[Top] [All Lists]

Re: [ietf-dkim] DKIM Key Sizes

2016-10-30 18:08:05
Hash: SHA256

On Oct 28, 2016, at 5:02 AM, Eliot Lear <lear(_at_)cisco(_dot_)com> wrote:

* PGP Signed by an unknown key

Hi Jon,

I really don't get your point.  Either the thing is worth signing or it
isn’t.  See below.

No problem. That you say, "either the thing is worth signing or it isn't" shows 
that you don't get it. I'll try to explain more.

That assurance comes at a cost of whatever integrity that assigns to a
message that might have been unintended. That message integrity is a
privacy loss by the users.

The real-world case in point is the leaked Podesta emails, where some
people have asserted that authenticity can be checked via the DKIM
signatures. I've raised an eyebrow on that, and the bottom line is
that you're presuming that attackers were sophisticated enough to
steal the email, yet unsophisticated enough that stealing the DKIM key
from an MTA is out of the question.

If it’s the same conversation I was engaged in, I think the person was
just confused.  A simpler approach to plausible deniability is simply
not to sign.  And further, just because a person can hack someone’s
personal message store, as seems to have happened to Podesta, doesn’t
mean that they can hack the private key out of Google.  Furthermore,
using the example above, the key would actually have to be around for a
while (e.g., not rotate out of DNS).  In fact, if the key really is
intended to verify the source over a longer period of time than, say,
delivery, certainly 512 is not enough.

Yup! I agree completely. If the key is intended to verify the source longer 
than delivery, then 512 is not enough.

And -- the key is *PRECISELY* intended to verify the source *NO* *LONGER* than 
delivery. 512 probably still isn't quite enough, since you can break a 512-bit 
key too easily. If I lick my finger and stick it into the wind, 600-700 bits is 
about where they should be.

512 is slight exaggeration for effect. You can make it work, but you really 
need to put some infrastructure in place to make it work well. But I belive 
that 512 is a better choice than 8K. Or even 4K. But not for the same reasons.

The full discussion is pretty nuanced, and I think the relevant part
here is that if an administrative domain wants to protect the privacy
of its users, it should be using *smaller* DKIM keys, not larger ones.
I think I could convincingly argue that a privacy-friendly email
provider is better off using 512 bit keys (where there's a chance of
spam forgery) than 4K keys (where there's a chance of ruining the
privacy of the customers). Now actually, I think the sweet spot for
key size is above 512, but it's lower than 768. For maximum balance,
you want the key to take longer than a week to crack, but less than a
year, and change it every month. Or something like that. You get the idea.

Color me obtuse, but are you worried about the incentives to break into
the user's account or are you worried about the provider wanting to peer
into users' email in order to assure itself that it’s not spam?

Ah, I see I'm going to have to at least snuggle, if not embrace the fuller 

My issue -- not a worry -- is the *semantic* value of the signature.

Many, if not most protocols we deal with in the IETF are strictly syntactic. We 
don't look at the meaning, we don't look at the internals of them. Often this 
is both good and intended. Look at IPsec, TLS, SSH, OpenPGP, S/MIME, and on and 
on, these are syntactic standards.

In some of them, the unspoken semantic issues are a wart. In S/MIME and 
OpenPGP, there's a huge semantic issue as to what a signed message *means*. I 
had a co-worker (at both Pretty Good Privacy, Inc. and PGP Corporation) who 
would not sign his emails. The reason he wouldn't is that he said that we 
didn't know what signed email meant and that he didn't want to end up finding 
out twenty years later that his signing a message was suddenly a legal 
commitment that he didn't intend.

This is also an interesting long discussion. Is a signature on an email nothing 
more than an source-directed integrity check? Does it mean a little more, such 
as the digital equivalent of letterhead? Does it mean even more than that, that 
it's 'official'? Or more than that, that it's certified true and accurate? I 
know where I stand on it, which is far more towards the integrity check or at 
most letterhead. I might even roll my eyes, but it's a not unreasonable concern 
that I simply happen to disagree with.

My last bit if setup and groundwork is to gesture towards Phil Rogaway's essay 
of last year on crypto and its social aspects, which is such a big thing that 
I'm just going to say, "Thanks, Phil," and move on.

All of this setup is relevant here because the mission of DKIM is to stop a 
certain class of message forgeries. These message forgeries are / were a danger 
to the public health of the Internet. DKIM addresses them. Note that I used the 
word "address" not "solve." SPF also addresses them, in a different way. So 
does a lot of other network infrastructure, as well, as well as the upper parts 
of the DKIM stack like DMARC.

I think it was Mark Delaney's words in his mission statement for the DK part of 
DKIM that it lets Alice's ISP know that a message from her from her bank 
actually came from her bank, even when forwarded through her alumni 
association. DKIM gives source attribution that travels with the message.

However, that source attribution has another side to it. That other side is 
that at its core, one might say a DKIM signature is in a similar moral space as 
the "super cookies" that some providers put into web connections (like Verizon, 
who the FTC fined $1.3M). It is a tracking thing that is put without the user's 
consent into their email that can be used for identification later without 
their knowledge. Worse, it's not merely a cookie that's a constant, it's a 
content-specific digital signature. Shock horror.

