[OPSAWG] Eric Rescorla's Discuss on draft-mm-wg-effect-encrypt-17: (with DISCUSS and COMMENT)

[Eric Rescorla <[e-mail]>]

Eric Rescorla has entered the following ballot position for
draft-mm-wg-effect-encrypt-17: 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-mm-wg-effect-encrypt/



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

I have completely re-reviewed the latest version. First, I want to
thank you for toning down some of the material that I was concerned
about.

Unfortunately, the fundamental concern that motivated my DISCUSS
remains: I do not believe that this document matches the consensus
of the IETF community.

Specifically, this document takes at face value a large number of
claims about the necessity/value of certain practices that either are
controversial within the IETF or that there is, I believe, rough
consensus to be actively bad, and that in many cases, encryption is
specifically designed to prevent. I have noted a number of these
below, but I do not believe that this is an exhaustive list (going
through my previous DISCUSS, I see that I noted a number of these but
that still appear to be unaddressed.)


DISCUSS
   session encryption that deployed more easily instead of no
   encryption.

I think I understand what you are saying here, but the term
"breakable" seems very unfortunate, especially in the context of this
document. The only such shift I am aware of that has received
acceptance in IETF is one from always requiring fully authenticated
encryption to allowing unauthenticated encryption, which you document
in the next paragraph. I recognize that there are other perspectives
(e.g., those in draft-rhrd) but those are pretty far from IETF
consenus. Accordingly, I think you should remove this sentence.


   motivation outside of privacy and pervasive monitoring that are
   driving end-to-end encryption for these content providers.

This section seems kind of confusing, at least as applied to
HTTP. There are three kinds of cache in HTTP:

- Reverse proxy caches (the kind CDNs run)
- Explicit forward proxy caches
- "Transparent" intercepting caches

The first of these move to HTTPS just fine and that's typically how
CDNs do it.  Explicit forward proxy caches are largely going away, but
part of the point of this kind of end-to-end encryption is to hide
data from those caches.  Transparent intercepting caches that the user
didn't opt into are a bad thing, so having them go away is a positive.

   document existing practices, the development of IETF protocols
   follows the guiding principles of [RFC1984] and [RFC2804].

This is pretty opaque in this context. It should explicitly state that
the IETF's position is that we do not engineer for these use cases,
not just to cite the RFCs.


   based billing, or for other reasons, possibly considered
   inappropriate by some.  See RFC7754 [RFC7754] for a survey of
   Internet filtering techniques and motivations.  This section is

I don't think this accurately represents the RFC, which makes clear
that the IAB consensus is that filtering is bad:

" From a technical perspective, there are no perfect or even good
solutions -- there is only least bad.  On that front, we posit that a
hybrid approach that combines endpoint-based filtering with network
filtering may prove least damaging.  An endpoint may choose to
participate in a filtering regime in exchange for the network
providing broader unfiltered access."


   detect or prevent attacks as well as to guarantee service level
   expectations.  In some cases, alternate options are available when
   encryption is in use, but techniques like that of data leakage

These certainly are use cases, but you really need to acknowledge that
it's also a form of user surveillance.


   Some DLP tools intercept traffic at the Internet gateway or proxy
   services with the ability to man-in-the-middle (MiTM) encrypted
   session traffic (HTTP/TLS).  These tools may use key words important
   to the enterprise including business sensitive information such as
   trade secrets, financial data, personally identifiable information
   (PII), or personal health information (PHI).  Various techniques are
   used to intercept HTTP/TLS sessions for DLP and other purposes, and
   are described in "Summarizing Known Attacks on TLS and DTLS"
   [RFC7457].

As far as I know, the MITM techniques used by middleboxes are not
described in 7457.

You also need to cover the fact that these MITM are a threat to user
security because they are often so badly implemented.


S 5.4.
It's pretty odd to talk about phishing without acknowledging that by
far the largest anti-phishing platform (Safe Browsing) operates in the
client, not be network interception.


