since the LNS filter definition must open his
ports up to ANY, then non-L2TP traffic can be sent to the LNS and IPSec
would not protect against this.  The only thing that this proposal insures
is that the LAC must originate the traffic on a source port of 1701, but
this does not necessary mean that it is L2TP traffic.  Clients are free to
choose any port they want.  It opens the LNS up to receiving any UDP
traffic such as TFTP, SNMP, Routing updates via RIP, etc.  Although you can
argue that for someone to do this, they have to be a) authenticated via IKE
which implies a level of trust, b) savy and malicious enough to know how to
send non L2TP traffic to the LNS with their source port set to 1701.
Although this is unlikely if there are better solutions present I don't
think this proposal should be our first choice.

Summary:
Neither Proposal 1 or 2 cover all of the cases.  IMO Proposal 1 covers the
more common cases and has no security holes while proposal 2 does.  Thus if
we want to define the simple, static filter case we should go with proposal
1.  I would also like to consider the following proposal which covers all
the cases at the expense of being somewhat more complex.

Proposal 3 (New and proposed by Me, although this follows closely what Bernard
has proposed)
----------

Definitions:
LACPort == Port LAC chooses to originate/receive L2TP traffic on.  This can
be a static port such as 1701 or a dynamic one assigned by the socket.  In
either case only L2TP is registered on this port.

LNSPort ==  Port LNS chooses to originate/receive L2TP traffic on.  This
can be the IANA assigned L2TP port number 1701 or can move to a new port
number on the SCCRP.

LNSAddr1 == New IP Address the LNS chooses before sending the SCCRP.  This
address should now be used by the LAC for all future traffic to the LNS.

ANY Port == Accept any port or a specific port during IKE negotiations.

Step 1:
-------
Protect the SCCRQ from the LAC to the LNS.  

LAC Implementation notes:
It is up to the implementation on whether the LAC filter needs to
be dynamically injected or defined statically.  Regardless the LAC MUST 
negotiate the specific port it has chosen for it's source port for L2TP.

LNS Implementation notes:
The LNS filter MAY be statically configured and MUST only be used to
respond to phase II quick mode negotiations.  Since the remote address is
unknown this filter MUST NOT be used to initiate a phase II.

LAC: Initiator, LNS: Responder

	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | LNS Addr  |    |  LNS Addr  |  Any Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | LACPort   |   1701    |    |    1701    |  ANY Port  |
	 +-----------------------+----+-------------------------+

A set of phase II SAs will exist once the quick mode negotiation finishes
and the resulting dynamic filter would be present at the LNS:

Prot 17, LACAddr, LACPort --> LNS Addr, 1701
Prot 17, LNSAddr, 1701    --> LAC Addr, LACPort

Step 2: 
-------
Protect the SCCRP sent from the LNS to the LAC.  In this case a new
phase II might need to be negotiated (port change or LNS addr changes) and
potentially a new phase 1 (if LNSAddr1 != LNSAddr).  If renegotiation occurs
the LNS now becomes the initiator.  The original Phase II and Phase I
SHOULD NOT be deleted since they may still be reused for new L2TP tunnels
between these peers.

LAC Implementation NOTES:

This should be a phase II responder only policy.  No phase II's should be
initiated from via a match on this filter.  It also MUST have a lower
priority than the filter defined in step 1 for the LAC since it is more
generic.  If a LAC chooses a source port of 1701 (LACPort == 1701), then
this filter MAY be statically defined.

LNS Implementation NOTES:
If the LNS decides to move to a new LNSAddr1 or LNSPort then L2TP MUST
inject a new filter with the new information.  This would result in a new
phase II being negotiated.  If LNSAddr1 != LNSAddr then a new phase 1 MUST
be negotiated with the LAC. If everything remains the same then the phase
I and phase II established in step 1 will be used to protect the SCCRP.  If
a new filter is injected this filter must have a higher priority than the
filter defined in step 1 since it is more specific.

LAC: Responder, LNS: Initiator

	    LAC Phase 2 Proxy		  LNS Phase 2 Proxy 
	    -----------------		  -----------------
	    Local	Remote            Local       Remote
	 +-----------------------+----+-------------------------+
Address  | LAC Addr  | ANY Addr  |    |  LNSAddr1  |  LAC Addr  |
	 +-----------------------+    +-------------------------+
Protocol |          17		 |    |           17            |
	 +-----------------------+    +-------------------------+
Port     | LACPort   |   ANY     |    |   LNSPort  |  LACPort   |
	 +-----------------------+----+-------------------------+

If the LNS changes either their source address or their source port, a new
set of phase II SAs will exist once the quick mode negotiation finishes
and the resulting dynamic filter would be present at the LNS:

Prot 17, LACAddr,  LACPort --> LNSAddr1, LNSPort
Prot 17, LNSAddr1, LNSPort --> LACAddr,  LACPort

NOTE: Any dynamically injected filters MUST be present until after L2TP
tears down the tunnel by sending a STOPCCN.  This is must be so to handle
the occasional phase 1/phase 2 renegotiation IKE performs.

Examples using proposal 3 (Both LAC and LNS chose ports):

SCCRQ (10.0.0.1,  5000 -> 10.0.0.2, 1701)
SCCRP (10.0.0.2,  6000 -> 10.0.0.1, 5000)
SCCCN (10.0.0.1,  5000 -> 10.0.0.2, 6000)

1) LAC send SCCRQ to LNS.  Matches outbound filter on LAC. Packet queued
or dropped by IPSec Engine and Phase 1 and 2 request are started.  L2TP on
the LAC installs a new filter into the IPSec filter database.  The filter
will allow IPSec to respond to quick mode negotiations to the LAC IP
address, with UDP packets sent to port 5000.

2) Phase 1 establishes between 10.0.0.1 and 10.0.0.2
3) Phase 2 is initiated from LAC to LNS with the proxy as follows:

	Local:	Addr = 10.0.0.1
		Prot = 17
		Port = 5000
	Remote: Addr = 10.0.0.2
		Prot = 17
		Port = 1701

This will be matched by the LNS and quick mode completes.

4) The SCCRQ is retransmitted from the LAC, protected by the SA established
in step 3 and received by the LNS.  L2TP chooses a new port number of 6000
and injects a filter into the IPSec filter database.  

5) The SCCRP is sent out to the LAC.  This packet matches the new filter
injected in step 4 and the LNS initiates a new phase 2 to the LAC.  The
proxy would be as follows:

	Local:	Addr = 10.0.0.2
		Prot = 17
		Port = 6000

	Remote: Addr = 10.0.0.1
		Prot = 17
		Port = 5000

This phase 2 exchange would match the newly injected filter which the LAC
install in step 1.  The phase 2 would complete successfully and all UDP traffic
between 10.0.0.1/5000 and 10.0.0.2/6000 would be protected by ipsec.

On Thu, 25 Nov 1999, Bernard Aboba wrote:

> a. The LNS IKE MM policy will look like: "accept MM from any IP addr".
> b. The LNS IKE QM policy will look like:
> 
> LNS Filters (static):
> 
> From <Any IP> to <LNS IP>, protocol UDP, source port Any, dest port 1701
> 
> Then, when the LNS completes the IKE MM negotiation, receives the first
> packet from the LAC, and chooses the port to reply on, but before sending
> the reply, it can plumb the following dynamic policy:
> 
> LNS Filters (dynamic):
> 
> From <LNS IP> to <LAC IP>, protocol UDP, source port <chosen>, dest port
> <chosen>
> 
> To avoid a race condition, the dynamic policy MUST be effective before the
> LNS sends its reply. Otherwise, the reply might not be IPSEC protected. It
> is possible that this could take so long that the LAC could time out and
> retransmit.
> 
> 
> 
> -----Original Message-----
> From: William Dixon [mailto:wdixon@microsoft.com]
> Sent: Wednesday, November 24, 1999 5:29 PM
> To: 'aboba@internaut.com'; adagio_v@pacbell.net; l2tp@ipsec.org
> Subject: RE: securing L2TP using IPSec
> 
> 
> Yes in the dynamic case the LAC can plumb filters based on it's chosen port.
> But the LAC then initiates IKE to the LNS (before sending any L2TP traffic).
> How will the LNS know to agree to this chosen source port?  Normally it
> can't.  The LNS policy would have to say:
>    From LAC IP, to LNS IP, Any source, Any dest port
> 
> You could configure each LAC with a source port range, so LNS policy is
> specific to the LAC IP for that range, or just specific to a source port
> range.
> 
> When many LACs are involved, inbound policy on the LNS just agrees to secure
> all UDP to/from any IP address.  Now for dedicated boxes, this might be ok.
> But the IPSec policy on the LNS would interfere with other IKE requests to
> secure different UDP traffic.
> 
> I should have also said that the solution presented dealt only with using
> IKE MM/QM.  Because the vendors at that discusion had all implemented MM/QM.
> Having multiple uses of IPSec to secure different UDP port applications
> brings up the issue that there is already a conflict with this approach -
> namely that IKE main mode policy can at best be IP address specific, not
> port specific.  The configuration presented has one main mode policy used to
> reply to all LACs.  There's nothing we can do about this in the RFC2401
> series documents with Main Mode & Quick Mode, except not to use those modes,
> instead use Aggressive Mode.
> 
> I'll have to go back to check, but I didn't think Aggressive Mode allowed
> the use of Certificate Request Payloads, just allows a cert to be presented.
> So this may work, but none of us had experience with this configuration.
> 
> Has anyone on the list used an Aggressive Mode configuration to secure L2TP?
> For the purpose of allowing other UDP services to be protected differently ?
> 
> Wm
> 
> ---------------
> >From <LAC IP> to <LNS IP>, protocol UDP, source port <chosen>, dest port
> Any
> >From <LNS IP> to <LAC IP>, protocol UDP, source port Any, dest port
> <chosen>
> 
> This will cover the case where the LNS and LAC *both* do dynamic port
> assignment.
> 
> -----Original Message-----
> From: Bernard Aboba [mailto:aboba@internaut.com]
> Sent: Wednesday, November 24, 1999 12:16 AM
> To: adagio_v@pacbell.net; l2tp@ipsec.org
> Subject: RE: securing L2TP using IPSec
> 
> 
> >How is it done?
> 
> Here's are some rough thoughts on how this might work:
> 
> LAC Filters (static):
> 
> >From <LAC IP> to <LNS IP>, protocol UDP, source port Any, dest port 1701
> >From <LNS IP> to <LAC IP>, protocol UDP, source port 1701, dest port Any
> 
> These filters will cause IKE to be kicked off when the LAC initiates
> L2TP to the LNS, and will also protect the return traffic if the LNS
> is also using static port assignment.
> 
> Prior to sending the first packet, the LAC can also plumb dynamic
> filter policy once it knows what port it will be sending on:
> 
> LAC Filters (dynamic):
> 
> >From <LAC IP> to <LNS IP>, protocol UDP, source port <chosen>, dest port
> Any
> >From <LNS IP> to <LAC IP>, protocol UDP, source port Any, dest port
> <chosen>
> 
> This will cover the case where the LNS and LAC *both* do dynamic port
> assignment.
> 
> Note that to avoid race conditions it is safest if the dynamic filter policy
> takes effect before the initial L2TP packet is sent by the LAC.
> This will ensure that after the initial IKE negotiation completes setting
> up the first IPSEC SA and the LNS responds, thereby kicking of a QM
> negotiation for the additional (dynamic) SA, that the LAC will be
> prepared for it. Otherwise, the second QM negotiation could fail.
> 
> 
> 




From owner-l2tp@ipsec.org  Fri Jan 14 18:44:22 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA23975
	for <l2tpext-archive@odin.ietf.org>; Fri, 14 Jan 2000 18:44:21 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id OAA00970; Fri, 14 Jan 2000 14:38:02 -0800 (PST)
Received: from wacko.redbacknetworks.com (wacko.redbacknetworks.com [155.53.130.200]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id OAA00966 for <l2tp@ipsec.org>; Fri, 14 Jan 2000 14:37:51 -0800 (PST)
Received: from tradrat.redbacknetworks.com (tradrat.redbacknetworks.com [155.53.190.107])
	by wacko.redbacknetworks.com (8.9.3/8.9.3) with ESMTP id OAA06643;
	Fri, 14 Jan 2000 14:37:20 -0800 (PST)
Received: (from tor@localhost) by tradrat.redbacknetworks.com (8.8.8/8.6.9) id OAA25458; Fri, 14 Jan 2000 14:37:20 -0800 (PST)
Date: Fri, 14 Jan 2000 14:37:19 -0800
From: Rene Tio <tor@redback.com>
To: David Wong <David_Wong@notes.teradyne.com>
Cc: l2tp@ipsec.org
Subject: Re: L2TP over IPSec
Message-ID: <20000114143719.C12237@tradrat.redbacknetworks.com>
References: <85256866.00774B2E.00@notes.teradyne.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.95.4us
In-Reply-To: <85256866.00774B2E.00@notes.teradyne.com>; from David Wong on Fri, Jan 14, 2000 at 04:40:14PM -0500
Sender: owner-l2tp@sailpix.com
Precedence: bulk

At the last WG meeting it was decided to use transport only.

On Fri, Jan 14, 2000 at 04:40:14PM -0500, David Wong wrote:
> >From what I've read, it sounds like IPSec transport mode is preferred over
> tunnel mode when one wants to protect L2TP traffic. Which IPSec mode is L2TP
> "supposed" to operate over, or is it really dependent on the situation? Does
> this mean that there will be quite a few implementations of both flavors out
> there or will transport mode be encouraged?
> 
> 


From owner-l2tp@ipsec.org  Mon Jan 17 01:31:51 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id BAA18201
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 01:31:50 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id VAA06530; Sun, 16 Jan 2000 21:28:15 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id VAA06526 for <l2tp@ipsec.org>; Sun, 16 Jan 2000 21:27:52 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id VAA20048
	for <l2tp@ipsec.org>; Sun, 16 Jan 2000 21:44:44 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-18.cisco.com [171.68.183.18]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id AAA11735 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 00:27:20 -0500 (EST)
Message-ID: <3882928E.44D4219E@cisco.com>
Date: Sun, 16 Jan 2000 22:54:54 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: l2tp@ipsec.org
Subject: Connect Speed AVP
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


An issue I forgot to cover in the town meeting is clarification of what
a Connect Speed of 0 means. Certainly, one should be liberal in
acceptance of this value. However, it would be better to define (1) if
it is legal to send the value, and (2) what is really means if it is
sent.

Some of my thoughts on this:

The Rx Connect Speed AVP MUST be sent, which I do not think should be
changed. Zero bandwidth could simply be defined as the value that should
be sent if/when the LAC is unable to determine the connect speed
properly.

That said, there is an edge case where perhaps zero is valid. e.g. a
Hybrid network where one stream is over the VPN an the other over
another path (such as a satellite or cable return feed). 

Input appreciated.

- Mark




From owner-l2tp@ipsec.org  Mon Jan 17 01:37:51 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id BAA18200
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 01:31:51 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id VAA06531; Sun, 16 Jan 2000 21:28:32 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id VAA06522 for <l2tp@ipsec.org>; Sun, 16 Jan 2000 21:27:51 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id VAA12482
	for <l2tp@ipsec.org>; Sun, 16 Jan 2000 21:30:05 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-18.cisco.com [171.68.183.18]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id AAA11731 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 00:27:06 -0500 (EST)
Message-ID: <3882902C.CBD81560@cisco.com>
Date: Sun, 16 Jan 2000 22:44:44 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: l2tp@ipsec.org
Subject: L2TP Minutes from VPN interoperability workshop town meeting
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


In general, L2TP interoperability was quite good. There are some issues 
for clarification:

Recommend at least 16 byte challenge in tunnel authentication

No one admitted to using Proxy LCP Initial Confreq AVP to change PPP
behavior now. 

When dropping out of order packets, it is a good idea to update Nr
first.

There was some question about whether or not to send a StopCCN when
Tunnel Authentication fails. Recommendation is to send the StopCCN.

Meaning of Framing Capabilities needs some clarification when sent from
an LNS that does not have any physical interfaces available for dialout.
The RFC currently makes a general statement disallowing any bit in the
Framing or Bearer Type AVPs to be set if the corresponding bit is not
set in the Capabilities AVP advertised by a peer. For outgoing calls
this makes good sense as framing is performed at the physical PPP link
connection point. For incoming calls, it does not make sense to be so
strict as, at best, the Framing Type is used to decide whether or not to
negotiate PPP options such as ACCM. While it may be important to dictate
whether or not to negotiate such link options, it is a bit strict to not
allow the call to even be placed.

It was noted that liberal implementations currently ignore the Framing
Capabilities AVP from an LNS when placing an incoming call request. Some
more strict implementations will not allow the call to be placed if the
corresponding bit is not set. The former promotes better
interoperability. In order to work with the more strict implementations,
it would be best for the LNS to always set both framing capabilities
bits.

The crux of the problem is that the text regarding the framing
capacities advertisement is overloaded for dialin and dialout.

Voluntary tunneling is even more confusing here as there is really no
physical connection point.

More clarification is necessary here. We will solicit opinions on what
is necessary now and for future extension.

Text will be clarified to allow only one bit to be set in the Framing
type and Bearer type AVPs.

SLI/ACCM race condition
Upon receipt of an SLI, an LAC might apply a new ACCM value before it
finishes tunneling LCP packets to the PPP client. If the PPP client is
strict about ACCM (as some are), the LCP packet could be dropped due to
invalid escaping, causing PPP negotiation to never converge.

Proposed fix for the SLI/ACCM race condition:
An SLI that is not clearing any ACCM bits at the LAC can always be
immediately applied If an SLI is received that will clear any ACCM bits,
the LAC goes into ACCM Pending state.  It sniffs all outgoing packets
until it sees a packet (from the LNS) with a Protocol ID less than
0xC000. LCP and the authentication protocols are all greater than
0xC000. Likewise, all of the NCPs and data protocols are less than
0xC000. Thus a packet with a Protocol ID less than 0xC000 indicates that
LCP is Open at both ends of the connection. At this point, the new ACCM
mask may be applied

Noted that we should still allow the source address to float during
tunnel setup. Need to pass this by the IESG. 

L2TP+IPSec

Skip Booth of cisco gave a presentation on IKE negotiation with L2TP, 
particularly in light of UDP port and IP address changes during L2TP 
tunnel establishment. 

Not a lot of comment as this was a lot to digest in this short meeting. 
Authors of draft-ietf-l2tp-security have agreed on this proposal and it 
will be included in the next rev of the draft to be posted shortly. 
Request was made to have the IPSec WG review what we are proposing. 
Further, this behavior is not unique to L2TP, there are other protocols 
which float UDP ports and the like which would exhibit the same
problems. 
Could this be handled more generically in IKE?




From owner-l2tp@ipsec.org  Mon Jan 17 13:15:53 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA05082
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 13:15:52 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA08655; Mon, 17 Jan 2000 09:23:56 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA08646 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 09:23:45 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA22569;
	Mon, 17 Jan 2000 09:40:39 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-28.cisco.com [171.68.183.28]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA12144; Mon, 17 Jan 2000 12:23:12 -0500 (EST)
Message-ID: <3883CBDC.BCAC6A51@cisco.com>
Date: Mon, 17 Jan 2000 21:11:40 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Sabrina Minshall <sabrina@accesscom.com>
CC: l2tp@ipsec.org
Subject: Re: ZLB
References: <200001171707.JAA21740@shell.accesscom.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


No.

Sabrina Minshall wrote:
> 
>    Do ZLB's require to be enqueued for retransmission?
> 
>    Thanks,
> 
>    Sabrina


From owner-l2tp@ipsec.org  Mon Jan 17 13:17:01 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA05095
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 13:17:00 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id JAA08616; Mon, 17 Jan 2000 09:08:00 -0800 (PST)
Received: from ns2.accesscom.com (root@ns2.accesscom.com [205.226.156.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA08612 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 09:07:50 -0800 (PST)
Received: from shell.accesscom.com (root@shell.accesscom.com [205.226.156.10])
	by ns2.accesscom.com (8.9.3/8.9.3/Debian/GNU) with ESMTP id JAA16613
	for <l2tp@ipsec.org>; Mon, 17 Jan 2000 09:07:36 -0800
Received: (from sabrina@localhost)
	by shell.accesscom.com (8.9.3/8.9.3/Debian/GNU) id JAA21740
	for l2tp@ipsec.org; Mon, 17 Jan 2000 09:07:35 -0800
From: Sabrina Minshall <sabrina@accesscom.com>
Message-Id: <200001171707.JAA21740@shell.accesscom.com>
Subject: ZLB
To: l2tp@ipsec.org
Date: Mon, 17 Jan 2000 09:07:35 -0800 (PST)
X-Mailer: ELM [version 2.4ME+ PL65 (25)]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

   Do ZLB's require to be enqueued for retransmission?


   Thanks,


   Sabrina


From owner-l2tp@ipsec.org  Mon Jan 17 14:26:31 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA06492
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 14:26:30 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA08678; Mon, 17 Jan 2000 09:28:28 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA08674 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 09:28:13 -0800 (PST)
Received: from comet (209.6.112.100) by mcfeeley.indusriver.com (Worldmail 1.3.167); 17 Jan 2000 12:26:34 -0500
Message-Id: <4.2.0.58.20000117122750.00a2dcc0@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Mon, 17 Jan 2000 12:31:38 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Connect Speed AVP
Cc: l2tp@ipsec.org
In-Reply-To: <3882928E.44D4219E@cisco.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

We implement LAC in the host without any WAN devices.
The PPP/L2TP is on stack.
The value for connect speed avp is set to 0 currently,
since it is not communication speed but function call speed.
--- David


At 10:54 PM 1/16/00 -0500, you wrote:

>An issue I forgot to cover in the town meeting is clarification of what
>a Connect Speed of 0 means. Certainly, one should be liberal in
>acceptance of this value. However, it would be better to define (1) if
>it is legal to send the value, and (2) what is really means if it is
>sent.
>
>Some of my thoughts on this:
>
>The Rx Connect Speed AVP MUST be sent, which I do not think should be
>changed. Zero bandwidth could simply be defined as the value that should
>be sent if/when the LAC is unable to determine the connect speed
>properly.
>
>That said, there is an edge case where perhaps zero is valid. e.g. a
>Hybrid network where one stream is over the VPN an the other over
>another path (such as a satellite or cable return feed).
>
>Input appreciated.
>
>- Mark




From owner-l2tp@ipsec.org  Mon Jan 17 20:09:21 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA10877
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 20:09:19 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id QAA09537; Mon, 17 Jan 2000 16:11:08 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id QAA09533 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 16:10:58 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id PAA27987;
	Mon, 17 Jan 2000 16:00:09 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id TAA12251; Mon, 17 Jan 2000 19:07:11 -0500 (EST)
Message-ID: <388454B9.8D0293E2@cisco.com>
Date: Tue, 18 Jan 2000 03:55:37 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: David Wong <David_Wong@notes.teradyne.com>
CC: l2tp@ipsec.org
Subject: Re: L2TP over IPSec
References: <85256866.00774B2E.00@notes.teradyne.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


Transport mode is encouraged. 

Can you identify a case where tunnel mode would be absolutely necessary?

David Wong wrote:
> 
> >From what I've read, it sounds like IPSec transport mode is preferred over
> tunnel mode when one wants to protect L2TP traffic. Which IPSec mode is L2TP
> "supposed" to operate over, or is it really dependent on the situation? Does
> this mean that there will be quite a few implementations of both flavors out
> there or will transport mode be encouraged?


From owner-l2tp@ipsec.org  Mon Jan 17 20:10:05 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA10892
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 20:10:02 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id QAA09547; Mon, 17 Jan 2000 16:12:06 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id QAA09543 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 16:11:43 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id QAA29198;
	Mon, 17 Jan 2000 16:02:26 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id TAA12247; Mon, 17 Jan 2000 19:04:38 -0500 (EST)
Message-ID: <38845420.B223DB43@cisco.com>
Date: Tue, 18 Jan 2000 03:53:04 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: "Waters, Stephen" <Stephen.Waters@cabletron.com>
CC: l2tp@ipsec.org
Subject: Re: fragmentation
References: <29752A74B6C5D211A4920090273CA3DCCDE8B2@new-exc1.ctron.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit





"Waters, Stephen" wrote:
> 
> There used to be a lot of text somewhere in the L2TP drafts on fragmentation
> and avoiding it by setting the virtual PPP MRU size appropriately.  Is this
> being progressed?

Draft 12 and RFC2661 both say:

   IP fragmentation may occur as the L2TP packet travels over the IP
   substrate.  L2TP makes no special efforts to optimize this.  A LAC
   implementation MAY cause its LCP to negotiate for a specific MRU,
   which could optimize for LAC environments in which the MTU's of the
   path over which the L2TP packets are likely to travel have a
   consistent value.

> 
> Ideally, L2TP should detect the MTU of the underlying IP link (or ATM,
> or...).
> 
> In the case where L2TP is protected with IPSEC, the IPSEC architecture calls
> for the support of DF propagation and PMTU discovery support. This is not
> possible if L2TP has hidden the DF bit in side L2TP and PPP.
> 
> In our implementation, L2TP calculates the MTU on the outbound network, and
> causes the virtual PPP instances to use an appropriate MRU size - it also
> 'asks' IPSEC how much overhead will be applied by the request IPSEC
> proposal.
> 
> IPSEC can set the DF independently of what L2TP has done, I guess, but
> processing 'Too Big' ICMP messages will be hard without the participation of
> L2TP.
> 
> Is this issue being worked in new drafts?

Not that I know of. In your case you were able to do all of this without
the use of new protocols, etc. This *could* be an additional
recommendation in the spec, but I see no need for it from an
interoperability standpoint as it is all done locally and with existing
mechanisms.

I would be happy to hear suggestions for a more dynamic method as well.
e.g. It seems that your use of PPP would require a renegotiation of LCP
in order to alter the MRU if the PMTU changed during the life of the
session (arguably something that likely does not occur all that often). 

Perhaps there is something that can be done via projection of the DF bit
from the outer to the inner header...

> 
> Cheers, Steve.


From owner-l2tp@ipsec.org  Mon Jan 17 20:41:16 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA11019
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 20:41:13 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id PAA09433; Mon, 17 Jan 2000 15:33:12 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA09429 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 15:32:57 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id PAA05109
	for <l2tp@ipsec.org>; Mon, 17 Jan 2000 15:34:27 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id SAA12222 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 18:31:21 -0500 (EST)
Message-ID: <38844C53.52D662ED@cisco.com>
Date: Tue, 18 Jan 2000 03:19:47 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: l2tp@ipsec.org
Subject: Vendor Specific Message Type 
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


I would like to propose additional text to indicate that the Message
Type AVP may be used to indicate a Vendor Specific message. I believe
this was the original intent, but the current text does not accurately
reflect this.

Early on, there was discussion of reserving zero for each vendor
specific AVP number space for use as a numeric identifier for Vendor
Type. The prevailing argument against this was that such a definitive
value would foster implementations to make operational decisions based
upon it, increasing complexity and reducing overall interoperabilty. 

Just to be clear, the idea is that if I want to create my own vendor
specific message, I simply include my SMI Enterprise Code value in the
Vendor ID field of the Message Type AVP. Then I can construct whatever
vendor specific AVPs (if any) I wish following it.

My questions are:

(1) Are there any implementations that currently define a vendor
specific AVP of zero?
(2) Would any of you break if you received a control message with a
nonzero Vendor ID in the Message Type AVP? 

Thanks,
 
- Mark


From owner-l2tp@ipsec.org  Mon Jan 17 22:10:08 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA12310
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 22:10:07 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id SAA09782; Mon, 17 Jan 2000 18:14:53 -0800 (PST)
Received: from bast.livingston.com (bast.livingston.com [149.198.247.2]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id SAA09778 for <l2tp@SailPix.com>; Mon, 17 Jan 2000 18:14:39 -0800 (PST)
Received: from server.livingston.com (server.livingston.com [149.198.1.70])
	by bast.livingston.com (8.9.3/8.9.3) with ESMTP id SAA03001
	for <l2tp@SailPix.com>; Mon, 17 Jan 2000 18:14:02 -0800 (PST)
Received: from kc.livingston.com ([149.198.1.4])
	by server.livingston.com (8.9.3/8.9.3/0.5) with ESMTP id SAA24793
	for <l2tp@SailPix.com>; Mon, 17 Jan 2000 18:14:08 -0800 (PST)
From: Jacob Teplitsky <jacobt@livingston.com>
Message-Id: <200001180214.SAA17818@kc.livingston.com>
Subject: Re: l2tp-digest V1000 #171
To: l2tp@SailPix.com
Date: Mon, 17 Jan 2000 18:14:08 -0800 (PST)
In-Reply-To: <200001171118.DAA08286@www.sailpix.com> from "l2tp-digest" at Jan 17, 2000 03:18:36 AM
X-Mailer: ELM [version 2.5 PL2]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

> Date: Sun, 16 Jan 2000 22:44:44 -0500
> From: "W. Mark Townsley" <townsley@cisco.com>
> Subject: L2TP Minutes from VPN interoperability workshop town meeting
> 
> 
> SLI/ACCM race condition
> Upon receipt of an SLI, an LAC might apply a new ACCM value before it
> finishes tunneling LCP packets to the PPP client. If the PPP client is
> strict about ACCM (as some are), the LCP packet could be dropped due to
> invalid escaping, causing PPP negotiation to never converge.
> 
> Proposed fix for the SLI/ACCM race condition:
> An SLI that is not clearing any ACCM bits at the LAC can always be
> immediately applied If an SLI is received that will clear any ACCM bits,
> the LAC goes into ACCM Pending state.  It sniffs all outgoing packets
> until it sees a packet (from the LNS) with a Protocol ID less than
> 0xC000. LCP and the authentication protocols are all greater than
> 0xC000. Likewise, all of the NCPs and data protocols are less than
> 0xC000. Thus a packet with a Protocol ID less than 0xC000 indicates that
> LCP is Open at both ends of the connection. At this point, the new ACCM
> mask may be applied
> 
Why LNS/PPP API can't delay sending SLI with ACCM bits until
it sends/receives a packet with a Protocol ID less than  0xC000 ?

