From christian.vogt at nomadiclab.com  Tue Jan  8 21:11:24 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Tue, 08 Jan 2008 15:11:24 +0200
Subject: [SAVA] SAVI Charter Revision after Vancouver
In-Reply-To: <476F36D5.1000302@juniper.net>
References: <4762FEF2.2020502@nomadiclab.com>	<A5C203E6-A090-4833-86D6-9D699255CB71@tcb.net>
	<476F36D5.1000302@juniper.net>
Message-ID: <4783767C.2000306@nomadiclab.com>

Mark -

> Hmm. I think transit packets are out-of-scope for SAVI.

That's right.  I suggest rephrasing bullet 2 in the current charter 
proposal from...

- The solutions verify the addresses of hosts; verification of router
   addresses is out of scope

...to...

- The solutions verify the IP source addresses in packets on the first
   hop; IP source address validation for packets forwarded by a router is
   out of scope

Note that this change will allow a SAVI solution to validate the IP
source addresses in packets sent by a router -- although this validation
will be limited to the packets' first hop.

> Tend to agree here. It's the way it might end up, but at this stage I 
> don't like to exclude any address-provisioning method.

The current charter proposal does not exclude any address provisioning
method as far as I see.

- Christian



From christian.vogt at nomadiclab.com  Tue Jan  8 21:49:55 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Tue, 08 Jan 2008 15:49:55 +0200
Subject: [SAVA] Limiting SAVI Solutions to Specific Network Entities
Message-ID: <47837F83.4030204@nomadiclab.com>

Jun -

in your recent emails you indicate a preference for limiting the first
SAVI solution to one that is to be implemented on switches, and to
specify separate solutions for implementation on other network entities
after rechartering.  While I agree with you that SAVI should consider
rechartering and the specification of advanced solutions at some point,
I do not agree that the differentiating criterion between the first SAVI
solution and those advanced solutions should be the entity on which it
could be implemented.

A good protocol is flexible WRT the network entity on which it gets
implemented, and I would like to uphold this principle also for our
prospective SAVI solution.  A more appropriate criterion to distinguish
a first from possible future SAVI solutions would, IMO, be a different
complexity-benefit trade-off -- more benefits at higher costs.  And
[draft-bi-sava-solution-ipv6-edge-network-signature] and
[draft-wu-sava-solution-firsthop-eap] might be good starting points for
such work.

But for now, let's concentrate on the first charter. ;-)

- Christian





From christian.vogt at nomadiclab.com  Wed Jan  9 04:17:25 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Tue, 08 Jan 2008 22:17:25 +0200
Subject: [SAVA] Review of draft-baker-sava-implementation-00
In-Reply-To: <F19210DD-90FC-4DE5-8827-0FB1CB64623C@cisco.com>
References: <474724DD.4050202@nomadiclab.com>
	<F19210DD-90FC-4DE5-8827-0FB1CB64623C@cisco.com>
Message-ID: <4783DA55.40509@nomadiclab.com>

Hi Fred,

getting back to our discussion on draft-baker-sava-implementation-00...


> Would adding this paragraph meet your needs, given that what you say 
> above is repeated in the next section?
> 
> The key point is the association of an IPv6 source address with a lower 
> layer identifier, which might be purely local in significance like a 
> port identifier on a switch, might be visible in the LAN like a MAC 
> address, and might be a combination of lower layer identifiers like "the 
> system on one of two ports using a stated MAC address". The association 
> is called a "trust binding", and is fundamental to the solution. The 
> implementation details are left to the implementor.

Right, the proposed solution should be introduced at such a high level
first, IMO.  However, I do not agree with the last sentence in the above
paragraph.  The specification should be much more detailed than it
currently is.  The current version leaves implementors too much freedom.
We should provide a clear algorithm for each address configuration mode.


>> - Means for address configuration we need to consider:  DHCP,
>>   autonomous configuration
> 
> how, precisely, does one snoop SAA, given that it sends no "Hey, 
> everybody, I have chosen to use this address" message?  The reason I went 
> to what you call "double checking" is that it is the first time we see 
> the device assert that it is in fact using a given address in SAA. The 
> DAD check asserts that it *might* start using it and gives an 
> opportunity for someone else to say "please don't".

