[LACNIC/Seguridad] Identificadores numericos predecibles (Fwd: Re: Last Call: <draft-gont-numeric-ids-sec-considerations-06.txt> (Security Considerations for Transient Numeric Identifiers Employed in Network Protocols) to Best Current Practice)
fgont en si6networks.com
Mar Dic 15 20:50:41 -03 2020
En ocasiones, cuando un mismo problema es recurrente en el tiempo, uno
se pregunta "como diablos puede ser que el mismo problema siga
ocurriendo una y otra vez para diferentes protocolos).
Junto a Ivan Arce, decidimos abordar uno de dichos problemas: el de uso
de "identificadores numericos temporales predecibles". Publicamos tres
documentos al respecto:
El primero estudia la historia de los IDs predecibles -- y quien sepa
leer entre lineas, tal vez pueda sacar alguna conclusión extra :-)
El segundo hace un analisis tecnico de estos identificadores, analiza
las implicancias de seguridad y privacidad, y propone algoritmos que
permiten generar identificadores numericos sin dichos problemas.
Finalmente, el tercer documento, apunta a recomendar que todas las
especificaciones hagan un analisis de seguridad y privacidad adecuado,
apra poder evitar que estos problemas ocurran en el futuro.
El nivel de ridiculez de las objeciones recibidas no ha tenido limite.
Aqui envio una, en la cual el autor argumento que como esta a punto de
publicarse la especificación de QUIC, y la misma *no* hace una analisis
de seguridad adecuado de sus identificadores numericos, entonces nuestro
documento #3 no deberia publicarse.
Dicho de otro modo: "estamos haciendo las cosas mal. Asi que no queremos
que se publique un documento con recomendaciones, ya que nos dejaría en
Abajo reenvio el mail en cuestión. Y quien esté interesado puede leer el
thread cometo en:
-------- Forwarded Message --------
Subject: Re: Last Call:
Considerations for Transient Numeric Identifiers Employed in Network
Protocols) to Best Current Practice
Date: Sun, 13 Dec 2020 14:06:39 -0800
From: Eric Rescorla <ekr en rtfm.com>
To: Iván Arce (Quarkslab) <iarce en quarkslab.com>
CC: Fernando Gont <fgont en si6networks.com>, last-call en ietf.org,
draft-gont-numeric-ids-sec-considerations en ietf.org
On Sun, Dec 13, 2020 at 1:03 PM Iván Arce (Quarkslab)
<iarce en quarkslab.com <mailto:iarce en quarkslab.com>> wrote:
Hello Eric, Fernando
On 12/13/20 5:38 AM, Fernando Gont wrote:
> Hello, Eric,
> Thanks for your comments! In-line....
> On 12/12/20 18:44, Eric Rescorla wrote:
>> At a high level, many of the attacks listed here (especially in
>> result from the reliance (potentially accidental) on unpredictable
>> identifiers as a countermeasure against various forms of
>> instance, TCP is subject to a variety of off-path injection attacks
>> which are partly mitigated by randomizing sequence numbers and port
>> numbers. However, the more modern practice is to cryptographically
>> authenticate datagrams, thus preventing injection attacks even
>> port and sequence number are predictable.
> I don't think the two are mutually-exclusive. Nobody is arguing that
> randomizing numeric IDs is an alternative to cryptographic
Indeed. I would further argue that use of authenticated encryption is
not warranted on all protocol designs so dismissing the need for
security and privacy considerations for transient IDs on the basis that
"we encrypt and authenticate the packets anyway" is not a good
Fortunately, that's not the position I am taking. Rather, I am saying
that authenticated encryption is an increasingly common practice and
that we should not publish a set of recommendations in this area which
does not engage with that properly.
>> Of course, it might be the case that these defenses are
>> (that would be useful to know!) but this document does not provide
>> much assistance in making that evaluation.
The intend of the document is not to provide cryptoanalysis guidance to
evaluate specific protocol designs but to provide general guidance on
how to deal with use of transient numeric identifiers in protocol
Indeed, the two protocols that you singled out do not follow our
guidance (they hardly could as the document we are reviewing is not yet
officially published) and assessing if the lack of precision for the
generation of transient IDs weakens the protocols would be a matter for
the respective authors to deal with.
This is an odd argument, as the authors of these documents could certainly
have read your documents and adopted your recommendations. But my
point here is different. As I said to Fernando, we are about to publish
protocols as PS and yet they appear to violate the guidance here, which
like a bad situation if we are about to publish this guidance as BCP. As
I think the problem is that the guidance is not well suited to this type
which means that this document ought to be adjusted. However, if you
have an argument that it is these protocols that should be revised, that
be very good to know.
>> Connection IDs MUST NOT contain any information that can be
>> an external observer (that is, one that does not cooperate
>> issuer) to correlate them with other connection IDs for the
>> connection. As a trivial example, this means the same
>> MUST NOT be issued more than once on the same connection.
> I believe that not recommending a good/safe choice for
> has been proved to be problematic.
Indeed, that has been the case even in documents that describe the use
of cryptographic algorithms in protocols.
See, for example, the case of "RFC 25288 - AES Galois Counter Mode
Cipher Suites for TLS" which in section 3 states:
The nonce_explicit is the "explicit" part of the nonce. It is
by the sender and is carried in each TLS record in the
GenericAEADCipher.nonce_explicit field. The nonce_explicit length
(SecurityParameters.record_iv_length) is 8 octets.
Each value of the nonce_explicit MUST be distinct for each distinct
invocation of the GCM encrypt function for any fixed key.
meet this uniqueness requirement can significantly degrade
The nonce_explicit MAY be the 64-bit sequence number.
The case below is quite similar and even more specific (it hints at a
particular way of implementing it) than the QUIC example and yet it
to discovery 8 years later of at least 4 vulnerable implementations
deployed on the Internet, as described by Bock at. al. in
"Nonce-Disrespecting Adversaries: Practical Forgery Attacks on GCM
Note that in the AES-GCM in TLS case the need for a counter was
warranted but not explicitly called for and such underspec'd text lead
to flawed implementations. In other cases, a random nonce/ID is better
than a numeric sequential value, we argue that authors should assess
what is appropriate in their particular case and write their spec and
rationale for it accordingly.
Well I'm certainly not arguing that it's not important to describe good
practices for cryptographic protocols, but: it seems to me that this RFC
did precisely what your document recommends in S 5.
- It specified the interop requirements (none)
- It provided the security and privacy analysis (they must be unique and
it's bad if they're not)
- It recommended an algorithm (the sequence number)
I would note that the eventual outcome in this case in TLS 1.3 was to
remove the explicit nonce entirely and to replace it with a value
generated from the sequence number.
>> I'm not saying that cryptographic protocols don't need to take
>> identifer selection, but the guidance in this document does not
>> very helpful in that respect.
I'd like to dig deeper into that feedback. In which way do you think it
does not seem helpful. What would you expect it to say to be helpful?
See above for a specific example.
Baring a more detailed proposal I propose to include text that
explicitly calls out that in cases where protocols require
algorithms to provide confidentiality and integrity (ie. authenticated
encryption) of the transient identifier fields some of the inherent
weaknesses in transient ID generation may be mitigated.
Does that sound like a sensible caveat?
Not really, no. I believe substantial rework is needed to address the
role that identifiers work in cryptographic protocols.
Más información sobre la lista de distribución Seguridad