Re: [tram] Secdir last call review of draft-ietf-tram-turnbis-27

[Benjamin Kaduk <[e-mail]>]

Hi Tiru,

Thanks for all the updates, including further downthread with Chris.
Just a couple notes inline...

On Wed, Jul 10, 2019 at 12:40:37PM +0000, Konda, Tirumaleswar Reddy wrote:
> Hi Ben,
> 
> Thanks for the review. Please see inline.
> 
> > -----Original Message-----
> > From: Benjamin Kaduk <[e-mail]>
> > Sent: Wednesday, July 10, 2019 2:01 AM
> > To: Konda, Tirumaleswar Reddy <[e-mail]>
> > Cc: Christopher Wood <[e-mail]>; [e-mail]; draft-ietf-
> > [e-mail]; [e-mail]; [e-mail]
> > Subject: Re: [tram] Secdir last call review of draft-ietf-tram-turnbis-27
> > 
> > This email originated from outside of the organization. Do not click links or
> > open attachments unless you recognize the sender and know the content is
> > safe.
> > 
> > Chris, thanks for the review.  Some more questions/comments inline as I
> > prepare to ballot on this document...
> > 
> > On Mon, Jul 08, 2019 at 02:18:26PM +0000, Konda, Tirumaleswar Reddy wrote:
> > > Hi Christopher,
> > >
> > > Thanks for the detailed review. Please see inline
> > >
> > > > -----Original Message-----
> > > > From: tram <[e-mail]> On Behalf Of Christopher Wood via
> > > > Datatracker
> > > > Sent: Monday, July 8, 2019 1:59 AM
> > > > To: [e-mail]
> > > > Cc: [e-mail]; [e-mail];
> > > > [e-mail]
> > > > Subject: [tram] Secdir last call review of
> > > > draft-ietf-tram-turnbis-27
> > > >
> > > > This email originated from outside of the organization. Do not click
> > > > links or open attachments unless you recognize the sender and know
> > > > the content is safe.
> > > >
> > > > Reviewer: Christopher Wood
> > > > Review result: Has Nits
> > > >
> > > > I have reviewed this document as part of the security directorate's
> > > > ongoing effort to review all IETF documents being processed by the
> > > > IESG. These comments were written primarily for the benefit of the
> > > > security area directors.
> > > > Document editors and WG chairs should treat these comments just like
> > > > any other last call comments.
> > > >
> > > > The summary of the review is: Ready with nits
> > > >
> > > > Summary:
> > > >
> > > > In general, the document is well written and clearly founded in
> > > > operational experience. The security considerations are thorough,
> > > > providing examples where necessary to highlight important problem
> > > > areas. It draws a clear line between issues best addressed by
> > > > applications outside of TURN, and provides sufficient detail for
> > > > those issues in scope. My comments and questions are, hopefully, mostly
> > nits.
> > > >
> > > > General comments:
> > > >
> > > > - TURN server authentication in the case of (D)TLS is
> > > > underspecified. Are servers assumed to have WebPKI certificates,
> > > > OOB-trusted raw public keys, or something else? Is there a preferred
> > form of authentication?
> > > > Relatedly, in
> > > > Section 3.2, how do clients authenticate the server?
> > >
> > > Server certificate validation is discussed in https://tools.ietf.org/html/draft-
> > ietf-tram-stunbis-21#section-6.2.3.  I have modified the text as follows:
> > >
> > > If (D)TLS transport is used between the TURN client and the TURN
> > > server, the cipher suites, server certificate validation and
> > > authentication of TURN server are discussed in Section 6.2.3 of
> > > [I-D.ietf-tram-stunbis]
> > 
> > That helps, but it would still be nice to give some indication of what certificate
> > PKI(s) are expected to be used.  Is anything other than the Web PKI under
> > consideration ?
> 
> No. Please see https://tools.ietf.org/html/draft-ietf-tram-stunbis-21#section-6.2.3, It discusses to verify the certificate path and identity of the server.
> 
> > 
> > > >- Section 3.7: Could
> > > > TURN servers not chunk data from stream-oriented transports (TCP or
> > > >TLS)  to a preferred MTU size before relaying to peers?
> > > > Specifically, there are likely
> > > > some cases wherein the server could deal with the client data
> > > >messages  larger than the recommended 500B limit. On that point,
> > > >should servers even  accept data larger than this recommended size ?
> > > >-
> > >
> > > Please see
> > > https://tools.ietf.org/html/draft-ietf-tram-turnbis-27#section-15 for
> > > TCP-to-UDP relay. If the PMTU is not known, and on legacy or otherwise
> > > unusual networks,
> > > 500 byte limit for application data is recommended.
> > >
> > > > Section 3.9: There may be
> > > > cases where the TLS connection post TCP connection establishment
> > > > fails. It would therefore seems prudent to not declare victory for
> > > > one connection over the other until TLS connects, too. -
> > >
> > > Agreed, Eric (as part of ISEG review) suggested to simplify the text. I have
> > updated the text as follows:
> > >
> > >    o  For TCP or TLS-over-TCP, the results of the Happy Eyeballs
> > >       procedure [RFC8305] are used by the TURN client for sending its
> > >       TURN messages to the server.
> > 
> > Is the editor's copy available for viewing somewhere?  It would be good to
> > see this (and other changes) in context.
> 
> Yes, Please see https://github.com/tireddy2/TURNbis 
> 
> > 
> > > > Section 3 could benefit from a
> > > > subsection on replays and the nonce role. In particular, as later
> > > > discussed in the security considerations, some of these attacks are
> > > > not new to TURN and should therefore be dealt with by the
> > > > application protocol (SRTP). This section might also describe nonce
> > > > rotation policies with more specificity, as this is underspecified in the
> > document.
> > >
> > > It is discussed in Section 5 in the specification, the server should expire the
> > nonce at least once every hour during the lifetime of the allocation.
> > >
> > > >- Section 3 could also benefit from
> > > > discussion about cleartext versus encryption transports between
> > > >clients and  servers.
> > >
> > > It is discussed in https://tools.ietf.org/html/draft-ietf-tram-turnbis-
> > 27#section-21.1.6.
> > 
> > There's not much discussion about potential information  leakage via (e.g.)
> > USERNAME/USERHASH, and really just a claim that the peer address is the
> > "primary protocol content".
> 
> Agreed, added the following paragraph :
> 
> If the TURN client and server use the STUN Extension for Third-Party Authorization [RFC7635], the username does not reveal the
> real user's identity, the USERNAME attribute carries an ephemeral and unique key identifier, and the key identifier can change per call.
> If the TURN client and server use the STUN long-term credential mechanism and the username reveals the
> real user's identity, the client need to either use (D)TLS transport between the client and the server or use 
> the USERHASH attribute instead of the USERNAME attribute to anonynmize the username.
> 
> > 
> > > > Encrypting the nonce, username, realm, etc., among other things, has
> > > > obvious benefits. - Why are SOFTWARE and FINGERPRINT attributes
> > > > recommended?  It seems like clients should be discouraged from
> > > > sending these if anything, especially if not used to make allocation
> > > > decisions on the server.
> > >
> > > Username may not be the user identity, Please see
> > https://tools.ietf.org/html/rfc7635), It could be an ephemeral and unique
> > key identifier. Further, username can be anonymized (please see
> > https://tools.ietf.org/html/draft-ietf-tram-stunbis-21#section-14.4).
> > > SOFTWARE and FINGERPRINT attributes are defined in the STUN
> > specification, see https://tools.ietf.org/html/draft-ietf-tram-stunbis-21.
> > 
> > These mechanisms  are partial mitigations but can still be susceptible to cross-
> > connection correlation attacks.
> 
> The use of FINGERPRINT is not mandatory, the TURN client and server may include the FINGERPRINT attribute.  Note that ICE (https://tools.ietf.org/html/rfc8445) mandates the use of FINGERPRINT attribute. Added the following line for SOFTWARE attribute (TURNbis is using the same behavior defined in RFC5766 to use SOFTWARE attribute)  :
> 
> SOFWATE attribute can reveal the specific software version of the TURN client and server to eavesdropper and it might possibly allow attacks against vulnerable software that is known to contain security holes. If it is important to prevent an eavesdropper from learning the software version, TURN can be run over (D)TLS.