    The transport header encryption prevents trusted transit proxies.  It
   may be that the benefits of such proxies could be achieved by end to

You don't define a "trusted transit proxy" so I don't know what this
means, and whether they provide any benefit, but this certainly sounds
euphemistic. Generally, "trusted" is not an adjective we associate
with network proxies operated by third parties.


   In the best case scenario, engineers and other innovators would work
   to solve the problems at hand in new ways rather than prevent the use
   of encryption.  As stated in [RFC7258], "an appropriate balance
   (between network management and PM mitigations) will emerge over time
   as real instances of this tension are considered."

Much of the context of this debate is about whether operators not
being able to do the things in this document is a problem, and this
seems to presume the answer.


   This optimization at network edges measurably improves real-time
   transmission over long delay Internet paths or networks with large

Do you have a citation for this claim?


   Web proxies are sometimes used to filter traffic, allowing only
   access to well-known sites found to be legitimate and free of malware
   on last check by a proxy service company.  This type of restriction
   is usually not noticeable in a corporate setting as the typical
   corporate user does not access sites that are not well-known to these
   tools, but may be noticeable to those in research who are unable to
   access colleague's individual sites or new web sites that have not
   yet been screened.  In situations where new sites are required for
   access, they can typically be added after notification by the user or
   proxy log alerts and review.  Home mail account access may be blocked
   in corporate settings to prevent another vector for malware to enter
   as well as for intellectual property to leak out of the network.
   This method remains functional with increased use of encryption and
   may be more effective at preventing malware from entering the
   network.  Web proxy solutions monitor and potentially restrict access
   based on the destination URL or the DNS name.  A complete URL may be
   used in cases where access restrictions vary for content on a
   particular site or for the sites hosted on a particular server.

If you are filtering on URL and there is HTTPS involved, then you are
a MITM, and this is potentially noticeable to end-users. We encounter
this regularly when MITM proxies make mistakes in getting into our
trust anchor store.


Re: 0-rating.
   user.  This feature is impacted if encryption hides the details of
   the content domain from the network.

Well, maybe. Facebook's zero rating, for instance, is IP-based.
https://info.internet.org/en/blog/2015/09/24/enhancing-security-and-privacy-of-free-basics/

S 2.2.2.
The presentation here seems biased given that it does not acknowledge
that one of the reasons that ISPs do traffic class discrimination is
to prioritize favored rather than disfavored traffic, regardless of
user preferences. I don't believe that the IETF has taken a position
for net neutrality, but I'm also pretty sure we don't have consensus
against it.


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

   Pervasive Monitoring (PM) attacks on the privacy of Internet users is
   of serious concern to both the user and the operator communities.
are of serious concern

   Traditional network management, planning, security operations, and
   performance optimization have been developed in an Internet where a
   large majority of data traffic flows without encryption.  While
you have a plural/singular mismatch here.

   Authentication with IPsec [RFC7619] and there are a number of
   infrastructure use cases such as server to server encryption, where
   this mode is deployed.  While OS is helpful in reducing pervassive
Nit, no comma before "where"

   recommendations from these documents were were built upon for TLS 1.3
   to provide a more inherently secure end-to-end protocol.
"were were"

   to-user content encryption schemes, such as S/MIME and PGP for email
   and encryption (e.g.  Off-the-Record (OTR)) for XMPP are used by
   those interested to protect their data as it crosses intermediary
Not, "e.g.,"

   been impacted by increases in encrypted traffic.  Only methods
   keeping with the goal of balancing network management and PM
   mitigation in [RFC7258] should be considered in solution work
   resulting from this document.
I don't understand what the normative content of this was.

   upgrades before user experience declines.  For example, findings of
   loss and jitter in VoIP traffic can be a predictor of future customer
   dissatisfaction (supported by metadata from the RTP/RTCP protocol )
   [RFC3550], or increases in DNS response time can generally make
   interactive web browsing appear sluggish.  But to detect such
   problems, the application or service stream must first be
   distinguished from others.

This is a little hard to read, but generally this information is not
encrypted for SRTP/SRTCP.

