[BEHAVE] Review Part 1 of draft-ietf-behave-v6v4-xlate-stateful-03

[George Tsirtsis <[e-mail]>]

Hi all,

I have been reviewing this draft in considerable detail. I think the
document is on the right track but I also I think it needs a fair
amount of work to make it comprehensible and clear enough to be
implementable.

Specifically I propose:
-  to more clearly identify the type of mapping and filtering
(Endpoint-Independent., Address-Dependent, Address and Port-Dependent)
supported in general and/or described in particular in certain parts
of the document. This will help put context and reason in some of the
processing described.
- to utilise the notation (X',x) <--> (T,t) and  (X',x),(Y',y) <-->
(T,t),(Z,z) when describing the searching or setting of the BIB and
session table entries. This will help make descriptions much easier to
understand, because some of them at the moment are little
mind-twisters.

I also provide a number of other comments/clarifications/corrections.

Due to the extend of the comments and description style changes I
propose for Section 3 and below, I stopped the review at section
3.2.2. for now, to get some feedback from the WG and the authors. If
the authors and WG think the comments are reasonable and useful, I
will continue the review to the rest of the document, and will work
with the authors as needed.

Regards
George
-----------


                draft-ietf-behave-v6v4-xlate-stateful-03

1.  Introduction

...

1.1.  Features of NAT64

GT> I think it would be beneficial to state clearly the type of NAT
mapping and filtering that is supported in this document. The document
seems to specifically support ONLY Endpoint-Independent Mappings,
while WRT to filtering types it says it adheres to RFC4787 and
RFC5382. In practice, however, it is not clear to what extend it does
that. see below.

   The features of NAT64 and DNS64 are:

   o  It enables IPv6-only nodes to initiate a client-server connection
      with an IPv4-only server, without needing any changes on either
      IPv4 or IPv6 nodes.  This works for roughly the same class of
      applications that work through IPv4-to-IPv4 NATs.

   o  It supports peer-to-peer communication between IPv4 and IPv6
      nodes, including the ability for IPv4 nodes to initiate
      communication with IPv6 nodes using peer-to-peer techniques (i.e.,
      using a rendezvous server and ICE).  To this end, NAT64 is
      compliant with the recommendations for how NATs should handle UDP
      [RFC4787], TCP [RFC4787], and ICMP [RFC5508].

GT> //TCP [RFC4787]/TCP [RFC5382]

   o  NAT64 supports IPv4 initiated communications to a subset of the
      IPv6 hosts through statically configured bindings in the NAT64
      device.

   o  Compatible with ICE.


GT> It would be better to spell things out a bit more clearly:
  a) in the *absence* of any state in NAT64 regarding a given IPv6
node, only said IPv6 node can initiate sessions to IPv4 nodes
  b) depending on the filtering used (Endpoint-Independent,
Address-Dependent, Address and Port-Dependent), IPv4-nodes MAY be able
to initiate sessions to a given IPv6 node, if the NAT64 somehow has an
appropriate mapping (i.e.,state) for said IPv6 node, via one of the
following mechanism.
          i) the IPv6 node has recently initiated a session to the
same or other external-IPv4 node
          ii) The IPv6 node has used ICE which essentially results in (i)
          iii) if static configuration (i.e., mapping) exists
regarding said IPv6 node

   o  Supports additional features with some changes on nodes.  These
      features include:

      *  Support for DNSSEC

GT> in some cases only though, no?

      *  Some forms of IPsec support

1.2.  Overview

   This section provides a non-normative introduction to the mechanisms
   of NAT64.

   NAT64 mechanism is implemented in an NAT64 box which has two
   interfaces, an IPv4 interface connected to the the IPv4 network, and
   an IPv6 interface connected to the IPv6 network.

GT> I think this is an overly simplistic view of the world. If for the
purpose of specification this makes descriptions easier then so be it
but there should at least be a section towards the end of the document
dealing with more complex scenarios where more than two interfaces
exist, IPv4-only, IPv6-only, and dual stack nodes are mixed in a given
site; identification and handling of packets not needing translation
etc, if there is interest in clarifying any of this we can discuss
separately.

   Packets generated
   in the IPv6 network for a receiver located in the IPv4 network will
   be routed within the IPv6 network towards the NAT64 box.  The NAT64
   box will translate them and forward them as IPv4 packets through the
   IPv4 network to the IPv4 receiver.

GT> We should not refer to NAT64 as a "box". Instead it is a function
that may operate on any interface of any router device. As a function,
NAT64 may only consider two IFs but a "box" may have many more IFs
etc.

The reverse takes place for
   packets generated in the IPv4 network for an IPv6 receiver.  NAT64,
   however, is not symmetric.  In order to be able to perform IPv6 -
   IPv4 translation NAT64 requires state, binding an IPv6 address and
   port (hereafter called an IPv6 transport address) to an IPv4 address
   and port (hereafter called an IPv4 transport address).

