[sipcore] Benjamin Kaduk's Discuss on draft-ietf-sipcore-sip-token-authnz-13: (with DISCUSS and COMMENT)

[Benjamin Kaduk via Datatracker <[e-mail]>]

Benjamin Kaduk has entered the following ballot position for
draft-ietf-sipcore-sip-token-authnz-13: Discuss

When responding, please keep the subject line intact and reply to all
email addresses included in the To and CC lines. (Feel free to cut this
introductory paragraph, however.)


Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
for more information about IESG DISCUSS and COMMENT positions.


The document, along with other ballot positions, can be found here:
https://datatracker.ietf.org/doc/draft-ietf-sipcore-sip-token-authnz/



----------------------------------------------------------------------
DISCUSS:
----------------------------------------------------------------------

I support Roman's Discuss.


The "Bearer" authentication challenge includes the address (or name?) of an
authorization server to contact to obtain tokens, as mentioned in multiple
places in the document (noted in the COMMENT section).  Our experience in
the OAuth world has shown that several classes of vulnerabilities are
possible when the client blindly attempt to use any provided AS, and that a
whitelist of "allowed" or "trusted" ASes is needed for secure operation.  I
believe that the same is true for the SIP usage, and we should mention this
requirement explicitly.


Section 1.2 tries to apply the OAuth confidential/public client distinction
to SIP UACs, but it does so in a non-analogous fashion: the OAuth
distinction is for the client's ability to protect credentials that identify
the client itself; the usage in this document refers to protecting *user*
credentials and obtained tokens.  I don't think that it's appropriate to
invoke the OAuth terminology when using it for a different meaning.
Both Public and Confidential OAuth clients are capable of providing the
necessary protections for *user* credentials (though they are of course not
guaranteed to do so), which leaves me unclear on what the intended
requirements actually are.


Section 2.3 states that:

   When a proxy wishes to authenticate a received request, it MUST
   search the request for Proxy-Authorization header fields with 'realm'
   parameters that match its realm.  It then MUST successfully validate

https://tools.ietf.org/html/rfc7235#section-4.4 suggests that it is not
expected to have a sequence or list of Proxy-Authorization header fields
present in a single request that are intended to be interpreted by different
proxies.  Is this text compatible with that part of RFC 7235?  Furthermore,
I didn't find much guidance in 7235 or 3261 about when to include the
"realm" parameter in Proxy-Authorization; do we want to give any guidance
here?  (That is to say, I almost didn't find where it was even defined as
possible to do so...)


I'm also not sure if we're attempting to profile RFC 6749 and always require
a refresh token to be issued, or just have some editorial tweaks to make to
avoid suggesting that we do have such a requirement (noted in the COMMENT).


----------------------------------------------------------------------
COMMENT:
----------------------------------------------------------------------

Section 1

   The OpenID Connect 1.0 specification [OPENID] defines a simple
   identity layer on top of the OAuth 2.0 protocol, which enables
   clients to verify the identity of the user based on the

Please make it clear that these are OAuth/OIDC clients, not SIP clients.

Section 1.3

OAuth 2.0 doesn't require the issuance of a Refresh token, but the
discussion here implies that for the SIP case there will always be a refresh
token.  If we're profiling 6749 in this manner, we should be more explicit
about the requirement to issue refresh tokens.

   *  ID Token: this token contains the SIP URI and other user-specific
      details that will be consumed by the UAC.

nit: I don't think we should use the definite article in "the SIP URI" since
we haven't discussed any such SIP URI usage yet.  That is, we should say
what it's used for, e.g., "a SIP URI associated with the user".

   *  Structured Token: a token that consists of a structured object
      that contains the claims associated with the token, e.g.  JWT as
      defined in [RFC7519].

nit: comma after "e.g.".

   *  Reference Token: a token that consists of a random string that is
      used to obtain the details of the token and its associated claims,
      as defined in [RFC6749].

It doesn't technically have to be random (though in practice it should
contain signficant random elements); "opaque" might be better wording.