Thanks
- Jacob


From owner-l2tp@ipsec.org  Mon Jan 17 22:10:31 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA12321
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 22:10:30 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id SAA09792; Mon, 17 Jan 2000 18:18:17 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id SAA09788 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 18:18:01 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id SAA20664;
	Mon, 17 Jan 2000 18:09:35 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id VAA12305; Mon, 17 Jan 2000 21:16:49 -0500 (EST)
Message-ID: <38851E95.C8C6E8AE@cisco.com>
Date: Tue, 18 Jan 2000 18:16:53 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Pyda Srisuresh <srisuresh@yahoo.com>
CC: David Wong <David_Wong@notes.teradyne.com>, l2tp@ipsec.org,
        ipsec@lists.tislabs.com
Subject: Re: L2TP over IPSec
References: <20000118021144.23584.qmail@web1405.mail.yahoo.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


Copying ipsec.

Pyda Srisuresh wrote:
> 
> --- "W. Mark Townsley" <townsley@cisco.com> wrote:
> >
> > Transport mode is encouraged.
> >
> > Can you identify a case where tunnel mode would be absolutely necessary?
> >
> 
> When L2TP tunnel end point is a tunnel mode IPsec gateway.
> It is easy for a VPN/IPsec gateway to implement tunnel mode IPsec for L2TP
> tunnels. Having to implement transport mode for L2TP is a bit of stretch
> for the security gateway.

Details please? Why is transport mode more difficult than tunnel mode? 

> 
> cheers,
> suresh
> 
> __________________________________________________
> Do You Yahoo!?
> Talk to your friends online with Yahoo! Messenger.
> http://im.yahoo.com


From owner-l2tp@ipsec.org  Mon Jan 17 22:10:54 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA12332
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 22:10:53 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id SAA09772; Mon, 17 Jan 2000 18:11:57 -0800 (PST)
Received: from web1405.mail.yahoo.com (web1405.mail.yahoo.com [128.11.23.169]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id SAA09768 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 18:11:45 -0800 (PST)
Received: (qmail 23585 invoked by uid 60001); 18 Jan 2000 02:11:44 -0000
Message-ID: <20000118021144.23584.qmail@web1405.mail.yahoo.com>
Received: from [38.186.47.232] by web1405.mail.yahoo.com; Mon, 17 Jan 2000 18:11:44 PST
Date: Mon, 17 Jan 2000 18:11:44 -0800 (PST)
From: Pyda Srisuresh <srisuresh@yahoo.com>
Subject: Re: L2TP over IPSec
To: "W. Mark Townsley" <townsley@cisco.com>,
        David Wong <David_Wong@notes.teradyne.com>
Cc: l2tp@ipsec.org
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Sender: owner-l2tp@sailpix.com
Precedence: bulk



--- "W. Mark Townsley" <townsley@cisco.com> wrote:
> 
> Transport mode is encouraged. 
> 
> Can you identify a case where tunnel mode would be absolutely necessary?
> 

When L2TP tunnel end point is a tunnel mode IPsec gateway.
It is easy for a VPN/IPsec gateway to implement tunnel mode IPsec for L2TP
tunnels. Having to implement transport mode for L2TP is a bit of stretch 
for the security gateway.

cheers,
suresh




__________________________________________________
Do You Yahoo!?
Talk to your friends online with Yahoo! Messenger.
http://im.yahoo.com


From owner-l2tp@ipsec.org  Mon Jan 17 23:32:15 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA13944
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 23:32:15 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id TAA09934; Mon, 17 Jan 2000 19:10:10 -0800 (PST)
Received: from databus.databus.com (databus.databus.com [198.186.154.34]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id TAA09927 for <l2tp@SailPix.com>; Mon, 17 Jan 2000 19:09:48 -0800 (PST)
From: Barney Wolff <barney@databus.com>
To: l2tp@SailPix.com
Date: Mon, 17 Jan 2000 21:43 EST
Subject: Re: l2tp-digest V1000 #171
Content-Type: text/plain
Message-ID: <3883d96f0.246@databus.databus.com>
Sender: owner-l2tp@SailPix.com
Precedence: bulk

Um, well, ordering of data vs control packets is not guaranteed by
the intervening transport medium.

There are really only three ways to go (that I can think of :)

1.  LAC snoops LCP, knows the right ACCM, snoops the protocol field,
applies appropriate ACCM depending on protocol.  Expensive for the LAC,
but it can ignore SLI.

2.  LNS sends SLI, waits for ack before sending first data with
new ACCM.  Hard, without an output queue that can be stalled.
Expensive, if you have to stall output every time you send an
LCP Echo.

3.  LNS sends SLI, sends data to which new ACCM should be applied,
and hopes packets don't get re-ordered.  LAC makes sure it does not
itself re-order packets, and hopes the transport medium didn't lose
or re-order the SLI.  The LAC must not be so dumb as to act on the
SLI before prior data packets have been sent out.  Maybe the LAC
is smart enough to send LCP with the default ACCM, even without
an SLI to tell it to.  I would have thought this would be a
requirement, to avoid the silliness of having to surround an LCP
Echo by SLIs, but I can't find it in 2661.  I seem to recall the
silicon-based life forms protesting.

After all, the worst that happens if the wrong ACCM is applied is
that the packet is ignored, and packets get lost once in a while
anyway.

Regards,
Barney Wolff

> From: Jacob Teplitsky <jacobt@livingston.com>
> Date: Mon, 17 Jan 2000 18:14:08 -0800 (PST)
> 
> Why LNS/PPP API can't delay sending SLI with ACCM bits until
> it sends/receives a packet with a Protocol ID less than  0xC000 ?


From owner-l2tp@ipsec.org  Mon Jan 17 23:49:00 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA14104
	for <l2tpext-archive@odin.ietf.org>; Mon, 17 Jan 2000 23:48:59 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id TAA10027; Mon, 17 Jan 2000 19:52:28 -0800 (PST)
Received: from ns2.accesscom.com (root@ns2.accesscom.com [205.226.156.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id TAA10023 for <l2tp@ipsec.org>; Mon, 17 Jan 2000 19:52:19 -0800 (PST)
Received: from shell.accesscom.com (root@shell.accesscom.com [205.226.156.10])
	by ns2.accesscom.com (8.9.3/8.9.3/Debian/GNU) with ESMTP id TAA10425
	for <l2tp@ipsec.org>; Mon, 17 Jan 2000 19:52:19 -0800
Received: (from sabrina@localhost)
	by shell.accesscom.com (8.9.3/8.9.3/Debian/GNU) id TAA28536
	for l2tp@ipsec.org; Mon, 17 Jan 2000 19:52:18 -0800
From: Sabrina Minshall <sabrina@accesscom.com>
Message-Id: <200001180352.TAA28536@shell.accesscom.com>
Subject: sequence num. synchronisation
To: l2tp@ipsec.org
Date: Mon, 17 Jan 2000 19:52:18 -0800 (PST)
X-Mailer: ELM [version 2.4ME+ PL65 (25)]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

How does one handle sequence number synchronisation or recovery
when the user space l2tp daemon dies and comes backup? Or, does 
it reset all the sessions?


Thanks,

sabrina


From owner-l2tp@ipsec.org  Tue Jan 18 03:32:04 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id DAA26987
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 03:32:03 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id XAA10302; Mon, 17 Jan 2000 23:30:52 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id XAA10298 for <l2tp@SailPix.com>; Mon, 17 Jan 2000 23:30:28 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id XAA26292;
	Mon, 17 Jan 2000 23:22:43 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-25.cisco.com [171.68.183.25]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id CAA12348; Tue, 18 Jan 2000 02:29:52 -0500 (EST)
Message-ID: <388567F4.5417F89D@cisco.com>
Date: Tue, 18 Jan 2000 23:29:56 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Barney Wolff <barney@databus.com>
CC: l2tp@SailPix.com
Subject: Re: l2tp-digest V1000 #171
References: <3883d96f0.246@databus.databus.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Barney Wolff wrote:
> 
> Um, well, ordering of data vs control packets is not guaranteed by
> the intervening transport medium.
> 
> There are really only three ways to go (that I can think of :)
> 
> 1.  LAC snoops LCP, knows the right ACCM, snoops the protocol field,
> applies appropriate ACCM depending on protocol.  Expensive for the LAC,
> but it can ignore SLI.

If you look at my proposal, it is essentially this, except:

1. The LAC never has to sniff if an SLI is not received
2. The LAC only sniffs a few packets, only after the SLI is received,
and only until it determines no more LCP packets will be sent, which is
likely for only a few packets and likely not during heavy traffic.

> 
> 2.  LNS sends SLI, waits for ack before sending first data with
> new ACCM.  Hard, without an output queue that can be stalled.
> Expensive, if you have to stall output every time you send an
> LCP Echo.

Uggh.

> 
> 3.  LNS sends SLI, sends data to which new ACCM should be applied,
> and hopes packets don't get re-ordered.  LAC makes sure it does not
> itself re-order packets, and hopes the transport medium didn't lose
> or re-order the SLI.  The LAC must not be so dumb as to act on the
> SLI before prior data packets have been sent out.  Maybe the LAC
> is smart enough to send LCP with the default ACCM, even without
> an SLI to tell it to.  I would have thought this would be a
> requirement, to avoid the silliness of having to surround an LCP
> Echo by SLIs, but I can't find it in 2661.  I seem to recall the
> silicon-based life forms protesting.

Not guaranteed. I am not 100% sure about ACCM treatment on LCP echos by
these "strict" PPP implementations. My thought is that they can handle
this (e.g. once ACCM negotiation is completed, it is OK), but you're
right, you can be completely sure. Trying to surround LCP echos is truly
silly and not guaranteed to work.

> 
> After all, the worst that happens if the wrong ACCM is applied is
> that the packet is ignored, and packets get lost once in a while
> anyway.

No, the worst is that the packet is *always* dropped, so LCP never
completes, and the call itself is dropped. 

> 
> Regards,
> Barney Wolff
> 
> > From: Jacob Teplitsky <jacobt@livingston.com>
> > Date: Mon, 17 Jan 2000 18:14:08 -0800 (PST)
> >
> > Why LNS/PPP API can't delay sending SLI with ACCM bits until
> > it sends/receives a packet with a Protocol ID less than  0xC000 ?


From owner-l2tp@ipsec.org  Tue Jan 18 09:40:43 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id JAA00121
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 09:40:42 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id FAA11920; Tue, 18 Jan 2000 05:30:00 -0800 (PST)
Received: from sioux.hosts.netdirect.net.uk (sioux.hosts.netdirect.net.uk [195.7.224.43]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id FAA11916 for <l2tp@ipsec.org>; Tue, 18 Jan 2000 05:29:40 -0800 (PST)
Received: from john (gw.mobileinternet.net [195.7.242.21] (may be forged))
	by sioux.hosts.netdirect.net.uk (8.9.3/8.9.3/Debian/GNU) with SMTP id NAA30008
	for <l2tp@ipsec.org>; Tue, 18 Jan 2000 13:28:27 GMT
Message-ID: <002501bf61b7$9e7dd7a0$0101010a@mobileinternet.com>
Reply-To: "John Bourke" <John.Bourke@MobileInternet.com>
From: "John Bourke" <John.Bourke@MobileInternet.com>
To: <l2tp@ipsec.org>
References: <200001180352.TAA28536@shell.accesscom.com>
Subject: LNS and Radius Serverr interactions
Date: Tue, 18 Jan 2000 13:26:15 -0000
Organization: Mobile Internet
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
X-Priority: 3
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook Express 5.00.2314.1300
X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2314.1300
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

Folks,

If I have many LACs and many LNS's, I may want to keep the shared secrets on
a Radius server.

Is there any provision in the current drafts to have the LNS pass shared
secret exchanges off to a Radius server ?

Has anyone implemented this ?

Once a L2TP  tunnel and session are up, and a ppp session has started,
during PPP authentication at the LNS, the LNS can use a Radius server to
check PPP user/password details.  Is there anything in this radius
interaction whihc allows me to identify the tunnel and session ?

The LNS then sends accounting information to the Radius server during the
PPP session.  The accounting information contains a reference to the L2TP
tunnel and session carrying the PPP session :-

4.1.  Acct-Tunnel-Connection
   Description
      This Attribute indicates the identifier assigned to the tunnel
      session.  It SHOULD be included in Accounting-Request packets
      which contain an Acct-Status-Type attribute having the value
      Start, Stop or any of the values described above.  This attribute,
      along with the Tunnel-Client-Endpoint and Tunnel-Server-Endpoint
      attributes [3], may be used to provide a means to uniquely
      identify a tunnel session for auditing purposes.
[...]
      The format of the identifier represented by the String field
      depends upon the value of the Tunnel-Type attribute [3].  For
      example, to fully identify an L2TP tunnel connection, the L2TP
      Tunnel ID and Call ID might be encoded in this field.  The exact
      encoding of this field is implementation dependent.

From the accounting information I could potentially have tunnel and session
is, but I cannot tie these back to the PPP authentication.

Should a mapping between the AVP's in the L2TP ICRQ and LNS Radius
interactions be defined ?


john




From owner-l2tp@ipsec.org  Tue Jan 18 11:13:27 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA01794
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 11:13:24 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id GAA12025; Tue, 18 Jan 2000 06:50:10 -0800 (PST)
Received: from databus.databus.com (databus.databus.com [198.186.154.34]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id GAA12018 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 06:49:52 -0800 (PST)
From: Barney Wolff <barney@databus.com>
To: l2tp@SailPix.com
Date: Tue, 18 Jan 2000 09:34 EST
Subject: Re: l2tp-digest V1000 #171
Content-Type: text/plain
Message-ID: <38847d8d0.83b@databus.databus.com>
Sender: owner-l2tp@SailPix.com
Precedence: bulk

> Date: Tue, 18 Jan 2000 23:29:56 -0800
> From: "W. Mark Townsley" <townsley@cisco.com>
> 
> Not guaranteed. I am not 100% sure about ACCM treatment on LCP echos by
> these "strict" PPP implementations. My thought is that they can handle
> this (e.g. once ACCM negotiation is completed, it is OK), but you're
> right, you can be completely sure. Trying to surround LCP echos is truly
> silly and not guaranteed to work.

RFC 1661 says (in section 5) that only LCP codes 1-7 must be sent with
the default options in effect, but I've met implementations that send
Echo Response with default ACCM, and have always worried that those
might not recognize an Echo Request with non-default ACCM.  But I
agree, surrounding an LCP Echo with SLIs feels silly.

> > After all, the worst that happens if the wrong ACCM is applied is
> > that the packet is ignored, and packets get lost once in a while
> > anyway.
> 
> No, the worst is that the packet is *always* dropped, so LCP never
> completes, and the call itself is dropped. 

I don't think so, because the SLI will eventually be received, unless
the transport has failed.  Of course an implementation with a small
retry counter may have given up, but that's always a risk - the
recommended timeout/retry values in 1661 result is giving up sooner
than the values in 2661 would.

Of course if packets are consistently re-ordered, there's no hope.

Barney Wolff


From owner-l2tp@ipsec.org  Tue Jan 18 11:25:30 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA02157
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 11:25:29 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id HAA12058; Tue, 18 Jan 2000 07:13:34 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id HAA12054 for <l2tp@ipsec.org>; Tue, 18 Jan 2000 07:13:18 -0800 (PST)
Received: from comet (209.6.112.100) by mcfeeley.indusriver.com (Worldmail 1.3.167); 18 Jan 2000 10:11:42 -0500
Message-Id: <4.2.0.58.20000118101416.00b16400@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Tue, 18 Jan 2000 10:16:49 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Vendor Specific Message Type 
Cc: l2tp@ipsec.org
In-Reply-To: <38844C53.52D662ED@cisco.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Our implementation ignores the message if
the VendorID is not in the vendor id list (include IETF's 0 )
However, if the M bit is set and the Vendor Id is not in the list then
the connection is dropped.
--- David

At 03:19 AM 1/18/00 -0800, you wrote:

>I would like to propose additional text to indicate that the Message
>Type AVP may be used to indicate a Vendor Specific message. I believe
>this was the original intent, but the current text does not accurately
>reflect this.
>
>Early on, there was discussion of reserving zero for each vendor
>specific AVP number space for use as a numeric identifier for Vendor
>Type. The prevailing argument against this was that such a definitive
>value would foster implementations to make operational decisions based
>upon it, increasing complexity and reducing overall interoperabilty.
>
>Just to be clear, the idea is that if I want to create my own vendor
>specific message, I simply include my SMI Enterprise Code value in the
>Vendor ID field of the Message Type AVP. Then I can construct whatever
>vendor specific AVPs (if any) I wish following it.
>
>My questions are:
>
>(1) Are there any implementations that currently define a vendor
>specific AVP of zero?
>(2) Would any of you break if you received a control message with a
>nonzero Vendor ID in the Message Type AVP?
>
>Thanks,
>
>- Mark




From owner-l2tp@ipsec.org  Tue Jan 18 12:20:22 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA03116
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 12:20:22 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id IAA12152; Tue, 18 Jan 2000 08:15:21 -0800 (PST)
Received: from springtide-t1-gw.mbo.ma.ultra.net (springtide-t1-gw.mbo.ma.ultra.net [209.6.2.90]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id IAA12148 for <l2tp@ipsec.org>; Tue, 18 Jan 2000 08:15:03 -0800 (PST)
Received: (qmail+freegate 22853 invoked by alias); 18 Jan 2000 16:15:01 -0000
Received: from unknown (HELO sol.springtidenet.com) (146.115.54.11)
  by 146.115.54.8 with SMTP; 18 Jan 2000 16:15:01 -0000
Received: by SOL with Internet Mail Service (5.5.2650.21)
	id <C9BCQ9WL>; Tue, 18 Jan 2000 11:14:33 -0500
Message-ID: <F10A2DA74539D311BD67009027719FD24B92@SOL>
From: "Phaneuf, Brian" <bphaneuf@springtidenet.com>
To: "'W. Mark Townsley'" <townsley@cisco.com>
Cc: l2tp@ipsec.org
Subject: RE: Vendor Specific Message Type 
Date: Tue, 18 Jan 2000 11:14:31 -0500
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2650.21)
Content-Type: text/plain;
	charset="iso-8859-1"
Sender: owner-l2tp@sailpix.com
Precedence: bulk

If we do not know the vendor or the message, then we react
according to the MBit in the AVP sent by the peer.
Close the connection if mandatory, ignore if it is optional.

-----Original Message-----
From: W. Mark Townsley [mailto:townsley@cisco.com]
Sent: Tuesday, January 18, 2000 6:20 AM
To: l2tp@ipsec.org
Subject: Vendor Specific Message Type 



I would like to propose additional text to indicate that the Message
Type AVP may be used to indicate a Vendor Specific message. I believe
this was the original intent, but the current text does not accurately
reflect this.

Early on, there was discussion of reserving zero for each vendor
specific AVP number space for use as a numeric identifier for Vendor
Type. The prevailing argument against this was that such a definitive
value would foster implementations to make operational decisions based
upon it, increasing complexity and reducing overall interoperabilty. 

Just to be clear, the idea is that if I want to create my own vendor
specific message, I simply include my SMI Enterprise Code value in the
Vendor ID field of the Message Type AVP. Then I can construct whatever
vendor specific AVPs (if any) I wish following it.

My questions are:

(1) Are there any implementations that currently define a vendor
specific AVP of zero?
(2) Would any of you break if you received a control message with a
nonzero Vendor ID in the Message Type AVP? 

Thanks,
 
- Mark


From owner-l2tp@ipsec.org  Tue Jan 18 13:26:43 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA04267
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 13:26:42 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id JAA12307; Tue, 18 Jan 2000 09:23:06 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA12303 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 09:22:47 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA04768;
	Tue, 18 Jan 2000 09:13:12 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-111.cisco.com [171.68.183.111]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA12609; Tue, 18 Jan 2000 12:20:26 -0500 (EST)
Message-ID: <3885F25D.1068D786@cisco.com>
Date: Wed, 19 Jan 2000 09:20:29 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Barney Wolff <barney@databus.com>
CC: l2tp@SailPix.com
Subject: Re: l2tp-digest V1000 #171
References: <38847d8d0.83b@databus.databus.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Barney Wolff wrote:
> 
> > Date: Tue, 18 Jan 2000 23:29:56 -0800
> > From: "W. Mark Townsley" <townsley@cisco.com>
> >
> > Not guaranteed. I am not 100% sure about ACCM treatment on LCP echos by
> > these "strict" PPP implementations. My thought is that they can handle
> > this (e.g. once ACCM negotiation is completed, it is OK), but you're
> > right, you can be completely sure. Trying to surround LCP echos is truly
> > silly and not guaranteed to work.
> 
> RFC 1661 says (in section 5) that only LCP codes 1-7 must be sent with
> the default options in effect, but I've met implementations that send
> Echo Response with default ACCM, and have always worried that those
> might not recognize an Echo Request with non-default ACCM.  But I
> agree, surrounding an LCP Echo with SLIs feels silly.
> 
> > > After all, the worst that happens if the wrong ACCM is applied is
> > > that the packet is ignored, and packets get lost once in a while
> > > anyway.
> >
> > No, the worst is that the packet is *always* dropped, so LCP never
> > completes, and the call itself is dropped.
> 
> I don't think so, because the SLI will eventually be received, unless

The problem is that the SLI *is* received. That is, the LNS negotiates
something other than the default map. When the LAC receives the SLI, it
blindly applies the new map, causing escaping to cease for some or all
characters. From this point forward, the strict PPP client will drop all
packets that do not have default escaping.

> the transport has failed.  Of course an implementation with a small
> retry counter may have given up, but that's always a risk - the
> recommended timeout/retry values in 1661 result is giving up sooner
> than the values in 2661 would.
> 
> Of course if packets are consistently re-ordered, there's no hope.
> 
> Barney Wolff


From owner-l2tp@ipsec.org  Tue Jan 18 13:55:19 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA04571
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 13:55:18 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA12389; Tue, 18 Jan 2000 09:58:05 -0800 (PST)
Received: from databus.databus.com (databus.databus.com [198.186.154.34]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id JAA12385 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 09:57:49 -0800 (PST)
From: Barney Wolff <barney@databus.com>
To: l2tp@SailPix.com
Date: Tue, 18 Jan 2000 12:44 EST
Subject: Re: l2tp-digest V1000 #171
Content-Type: text/plain
Message-ID: <3884a9990.b53@databus.databus.com>
Sender: owner-l2tp@SailPix.com
Precedence: bulk

I guess I don't understand.  You seem to be saying that the LAC acts
on the SLI when received, even though it has an output queue of PPP
frames that were received before the SLI, but have not yet been sent
out on the serial port (or equivalent).  I would call such behavior
a bug.  At any rate, there can be at most one output frame on queue
in the worrisome case, an LCP Conf-Ack, right?  Again, unless I
really don't understand the problem.

Barney

> Date: Wed, 19 Jan 2000 09:20:29 -0800
> From: "W. Mark Townsley" <townsley@cisco.com>
> 
> The problem is that the SLI *is* received. That is, the LNS negotiates
> something other than the default map. When the LAC receives the SLI, it
> blindly applies the new map, causing escaping to cease for some or all
> characters. From this point forward, the strict PPP client will drop all
> packets that do not have default escaping.


From owner-l2tp@ipsec.org  Tue Jan 18 16:15:27 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA06045
	for <l2tpext-archive@odin.ietf.org>; Tue, 18 Jan 2000 16:15:26 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id LAA12806; Tue, 18 Jan 2000 11:50:22 -0800 (PST)
Received: from bast.livingston.com (bast.livingston.com [149.198.247.2]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA12802 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 11:50:08 -0800 (PST)
Received: from server.livingston.com (server.livingston.com [149.198.1.70])
	by bast.livingston.com (8.9.3/8.9.3) with ESMTP id LAA16530
	for <l2tp@SailPix.com>; Tue, 18 Jan 2000 11:49:31 -0800 (PST)
Received: from kc.livingston.com ([149.198.1.4])
	by server.livingston.com (8.9.3/8.9.3/0.5) with ESMTP id LAA00717
	for <l2tp@SailPix.com>; Tue, 18 Jan 2000 11:49:37 -0800 (PST)
From: Jacob Teplitsky <jacobt@livingston.com>
Message-Id: <200001181949.LAA00971@kc.livingston.com>
Subject: SLI/ACCM race condition
To: l2tp@SailPix.com
Date: Tue, 18 Jan 2000 11:49:37 -0800 (PST)
In-Reply-To: <200001181118.DAA11727@www.sailpix.com> from "l2tp-digest" at Jan 18, 2000 03:18:33 AM
X-Mailer: ELM [version 2.5 PL2]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

> Date: Tue, 18 Jan 2000 23:29:56 -0800
> From: "W. Mark Townsley" <townsley@cisco.com>
> Subject: Re: l2tp-digest V1000 #171
> 
> Barney Wolff wrote:
> > 
> > Um, well, ordering of data vs control packets is not guaranteed by
> > the intervening transport medium.
> > 
> > There are really only three ways to go (that I can think of :)

Well, we can add a new AVP to SLI:
- data packet AVP. The last LCP ACK that finish the LCP negotiations
 can be sent with the SLI message.
................

> > After all, the worst that happens if the wrong ACCM is applied is
> > that the packet is ignored, and packets get lost once in a while
> > anyway.
> 
> No, the worst is that the packet is *always* dropped, so LCP never
> completes, and the call itself is dropped. 

If we use data packet AVP, this can't happen.

Thanks
- Jacob


From owner-l2tp@ipsec.org  Wed Jan 19 14:37:16 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA04290
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 14:37:11 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id MAA12845; Tue, 18 Jan 2000 12:33:01 -0800 (PST)
Received: from coredump.ae.usr.com (coredump.ae.usr.com [207.24.169.214]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id MAA12841 for <l2tp@ipsec.org>; Tue, 18 Jan 2000 12:32:50 -0800 (PST)
Received: from otherland (otherland.ae.usr.com [207.24.169.127])
	by coredump.ae.usr.com (8.9.2/8.9.2) with SMTP id OAA20996
	for <l2tp@ipsec.org>; Tue, 18 Jan 2000 14:32:18 -0600 (CST)
From: "Mike Wronski" <mike@coredump.ae.usr.com>
To: <l2tp@ipsec.org>
Subject: Training?
Date: Tue, 18 Jan 2000 14:32:15 -0600
Message-ID: <000901bf61f3$1b2acc00$7fa918cf@otherland.ae.usr.com>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
X-Priority: 3 (Normal)
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook 8.5, Build 4.71.2173.0
Importance: Normal
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

Does anyone have recommendations for training courses that cover VPN, L2TP,
IPSEC and policies at a low(bit) level? Something comprehensive for the
engineer and not the Executive.

-
Mike Wronski
3Com Network Systems Engineer



From owner-l2tp@ipsec.org  Wed Jan 19 15:09:44 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA04626
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 15:09:43 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id KAA00218; Wed, 19 Jan 2000 10:46:40 -0800 (PST)
Received: from bast.livingston.com (bast.livingston.com [149.198.247.2]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id KAA00209 for <l2tp@SailPix.com>; Wed, 19 Jan 2000 10:44:50 -0800 (PST)
Received: from server.livingston.com (server.livingston.com [149.198.1.70])
	by bast.livingston.com (8.9.3/8.9.3) with ESMTP id OAA18969;
	Tue, 18 Jan 2000 14:09:04 -0800 (PST)
Received: from kc.livingston.com ([149.198.1.4])
	by server.livingston.com (8.9.3/8.9.3/0.5) with ESMTP id OAA09095;
	Tue, 18 Jan 2000 14:09:10 -0800 (PST)
From: Jacob Teplitsky <jacobt@livingston.com>
Message-Id: <200001182209.OAA03806@kc.livingston.com>
Subject: Re: SLI/ACCM race condition
To: townsley@cisco.com (W. Mark Townsley)
Date: Tue, 18 Jan 2000 14:09:10 -0800 (PST)
Cc: jacobt@livingston.com (Jacob Teplitsky), l2tp@SailPix.com
In-Reply-To: <38862CAD.4C8F9AA4@cisco.com> from "W. Mark Townsley" at Jan 19, 2000 01:29:17 PM
X-Mailer: ELM [version 2.5 PL2]
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

In any case I would like to add the requirement that the LNS 
SHOULD send
the last LCP ACK that finish the LCP negotiations and the SLI message
in the right order.
- Jacob
> 
> 
> 
> Jacob Teplitsky wrote:
> > 
> > > Date: Tue, 18 Jan 2000 23:29:56 -0800
> > > From: "W. Mark Townsley" <townsley@cisco.com>
> > > Subject: Re: l2tp-digest V1000 #171
> > >
> > > Barney Wolff wrote:
> > > >
> > > > Um, well, ordering of data vs control packets is not guaranteed by
> > > > the intervening transport medium.
> > > >
> > > > There are really only three ways to go (that I can think of :)
> > 
> > Well, we can add a new AVP to SLI:
> > - data packet AVP. The last LCP ACK that finish the LCP negotiations
> >  can be sent with the SLI message.
> 
> This corrects the race condition at its core by packaging the two
> contenders into a single packet so that one cannot beat the other to the
> finish. However, it is a considerable protocol change which would set us
> back interoperabilty-wise. Sniffing a few packets at the LAC, only when
> the SLI is received, and only for a short time, may be better given the
> potential disruption. Further, if we are going to change RFCs, I would
> like to see changes in the PPP RFC to reflect what conventional wisdom
> in the PPP group already seems to be (e.g. be liberal in acceptance of
> unescaped characters). Either protocol change would fix the problem
> theoretically, but niether would help our situation much given the
> current level of deployment. Sniffing a few packets at the LAC starts to
> seem less painful... But I am still open to more suggestions and
> concerns...
> 
> > ................
> > 
> > > > After all, the worst that happens if the wrong ACCM is applied is
> > > > that the packet is ignored, and packets get lost once in a while
> > > > anyway.
> > >
> > > No, the worst is that the packet is *always* dropped, so LCP never
> > > completes, and the call itself is dropped.
> > 
> > If we use data packet AVP, this can't happen.
> > 
> > Thanks
> > - Jacob
> 



From owner-l2tp@ipsec.org  Wed Jan 19 15:51:48 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA05031
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 15:51:48 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id NAA12900; Tue, 18 Jan 2000 13:20:07 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id NAA12893 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 13:19:51 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id NAA27390;
	Tue, 18 Jan 2000 13:36:52 -0800 (PST)
Received: from cisco.com (dhcp-sjc16-115.cisco.com [171.70.96.115]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id QAA12680; Tue, 18 Jan 2000 16:19:12 -0500 (EST)
Message-ID: <38862A54.33C09613@cisco.com>
Date: Wed, 19 Jan 2000 13:19:16 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Barney Wolff <barney@databus.com>
CC: l2tp@SailPix.com
Subject: Re: l2tp-digest V1000 #171
References: <3884a9990.b53@databus.databus.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Barney Wolff wrote:
> 
> I guess I don't understand.  You seem to be saying that the LAC acts
> on the SLI when received, even though it has an output queue of PPP
> frames that were received before the SLI, but have not yet been sent
> out on the serial port (or equivalent).  I would call such behavior
> a bug.  At any rate, there can be at most one output frame on queue
> in the worrisome case, an LCP Conf-Ack, right?  Again, unless I
> really don't understand the problem.

