Re: [saag] AD sponsoring Effect of Ubiquitous Encryption

[Stephen Farrell <[e-mail]>]

Hi again,

First, thanks to Paul Hoffman for being willing to be the
document shepherd for this. And to the authors for what
will be a fine RFC when it's done.

Second, I've done my AD review of this and there are a set
of things that need fixing before we'll be ready for IETF
last call. At minimum there are still a few placeholder
sections of text (2.3,2.3.1,3.2,2), but I also have many
comments that cumulatively raise to the level where we
should address a bunch of those before IETF LC, even
though many (but not all) of my comments are relatively
nitty. I think that almost all of those should be pretty
easy to resolve, so we should still be able to get this
done before IETF98 I hope.

My AD review is below.

Cheers,
S.

general: I wonder if we're dealing with all possible effects
or only with those relevant to network management, security
ad privacy? For example, more encryption does prevent some
advertising insertion, but whether or not that is in scope
here is unclear to me. Should it be?

- title: I wonder would "Effects of Much More Common
Encryption" be a better title? Ubiquitous is never actually
going to happen, nor is it required for many of the effects
to be seen. (And there is defo. >1 effect to be seen.) Or,
you could keep the title, but explain in the text that you
really mean "much more common."

- abstract: "solve these problems" is a tad one-sided in that
it doesn't recognise the valid reasons for more
confidentiality. I think it's important that the abstract
recognise that we're here documenting one set of consequences
of what is a valid set of trade-offs and not describing a set
of problems caused for no good reason.  I think that it's
important to set the right tone early for this one - if we
don't then some readers who want better privacy may be more
likely to be dismissive, which'd be a shame.

- abstract: "will impact" is maybe no longer the correct
tense - we've seen some of these impacts already I think, so
maybe "are impacting" or just "impact" is better?

- abstract: "more drastic" seems unfounded - what do you
really want to say there?

- intro: I'm not sure "primarily a passive attack" is right.
Maybe easiest to just delete that clause as I don't think
it's needed here. You could add a reference to RFC7624 for
those who'd like more detail.

- intro, 2nd para: I'm not sure what you mean referring to
"many attackers"? That sentence doesn't seem to make sense
as-is.

- intro: the UTA document references need to be updated

- intro: the 30% figure is probably out of date, please check
again (or add a date as done with the others).

- intro: "for corporate users" would be better removed - we
care as much about non-corporate users after all

- intro: "TLS over HTTP" is backwards, as is "TLS over XMPP"

- intro: "Although this does provide..." - that sentence
isn't right, passive is not the opposite of e2e, I think you
want s/passive wiretapping/transport layer attacks/

- intro: You mention OTR, but the cooler kids these days use
signal/telegram (though I don't for platform reasons;-).
Might be worth checking out what to say about those, or just
say "e.g. OTR" instead of making it seem like OTR is the only
way to do IM e2e.

- intro, 2nd last para: that one is very good

- intro, last para: you don't mention enterprise networks
here, worth a mention?

- section 2, 2nd para: I think you could do with a forward
reference or example to the "opterational practices that
require cleartext" or maybe better is to s/require/previously
depended on/

- 2: "TLS over SMTP" - backwards again

- 2: "The EFF reported..." that needs a reference

- 2: I'm not at all sure encryption "prevents" load
balancing, in fact I'm pretty sure it does not prevent that

- 2: "on the decline now that they are exposed through the
media" - I think it'd be better to remove that sentence - it
doesn't add much and there are more details better described
later I'm sure.

- 2.1: this sections seems mixed between middleboxes and
mobile, probably as a result of iterative editing. Not sure
how best to fix.

- 2.1, last para: Is that really a middlebox function? I
thought most of the examples given were cases of end systems
(web servers) impinging on privacy and not middleboxes (and
the naughty web servers are not affected by encryption so
aren't really in scope)

- 2.1.3: I don't think DPI aims to improve "features and
efficiency"

- 2.1.3: "must be known" is overstated - again I think we
want to always phrase it as "has previously been assumed to
be required" or similar

- 2.1.4: I think compression could (and maybe should) be
separated out - one of the things CRIME has shown us is that
we had more to learn about how to combine compression and
encryption. I also think most compression is not done/undone
at middleboxes, nor does it relate to monitoring so
compression doesn't seem to fit well in 2.1. (Transcoding
however might fit in some middlebox section, but is not
compression.)