Section 1.4.1

Perhaps Figure 1 could include some indication that steps 5 and 6 are
optional/do not always occur (in the case when the access token is a
self-contained JWT)?

   In step [2], the registrar challenges the UA, by sending a SIP 401
   (Unauthorized) response to the REGISTER request.  In the response,
   the registrar includes information about the AS to contact in order
   to obtain a token.

The UAC needs to have a preconfigured whitelist of acceptable ASes in order
to avoid directing the user's credentials to malicious sites.

   The registrar validates the access token.  If the access token is a
   reference token, the registrar MAY perform an introspection
   [RFC7662], as in steps [5] and [6], in order to obtain more
   information about the access token and its scope, per [RFC7662].

I think we could tighten up the normative language a bit here.
In particular, the registrar MUST validate the token in some fashion; for
reference tokens, the main ways to do that are either inspection or
(essentially) being the party that issued the token in the first place.

   Otherwise, after the registrar validates the token to make sure it
   was signed by a trusted entity, it inspects its claims and acts upon
   it.

I think we can also be more specific than "a trusted entity" -- in this
flow, it looks like the registrar should know exactly which entity should
have signed the token in question (for the user in question), and should not
accept a signed token from a different entity that happens to be trusted to
issue other tokens.

Section 1.4.2

(Similarly, Figure 2 could note that steps 3 and 4 do not always occur, and
the text about token validation should remain parallel between examples.)

Section 2

I note that RFC 3261 explicitly disclaims any definition of authorization
systems, which is not true for this document, given that OAuth is an
authorization system :)

Section 2.1.1

   Required) response with a Proxy-Authenticate header field.  If the
   WWW-Authenticate or Proxy-Authenticate header field indicates
   "Bearer" scheme authentication and contains an address to an
   authorization server, the UAC contacts the authorization server in
   order to obtain tokens, and includes the requested scopes, based on a
   local configuration (Figure 1).

[whitelist of allowed ASes, again]

   The detailed OAuth2 procedure to authenticate the user and obtain
   these tokens is out of scope of this document.  The address of the
   authorization server might already be known to the UAC via
   configuration.  In which case, the UAC can contact the authorization
   server for tokens before it sends a SIP request (Figure 2).

nit: I think that "in which case" functions as a conjunction, which makes
this an incomplete sentence.

   The tokens returned to the UAC depend on the type of authorization
   server (AS): an OAuth AS provides an access token and refresh token
   [RFC6749].  The UAC provides the access token to the SIP servers to

Again, this implies that refresh tokens are always issued; RFC 6749 is quite
clear that "issuing a refresh token is optional at the discretion of the
authorization server".

   authorize UAC's access to the service.  The UAC uses the refresh
   token only with the AS to get a new access token and refresh token
   before the expiry of the current access token (see [RFC6749], section

(New refresh token is also optional.)

   1.5 Refresh Token for details).  An OpenID Connect server returns an
   additional ID-Token containing the SIP URI and other user-specific
   details that will be consumed by the UAC.

[same comment as above about "the SIP URI".]

   If the UAC receives a 401/407 response with multiple WWW-
   Authenticate/Proxy-Authenticate header fields, providing challenges
   using different authentication schemes for the same realm, the UAC
   selects one or more of the provided schemes (based on local policy)
   and provides credentials for those schemes.

RFC 3261 seems to be written in a world that assumed that Basic and Digest
are the only defined HTTP authentication schemes, and since it prohibits
Basic, that there would only be one authentication challenge (type)
possible.  This text righly acknowledges that we are opening the field up
and could have cases where multiple authentication schemes are possible;
however, it goes even further and admits the possibility of using them
simultaneously.  It seems like this places an onus on us to give some
indication of the semantics when multiple schemes are used at the same time.
Do the authenticated identities need to match?  Or are we asserting that
both/all identities are consenting to the request in question?  (If we don't
have a concrete use case in mind, perhaps the "or more" is just opening a
box that we don't need to open right now.)