   Such binding state is created when the first packet flowing from the



Bagnulo, et al.           Expires May 25, 2010                  [Page 5]

Internet-Draft                    NAT64                    November 2009


   IPv6 network to the IPv4 network is translated.  After the binding
   state has been created, packets flowing in either direction on that
   particular flow are translated.  The result is that NAT64 only
   supports communications initiated by the IPv6-only node towards an
   IPv4-only node.  Some additional mechanisms, like ICE, can be used in
   combination with NAT64 to provide support for communications
   initiated by the IPv4-only node to the IPv6-only node.  The
   specification of such mechanisms, however, is out of the scope of
   this document.

GT> I think the above is inaccurate in multiple levels.
- static configuration is not mentioned
- assuming filtering allows it, endpoint independent mappings mean
that even w/o ICE, an IPv4 node may be able to contact an IPv6 node
itf the latter has initiated a session to the same or other IPv4 node.


1.2.1.  NAT64 solution elements

   In this section we describe the different elements involved in the
   NAT64 approach.

   The main component of the proposed solution is the translator itself.
   The translator has essentially two main parts, the address
   translation mechanism and the protocol translation mechanism.

   Protocol translation from IPv4 packet header to IPv6 packet header
   and vice-versa is performed according to IP/ICMP Translation
   Algorithm [I-D.ietf-behave-v6v4-xlate].

   Address translation maps IPv6 transport addresses to IPv4 transport
   addresses and vice-versa.  In order to create these mappings the
   NAT64 box has two pools of addresses i.e. an IPv6 address pool (to
   represent IPv4 addresses in the IPv6 network) and an IPv4 address
   pool (to represent IPv6 addresses in the IPv4 network).  Since there
   is enough IPv6 address space, it is possible to map every IPv4
   address into a different IPv6 address.

   NAT64 creates the required mappings by using as the IPv6 address pool
   an IPv6 IPv6 prefix (hereafter called Pref64::/n).  This allows each
   IPv4 address to be mapped into a different IPv6 address by simply
   concatenating the Pref64::/n prefix assigned as the IPv6 address pool
   of the NAT64, with the IPv4 address being mapped and a suffix (i.e.
   an IPv4 address X is mapped into the IPv6 address Pref64:X:SUFFIX).
   The NAT64 prefix Pref64::/n is assigned by the administrator of the
   NAT64 box from the global unicast IPv6 address block assigned to the
   site.

   The IPv4 address pool is a set of IPv4 addresses, normally a small
   prefix assigned by the local administrator.  Since IPv4 address space
   is a scarce resource, the IPv4 address pool is small and typically
   not sufficient to establish permanent one-to-one mappings with IPv6
   addresses.  So, mappings using the IPv4 address pool will be created
   and released dynamically.  Moreover, because of the IPv4 address
   scarcity, the usual practice for NAT64 is likely to be the mapping of



Bagnulo, et al.           Expires May 25, 2010                  [Page 6]

Internet-Draft                    NAT64                    November 2009


   IPv6 transport addresses into IPv4 transport addresses, instead of
   IPv6 addresses into IPv4 addresses directly, which enable a higher
   utilization of the limited IPv4 address pool.

   Because of the dynamic nature of the IPv6 to IPv4 address mapping and
   the static nature of the IPv4 to IPv6 address mapping, it is easy to
   understand that it is far simpler to allow communication initiated
   from the IPv6 side toward an IPv4 node, which address is permanently

GT> instead of "permanently" maybe use the term "deterministically" or
maybe  "algorithmically"?

   mapped into an IPv6 address, than communications initiated from IPv4-
   only nodes to an IPv6 node in which case IPv4 address needs to be
   associated with it dynamically.  For this reason NAT64 supports only
   communications initiated from the IPv6 side.

GT> Delete the last statement, which is NOT true given possible static
configuration as well as some other rather peculiar tricks the
document supports below for p2p apps etc.

   An IPv6 initiator can know or derive in advance the IPv6 address
   representing the IPv4 target and send packets to that address.  The
   packets are intercepted by the NAT64 device, which associates an IPv4
   transport address of its IPv4 pool to the IPv6 transport address of
   the initiator, creating binding state, so that reply packets can be
   translated and forwarded back to the initiator.  The binding state is
   kept while packets are flowing.  Once the flow stops, and based on a
   timer, the IPv4 transport address is returned to the IPv4 address
   pool so that it can be reused for other communications.