- 2.1.5: s/law agencies/law enforcement agencies/ is the
usual term, and LEAs are not the only entities that insist on
this kind of thing - considering censorship.

- 2.1.5: "sites promoting anorexia" seems like an odd
example, but I think in any case it'd be better to generalise
the examples a bit

- 2.1.5: DNS queries do not "reveal" URLs. And the described
mechanism seems quite odd.

- 2.1.5: The 3rd last para about  "TLS proxies" and "future
versions of HTTP and TCP" is not clear to me. Is that really
needed?

- 2.1.5: Parental controls might deserve it's own subsection
- it differs in that the endopints are involved, which I
think means that the effects of e.g. TLS on the changes
required to the service are different from cases where an
external party (govt/SP/whoever) imposes the filtering.

- 2.1.6.3: I think you need a reference to RFC2804 here. The
IETF does not to LI in standards, for the (good) reasons
stated therin. So this may be a case where the effects of
crypto do enhance existing IETF policy but also counter
existing policies of other entities. But we should not ignore
the former (IETF) policy - we ought not claim here that one
or the other are better, but just ack the tussle.

- 2.1.6.4: MTAs do trypically have cleartext access, so what
kind of spam/malware filtering do you mean there?  How does
that relate to the mail architecture? (There's an RFC for
that that ought be ref'd.) If section 5 does cover this well,
then I'd say removing this section would be good as it add
nothing much that I can see.

- 2.2: Do we have evidence that the decrease of measurement
accuracy has actually lead to a decrease in repair
efficiency? I'm not convinced - it doesn't seem like a crazy
proposition, but nor does it seem like a tautology. (What
metric for efficiency of repair is there that could be mapped
to visibility of different protocol layers?)

- 2.2, 2nd para: This is repetitive. Making the point once is
enough.

- 2.3 and 2.3.1: What is the intent here? These sections seem
to be content free. Looks like a placeholder that never got
filled in? (Not unreasonable as it's hard to see real
downsides to encryption of inter-DC traffic.) These need to
be fixed or deleted. Probably deleted is better. (Certainly
easier.) That needs fixing before IETF LC.

3.1.1, 1st para: I'm not clear how session monitoring of
cleartext is needed for SLA evaluation. Maybe an example of
some specific thing needed would helpa but I'm not seeing an
input to SLA evaluation here. (I do see how logs can be
useful there.)

3.1.1 (and elsewhere): s/mobile clients/clients/ - while some
of thie text relates to the marnew w/s, it ought only specify
mobile clients where those are specifically relevant, an in
this case, and others, they're not.

3.1.2: "cyber-security" and "cyber-attack" please avoid
thosse ill-defined marketing terms unless you really need and
can justify them

3.1.1: Is "malware explosion" a generally known term? I think
detection is sufficient and searing for "malware explosion"
doesn't turn up what I think you mean. (I'm guessing you mean
unzipping etc.)

3.1.2, last para: I'm not clear how this para fits here or in
this document in general. Not sure if that's just me or if
some better linkage is needed, or even if the text would be
better moved or deleted.

3.2.1: I'm not sure I agree with this, at least as stated.
Wouldn't passive monitoring of ciphertext be just as
effective for the SLA metrics mentioned in almost all cases?

3.2.2: "[Need descriptions for..." The rest of that para
seems like placeholder text that needs fixing before IETF LC.

3.2.2: I think it'd be worth saying somewhere here that
STARTTLS ought have zero effect on anti-SPAM etc. for SMTP
traffic but is only an issue in corner-cases for e.g. hosters
who want to scan outbound port 25 traffic in case their
cusomters have suffered breaches and start sending spam.  (At
least that's my understanding.) PGP or S/MIME do of course.

3.2.2: What header encryption efforts are meant here? I'm not
familiar with anything doing exactly that. Or maybe it's just
a phrasing issue? Or perhaps this text is a bit old and OBE?
(I mean that initial hopes that we initiate work to to
encrypt e.g. Subject seem to be currently moribund.)

