Hi Daniel,
(...)
As I'm sure you have noticed, we have updated Section 3 following
your last input. It now explicitly says:
Attackers can collaborate to reach a common goal.
It also says
Note that in this attacker model, an attacker (see A1) can be a RO or
act as one. For example, an attacker can use his own browser to
replay tokens or authorization codes obtained by any of the attacks
described above at the client or RS.
Your scenario is therefore covered. It was already before, but that
was obviously too implicit, so we made it more clear with the recent
update.
[Denis] I don't believe that the scenario is covered with the above
sentences.
I don't understand. This is not about believing, it is written very
clearly and conclusively in the text of the current draft.
We've had this discussion before, and, although irrelevant for the
meaning of the BCP itself, I would like to refer you again to the
formal model in our research paper, which the BCP attacker model is
based on. It has a *very* precise definition of the attacker model
that does not leave room for interpretation. The natural-language
description in the BCP, as usual, is more fuzzy than the formal
definition, but both attacker models include clients cooperating.
[Denis]. Your very last sentence is finally using two magic words that
are not present anywhere in the BCP: "*clients cooperating*".
However, *clients collusion* or *clients collaboration* would be more
accurate.
Finally, section 4 (Attacks and Mitigations) should include an
additional subsection, e.g. section 4.16, addressing the case of a
collaboration attack
between clients against a RS.
If I remember correctly, you first presented this attack at the
OAuth Security Workshop in 2017.
Since then, it has been brought up countless times on this mailing
list, both with regards to the Security BCP as well as for the JWT
Token draft.
There has been practically no positive resonance at the meeting 2017
or here on the mailing list as to including this in either of the
drafts.
A number of reasons come to mind, but first and foremost, I think
that what you describe is not perceived as an attack, or, worded
differently,
it is obvious that what you describe in the "attack" is possible.
[Denis] Here after comes the important sentence which is wrong:
*There is no expectation that OAuth would defend against this kind
of thin**g*,
just as there is no mitigation against password sharing in
password-based authentication.
[Denis] In the section 4.16.2 there is a clear proposal that explains
how *"OAuth can successfully defend against this kind of thin**g"*.
*So* *there **IS **a solution*.
I did not say that there is no solution.
[Denis] Well, ... ask anybody else how he would interpret your statement.
Currently, when reading the text, an implementer might consider to
deliver an access token that contains a claim such as "older the 18"
without any "sub" or equivalent claim.
Such an access token would be transferable to anyone and the RS would
not be able to identify the attack.
Your proposed solution does not enable an RS to identify the attack
unless used together with some form of authentication way outside the
scope of OAuth.
[Denis] I never said that there is "some form of authentication". The
word "authentication" is not present in my text proposal.
At the moment (/and this is a topic of its own that could be discussed
later on/), a "sub" claim present in an access token is unrelated to any
identifier
used during an authentication exchange between an end-user and a RS.
This means that your statement is incorrect.
*"OAuth can successfully defend against this kind of thin**g" and, since
countermeasures exist, they should be described.
*
Again, this also goes deeply into OIDC territory.
I therefore propose to proceed with the Security BCP *with the
inclusion of this attack*.
Even though the Security BCP attacker model includes the general
setting required for the attack, the attack does not violate an
expected security property.
I therefore propose to proceed with the Security BCP without
including this attack.
-Daniel
[Denis] Since you have deleted the remaining of my email, I copy it
again. If you respond to this email, please DO NOT delete it.
I deleted the remainder of the mail as it was not relevant to my
answer (see RFC1855, Section 3.1.1). Everybody can access your
original mail on the mailing list.
-Daniel
I re-established the remainder of the mail as it is relevant to *my
*answer.
However, reading it again, the "Attack description" does not refer to a
JWT access token whereas it is not the case for the two other sub-sections.
Nevertheless, these two sub-sections could be easily generalized to also
address the case where JWT access tokens are not being used.
Finally, section 4 (Attacks and Mitigations) should include an
additional subsection, e.g. section 4.16, addressing the case of a
collaboration attack
between clients against a RS.
This sub-section would need to include to other sub-sections:
4.16.1Attack Description
4.16.2Countermeasures
The following text is a skeleton proposed for these subsections:
*4.16****Collaboration attack between clients against a RS*
The goal of the attack is for an illegitimate client to obtain an
access token from an authorization server with the help of a client
of the authorization server.
*4.16.1****Attack Description*
The legitimate client performs in real time all the cryptographic
computations needed by the illegitimate client to get an access
token and to present it to a RS.
This attack is not a replay of a access token, but the use of a
legitimate access token by an illegitimate client with the
complicity of the legitimate client.
It should be observed that protecting some private keys into a
secure element is ineffective to counter this kind of attack, since
the legitimate client can perform
all the computations needed by the illegitimate client, without the
need to know or to transfer the values of these private keys.
*4.16.2****Countermeasures*
This attack may be countered by using a "sub" claim into the access
token. It should be observed that the "sub" claim is a REQUIRED
claim in the JWT access token
data structure. See section 2.2 from JSON Web Token (JWT) Profile
for OAuth 2.0 Access Tokens.
Section 5 (Security Considerations) from JSON Web Token (JWT)
Profile for OAuth 2.0 Access Tokens states:
Authorization servers should prevent scenarios where clients can
affect the value of the "sub" claim in ways that could confuse
resource servers.
This statement is correct but insufficient, since it does not say
how resources servers cannot get confused.
Section 6(Privacy Considerations) states:
This profile mandates the presence of the "sub" claim in every
JWT access token, making it possible for resource servers to rely on
that information
for correlating incoming requests with data stored locally for
the authenticated principal.
This statement is correct but is unclear. To be more precise, in
order to counter the collaboration attack between clients against a
RS, the RS should manage
user accounts associated either with a globally unique identifier or
an identifier specific to an AS-RS pair while the "sub" claim shall
contain either
a globally unique identifier or an identifier specific to an AS-RS
pair which shall be compared with the identifier of the user
account. If there is no match,
the access token shall be discarded.
Denis
PS. I re-read RFC1855, Section 3.1.1, but there is nothing in the
Netiquette to delete or maintain some parts of a received message.
--
https://danielfett.de
_______________________________________________
OAuth mailing list
OAuth@ietf.org
https://www.ietf.org/mailman/listinfo/oauth