   To allow an IPv6 initiator to do the standard DNS lookup to learn the
   address of the responder, DNS64 [I-D.ietf-behave-dns64] is used to
   synthesize an AAAA RR from the A RR (containing the real IPv4 address
   of the responder).  DNS64 receives the DNS queries generated by the
   IPv6 initiator.  If there is no AAAA record available for the target
   node (which is the normal case when the target node is an IPv4-only
   node), DNS64 performs a query for the A record.  If an A record is
   discovered, DNS64 creates a synthetic AAAA RR that includes the IPv6
   representations of the IPv4 address created by concatenating the
   Pref64::/n of a NAT64 to the responder's IPv4 address and a suffix
   (i.e. if the IPv4 node has IPv4 address X, then the synthetic AAAA RR
   will contain the IPv6 address formed as Pref64:X:SUFFIX).  The
   synthetic AAAA RR is passed back to the IPv6 initiator, which will
   initiate an IPv6 communication with the IPv6 address associated to
   the IPv4 receiver.  The packet will be routed to the NAT64 device,
   which will create the IPv6 to IPv4 address mapping as described
   before.

1.2.2.  Walkthrough

   In this example, we consider an IPv6 node located in a IPv6-only site
   that initiates a communication to a IPv4 node located in the IPv4
   network.

   The notation used is the following: upper case letters are IPv4



Bagnulo, et al.           Expires May 25, 2010                  [Page 7]

Internet-Draft                    NAT64                    November 2009


   addresses; upper case letters with a prime(') are IPv6 addresses;
   lower case letters are ports; prefixes are indicated by "P::X", which
   is a IPv6 address built from an IPv4 address X by adding the prefix
   P, mappings are indicated as "(X,x) <--> (Y',y)".

   The scenario for this case is depicted in the following figure:


     +---------------------------------------+       +---------------+
     |IPv6 network    +-------------+        |       |               |
     |  +----+        | Name server |   +-------+    |   IPv4        |
     |  | H1 |        | with DNS64  |   | NAT64 |----| Network       |
     |  +----+        +-------------+   +-------+    |               |
     |    |IP addr: Y'     |              |  |       |  IP addr: X   |
     |    ---------------------------------  |       |  +----+       |
     +---------------------------------------+       |  | H2 |       |
                                                     |  +----+       |
                                                     +---------------+

   The figure shows a IPv6 node H1 which has an IPv6 address Y' and an
   IPv4 node H2 with IPv4 address X.

   A NAT64 connects the IPv6 network to the IPv4 network.  This NAT64
   has a /n prefix (called Pref64::/n) that it uses to represent IPv4
   addresses in the IPv6 address space and an IPv4 address T assigned to
   its IPv4 interface. the routing is configured in such a way, that the
   IPv6 packets addressed to a destination address containing Pref64::/n
   are routed to the IPv6 interface of the NAT64 box.

GT> I think it might be good to indicate that for this site,
Pref64::/n is uniquely used for mapping the IPv4 address space  to
IPv6 i.e., no address from Pref64::/n is to be assigned to a native
IPv6 interface. It should also be said that Pref64::/ should only be
given to a single NAT64 in a given site. Anything other than that
requires a lot more explanation.


   Also shown is a local name server with DNS64 functionality.  The
   local name server needs to know the /n prefix assigned to the local
   NAT64 (Pref64::/n).  For the purpose of this example, we assume it
   learns this through manual configuration.

   For this example, assume the typical DNS situation where IPv6 hosts
   have only stub resolvers and the local name server does the recursive
   lookups.

   The steps by which H1 establishes communication with H2 are:

   1.  H1 performs a DNS query for FQDN(H2) and receives the synthetic
       AAAA RR from the local name server that implements the DNS64
       functionality.  The AAAA record contains an IPv6 address formed
       by the Pref64::/n associated to the NAT64 box and the IPv4
       address of H2 and a suffix.

   2.  H1 sends a packet to H2.  The packet is sent from a source
       transport address of (Y',y) to a destination transport address of



Bagnulo, et al.           Expires May 25, 2010                  [Page 8]

Internet-Draft                    NAT64                    November 2009


       (Pref64:X:SUFFIX,x), where y and x are ports set by H1.

   3.  The packet is routed to the IPv6 interface of the NAT64 (since
       the IPv6 routing is configured that way).

   4.  The NAT64 receives the packet and performs the following actions:

       *  The NAT64 selects an unused port t on its IPv4 address T and
          creates the mapping entry (Y',y) <--> (T,t)

       *  The NAT64 translates the IPv6 header into an IPv4 header using
          IP/ICMP Translation Algorithm [I-D.ietf-behave-v6v4-xlate].

       *  The NAT64 includes (T,t) as source transport address in the
          packet and (X,x) as destination transport address in the
          packet.  Note that X is extracted directly from the
          destination IPv6 address of the received IPv6 packet that is
          being translated.

   5.  The NAT64 sends the translated packet out its IPv4 interface and
       the packet arrives at H2.

   6.  H2 node responds by sending a packet with destination transport
       address (T,t) and source transport address (X,x).