   When using increased encryption, operators lose a source of data that
   may be used to debug user issues.  Because of this, application
   server operators using increased encryption might be called upon more
   frequently to assist with debugging and troubleshooting, and thus may
   want to consider what tools can be put in the hands of their clients
   or network operators.

This is phrased as hypothetical, but is there any actual data to
support this? We have a lot of experience now with encrypted voice and
video as well as Web. Do those providers report any increased level of requests.

   the new flow and would not be able to route it back to the original
   POP.

I'm having a little trouble understanding this paragraph. Why is this
about mobile operators? It seems like it's an issue for anyone who has
a service that might have mobile clients.

   effective at providing data offload when there is a network element
   close to the receiver that has access to see all the content.
This section is also confusing in that it fails to distinguish between proxies
of this type that the user opts into from transparent proxies that just do this
service without the user's consent. Part of the purpose of encryption is to
preclude that.

   created from the intercepting device to the client's destination,
   this is an opt-in strategy for the client.  Changes to TLSv1.3 do not
   impact this more invasive method of interception, where this has the
I don't understand the cite to TLS 1.3 here. None of this is presently affected
by 1.3.

   endpoints as a service reduces providers' ability to offer parental
   control.
This section makes it seem like there are no other methods that allow for
parental control, but that's not true. There are parental control mechanisms
which rely on MITM proxies or application interfacing.

   HTTP headers and content are encrypted, this prevents mobile carriers
   from intercepting the traffic and performing an HTTP redirect.  As a
   result, some mobile carriers block customer's encrypted requests,
Yes, stopping traffic redirection is actually one of the major design purposes
of HTTPS.

   to the customer and additional overhead to the mobile network service
   provider.
This section is basically a commercial for this practice. It needs to
acknowledge that it's actually much closer to IETF consensus that it's a bad
practice.

   headers to accomplish the, sometimes considered controversial,
   functions above.
Again, this is precisely the form of attack that HTTPS is intended to prevent.

   perform SSL/TLS decryption are impacted by the increased use of
   encryption that prevents interception.
I don't really understand what this text means. What is "use of encryption that
prevents interception" in the context of tools which already do decryption?

   [DarkMail].  Of course, SPAM filtering will not be possible on
   encrypted content.
This is not strictly correct, as you can still header filter, etc.

   over the Internet.  Options for monitoring will vary with each
   encryption approach described below.  In most cases, solution have
   been identified to provide encryption while ensuring management
Nit: "solutions have"

   allow for multiple end user sessions to share the same TCP
   connection.  This rasises several points of interest with an
   increased use of encryption.  TCP session multiplexing should still
Typos: rasises

   be possible when TLS or TCPcrypt is in use since the TCP header
   information is exposed leaving the 5-tuple accessible.  The use TCP
   session multiplexing of an IP layer encyption, e.g.  IPsec, that only
I'm not sure if this is correct. What precisely do you mean by "TCP
session multiplexing"? A cite would be helpful here.

   center.  HTTP pipelining requires both the client and server to
   participate; visibilty of packets once encrypted will hide the use of
   HTTP pipelining for any monitoring that takes place outside of the
Typo: visibility

   traffic cannot be intercepted, encouraging alternate options such as
   performing these functions at the edge.
I'm not seeing the relevance of this. Who is proposing solutions in which
encrypted traffic cannot be intercepted at the outgoing network edge.

   owing to concealment of critical headers and payloads.  Many forms of
   enterprise performance management may be similarly affected.
Again, this is sort of transparent caching is an anti-pattern, so this document
ought to acknowledge that,

   parts of the address assignment/management protocols, critical for
   SAVI mechanisms, can result in a dysfunction of the SAVI mechanisms.
Does anyone actually deploy SAVI? If not, encryption isn't having much of an
impact.

   network filters look out for seeing a Server Name Indication
   extension, they may not find one.  The SNI Encryption in TLS Through
   Tunneling [I-D.ietf-tls-sni-encryption] draft has been adopted by the
This doesn't really follow. I mean they "may not" but overwhelmingly they will