I think what is relevant here is that the existing message that says
"Hey, everybody, I have chosen to use this address" -- namely the NS of
the statelessly configuring entity -- is insufficient as a trigger to
create a trust binding.  This is so regardless of whether a host, a
router, or a switch is validating an address, because a potential
defending NA would be sent by unicast and thus may not be received by a
snooping entity that sits behind a switch.  This is why another NS/NA
exchange between the validating and validated entities is needed.


>> - Reactive verification of address binding (double-checking):  If no
>>   binding exists for a packet, double check addresses prior to dropping
>>   the packet, e.g., based on Neighbor Unreachability Detection.
> 
> A host might check the address before dropping it. A router will simply 
> drop the packet if it has no neighbor binding.

If a router drops a packet in case it has no binding, wouldn't this be
problematic if that router is behind a switch (see my previous comment)?


>> - What happens if a filter misses part of the address configuration
>>   signaling before it boots up or appears on a link?  (E.g., what about
>>   filters on mobile hosts?)
> 
> I'm not sure I follow. When a host decides to use a new address and 
> sends a datagram using it to a neighboring host, RFC 4862 has the 
> neighboring host send the Neighbor Solicit sometime between receiving 
> the datagram and sending a response. [...]

My comment addressed the situation where the SAVI filter is implemented
on a router or switch, and a host configures its address at a time or at
a place where the SAVI filter cannot snoop the address configuration
traffic.  This may happen if the filter becomes enabled on a link where
hosts already have their addresses, or when a mobile host takes an
address from one switch to another switch on the same link.  If we
require routers and switches to pursue an NS/NA exchange with the
validated host before dropping a packet, as suggested in my previous
comment, then this issue is solved as well.


>> - How long should refresh triggers and timeout values for soft-state
>>   bindings be?
> 
> what is the refresh timeout for ND?

Dimensioning the trust binding lifetime in accordance with Neighbor
Unreachability Detection timeouts is a good idea.

Neighbor Unreachability Detection uses two ways to refresh state (see
RFC 4861, section 7.3.1), of which only one is applicable without
restrictions to our SAVI solution:

- "Forward progress" of an application can only be determined by a host
   engaged in active communications with a peer, hence it is not
   applicable when the SAVI filter is not on a host.

- Periodic exchanges of NS/NA messages happen by default every 35 s
   (REACHABLE_TIME + DELAY_FIRST_PROBE_TIME).  Such exchanges could be
   applied by any kind of SAVI filter.

I would recommend making the timeout interval in SAVI configurable with
a default of 30 s (REACHABLE_TIME).

- Christian






From fred at cisco.com  Wed Jan  9 04:33:35 2008
From: fred at cisco.com (Fred Baker)
Date: Tue, 8 Jan 2008 12:33:35 -0800
Subject: [SAVA] SAVI Charter Revision after Vancouver
In-Reply-To: <4783767C.2000306@nomadiclab.com>
References: <4762FEF2.2020502@nomadiclab.com>	<A5C203E6-A090-4833-86D6-9D699255CB71@tcb.net>
	<476F36D5.1000302@juniper.net> <4783767C.2000306@nomadiclab.com>
Message-ID: <41B07028-530F-4FAB-A412-6F5C2F49368F@cisco.com>

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


On Jan 8, 2008, at 5:11 AM, Christian Vogt wrote:

> That's right.  I suggest rephrasing bullet 2 in the current charter  
> proposal from...
>
> - The solutions verify the addresses of hosts; verification of router
>    addresses is out of scope
>
> ...to...
>
> - The solutions verify the IP source addresses in packets on the first
>    hop; IP source address validation for packets forwarded by a  
> router is
>    out of scope
>
> Note that this change will allow a SAVI solution to validate the IP  
> source addresses in packets sent by a router -- although this  
> validation will be limited to the packets' first hop.

