Re: [radext] OPS-DIR review of draft-ietf-radext-ip-port-radius-ext-11

[<[e-mail]>]

Hi Tim,

(Apologies for the delay to answer the message)

Thank you for the review.

Please see inline.

Cheers,
Med

De : Tim Chown [mailto:[e-mail]]
Envoyé : mercredi 17 août 2016 12:23
À : [e-mail]; [e-mail]
Objet : OPS-DIR review of draft-ietf-radext-ip-port-radius-ext-11

Hi,

I have reviewed this document as part of the Operational directorate's
ongoing effort to review all IETF documents being processed by the IESG.  These
comments were written with the intent of improving the operational aspects of the
IETF drafts. Comments that are not addressed in last call may be included in AD reviews
during the IESG review.  Document editors and WG chairs should treat these comments
just like any other last call comments.

This draft defines three new RADIUS attributes to be used when communicating with a
RADIUS server to facilitate the configuration or reporting of IP and port ranges used
with a network appliance, typically a CGN, where there is a need to constrain
the ports available per customer where IP address sharing is in use.

The three RADIUS attributes are:
IP-Port-Limit-Info - defines the maximum number of ports available
IP-Port-Range - the specific range of port numbers available
IP-Port-Forwarding-Map - to configure port forwarding on a NAT/CGN device

I would consider the document to be "Ready with Issues".

I have some general comments, followed by some specific comments. Note that while I
am familiar with RADIUS (from an eduroam context) the draft is not one I was
familiar with or followed prior to this review. Thus these comments may have already
been addressed.

General comments:

There are at least two areas in which this document has "creep". One is that it is
providing an alternative method to PCP to define port forwarding mappings on a device.
So there is an open question as to whether PCP should be the method of choice for
this function, or whether we wish to create a new way to establish such mappings.

[Med] PCP targets a different deployment model where an application controls its mapping in a CPE, CGN, Firewall, etc. This document targets a deployment where a AAA server is responsible de instructing NAT mappings for a given subscriber.

Secondly, two of the new attributes support inclusion of a new TLV, IP-Port-Local-Id,
which allows user/device-specific information to be transmitted via RADIUS, such as
MAC address or VLAN ID. While this is intended to allow differentiation of users for
accounting/identification r, in doing so it adds an additional potential privacy
concern into a new RADIUS attribute, depending on specific use cases of the TLV.

[Med] The main use case is the deployment where the internal IP address is not sufficient to demux users. We introduced this NEW text in -12:


   The primary issue addressed by this TLV is that there are CGN

   deployments that do not distinguish internal hosts by their internal

   IP address alone, but use further identifiers for unique subscriber

   identification.  For example, this is the case if a CGN supports

   overlapping private or shared IP address spaces (refer to [RFC1918<https://tools.ietf.org/html/rfc1918>]

   and [RFC6598<https://tools.ietf.org/html/rfc6598>]) for internal hosts of different subscribers.  In such

   cases, different internal hosts are identified and mapped at the CGN

   by their IP address and/or another identifier, for example, the

   identifier of a tunnel between the CGN and the subscriber.  In these

   scenarios (and similar ones), the internal IP address is not

   sufficient to demultiplex connections from internal hosts.  An

   additional identifier needs to be present in the IP-Port-Range

   Attribute and IP-Port-Forwarding-Mapping Attribute in order to

   uniquely identify an internal host.  The IP-Port-Local-Id TLV is used

   to carry this identifier.


This is not discussed in the Security Considerations section, but probably should be.
[Med] Fair point. We added this text:


   The IP-Port-Local-Id TLV includes an identifier of which the type and

   length is deployment and implementation dependent.  This identifier

   might carry privacy sensitive information.  It is therefore

   RECOMMENDED to utilize identifiers that do not have such privacy

   concerns.

I note the new attributes use a number of IPFIX information elements; has the draft
considered its relationship to draft-ietf-behave-ipfix-nat-logging-09, which says
the "lack of a consistent way to log the data makes it difficult to write the
collector applications that would receive this data and process it to present useful
information"? This draft is introducing a new method to log such elements; is this
a concern at all?