   7.  The packet is routed to the NAT64 box, which will look for an
       existing mapping containing (T,t).  Since the mapping (Y',y) <-->
       (T,t) exists, the NAT64 performs the following operations:

       *  The NAT64 translates the IPv4 header into an IPv6 header using
          IP/ICMP Translation Algorithm [I-D.ietf-behave-v6v4-xlate].

       *  The NAT64 includes (Y',y) as destination transport address in
          the packet and (Pref64:X:SUFFIX,x) as source transport address
          in the packet.  Note that X is extracted directly from the
          source IPv4 address of the received IPv4 packet that is being
          translated.

   8.  The translated packet is sent out the IPv6 interface to H1.

   The packet exchange between H1 and H2 continues and packets are
   translated in the different directions as previously described.

   It is important to note that the translation still works if the IPv6
   initiator H1 learns the IPv6 representation of H2's IPv4 address
   (i.e.  Pref64:X:SUFFIX) through some scheme other than a DNS look-up.
   This is because the DNS64 processing does NOT result in any state
   installed in the NAT64 box and because the mapping of the IPv4



Bagnulo, et al.           Expires May 25, 2010                  [Page 9]

Internet-Draft                    NAT64                    November 2009


   address into an IPv6 address is the result of concatenating the
   prefix defined within the site for this purpose (called Pref64::/n in
   this document) to the original IPv4 address and a suffix.

1.2.3.  Filtering

   A NAT64 box may do filtering, which means that it only allows a
   packet in through an interface if the appropriate permission exists.
   A NAT64 may do no filtering, or it may filter on its IPv4 interface.
   Filtering on the IPv6 interface is not supported, as mappings are
   only created by packets traveling in the IPv6 --> IPv4 direction.

GT> Again this NOT strictly true given the rest of the document. What
is it that you mean here by "filtering on the IPv6 interface"?

   If a NAT64 performs address-dependent filtering according to RFC4787
   [RFC4787] on its IPv4 interface, then an incoming packet is dropped
   unless a packet has been recently sent out the interface with a
   source transport address equal to the destination transport address
   of the incoming packet and destination IP address equal to the source
   IP address of the incoming packet.

   NAT64 filtering is consistent with the recommendations of RFC 4787
   [RFC4787], and the ones of RFC 5382 [RFC5382]


GT> I think this section should be re-written. RFCs4787/5382 provide
very detailed requirements and reasoning for each type of filtering,
most of which is lost in this section. The brief description of only
some cases is misleading and over-simplified. Since it is rather
pointless to repeat the contents of these RFCs, It would be beneficial
to simply list the known filtering behaviours ("Endpoint-Independent
Filtering", "Address-Dependent Filtering", "Address and Port-Dependent
Filtering") and then summarize what the quoted RFCs recommend i.e. use
Endpoint-Independent Filtering whenever possible, use
Address-Dependent Filtering when you have to, and avoid using Address
and Port-Dependent Filtering if at all possible. Then the reader
should be encourage to really read these quoted RFCs to get all the
details, w/o which it is actually very difficult to understand the
rest of this document.

2.  Terminology

   This section provides a definitive reference for all the terms used
   in document.

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in RFC 2119 [RFC2119].

   The following terms are used in this document:

   3-Tuple:  The tuple (source IP address, destination IP address, Query
      Identifier).  A 3-tuple uniquely identifies an ICMP Query session.
      When an ICMP Query session flows through a NAT64, each session has
      two different 3-tuples: one with IPv4 addresses and one with IPv6
      addresses.

   5-Tuple:  The tuple (source IP address, source port, destination IP
      address, destination port, transport protocol).  A 5-tuple
      uniquely identifies a UDP/TCP session.  When a UDP/TCP session
      flows through a NAT64, each session has two different 5-tuples:
      one with IPv4 addresses and one with IPv6 addresses.






Bagnulo, et al.           Expires May 25, 2010                 [Page 10]

Internet-Draft                    NAT64                    November 2009


   BIB:  Binding Information Base.  A table of mappings kept by a NAT64.
      Each NAT64 has three BIBs, one for TCP, one for UDP and one for
      ICMP Queries.

   DNS64:  A logical function that synthesizes AAAA Resource Records
      (containing IPv6 addresses) from A Resource Records (containing
      IPv4 addresses).

   Endpoint-Independent Mapping:  In NAT64, using the same mapping for
      all the sessions involving a given IPv6 transport address of an
      IPv6 host (irrespectively of the transport address of the IPv4
      host involved in the communication).  Endpoint-independent mapping
      is important for peer-to-peer communication.  See [RFC4787] for
      the definition of the different types of mappings in IPv4-to-IPv4
      NATs.

GT> It would probably be useful to also include the terms "Address-
      Dependent" and "Address and Port-Dependent"

   Hairpinning:  Having a packet do a "U-turn" inside a NAT and come
      back out the same interface as it arrived on.  Hairpinning support
      is important for peer-to-peer applications, as there are cases
      when two different hosts on the same side of a NAT can only
      communicate using sessions that hairpin through the NAT.