It presumes that we can distinguish between datagrams sent by the  
layers above IP in a router from those that IP forwarded. Let me draw  
you a picture:

                +----------------+
                |                |
                |+-------+------+|
                || SNMP  |Trojan||
                ||Process|Module||
                |+-------+------+|
                |+--------------+|
                ||   IP Protocol||
                |+-+--------+---+|
                +--+--------+----+
                   |        |
           ---+----+---   --+-----+-------
              |                   |
           +--+---+            +--+-+
           |Remote|            |SAVI|
           | Host |            |Host|
           +------+            +----+

Here we have three hypothetical systems - a router with at least two  
interfaces and a host on either side. The "SAVI Host" receives four  
separate datagrams, each containing the MAC address of the router  
interface on its side and one of four IP addresses:

  - a datrgam correctly initiated by the remote host and forwarded by  
the router
  - an SNMP response correctly originated from the router's "remote"  
interface
  - an SNMP response correctly originated from the router's "local"  
interface
  - some other message originated with a spoofed address by the  
Trojan in the router

By what set of rules does one disambiguate this?

SeND gives us a way for the SAVI Host and its Router to manage  
address exchanges for the local LAN. The set of requirements I  
described similarly enable the host to be sure of the routers local  
LAN address when standard ND is used, but with a lower level of  
assurance. Since neither ND nor SeND are negotiated for remote  
addresses, they don't help with the Trojan, the remote host, or the  
non-local address on the router. Standard uRPF or filter-based  
ingress filtering can be used to provide a reasonableness check on  
the other addresses, but this is what Wu's problem statement asserts  
is inadequate - if the Trojan guessed a prefix that happens to match  
the uRPF criteria, it will not be detected.

I don't see how one distinguishes between a datagram originated by a  
router and one forwarded by a router. If you say that the TTL on a  
datagram originated by a router must be 255, then having the Trojan  
set its TTL to 254 means that the test isn't applied to its traffic.

I think datagrams that go through a router, whether forwarded or  
originated there, have to be ruled out of scope, unless someone can  
give me a set of rules to address this.

I also think that this isn't properly a charter question, it is a  
technology question. If the working group is smarter than I and comes  
up with a way, the charter shouldn't say that we're not allowed to  
propose it. I think that the charter should simply say that we are  
working on the local LAN variation on ingress filtering, and that  
further discussion of routing-based ingress filtering is either out  
of scope or is a second topic.
-----BEGIN PGP SIGNATURE-----

iD8DBQFHg94fbjEdbHIsm0MRAsckAJ4xWCc32mFLPjHrIvcXeIVvqlTvYQCfW4yz
DBt56GkyhpZYT5mUficrAJA=
=icRO
-----END PGP SIGNATURE-----


From christian.vogt at nomadiclab.com  Wed Jan  9 18:56:29 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Wed, 09 Jan 2008 12:56:29 +0200
Subject: [SAVA] SAVI Charter Revision after Vancouver
In-Reply-To: <41B07028-530F-4FAB-A412-6F5C2F49368F@cisco.com>
References: <4762FEF2.2020502@nomadiclab.com>	<A5C203E6-A090-4833-86D6-9D699255CB71@tcb.net>
	<476F36D5.1000302@juniper.net> <4783767C.2000306@nomadiclab.com>
	<41B07028-530F-4FAB-A412-6F5C2F49368F@cisco.com>
Message-ID: <4784A85D.8010608@nomadiclab.com>

Fred -

if we knew that hosts on the same link never exchange packets via a
router, then a receiving host could judge from the IP source address
prefix of a packet received from a router whether the packet was
originated or forwarded by the router:

- If the prefix is on-link, the receiving host would initiate an NS/NA
   exchange to verify the packet's IP source address.
- If the prefix is off-link, no IP address validation would occur.

The special case where the host receives a packet from the router's
remote interface would then be considered a forwarding event.

However, these IP source address validation rules yield a false negative
when hosts on the same link exchange packets via a router:  The router
would in this case be the packet sender, and the NS/NA exchange would
falsely indicate that the packet's IP source address is false.

It therefore indeed seems that IP source address validation should be
omitted for packets received from a router -- be it packets that were
forwarded or originated by the router.  A host could check its
default-router list [RFC 4861] to decide whether an on-link packet
sender is a router or a host.