[Med] FWIW, one of the co-authors is the main editor of draft-ietf-behave-ipfix-nat-logging. As a reminder, this draft reuses the IPFIX registry because it offers a rich set of elements. Further, this draft does not focues on NAT logging. Of course, some TLV can be used for notification purposes but this is not logging per se.

The examples of use cases of the new attributes include both NAT44 devices and CGNs.
The document could state more clearly the address sharing scenarios, perhaps with a
simplified network element diagram for each example, showing the user/host, CPE/NAT44,
and NAT444/CGN?

[Med] I'm sure if this will add more clarity to the draft because there are many deployment cases that can be cited. The draft explicitly calls out the CPE case and CGN.

Some additional clarity here would be useful (see also comments below).
Also, the term "the user" is used in many places in the document where in practice
"the customer's CPE" would be more appropriate.

[Med] I agree that some occurrences can be replaced easily by subscriber or customer but still "user" is accurate. For example, it is not a customer that initiates a connection but a user (that may be the customer itself or another person that uses the right to invoke the service). In order to avoid overloading the document with such subtleties, we prefer to use "user" in the whole document.


Specific comments:

NAT64 is mentioned as a use case at the start, but no example is given later in the
document. This might add useful value.

[Med] Happily! We will update the text accordingly.

In Sections 3.2.6, 3.2.7, 3.2.9 and 3.2.10, the IPFIX information elements in the
TLV are 16 bit values, but 32 bits are reserved for the element. Similarly the
NatEvent element is 8-bit, but has 32 bits reserved. It would be useful if the
document stated why these elements are being padded out to 32 bits.

[Med] Please check https://tools.ietf.org/html/rfc6158#appendix-A.2.1

In Section 4.1.1, I don't think NAT64 is specifically designed to multiplex users
over a smaller number of shared IPv4 addresses, rather its primary design goal
is to facilitate access to legacy IPv4 content from IPv6-only networks.  The
text should be clarified.


[Med] I guess you are referring to "over a smaller number of shared IPv4 addresses". "Smaller" is used here to indicate that the external IP address pool is not sufficient to assign a full IPv4 address to each internal host. NAT64 falls into such category. RFC6146 says: "One or more public IPv4 addresses assigned to a NAT64

   translator are shared among several IPv6-only clients."

Also in 4.1.1, do users really have service agreements that state port limits?
[Med] There is no assumption in the draft that SLAs include port limits.

If they do, I doubt users are aware of them (or care...), and the issue is beyond
the scope of this document.
[Med] Agree.

In 4.1.2, I think you mean "block", not "bulk"?
And the comment on "randomization" might fit better in the Security Considerations
section if you discuss privacy there (which is presumably what you mean?)

[Med] Can be added, indeed.

Also in 4.1.2 you discuss the scenario as if it's CGN, but the flow diagram shows
only the NAT44 (presumably in the CPE) and not an ISP CGN.

[Med] No, the diagram shows a NAT that is collocated with a BNG. This is a CGN case.

The same happens in 4.1.3; discussion of CGN and NAT44 interchangeably, without
the diagram showing there may (presumably) be mappings to establish at both the
user's CPE and the ISP's CGN.

[Med] Idem as above.

And in 4.1.4 the example talks of NAT44 for Joe's CPE, but then also about a CGN
allocating more ports; is that at the NAT44, or at the CGN?

[Med] The intent is CGN here. The text can be clarified.

(These specific NAT44/CGN comments are examples of the general comment I made earlier.)

In Section 5, I found the format of the table with 0 and 0+ a little unintuitive.
[Med] This is not specific to this document. We are following the format in https://tools.ietf.org/html/rfc2865#section-5.44. You can see, e.g.,:

·         https://tools.ietf.org/html/rfc6911#section-3.6

·         https://tools.ietf.org/html/rfc7268#section-3

·         ...


--

Tim