   Mapping:  A mapping between an IPv6 transport address and a IPv4
      transport address.  Used to translate the addresses and ports of
      packets flowing between the IPv6 host and the IPv4 host.  In
      NAT64, the IPv4 transport address is always a transport address
      assigned to the NAT64 itself, while the IPv6 transport address
      belongs to some IPv6 host.

GT> What about the term  "Filtering"?

   NAT64:  A device that translates IPv6 packets to IPv4 packets and
      vice-versa, with the provision that the communication must be
      initiated from the IPv6 side.  The translation involves not only
      the IP header, but also the transport header (TCP or UDP).

   Session:  A TCP, UDP or ICMP Query session.  In other words, the bi-
      directional flow of packets between two ports on two different
      hosts.  In NAT64, typically one host is an IPv4 host, and the
      other one is an IPv6 host.

   Session table:  A table of sessions kept by a NAT64.  Each NAT64 has
      three session tables, one for TCP, one for UDP and one for ICMP
      Queries.

   Synthetic RR:  A DNS Resource Record (RR) that is not contained in
      any zone data file, but has been synthesized from other RRs.  An
      example is a synthetic AAAA record created from an A record.





Bagnulo, et al.           Expires May 25, 2010                 [Page 11]

Internet-Draft                    NAT64                    November 2009


   Transport Address:  The combination of an IPv6 or IPv4 address and a
      port.  Typically written as (IP address, port); e.g. (192.0.2.15,
      8001).

   Tuple:  Refers to either a 3-Tuple or a 5-tuple as defined above.

   For a detailed understanding of this document, the reader should also
   be familiar with DNS terminology [RFC1035] and current NAT
   terminology [RFC4787].


3.  NAT64 Normative Specification

   A NAT64 is a device with at least one IPv6 interface and at least one
   IPv4 interface.  Each NAT64 device MUST have one unicast /n IPv6
   prefix assigned to it, denoted Pref64::/n.  Additional consideration
   about the Pref64::/n are presented in Section 3.2.5.  Each NAT64 box
   MUST have one or more unicast IPv4 addresses assigned to it.

   A NAT64 uses the following dynamic data structures:

   o  UDP Binding Information Base

   o  UDP Session Table

   o  TCP Binding Information Base

   o  TCP Session Table

   o  ICMP Query Binding Information Base

   o  ICMP Query Session Table

   A NAT64 has three Binding Information Bases (BIBs): one for TCP, one
   for UDP and one for ICMP Queries.  In the case of UDP and TCP BIBs,
   each BIB entry specifies a mapping between an IPv6 transport address
   and an IPv4 transport address:

      (X',x) <--> (T,t)

GT> This notation is very useful, I wish you used it much more in the
descriptions below...see more on that later.

   where X' is some IPv6 address, T is an IPv4 address, and x and t are
   ports.  T will always be one of the IPv4 addresses assigned to the
   NAT64.  A given IPv6 or IPv4 transport address can appear in at most
   one entry in a BIB: for example, (2001:db8::17, 4) can appear in at
   most one TCP and at most one UDP BIB entry.

GT> Maybe it should be noted that this is what implements/ensure
"Endpoint-Independent mappings"?.

   TCP and UDP have
   separate BIBs because the port number space for TCP and UDP are
   distinct.




Bagnulo, et al.           Expires May 25, 2010                 [Page 12]

Internet-Draft                    NAT64                    November 2009


   In the case of the ICMP Query BIB, each ICMP Query BIB entry specify
   a mapping between an (IPv6 address, Query Identifier) pair and an
   (IPv4 address, Query Identifier pair).

      (X',I1) <--> (T,I2)

   where X' is some IPv6 address, T is an IPv4 address, and I1 and I2
   are Query Identifiers.  T will always be one of the IPv4 addresses
   assigned to the NAT64.  A given (IPv6 or IPv4 address, Query Id) pair
   can appear in at most one entry in the ICMP Query BIB.

   Entries in any of the three BIBs can be created dynamically as the
   result of the flow of packets as described in the section Section 3.2
   but the can also can be created manually by the system administrator.

GT> //can also can/can also

   NAT64 implementations SHOULD support manually configured BIB entries
   for any of the three BIBs.  Dynamically-created entries are deleted
   from the corresponding BIB when the last session associated to the
   BIB entry is removed from the session table.

GT> MUST/SHOULD/MAY??? sounds like a SHOULD to me. or is a MUST if
associated with a max-wait time.

   Manually-configured BIB
   entries are not deleted when there is no corresponding session table
   entry and can only be deleted by the administrator.