3.2.2: I'm not sure the bullet list here is a good idea.
E.g. it omits DKIM related things and SPF related things.
Maybe check with some anti-spam folks if keeping this list?

3.3: I wondered if this section would be better reorganised
into a structure based on where the keys are located. Not
sure but that might be a clearer way to distinguish the
various kinds of encrypted storage. (Feel free to ignore this
suggestion if it doesn't resonate.)

3.3.1: I found this sub-section confusing and not useful.
Would suggest simplifying it a lot.  The term "burst data"
isn't that clear.

- section 4: Is "for Enterprise Users" right here?

- 4: "harder to break" is not relevant here - at least in
terms of algorithms, which is what that'd be interpreted as
far as to mean. If you mean that non FS mechanisms (e.g. RSA
key transport) provide more opportunity for decryption at
places other than the endpoint at which the applications
calls for decryption, but that we increasingly do want to see
more use of FS mechanisns, and that that's a tension, then
maybe say that.

- 4.1: Not sure "static" is quire right - I think what you
mean is replicated TLS server RSA private keys.

- 4.1.1: "Enterprise users are subject to the policies of
their organization." That is jurisdiction-dependent - at
least in some countries (e.g. Ireland) the enterprise cannot
breach employment laws that can touch on employee privacy.
Such jurisdictional issues can work both ways of course,
either re-inforcing privacy or calling for privacy to be
eroded even more than the enterprise would wish.  Suggest
adding "... and the jurisdictions in which the enterprise
operates" to cover both;-)

- 4.1.1: "These functions are critical to security and fraud
monitoring." I don't accept that this is true of all the
numbered elements in the list.  Supposed countermeasures for
IP leakage for example are often ineffective (though vendors
who sell such things will doubtless disagree;-). I'd say
toning that down a bit would be right, e.g. s/are criticial/
are often consided important/.

- 4.1.2: Mention of PDM reminds me that that has had a pretty
thorough discussion related to the secdir review and issues
relevant to this draft. Incorporating a description of that
history would maybe be useful so that we have less pain when
we re-do that in future for other protocols.

- 4.1.3: I don't accept "impossible" is correct. If we wanted
to say that, we'd need evidence, e.g. via a set of
references. "More onerous" would be fine if that's all we
need to say. A number of other similar changes are needed for
the text in 4.1.3.x as well, e.g.:

- 4.1.3.1: "requires" is overstated, and what evidence is
there for "frequently used"? I don't accept either assertion
without evidence.

- 4.1.3.x: there are more of the above that need fixing

- 4.2, 2nd para: A lot of this is duplicative of earlier
text.

- 4.2: "unique and efficient vantage point" that's sales
language and doesn't belong here.

- 5.1: "TLS over SMTP" again a couple of times

- 5.2: Is I-D.teague.. the right reference here? I'd have
thought there was an ipfix RFC already that'd be good here.

- 5.6: "has quickly reacted to" is sales language and doesn't
belong.

- 5.6: What is "[not BCP84]" as a reference? And I'm not sure
why RFC2504 is a good reference here either. Are we talking
about BCP38 but someone forgot to update text written from
memory or something? :-)

- 5.6: SAVI was only chartered for enterprise networks iirc.

- 5.7: THe [SACM] reference is odd and has no entry in
section 12.

- 6.1: Are those "notable exceptions" still relevant? I'd say
maybe not.

- 6.1: Is the Alt-SVC description correct? I'm not sure. I
suggest you ask an HTTP person maybe. That could also do with
some references.

- 6.1: "... to be blocked" makes some assumptions about
what's ok/right that are not needed, "being requested" would
be better.

- section 7: "efforts to prevent it" - which it? And I don't
think prevent is what you mean, unless you're moving into the
"what the spooks want" territory and away from "what network
managers need" which'd seem out of scope.  I think that needs
rephrasing.

- section 7: "National surveillance programs" is not the
right phrase. Maybe just "Law enforcement agencies..."

- 7: please drop the various uses of "cyber" again;-)

- 7: David Cameron seems to no longer act in that role.