Ensuring your output queue is empty before applying the SLI is a good,
cheap defense, and would probably handle many (if not most) cases. But
we still have the reordering problem. If the SLI is received at the LAC
before the last LCP packet, then it could get applied first. This is the
unrecoverable case.

> 
> Barney
> 
> > Date: Wed, 19 Jan 2000 09:20:29 -0800
> > From: "W. Mark Townsley" <townsley@cisco.com>
> >
> > The problem is that the SLI *is* received. That is, the LNS negotiates
> > something other than the default map. When the LAC receives the SLI, it
> > blindly applies the new map, causing escaping to cease for some or all
> > characters. From this point forward, the strict PPP client will drop all
> > packets that do not have default escaping.


From owner-l2tp@ipsec.org  Wed Jan 19 15:52:09 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA05043
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 15:52:08 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id WAA00182; Tue, 18 Jan 2000 22:02:22 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id WAA00174 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 22:01:53 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id NAA04815;
	Tue, 18 Jan 2000 13:46:52 -0800 (PST)
Received: from cisco.com (dhcp-sjc16-115.cisco.com [171.70.96.115]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id QAA12684; Tue, 18 Jan 2000 16:29:14 -0500 (EST)
Message-ID: <38862CAD.4C8F9AA4@cisco.com>
Date: Wed, 19 Jan 2000 13:29:17 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Jacob Teplitsky <jacobt@livingston.com>
CC: l2tp@SailPix.com
Subject: Re: SLI/ACCM race condition
References: <200001181949.LAA00971@kc.livingston.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Jacob Teplitsky wrote:
> 
> > Date: Tue, 18 Jan 2000 23:29:56 -0800
> > From: "W. Mark Townsley" <townsley@cisco.com>
> > Subject: Re: l2tp-digest V1000 #171
> >
> > Barney Wolff wrote:
> > >
> > > Um, well, ordering of data vs control packets is not guaranteed by
> > > the intervening transport medium.
> > >
> > > There are really only three ways to go (that I can think of :)
> 
> Well, we can add a new AVP to SLI:
> - data packet AVP. The last LCP ACK that finish the LCP negotiations
>  can be sent with the SLI message.

This corrects the race condition at its core by packaging the two
contenders into a single packet so that one cannot beat the other to the
finish. However, it is a considerable protocol change which would set us
back interoperabilty-wise. Sniffing a few packets at the LAC, only when
the SLI is received, and only for a short time, may be better given the
potential disruption. Further, if we are going to change RFCs, I would
like to see changes in the PPP RFC to reflect what conventional wisdom
in the PPP group already seems to be (e.g. be liberal in acceptance of
unescaped characters). Either protocol change would fix the problem
theoretically, but niether would help our situation much given the
current level of deployment. Sniffing a few packets at the LAC starts to
seem less painful... But I am still open to more suggestions and
concerns...

> ................
> 
> > > After all, the worst that happens if the wrong ACCM is applied is
> > > that the packet is ignored, and packets get lost once in a while
> > > anyway.
> >
> > No, the worst is that the packet is *always* dropped, so LCP never
> > completes, and the call itself is dropped.
> 
> If we use data packet AVP, this can't happen.
> 
> Thanks
> - Jacob


From owner-l2tp@ipsec.org  Wed Jan 19 16:29:00 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA05527
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 16:28:59 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id WAA00172; Tue, 18 Jan 2000 22:01:19 -0800 (PST)
Received: from databus.databus.com (databus.databus.com [198.186.154.34]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id WAA00168 for <l2tp@SailPix.com>; Tue, 18 Jan 2000 22:00:55 -0800 (PST)
From: Barney Wolff <barney@databus.com>
To: l2tp@SailPix.com
Date: Tue, 18 Jan 2000 17:05 EST
Subject: Re: l2tp-digest V1000 #171
Content-Type: text/plain
Message-ID: <3884e6d70.fba@databus.databus.com>
Sender: owner-l2tp@SailPix.com
Precedence: bulk

Sorry to speak up yet again, but I don't understand why this case is
unrecoverable.  The PPP client may indeed throw away the last LCP frame.
But the PPP client is then bound to retransmit its LCP Conf-Req, and the
LNS must therefore send an SLI with the default ACCMs in it before
responding.  At the end of the second round of LCP, unless the
re-ordering happens again, things will be fine.  Only if the SLI
consistently passes the LCP Conf-Ack will negotiation fail.  A transport
that consistently manages to have the SLI pass the Conf-Ack is both very
smart and very evil.

The LNS does have to send the SLI with default ACCMs, but this is not
a new requirement - PPP peers are free to renegotiate LCP at any time,
although few actually do.  If 2661 says that an L2TP "call" can have
only a single LCP negotiation, I'm not aware of it.  RFC 1661 is clear
that LCP types 1-7 must always be sent using default options, so if
tunneling via L2TP breaks that, we have a fight with the PPP folks.

Barney

> Date: Wed, 19 Jan 2000 13:19:16 -0800
> From: "W. Mark Townsley" <townsley@cisco.com>
> 
> Ensuring your output queue is empty before applying the SLI is a good,
> cheap defense, and would probably handle many (if not most) cases. But
> we still have the reordering problem. If the SLI is received at the LAC
> before the last LCP packet, then it could get applied first. This is the
> unrecoverable case.


From owner-l2tp@ipsec.org  Wed Jan 19 19:48:57 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA06870
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 19:48:56 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id PAA01337; Wed, 19 Jan 2000 15:32:27 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA01333 for <l2tp@SailPix.com>; Wed, 19 Jan 2000 15:32:04 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id PAA24356;
	Wed, 19 Jan 2000 15:24:10 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id SAA13209; Wed, 19 Jan 2000 18:31:35 -0500 (EST)
Message-ID: <38879ADB.C62CCC85@cisco.com>
Date: Thu, 20 Jan 2000 15:31:39 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Barney Wolff <barney@databus.com>
CC: l2tp@SailPix.com
Subject: Re: l2tp-digest V1000 #171
References: <3884e6d70.fba@databus.databus.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@SailPix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Barney Wolff wrote:
> 
> Sorry to speak up yet again, but I don't understand why this case is

No apology necessary... Please, this is what makes a Working Group work.

> unrecoverable.  The PPP client may indeed throw away the last LCP frame.
> But the PPP client is then bound to retransmit its LCP Conf-Req, and the
> LNS must therefore send an SLI with the default ACCMs in it before
> responding.  At the end of the second round of LCP, unless the
> re-ordering happens again, things will be fine.  Only if the SLI
> consistently passes the LCP Conf-Ack will negotiation fail.  A transport
> that consistently manages to have the SLI pass the Conf-Ack is both very
> smart and very evil.
>
> The LNS does have to send the SLI with default ACCMs, but this is not
> a new requirement - PPP peers are free to renegotiate LCP at any time,

No, not a new requirement, but something that is not obvious in the
spec. I suspect most implementations do not do this today.

> although few actually do.  If 2661 says that an L2TP "call" can have
> only a single LCP negotiation, I'm not aware of it.  RFC 1661 is clear
> that LCP types 1-7 must always be sent using default options, so if
> tunneling via L2TP breaks that, we have a fight with the PPP folks.

I think I like this approach better. So, we have the following
clarifications to make:

1. The LAC must send all PPP packets in its queue before applying the
ACCM received in an SLI

2. The LNS must send an SLI of 0xFFFFFFFF when LCP renegotiation begins,
and an SLI with the new value when LCP negotiation completes (this last
point is already fairly obvious in the spec).

3. The LNS should optimize sending of the SLI when possible. e.g. an SLI
should not be sent if the value within is the same as the last SLI sent.

4. It should be noted that if the SLI and last conf ack at the LNS are
reordered on the intervening network, there is the potential for an
incorrectly escaped LCP packet to be sent to the PPP client. This may
cause some PPP clients to timeout and reinitialize LCP.


> 
> Barney
> 
> > Date: Wed, 19 Jan 2000 13:19:16 -0800
> > From: "W. Mark Townsley" <townsley@cisco.com>
> >
> > Ensuring your output queue is empty before applying the SLI is a good,
> > cheap defense, and would probably handle many (if not most) cases. But
> > we still have the reordering problem. If the SLI is received at the LAC
> > before the last LCP packet, then it could get applied first. This is the
> > unrecoverable case.


From owner-l2tp@ipsec.org  Wed Jan 19 19:55:03 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA06939
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 19:55:03 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id PAA01361; Wed, 19 Jan 2000 15:41:59 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA01357 for <l2tp@ipsec.org>; Wed, 19 Jan 2000 15:41:51 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id PAA11752
	for <l2tp@ipsec.org>; Wed, 19 Jan 2000 15:44:25 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id SAA13225 for <l2tp@ipsec.org>; Wed, 19 Jan 2000 18:41:28 -0500 (EST)
Message-ID: <38879D2D.6BDE251A@cisco.com>
Date: Thu, 20 Jan 2000 15:41:33 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: l2tp@ipsec.org
Subject: Nr in data packets
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


While we are all so active on the list, I thought I might bring up the
subject of Nr in data packets. Without flow control, this field is
pretty much useless. If there are no objections, I was planning on
stating that its use is deprecated and that its value should be ignored. 

- Mark


From owner-l2tp@ipsec.org  Wed Jan 19 19:55:12 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA06950
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 19:55:12 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id PAA01374; Wed, 19 Jan 2000 15:52:53 -0800 (PST)
Received: from mercury.Sun.COM (mercury.Sun.COM [192.9.25.1]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA01370 for <l2tp@ipsec.org>; Wed, 19 Jan 2000 15:52:42 -0800 (PST)
Received: from engmail2.Eng.Sun.COM ([129.146.1.25])
	by mercury.Sun.COM (8.9.3+Sun/8.9.3) with ESMTP id NAA04341;
	Tue, 18 Jan 2000 13:34:38 -0800 (PST)
Received: from nasnfs.eng.sun.com (nasnfs-201.Eng.Sun.COM [129.146.201.28])
	by engmail2.Eng.Sun.COM (8.9.1b+Sun/8.9.1/ENSMAIL,v1.6) with ESMTP id NAA18149;
	Tue, 18 Jan 2000 13:34:38 -0800 (PST)
Received: from nasnfs.Eng.Sun.COM (centralapp.Central.Sun.COM [129.147.34.61])
	by nasnfs.eng.sun.com (8.9.3+Sun/8.9.1) with ESMTP id NAA19384;
	Tue, 18 Jan 2000 13:34:24 -0800 (PST)
From: Patrice Calhoun <Pat.Calhoun@eng.sun.com>
Message-Id: <200001182134.NAA19384@nasnfs.eng.sun.com>
Date: Tue, 18 Jan 2000 13:30:05 -0800
To: "W. Mark Townsley" <townsley@cisco.com>, <l2tp@ipsec.org>
Reply-To: <pcalhoun@eng.sun.com>
Subject: Re: L2TP Minutes from VPN interoperability workshop town meeting
X-Mailer: Sun NetMail 2.3
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

[snip]
>When dropping out of order packets, it is a good idea to update Nr
>first.
>
[snip]

I am not sure that I really understand this one. If I am expecting 4,5,6 and 7
and receive 6, I shouldn't update Nr to reflect it since I still need 4 and 5.
Updating it would tell my peer NOT to retransmit 4 and 5.

Am I mis-understanding this?

PatC



From owner-l2tp@ipsec.org  Wed Jan 19 20:42:17 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA07257
	for <l2tpext-archive@odin.ietf.org>; Wed, 19 Jan 2000 20:42:13 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id PAA01347; Wed, 19 Jan 2000 15:40:00 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA01343 for <l2tp@ipsec.org>; Wed, 19 Jan 2000 15:39:47 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id PAA27801;
	Wed, 19 Jan 2000 15:31:41 -0800 (PST)
Received: from cisco.com (dhcp-n-121.cisco.com [171.70.28.121]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id SAA13221; Wed, 19 Jan 2000 18:38:42 -0500 (EST)
Message-ID: <38879C87.27CC58DE@cisco.com>
Date: Thu, 20 Jan 2000 15:38:47 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en
MIME-Version: 1.0
To: pcalhoun@eng.sun.com
CC: l2tp@ipsec.org
Subject: Re: L2TP Minutes from VPN interoperability workshop town meeting
References: <200001182134.NAA19384@nasnfs.eng.sun.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Patrice Calhoun wrote:
> 
> [snip]
> >When dropping out of order packets, it is a good idea to update Nr
> >first.
> >
> [snip]
> 
> I am not sure that I really understand this one. If I am expecting 4,5,6 and 7
> and receive 6, I shouldn't update Nr to reflect it since I still need 4 and 5.
> Updating it would tell my peer NOT to retransmit 4 and 5.

The point is to update the Nr received in the misordered packet. So, it
is to update what I am supposed to retransmit, not what my peer is
supposed to retransmit. The text was probably not clear. 

> 
> Am I mis-understanding this?
> 
> PatC


From owner-l2tp@ipsec.org  Thu Jan 20 06:35:56 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id GAA24700
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 06:35:55 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id CAA02185; Thu, 20 Jan 2000 02:11:17 -0800 (PST)
Received: from dfintra.datafellows.com (dfintra.datafellows.com [194.197.29.8]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id CAA02181 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 02:10:55 -0800 (PST)
Received: (qmail 12142 invoked from network); 20 Jan 2000 10:10:16 -0000
Received: from df129-180.datafellows.com (HELO F-Secure.com) (10.128.129.180)
  by dfintra.datafellows.com with SMTP; 20 Jan 2000 10:10:16 -0000
Message-ID: <3886DF43.66641B12@F-Secure.com>
Date: Thu, 20 Jan 2000 12:11:15 +0200
From: Ari Huttunen <Ari.Huttunen@F-Secure.com>
Organization: F-Secure Oyj
X-Mailer: Mozilla 4.51 [en] (WinNT; I)
X-Accept-Language: en
MIME-Version: 1.0
To: l2tp <l2tp@ipsec.org>
Subject: Win2000 Netmon - L2TP protocol parser?
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

The Network Monitor documentation that comes with Windows 2000 Server
(released to manufacturing version) states that it understands
the L2TP protocol. Alas, l2tp.dll is nowhere to be found.

Would anyone have a clue if that functionality should be
available or not? I know this is not a Windows 2000 support
mailing list, but that monitor is actually pretty good otherwise!

-- 
Ari Huttunen                   phone: +358 9 859 900
Senior Software Engineer       fax  : +358 9 8599 0452

F-Secure Corporation       http://www.F-Secure.com 

F-Secure products: Integrated Solutions for Enterprise Security


From owner-l2tp@ipsec.org  Thu Jan 20 11:05:16 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id LAA04671
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 11:05:13 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id GAA03772; Thu, 20 Jan 2000 06:48:24 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id GAA03768 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 06:48:11 -0800 (PST)
Received: from comet (209.6.112.100) by mcfeeley.indusriver.com (Worldmail 1.3.167); 20 Jan 2000 09:46:23 -0500
Message-Id: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Thu, 20 Jan 2000 09:50:54 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <38879D2D.6BDE251A@cisco.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Does PPP needs L2TP's guarantee of packet ordering in data channel?
It is due to PPP uses stateful compression.
--- David



At 03:41 PM 1/20/00 -0800, you wrote:

>While we are all so active on the list, I thought I might bring up the
>subject of Nr in data packets. Without flow control, this field is
>pretty much useless. If there are no objections, I was planning on
>stating that its use is deprecated and that its value should be ignored.
>
>- Mark

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Thu Jan 20 13:20:15 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA10898
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 13:20:14 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id JAA03938; Thu, 20 Jan 2000 09:14:09 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA03934 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 09:13:55 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA20391;
	Thu, 20 Jan 2000 09:15:56 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-77.cisco.com [171.68.183.77]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA13586; Thu, 20 Jan 2000 12:12:41 -0500 (EST)
Message-ID: <3888938C.5AF5FA7A@cisco.com>
Date: Fri, 21 Jan 2000 09:12:44 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: David Chen <dchen@indusriver.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



David Chen wrote:
> 
> Does PPP needs L2TP's guarantee of packet ordering in data channel?

Ns is used for packet ordering, I am proposing deprecation of Nr.

> It is due to PPP uses stateful compression.
> --- David
> 
> At 03:41 PM 1/20/00 -0800, you wrote:
> 
> >While we are all so active on the list, I thought I might bring up the
> >subject of Nr in data packets. Without flow control, this field is
> >pretty much useless. If there are no objections, I was planning on
> >stating that its use is deprecated and that its value should be ignored.
> >
> >- Mark
> 
> ========================================
> David Chen
> Indus River Networks, Inc.
> 31 Nagog Park
> Acton, MA 01720
> U.S.A.
> dchen@indusriver.com
> (978) 266-8141
> ========================================


From owner-l2tp@ipsec.org  Thu Jan 20 13:57:38 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA12059
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 13:57:38 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA03983; Thu, 20 Jan 2000 09:56:15 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA03979 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 09:55:52 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA19218;
	Thu, 20 Jan 2000 09:57:56 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-77.cisco.com [171.68.183.77]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA13660; Thu, 20 Jan 2000 12:54:49 -0500 (EST)
Message-ID: <38889D6D.5D3F38F2@cisco.com>
Date: Fri, 21 Jan 2000 09:54:53 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: David Chen <dchen@indusriver.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com> <4.2.0.58.20000120124250.00b09480@pop3.indusriver.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



David Chen wrote:
> 
> Is the Nr for acknowledging to the sender by the receiver what
> has been received  so far?

Yes, that was its intent with flow control.

> Similar to HDLC if the difference between Ns and Nr is larger than
> presetted window
> then it will require peers to resync/retransmit packets.

L2TP does not provide lossless transmission of data packets. If you
really, really, want this, do PPP reliable transmission (RFC1663).

> Does the PPP's compression require lossless compression?

Some PPP compression is stateful, thus losing a packet will cause a
reset for all packets already in transit. Best idea is to simply
disallow stateful compression at the LNS.

> If so, we may not only need Ns for ordering but Nr for lossless transmission.
> --- David
> 
> At 09:12 AM 1/21/00 -0800, you wrote:
> 
> >David Chen wrote:
> > >
> > > Does PPP needs L2TP's guarantee of packet ordering in data channel?
> >
> >Ns is used for packet ordering, I am proposing deprecation of Nr.
> >
> > > It is due to PPP uses stateful compression.
> > > --- David
> > >
> > > At 03:41 PM 1/20/00 -0800, you wrote:
> > >
> > > >While we are all so active on the list, I thought I might bring up the
> > > >subject of Nr in data packets. Without flow control, this field is
> > > >pretty much useless. If there are no objections, I was planning on
> > > >stating that its use is deprecated and that its value should be ignored.
> > > >
> > > >- Mark
> > >
> > > ========================================
> > > David Chen
> > > Indus River Networks, Inc.
> > > 31 Nagog Park
> > > Acton, MA 01720
> > > U.S.A.
> > > dchen@indusriver.com
> > > (978) 266-8141
> > > ========================================
> 
> ========================================
> David Chen
> Indus River Networks, Inc.
> 31 Nagog Park
> Acton, MA 01720
> U.S.A.
> dchen@indusriver.com
> (978) 266-8141
> ========================================


From owner-l2tp@ipsec.org  Thu Jan 20 14:03:34 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA12230
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 14:03:33 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id JAA03963; Thu, 20 Jan 2000 09:42:53 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA03959 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 09:42:32 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167); 20 Jan 2000 12:40:40 -0500
Message-Id: <4.2.0.58.20000120124250.00b09480@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Thu, 20 Jan 2000 12:48:55 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <3888938C.5AF5FA7A@cisco.com>
References: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Is the Nr for acknowledging to the sender by the receiver what
has been received  so far?
Similar to HDLC if the difference between Ns and Nr is larger than 
presetted window
then it will require peers to resync/retransmit packets.
Does the PPP's compression require lossless compression?
If so, we may not only need Ns for ordering but Nr for lossless transmission.
--- David

At 09:12 AM 1/21/00 -0800, you wrote:


>David Chen wrote:
> >
> > Does PPP needs L2TP's guarantee of packet ordering in data channel?
>
>Ns is used for packet ordering, I am proposing deprecation of Nr.
>
> > It is due to PPP uses stateful compression.
> > --- David
> >
> > At 03:41 PM 1/20/00 -0800, you wrote:
> >
> > >While we are all so active on the list, I thought I might bring up the
> > >subject of Nr in data packets. Without flow control, this field is
> > >pretty much useless. If there are no objections, I was planning on
> > >stating that its use is deprecated and that its value should be ignored.
> > >
> > >- Mark
> >
> > ========================================
> > David Chen
> > Indus River Networks, Inc.
> > 31 Nagog Park
> > Acton, MA 01720
> > U.S.A.
> > dchen@indusriver.com
> > (978) 266-8141
> > ========================================

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Thu Jan 20 14:26:38 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA12709
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 14:26:36 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id KAA04011; Thu, 20 Jan 2000 10:12:37 -0800 (PST)
Received: from enserv.ensemblecom.com (fwuser@[205.163.11.194]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id KAA04007 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 10:11:56 -0800 (PST)
Message-ID: <7F8199C3CF1DD2119D1400104B2F099243D1C3@ENSERV>
From: John Thomes <john@ensemblecom.com>
To: l2tp <l2tp@ipsec.org>
Subject: RE: Win2000 Netmon - L2TP protocol parser?
Date: Thu, 20 Jan 2000 10:06:54 -0800
MIME-Version: 1.0
Content-Type: multipart/alternative;
	boundary="----_=_NextPart_001_01BF6371.227DB542"
Sender: owner-l2tp@sailpix.com
Precedence: bulk

This message is in MIME format. Since your mail reader does not understand
this format, some or all of this message may not be legible.

------_=_NextPart_001_01BF6371.227DB542
Content-Type: text/plain



	If you don't mind using Linux or Sun, there is a free network
protocol analyzer that supports L2TP @ http://ethereal.zing.org/.

	- John Thomes

> -----Original Message-----
> From:	Ari Huttunen [SMTP:Ari.Huttunen@F-Secure.com]
> Sent:	Thursday, January 20, 2000 2:11 AM
> To:	l2tp
> Subject:	Win2000 Netmon - L2TP protocol parser?
> 
> The Network Monitor documentation that comes with Windows 2000 Server
> (released to manufacturing version) states that it understands
> the L2TP protocol. Alas, l2tp.dll is nowhere to be found.
> 
> Would anyone have a clue if that functionality should be
> available or not? I know this is not a Windows 2000 support
> mailing list, but that monitor is actually pretty good otherwise!
> 
> -- 
> Ari Huttunen                   phone: +358 9 859 900
> Senior Software Engineer       fax  : +358 9 8599 0452
> 
> F-Secure Corporation       http://www.F-Secure.com 
> 
> F-Secure products: Integrated Solutions for Enterprise Security

------_=_NextPart_001_01BF6371.227DB542
Content-Type: text/html
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV=3D"Content-Type" CONTENT=3D"text/html; =
charset=3Dus-ascii">
<META NAME=3D"Generator" CONTENT=3D"MS Exchange Server version =
5.5.2232.0">
<TITLE>RE: Win2000 Netmon - L2TP protocol parser?</TITLE>
</HEAD>
<BODY>
<BR>
<BR>
<UL>
<P><FONT COLOR=3D"#0000FF" SIZE=3D2 FACE=3D"Arial">If you don't mind =
using Linux or Sun, there is a free network protocol analyzer that =
supports L2TP @<U> <A HREF=3D"http://ethereal.zing.org/" =
TARGET=3D"_blank">http://ethereal.zing.org/</A></U>.</FONT></P>

<P><FONT COLOR=3D"#0000FF" SIZE=3D2 FACE=3D"Arial">- John Thomes</FONT>
</P>

<P><FONT SIZE=3D1 FACE=3D"Arial">-----Original Message-----</FONT>
<BR><B><FONT SIZE=3D1 FACE=3D"Arial">From:&nbsp;&nbsp;</FONT></B> <FONT =
SIZE=3D1 FACE=3D"Arial">Ari Huttunen =
[SMTP:Ari.Huttunen@F-Secure.com]</FONT>
<BR><B><FONT SIZE=3D1 FACE=3D"Arial">Sent:&nbsp;&nbsp;</FONT></B> <FONT =
SIZE=3D1 FACE=3D"Arial">Thursday, January 20, 2000 2:11 AM</FONT>
<BR><B><FONT SIZE=3D1 =
FACE=3D"Arial">To:&nbsp;&nbsp;&nbsp;&nbsp;</FONT></B> <FONT SIZE=3D1 =
FACE=3D"Arial">l2tp</FONT>
<BR><B><FONT SIZE=3D1 =
FACE=3D"Arial">Subject:&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</FONT>=
</B> <FONT SIZE=3D1 FACE=3D"Arial">Win2000 Netmon - L2TP protocol =
parser?</FONT>
</P>

<P><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">The Network Monitor =
documentation that comes with Windows 2000 Server</FONT>
<BR><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">(released to =
manufacturing version) states that it understands</FONT>
<BR><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">the L2TP protocol. =
Alas, l2tp.dll is nowhere to be found.</FONT>
</P>

<P><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">Would anyone have a =
clue if that functionality should be</FONT>
<BR><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">available or not? I =
know this is not a Windows 2000 support</FONT>
<BR><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">mailing list, but =
that monitor is actually pretty good otherwise!</FONT>
</P>

<P><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">-- </FONT>
<BR><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">Ari =
Huttunen&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbs=
p;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; phone: +358 9 859 =
900</FONT>
<BR><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">Senior Software =
Engineer&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; fax&nbsp; : +358 9 8599 =
0452</FONT>
</P>

<P><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">F-Secure =
Corporation&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</FONT><U> <FONT =
COLOR=3D"#0000FF" SIZE=3D2 FACE=3D"Arial"><A =
HREF=3D"http://www.F-Secure.com" =
TARGET=3D"_blank">http://www.F-Secure.com</A></FONT></U><FONT =
COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial"> </FONT>
</P>

<P><FONT COLOR=3D"#000000" SIZE=3D2 FACE=3D"Arial">F-Secure products: =
Integrated Solutions for Enterprise Security</FONT>
</P>
</UL>
</BODY>
</HTML>
------_=_NextPart_001_01BF6371.227DB542--


From owner-l2tp@ipsec.org  Thu Jan 20 16:14:18 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA15539
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 16:14:16 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id LAA04227; Thu, 20 Jan 2000 11:51:33 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA04223 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 11:51:07 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167); 20 Jan 2000 14:49:22 -0500
Message-Id: <4.2.0.58.20000120145706.00b184b0@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Thu, 20 Jan 2000 14:57:33 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <38889D6D.5D3F38F2@cisco.com>
References: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com>
 <4.2.0.58.20000120124250.00b09480@pop3.indusriver.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

If PPP wants flow control in data channel,
then it should implement rfc1663 and...
L2tp can drop both Ns and Nr in data channel?
--- David



At 09:54 AM 1/21/00 -0800, you wrote:


>David Chen wrote:
> >
> > Is the Nr for acknowledging to the sender by the receiver what
> > has been received  so far?
>
>Yes, that was its intent with flow control.
>
> > Similar to HDLC if the difference between Ns and Nr is larger than
> > presetted window
> > then it will require peers to resync/retransmit packets.
>
>L2TP does not provide lossless transmission of data packets. If you
>really, really, want this, do PPP reliable transmission (RFC1663).
>
> > Does the PPP's compression require lossless compression?
>
>Some PPP compression is stateful, thus losing a packet will cause a
>reset for all packets already in transit. Best idea is to simply
>disallow stateful compression at the LNS.
>
> > If so, we may not only need Ns for ordering but Nr for lossless 
> transmission.
> > --- David
> >
> > At 09:12 AM 1/21/00 -0800, you wrote:
> >
> > >David Chen wrote:
> > > >
> > > > Does PPP needs L2TP's guarantee of packet ordering in data channel?
> > >
> > >Ns is used for packet ordering, I am proposing deprecation of Nr.
> > >
> > > > It is due to PPP uses stateful compression.
> > > > --- David
> > > >
> > > > At 03:41 PM 1/20/00 -0800, you wrote:
> > > >
> > > > >While we are all so active on the list, I thought I might bring up the
> > > > >subject of Nr in data packets. Without flow control, this field is
> > > > >pretty much useless. If there are no objections, I was planning on
> > > > >stating that its use is deprecated and that its value should be 
> ignored.
> > > > >
> > > > >- Mark
> > > >
> > > > ========================================
> > > > David Chen
> > > > Indus River Networks, Inc.
> > > > 31 Nagog Park
> > > > Acton, MA 01720
> > > > U.S.A.
> > > > dchen@indusriver.com
> > > > (978) 266-8141
> > > > ========================================
> >
> > ========================================
> > David Chen
> > Indus River Networks, Inc.
> > 31 Nagog Park
> > Acton, MA 01720
> > U.S.A.
> > dchen@indusriver.com
> > (978) 266-8141
> > ========================================



From owner-l2tp@ipsec.org  Thu Jan 20 16:45:02 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA16015
	for <l2tpext-archive@odin.ietf.org>; Thu, 20 Jan 2000 16:44:58 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id MAA04287; Thu, 20 Jan 2000 12:25:38 -0800 (PST)
Received: from dfssl.exchange.microsoft.com ([131.107.88.59]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id MAA04280 for <l2tp@ipsec.org>; Thu, 20 Jan 2000 12:24:21 -0800 (PST)
Received: from 127.0.0.1 by dfssl.exchange.microsoft.com (InterScan E-Mail VirusWall NT); Thu, 20 Jan 2000 12:14:32 -0800 (Pacific Standard Time)
Received: by dfssl with Internet Mail Service (5.5.2650.21)
	id <DFRQ6S93>; Thu, 20 Jan 2000 12:14:31 -0800
Message-ID: <4992824A0863D211964B0008C7B1ACB808AF1CFB@fifi.platinum.corp.microsoft.com>
From: William Dixon <wdixon@Exchange.Microsoft.com>
To: "'John Thomes'" <john@ensemblecom.com>, l2tp <l2tp@ipsec.org>
Subject: RE: Win2000 Netmon - L2TP protocol parser?
Date: Thu, 20 Jan 2000 12:14:35 -0800
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2650.21)
Content-Type: multipart/alternative;
	boundary="----_=_NextPart_001_01BF6382.F5E9A3B2"
Sender: owner-l2tp@sailpix.com
Precedence: bulk

This message is in MIME format. Since your mail reader does not understand
this format, some or all of this message may not be legible.