The later suggestion was:

% We want to avoid double encryption of application data, and SOFTWARE attribute has  
% been sent in clear text all these years in the field, and no attack has been        
% detected based on SOFTWARE field. I can revise the text as follows:                 
% If the software version is known to contain security vulnerabilities, TURN SHOULD be
% run over (D)TLS to prevent leaking the SOFTWARE attribute in clear text.

I think some of the concern is over the risk of so-called "zero-day"
exploits, where the attacker knows of a version-specific flaw but the
defenders do not (and have zero days notice to protect against it).  Such a
concern would suggest to avoid sending SOFTWARE over enencrypted transport
regardless of the state of known vulnerabilities, but in the end it is a
matter of local policy.


> 
> > 
> > > - Section 5: When servers receive data that exceeds an allocation’s
> > > > bandwidth quota, servers silently discard this data. This means
> > > > there’s no allocation-based flow control mechanism between client
> > > > and server beyond what’s provided by the transport protocol, right?
> > >
> > > Yes, UDP does not include a congestion control mechanism.
> > >
> > > > This seems fine, though
> > > > perhaps some discussion of why this design decision was made would
> > > > be helpful. For example, I could imagine explicit signals from
> > > > servers to clients that indicate server state would reveal
> > > > information about other allocations on the server, which might be
> > > > problematic. -
> > >
> > > The errors 486 (Allocation Quota Exceeded) or 508 (Insufficient Capacity)
> > do not reveal information of which other users are using the TURN server.
> > 
> > At least the latter seems to give some indication that many other users are
> > using the server at the  time (so that it's overloaded), though not information
> > about *which* users that is.
> 
> Yes, similar to 503 HTTP error response.
> 
> > 
> > > > Section 7.2 suggests that
> > > > servers can redirect client allocation requests to other servers.
> > > > Can this create a loop, wherein two TURN servers redirect to one another?
> > >
> > > The client detect and stop the loop, it is similar to HTTP redirection.
> > 
> > Where is this specified?
> 
> It is specified in the last paragraph of https://tools.ietf.org/html/draft-ietf-tram-stunbis-21#section-10 
> 
> <snip>
> If the client has been redirected to a
> server to which it has already sent this request within the last five
> minutes, it MUST ignore the redirection and consider the transaction
> to have failed.  This prevents infinite ping-ponging between servers
> in case of redirection loops.
> </snip>

Thanks for the pointer; as you saw in my ballot position I was quite
pressed for time this week and only was able to look at the diff.

-Ben