[Gen-art] Gen-art Telechat review of draft-ietf-nfsv4-rpcsec-gssv3-15

[Elwyn Davies <[e-mail]>]

I am the assigned Gen-ART reviewer for this draft. The General Area
Review Team (Gen-ART) reviews all IETF documents being processed
by the IESG for the IETF Chair. Please wait for direction from your
document shepherd or AD before posting a new version of the draft.

For more information, please see the FAQ at

<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.

Document: draft-ietf-nfsv4-rpcsec-gssv3-15.txt
Reviewer: Elwyn Davies
Review Date: 2016/01/16
IETF LC End Date: 2015/12/09
IESG Telechat date: 2016/01/21

Summary: Almost ready.  Thank you for addressing my comments from the 
last call review.  For the record there are a couple of other points 
that have been raised elsewhere that need to be addressed.

Major issues:
s2.7.1/s4: There is a security issue with RPCSEC_GSS_CREATE when 
multi-principal authentication is required.  See mail [1] below. This 
may have a (minor) knock-on effect of the NFSv4.2 specification where 
this is used; as currently specified (draft-ietf-nfsv4-minorversion2-40) 
use of privacy is already mandated for at least some of the relevant 
uses of RPCSEC_GSS_CREATE in NFSv4.2 which will mitigate the problem.  I 
have raised this issue in the Telechat review of the NFSv4.2 draft to 
ensure that privacy is mandated in *all* relevant cases - and to ensure 
changes are coordinated.

Minor issues:
s2.7.1.4: Some refinement of the constraints on the rp_name string 
marked as 'human readable' would be desirable.

Nits/editorial comments:
None

========================================
[1]
 > From: Nico Williams <[e-mail]>
 > Subject: Re: rpcsec-gssv3
 > Date: January 11, 2016 at 8:01:52 PM EST
 > To: Benjamin Kaduk <[e-mail]>
 > Cc: Stephen Farrell <[e-mail]>, "[e-mail]" 
<[e-mail]>, Spencer Dawkins <[e-mail]>, 
Martin Stiemerling <[e-mail]>, "Adamson, Andy" 
<[e-mail]>
 >
 > On Thu, Jan 07, 2016 at 02:14:01PM -0600, Nico Williams wrote:
 >> Ok, thanks.  I'll post a write up this Saturday.
 >
 > Or on Monday.
 >
 > OK, so, a while back Ben Kaduk noticed that there is a security problem
 > with the RPCSEC_GSS_CREATE procedure with multi-principal
 > authentication.
 >
 > The attack varies from easy to difficult to mount depending on whether
 > the server implements a global or per-connection GSS context handle
 > namespace, and whether it assigns them in a way that the attacker can
 > get a specific context handle number assigned to it.
 >
 > There are two fixes, the simplest of which is to require that the
 > RPCSEC_GSS_CREATE procedure be called with privacy protection when using
 > the multi-principal authentication feature.
 >
 > To recap how the RPCSEC_GSS_CREATE procedure with multi-principal
 > authentication feature works, the client first makes a MIC token with a
 > GSS context that authenticates the user to the server, then it it uses
 > that MIC in the payload of the RPCSEC_GSS_CREATE procedure. The
 > RPCSEC_GSS_CREATE procedure is itself protected with a GSS context that
 > authenticates the client.
 >
 > The problem is that the data that the user GSS context MICs is
 > insufficiently strong a statement of intent because it only identifies
 > the client host GSS context by its RPCSEC_GSS context handle ID and
 > nothing more.  An attacker that can intercept an RPCSEC_GSS_CREATE
 > procedure and cause the RPCSEC_GSS context handle to get re-assigned
 > will be able to steal the victim's identity.
 >
 > There are a number of solid fixes that fall into two classes:
 >
 > 1) Add to the data that the user context MICs in order to improve the
 >   quality of the statement of intent.
 >
 >   E.g., add the client host principal's name.  Or add a MIC made with
 >   the client host's GSS context.
 >
 > 2) Make it so the attacker cannot steal the MIC made with the user GSS
 >   context.
 >
 >   E.g., use privacy protection for the RPCSEC_GSS_CREATE procedure,
 >   thus the MIC made with the user GSS context cannot be obtained by the
 >   attacker without having the client host's or server's credentials.
 >   Stephen proposed this.
 >
 >   (The OpenAFS rxgk protocol uses GSS_Pseudo_random() [RFC4401] to
 >   similar effect.)
 >
 > At this stage the simplest thing to do is to require that clients always
 > use privacy protection for the RPCSEC_GSS_CREATE procedure. Note
 > that there is nothing that the server can do to prevent this attack if
 > clients don't use privacy protection for this, but the server should
 > still reject RPCSEC_GSS_CREATE procedure calls without privacy
 > protection.  (All of this is only for the multi-principal authentication
 > case.)
 >
 > Nico
 > --