------_=_NextPart_001_01BF6382.F5E9A3B2
Content-Type: text/plain

I'll check on what Netmon can parse in Windows 2000.  Normally L2TP is IPSec
encrypted, so...  However, in Win2k if you set the L2TP tunnel encryption
type to "no encryption", it gets interpreted as a multiple suite proposal
from the client for 
 
1. ESP null encryption, with integrity & authentication
2. AH
 
LNS server side can then choose what it wants.
 
Netmon does parse AH into upper layer protocols, but it doesn't parse ESP
past a few fields in the ESP header.
 
By default, Win2k's L2TP uses predefined IPSec policy to do certificate
auth.  If you have implemented a client of Cisco's CEP, the Microsoft's CA
will support Cisco's Simple Certificate Enrollment protocol, available as an
update pack to the Windows 2000 Server. More info available at:
http://sectestca3.rte.microsoft.com/certsrv/mscep/mscep.dll
<http://sectestca3.rte.microsoft.com/certsrv/mscep/mscep.dll>   We didn't
implement a client for that ourselves.  We will be working on client and
server sides of CMC for the future.
 
The Win2k Resource Kit has this text for testing non-IPSec L2TP, clear text
or custom IPSec policy for L2TP:
 
To prevent the Routing and Remote Access service from automatically creating
an IPSec policy for L2TP traffic, set the value of ProhibitIpSec to 1
(HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\RasMan \Parameters).
By default, ProhibitIpSec is set to 0. When ProhibitIpSec is set to 1, the
encryption settings on the demand-dial interface configured on the calling
router are ignored in favor of the encryption settings of the manually
configured IPSec policy. The computer must be restarted for the changes to
this registry setting to take effect
 
If you do a custom policy on the Win2k server side, for the server to
properly accept L2TP tunnels, the IKE proposal needs to be what we use:
From source port 1701, dest port Any.
 
Wm

-----Original Message-----
From: John Thomes [mailto:john@ensemblecom.com]
Sent: Thursday, January 20, 2000 10:07 AM
To: l2tp
Subject: RE: Win2000 Netmon - L2TP protocol parser?





	If you don't mind using Linux or Sun, there is a free network
protocol analyzer that supports L2TP @ http://ethereal.zing.org/
<http://ethereal.zing.org/> .

	- John Thomes 

	-----Original Message----- 
From:   Ari Huttunen [SMTP:Ari.Huttunen@F-Secure.com] 
Sent:   Thursday, January 20, 2000 2:11 AM 
To:     l2tp 
Subject:        Win2000 Netmon - L2TP protocol parser? 

	The Network Monitor documentation that comes with Windows 2000
Server 
(released to manufacturing version) states that it understands 
the L2TP protocol. Alas, l2tp.dll is nowhere to be found. 

	Would anyone have a clue if that functionality should be 
available or not? I know this is not a Windows 2000 support 
mailing list, but that monitor is actually pretty good otherwise! 

	-- 
Ari Huttunen                   phone: +358 9 859 900 
Senior Software Engineer       fax  : +358 9 8599 0452 

	F-Secure Corporation       http://www.F-Secure.com
<http://www.F-Secure.com>  

	F-Secure products: Integrated Solutions for Enterprise Security 


------_=_NextPart_001_01BF6382.F5E9A3B2
Content-Type: text/html
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META HTTP-EQUIV=3D"Content-Type" CONTENT=3D"text/html; =
charset=3Dus-ascii">
<TITLE>RE: Win2000 Netmon - L2TP protocol parser?</TITLE>

<META content=3D"MSHTML 5.00.2920.0" name=3DGENERATOR></HEAD>
<BODY>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>I'll=20
check on what Netmon can parse in Windows 2000.&nbsp; Normally L2TP is =
IPSec=20
encrypted, so...&nbsp; However, in Win2k if you set the L2TP tunnel =
encryption=20
type to "no encryption", it gets interpreted as a multiple suite =
proposal from=20
the client for </SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000></SPAN></FONT>&nbsp;</DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>1. ESP=20
null encryption, with integrity &amp; =
authentication</SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>2.=20
AH</SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000></SPAN></FONT>&nbsp;</DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>LNS=20
server side can then choose what it wants.</SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000></SPAN></FONT>&nbsp;</DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>Netmon=20
does parse AH into upper layer protocols, but it doesn't parse ESP past =
a few=20
fields in the ESP header.</SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000></SPAN></FONT>&nbsp;</DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>By=20
default, Win2k's L2TP uses predefined IPSec policy to do certificate =
auth.&nbsp;=20
</SPAN></FONT><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000>If you have implemented a client of Cisco's =
CEP, the=20
Microsoft's CA will support Cisco's Simple Certificate Enrollment =
protocol,=20
available as an update pack to the Windows 2000 Server. More info =
available at:=20
<A=20
href=3D"http://sectestca3.rte.microsoft.com/certsrv/mscep/mscep.dll">htt=
p://sectestca3.rte.microsoft.com/certsrv/mscep/mscep.dll</A>&nbsp;=20
We didn't implement a client for that ourselves.&nbsp; We will be =
working on=20
client and server sides of CMC for the future.</SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000></SPAN></FONT>&nbsp;</DIV>
<DIV><SPAN class=3D486175619-20012000>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>The=20
Win2k Resource Kit has this text for testing non-IPSec L2TP, clear text =
or=20
custom IPSec policy for L2TP:</SPAN></FONT></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000></SPAN></FONT>&nbsp;</DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000><SPAN=20
style=3D"FONT-FAMILY: 'Times New Roman'; FONT-SIZE: 10.5pt; =
mso-bidi-font-size: 10.0pt; mso-fareast-font-family: 'Times New Roman'; =
mso-ansi-language: EN-US; mso-fareast-language: EN-US; =
mso-bidi-language: AR-SA">To=20
prevent the Routing and Remote Access service from automatically =
creating an=20
IPSec policy for L2TP traffic, set the value of ProhibitIpSec to 1=20
(HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\RasMan =
\Parameters). By=20
default, ProhibitIpSec is set to 0. When ProhibitIpSec is set to 1, the =

encryption settings on the demand-dial interface configured on the =
calling=20
router are ignored in favor of the encryption settings of the manually=20
configured IPSec policy. The computer must be restarted for the changes =
to this=20
registry setting to take effect</SPAN></SPAN></FONT></DIV>
<DIV><SPAN class=3D486175619-20012000></SPAN>&nbsp;</DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN =
class=3D486175619-20012000>If you=20
do a custom policy on the Win2k server side, for the server to properly =
accept=20
L2TP tunnels, the IKE proposal needs to be what we use:&nbsp; From =
source port=20
1701, dest port Any.</SPAN></FONT></DIV>
<DIV><SPAN class=3D486175619-20012000></SPAN>&nbsp;</DIV></SPAN></DIV>
<DIV><FONT color=3D#0000ff face=3DArial size=3D2><SPAN=20
class=3D486175619-20012000>Wm</SPAN></FONT></DIV>
<BLOCKQUOTE>
  <DIV align=3Dleft class=3DOutlookMessageHeader dir=3Dltr><FONT =
face=3DTahoma=20
  size=3D2>-----Original Message-----<BR><B>From:</B> John Thomes=20
  [mailto:john@ensemblecom.com]<BR><B>Sent:</B> Thursday, January 20, =
2000 10:07=20
  AM<BR><B>To:</B> l2tp<BR><B>Subject:</B> RE: Win2000 Netmon - L2TP =
protocol=20
  parser?<BR><BR></DIV></FONT><BR><BR>
  <UL>
    <P><FONT color=3D#0000ff face=3DArial size=3D2>If you don't mind =
using Linux or=20
    Sun, there is a free network protocol analyzer that supports L2TP =
@<U> <A=20
    href=3D"http://ethereal.zing.org/"=20
    target=3D_blank>http://ethereal.zing.org/</A></U>.</FONT></P>
    <P><FONT color=3D#0000ff face=3DArial size=3D2>- John Thomes</FONT> =
</P>
    <P><FONT face=3DArial size=3D1>-----Original Message-----</FONT> =
<BR><B><FONT=20
    face=3DArial size=3D1>From:&nbsp;&nbsp;</FONT></B> <FONT =
face=3DArial size=3D1>Ari=20
    Huttunen [SMTP:Ari.Huttunen@F-Secure.com]</FONT> <BR><B><FONT =
face=3DArial=20
    size=3D1>Sent:&nbsp;&nbsp;</FONT></B> <FONT face=3DArial =
size=3D1>Thursday,=20
    January 20, 2000 2:11 AM</FONT> <BR><B><FONT face=3DArial=20
    size=3D1>To:&nbsp;&nbsp;&nbsp;&nbsp;</FONT></B> <FONT face=3DArial=20
    size=3D1>l2tp</FONT> <BR><B><FONT face=3DArial=20
    =
size=3D1>Subject:&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</FONT></B> =
<FONT=20
    face=3DArial size=3D1>Win2000 Netmon - L2TP protocol parser?</FONT> =
</P>
    <P><FONT color=3D#000000 face=3DArial size=3D2>The Network Monitor =
documentation=20
    that comes with Windows 2000 Server</FONT> <BR><FONT =
color=3D#000000=20
    face=3DArial size=3D2>(released to manufacturing version) states =
that it=20
    understands</FONT> <BR><FONT color=3D#000000 face=3DArial =
size=3D2>the L2TP=20
    protocol. Alas, l2tp.dll is nowhere to be found.</FONT> </P>
    <P><FONT color=3D#000000 face=3DArial size=3D2>Would anyone have a =
clue if that=20
    functionality should be</FONT> <BR><FONT color=3D#000000 =
face=3DArial=20
    size=3D2>available or not? I know this is not a Windows 2000 =
support</FONT>=20
    <BR><FONT color=3D#000000 face=3DArial size=3D2>mailing list, but =
that monitor is=20
    actually pretty good otherwise!</FONT> </P>
    <P><FONT color=3D#000000 face=3DArial size=3D2>-- </FONT><BR><FONT =
color=3D#000000=20
    face=3DArial size=3D2>Ari=20
    =
Huttunen&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbs=
p;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;=20
    phone: +358 9 859 900</FONT> <BR><FONT color=3D#000000 face=3DArial =

    size=3D2>Senior Software =
Engineer&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;=20
    fax&nbsp; : +358 9 8599 0452</FONT> </P>
    <P><FONT color=3D#000000 face=3DArial size=3D2>F-Secure=20
    Corporation&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</FONT><U> <FONT=20
    color=3D#0000ff face=3DArial size=3D2><A =
href=3D"http://www.F-Secure.com"=20
    target=3D_blank>http://www.F-Secure.com</A></FONT></U><FONT =
color=3D#000000=20
    face=3DArial size=3D2> </FONT></P>
    <P><FONT color=3D#000000 face=3DArial size=3D2>F-Secure products: =
Integrated=20
    Solutions for Enterprise Security</FONT> =
</P></UL></BLOCKQUOTE></BODY></HTML>

------_=_NextPart_001_01BF6382.F5E9A3B2--


From owner-l2tp@ipsec.org  Fri Jan 21 12:48:45 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA14632
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 12:48:44 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id HAA01954; Fri, 21 Jan 2000 07:32:00 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id HAA01950 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 07:31:20 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167); 21 Jan 2000 10:29:35 -0500
Message-Id: <4.2.0.58.20000121103719.00b0bc40@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 21 Jan 2000 10:37:51 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <3888D7DD.74C8A98D@cisco.com>
References: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com>
 <4.2.0.58.20000120124250.00b09480@pop3.indusriver.com>
 <4.2.0.58.20000120145433.00b0bd60@pop3.indusriver.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Since PPP was evolved from serial link connection,
the PPP assumes the underlining protocol layer/media is serial.
However, the PPP is for encapsulating protocol's packet.
These protocols do not demand PPP or underlining layer's transmission is 
serial.
It seems that PPP does not require the underlining layer transmission is 
serial unless
other reason as compress which the PPP is aware of it and can use flow 
control within
itself.
In this view, the Ns for emulate the serial link connection is rational but
is not used by PPP.
What is the circumstance that the Ns is demanded by PPP?
--- David


At 02:04 PM 1/21/00 -0800, you wrote:


>David Chen wrote:
> >
> > If PPP wants flow control in data channel,
> > then it should implement rfc1663 and...
>
>There is no officially sactioned way to do flow control with L2TP now.
>1663 is PPP reliablility. two different things.
>
> > L2tp can drop both Ns and Nr in data channel?
>
>Ns is still useful for sequencing (unreliable sequencing).
>
> > --- David
> >
> > At 09:54 AM 1/21/00 -0800, you wrote:
> >
> > >David Chen wrote:
> > > >
> > > > Is the Nr for acknowledging to the sender by the receiver what
> > > > has been received  so far?
> > >
> > >Yes, that was its intent with flow control.
> > >
> > > > Similar to HDLC if the difference between Ns and Nr is larger than
> > > > presetted window
> > > > then it will require peers to resync/retransmit packets.
> > >
> > >L2TP does not provide lossless transmission of data packets. If you
> > >really, really, want this, do PPP reliable transmission (RFC1663).
> > >
> > > > Does the PPP's compression require lossless compression?
> > >
> > >Some PPP compression is stateful, thus losing a packet will cause a
> > >reset for all packets already in transit. Best idea is to simply
> > >disallow stateful compression at the LNS.
> > >
> > > > If so, we may not only need Ns for ordering but Nr for lossless
> > > transmission.
> > > > --- David
> > > >
> > > > At 09:12 AM 1/21/00 -0800, you wrote:
> > > >
> > > > >David Chen wrote:
> > > > > >
> > > > > > Does PPP needs L2TP's guarantee of packet ordering in data channel?
> > > > >
> > > > >Ns is used for packet ordering, I am proposing deprecation of Nr.
> > > > >
> > > > > > It is due to PPP uses stateful compression.
> > > > > > --- David
> > > > > >
> > > > > > At 03:41 PM 1/20/00 -0800, you wrote:
> > > > > >
> > > > > > >While we are all so active on the list, I thought I might 
> bring up the
> > > > > > >subject of Nr in data packets. Without flow control, this field is
> > > > > > >pretty much useless. If there are no objections, I was planning on
> > > > > > >stating that its use is deprecated and that its value should be
> > > ignored.
> > > > > > >
> > > > > > >- Mark
> > > > > >
> > > > > > ========================================
> > > > > > David Chen
> > > > > > Indus River Networks, Inc.
> > > > > > 31 Nagog Park
> > > > > > Acton, MA 01720
> > > > > > U.S.A.
> > > > > > dchen@indusriver.com
> > > > > > (978) 266-8141
> > > > > > ========================================
> > > >
> > > > ========================================
> > > > David Chen
> > > > Indus River Networks, Inc.
> > > > 31 Nagog Park
> > > > Acton, MA 01720
> > > > U.S.A.
> > > > dchen@indusriver.com
> > > > (978) 266-8141
> > > > ========================================
> >
> > ========================================
> > David Chen
> > Indus River Networks, Inc.
> > 31 Nagog Park
> > Acton, MA 01720
> > U.S.A.
> > dchen@indusriver.com
> > (978) 266-8141
> > ========================================

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Fri Jan 21 13:48:05 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA15503
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 13:48:02 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA02174; Fri, 21 Jan 2000 09:00:09 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id IAA02162 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 08:59:22 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id IAA26638;
	Fri, 21 Jan 2000 08:51:01 -0800 (PST)
Received: from cisco.com (sjck-dial-gw5-64.cisco.com [171.68.183.64]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id LAA14062; Fri, 21 Jan 2000 11:58:26 -0500 (EST)
Message-ID: <3889E1B6.66D21A13@cisco.com>
Date: Sat, 22 Jan 2000 08:58:30 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: David Chen <dchen@indusriver.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <4.2.0.58.20000120094819.00b0be80@pop3.indusriver.com>
	 <4.2.0.58.20000120124250.00b09480@pop3.indusriver.com>
	 <4.2.0.58.20000120145433.00b0bd60@pop3.indusriver.com> <4.2.0.58.20000121103719.00b0bc40@pop3.indusriver.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



David Chen wrote:
> 
> Since PPP was evolved from serial link connection,
> the PPP assumes the underlining protocol layer/media is serial.
> However, the PPP is for encapsulating protocol's packet.
> These protocols do not demand PPP or underlining layer's transmission is
> serial.
> It seems that PPP does not require the underlining layer transmission is
> serial unless
> other reason as compress which the PPP is aware of it and can use flow
> control within
> itself.
> In this view, the Ns for emulate the serial link connection is rational but
> is not used by PPP.
> What is the circumstance that the Ns is demanded by PPP?

RFC 1661 states that all packets are assummed to be delivered in order.
Some protocols (VJ header compression, PPP payload compression and
encryption) were designed with this in mind. Further, the CPs that run
on PPP may have a problem with out of order packets, though given their
stop and wait nature and relatively long default timeout between
retransmissions, it is unlikely. 

The bottom line is, PPP has an in-order delivery assumption, thus L2TP
must have some way of not breaking this. It is optional, so if the user
understands the risks then sequence numbers may be omitted. We have
found that, w/o the stateful protocols listed above, operating w/o
guaranteed delivery order reduces L2TP overhead, and does not adversly
affect the PPP stream.


> --- David
> 
> At 02:04 PM 1/21/00 -0800, you wrote:
> 
> >David Chen wrote:
> > >
> > > If PPP wants flow control in data channel,
> > > then it should implement rfc1663 and...
> >
> >There is no officially sactioned way to do flow control with L2TP now.
> >1663 is PPP reliablility. two different things.
> >
> > > L2tp can drop both Ns and Nr in data channel?
> >
> >Ns is still useful for sequencing (unreliable sequencing).
> >
> > > --- David
> > >
> > > At 09:54 AM 1/21/00 -0800, you wrote:
> > >
> > > >David Chen wrote:
> > > > >
> > > > > Is the Nr for acknowledging to the sender by the receiver what
> > > > > has been received  so far?
> > > >
> > > >Yes, that was its intent with flow control.
> > > >
> > > > > Similar to HDLC if the difference between Ns and Nr is larger than
> > > > > presetted window
> > > > > then it will require peers to resync/retransmit packets.
> > > >
> > > >L2TP does not provide lossless transmission of data packets. If you
> > > >really, really, want this, do PPP reliable transmission (RFC1663).
> > > >
> > > > > Does the PPP's compression require lossless compression?
> > > >
> > > >Some PPP compression is stateful, thus losing a packet will cause a
> > > >reset for all packets already in transit. Best idea is to simply
> > > >disallow stateful compression at the LNS.
> > > >
> > > > > If so, we may not only need Ns for ordering but Nr for lossless
> > > > transmission.
> > > > > --- David
> > > > >
> > > > > At 09:12 AM 1/21/00 -0800, you wrote:
> > > > >
> > > > > >David Chen wrote:
> > > > > > >
> > > > > > > Does PPP needs L2TP's guarantee of packet ordering in data channel?
> > > > > >
> > > > > >Ns is used for packet ordering, I am proposing deprecation of Nr.
> > > > > >
> > > > > > > It is due to PPP uses stateful compression.
> > > > > > > --- David
> > > > > > >
> > > > > > > At 03:41 PM 1/20/00 -0800, you wrote:
> > > > > > >
> > > > > > > >While we are all so active on the list, I thought I might
> > bring up the
> > > > > > > >subject of Nr in data packets. Without flow control, this field is
> > > > > > > >pretty much useless. If there are no objections, I was planning on
> > > > > > > >stating that its use is deprecated and that its value should be
> > > > ignored.
> > > > > > > >
> > > > > > > >- Mark
> > > > > > >
> > > > > > > ========================================
> > > > > > > David Chen
> > > > > > > Indus River Networks, Inc.
> > > > > > > 31 Nagog Park
> > > > > > > Acton, MA 01720
> > > > > > > U.S.A.
> > > > > > > dchen@indusriver.com
> > > > > > > (978) 266-8141
> > > > > > > ========================================
> > > > >
> > > > > ========================================
> > > > > David Chen
> > > > > Indus River Networks, Inc.
> > > > > 31 Nagog Park
> > > > > Acton, MA 01720
> > > > > U.S.A.
> > > > > dchen@indusriver.com
> > > > > (978) 266-8141
> > > > > ========================================
> > >
> > > ========================================
> > > David Chen
> > > Indus River Networks, Inc.
> > > 31 Nagog Park
> > > Acton, MA 01720
> > > U.S.A.
> > > dchen@indusriver.com
> > > (978) 266-8141
> > > ========================================
> 
> ========================================
> David Chen
> Indus River Networks, Inc.
> 31 Nagog Park
> Acton, MA 01720
> U.S.A.
> dchen@indusriver.com
> (978) 266-8141
> ========================================


From owner-l2tp@ipsec.org  Fri Jan 21 13:58:28 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA15640
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 13:58:26 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA02249; Fri, 21 Jan 2000 09:57:09 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA02245 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 09:56:41 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id KAA14797
	for l2tp@ipsec.org  env-from <vjs>;
	Fri, 21 Jan 2000 10:56:35 -0700 (MST)
Date: Fri, 21 Jan 2000 10:56:35 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001211756.KAA14797@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: "W. Mark Townsley" <townsley@cisco.com>

> ...
> RFC 1661 states that all packets are assummed to be delivered in order.
> Some protocols (VJ header compression, PPP payload compression and
> encryption) were designed with this in mind.

You must also add RFC 1990 (MP) to that list of PPP protocols that assume
that the order of the stream of packets is preserved.  Re-ordering MP
fragments is not catastrophic in PPP systems that handle MP sequence number
roll-over robustly, but re-ordering can stall the delivery of the data
(e.g. IP) until the non-standard timers in robust MP implementations fire.


>                                              Further, the CPs that run
> on PPP may have a problem with out of order packets, though given their
> stop and wait nature and relatively long default timeout between
> retransmissions, it is unlikely. 

I think that is wrong.  The request-response nature of CP's makes
re-ordering unlikely on paths that does re-order packets, but does not
protect against re-ordering when it does happen.  Think about the
many places in the RFC 1661 state machine that break catastrophically
with any re-ordering.
There are also interactions among CP's that would be broken by packet
reordering, such between authentication and LCP, and between LCP and
everything else.


> The bottom line is, PPP has an in-order delivery assumption, thus L2TP
> must have some way of not breaking this. It is optional, so if the user
> understands the risks then sequence numbers may be omitted.

Who is the "user" that is doing that understanding of risks?  The typical
Windows 98 or even NT user does not have a clue about what an packet stream
might be.  Many technical people at ISP's are as well informed, although
they don't realize it and certainly won't admit it.

>                                                             We have
> found that, w/o the stateful protocols listed above, operating w/o
> guaranteed delivery order reduces L2TP overhead, and does not adversly
> affect the PPP stream.

What measurements have you made of the results of packet reordering
on PPP?  That reasonable PPP implementations will probably recover from
any single instance of reordered packets, possibly by tearing down the
link and restarting, is not the same as "does not adversely affect the
PPP stream."

Have you considered of bad effects that packet reordering has on TCP
performance?  Re-ordered TCP/IP segments cause retransmissions and reduce
the congestion window.  Those bad effects of re-ordered IP datagrams are
why official PPP multilink in RFC 1990 is much better than "BF&I" or "load
sharing" multilink, despite the greatly reduced complexity and theoretical
reliability of the latter.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Fri Jan 21 15:39:32 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA17045
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 15:39:30 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id LAA02439; Fri, 21 Jan 2000 11:34:11 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA02435 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 11:33:42 -0800 (PST)
Received: from mercurynt40 (192.168.211.161) by mcfeeley.indusriver.com (Worldmail 1.3.167) for l2tp@ipsec.org; 21 Jan 2000 14:32:02 -0500
Message-Id: <4.2.0.58.20000121143347.00a62af0@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 21 Jan 2000 14:40:13 -0500
To: l2tp@ipsec.org
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
In-Reply-To: <200001211756.KAA14797@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Another related topic is "PPP over ethernet" that
the underline layer is not lossless and not serial transmission.
Also, MP has to re-ordering frames anyway when fan-in (reverse multiplexing)
all individual links.
--- David

At 10:56 AM 1/21/00 -0700, you wrote:
> > From: "W. Mark Townsley" <townsley@cisco.com>
>
> > ...
> > RFC 1661 states that all packets are assummed to be delivered in order.
> > Some protocols (VJ header compression, PPP payload compression and
> > encryption) were designed with this in mind.
>
>You must also add RFC 1990 (MP) to that list of PPP protocols that assume
>that the order of the stream of packets is preserved.  Re-ordering MP
>fragments is not catastrophic in PPP systems that handle MP sequence number
>roll-over robustly, but re-ordering can stall the delivery of the data
>(e.g. IP) until the non-standard timers in robust MP implementations fire.
>
>
> >                                              Further, the CPs that run
> > on PPP may have a problem with out of order packets, though given their
> > stop and wait nature and relatively long default timeout between
> > retransmissions, it is unlikely.
>
>I think that is wrong.  The request-response nature of CP's makes
>re-ordering unlikely on paths that does re-order packets, but does not
>protect against re-ordering when it does happen.  Think about the
>many places in the RFC 1661 state machine that break catastrophically
>with any re-ordering.
>There are also interactions among CP's that would be broken by packet
>reordering, such between authentication and LCP, and between LCP and
>everything else.
>
>
> > The bottom line is, PPP has an in-order delivery assumption, thus L2TP
> > must have some way of not breaking this. It is optional, so if the user
> > understands the risks then sequence numbers may be omitted.
>
>Who is the "user" that is doing that understanding of risks?  The typical
>Windows 98 or even NT user does not have a clue about what an packet stream
>might be.  Many technical people at ISP's are as well informed, although
>they don't realize it and certainly won't admit it.
>
> >                                                             We have
> > found that, w/o the stateful protocols listed above, operating w/o
> > guaranteed delivery order reduces L2TP overhead, and does not adversly
> > affect the PPP stream.
>
>What measurements have you made of the results of packet reordering
>on PPP?  That reasonable PPP implementations will probably recover from
>any single instance of reordered packets, possibly by tearing down the
>link and restarting, is not the same as "does not adversely affect the
>PPP stream."
>
>Have you considered of bad effects that packet reordering has on TCP
>performance?  Re-ordered TCP/IP segments cause retransmissions and reduce
>the congestion window.  Those bad effects of re-ordered IP datagrams are
>why official PPP multilink in RFC 1990 is much better than "BF&I" or "load
>sharing" multilink, despite the greatly reduced complexity and theoretical
>reliability of the latter.
>
>
>Vernon Schryver    vjs@rhyolite.com

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Fri Jan 21 16:07:46 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA17350
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 16:07:42 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id LAA02462; Fri, 21 Jan 2000 11:54:15 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA02458 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 11:53:59 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id MAA20855
	for l2tp@ipsec.org  env-from <vjs>;
	Fri, 21 Jan 2000 12:53:57 -0700 (MST)
Date: Fri, 21 Jan 2000 12:53:57 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001211953.MAA20855@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> Another related topic is "PPP over ethernet" that
> the underline layer is not lossless and not serial transmission.

Most layers beneath PPP can suffer losses.  PPP itself is supposed to
throw away packets.  That's the purpose of the FCS in async framed PPP.

I do not know what is meant by "serial transmission".  Ethernet or
IEEE-802.3 seems pretty serial to me.  Even if you throw in bridges,
re-ordering is very rare, as required by many non-IP protocols.  The only
network media that I can think of off hand that are not serial as I
understand the word are old-style HIPPI (i.e. not the new fangled
serial-HIPPI) and some of the memory-to-memory and bus extension schemes.
Some people might object that the latter are not network media.  They are
certainly very short range, but some of them have carried TCP/IP.

If "serial transmission" is intended to mean "preserves order," then of
course, there are many networks (but that are not link layer media) that
fit that bill.


> Also, MP has to re-ordering frames anyway when fan-in (reverse multiplexing)
> all individual links.
> ...

No, MP as defined in RFC 1990 does not re-order frames.  MP is required
to deliver packets in the order implied by the MP fragment sequence
numbers, but the multilink protocol assumes the layers beneath each link
in the MP bundle delivers packets to the receiver in the same order as
the packets were put on the link by the transmitter.
Consider the implications of the non-decreasing sequence number rule
for receivers in RFC 1990.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Fri Jan 21 17:11:39 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA17985
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 17:11:37 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id MAA02537; Fri, 21 Jan 2000 12:33:38 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id MAA02533 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 12:33:16 -0800 (PST)
Received: from mercurynt40 (192.168.211.161) by mcfeeley.indusriver.com (Worldmail 1.3.167) for l2tp@ipsec.org; 21 Jan 2000 15:31:36 -0500
Message-Id: <4.2.0.58.20000121153211.00a5db40@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 21 Jan 2000 15:39:48 -0500
To: l2tp@ipsec.org
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
In-Reply-To: <200001211953.MAA20855@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Why PPP demanding the lower layer be serial transmission? (frame in sequence)
when its encapsulate IP(X)...that do not demand this.
If PPP needs this other then its encapsulate protocol then
it can use RFC1663.
--- David

At 12:53 PM 1/21/00 -0700, you wrote:
> > From: David Chen <dchen@indusriver.com>
>
> > Another related topic is "PPP over ethernet" that
> > the underline layer is not lossless and not serial transmission.
>
>Most layers beneath PPP can suffer losses.  PPP itself is supposed to
>throw away packets.  That's the purpose of the FCS in async framed PPP.
>
>I do not know what is meant by "serial transmission".  Ethernet or
>IEEE-802.3 seems pretty serial to me.  Even if you throw in bridges,
>re-ordering is very rare, as required by many non-IP protocols.  The only
>network media that I can think of off hand that are not serial as I
>understand the word are old-style HIPPI (i.e. not the new fangled
>serial-HIPPI) and some of the memory-to-memory and bus extension schemes.
>Some people might object that the latter are not network media.  They are
>certainly very short range, but some of them have carried TCP/IP.
>
>If "serial transmission" is intended to mean "preserves order," then of
>course, there are many networks (but that are not link layer media) that
>fit that bill.
>
>
> > Also, MP has to re-ordering frames anyway when fan-in (reverse 
> multiplexing)
> > all individual links.
> > ...
>
>No, MP as defined in RFC 1990 does not re-order frames.  MP is required
>to deliver packets in the order implied by the MP fragment sequence
>numbers, but the multilink protocol assumes the layers beneath each link
>in the MP bundle delivers packets to the receiver in the same order as
>the packets were put on the link by the transmitter.
>Consider the implications of the non-decreasing sequence number rule
>for receivers in RFC 1990.
>
>
>Vernon Schryver    vjs@rhyolite.com

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Fri Jan 21 18:02:01 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA18503
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 18:02:00 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id NAA02686; Fri, 21 Jan 2000 13:55:25 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id NAA02682 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 13:55:03 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id OAA25617
	for l2tp@ipsec.org  env-from <vjs>;
	Fri, 21 Jan 2000 14:55:02 -0700 (MST)
Date: Fri, 21 Jan 2000 14:55:02 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001212155.OAA25617@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> Why PPP demanding the lower layer be serial transmission? (frame in sequence)
> when its encapsulate IP(X)...that do not demand this.