GT> All of the text above regarding BIBs, says what BIBs are but it
does not say what BIBs are for. e.g., BIBs implement
"Endpoint-Independent filtering"....other types of filtering can be
implemented with session entries. All session table entries MUST have
corresponding BIB entries. right?

   A NAT64 also has three session tables: one for TCP sessions, one for
   UDP sessions and one for ICMP Query sessions.  Each entry keeps
   information on the state of the corresponding session.  In the TCP
   and UDP session tables, each entry specifies a mapping between a pair
   of IPv6 transport address and a pair of IPv4 transport address:

      (X',x),(Y',y) <--> (T,t),(Z,z)

   where X' and Y' are IPv6 addresses, T and Z are IPv4 addresses, and
   x, y, z and t are ports.  T will always be one of the IPv4 addresses
   assigned to the NAT64.  Y' is always the IPv6 representation of the
   IPv4 address Z, so Y' is obtained from Z using the algorithm applied
   by the NAT64 to create IPv6 representations of IPv4 addresses. y is
   always equal to z.  In addition, each session table entry has a
   lifetime.

GT> Instead of "is" this and "will always be" that, consider using
normative MUSTs.


   In the ICMP query session table, each entry specifies a mapping
   between a 3-tuple of IPv6 source address, IPv6 destination address
   and ICMPv6 Query Id and a 3-tuple of IPv4 source address, IPv4
   destination address and ICMPv4 Query Id:

      (X',Y',I1) <--> (T,Z,I2)

   where X' and Y' are IPv6 addresses, T and Z are IPv4 addresses, and
   I1 and I2 are ICMP query Ids. T will always be one of the IPv4
   addresses assigned to the NAT64.  Y' is always the IPv6
   representation of the IPv4 address Z, so Y' is obtained from Z using



Bagnulo, et al.           Expires May 25, 2010                 [Page 13]

Internet-Draft                    NAT64                    November 2009


   the algorithm applied by the NAT64 to create IPv6 representations of
   IPv4 addresses.  In addition, each session table entry has a
   lifetime.

   The NAT64 uses the session state information to determine when the
   session is completed, and also uses session information for ingress
   filtering.

GT> what kind? Address-Dependent? Address and Port-Dependent
Filtering? both? other?

 A session can be uniquely identified by either an
   incoming tuple or an outgoing tuple.

GT> what is "ingress" in this context?

   For each session, there is a corresponding BIB entry, uniquely
   specified by either the source IPv6 transport address or the source
   IPv6 address and ICMPv6 Query Id (in the IPv6 --> IPv4 direction) or
   the destination IPv4 transport address or the destination IPv4
   address and the ICMPv4 Query Id (in the IPv4 --> IPv6 direction).

GT> It is not clear why the above is ICMP specific and why the BIB
entry can be identified in so many different ways. I am a bit
confused.

   However, a single BIB entry can have multiple corresponding sessions.
   When the last corresponding session is deleted, if the BIB entry was
   dynamically created, the BIB entry is deleted.

GT> this was said also earlier...see if one instance can be removed.

   The processing of an incoming IP packet takes the following steps:

   1.  Determining the incoming tuple

   2.  Filtering and updating binding and session information

   3.  Computing the outgoing tuple

   4.  Translating the packet

   5.  Handling hairpinning

   The details of these steps are specified in the following
   subsections.

   This breakdown of the NAT64 behavior into processing steps is done
   for ease of presentation.  A NAT64 MAY perform the steps in a
   different order, or MAY perform different steps, as long as the
   externally visible outcome is the same.

3.1.  Determining the Incoming tuple

GT> It would be helpful to make clear what "incoming IP packet" the
text refers to. Since no IP version is indicated, I assume that this
section is common to all IPv4, IPv6, and ICMP packets received by the
NAT64 function.


   This step associates a incoming tuple with every incoming IP packet
   for use in subsequent steps.  In the case of TCP, UDP and ICMP error
   packets, the tuple is a 5-tuple consisting of source IP address,
   source port, destination IP address, destination port, transport
   protocol.  In case of ICMP Queries, the tuple is a 3-tuple consisting
   of the source IP address, destination IP address and Query
   Identifier.




Bagnulo, et al.           Expires May 25, 2010                 [Page 14]

Internet-Draft                    NAT64                    November 2009


   If the incoming IP packet contains a complete (un-fragmented) UDP or
   TCP protocol packet, then the 5-tuple is computed by extracting the
   appropriate fields from the packet.

   If the incoming packet is an ICMP query message (i.e. an ICMPv4 Query
   message or an ICMPv6 Informational message), the 3-tuple is the
   source IP address. the destination IP address and the ICMP Query
   Identifier.

   If the incoming IP packet contains a complete (un-fragmented) ICMP
   error message, then the 5-tuple is computed by extracting the
   appropriate fields from the IP packet embedded inside the ICMP error
   message.  However, the role of source and destination is swapped when
   doing this: the embedded source IP address becomes the destination IP
   address in the 5-tuple, the embedded source port becomes the
   destination port in the 5-tuple, etc.  If it is not possible to
   determine the 5-tuple (perhaps because not enough of the embedded
   packet is reproduced inside the ICMP message), then the incoming IP
   packet is silently discarded.