- Christian






From christian.vogt at nomadiclab.com  Mon Jan 14 19:15:06 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Mon, 14 Jan 2008 13:15:06 +0200
Subject: [SAVA] SAVI Charter Revision after Vancouver, Take 3
Message-ID: <478B443A.8060206@nomadiclab.com>

Folks,

enclosed please find the 3rd take of the proposed SAVI charter.  Please
send your comments no later than Friday, January 18, so that we can nail
down the final charter proposal early next week.

Differences between take 2 and 3 of the charter are limited to a
revision of bullet 2, which now more clearly states that the validation
of addresses in packets sent/forwarded by routers is out of scope.

- Christian



-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: savi-charter-2008-01-14.txt
Url: http://mail.nrc.tsinghua.edu.cn/pipermail/sava/attachments/20080114/a631e8b4/attachment.txt 

From fred at cisco.com  Tue Jan 15 03:34:41 2008
From: fred at cisco.com (Fred Baker)
Date: Mon, 14 Jan 2008 11:34:41 -0800
Subject: [SAVA] SAVI Charter Revision after Vancouver
In-Reply-To: <4784A85D.8010608@nomadiclab.com>
References: <4762FEF2.2020502@nomadiclab.com>	<A5C203E6-A090-4833-86D6-9D699255CB71@tcb.net>
	<476F36D5.1000302@juniper.net> <4783767C.2000306@nomadiclab.com>
	<41B07028-530F-4FAB-A412-6F5C2F49368F@cisco.com>
	<4784A85D.8010608@nomadiclab.com>
Message-ID: <80DF4BB3-206F-45D2-8AB2-CE7F7953C136@cisco.com>

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Sorry, for whatever reason I am just seeing this note.

Yes. What I would suggest is that any system (switch or host/router)  
have an exception list that identifies systems, ports, etc that it  
should not attempt to verify. The default-router list would be an  
obvious set of systems that a host should not attempt to verify. On a  
switch, if it is tying IP addresses to ports, it should be able to  
exclude trunk ports (ports to other switches), as there is  
potentially an arbitrarily large number of addresses to associate  
with them.

On Jan 9, 2008, at 2:56 AM, Christian Vogt wrote:

> Fred -
>
> if we knew that hosts on the same link never exchange packets via a
> router, then a receiving host could judge from the IP source address
> prefix of a packet received from a router whether the packet was
> originated or forwarded by the router:
>
> - If the prefix is on-link, the receiving host would initiate an NS/NA
>   exchange to verify the packet's IP source address.
> - If the prefix is off-link, no IP address validation would occur.
>
> The special case where the host receives a packet from the router's
> remote interface would then be considered a forwarding event.
>
> However, these IP source address validation rules yield a false  
> negative
> when hosts on the same link exchange packets via a router:  The router
> would in this case be the packet sender, and the NS/NA exchange would
> falsely indicate that the packet's IP source address is false.
>
> It therefore indeed seems that IP source address validation should be
> omitted for packets received from a router -- be it packets that were
> forwarded or originated by the router.  A host could check its
> default-router list [RFC 4861] to decide whether an on-link packet
> sender is a router or a host.
>
> - Christian
>
>
>
>

-----BEGIN PGP SIGNATURE-----

iD8DBQFHi7lSbjEdbHIsm0MRAlkZAJ9WYSOTZmwgEzvRbCJrHBRleqNzmQCfflgI
v9mITkw7TXB5hTlUezEl+Ss=
=1ljB
-----END PGP SIGNATURE-----


From christian.vogt at nomadiclab.com  Tue Jan 15 07:24:05 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Tue, 15 Jan 2008 01:24:05 +0200
Subject: [SAVA] SAVI Charter Revision after Vancouver
In-Reply-To: <80DF4BB3-206F-45D2-8AB2-CE7F7953C136@cisco.com>
References: <4762FEF2.2020502@nomadiclab.com>	<A5C203E6-A090-4833-86D6-9D699255CB71@tcb.net>
	<476F36D5.1000302@juniper.net> <4783767C.2000306@nomadiclab.com>
	<41B07028-530F-4FAB-A412-6F5C2F49368F@cisco.com>
	<4784A85D.8010608@nomadiclab.com>
	<80DF4BB3-206F-45D2-8AB2-CE7F7953C136@cisco.com>
Message-ID: <478BEF15.4040107@nomadiclab.com>

> Yes. What I would suggest is that any system (switch or host/router) 
> have an exception list that identifies systems, ports, etc that it 
> should not attempt to verify. The default-router list would be an 
> obvious set of systems that a host should not attempt to verify. On a 
> switch, if it is tying IP addresses to ports, it should be able to 
> exclude trunk ports (ports to other switches), as there is potentially 
> an arbitrarily large number of addresses to associate with them.