Why?--because of history, as someone (perhaps you) observed early in
this thread.  PPP originally came from the fusion of two groups in the
IETF.  One was working on a more robust replacement for SLIP, RFC 1055.
The other was working on an interoperable, open-standard replacement
for proprietary router-to-router protocols over leased lines.  In both
cases, the underlying media never re-ordered frames, and so it made no
sense to add the major complications that would be required to deal with
link layers that might deliver out of order.

Even today, the overwhelming majority of all existing and likely
PPP links are on link layers that do not re-order frames.

You are mistaken about the requirements of the upper layers.  Many
protocols carried in PPP frames do not tolerate out-of-order packets.
Doesn't IPX require that its link layer not re-order packets?  The packets
supported by the PPP bridging protocol certainly must not be re-ordered.
Not even IPv4 is tolerant of re-ordered packets.  Some popular IP
implementations (e.g. for PC's) died when given out-of-order IP packets.
As I said earlier today, TCP/IP survives out-of-order segments, but does
not tolerate them, unless you define reduced speed as well as bandwidth
wasted on unnecessary retransmission counts as tolerance.


> If PPP needs this other then its encapsulate protocol then
> it can use RFC1663.

I do not understand that.  Before RFC 1663 can start working, LCP
must get to the OPENED state, and that requires that packets not
be re-ordered.  In other words, I do not see how RFC 1663 might
allow you to run PPP over a link that re-orders packets.

If you look in the archives of the PPP Compression mailing list and
probably also in the main PPP mailing list archives, you'll find that the
justification for RFC 1663 was to recover from lost packets for PPP payload
compression.  RFC 1663 was not intended to deal with packets re-ordered
by the lower layer, although I assume it would, if you could somehow also
ensure that LCP packets would not be re-ordered.

Another problem is the very limited support in the real world for RFC 1663.


PPP does not require a lossless, correctly ordered stream from the
lower layer.  The lower layer need do no more than detect out-of-order
packets and discard them instead of delivering them to upper layer.
To me, that requirement sounds trivial.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Fri Jan 21 22:04:20 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA21532
	for <l2tpext-archive@odin.ietf.org>; Fri, 21 Jan 2000 22:04:19 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id RAA02957; Fri, 21 Jan 2000 17:56:54 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id RAA02953 for <l2tp@ipsec.org>; Fri, 21 Jan 2000 17:56:27 -0800 (PST)
Received: from mercurynt40 (192.168.211.164) by mcfeeley.indusriver.com (Worldmail 1.3.167); 21 Jan 2000 20:54:39 -0500
Message-Id: <4.2.0.58.20000121204528.00a60c60@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 21 Jan 2000 21:02:52 -0500
To: Vernon Schryver <vjs@calcite.rhyolite.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <200001212155.OAA25617@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk


Does PPP recognized the difference of incoming packet has the 1,2,4,3,5 
(out of sequence)
order or 1,2,4,5 (missing packet).
If it does not, then it does not stop PPP function correctly even the lower 
layer does
not guarantee the frame transmitting sequence.
However, the performance could suffer.
--- David


At 02:55 PM 1/21/00 -0700, you wrote:
> > From: David Chen <dchen@indusriver.com>
>
> > Why PPP demanding the lower layer be serial transmission? (frame in 
> sequence)
> > when its encapsulate IP(X)...that do not demand this.
>
>Why?--because of history, as someone (perhaps you) observed early in
>this thread.  PPP originally came from the fusion of two groups in the
>IETF.  One was working on a more robust replacement for SLIP, RFC 1055.
>The other was working on an interoperable, open-standard replacement
>for proprietary router-to-router protocols over leased lines.  In both
>cases, the underlying media never re-ordered frames, and so it made no
>sense to add the major complications that would be required to deal with
>link layers that might deliver out of order.
>
>Even today, the overwhelming majority of all existing and likely
>PPP links are on link layers that do not re-order frames.
>
>You are mistaken about the requirements of the upper layers.  Many
>protocols carried in PPP frames do not tolerate out-of-order packets.
>Doesn't IPX require that its link layer not re-order packets?  The packets
>supported by the PPP bridging protocol certainly must not be re-ordered.
>Not even IPv4 is tolerant of re-ordered packets.  Some popular IP
>implementations (e.g. for PC's) died when given out-of-order IP packets.
>As I said earlier today, TCP/IP survives out-of-order segments, but does
>not tolerate them, unless you define reduced speed as well as bandwidth
>wasted on unnecessary retransmission counts as tolerance.
>
>
> > If PPP needs this other then its encapsulate protocol then
> > it can use RFC1663.
>
>I do not understand that.  Before RFC 1663 can start working, LCP
>must get to the OPENED state, and that requires that packets not
>be re-ordered.  In other words, I do not see how RFC 1663 might
>allow you to run PPP over a link that re-orders packets.
>
>If you look in the archives of the PPP Compression mailing list and
>probably also in the main PPP mailing list archives, you'll find that the
>justification for RFC 1663 was to recover from lost packets for PPP payload
>compression.  RFC 1663 was not intended to deal with packets re-ordered
>by the lower layer, although I assume it would, if you could somehow also
>ensure that LCP packets would not be re-ordered.
>
>Another problem is the very limited support in the real world for RFC 1663.
>
>
>PPP does not require a lossless, correctly ordered stream from the
>lower layer.  The lower layer need do no more than detect out-of-order
>packets and discard them instead of delivering them to upper layer.
>To me, that requirement sounds trivial.
>
>
>Vernon Schryver    vjs@rhyolite.com



From owner-l2tp@ipsec.org  Sat Jan 22 13:09:02 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA09908
	for <l2tpext-archive@odin.ietf.org>; Sat, 22 Jan 2000 13:09:01 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA04854; Sat, 22 Jan 2000 09:05:11 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA04850 for <l2tp@ipsec.org>; Sat, 22 Jan 2000 09:04:54 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id KAA16879
	for l2tp@ipsec.org  env-from <vjs>;
	Sat, 22 Jan 2000 10:04:52 -0700 (MST)
Date: Sat, 22 Jan 2000 10:04:52 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001221704.KAA16879@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> Does PPP recognized the difference of incoming packet has the 1,2,4,3,5 
> (out of sequence)
> order or 1,2,4,5 (missing packet).
> If it does not, then it does not stop PPP function correctly even the lower 
> layer does
> not guarantee the frame transmitting sequence.
> However, the performance could suffer.

It makes no sense to talk about PPP recognizing out of order packets.
Once again, for the umteenth time, the sequence 1,2,4,3,5 can break a PPP
link by confusing the state machine.

Please read RFC 1661 and understand the state machine.  Please look at
PPP packets and notice that they do not have any sequence numbers, and so
PPP cannot detect an out-of-order packet.  (The ID field in many PP packets
is not a sequence number and cannot serve as one for obvious reasons.)
Please accept the fact that there are upper layer protocols such as those
carried by PPP bridging packets that also require that their data never
be re-ordered, that cannot recognize a re-ordering and discard delayed
packets.

Yes, you could produce a garbage PPP implementation that re-ordered
packets.  It would work most of the time.  Most of the problems it would
cause would not be attributed to that bug and you might sell a lot of
it, because many people lack the skills, tools, and interesting to
determine the nature of bugs However, there are parts of the computer
industry where such garbage is not accepted.  Worse, such garbage would
eventually be figured out and laid at your doorstep (especially given
this public discussion), and then even the hordes that are not bothered
by the blue screen of death several times per day would shun your product.
Only the biggest vendors can get away with selling garbage, and even
they can't sell it forever.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Mon Jan 24 02:04:18 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id CAA26497
	for <l2tpext-archive@odin.ietf.org>; Mon, 24 Jan 2000 02:04:17 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id VAA08197; Sun, 23 Jan 2000 21:32:59 -0800 (PST)
Received: from kickme.cisco.com (kickme.cisco.com [198.92.30.42]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id VAA08193 for <l2tp@ipsec.org>; Sun, 23 Jan 2000 21:32:29 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by kickme.cisco.com (8.9.1a/8.9.1) with ESMTP id VAA05614;
	Sun, 23 Jan 2000 21:24:13 -0800 (PST)
Received: from cisco.com (rtp-dial-1-24.cisco.com [161.44.116.24]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id AAA14864; Mon, 24 Jan 2000 00:31:59 -0500 (EST)
Message-ID: <388B91EB.6546F80A@cisco.com>
Date: Sun, 23 Jan 2000 15:42:35 -0800
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Vernon Schryver <vjs@calcite.rhyolite.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001212155.OAA25617@calcite.rhyolite.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

Vernon Schryver wrote:
> 
> > From: David Chen <dchen@indusriver.com>
> 
> > Why PPP demanding the lower layer be serial transmission? (frame in sequence)
> > when its encapsulate IP(X)...that do not demand this.
> 
> Why?--because of history, as someone (perhaps you) observed early in
> this thread.  PPP originally came from the fusion of two groups in the
> IETF.  One was working on a more robust replacement for SLIP, RFC 1055.
> The other was working on an interoperable, open-standard replacement
> for proprietary router-to-router protocols over leased lines.  In both
> cases, the underlying media never re-ordered frames, and so it made no
> sense to add the major complications that would be required to deal with
> link layers that might deliver out of order.
> 
> Even today, the overwhelming majority of all existing and likely
> PPP links are on link layers that do not re-order frames.
> 
> You are mistaken about the requirements of the upper layers.  Many
> protocols carried in PPP frames do not tolerate out-of-order packets.

That is true, which is why we can turn on sequencing when necessary. 

> Doesn't IPX require that its link layer not re-order packets?  The packets
> supported by the PPP bridging protocol certainly must not be re-ordered.
> Not even IPv4 is tolerant of re-ordered packets.  Some popular IP
> implementations (e.g. for PC's) died when given out-of-order IP packets.
> As I said earlier today, TCP/IP survives out-of-order segments, but does
> not tolerate them, unless you define reduced speed as well as bandwidth
> wasted on unnecessary retransmission counts as tolerance.

We all know that, by design, IP may reorder packets. I see no reason why
it is L2TP's business to correct this for tunneled PPP/IP frames. If the
reordering of IP packets being detrimental to some broken stacks is
cause for enabling sequence mode by L2TP, then why limit L2TP traffic to
this new-found benefit for IP? We should start using L2TP in sequence
mode for all IP transmissions. L2TP becomes the packet sequencing
protocol for the Internet! This is of course obsurd, but it illustrates
the point that, when carrying IP within L2TP/PPP over IP, it is safe to
not bother with sequencing...  

> 
> > If PPP needs this other then its encapsulate protocol then
> > it can use RFC1663.
> 
> I do not understand that.  Before RFC 1663 can start working, LCP
> must get to the OPENED state, and that requires that packets not
> be re-ordered.  In other words, I do not see how RFC 1663 might
> allow you to run PPP over a link that re-orders packets.
> 
> If you look in the archives of the PPP Compression mailing list and
> probably also in the main PPP mailing list archives, you'll find that the
> justification for RFC 1663 was to recover from lost packets for PPP payload
> compression.  RFC 1663 was not intended to deal with packets re-ordered
> by the lower layer, although I assume it would, if you could somehow also
> ensure that LCP packets would not be re-ordered.
> 
> Another problem is the very limited support in the real world for RFC 1663.
> 
> PPP does not require a lossless, correctly ordered stream from the
> lower layer.  The lower layer need do no more than detect out-of-order
> packets and discard them instead of delivering them to upper layer.
> To me, that requirement sounds trivial.

In fact, there are some cases where sequence checking can actually be
dangerous on an IP link. Consider the following actual occurance that we
witnessed at a recent bakeoff. One implementation consisently reordered
fragmented L2TP packets. This UDP/IP stack was operating properly as IP
stacks are certainly allowed to reorder packets (and fragmentation is a
common cause of this). L2TP Sequence checking was enabled by the
receiving peer, thus dropping all out of order packets just as you
suggest. The end result was that no fragmented L2TP packets were ever
received. 

At any rate, yes there are indeed cases where preservation of order was
an important underlying assumption for PPP. For these, L2TP has a method
for enabling (and disabling) sequence numbers at any time during the
life of the session. 


> 
> Vernon Schryver    vjs@rhyolite.com




From owner-l2tp@ipsec.org  Mon Jan 24 02:28:38 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id CAA29213
	for <l2tpext-archive@odin.ietf.org>; Mon, 24 Jan 2000 02:28:37 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id WAA08265; Sun, 23 Jan 2000 22:28:59 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id WAA08250 for <l2tp@ipsec.org>; Sun, 23 Jan 2000 22:28:11 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id XAA07458
	for l2tp@ipsec.org  env-from <vjs>;
	Sun, 23 Jan 2000 23:28:10 -0700 (MST)
Date: Sun, 23 Jan 2000 23:28:10 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001240628.XAA07458@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: "W. Mark Townsley" <townsley@cisco.com>

> > You are mistaken about the requirements of the upper layers.  Many
> > protocols carried in PPP frames do not tolerate out-of-order packets.
>
> That is true, which is why we can turn on sequencing when necessary. 

Who is that "we," Kemo Sabe?
"We" cannot be end users configuring their PC's, since they don't
understand out-of-order from in-order.  It cannot even be system
administrators employed by ISP's, since many of them understand the issue
no better than typical end users (i.e. not at all), and the administrators
who do understand will have neither the time to figure out whether sequence
numbers are necessary, nor the tools to determine that necessity.


> ...
> We all know that, by design, IP may reorder packets. I see no reason why
> it is L2TP's business to correct this for tunneled PPP/IP frames.

That is a straw man.  No one said anything about correcting the
order of already mis-ordered IP frames.  
Never mind that in real life, out-of-order UDP/IP packets, out-of-order
TCP/IP segments, and out-of-order IP fragments are all rare.

>                                                                   If the
> reordering of IP packets being detrimental to some broken stacks is
> cause for enabling sequence mode by L2TP, then why limit L2TP traffic to
> this new-found benefit for IP?  We should start using L2TP in sequence
> mode for all IP transmissions. L2TP becomes the packet sequencing
> protocol for the Internet! This is of course obsurd, but it illustrates
> the point that, when carrying IP within L2TP/PPP over IP, it is safe to
> not bother with sequencing...  

I don't understand that either, even after discounting the hyperbolic 
rhetoric

It's true that some PC code is so broken that it does not merely
suffer retransmissions and closed congestion windows from out-of-order
TCP/IP segments, but goes wheels-up.  If you were saying that such
broken systems should not constrain L2TP, I would agree.

It is well known among that miss-ordering TCP/IP packets harms RFC
compliant TCP/IP implementations by significantly reducing throughput,
significantly increasing latency, and increasing wasted bandwidth.  Out
of order TCP/IP segments are required to trigger duplicate ACKs, and
duplicate ACKs cause both retransmissions and partially close the sender's
TCP congestion window.  None of this has anything to do with "broken
stacks," but with the opposite, complaint, good TCP code.

As I said before, a major reason that RFC 1990 MP is far more popular than
"load balancing" or "BF&I" multilink despite the vastly increased
complexity and reduced raw throughput and latency of MP is that TCP/IP
throughput suffers due to the inevitable mis-ordering of TCP/IP segments.
In other words, if it made sense in the real world to have L2TP not
preserve order, then it would also make sense to transmit IP packets in
MP bundles without MP headers.  Do you know of any routers that do that,
and if not, why not?

Unless you define "safe" as reducing performance and increasing wasted
bandwidth, allowing L2TP to mis-order IP packets is a not safe.


> ...
> In fact, there are some cases where sequence checking can actually be
> dangerous on an IP link. Consider the following actual occurance that we
> witnessed at a recent bakeoff. One implementation consisently reordered
> fragmented L2TP packets. This UDP/IP stack was operating properly as IP
> stacks are certainly allowed to reorder packets (and fragmentation is a
> common cause of this).

Since when is IP fragmentation of mis-ordered UDP/IP packets?  Either I
don't understand what you mean, or you are wrong.  Are you assuming that
a lost IP fragment in a first UDP transmission combined with a UDP
retransmission can produce a delayed UDP packet?  If so, perhaps in non-RFC
compliant UDP stacks+applications that cheat on changing the IP ID when
the application sends a new UDP packet (e.g. a retransmission), but I
think such systems are rare.


>                        L2TP Sequence checking was enabled by the
> receiving peer, thus dropping all out of order packets just as you
> suggest. The end result was that no fragmented L2TP packets were ever
> received. 

So one of the L2TP peers was badly broken and was transmitting packets
out of order?  And you infer somehow that preserving order is bad?
That reasoning does not make sense to me.


> At any rate, yes there are indeed cases where preservation of order was
> an important underlying assumption for PPP. For these, L2TP has a method
> for enabling (and disabling) sequence numbers at any time during the
> life of the session. 

The standard problem of standards committee protocols is bloat with
optional features and switches switches that no one will ever use because
they cannot be used in real life.  As far as I can see, any switch that
allows L2TP to mis-order either PPP or upper layer packets is a classic
example of a standards committe misfeature and bloat.  In real life, the
very few if any circumstances in which mis-ordered L2TP frames can be
turned on cannot or will be determined by the people who would have to be
doing the determining.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Mon Jan 24 05:06:46 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id FAA00599
	for <l2tpext-archive@odin.ietf.org>; Mon, 24 Jan 2000 05:06:45 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id AAA08542; Mon, 24 Jan 2000 00:39:05 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id AAA08538 for <l2tp@ipsec.org>; Mon, 24 Jan 2000 00:38:09 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id AAA08187;
	Mon, 24 Jan 2000 00:40:46 -0800 (PST)
Received: from cisco.com (rtp-dial-1-24.cisco.com [161.44.116.24]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id DAA15081; Mon, 24 Jan 2000 03:37:39 -0500 (EST)
Message-ID: <388C0F54.BDE34CA2@cisco.com>
Date: Mon, 24 Jan 2000 03:37:40 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Vernon Schryver <vjs@calcite.rhyolite.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001240628.XAA07458@calcite.rhyolite.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Vernon Schryver wrote:
> 
> > From: "W. Mark Townsley" <townsley@cisco.com>
> 
> > > You are mistaken about the requirements of the upper layers.  Many
> > > protocols carried in PPP frames do not tolerate out-of-order packets.
> >
> > That is true, which is why we can turn on sequencing when necessary.
> 
> Who is that "we," Kemo Sabe?

Those who implement the protocol. The LNS is in control of sequencing.
Thus the LNS can then make a determination of whether sequencing is
necessary via heuristics, not necessarily via config. So, the protocol
allows for the programmers to design in sequence checking when they deem
necessary. The "dialup user" does not have to be cognizant of this at
all.

> "We" cannot be end users configuring their PC's, since they don't
> understand out-of-order from in-order.  It cannot even be system
> administrators employed by ISP's, since many of them understand the issue
> no better than typical end users (i.e. not at all), and the administrators
> who do understand will have neither the time to figure out whether sequence
> numbers are necessary, nor the tools to determine that necessity.

I have more faith in the network admins than you do. I have worked with
many very bright network admins that certainly understand what packet
ordering is and is not. Perhaps it is a misleading artifact of who my
customer base is, but when I refer to a "user", I am often referring to
a customer who actually understands networking (e.g. *not* the average
DUN user at his windows box, but the network admin configuring the
router/server/etc). 

> 
> > ...
> > We all know that, by design, IP may reorder packets. I see no reason why
> > it is L2TP's business to correct this for tunneled PPP/IP frames.
> 
> That is a straw man.  No one said anything about correcting the
> order of already mis-ordered IP frames.

Then I misunderstood you.

> Never mind that in real life, out-of-order UDP/IP packets, out-of-order
> TCP/IP segments, and out-of-order IP fragments are all rare.
> 
> >                                                                   If the
> > reordering of IP packets being detrimental to some broken stacks is
> > cause for enabling sequence mode by L2TP, then why limit L2TP traffic to
> > this new-found benefit for IP?  We should start using L2TP in sequence
> > mode for all IP transmissions. L2TP becomes the packet sequencing
> > protocol for the Internet! This is of course obsurd, but it illustrates
> > the point that, when carrying IP within L2TP/PPP over IP, it is safe to
> > not bother with sequencing...
> 
> I don't understand that either, even after discounting the hyperbolic
> rhetoric
> 
> It's true that some PC code is so broken that it does not merely
> suffer retransmissions and closed congestion windows from out-of-order
> TCP/IP segments, but goes wheels-up.  If you were saying that such
> broken systems should not constrain L2TP, I would agree.

Yes. I was refuting using broken TCP/IP stacks as a design constraint on
L2TP.

> 
> It is well known among that miss-ordering TCP/IP packets harms RFC
> compliant TCP/IP implementations by significantly reducing throughput,
> significantly increasing latency, and increasing wasted bandwidth.  Out
> of order TCP/IP segments are required to trigger duplicate ACKs, and
> duplicate ACKs cause both retransmissions and partially close the sender's
> TCP congestion window.  None of this has anything to do with "broken
> stacks," but with the opposite, complaint, good TCP code.
> 
> As I said before, a major reason that RFC 1990 MP is far more popular than
> "load balancing" or "BF&I" multilink despite the vastly increased
> complexity and reduced raw throughput and latency of MP is that TCP/IP
> throughput suffers due to the inevitable mis-ordering of TCP/IP segments.
> In other words, if it made sense in the real world to have L2TP not
> preserve order, then it would also make sense to transmit IP packets in
> MP bundles without MP headers.  Do you know of any routers that do that,
> and if not, why not?
> 
> Unless you define "safe" as reducing performance and increasing wasted
> bandwidth, allowing L2TP to mis-order IP packets is a not safe.

I don't believe L2TP should mis-order packets, but it should not have to
go out of its way to preserve the order of already mis-ordered IP
frames.

> 
> > ...
> > In fact, there are some cases where sequence checking can actually be
> > dangerous on an IP link. Consider the following actual occurance that we
> > witnessed at a recent bakeoff. One implementation consisently reordered
> > fragmented L2TP packets. This UDP/IP stack was operating properly as IP
> > stacks are certainly allowed to reorder packets (and fragmentation is a
> > common cause of this).
> 
> Since when is IP fragmentation of mis-ordered UDP/IP packets?  Either I
> don't understand what you mean, or you are wrong.  Are you assuming that

I don't think you understood. I was referring to fragmentation of an IP
packet transmitted from the client. The circumstance causes the two IP
fragments, each framed as individual PPP packets, to be repeatedly
misordered exiting the LAC (due to fragmentation of one of the two
already fragmented packets at the LAC). It was not actually L2TP that
was misordering the packets, it was the IP stack on the LAC. It did so
because of fragmentation of one of the two initial IP fragments.

Perhaps more detail would help: The smaller of the two IP fragments
coming from the PPP client was getting switched out before the
fragmentation of the larger packet could occur, causing the L2TP packet
to be consistently out of order from the LAC. The key here is that
everyone was operating within spec, and the ultimate result was that the
the packet would never, ever, be received on the other end. Basic
sequence checking of IP assumes that reordering is a random phenomena,
when in fact it is not. IP sequence checking in the face of a consistent
packet misordering situation results in an unrecoverable denial of
service (unless a route change or some such actually restores the order
by chance). 

> a lost IP fragment in a first UDP transmission combined with a UDP
> retransmission can produce a delayed UDP packet?  If so, perhaps in non-RFC
> compliant UDP stacks+applications that cheat on changing the IP ID when
> the application sends a new UDP packet (e.g. a retransmission), but I
> think such systems are rare.
> 
> >                        L2TP Sequence checking was enabled by the
> > receiving peer, thus dropping all out of order packets just as you
> > suggest. The end result was that no fragmented L2TP packets were ever
> > received.
> 
> So one of the L2TP peers was badly broken and was transmitting packets
> out of order?  And you infer somehow that preserving order is bad?
> That reasoning does not make sense to me.

No one was broken. Preserving order by dropping packets caused a bad
problem.

> 
> > At any rate, yes there are indeed cases where preservation of order was
> > an important underlying assumption for PPP. For these, L2TP has a method
> > for enabling (and disabling) sequence numbers at any time during the
> > life of the session.
> 
> The standard problem of standards committee protocols is bloat with
> optional features and switches switches that no one will ever use because
> they cannot be used in real life.  As far as I can see, any switch that

There is at least some notion of removal of features that never get used
as the document advances the standards track.

What you say about committee-born protocols is true in many regards, and
L2TP is niether immune nor alone in this arena. The upside in our case
is a fairly large number of interoperable implementations, and general
industry acceptance. 

> allows L2TP to mis-order either PPP or upper layer packets is a classic
> example of a standards committe misfeature and bloat.  In real life, the
> very few if any circumstances in which mis-ordered L2TP frames can be
> turned on cannot or will be determined by the people who would have to be
> doing the determining.

Either the code or the network admins will be doing this. I would not
expect the "average dialup user" to understand the implications of
packet sequencing. 

- Mark

> 
> Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Mon Jan 24 12:56:48 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA16395
	for <l2tpext-archive@odin.ietf.org>; Mon, 24 Jan 2000 12:56:48 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id IAA10378; Mon, 24 Jan 2000 08:05:03 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id IAA10374 for <l2tp@ipsec.org>; Mon, 24 Jan 2000 08:04:37 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167) for l2tp@ipsec.org; 24 Jan 2000 11:02:48 -0500
Message-Id: <4.2.0.58.20000124110128.00b1cd40@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Mon, 24 Jan 2000 11:11:05 -0500
To: l2tp@ipsec.org
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
In-Reply-To: <200001221704.KAA16879@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

The design principle of the state machine is that
if you are in state that have seen packet #1,2 and not seeing #3 then you
goto other state otherwise goto yet another state.

The rational here is that PPP state machine does not differentiate the 
difference of ....
unless the flow control is implemented in PPP.

As far as TCP, one of its design drawback is that it does not handle out of 
sequence packet.
The L2tp or any underlining layers of (TCP) are not obligated to make up 
its design deficiency.

--- David


At 10:04 AM 1/22/00 -0700, you wrote:
> > From: David Chen <dchen@indusriver.com>
>
> > Does PPP recognized the difference of incoming packet has the 1,2,4,3,5
> > (out of sequence)
> > order or 1,2,4,5 (missing packet).
> > If it does not, then it does not stop PPP function correctly even the 
> lower
> > layer does
> > not guarantee the frame transmitting sequence.
> > However, the performance could suffer.
>
>It makes no sense to talk about PPP recognizing out of order packets.
>Once again, for the umteenth time, the sequence 1,2,4,3,5 can break a PPP
>link by confusing the state machine.
>
>Please read RFC 1661 and understand the state machine.  Please look at
>PPP packets and notice that they do not have any sequence numbers, and so
>PPP cannot detect an out-of-order packet.  (The ID field in many PP packets
>is not a sequence number and cannot serve as one for obvious reasons.)
>Please accept the fact that there are upper layer protocols such as those
>carried by PPP bridging packets that also require that their data never
>be re-ordered, that cannot recognize a re-ordering and discard delayed
>packets.
>
>Yes, you could produce a garbage PPP implementation that re-ordered
>packets.  It would work most of the time.  Most of the problems it would
>cause would not be attributed to that bug and you might sell a lot of
>it, because many people lack the skills, tools, and interesting to
>determine the nature of bugs However, there are parts of the computer
>industry where such garbage is not accepted.  Worse, such garbage would
>eventually be figured out and laid at your doorstep (especially given
>this public discussion), and then even the hordes that are not bothered
>by the blue screen of death several times per day would shun your product.
>Only the biggest vendors can get away with selling garbage, and even
>they can't sell it forever.
>
>
>Vernon Schryver    vjs@rhyolite.com

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Tue Jan 25 16:00:05 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA07713
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:00:03 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id JAA10591; Mon, 24 Jan 2000 09:47:25 -0800 (PST)
Received: from dirty.research.bell-labs.com (dirty.research.bell-labs.com [204.178.16.6]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id JAA10583 for <l2tp@ipsec.org>; Mon, 24 Jan 2000 09:47:01 -0800 (PST)
Received: from zubin.dnrc.bell-labs.com ([135.180.130.56]) by dirty; Mon Jan 24 12:41:45 EST 2000
Received: from dnrc.bell-labs.com (ex-vpn67.pa.bell-labs.com [135.250.1.67])
	by zubin.dnrc.bell-labs.com (8.9.3/8.9.3) with ESMTP id MAA03613;
	Mon, 24 Jan 2000 12:41:40 -0500 (EST)
Message-ID: <388C8F46.D57F4024@dnrc.bell-labs.com>
Date: Mon, 24 Jan 2000 09:43:34 -0800
From: Grenville Armitage <gja@dnrc.bell-labs.com>
Organization: Bell Laboratories, Lucent Technologies
X-Mailer: Mozilla 4.61 [en] (WinNT; U)
X-Accept-Language: en
MIME-Version: 1.0
To: "W. Mark Townsley" <townsley@cisco.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001212155.OAA25617@calcite.rhyolite.com> <388B91EB.6546F80A@cisco.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



"W. Mark Townsley" wrote:
	[..]
> We all know that, by design, IP may reorder packets.

Judging by your subsequent response to Vern, this concept is
being interpreted more liberally than I think is warranted.