- 7: "avoid the fears of terrorism" is wrong - fears will
exiat and be propogated by various actors regardless. And
understanding technology won't help there. I think the fear
you mean is that technolgoy will be abused by bad actors.

- 7: The text seems to assume that all governments are
equally ok and that they all like one another equally. I dont
believe that is the case, so there are additional categories
of bad actor that are missing here. I also find the frequency
of the occurrence of "terror" to be higher than needed. (That
is a scare-word best omitted when not needed.)

- 7: I think a reference to the workshop held in Wash. DC
last year (maybe June?) on whether or not magic solutions for
LEAs can exist here may be useful. Some IETFers attended, and
it covered these issues in a lot more depth than could be
fitted here but I don't have the reference to hand. (Spoiler:
there is no magic;-)

- 7, last para; I'd recommend deleting this. When and if PIR
methods are usable they will create as many issues as the
solve, from the perspective of this document.

- section 11: Is this useful? If not, it ought be deleted.
If so, then it should be a section of its own on mobile
networks and not an appendix. (I'll comment assuming that
keeping the text is better, but I'm not sure it is, given the
amount of context it may need.)

- 11.1: What is an ACK stream? A reference or description is
needed. "Prevents" needs justification (possibly via
reference)

- 11.1: eNodeB needs a reference, KPI/KQI need expansion, CEM
needs describing or a reference. What is a "sector"?  And (in
this context) what is "a server"? APN needs a reference.

- 11.1: "trusted parties" is unclear - you need to say what
is trusted by whom for what actions for this to be useful.

- 11.1: DRM/QOE need expansion and a reference to justify the
claim wrt *maximising* QOE.

- 11.1: "Trusted proxy" - see above wrt term trusted. But in
this case, the term is misleading and ought not be used as
there seems to be no way in which the entity who needs to
trust the proxy can know that that proxy exists. Please drop
that term entirely. (Call it a TLS-MitM attacker if you must,
as that is what it is and that is how such things were
referred to earlier in this document.) There are two
occurrences of that misldeading marketing term to remove.

- 11.1: RAN needs expansion and "RAN-aware pacing" needs
explanation and/or reference.

- 11.2: What is meant here by "Transport header"? TCP mostly
doesn't have this issue, so I assume this is looking forward
to QUIC. I'd argue that if so, this text should be deleted as
we now have a WG for QUIC which is specifically chartered to
address these issues.  If this is not about QUIC, then please
elaborate.

- 11.2: ANDSF needs a reference

- 11.2: SON and MLB need references

- 11.2: UPCON needs a reference, off-peak acceleration and
outbound roamers need explanation and/or reference

- 11.2: DSCP needs a reference

- 11.3: RCS, QCI, LTE need rererences

- 11.3: eMBMS needs a reference, see above wrt "trusted"

- 11.3: Asserting that TLS is not necessary for DRM'd content
is arguable and not blatently true, as asserted here.

- 11.4: see above wrt "trusted."

- 11.4: QAM, CS-Voice need references

- 11.4: PGW, GGSN need references

- 11.4; "GW" needs explanation and/or a reference

- 11.4: "Decreasing Client-Server control loop requires
deploying CDNs/Cloud functions that terminate encryption
closer to the edge." I don't accept that "requires" is
clearly true. Please justify e.g. via reference. Or maybe
re-phrase so that the statement is clearly true but doesn't
have gigantic implications for who is allowed to deploy what.
(Or maybe I'm confused by what is meant here?)

- 11.4: s/de-encryt/decrypt/

- 11.4: MEC needs a reference, see above wrt "trusted proxy"

- 12.1: I agree with the shepherd that all the references
here can be informative and you don't need 2119.

- 12.2: URLs like that used for CAIDA aren't great for RFCs -
better to find a better reference where that's easy (which it
is in that case). I'm less sure about how best to deal with
the Intercept and EFF, but if you search in
scholar.google.com you may find other references for that
content that are easier to use. (The issue here being
reference stability.)

- 12.2: [homomophic] needs a better reference if it stays,
e.g. Gentry's PhD or something.

- 12.2: various references need to be updated. I-D nits will
help you there

- 12.2: I'm sure you'll agree that [TOR] isn't right as-is:-)