Now of course, the counter-argument is about the semantic layer. Yeah, 
syntactically, there's a resemblance, but the reason they're different is 
many-fold and has to do with purpose, use, and so on. 

For one, the "super cookies" are a conversation between the network provider 
and an ad network that does not benefit the user in any way. They are an 
*exploitation* of the user. A DKIM signature benefits us all through that 
initial use case. They add to the public health of the Internet as a whole and 
the email infrastructure at large. I can mount a vigorous defense of DKIM, but 
I could also use DKIM in a defense of tracking cookies. I believe I could also 
shoot down that analogy.

I also think that we can improve DKIM in a lot of ways operationally that 
further breaks whatever improper analogy there is. We also ought to re-examine 
a lot of the email infrastructure with a post-Snowden eye, and that eye should 
include a post-Yahoo-scanning scandal as well. We really should look at how 
much the email ecosystem as a whole has become part of the broader surveillance 
structure of the Internet.

These include:

- - MDAs should strip out DKIM signatures. If they did, my whole silly comments 
about key sizes are moot. It means that they aren't part of the whole permanent 
record of email stores and their insecurity because of the peculiar legal 
status of email.

- - MDAs should also be stripping Received headers and a bunch of others. 
Received headers write in stone the geoposition someone was at when they 
pressed send. It's a 4-space pin in the map. You can track people in their 
business travels by looking through old mailing list archives. You can see what 
type of computer they run, what MUA they are using, and their upgrade patterns. 
All the better to spearphish them with. 

- - That goes double for you, Mailman. And any other list archiver. 

These and more are part of why email is so broken these days. It was made for 
an environment when we all used terminals on timesharing systems and things 
that made sense then (putting logging and debug information into a message) do 
not make sense in a world of mass surveillance. Cleaning up old messes is 
always hard, and this is another one.

But anyway, back to DKIM --

The problem here -- DKIM being a super cookie -- can be handled any number of 
ways. One of those is to rotate keys frequently. Really, big ISPs like Google 
and Yahoo and Apple and Microsoft ought to be putting out keys with one week of 
use and two weeks of life. That doesn't solve the whole problem because there's 
no reason why The Internet Archive or other entities can't just start archiving 
the DKIM public keys. Moreover, even if you rotate the keys, that merely 
downgrades the DKIM information from being a cryptographically authenticated 
super cookie to an unauthenticated super cookie (like the ones Verizon was 

The comments I made that started this -- cryptographically weak DKIM -- is also 
a partial solution at best. But I want to swing back to my point, which is that 
we crypto people have a tendency to look at crypto parameters and just blithely 
start using them everywhere, looking at the surface structure of what we're 
doing and not the deep structure.

Part of the deep structure is that DKIM had cryptographic authentication 
post-delivery as something between an non-goal and an anti-goal. We didn't want 
to have to force MUAs to have to play with them, and didn't even really *want* 
MUAs to do it, certainly not long term. Yahoo had plans to show DKIM status in 
their web client, but only for a short term. Yes, your email came from your 
bank. Moreover, this is why DKIM is really important for banks, and not at all 
important for domains like 

Back in pre-Snowden days, it was hard to argue this sort of privacy / metadata 
/ traffic analysis stuff without looking like a looney. I also realize that 
DKIM is not a *mere* tracking cookie, it serves and succeeds at improving the 
public health of the Internet. My wry points about weak-keyed DKIM are there in 
part to look at the whole issue. The good that DKIM provides is not improved by 
8K keys. It's not improved by 4K keys, or even 2K keys (at the time of this 
writing). That's because a DKIM key needs to be alive for no more than a week, 
and in the vast majority of cases not even an entire minute! The threat model 
of DKIM is also such that if someone managed to compromise a DKIM key, they are 
only able to authenticate a spam message. 

Yes, yes, I know that that "only" could empty someone's bank account, but the 
defense against that adversary is a defense-in-depth strategy which has a 
series of filters and checkpoints and the whole point of defense in depth is 
that each screen plays a part in the filter as a whole. My point is that DKIM 
is designed to be effective with lower levels of cryptographic security and 
that's part of why it's so cool! Increasing the crypto in just about *any* 
system is strengthening the strongest link in the chain. The DKIM link in the 
anti-fraud, anti-spam email defense does not benefit with even 2K keys. It 
benefits by having operations management rotate those keys and having 
operations management look at the broader problem of cruft in our basement of 
email -- like the way headers in general add to the surveillance-friendly 
problems of email stores and list archives.

Like many security issues, crypto is only part of the complete breakfast. DKIM 
headers are an issue because they provide long-term cryptographic surveillance. 
But the surveillance tags are there in non-cryptographic places, too, and it is 
wrong to say that we don't need to fix them because well, the park is still 
dirty after we clean up our litter. We must clean up our litter and get others 
to clean theirs up, too.


Version: PGP Universal 3.3.0 (Build 9060)
Charset: utf-8


NOTE WELL: This list operates according to

<Prev in Thread] Current Thread [Next in Thread>