It is a design goal for protocols running over IP that they
dont blue-screen when reordering occurs. But there's no
design goal that such protocols must work efficiently, or
even at all, in the presence of reordered packets. Reordering
'may' occur as a rare consequence of routing transients. This
doesn't justify IP stacks (either at the end or in the middle)
from *consistently* reordering packets (or fragments) - that's
just broken stack design (especially since such a broken stack in
a LAC or end-host will typically still be in the path after a routing
change)

So IP *may* reorder packets as a rare event. But any design
(network or node) that causes regular, correlated, and repeatable
reordering is broken.

cheers,
gja
________________________________________________________________________
Grenville Armitage                            http://mh005.infi.net/~gja
Bell Labs Research Silicon Valley


From owner-l2tp@ipsec.org  Tue Jan 25 16:12:16 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA07853
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:12:15 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id LAA00194; Tue, 25 Jan 2000 11:52:13 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA00183; Tue, 25 Jan 2000 11:51:24 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA28798;
	Tue, 25 Jan 2000 09:21:35 -0800 (PST)
Received: from cisco.com (chacha-async-37.cisco.com [171.69.123.137]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA15518; Tue, 25 Jan 2000 12:02:52 -0500 (EST)
Message-ID: <388DD711.BE32CB14@cisco.com>
Date: Tue, 25 Jan 2000 12:02:09 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Vernon Schryver <vjs@calcite.rhyolite.com>, admin@ipsec.org
CC: l2tp@ipsec.org
Subject: Re: list archives
References: <200001241829.LAA20154@calcite.rhyolite.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


Thanks.

The administrator has been notified of this, and is now being notified
again. If administration of the list is too burdensome for ipsec.org, I
will be happy to solicit volunteers from others who might be able to
host it.

I believe I have all messages since Novemember, and can provide them to
populate the archive to get it back up to date.

- Mark

Vernon Schryver wrote:
> 
> Where is the archive for this mailing list?
> 
> I mentioned the recent discussions to someone, and he pointed out that
> http://www.ipsec.org/email/l2tp/ seems to have died last November.
> 
> Vernon Schryver    vjs@rhyolite.com



From owner-l2tp@ipsec.org  Tue Jan 25 16:12:27 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA07864
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:12:25 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id LAA00206; Tue, 25 Jan 2000 11:52:40 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA00175 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 11:51:19 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA28750;
	Tue, 25 Jan 2000 09:21:27 -0800 (PST)
Received: from cisco.com (chacha-async-37.cisco.com [171.69.123.137]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA15514; Tue, 25 Jan 2000 12:02:48 -0500 (EST)
Message-ID: <388DD414.C4D401DF@cisco.com>
Date: Tue, 25 Jan 2000 11:49:24 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Grenville Armitage <gja@dnrc.bell-labs.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001212155.OAA25617@calcite.rhyolite.com> <388B91EB.6546F80A@cisco.com> <388C8F46.D57F4024@dnrc.bell-labs.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


I won't claim to be the foremost IP stack expert for a moment, but
common sense to me says that if I receive an IP packet that must be
fragmented followed by a bunch of smaller packets that do not, it makes
sense to go and and switch the smaller packets rather than queuing them
while I chew on the fragmentation. If you have hardware or even driver
assisted fragmentation then perhaps it is not an issue, but if your
processer has to sit there and allocate a bunch of buffers, chop up the
packet, copy data, etc. this could take a serious amount of time.
Meanwhile, your input queue is growing, and latency increasing. Even if
the packets are associated with the same TCP stream (which on a middle
node they are likely not) latency can be as much of a burden to
performance as packet loss (ask anyone in tcpsat about that). I am
curious though, which RFC states that IP packet reordering may only
occur during route transitions? 

- Mark 

Grenville Armitage wrote:
> 
> "W. Mark Townsley" wrote:
>         [..]
> > We all know that, by design, IP may reorder packets.
> 
> Judging by your subsequent response to Vern, this concept is
> being interpreted more liberally than I think is warranted.
> 
> It is a design goal for protocols running over IP that they
> dont blue-screen when reordering occurs. But there's no
> design goal that such protocols must work efficiently, or
> even at all, in the presence of reordered packets. Reordering
> 'may' occur as a rare consequence of routing transients. This
> doesn't justify IP stacks (either at the end or in the middle)
> from *consistently* reordering packets (or fragments) - that's
> just broken stack design (especially since such a broken stack in
> a LAC or end-host will typically still be in the path after a routing
> change)
> 
> So IP *may* reorder packets as a rare event. But any design
> (network or node) that causes regular, correlated, and repeatable
> reordering is broken.
> 
> cheers,
> gja
> ________________________________________________________________________
> Grenville Armitage                            http://mh005.infi.net/~gja
> Bell Labs Research Silicon Valley




From owner-l2tp@ipsec.org  Tue Jan 25 16:14:31 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA07907
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:14:30 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id KAA10664; Mon, 24 Jan 2000 10:29:15 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id KAA10660 for <l2tp@ipsec.org>; Mon, 24 Jan 2000 10:29:07 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id LAA20154
	for l2tp@ipsec.org  env-from <vjs>;
	Mon, 24 Jan 2000 11:29:05 -0700 (MST)
Date: Mon, 24 Jan 2000 11:29:05 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001241829.LAA20154@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: list archives
Sender: owner-l2tp@sailpix.com
Precedence: bulk

Where is the archive for this mailing list?

I mentioned the recent discussions to someone, and he pointed out that
http://www.ipsec.org/email/l2tp/ seems to have died last November.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 16:27:45 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA07940
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:27:44 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id MAA00342; Tue, 25 Jan 2000 12:04:53 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id MAA00336 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 12:04:17 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id PAA24576
	for l2tp@ipsec.org  env-from <vjs>;
	Mon, 24 Jan 2000 15:02:49 -0700 (MST)
Date: Mon, 24 Jan 2000 15:02:49 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001242202.PAA24576@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> IP service does not guarantee packet delivery in sequence.
> Clearly, UDP/IP does not neither and
> TCP treat out of sequence packet as if loss of packet. -
> as appose to buffering and re-ordering (packets).
> (therefore, TCP provides in sequence no-loss packet delivery)

I think I did not say that "TCP [treats an] out of sequence packet as if
[it were] loss of packet," and I think it is wrong.  Any reasonable TCP
implemention will save an out-of-order segment that is in the receive
window, and generate an ACK for the current edge of the window.  I did
say that out of order packets cause duplicate ACKs which cause
retransmissions and reduction of the congestion window.  The difference
between "treat out of order as loss" and what I said is technical, but I
think the difference is significant.


> As for L2tp, a packet such as PAYLOAD/TCP/IP/PPP/L2TP/UDP/IP on the 
> protocol stack,
> is clearly that L2TP does not need keeping the packet sequence and dropping 
> the packet for
> the same reason (keeping packet sequence).

Maybe that's clearly correct to you.  If I understand what you mean, the
opposite is clearly correct to me.  As I said before, PPP does not and
cannot drop control protocol packets when they are out of order.  Instead,
the PPP state machines must and do act on the miss-ordered stimuli.  TCP
also generally does not drop segments when they are out of order.  Also
as I keep pointing out, there are more PPP payloads than TCP/IP, and some
of them, such as bridging, do not tolerate reordering at all.


Have you read RFC 1661 and understood the state machine?  Do you know what
I mean by "congestion window" and "duplicate ACKs?"  Do you know what I
mean when I repeatedly mention bridging?  Perhaps you do, but I have the
nagging feeling that you are arguing based on how you think PPP and TCP
should or work or be specified instead of how they are specified and work.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 16:31:14 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA08025
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:31:13 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id LAA00213; Tue, 25 Jan 2000 11:53:08 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA00177 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 11:51:20 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id JAA28700;
	Tue, 25 Jan 2000 09:21:28 -0800 (PST)
Received: from cisco.com (chacha-async-37.cisco.com [171.69.123.137]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id MAA15510; Tue, 25 Jan 2000 12:02:42 -0500 (EST)
Message-ID: <388DD3D0.F48B278@cisco.com>
Date: Tue, 25 Jan 2000 11:48:16 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Vernon Schryver <vjs@calcite.rhyolite.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001241722.KAA18857@calcite.rhyolite.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


So, a router interface should queue all incoming IP packets while it
takes a large one and splits it into one or more IP fragments? 

Sigh, this discussion has veered away from the meaningful, and is headed
down a course of toward who misunderstood who like some bad comedy. 

The bottom line is this: L2TP data packets carrying PPP/IP is ultimately
an IP packet within an IP packet. Applying sequencing checking to a
tunneled IP packet would surely make a the original IP architects
shudder just as much as reordering PPP frames does for a PPP guru. Each
are fundamental, opposing, assumptions for each protocol, respectively. 

Together, we have identified cases where checking packet sequence and
not checking packet sequence can break the tunneled connection. We could
compare pathological examples all day. In practice, I have seen only one
of these occur in a repeatable fashion. If you want to avoid this, my
recommendation is to not check sequencing on tunneled IP packets. 

- Mark

Vernon Schryver wrote:
> 
> > From: "W. Mark Townsley" <townsley@cisco.com>
> 
> > ...
> > > > We all know that, by design, IP may reorder packets. I see no reason why
> > > > it is L2TP's business to correct this for tunneled PPP/IP frames.
> > >
> > > That is a straw man.  No one said anything about correcting the
> > > order of already mis-ordered IP frames.
> >
> > Then I misunderstood you.
> 
> I do not see how you could have honestly believed I was talking about
> fixing the order of already out of order IP packets.  That task is
> obviously impossible.  There is no way that any intermediate box (or in
> the case of L2TP, 2 or 3 boxes) can know the correct order of IP payloads.
> An intermediate box cannot rely on IP ID's, and there is no other clue.
> Only the sending and receiving hosts know the correct order of UDP packets.
> 
> > ...
> > Yes. I was refuting using broken TCP/IP stacks as a design constraint on
> > L2TP.
> 
> Ok, it's good we're not arguing about that either.  Let's also apply that
> idea to not using broken TCP/IP stacks inside L2TP implementations as a
> design constraint on L2TP or as a justification for doing something wrong.
> 
> > > It is well known among that miss-ordering TCP/IP packets harms RFC
> > > compliant TCP/IP implementations  ...
> 
> > I don't believe L2TP should mis-order packets, but it should not have to
> > go out of its way to preserve the order of already mis-ordered IP
> > frames.
> 
> Again, that makes no sense to me.  L2TP obviously cannot do anything
> about already mis-ordered IP frames.  I thought the sequence number
> inside L2TP is to ensure the L2TP does not increase mis-ordering
> as the result of mis-ordering from layers below L2TP, and also to
> keep L2TP from breaking things that do not tolerate any mis-ordering
> including PPP bridging and PPP CP packets.
> 
> > ...
> > I don't think you understood. I was referring to fragmentation of an IP
> > packet transmitted from the client. The circumstance causes the two IP
> > fragments, each framed as individual PPP packets, to be repeatedly
> > misordered exiting the LAC (due to fragmentation of one of the two
> > already fragmented packets at the LAC). It was not actually L2TP that
> > was misordering the packets, it was the IP stack on the LAC. It did so
> > because of fragmentation of one of the two initial IP fragments.
> >
> > Perhaps more detail would help: The smaller of the two IP fragments
> > sequence checking of IP assumes that reordering is a random phenomena,
> > when in fact it is not. IP sequence checking in the face of a consistent
> > packet misordering situation results in an unrecoverable denial of
> > service (unless a route change or some such actually restores the order
> > by chance).
> 
> So you saw a router that consistently moves smaller packets ahead of larger
> packets?  I think that's compliant, but generally a bad idea, and certainly
> not justification for turning off L2TP sequence numbers.  Note that I've
> shipped PPP code that optionally ordered its transmit queue to send most
> *but not all* tinggrams ahead of big packets as an ad hoc type-of-service
> differentiation.  I no longer recommend that hack.
> 
> You are justifying a bad idea (turning off L2TP sequence numbers) based
> on the bad behavior a broken box (one that consistently mis-orders IP
> packets).
> 
> Put asside short term sales considerations and think about the long term.
> Which do you think would be the better course, fixing the broken box to
> not mis-order IP packets or turning off L2TP sequence numbers?  The broken
> box will cause significantly reduced TCP throughput and increased latency
> regardless of whether it uses L2TP sequence numbers.  It will also stress
> the IP fragmentation reassembly mechanisms in receivers, and that will
> cause increased IP packet losses.  Eventually even people who are so inured
> to garbage that they tolerate daily blue screens of death will notice the
> difference between using this broken box and others.  Since they won't be
> constrained by the availability of applications like MS Office and IE,
> and since the primary criteria on which routers are evaluated is speed
> and reliability, they'll switch.
> 
> > ...
> > > The standard problem of standards committee protocols is bloat with
> > > optional features and switches switches that no one will ever use because
> > > they cannot be used in real life.  As far as I can see, any switch that
> >
> > There is at least some notion of removal of features that never get used
> > as the document advances the standards track.
> 
> That sentiment is always offered to justify bloating a protocol with
> optional bad ideas.  Because in real life standards committees specifically
> including PPP working groups, optional bad ideas that get as far Proposed
> Standard are *NEVER* removed makes the sentiment at best uninformed.
> Consider the many bad optional ideas in and among PPP protocols, and notice
> how many have been removed.  Some have been proposed for deletion, but
> somehow the removal efforts always stall.
> 
> > What you say about committee-born protocols is true in many regards, and
> > L2TP is niether immune nor alone in this arena. The upside in our case
> > is a fairly large number of interoperable implementations, and general
> > industry acceptance.
> 
> That bit about "general industry acceptance" is usually said of protocols
> for which it is in fact false.  I hope it is true in this case, but I
> can't tell.  It's probably too early to say.
> 
> Vernon Schryver    vjs@rhyolite.com




From owner-l2tp@ipsec.org  Tue Jan 25 16:32:56 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA08051
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:32:55 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA10582; Mon, 24 Jan 2000 09:46:56 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA10578 for <l2tp@ipsec.org>; Mon, 24 Jan 2000 09:46:39 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id KAA19404
	for l2tp@ipsec.org  env-from <vjs>;
	Mon, 24 Jan 2000 10:46:37 -0700 (MST)
Date: Mon, 24 Jan 2000 10:46:37 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001241746.KAA19404@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> The design principle of the state machine is that
> if you are in state that have seen packet #1,2 and not seeing #3 then you
> goto other state otherwise goto yet another state.
>
> The rational here is that PPP state machine does not differentiate the 
> difference of ....
> unless the flow control is implemented in PPP.

That so called design principle is how all state machines work.  It's
the definition of a state machine.

How did flow control get into this discussion?  We were talking about
the fact that the PPP state machine, which is used in at most slightly
modified form in all PPP protocols assumes control packets are not
re-ordered by the lower layers.


> As far as TCP, one of its design drawback is that it does not handle out of 
> sequence packet.
> The L2tp or any underlining layers of (TCP) are not obligated to make up 
> its design deficiency.

I do not understand any of that.  TCP throughput and latency suffer
significantly should the lower layers re-order packets.  However, the
primary design goal and description of TCP is to transport a reliable,
sequenced stream of bytes.  If TCP "does not handle out of sequence
packet," then the world needs to know.

Anyone who wants to run L2TP or any other underlining (underlying?) layers
just for their own sake is welcome to do so.  However, most people won't
use L2TP except to move TCP/IP.  If L2TP does not cater to the whims of
TCP/IP or otherwise allows TCP/IP to not do well, then people will blame
L2TP instead of TCP/IP.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 16:36:32 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA08093
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:36:29 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id MAA00343; Tue, 25 Jan 2000 12:05:07 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id MAA00334 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 12:04:15 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id LAA23121
	for l2tp@ipsec.org  env-from <vjs>;
	Tue, 25 Jan 2000 11:14:00 -0700 (MST)
Date: Tue, 25 Jan 2000 11:14:00 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001251814.LAA23121@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: "W. Mark Townsley" <townsley@cisco.com>

> So, a router interface should queue all incoming IP packets while it
> takes a large one and splits it into one or more IP fragments? 

Not exactly.  A router must be largely FIFO for a given UDP or TCP
stream.  An occassional reordering is tolerable and unavoidable,
but consistent re-ordering is broken, cannot be tolerated, and will
eventually be rejected by the market.

Why have some of Cisco routers for many years tried to keep a given TCP
stream associated with a given output port when load balancing over equal
paths?  Why is the standard Cisco response to observations about apparent
inconsistenties or flapping in paths reported by traceroute or ping -R
(LSRR) that ICMP packets get special handling but that TCP/IP packets
generally see a constant path?  Why go to that significant trouble and
expense to associate TCP streams with output ports (or routes) except to
avoid IP re-ordering, and why both about IP re-ordering?  My guess is that
some people at Cisco know of the Bad Things that happen with significant
packet reo-ordering.

Again, if you think that significant re-ordering TCP/IP packets by PPP
routers is tolerable, why doesn't Cisco support "load balancing" or "BF&I"
PPP multilink better?  BF&I requires less processing by the sending
router and an enormous amount less by the receiver than RFC 1990.  BF&I
also has better throughput and latency.


> ...
> The bottom line is this: L2TP data packets carrying PPP/IP is ultimately
> an IP packet within an IP packet. Applying sequencing checking to a
> tunneled IP packet would surely make a the original IP architects
> shudder just as much as reordering PPP frames does for a PPP guru. Each
> are fundamental, opposing, assumptions for each protocol, respectively. 

Maybe you're right, but I bet not.  Shall we ask some?

By the way, do you remember when a lot of IP packets of what is now called
the Internet were carried on x.25 circuits?  My vague recollection is also
that x.25 has a sequence number, albeit rather small.


> Together, we have identified cases where checking packet sequence and
> not checking packet sequence can break the tunneled connection. We could
> compare pathological examples all day. In practice, I have seen only one
> of these occur in a repeatable fashion. If you want to avoid this, my
> recommendation is to not check sequencing on tunneled IP packets. 

I don't understand.  Are you recommending that most L2TP boxes turn off
sequence numbers, only that broken box that can't IP fragment sanely,
or only that broken box and only some of the time?  If the last, when?

Aren't you still patching one pathology, a router that consistently
mis-orders some packets, with another, worse pathology, turning
off L2TP sequencing?

You have not explicitly addressed my implicit question, but I have
inferred what might not be your answer.  Let me make it explicit.
Do you recommend under any circumstances turning off L2TP sequencing for
non-IP packets such as bridging, IPX and PPP CP packets on that wonderful
box that consistently re-orders when IP fragmenting?


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 16:45:40 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA08191
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 16:45:39 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id MAA00612; Tue, 25 Jan 2000 12:36:35 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id MAA00607 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 12:36:12 -0800 (PST)
Received: from mercurynt40 (192.168.211.201) by mcfeeley.indusriver.com (Worldmail 1.3.167); 25 Jan 2000 10:36:49 -0500
Message-Id: <4.2.0.58.20000125104300.00b13780@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Tue, 25 Jan 2000 10:45:04 -0500
To: Vernon Schryver <vjs@calcite.rhyolite.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <200001241746.KAA19404@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

IP service does not guarantee packet delivery in sequence.
Clearly, UDP/IP does not neither and
TCP treat out of sequence packet as if loss of packet. -
as appose to buffering and re-ordering (packets).
(therefore, TCP provides in sequence no-loss packet delivery)
As for L2tp, a packet such as PAYLOAD/TCP/IP/PPP/L2TP/UDP/IP on the 
protocol stack,
is clearly that L2TP does not need keeping the packet sequence and dropping 
the packet for
the same reason (keeping packet sequence).
--- David


At 10:46 AM 1/24/00 -0700, you wrote:
> > From: David Chen <dchen@indusriver.com>
>
> > The design principle of the state machine is that
> > if you are in state that have seen packet #1,2 and not seeing #3 then you
> > goto other state otherwise goto yet another state.
> >
> > The rational here is that PPP state machine does not differentiate the
> > difference of ....
> > unless the flow control is implemented in PPP.
>
>That so called design principle is how all state machines work.  It's
>the definition of a state machine.
>
>How did flow control get into this discussion?  We were talking about
>the fact that the PPP state machine, which is used in at most slightly
>modified form in all PPP protocols assumes control packets are not
>re-ordered by the lower layers.
>
>
> > As far as TCP, one of its design drawback is that it does not handle 
> out of
> > sequence packet.
> > The L2tp or any underlining layers of (TCP) are not obligated to make up
> > its design deficiency.
>
>I do not understand any of that.  TCP throughput and latency suffer
>significantly should the lower layers re-order packets.  However, the
>primary design goal and description of TCP is to transport a reliable,
>sequenced stream of bytes.  If TCP "does not handle out of sequence
>packet," then the world needs to know.
>
>Anyone who wants to run L2TP or any other underlining (underlying?) layers
>just for their own sake is welcome to do so.  However, most people won't
>use L2TP except to move TCP/IP.  If L2TP does not cater to the whims of
>TCP/IP or otherwise allows TCP/IP to not do well, then people will blame
>L2TP instead of TCP/IP.
>
>
>Vernon Schryver    vjs@rhyolite.com



From owner-l2tp@ipsec.org  Tue Jan 25 17:07:36 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA08723
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 17:07:35 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA10561; Mon, 24 Jan 2000 09:23:37 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA10557 for <l2tp@ipsec.org>; Mon, 24 Jan 2000 09:22:58 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id KAA18857
	for l2tp@ipsec.org  env-from <vjs>;
	Mon, 24 Jan 2000 10:22:57 -0700 (MST)
Date: Mon, 24 Jan 2000 10:22:57 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001241722.KAA18857@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: "W. Mark Townsley" <townsley@cisco.com>

> ...
> > > We all know that, by design, IP may reorder packets. I see no reason why
> > > it is L2TP's business to correct this for tunneled PPP/IP frames.
> > 
> > That is a straw man.  No one said anything about correcting the
> > order of already mis-ordered IP frames.
>
> Then I misunderstood you.

I do not see how you could have honestly believed I was talking about
fixing the order of already out of order IP packets.  That task is
obviously impossible.  There is no way that any intermediate box (or in
the case of L2TP, 2 or 3 boxes) can know the correct order of IP payloads.
An intermediate box cannot rely on IP ID's, and there is no other clue.
Only the sending and receiving hosts know the correct order of UDP packets.

> ...
> Yes. I was refuting using broken TCP/IP stacks as a design constraint on
> L2TP.

Ok, it's good we're not arguing about that either.  Let's also apply that
idea to not using broken TCP/IP stacks inside L2TP implementations as a
design constraint on L2TP or as a justification for doing something wrong.


> > It is well known among that miss-ordering TCP/IP packets harms RFC
> > compliant TCP/IP implementations  ...

> I don't believe L2TP should mis-order packets, but it should not have to
> go out of its way to preserve the order of already mis-ordered IP
> frames.

Again, that makes no sense to me.  L2TP obviously cannot do anything
about already mis-ordered IP frames.  I thought the sequence number
inside L2TP is to ensure the L2TP does not increase mis-ordering
as the result of mis-ordering from layers below L2TP, and also to
keep L2TP from breaking things that do not tolerate any mis-ordering
including PPP bridging and PPP CP packets.

> ...
> I don't think you understood. I was referring to fragmentation of an IP
> packet transmitted from the client. The circumstance causes the two IP
> fragments, each framed as individual PPP packets, to be repeatedly
> misordered exiting the LAC (due to fragmentation of one of the two
> already fragmented packets at the LAC). It was not actually L2TP that
> was misordering the packets, it was the IP stack on the LAC. It did so
> because of fragmentation of one of the two initial IP fragments.
>
> Perhaps more detail would help: The smaller of the two IP fragments
> sequence checking of IP assumes that reordering is a random phenomena,
> when in fact it is not. IP sequence checking in the face of a consistent
> packet misordering situation results in an unrecoverable denial of
> service (unless a route change or some such actually restores the order
> by chance). 

So you saw a router that consistently moves smaller packets ahead of larger
packets?  I think that's compliant, but generally a bad idea, and certainly
not justification for turning off L2TP sequence numbers.  Note that I've
shipped PPP code that optionally ordered its transmit queue to send most
*but not all* tinggrams ahead of big packets as an ad hoc type-of-service
differentiation.  I no longer recommend that hack.

You are justifying a bad idea (turning off L2TP sequence numbers) based
on the bad behavior a broken box (one that consistently mis-orders IP
packets).

Put asside short term sales considerations and think about the long term.
Which do you think would be the better course, fixing the broken box to
not mis-order IP packets or turning off L2TP sequence numbers?  The broken
box will cause significantly reduced TCP throughput and increased latency
regardless of whether it uses L2TP sequence numbers.  It will also stress
the IP fragmentation reassembly mechanisms in receivers, and that will
cause increased IP packet losses.  Eventually even people who are so inured
to garbage that they tolerate daily blue screens of death will notice the
difference between using this broken box and others.  Since they won't be
constrained by the availability of applications like MS Office and IE,
and since the primary criteria on which routers are evaluated is speed
and reliability, they'll switch.


> ...
> > The standard problem of standards committee protocols is bloat with
> > optional features and switches switches that no one will ever use because
> > they cannot be used in real life.  As far as I can see, any switch that
>
> There is at least some notion of removal of features that never get used
> as the document advances the standards track.

That sentiment is always offered to justify bloating a protocol with
optional bad ideas.  Because in real life standards committees specifically
including PPP working groups, optional bad ideas that get as far Proposed
Standard are *NEVER* removed makes the sentiment at best uninformed.
Consider the many bad optional ideas in and among PPP protocols, and notice
how many have been removed.  Some have been proposed for deletion, but
somehow the removal efforts always stall.


> What you say about committee-born protocols is true in many regards, and
> L2TP is niether immune nor alone in this arena. The upside in our case
> is a fairly large number of interoperable implementations, and general
> industry acceptance. 

That bit about "general industry acceptance" is usually said of protocols
for which it is in fact false.  I hope it is true in this case, but I
can't tell.  It's probably too early to say.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 17:17:41 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA08803
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 17:17:40 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id MAA00678; Tue, 25 Jan 2000 12:52:24 -0800 (PST)
Received: from crufty.research.bell-labs.com (crufty.research.bell-labs.com [204.178.16.49]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id MAA00668 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 12:51:48 -0800 (PST)
Received: from zubin.dnrc.bell-labs.com ([135.180.130.56]) by crufty; Tue Jan 25 12:31:52 EST 2000
Received: from dnrc.bell-labs.com (ex-vpn68.pa.bell-labs.com [135.250.1.68])
	by zubin.dnrc.bell-labs.com (8.9.3/8.9.3) with ESMTP id MAA27693;
	Tue, 25 Jan 2000 12:31:38 -0500 (EST)
Message-ID: <388DDE6C.5BD7C09F@dnrc.bell-labs.com>
Date: Tue, 25 Jan 2000 09:33:32 -0800
From: Grenville Armitage <gja@dnrc.bell-labs.com>
Organization: Bell Laboratories, Lucent Technologies
X-Mailer: Mozilla 4.61 [en] (WinNT; U)
X-Accept-Language: en
MIME-Version: 1.0
To: "W. Mark Townsley" <townsley@cisco.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001212155.OAA25617@calcite.rhyolite.com> <388B91EB.6546F80A@cisco.com> <388C8F46.D57F4024@dnrc.bell-labs.com> <388DD414.C4D401DF@cisco.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



"W. Mark Townsley" wrote:
> 
> I won't claim to be the foremost IP stack expert for a moment, but
> common sense to me says that if I receive an IP packet that must be
> fragmented followed by a bunch of smaller packets that do not, it makes
> sense to go and and switch the smaller packets rather than queuing them
> while I chew on the fragmentation.

Then our definition of 'common sense' differs ;) Given a goal of
optimizing end to end performance, the above is not common sense
(although on the surface it is an appealing argument from the
perspective of the node facing the sequence of packets you describe).

If the packets do not belong to the same flow (or aren't even
targetted at the same IP destination) then process in whatever
order you like - but that's not "reordering" in the sense being
discussed here.

	[..]
> Even if
> the packets are associated with the same TCP stream (which on a middle
> node they are likely not) latency can be as much of a burden to
> performance as packet loss (ask anyone in tcpsat about that).

Latency happens. But reordering fragments of packet N
in order to expedite packets N+1, N+2, .. of the same TCP stream
is false economy. You'll just be screwing with TCP's mind, and
in the end your goodput will be lower than if you'd simply let
nature take its course and kept the stream's packets in order.

> I am
> curious though, which RFC states that IP packet reordering may only
> occur during route transitions?

My words were:
	[..]
> > Reordering
> > 'may' occur as a rare consequence of routing transients.

Noone said there's a rule mandating it only occur under those
circumstances. That's just the unpredictable circumstance IP apps
need to be designed to tolerate. But IP stacks that reorder
in some misguided attempt to reduce latency constitute an avoidable
circumstance.

cheers,
gja


From owner-l2tp@ipsec.org  Tue Jan 25 17:52:47 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA09067
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 17:52:46 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id NAA01025; Tue, 25 Jan 2000 13:36:41 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id NAA01021 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 13:35:54 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id NAA22517;
	Tue, 25 Jan 2000 13:38:33 -0800 (PST)
Received: from cisco.com (chacha-async-10.cisco.com [171.69.123.110]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id QAA15570; Tue, 25 Jan 2000 16:35:19 -0500 (EST)
Message-ID: <388E1716.B11445B7@cisco.com>
Date: Tue, 25 Jan 2000 16:35:18 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Vernon Schryver <vjs@calcite.rhyolite.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001251814.LAA23121@calcite.rhyolite.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

Vernon Schryver wrote:
> 
> > From: "W. Mark Townsley" <townsley@cisco.com>
> 
> > So, a router interface should queue all incoming IP packets while it
> > takes a large one and splits it into one or more IP fragments?
> 
> Not exactly.  A router must be largely FIFO for a given UDP or TCP
> stream.  An occassional reordering is tolerable and unavoidable,
> but consistent re-ordering is broken, cannot be tolerated, and will
> eventually be rejected by the market.
> 
> Why have some of Cisco routers for many years tried to keep a given TCP
> stream associated with a given output port when load balancing over equal
> paths?  Why is the standard Cisco response to observations about apparent
> inconsistenties or flapping in paths reported by traceroute or ping -R
> (LSRR) that ICMP packets get special handling but that TCP/IP packets
> generally see a constant path?  Why go to that significant trouble and
> expense to associate TCP streams with output ports (or routes) except to
> avoid IP re-ordering, and why both about IP re-ordering?  My guess is that
> some people at Cisco know of the Bad Things that happen with significant
> packet reo-ordering.