      NOTE: The transport protocol is always one of TCP or UDP, even if
      the IP packet contains an ICMP Error message.

GT> I do not understand this note.

   If the incoming IP packet contains a fragment, then more processing
   may be needed.  This specification leaves open the exact details of
   how a NAT64 handles incoming IP packets containing fragments, and
   simply requires that a NAT64 handle fragments arriving out-of-order.
   A NAT64 MAY elect to queue the fragments as they arrive and translate
   all fragments at the same time.  Alternatively, a NAT64 MAY translate
   the fragments as they arrive, by storing information that allows it
   to compute the 5-tuple for fragments other than the first.  In the
   latter case, the NAT64 will still need to handle the situation where
   subsequent fragments arrive before the first.

GT> Has the option of NAT64 reassembling packets before translating
been discussed at all?
GT> On a different note, this section is supposed to be about
"determining the incoming tuple". I think handling fragments should
only be discussed in as far as it is needed to determine the
tuple....other fragmentation concerns (i.e., most of the rest of the
text) should not be here IMO.

   Implementors of NAT64 should be aware that there are a number of
   well-known attacks against IP fragmentation; see [RFC1858] and
   [RFC3128].

   Assuming it otherwise has sufficient resources, a NAT64 MUST allow
   the fragments to arrive over a time interval of at least 10 seconds.
   A NAT64 MAY require that the UDP, TCP, or ICMP header be completely
   contained within the first fragment.

   Except from the retrieval of 5-tuple information from the incoming
   packets as discussed above, all other fragmentation and PMTUD related
   processing performed by the NAT64 device is performed as defined in
   [I-D.ietf-behave-v6v4-xlate], including the translation of all
   related fragmentation fields in the IP header, the determination of



Bagnulo, et al.           Expires May 25, 2010                 [Page 15]

Internet-Draft                    NAT64                    November 2009


   the outgoing packet size, the fragmentation of outgoing packets and
   the generation and processing of ICMP Packet Too Big errors.

   In particular, the NAT64 must generate ICMP Packet Too Big error
   messages in the case the outgoing packet does not fit in the outgoing
   MTU and needs to be discarded.

GT> Is this correct? Section 2.1 of [I-D.ietf-behave-v6v4-xlate]
indicates that IPv4 packets can be fragmented further to fit in IPv6
packets... why don't we just stay silent on the issue and just point
to [I-D.ietf-behave-v6v4-xlate]?

3.2.  Filtering and Updating Binding and Session Information

   This step updates binding and session information stored in the
   appropriate tables.  This step may also filter incoming packets, if
   desired.

   Irrespectively of the transport protocol used, the NAT64 must
   silently discard all incoming IPv6 packets containing a source
   address that contains the Pref64::/n.  This is required in order to
   prevent hairpinning loops as described in the Security Considerations
   section.

   The details of this step depend on the protocol (UDP TCP or ICMP
   Query).

3.2.1.  UDP Session Handling

   The state information stored for a UDP session in the UDP session
   table includes a timer that tracks the remaining lifetime of the UDP
   session.  The NAT64 decrements this timer at regular intervals.  When
   the timer expires, the UDP session is deleted.  If all the UDP
   sessions corresponding to a UDP BIB entry are deleted, then the UDP
   BIB entry is also deleted (only applies to the case of dynamically
   created entries).

GT> I commented on this earlier so align any changes you make earlier with this.

   An IPv6 incoming packet is processed as follows:

GT> Presumably such a packet is only processed if the destination
address is from Pref64::/n? It would help to state this.

      The NAT64 searches for a UDP BIB entry that matches the IPv6
      source transport address.  If such entry does not exists, a new
      entry is created.  As IPv6 address, the source IPv6 transport
      address of the packet is included and an IPv4 transport address
      allocated using the rules defined in Section 3.2.3 is included as
      IPv4 address.

GT> The last sentence does not read well. It would help to be more
exact in this and subsequent descriptions. For example: the new BIB
entry has the form of  (Y*,y)<->(T,t), where Y* is set to the source
IPv6 address of the packet, y is set to the source port number of the
packet, T is one of the NAT64's IPv4 addresses used for translation
and t is set to an unused port number on the selected IPv4 address T.

      The NAT64 searches for the session table entry corresponding to
      the incoming 5-tuple.

GT> This search, I assume is only meaningful if there was already a
BIB entry? If the BIB entry was just created based on the last
paragraph, there is no point in searching the session table, right?