Section 2.1.2

   access to it.  Endpoints that support this specification MUST support
   encrypted JSON Web Tokens (JWT) [RFC7519] for encoding and protecting
   access tokens when they are included in SIP requests, unless some
   other mechanism is used to guarantee that only authorized SIP
   endpoints have access to the access token.

I think we may want "MUST support" (always) and "MUST use, unless some other
mechanism".

I'd also suggest a quick note that TLS remains fine for protecting the
UAC/AS interaction where the refresh token is used.

Whatever is done, please harmonize with Section 5 that has essentially the
same text.

Section 2.1.3

   Based on local policy, the UAC MAY include an access token that has
   been used for another binding associated with the same Address Of
   Record (AOR) in the request.

Just to check: there's no real privacy considerations with such use, as its
the same parties interacting and there are other identifiers (e.g., IP
addresses) to confirm that it's the same client communicating to the server?

Also, is it clear what will be done (new token request) when the MAY is not
heeded?

Section 2.1.4

The text here just talks about "a valid access token" and similar, without
saying a whole lot about it being related in any way to the specifics of the
authentication challenge.  Is there something useful to say about, e.g., the
token in question having been issued by the authorization server indicated
in the challenge?

Section 2.2

   authorization credentials acceptable to it, it SHOULD challenge the
   request by sending a 401 (Unauthorized) response.  To indicate that
   it is willing to accept an access token as a credential, the UAS/
   Registrar MUST include a Proxy-Authentication header field in the
   response that indicates "Bearer" scheme and includes an address of an

nit(?): I'd consider just making this a declarative statement, a la "the
UAS/registrar includes a Proxy-Authentication header field with the "Bearer"
scheme to indicate that it is willing to accept an access token as a
credential"  (but that wording is incomplete and would need to be fleshed
out a bit more).

   authorization server from which the originator can obtain an access
   token.

nit: "address of" an AS, does that mean bare IP address only or is a DNS
name okay?

   [RFC7519].  If the token provided is an expired access token, then
   the UAS MUST reply with 401 Unauthorized, as defined in section 3 of
   [RFC6750].  If the validation is successful, the UAS/Registrar can
   continue to process the request using normal SIP procedures.  If the
   validation fails, the UAS/Registrar MUST reject the request.

Is "expired" the only case that should get a 401 vs. outright rejection, as
stated here?

Section 2.3

   sending a 407 (Proxy Authentication Required) response.  To indicate
   that it is willing to accept an access token as a credential, the
   proxy MUST include a Proxy-Authentication header field in the
   response that indicates "Bearer" scheme and includes an address to an

[same comment as above about normative vs. declarative statement]
Also, please keep the text parallel between sections -- this copy has at
least one nit ("address to an" vs. "address of an").

Section 3

   If an access token is encoded as a JWT, it might contain a list of
   claims [RFC7519], some registered and some application-specific.  The

I don't think there's a question of whether a JWT will contain a list of
claims :)
Maybe "If the access token is encoded as a JWT, it will contain a list of
claims [RFC7519], which may include both registered and application-specific
claims"?

Section 4

This section claims to cover WWW-Authenticate.  What specification will the
SIP use of Bearer for Authorization operate under?

       challenge  =/  ("Bearer" LWS bearer-cln *(COMMA bearer-cln))

side note: is there a mnemonic for the "cln" in "bearer-cln"?

       bearer-cln = realm / scope / authz-server / error /
                    auth-param

nit: realm is already included in auth-param, though I don't see a harm in
calling it out explicitly.

       realm = <defined in RFC3261>
       auth-param = <defined in RFC3261>

RFC 3261 defers to RFC 2617 (now obsoleted by 7235) for both of these; we
could perhaps short-circuit the chain and say "defined in RFC 7235".

Section 5

We should probably note that SIP makes much heavier use of proxies than is
common in the web world of OAuth.

I'm happy to see that some privacy considerations are being added in
response to Barry's review.

Section 8

I think RFCs 8252 and 8414 could be just informative.