It has more to do with what types of packets have optimized switching
paths, and which do not. Further, whether the packet is locally destined
or being switched through the router, and even which interface it is
being switched to. For example, since traceroute is a series of pings to
each node, the packet ends up in a different switching path when the
node is first discovered vs. being switched through to the next node.

Vernon, I'm not disputing that preserving order for a TCP connection is
not a Good Thing. Let me ask, at what point do you suggest we stop
peeking into a packet to determine the flow? I doubt any router looks
inside UDP, then L2TP, then PPP, then IP, then TCP to try and keep the
flow ordered (or even the UDP packet to preserve L2TP flow alone). At
some point, the cost of all of this lookup becomes more painful than the
gain. I would be happy to see someone take as an action item researching
this and providing the results. Even if it is proven that we should
perform all this flow checking, it still cannot be guaranteed (as it
certainly could be configured off).

> 
> Again, if you think that significant re-ordering TCP/IP packets by PPP
> routers is tolerable, why doesn't Cisco support "load balancing" or "BF&I"
> PPP multilink better?  BF&I requires less processing by the sending
> router and an enormous amount less by the receiver than RFC 1990.  BF&I
> also has better throughput and latency.
> 
> > ...
> > The bottom line is this: L2TP data packets carrying PPP/IP is ultimately
> > an IP packet within an IP packet. Applying sequencing checking to a
> > tunneled IP packet would surely make a the original IP architects
> > shudder just as much as reordering PPP frames does for a PPP guru. Each
> > are fundamental, opposing, assumptions for each protocol, respectively.
> 
> Maybe you're right, but I bet not.  Shall we ask some?

Go ahead. I would think that they would consider trying to sustain order
as noble when possible, but checking and dropping packets that were out
of order on receipt as a problem. The latter is what I am concerned
about in L2TP.

> 
> By the way, do you remember when a lot of IP packets of what is now called
> the Internet were carried on x.25 circuits?  My vague recollection is also
> that x.25 has a sequence number, albeit rather small.

Checking for out of sequence packets and dropping them is different than
providing a reliable transport. With the latter you may have some
windowing problems between the reliable transports' operation on top of
one another. The former means that there may be packets which never get
delivered.

A reproducible packet reordering circumstance + IP sequence checking =
packets that never will be received. I don't like the result of this
equation.

Take away either of the items in the sum above, and all packets will be
received. I cannot guarantee that all IP stacks will only have rare,
random, reordering. Or, that they will peek into my UDP packet to
determine L2TP flow.

> 
> > Together, we have identified cases where checking packet sequence and
> > not checking packet sequence can break the tunneled connection. We could
> > compare pathological examples all day. In practice, I have seen only one
> > of these occur in a repeatable fashion. If you want to avoid this, my
> > recommendation is to not check sequencing on tunneled IP packets.
> 
> I don't understand.  Are you recommending that most L2TP boxes turn off
> sequence numbers, only that broken box that can't IP fragment sanely,
> or only that broken box and only some of the time?  If the last, when?
> 
> Aren't you still patching one pathology, a router that consistently
> mis-orders some packets, with another, worse pathology, turning
> off L2TP sequencing?

I don't think that L2TP operating without sequencing in some
circumstances is a pathology. Also, I'm not 100% convinced that a router
consistently reordering under certain situations is a pathology either.

> 
> You have not explicitly addressed my implicit question, but I have
> inferred what might not be your answer.  Let me make it explicit.
> Do you recommend under any circumstances turning off L2TP sequencing for
> non-IP packets such as bridging, IPX and PPP CP packets on that wonderful
> box that consistently re-orders when IP fragmenting?

At risk of you trashing the text in RFC2661, I will give you my answer
(from section 5.4):

   The LNS may initiate disabling of sequencing at any time during the
   session (including the first data message sent). It is recommended
   that for connections where reordering or packet loss may occur,
   sequence numbers always be enabled during the initial negotiation
   stages of PPP and disabled only when and if the risk is considered
   acceptable. For example, if the PPP session being tunneled is not
   utilizing any stateful compression or encryption protocols and is
   only carrying IP (as determined by the PPP NCPs that are
   established), then the LNS might decide to disable sequencing as IP
   is tolerant to datagram loss and reordering.

Thus, if you are running IP only (and are running over an IP network as
opposed to L2TP over ATM or some such), then I support disabling
sequence checking after the CPs have finished. For other protocols,
particularly stateful compression, encryption and PPP CPs, I recommend
enabling sequence numbers. That said, most implementations I see today
simply have sequencing off from the start. 

> 
> Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 17:54:59 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id RAA09116
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 17:54:58 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id NAA01080; Tue, 25 Jan 2000 13:55:52 -0800 (PST)
Received: from sj-mailhub-3.cisco.com (sj-mailhub-3.cisco.com [171.68.224.215]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id NAA01075 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 13:55:23 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-3.cisco.com (8.9.1a/8.9.1) with ESMTP id OAA07478;
	Tue, 25 Jan 2000 14:13:25 -0800 (PST)
Received: from cisco.com (chacha-async-10.cisco.com [171.69.123.110]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id QAA15587; Tue, 25 Jan 2000 16:54:43 -0500 (EST)
Message-ID: <388E1BA2.457080EC@cisco.com>
Date: Tue, 25 Jan 2000 16:54:42 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Grenville Armitage <gja@dnrc.bell-labs.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001212155.OAA25617@calcite.rhyolite.com> <388B91EB.6546F80A@cisco.com> <388C8F46.D57F4024@dnrc.bell-labs.com> <388DD414.C4D401DF@cisco.com> <388DDE6C.5BD7C09F@dnrc.bell-labs.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



Grenville Armitage wrote:
> 
> "W. Mark Townsley" wrote:
> >
> > I won't claim to be the foremost IP stack expert for a moment, but
> > common sense to me says that if I receive an IP packet that must be
> > fragmented followed by a bunch of smaller packets that do not, it makes
> > sense to go and and switch the smaller packets rather than queuing them
> > while I chew on the fragmentation.
> 
> Then our definition of 'common sense' differs ;) Given a goal of
> optimizing end to end performance, the above is not common sense

There are all sorts of things that can be optimized. Host to host TCP
performance is not the only thing, and overanalyzing each packet
midstream can be very costly.

> (although on the surface it is an appealing argument from the
> perspective of the node facing the sequence of packets you describe).
> 
> If the packets do not belong to the same flow (or aren't even
> targetted at the same IP destination) then process in whatever
> order you like - but that's not "reordering" in the sense being
> discussed here.

Of course it is. When doing any sort of traffic shaping, there is
question of queuing, flow detection, etc. There are always tradeoffs.
The question here is whether it is worthwhile to peek into the UDP
packet to ensure in order delivery of L2TP, or even into the L2TP packet
to ensure inorder delievery of a TCP session being tunneled. The
"smarter" you are about the type of traffic you are switching, the
slower you are at switching it.

> 
>         [..]
> > Even if
> > the packets are associated with the same TCP stream (which on a middle
> > node they are likely not) latency can be as much of a burden to
> > performance as packet loss (ask anyone in tcpsat about that).
> 
> Latency happens. But reordering fragments of packet N
> in order to expedite packets N+1, N+2, .. of the same TCP stream

Sure, if part of the same TCP stream. But, you do not have that
information. So, the question becomes whether or not you expedite UDP
packets N+1, N+2, etc., not TCP packets of the same stream. These could
be intermixed L2TP, DNS, TFTP, voice, etc. packets, as well as multiple
streams within L2TP.

> is false economy. You'll just be screwing with TCP's mind, and
> in the end your goodput will be lower than if you'd simply let
> nature take its course and kept the stream's packets in order.
> 
> > I am
> > curious though, which RFC states that IP packet reordering may only
> > occur during route transitions?
> 
> My words were:
>         [..]
> > > Reordering
> > > 'may' occur as a rare consequence of routing transients.
> 
> Noone said there's a rule mandating it only occur under those
> circumstances. That's just the unpredictable circumstance IP apps
> need to be designed to tolerate. But IP stacks that reorder
> in some misguided attempt to reduce latency constitute an avoidable
> circumstance.

It's not just latency. It's overall throughput, CPU usage, queue and
memory limititations, etc. 

> 
> cheers,
> gja


From owner-l2tp@ipsec.org  Tue Jan 25 18:13:29 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA09553
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 18:13:28 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id OAA01111; Tue, 25 Jan 2000 14:03:52 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id OAA01107 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 14:03:27 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id PAA27652
	for l2tp@ipsec.org  env-from <vjs>;
	Tue, 25 Jan 2000 15:03:26 -0700 (MST)
Date: Tue, 25 Jan 2000 15:03:26 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001252203.PAA27652@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: Grenville Armitage <gja@dnrc.bell-labs.com>
> To: "W. Mark Townsley" <townsley@cisco.com>

> > I won't claim to be the foremost IP stack expert for a moment, but
> > common sense to me says that if I receive an IP packet that must be
> > fragmented followed by a bunch of smaller packets that do not, it makes
> > sense to go and and switch the smaller packets rather than queuing them
> > while I chew on the fragmentation.
>
> Then our definition of 'common sense' differs ;) Given a goal of
> optimizing end to end performance, the above is not common sense
> (although on the surface it is an appealing argument from the
> perspective of the node facing the sequence of packets you describe).
>
> If the packets do not belong to the same flow (or aren't even
> targetted at the same IP destination) then process in whatever
> order you like - but that's not "reordering" in the sense being
> discussed here.

Those are good responses, but a better is that any system that needs
significant time to IP fragment is broken.  IP fragmentation is a triviaL
operation, especially at the bit rates we are talking about.  IP
fragmentation consists of adding 20 bytes plus a lower layer header to
the middle of a packet, and doing thing else.  Those 20 bytes of new IP
header are identical to the previous IP header except for two or three
trivial adjustments that are independent of everything except the number
of intervening payload bytes.


> 	[..]
> > Even if
> > the packets are associated with the same TCP stream (which on a middle
> > node they are likely not) latency can be as much of a burden to
> > performance as packet loss (ask anyone in tcpsat about that).
>
> Latency happens. But reordering fragments of packet N
> in order to expedite packets N+1, N+2, .. of the same TCP stream
> is false economy. You'll just be screwing with TCP's mind, and
> in the end your goodput will be lower than if you'd simply let
> nature take its course and kept the stream's packets in order.
> ...

In other words, contrary to the claim, re-ordering a TCP stream as
described only hurts and never helps TCP latency.

Another way to look at that the issue is that if an L2TP box is
receiving 1500+ byte PPP frames and sending them over a UDP/IP link
with a smaller MTU cannot handle the inevitably implied heavy IP
fragmentation, then shuffling its output queue is not a solution.
If the router doesn't have the needed horsepower, rearranging its
output queue of deck chairs as it sinks isn't a good plan.


It is bogus to reason that since all IP protocols are required to
tolerate a little re-ordering, therefore turning off the L2TP
re-ordering protection is an ok idea.  Even if it were true that
IP protocols don't care about re-ordering (that is false), there
are other L2TP payloads that absolutely cannot be re-ordered.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Tue Jan 25 18:14:41 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA09584
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 18:14:40 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id NAA01090; Tue, 25 Jan 2000 13:58:12 -0800 (PST)
Received: from databus.databus.com (databus.databus.com [198.186.154.34]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id NAA01086 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 13:58:00 -0800 (PST)
From: Barney Wolff <barney@databus.com>
To: l2tp@ipsec.org
Date: Tue, 25 Jan 2000 16:45 EST
Subject: Re: Nr in data packets
Content-Type: text/plain
Message-ID: <388e1c620.5cc@databus.databus.com>
Sender: owner-l2tp@sailpix.com
Precedence: bulk

There is an experiment that I wish someone with the necessary equipment
would try, that I think would shed some useful light on this issue.

SRC---FDDI---Router---Ethernet---DEST

Send alternating big (>1500) and small UDP packets from SRC to DEST,
and see if DEST indeed receives things out of sequence.

My reason for asking is that this is quite different from a scenario
where it is the L2TP stack's own network interface that is doing the
re-ordering.  In that case, I have to side with Vernon and say that
emitting things out of order is a bug.  But if an intervening router
can consistently re-order, not under the control of either LAC or
LNS, then the problem is very serious.  Perhaps in that case the
right thing to do is mandate DF and PMTU discovery, and force the
fragmentation to occur before the packet reaches anything that
will re-order.

Barney Wolff  <barney@databus.com>


From owner-l2tp@ipsec.org  Tue Jan 25 19:03:50 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA10436
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 19:03:46 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id OAA01280; Tue, 25 Jan 2000 14:48:42 -0800 (PST)
Received: from crufty.research.bell-labs.com (crufty.research.bell-labs.com [204.178.16.49]) by www.sailpix.com (8.8.8/8.6.9) with SMTP id OAA01276 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 14:48:03 -0800 (PST)
Received: from zubin.dnrc.bell-labs.com ([135.180.130.56]) by crufty; Tue Jan 25 17:46:11 EST 2000
Received: from dnrc.bell-labs.com (dhcp-6-168.pa.bell-labs.com [135.250.6.168])
	by zubin.dnrc.bell-labs.com (8.9.3/8.9.3) with ESMTP id RAA05234;
	Tue, 25 Jan 2000 17:46:06 -0500 (EST)
Message-ID: <388E284B.E9CC42CC@dnrc.bell-labs.com>
Date: Tue, 25 Jan 2000 14:48:43 -0800
From: Grenville Armitage <gja@dnrc.bell-labs.com>
Organization: Bell Laboratories, Lucent Technologies
X-Mailer: Mozilla 4.7 [en] (WinNT; U)
X-Accept-Language: en
MIME-Version: 1.0
To: "W. Mark Townsley" <townsley@cisco.com>
CC: l2tp@ipsec.org
Subject: Re: Nr in data packets
References: <200001212155.OAA25617@calcite.rhyolite.com> <388B91EB.6546F80A@cisco.com> <388C8F46.D57F4024@dnrc.bell-labs.com> <388DD414.C4D401DF@cisco.com> <388DDE6C.5BD7C09F@dnrc.bell-labs.com> <388E1BA2.457080EC@cisco.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit



"W. Mark Townsley" wrote:
	[..]
> There are all sorts of things that can be optimized. Host to host TCP
> performance is not the only thing, and overanalyzing each packet
> midstream can be very costly.

My apologies, I thought global optimization was always a goal
of local design considerations. If someone wants to build
a box that tosses packets through really really fast, and
casually reorders in the process, good luck to them. They'll
ace the pps test, and screw their customer's network performance.
I guess its a free marketplace.

	[..]
> > If the packets do not belong to the same flow (or aren't even
> > targetted at the same IP destination) then process in whatever
> > order you like - but that's not "reordering" in the sense being
> > discussed here.
> 
> Of course it is.

Again we differ. Inter-flow packet reordering isn't
relevant to the e2e performance of any given transport
flow. Intra-flow reordering *is*. Which is what I said
above. (Which is arguably what Vern has been pointing out
too)

> When doing any sort of traffic shaping, there is
> question of queuing, flow detection, etc.

The distinction between inter- and intra-flow reordering 
(microflows, in diffserv terminology) still stands. Inter-flow
reordering is vastly less important than intra-flow reordering.
Thus your scenario of a router facing a sequence of small
packets behind a to-be-fragmented large packet really depends
on whether the packets belong to the same flow/stream or not.
If so, do not reorder. If not, reorder as desired. 

	[..]
> >         [..]
> > > Even if
> > > the packets are associated with the same TCP stream (which on a middle
> > > node they are likely not) latency can be as much of a burden to
> > > performance as packet loss (ask anyone in tcpsat about that).
> >
> > Latency happens. But reordering fragments of packet N
> > in order to expedite packets N+1, N+2, .. of the same TCP stream
> 
> Sure, if part of the same TCP stream.

Well, that's the case you mentioned (I even quoted it above
to establish context for my response).

> But, you do not have that
> information.

So you should assume it probably is, or at least probably
might be. Which means gratuitous reordering within
an IP stack is counter-productive (when viewing the system
as a whole, including the actions required of the endstations
to deal with the reordering).

> So, the question becomes whether or not you expedite UDP
> packets N+1, N+2, etc., not TCP packets of the same stream. These could
> be intermixed L2TP, DNS, TFTP, voice, etc. packets, as well as multiple
> streams within L2TP.

If packet N is UDP, and packet N+1 is TCP then these represent
distinct flows (or flow aggregates, if for some reason meaningful
port numbers aren't available or aren't being used for internal
classification purposes within the IP stack in question). You
can reorder the fragments of N relative to N+1 to your hearts
content since the UDP and TCP streams will still arrive inorder
at their respective destinations. 

But I think you knew that. 

Like I said when I joined in, the statement that IP permits
reordering isn't a carte blanche invitation to IP stack
implementors to introduce internal reordering. To be more
precise, intra-flow reordering is the main concern. This
has been acknowledged in the diffserv WG, and MPLS WG.

	[..]
> > Noone said there's a rule mandating it only occur under those
> > circumstances. That's just the unpredictable circumstance IP apps
> > need to be designed to tolerate. But IP stacks that reorder
> > in some misguided attempt to reduce latency constitute an avoidable
> > circumstance.
> 
> It's not just latency. It's overall throughput, CPU usage, queue and
> memory limititations, etc.

Re-sequencing of reordered packets at the endstation consumes cpu,
queues, and memory. Performing the re-sequencing in general adds
latency. The endstation is the customer - 'the network' needs
to deal with that fact. 

cheers,
gja


From owner-l2tp@ipsec.org  Tue Jan 25 19:44:10 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA11462
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 19:44:10 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id PAA01414; Tue, 25 Jan 2000 15:26:44 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA01407 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 15:26:13 -0800 (PST)
Received: from mercurynt40 (192.168.211.231) by mcfeeley.indusriver.com (Worldmail 1.3.167); 25 Jan 2000 18:24:32 -0500
Message-Id: <4.2.0.58.20000125181455.00a637a0@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Tue, 25 Jan 2000 18:32:46 -0500
To: Vernon Schryver <vjs@calcite.rhyolite.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: Nr in data packets
Cc: l2tp@ipsec.org
In-Reply-To: <200001242202.PAA24576@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

1) It seems need more illustration to my point
   For a L2TP packet such as PAYLOAD/TCP/IP1/PPP/L2TP/UDP/IP2 on the 
protocol stack:
The IP1, IP2, and UDP services does not guarantee packet delivery in sequence
yet, TCP and PPP with flow control does.
if L2TP service guarantee packet in sequence delivery then it does not
help the PAYLOAD or TCP due to that the IP1 service still does not 
guarantee packet
delivery in sequence.
2) The argument about LCP,NCP needs underlying protocol provide packet 
delivery in
sequence is not so clear:
When sender send out Config-Req, it will not send any packet (to move the 
statemachine)
untill the receiver responds with Config-XXX....
Can you illustrate how the PPP state machine send out more than one different
protocol frame in one state... thereby need the packet sequence delivery 
service
3) For TCP's argument about needs packet sequence delivery service from IP 
is impossible
and its design drawback to handle this issue, please refer to the
page 536, 3rd paragraph about "what to do with out of order segments" of 
book "Computer Networks" by Andrew S. Tanenbaum, ISBN 0-13-349945-6.

--- David



At 03:02 PM 1/24/00 -0700, you wrote:
> > From: David Chen <dchen@indusriver.com>
>
> > IP service does not guarantee packet delivery in sequence.
> > Clearly, UDP/IP does not neither and
> > TCP treat out of sequence packet as if loss of packet. -
> > as appose to buffering and re-ordering (packets).
> > (therefore, TCP provides in sequence no-loss packet delivery)
>
>I think I did not say that "TCP [treats an] out of sequence packet as if
>[it were] loss of packet," and I think it is wrong.  Any reasonable TCP
>implemention will save an out-of-order segment that is in the receive
>window, and generate an ACK for the current edge of the window.  I did
>say that out of order packets cause duplicate ACKs which cause
>retransmissions and reduction of the congestion window.  The difference
>between "treat out of order as loss" and what I said is technical, but I
>think the difference is significant.
>
>
> > As for L2tp, a packet such as PAYLOAD/TCP/IP/PPP/L2TP/UDP/IP on the
> > protocol stack,
> > is clearly that L2TP does not need keeping the packet sequence and 
> dropping
> > the packet for
> > the same reason (keeping packet sequence).
>
>Maybe that's clearly correct to you.  If I understand what you mean, the
>opposite is clearly correct to me.  As I said before, PPP does not and
>cannot drop control protocol packets when they are out of order.  Instead,
>the PPP state machines must and do act on the miss-ordered stimuli.  TCP
>also generally does not drop segments when they are out of order.  Also
>as I keep pointing out, there are more PPP payloads than TCP/IP, and some
>of them, such as bridging, do not tolerate reordering at all.
>
>
>Have you read RFC 1661 and understood the state machine?  Do you know what
>I mean by "congestion window" and "duplicate ACKs?"  Do you know what I
>mean when I repeatedly mention bridging?  Perhaps you do, but I have the
>nagging feeling that you are arguing based on how you think PPP and TCP
>should or work or be specified instead of how they are specified and work.
>
>
>Vernon Schryver    vjs@rhyolite.com

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Tue Jan 25 20:21:14 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id UAA12219
	for <l2tpext-archive@odin.ietf.org>; Tue, 25 Jan 2000 20:21:12 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id PAA01383; Tue, 25 Jan 2000 15:20:07 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id PAA01373 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 15:19:37 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id QAA29586
	for l2tp@ipsec.org  env-from <vjs>;
	Tue, 25 Jan 2000 16:19:36 -0700 (MST)
Date: Tue, 25 Jan 2000 16:19:36 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001252319.QAA29586@calcite.rhyolite.com>
To: l2tp@ipsec.org
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: "W. Mark Townsley" <townsley@cisco.com>

> ...
> being switched to. For example, since traceroute is a series of pings to

Just as an asside, traceroute is not a series of pings if that means ICMP
Echo-Requests.  Classic traceroute uses UDP packets, each to a different
UDP port number, and so apparently parts of different streams.  Microsoft's
`tracert` uses ICMP Echo-Requests.  I've heard but not verified that
FreeBSD's 4.? traceroute can optionally use TCP, GRE, and ICMP.


> ...
> Vernon, I'm not disputing that preserving order for a TCP connection is
> not a Good Thing. Let me ask, at what point do you suggest we stop
> peeking into a packet to determine the flow? I doubt any router looks
> inside UDP, then L2TP, then PPP, then IP, then TCP to try and keep the
> flow ordered (or even the UDP packet to preserve L2TP flow alone). At
> some point, the cost of all of this lookup becomes more painful than the
> gain. I would be happy to see someone take as an action item researching
> this and providing the results. Even if it is proven that we should
> perform all this flow checking, it still cannot be guaranteed (as it
> certainly could be configured off).

You keep implying I support obviously stupid things.  Just as I never
supported the silly notion of fixing the order of already miss-ordered
TCP/IP segments, I've said nothing about arbitrary computations on
payloads.  My point is not that IP reordering must never happen; that
would be stupid.  It is not even that an L2TP router must never re-order
IP/PPP packets; rare reordering is ok (modulo CCP, BP, ECP, etc), and even
frequent re-ordering across streams.  My point is that turning off L2TP
sequence numbers in a router with badly broken IP fragmentation machinery
is a not a work-around.  I don't care why or how the fragmentation
machinery is broken.  It doesn't matter if some packet parsing assumptions
that worked elsewhere don't work for L2TP.  The router is badly broken,
and there is no choice but fixing it.  If the fix has has costs (what
doesn't?), tough luck.

Worse and genuinely shocking, you seem to support L2TP implementations
that have sequence numbers turned off all of the time, even on
packets that "MUST NOT" be reordered.


> ...
> Thus, if you are running IP only (and are running over an IP network as
> opposed to L2TP over ATM or some such), then I support disabling
> sequence checking after the CPs have finished. For other protocols,
> particularly stateful compression, encryption and PPP CPs, I recommend
> enabling sequence numbers.

I think that's reasonable.

Separately, any router that does a lot of re-ordering of IP packets within
a stream is badly broken and will eventually be trashed in the market.
Turning off L2TP sequence numbers does not fix or work around that serious
bug.


>                            That said, most implementations I see today
> simply have sequencing off from the start. 

Do you think that is reasonable?  I not only think it's unreasonable,
but I hope never to have to work with any of the people who wrote that
code, unless they are even now busy fixing it.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Wed Jan 26 00:39:40 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id AAA16973
	for <l2tpext-archive@odin.ietf.org>; Wed, 26 Jan 2000 00:39:39 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id UAA01966; Tue, 25 Jan 2000 20:29:50 -0800 (PST)
Received: from sj-mailhub-2.cisco.com (sj-mailhub-2.cisco.com [171.69.43.88]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id UAA01962 for <l2tp@ipsec.org>; Tue, 25 Jan 2000 20:29:34 -0800 (PST)
Received: from townsley-ultra.cisco.com (townsley-ultra.cisco.com [161.44.49.31])
	by sj-mailhub-2.cisco.com (8.9.1a/8.9.1) with ESMTP id UAA17080;
	Tue, 25 Jan 2000 20:32:16 -0800 (PST)
Received: from cisco.com (chacha-async-39.cisco.com [171.69.123.139]) by townsley-ultra.cisco.com (8.8.4-Cisco.1/CISCO.WS.1.2) with ESMTP id XAA15657; Tue, 25 Jan 2000 23:29:03 -0500 (EST)
Message-ID: <388E780C.1BBB761F@cisco.com>
Date: Tue, 25 Jan 2000 23:29:00 -0500
From: "W. Mark Townsley" <townsley@cisco.com>
X-Mailer: Mozilla 4.7 [en] (Win95; U)
X-Accept-Language: en,pdf
MIME-Version: 1.0
To: Vernon Schryver <vjs@calcite.rhyolite.com>, gja@dnrc.bell-labs.com,
        barney@databus.com
CC: l2tp@ipsec.org
Subject: On packet reordering... (was Nr in data packets)
References: <200001252319.QAA29586@calcite.rhyolite.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


Trying to summarize.

Let's get away from whether an IP stack is broken or not when it decides
to reorder packets. I don't have the time to discuss it further here.
Even if we came up with a BCP for IP stack ordering in the face of
fragmentation, etc. it is not something we are going to include in the
L2TP documentation. I will put my trust in those whose livelihoods
depend on this, and who have probably been through the same arguments a
thousand times. 

Barney, I am in full support of PMTU discovery for L2TP, and there are
those that are working on this already. Be on the lookout for a
proposal. That said, we still have to operate in the wake of PMTU not. 

Thus, I still say, for IP traffic tunneled by L2TP:

"Be conservative in what you send..." e.g. _try_ not to reorder IP
packets
"and liberal in what you accept" e.g. pass the out of order packets to
IP on receipt and let it sort out what to do. 

For non-IP packets, and those with ECP/CCP/VJ etc., I suggest checking
sequence as RFC2661 already states. I *think* I even heard Vernon state
that this is reasonable.

Those are the relevant points for L2TP as I see them. If I missed any,
please let me know. 

- Mark


From owner-l2tp@ipsec.org  Thu Jan 27 18:03:58 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA16557
	for <l2tpext-archive@odin.ietf.org>; Thu, 27 Jan 2000 18:03:57 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id MAA00213; Thu, 27 Jan 2000 12:27:32 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id MAA00209 for <l2tp@ipsec.org>; Thu, 27 Jan 2000 12:26:57 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id NAA20497
	for l2tp@ipsec.org  env-from <vjs>;
	Thu, 27 Jan 2000 13:26:54 -0700 (MST)
Date: Thu, 27 Jan 2000 13:26:54 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001272026.NAA20497@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: Nr in data packets
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> 1) It seems need more illustration to my point
>    For a L2TP packet such as PAYLOAD/TCP/IP1/PPP/L2TP/UDP/IP2 on the 
> protocol stack:
> The IP1, IP2, and UDP services does not guarantee packet delivery in sequence
> yet, TCP and PPP with flow control does.
> if L2TP service guarantee packet in sequence delivery then it does not
> help the PAYLOAD or TCP due to that the IP1 service still does not 
> guarantee packet delivery in sequence.

What is "PPP with flow control" RFC 1663?  If so, please consider the
implications of the title of RFC 1663.

The issue is not whether rare re-ordering by L2TP significantly harms TCP/IP.

I think the following are facts:
  - delivering out-of-order PPP packets breaks much of PPP.
  - significant miss-ordering significantly harms TCP. 
  - Rare miss-ordering is not significant to TCP.
I do not see any inconsistency between those facts and your observation above.

The case at hand is an L2TP router that does a lot of re-ordering and the
claim is that turning off L2TP sequence numbers makes the system useful.
That is false, because the broken router would continue to miss-order many
TCP/IP1 segments, and that would be a Bad Thing(tm) for TCP.

Or maybe the case at hand is your statement many messages ago that seemed
to me to be "L2TP Sequence Number Checking Considered Unnecessary."
I think the problems of PPP and especially CCP and ECP with re-ordering
bear on that statement.


> 2) The argument about LCP,NCP needs underlying protocol provide packet 
> delivery in
> sequence is not so clear:
> When sender send out Config-Req, it will not send any packet (to move the 
> statemachine)
> untill the receiver responds with Config-XXX....
> Can you illustrate how the PPP state machine send out more than one different
> protocol frame in one state... thereby need the packet sequence delivery 
> service

Consider {Conf-Req1, Conf-Req2, Conf-Ack99} re-ordered into
{Conf-Req2, Conf-Ack99, Conf-Req1}.
Yes, and as I said before, reasonable PPP implementations will respond to
the out of order Conf-Req1 by tearing down the PPP connection and all
upper CPs (but not necessarily lower CPs or modems etc), and starting over
with Conf-Req3.  Maybe you consider that good or at least OK.  I don't,
partly because if Conf-Req1 can be re-ordered, then so can Conf-Req3,
Conf-Req5, and so on forever.
   
Consider what a re-ordering L2TP box does to most CCP and ECP variations.
Re-ordering CCP can lead to user data corruption, because even the best
CCP checksums are weak and intended only to supplement sequence numbers
that detect lost packets.


> 3) For TCP's argument about needs packet sequence delivery service from IP 
> is impossible
> and its design drawback to handle this issue, please refer to the
> page 536, 3rd paragraph about "what to do with out of order segments" of 
> book "Computer Networks" by Andrew S. Tanenbaum, ISBN 0-13-349945-6.