      If no such entry is found, a new entry is
      created.  The information included in the session table is as
      follows: the IPv6 transport source and destination addresses
      contained in the received IPv6 packet, the IPv4 transport source
      address is extracted from the corresponding UDP BIB entry and the
      IPv4 transport destination address contains the same port as the



Bagnulo, et al.           Expires May 25, 2010                 [Page 16]

Internet-Draft                    NAT64                    November 2009


      IPv6 destination transport address and the IPv4 address that is
      algorithmically generated from the IPv6 destination address using
      the reverse algorithm as specified in Section 3.2.5.

GT> Again, reading this makes my head hurt. I think this could be much
clearer if you use the (X',x),(Y',y) <--> (T,t),(Z,z) and say how
these values are set based on the incoming packet.

      The NAT64 sets or resets the timer in the session table entry to
      maximum session lifetime.  By default, the maximum session
      lifetime is 5 minutes, but for specific destination ports in the
      Well-Known port range (0..1023), the NAT64 MAY use a smaller
      maximum lifetime.

GT> It would be nice to indicate why a smaller mac lifetime is suggested here.

  The packet is translated and forwarded as
      described in the following sections.

   An IPv4 incoming packet is processed as follows:

      The NAT64 searches for a UDP BIB entry that matches the IPv4
      destination transport address.

GT> More accurately: Given the BIB entries are of the form
(Y*,y)<->(T,t), NAT64 attempts to match the destination IPv4 transport
address to the (T,t) part of the BIB entires.

     If such entry does not exists, the
      packet is dropped. An ICMP message MAY be sent to the original
      sender of the packet, unless the discarded packet is itself an
      ICMP message.  The ICMP message, if sent, has a type of 3
      (Destination Unreachable).

      If the NAT64 filters on its IPv4 interface,

GT> You have to define what this means. if I understand correctly
below you describes "Address and Port-Dependent Filtering" or is it
"Address-Dependent Filtering"? it is hard to tell given that you try
to match "addresses" with "transport addresses" which makes no sense.

then the NAT64 checks
      to see if the incoming packet is allowed according to the address-
      dependent filtering rule.  To do this, it searches for a session
      table entry with a source IPv4 transport address equal to the
      destination IPv4 transport address in the incoming 5-tuple and
      destination IPv4 address (in the session table entry) equal to the
      source IPv4 address in the incoming 5-tuple.

GT> This is really hard to parse.  If you use the (X',x),(Y',y) <-->
(T,t),(Z,z) notation it all becomes much clearer. For example, here
the NAT64 could implement "Address and Port-Dependent Filtering" by
trying to match the destination IPv4 transport address in the packet
with the (T,t) part, and the source IPv4 transport address of the
packet with the (Z,z)  part, of the entries in the session table. Is
this what you mean? If instead you do "Address-Dependent Filtering"
you simply drop the t and z from the above.

If such an entry is
      found (there may be more than one), packet processing continues.

GT> I do not get how there can be more than one of these.

      Otherwise, the packet is discarded.  If the packet is discarded,
      then an ICMP message MAY be sent to the original sender of the
      packet, unless the discarded packet is itself an ICMP message.
      The ICMP message, if sent, has a type of 3 (Destination
      Unreachable) and a code of 13 (Communication Administratively
      Prohibited).

GT> Now it gets really confusing. What follows is clearly not a
continuation from the above. It looks like its a case where a session
is allowed to be initiated from the IPv4 side; if that is the case
then it should be clearly stated! Also does the following describe
processing when NAT64 does NOT filter at all on IPv4? or when it does
"Endpoint-Independent Filtering"? or " Address-Dependent Filtering"?
It seems that the description below only assumes a matching BIB entry
so my guess is that the following is based on "Endpoint-Independent
Filtering"

      The NAT64 searches for the session table entry corresponding to
      the incoming 5-tuple.

GT> Again you should use the  (X',x),(Y',y) <--> (T,t),(Z,z)  notation
to make clear what "searching" means above, as well as to define how
the new entry is "created" below.

      If no such entry is found, a new entry is
      created.  The UDP session table entry contains the transport
      source and destination address contained in the IPv4 packet and
      the source IPv6 transport address (in the IPv6 --> IPv4 direction)
      contained in the existing UDP BIB entry.  The destination IPv6
      transport address contains the same port than the destination IPv4
      transport address and the IPv6 representation of the IPv4 address
      of the destination IPv4 transport address, generated using the
      algorithm described in Section 3.2.5.

      The NAT64 sets or resets the timer in the session table entry to
      maximum session lifetime.  By default, the maximum session



Bagnulo, et al.           Expires May 25, 2010                 [Page 17]

Internet-Draft                    NAT64                    November 2009


      lifetime is 5 minutes, but for specific destination ports in the
      Well-Known port range (0..1023), the NAT64 MAY use a smaller
      maximum lifetime.


GT> OK...I am stopping here...lets see if any of the above makes sense
to people and then I will continue.