Yep, this makes sense.

- Christian



From pekkas at netcore.fi  Thu Jan 24 21:03:37 2008
From: pekkas at netcore.fi (Pekka Savola)
Date: Thu, 24 Jan 2008 15:03:37 +0200 (EET)
Subject: [SAVA] SAVI Charter Revision after Vancouver, Take 3
In-Reply-To: <478B443A.8060206@nomadiclab.com>
References: <478B443A.8060206@nomadiclab.com>
Message-ID: <alpine.LRH.1.00.0801241455420.28308@netcore.fi>

On Mon, 14 Jan 2008, Christian Vogt wrote:
> enclosed please find the 3rd take of the proposed SAVI charter.  Please
> send your comments no later than Friday, January 18, so that we can nail
> down the final charter proposal early next week.
>
> Differences between take 2 and 3 of the charter are limited to a
> revision of bullet 2, which now more clearly states that the validation
> of addresses in packets sent/forwarded by routers is out of scope.

The date is already over, but it looks pretty good.

s/multipe/multiple/.

Looking at the minutes of the BOF (I wasn't in Vancouver), I believe 
Sam Hartman will at least think the following bullet points are not 
clear enough.

"- The solutions must support interfaces with multipe IP addresses
- The solutions must support hosts with multiple interfaces
- The solutions must interoperate with mobile hosts, which may change
   the point of network attachment for one of their interfaces"

Specifically, does "must support" and "must interoperate" require more 
than just 'administrator should turn off SAVI check on the affected 
ports'?  For example, are the mechanisms required to be robust enough 
that they 1) will work without administrator action even in these 
scenarios, and that 2) 'will work' does not imply 'turn off the SAVI 
check for that port' ?

I guess more clarity on this wording could go a long way in 
alleviating concerns about the mechanism being actively harmful if 
deployed (e.g., by restricting common and/or valid deployment 
scenarios, e.g., virtualization).

On the other hand, if we don't yet know if solutions could be made to 
work under these premises automatically, we're in a more difficult 
position..

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings


From christian.vogt at nomadiclab.com  Fri Jan 25 19:27:10 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Fri, 25 Jan 2008 12:27:10 +0100
Subject: [SAVA] SAVI Charter Revision after Vancouver, Take 4
Message-ID: <4799C78E.1000507@nomadiclab.com>

Folks,

our current charter proposal refers to the RIPE IP Anti-Spoofing task
force, but this was closed down in the meantime.  I therefore removed
the explicit reference.  Modified charter proposal attached.

- Christian


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: savi-charter-2008-01-25.txt
Url: http://mail.nrc.tsinghua.edu.cn/pipermail/sava/attachments/20080125/90f3f40e/attachment.txt 

From christian.vogt at nomadiclab.com  Mon Jan 28 16:55:51 2008
From: christian.vogt at nomadiclab.com (Christian Vogt)
Date: Mon, 28 Jan 2008 10:55:51 +0200
Subject: [SAVA] SAVI Charter Revision after Vancouver, Take 4
In-Reply-To: <alpine.LRH.1.00.0801241455420.28308@netcore.fi>
References: <478B443A.8060206@nomadiclab.com>
	<alpine.LRH.1.00.0801241455420.28308@netcore.fi>
Message-ID: <479D9897.4070607@nomadiclab.com>

Hi Pekka,

thanks for your charter review!  I clarified that mobile and multi-
interfaced hosts are to be supported "by default", i.e., without special
administrator action.  This modification affects bullets 7-9.

Revised charter version attached.

- Christian


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: savi-charter-2008-01-28.txt
Url: http://mail.nrc.tsinghua.edu.cn/pipermail/sava/attachments/20080128/8d552a8c/attachment.txt 