I don't have a clue what is meant by "TCP's argument about needs packet
sequence delivery service from IP is impossible and its design drawback."

Instead of arguing by authority, I'd rather consider the facts directly.

If I were looking for a book about TCP/IP, Prof. Tanenbaums' is not among
those I'd think of, because I've not seen his name in the RFC index, on
respected TCP/IP code, nor in the IETF or IRTF mailing lists devoted to
TCP/IP technical issues.

By the way, is that the book in which Prof. Tanenbaum proclaimed that an
Ethernet cannot carry more than about 30% because he confused ALOHA with
CSMA/CD?  (I hope you know that 30% sounds egregiously silly to people
with valid clues about CSMA/CD.)  I've heard that textbook is good and
not bad as that example suggests.  However, that error does demonstrate
a difference between the expertise obtained by reading about things and
that earned by doing.  I bet many people here know more about current
networking topics than Prof. Tanenbaum, although I bet he knows more about
SNA, x.25, and even ALOHA.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Fri Jan 28 12:21:35 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA20046
	for <l2tpext-archive@odin.ietf.org>; Fri, 28 Jan 2000 12:21:33 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id HAA02460; Fri, 28 Jan 2000 07:44:25 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id HAA02456 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 07:44:00 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167); 28 Jan 2000 10:42:14 -0500
Message-Id: <4.2.0.58.20000128103049.00b2a3e0@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 28 Jan 2000 10:50:33 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: On packet reordering... (was Nr in data packets)
Cc: l2tp@ipsec.org
In-Reply-To: <388E780C.1BBB761F@cisco.com>
References: <200001252319.QAA29586@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk


My argument is that "the Ns bit in L2tp data channel is not needed".
Here is my analysis:
 From LAC to LNS:
For those PPP (out of order) packet sequence 1,2,4,5,6,7,8,3,9,10 arriving 
at LAC if
1) L2tp Ns bit is off
then the PPP in LNS will receive 1,2,4,5,6,7,8,3,9,10  ---> same sequence
2) L2tp Ns bit is on
then the PPP in LNS will receive 1,2,4,5,6,7,8,9,10 ---> packet 3 is droped

The question is "Does any protocol that is encapsulated by the PPP
will not break if a packet is dropped (case 2)
but will break if the packet delivery sequence is out of order (case 1) ?"

Are you suggesting those protocol other than IP that encapsulated by PPP
satisfy this question?

--- David



At 11:29 PM 1/25/00 -0500, you wrote:

>Trying to summarize.
>
>Let's get away from whether an IP stack is broken or not when it decides
>to reorder packets. I don't have the time to discuss it further here.
>Even if we came up with a BCP for IP stack ordering in the face of
>fragmentation, etc. it is not something we are going to include in the
>L2TP documentation. I will put my trust in those whose livelihoods
>depend on this, and who have probably been through the same arguments a
>thousand times.
>
>Barney, I am in full support of PMTU discovery for L2TP, and there are
>those that are working on this already. Be on the lookout for a
>proposal. That said, we still have to operate in the wake of PMTU not.
>
>Thus, I still say, for IP traffic tunneled by L2TP:
>
>"Be conservative in what you send..." e.g. _try_ not to reorder IP
>packets
>"and liberal in what you accept" e.g. pass the out of order packets to
>IP on receipt and let it sort out what to do.
>
>For non-IP packets, and those with ECP/CCP/VJ etc., I suggest checking
>sequence as RFC2661 already states. I *think* I even heard Vernon state
>that this is reasonable.
>
>Those are the relevant points for L2TP as I see them. If I missed any,
>please let me know.
>
>- Mark

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Fri Jan 28 12:26:01 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA20203
	for <l2tpext-archive@odin.ietf.org>; Fri, 28 Jan 2000 12:25:59 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id HAA02476; Fri, 28 Jan 2000 07:54:18 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id HAA02472 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 07:53:54 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167); 28 Jan 2000 10:52:12 -0500
Message-Id: <4.2.0.58.20000128105641.00b27b40@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 28 Jan 2000 11:00:31 -0500
To: "W. Mark Townsley" <townsley@cisco.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: On packet reordering... (was Nr in data packets)
Cc: l2tp@ipsec.org
In-Reply-To: <388E780C.1BBB761F@cisco.com>
References: <200001252319.QAA29586@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk

This is a correction of the following analysis of my last posting,
I apologize the confusion:

My argument is that "the Ns bit in L2tp data channel is not needed".
Here is my analysis:
 From LAC to LNS:
For those PPP packet sequence 1,2,3,4,5,6,7,8,9,10 arriving at LAC,
however the delivery from LAC to LNS is out of sequence, if
1) L2tp Ns bit is off
then the PPP in LNS will receive 1,2,4,5,6,7,8,3,9,10  ---> out of sequence
2) L2tp Ns bit is on
then the PPP in LNS will receive 1,2,4,5,6,7,8,9,10 ---> packet 3 is droped

The question is "Does any protocol that is encapsulated by the PPP
will not break if a packet is dropped (case 2)
but will break if the packet delivery sequence is out of order (case 1) ?"

Are you suggesting those protocol other than IP that encapsulated by PPP
satisfy this question?

--- David

At 11:29 PM 1/25/00 -0500, you wrote:

>Trying to summarize.
>
>Let's get away from whether an IP stack is broken or not when it decides
>to reorder packets. I don't have the time to discuss it further here.
>Even if we came up with a BCP for IP stack ordering in the face of
>fragmentation, etc. it is not something we are going to include in the
>L2TP documentation. I will put my trust in those whose livelihoods
>depend on this, and who have probably been through the same arguments a
>thousand times.
>
>Barney, I am in full support of PMTU discovery for L2TP, and there are
>those that are working on this already. Be on the lookout for a
>proposal. That said, we still have to operate in the wake of PMTU not.
>
>Thus, I still say, for IP traffic tunneled by L2TP:
>
>"Be conservative in what you send..." e.g. _try_ not to reorder IP
>packets
>"and liberal in what you accept" e.g. pass the out of order packets to
>IP on receipt and let it sort out what to do.
>
>For non-IP packets, and those with ECP/CCP/VJ etc., I suggest checking
>sequence as RFC2661 already states. I *think* I even heard Vernon state
>that this is reasonable.
>
>Those are the relevant points for L2TP as I see them. If I missed any,
>please let me know.
>
>- Mark

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Fri Jan 28 14:29:06 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id OAA24632
	for <l2tpext-archive@odin.ietf.org>; Fri, 28 Jan 2000 14:29:03 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id KAA02651; Fri, 28 Jan 2000 10:00:53 -0800 (PST)
Received: from calcite.rhyolite.com (calcite.rhyolite.com [38.159.140.3]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id KAA02647 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 10:00:37 -0800 (PST)
Received: (from vjs@localhost)
	by calcite.rhyolite.com (8.9.3/calcite) id LAA14354
	for l2tp@ipsec.org  env-from <vjs>;
	Fri, 28 Jan 2000 11:00:36 -0700 (MST)
Date: Fri, 28 Jan 2000 11:00:36 -0700 (MST)
From: Vernon Schryver <vjs@calcite.rhyolite.com>
Message-Id: <200001281800.LAA14354@calcite.rhyolite.com>
To: l2tp@ipsec.org
Subject: Re: On packet reordering... (was Nr in data packets)
Sender: owner-l2tp@sailpix.com
Precedence: bulk

> From: David Chen <dchen@indusriver.com>

> ...
> My argument is that "the Ns bit in L2tp data channel is not needed".
> Here is my analysis:
>  From LAC to LNS:
> For those PPP packet sequence 1,2,3,4,5,6,7,8,9,10 arriving at LAC,
> however the delivery from LAC to LNS is out of sequence, if
> 1) L2tp Ns bit is off
> then the PPP in LNS will receive 1,2,4,5,6,7,8,3,9,10  ---> out of sequence
> 2) L2tp Ns bit is on
> then the PPP in LNS will receive 1,2,4,5,6,7,8,9,10 ---> packet 3 is droped
>
> The question is "Does any protocol that is encapsulated by the PPP
> will not break if a packet is dropped (case 2)
> but will break if the packet delivery sequence is out of order (case 1) ?"

Yes, as I've repeatedly pointed out, LCP, CCP, ECP, BCP, and a host
of others will break.


> Are you suggesting those protocol other than IP that encapsulated by PPP
> satisfy this question?

I'm not just suggesting it, I'm saying it,
and yet again for the umpteenth time.


With this, I'll stop repeating myself.


Vernon Schryver    vjs@rhyolite.com


From owner-l2tp@ipsec.org  Fri Jan 28 15:36:32 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA26443
	for <l2tpext-archive@odin.ietf.org>; Fri, 28 Jan 2000 15:36:29 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id KAA02727; Fri, 28 Jan 2000 10:47:51 -0800 (PST)
Received: from zev.net (zev.net [206.111.135.146]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id KAA02722 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 10:47:24 -0800 (PST)
Received: from zev.net (localhost.zev.net [127.0.0.1] (may be forged))
	by zev.net (8.9.3/8.9.3) with ESMTP id KAA21434;
	Fri, 28 Jan 2000 10:47:16 -0800 (PST)
Message-Id: <200001281847.KAA21434@zev.net>
To: David Chen <dchen@indusriver.com>
cc: Vernon Schryver <vjs@calcite.rhyolite.com>, l2tp@ipsec.org
Subject: Re: On packet reordering... (was Nr in data packets) 
In-reply-to: Your message of "Fri, 28 Jan 2000 13:46:01 EST."
             <4.2.0.58.20000128133211.00b2e650@pop3.indusriver.com> 
Date: Fri, 28 Jan 2000 10:47:16 -0800
From: Bill Palter <palter@zev.net>
Sender: owner-l2tp@sailpix.com
Precedence: bulk


> At 11:00 AM 1/28/00 -0700, you wrote:
> > > From: David Chen <dchen@indusriver.com>
> >
> > > ...
> > > My argument is that "the Ns bit in L2tp data channel is not needed".
> > > Here is my analysis:
> > >  From LAC to LNS:
> > > For those PPP packet sequence 1,2,3,4,5,6,7,8,9,10 arriving at LAC,
> > > however the delivery from LAC to LNS is out of sequence, if
> > > 1) L2tp Ns bit is off
> > > then the PPP in LNS will receive 1,2,4,5,6,7,8,3,9,10  ---> out of sequence
> > > 2) L2tp Ns bit is on
> > > then the PPP in LNS will receive 1,2,4,5,6,7,8,9,10 ---> packet 3 is droped
> > >
> > > The question is "Does any protocol that is encapsulated by the PPP
> > > will not break if a packet is dropped (case 2)
> > > but will break if the packet delivery sequence is out of order (case 1) ?"
> >
> >Yes, as I've repeatedly pointed out, LCP, CCP, ECP, BCP, and a host
> >of others will break.
> 
> All these protocol you mentioned here are not the protocols that encapsulated
> by PPP but PPP's control protocol.
> 

  PPP control protocols are encapsulated just like any data packets running 
over PPP. Also I beleive the that in the cases of CCP, ECP and BCP the mention
referes to the data packets as well as the the control protocol.


> Although this is not my question, It is interest to me to
> understand the LCP will break if the packet delivery is out of sequence but 
> will not
> if a packet(s) is dropped in the sequence.
> 
> ?In what state that the LCP state machine (p.12 rfc1661) will break
> (i.e. can not handle the incoming event)
> if the incoming packets (must be more than one) is out of sequence.
> Also, it will not break if the incoming packets has at least one of them is 
> dropped.
> Please note that in some state can send out two packets but
> line 6 page 2. says that "multiple actions may be implemented in any 
> convenient order".
> 
> 
> Let's look at LCP (I assume you mean
> 
> 
> 
> > > Are you suggesting those protocol other than IP that encapsulated by PPP
> > > satisfy this question?
> >
> >I'm not just suggesting it, I'm saying it,
> >and yet again for the umpteenth time.
> >
> >
> >With this, I'll stop repeating myself.
> >
> >
> >Vernon Schryver    vjs@rhyolite.com
> 
> ========================================
> David Chen
> Indus River Networks, Inc.
> 31 Nagog Park
> Acton, MA 01720
> U.S.A.
> dchen@indusriver.com
> (978) 266-8141
> ========================================
--
Bill Palter (Palter@zev.net)


From owner-l2tp@ipsec.org  Fri Jan 28 16:54:48 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id QAA27830
	for <l2tpext-archive@odin.ietf.org>; Fri, 28 Jan 2000 16:54:47 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id KAA02712; Fri, 28 Jan 2000 10:39:47 -0800 (PST)
Received: from mcfeeley.indusriver.com (mcfeeley.indusriver.com [146.115.206.207]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id KAA02708 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 10:39:22 -0800 (PST)
Received: from mercurynt40 (209.6.112.250) by mcfeeley.indusriver.com (Worldmail 1.3.167); 28 Jan 2000 13:37:42 -0500
Message-Id: <4.2.0.58.20000128133211.00b2e650@pop3.indusriver.com>
X-Sender: dchen@pop3.indusriver.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.0.58 
Date: Fri, 28 Jan 2000 13:46:01 -0500
To: Vernon Schryver <vjs@calcite.rhyolite.com>
From: David Chen <dchen@indusriver.com>
Subject: Re: On packet reordering... (was Nr in data packets)
Cc: :@mail2.sailpix.com
In-Reply-To: <200001281800.LAA14354@calcite.rhyolite.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-l2tp@sailpix.com
Precedence: bulk


At 11:00 AM 1/28/00 -0700, you wrote:
> > From: David Chen <dchen@indusriver.com>
>
> > ...
> > My argument is that "the Ns bit in L2tp data channel is not needed".
> > Here is my analysis:
> >  From LAC to LNS:
> > For those PPP packet sequence 1,2,3,4,5,6,7,8,9,10 arriving at LAC,
> > however the delivery from LAC to LNS is out of sequence, if
> > 1) L2tp Ns bit is off
> > then the PPP in LNS will receive 1,2,4,5,6,7,8,3,9,10  ---> out of sequence
> > 2) L2tp Ns bit is on
> > then the PPP in LNS will receive 1,2,4,5,6,7,8,9,10 ---> packet 3 is droped
> >
> > The question is "Does any protocol that is encapsulated by the PPP
> > will not break if a packet is dropped (case 2)
> > but will break if the packet delivery sequence is out of order (case 1) ?"
>
>Yes, as I've repeatedly pointed out, LCP, CCP, ECP, BCP, and a host
>of others will break.

All these protocol you mentioned here are not the protocols that encapsulated
by PPP but PPP's control protocol.

Although this is not my question, It is interest to me to
understand the LCP will break if the packet delivery is out of sequence but 
will not
if a packet(s) is dropped in the sequence.

?In what state that the LCP state machine (p.12 rfc1661) will break
(i.e. can not handle the incoming event)
if the incoming packets (must be more than one) is out of sequence.
Also, it will not break if the incoming packets has at least one of them is 
dropped.
Please note that in some state can send out two packets but
line 6 page 2. says that "multiple actions may be implemented in any 
convenient order".


Let's look at LCP (I assume you mean



> > Are you suggesting those protocol other than IP that encapsulated by PPP
> > satisfy this question?
>
>I'm not just suggesting it, I'm saying it,
>and yet again for the umpteenth time.
>
>
>With this, I'll stop repeating myself.
>
>
>Vernon Schryver    vjs@rhyolite.com

========================================
David Chen
Indus River Networks, Inc.
31 Nagog Park
Acton, MA 01720
U.S.A.
dchen@indusriver.com
(978) 266-8141
========================================


From owner-l2tp@ipsec.org  Fri Jan 28 19:54:00 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id TAA00514
	for <l2tpext-archive@odin.ietf.org>; Fri, 28 Jan 2000 19:53:59 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id OAA02996; Fri, 28 Jan 2000 14:56:33 -0800 (PST)
Received: from showboat.teradyne.com (showboat.teradyne.com [198.51.251.10]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id OAA02992 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 14:56:22 -0800 (PST)
Received: from chorus.teradyne.com (chorus.teradyne.com [131.101.1.195])
	by showboat.teradyne.com (8.8.8+Sun/8.8.8) with ESMTP id RAA00469
	for <l2tp@ipsec.org>; Fri, 28 Jan 2000 17:56:10 -0500 (EST)
Received: from notes.teradyne.com (jaypeak.corp.teradyne.com [131.101.17.23]) by chorus.teradyne.com (8.8.8+Sun/8.7.1) with SMTP id RAA14742 for <l2tp@ipsec.org>; Fri, 28 Jan 2000 17:56:16 -0500 (EST)
Received: by notes.teradyne.com(Lotus SMTP MTA v4.6.3  (733.2 10-16-1998))  id 85256874.007E63C0 ; Fri, 28 Jan 2000 18:00:30 -0500
X-Lotus-FromDomain: TERADYNE
From: "David Wong" <David_Wong@notes.teradyne.com>
To: l2tp@ipsec.org
Message-ID: <85256874.007E5CF4.00@notes.teradyne.com>
Date: Fri, 28 Jan 2000 17:55:41 -0500
Subject: L2TP over IPSec
Mime-Version: 1.0
Content-type: text/plain; charset=us-ascii
Content-Disposition: inline
Sender: owner-l2tp@sailpix.com
Precedence: bulk

I was curious what types of access routers/configurations people have been
successful with configuring/testing L2TP over IPSec.

Also, has anyone used a Cisco1600 series (1605) or a Win2000 Server as the LNS
while running L2TP over IPSec? The 1605 with the 12.05(T), 56-bit encryption
image does not seem to like an encrypted ICMP Echo Request I send to it in
transport mode (something about bad pad length). I know the Free S/WAN
implementation of IPSec will actually reply back to an encrypted Echo Request
that I send it.

In terms of using Windows 2000, I've tried installing the Certificate Authority
service on the machine, without luck, and even if I were to, it's not clear to
me how to use a certificate if the client I'm using needs the authentication and
encryption key information in text form.




From owner-l2tp@ipsec.org  Sun Jan 30 23:33:44 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id XAA16308
	for <l2tpext-archive@odin.ietf.org>; Sun, 30 Jan 2000 23:33:43 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id TAA07940; Sun, 30 Jan 2000 19:12:40 -0800 (PST)
Received: from mail1.cisco.com (mail1.cisco.com [171.68.225.60]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id TAA07936 for <l2tp@ipsec.org>; Sun, 30 Jan 2000 19:12:23 -0800 (PST)
Received: from GEDAVISVPN3 ([10.19.12.76] (may be forged)) by mail1.cisco.com (8.8.6 (PHNE_14041)/CISCO.SERVER.1.2) with SMTP id TAA14304; Sun, 30 Jan 2000 19:11:42 -0800 (PST)
From: "gedavis" <gedavis@cisco.com>
To: "David Wong" <David_Wong@notes.teradyne.com>, <l2tp@ipsec.org>
Cc: <gedavis@cisco.com>
Subject: RE: L2TP over IPSec
Date: Sun, 30 Jan 2000 19:11:11 -0800
Message-ID: <NBEILNFKKPPCGDCAMJCLKEFBCFAA.gedavis@cisco.com>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: 7bit
X-Priority: 3 (Normal)
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook IMO, Build 9.0.2416 (9.0.2910.0)
Importance: Normal
X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2919.6700
In-Reply-To: <85256874.007E5CF4.00@notes.teradyne.com>
Disposition-Notification-To: "gedavis" <gedavis@cisco.com>
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

David,
I have been working some with the MC-CA and what manner of client are you
referring to? router? pc? maybe I can help

Gerald


-----Original Message-----
From: owner-l2tp@sailpix.com [mailto:owner-l2tp@sailpix.com]On Behalf Of
David Wong
Sent: Friday, January 28, 2000 2:56 PM
To: l2tp@ipsec.org
Subject: L2TP over IPSec


I was curious what types of access routers/configurations people have been
successful with configuring/testing L2TP over IPSec.

Also, has anyone used a Cisco1600 series (1605) or a Win2000 Server as the
LNS
while running L2TP over IPSec? The 1605 with the 12.05(T), 56-bit encryption
image does not seem to like an encrypted ICMP Echo Request I send to it in
transport mode (something about bad pad length). I know the Free S/WAN
implementation of IPSec will actually reply back to an encrypted Echo
Request
that I send it.

In terms of using Windows 2000, I've tried installing the Certificate
Authority
service on the machine, without luck, and even if I were to, it's not clear
to
me how to use a certificate if the client I'm using needs the authentication
and
encryption key information in text form.






From owner-l2tp@ipsec.org  Mon Jan 31 12:43:22 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA10802
	for <l2tpext-archive@odin.ietf.org>; Mon, 31 Jan 2000 12:43:21 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id IAA09858; Mon, 31 Jan 2000 08:23:22 -0800 (PST)
Received: from showboat.teradyne.com (showboat.teradyne.com [198.51.251.10]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id IAA09854 for <l2tp@ipsec.org>; Mon, 31 Jan 2000 08:22:58 -0800 (PST)
Received: from chorus.teradyne.com (chorus.teradyne.com [131.101.1.195])
	by showboat.teradyne.com (8.8.8+Sun/8.8.8) with ESMTP id LAA29108;
	Mon, 31 Jan 2000 11:22:45 -0500 (EST)
Received: from notes.teradyne.com (jaypeak.corp.teradyne.com [131.101.17.23]) by chorus.teradyne.com (8.8.8+Sun/8.7.1) with SMTP id LAA01196; Mon, 31 Jan 2000 11:22:53 -0500 (EST)
Received: by notes.teradyne.com(Lotus SMTP MTA v4.6.3  (733.2 10-16-1998))  id 85256877.005A68FE ; Mon, 31 Jan 2000 11:27:31 -0500
X-Lotus-FromDomain: TERADYNE
From: "David Wong" <David_Wong@notes.teradyne.com>
To: "gedavis" <gedavis@cisco.com>
cc: l2tp@ipsec.org
Message-ID: <85256877.005A6698.00@notes.teradyne.com>
Date: Mon, 31 Jan 2000 11:22:33 -0500
Subject: RE: L2TP over IPSec
Mime-Version: 1.0
Content-type: text/plain; charset=us-ascii
Content-Disposition: inline
Sender: owner-l2tp@sailpix.com
Precedence: bulk

The client is a Linux workstation, which currently reads in the authentication
and encryption key values from a text file. So, right now it only supports
manual keying... I guess what I'm also confused about is while installing the
Certificate Authority service, I'm asked to enter in the name/address of the
root authority to request a certificate from (even though I've said that the
machine itself will be the root authority).





"gedavis" <gedavis@cisco.com> on 01/30/2000 10:11:11 PM

To:   David Wong/Bos/Teradyne@Teradyne, l2tp@ipsec.org
cc:   gedavis@cisco.com

Subject:  RE: L2TP over IPSec




David,
I have been working some with the MC-CA and what manner of client are you
referring to? router? pc? maybe I can help

Gerald


-----Original Message-----
From: owner-l2tp@sailpix.com [mailto:owner-l2tp@sailpix.com]On Behalf Of
David Wong
Sent: Friday, January 28, 2000 2:56 PM
To: l2tp@ipsec.org
Subject: L2TP over IPSec


I was curious what types of access routers/configurations people have been
successful with configuring/testing L2TP over IPSec.

Also, has anyone used a Cisco1600 series (1605) or a Win2000 Server as the
LNS
while running L2TP over IPSec? The 1605 with the 12.05(T), 56-bit encryption
image does not seem to like an encrypted ICMP Echo Request I send to it in
transport mode (something about bad pad length). I know the Free S/WAN
implementation of IPSec will actually reply back to an encrypted Echo
Request
that I send it.

In terms of using Windows 2000, I've tried installing the Certificate
Authority
service on the machine, without luck, and even if I were to, it's not clear
to
me how to use a certificate if the client I'm using needs the authentication
and
encryption key information in text form.










From owner-l2tp@ipsec.org  Mon Jan 31 13:47:31 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id NAA12523
	for <l2tpext-archive@odin.ietf.org>; Mon, 31 Jan 2000 13:47:29 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id JAA09950; Mon, 31 Jan 2000 09:40:32 -0800 (PST)
Received: from gateway.assuredaccess.com (gateway.assuredaccess.com [206.169.126.4]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id JAA09946 for <l2tp@ipsec.org>; Mon, 31 Jan 2000 09:40:23 -0800 (PST)
Received: from apollo.assuredaccess.com ([10.193.17.102])
          by gateway.assuredaccess.com (Netscape Mail Server v2.0)
          with SMTP id AAA240 for <l2tp@ipsec.org>;
          Mon, 31 Jan 2000 09:41:52 -0800
Received: from assuredaccess.com by apollo.assuredaccess.com (SMI-8.6/SMI-SVR4)
	id JAA09752; Mon, 31 Jan 2000 09:36:08 -0800
Message-ID: <3895C907.2CB0B4A@assuredaccess.com>
Date: Mon, 31 Jan 2000 09:40:23 -0800
From: Jeff Enderwick <jeff@assuredaccess.com>
X-Mailer: Mozilla 4.07 [en] (X11; I; Linux 2.0.36 i686)
MIME-Version: 1.0
To: l2tp@ipsec.org
Subject: tcpdump and l2tp?
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

Yo,

Has anyone written a tcpdump packet dumper for L2TP?
Are there any freeware sniffers that run on Linux or BSD?

Thanks,

jeff@assuredaccess.com


From owner-l2tp@ipsec.org  Mon Jan 31 15:52:55 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA15400
	for <l2tpext-archive@odin.ietf.org>; Mon, 31 Jan 2000 15:52:54 -0500 (EST)
Received: (from majordomo@localhost) by www.sailpix.com (8.8.8/8.6.9) id LAA10211; Mon, 31 Jan 2000 11:35:08 -0800 (PST)
Received: from enserv.ensemblecom.com (fwuser@[205.163.11.194]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id LAA10207 for <l2tp@ipsec.org>; Mon, 31 Jan 2000 11:34:54 -0800 (PST)
Message-ID: <3895E394.E334A329@ensemblecom.com>
Date: Mon, 31 Jan 2000 11:33:41 -0800
From: John Thomes <john@ensemblecom.com>
MIME-Version: 1.0
To: Jeff Enderwick <jeff@assuredaccess.com>
CC: l2tp@ipsec.org
Subject: Re: tcpdump and l2tp?
References: <3895C907.2CB0B4A@assuredaccess.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit


Jeff,

I created an L2TP filter for the "Ethereal" Network Analyzer.

See ---> http://ethereal.zing.org/.

Ethereal runs on Linux, BSD, AIX.... etc.

Source code is available, licensed under GNU.

Hope this helps.

-Mr. John Thomes



Ensemble Communications
San Diego

Jeff Enderwick wrote:

> Yo,
>
> Has anyone written a tcpdump packet dumper for L2TP?
> Are there any freeware sniffers that run on Linux or BSD?
>
> Thanks,
>
> jeff@assuredaccess.com



From owner-l2tp@ipsec.org  Mon Jan 31 18:56:26 2000
Received: from www.sailpix.com (mail2.sailpix.com [155.53.200.203])
	by ietf.org (8.9.1a/8.9.1a) with ESMTP id SAA18580
	for <l2tpext-archive@odin.ietf.org>; Mon, 31 Jan 2000 18:56:25 -0500 (EST)
Received: from localhost (localhost [[UNIX: localhost]]) by www.sailpix.com (8.8.8/8.6.9) id OAA10458; Mon, 31 Jan 2000 14:41:14 -0800 (PST)
Received: from gate.ascend.co.jp (root@gate.ascend.co.jp [202.246.11.253]) by www.sailpix.com (8.8.8/8.6.9) with ESMTP id OAA10454 for <l2tp@ipsec.org>; Mon, 31 Jan 2000 14:40:57 -0800 (PST)
Received: from home.ascend.co.jp (home.ascend.co.jp [202.246.11.1])
	by gate.ascend.co.jp (8.9.1a/3.7W/09/09/99) with ESMTP id HAA19206;
	Tue, 1 Feb 2000 07:40:52 +0900 (JST)
Received: from toshiba.ober.net (max-house-pool-57 [134.112.245.57])
	by home.ascend.co.jp (8.8.8/3.7W/10/14/99) with ESMTP id HAA10784;
	Tue, 1 Feb 2000 07:40:48 +0900 (JST)
Received: from localhost (localhost [127.0.0.1])
	by toshiba.ober.net (8.9.3/3.6W/home.ascend.co.jpevision: 1.1 $) with ESMTP id HAA01612;
	Tue, 1 Feb 2000 07:43:39 +0900 (JST)
To: jeff@assuredaccess.com
Cc: l2tp@ipsec.org
Subject: Re: tcpdump and l2tp?
From: Motonori Shindo <mshindo@ascend.co.jp>
In-Reply-To: <3895C907.2CB0B4A@assuredaccess.com>
References: <3895C907.2CB0B4A@assuredaccess.com>
X-Mailer: Mew version 1.94b47 on XEmacs 20.4 (Emerald)
X-PGP-fingerprint: 06 B0 B1 A4 06 C1 6A 14  63 C0 D7 18 01 CD D9 83
Mime-Version: 1.0
Content-Type: Text/Plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-Id: <20000201074338B.mshindo@ascend.co.jp>
Date: Tue, 01 Feb 2000 07:43:38 +0900
X-Dispatcher: imput version 990731(IM118)
Lines: 32
Sender: owner-l2tp@sailpix.com
Precedence: bulk
Content-Transfer-Encoding: 7bit

Jeff,

From: Jeff Enderwick <jeff@assuredaccess.com>
Subject: tcpdump and l2tp?
Date: Mon, 31 Jan 2000 09:40:23 -0800

> Yo,
> 
> Has anyone written a tcpdump packet dumper for L2TP?
> Are there any freeware sniffers that run on Linux or BSD?

I did. I placed it at

   ftp://ftp.mbuf.org/pub/l2tp/tcpdump-l2tp.patch 

This URL is tentative. Once I place it to more permanent place, I will
announce it to this mailing list again. So please stay tuned:-)

Hope this helps.

> Thanks,
> 
> jeff@assuredaccess.com

=====================================================
Motonori Shindo
   Systems Engineer     
   Ascend Communications Japan K.K.
   A wholly owned subsidiary of Lucent Technologies.
   email: mshindo@ascend.co.jp or mshindo@lucent.com
   TEL: +81-3-5325-7306 
=====================================================


