MIME-Version: 1.0
From: Riccardo Spagni <ric@spagni.net>
Date: Fri, 17 Jul 2015 10:00:13 +0200
Message-ID: <CADhj2oQ-4oOAYkVpSRV6nCFN12AFkZ=O6Fof8YVNeKtrQ-t28Q@mail.gmail.com>
To: bitcoin-dev@lists.linuxfoundation.org
Content-Type: text/plain; charset=UTF-8
Subject: Re: [bitcoin-dev] Proposal: extend bip70 with OpenAlias
Precedence: list

> I appreciate the thought :)  I think where we differ is on where we believe the
> trade offs should be on perceived privacy versus censorship resistance and
> centralization.
>
> By having a limited number of proxies people need to go through to easily
> implement, be it the 4 you recommend, or 53, you actually have a very limited
> number of actors for an authority or hacker to go to in order to be able to
> install/force logging, or censorship.  This very centralization forces us back
> to a model where we need to trust a very small number of actors in order for
> the system to operate as designed.  This, to me, appears to be the opposite of
> the goals of the bitcoin ecosystem.  To ensure this point is clear, I strongly
> believe recommending people focus all lookups through 4 centralized "proxies"
> is a bad idea and counter to bitcoin's ideals.
>
> The fact that hackers or state actors need to corrupt only a small number of
> servers/services in order to gain global visibility into all queries, I
> believe, breaks any perceived privacy gains from using DNSCrypt.  A very small
> number of hacks or subpoenas and everyone's records are fair game in one place.

You're misstating (or not understanding) the attack surface.

State-level attackers won't compromise 50+ DNSCrypt servers, they can get the
information on lookups a lot more trivially. Censorship resistance and
protection from state-level attackers comes from the decentralised side of
OpenAlias (ie. Namecoin resolution, preferably done using a local copy of the
NMC blockchain). Since Netki supports Namecoin resolution too there is no need
to worry about protecting end users from that.

There is, however, a need to protect users from man-in-the-middle attacks where
the data is not modified en-route, but it is sniffed. Who you pay in a financial
transaction is, and should be, privileged information between yourself and that
person. By encouraging open DNS lookups you're effectively hanging that
information out for all to see.

It is true that there are only 4 DNSCrypt servers we are comfortable
recommending. It is also true that there were, at one stage, only 4 Electrum
servers. There were also only 4 Bitcoin nodes. As something grows and becomes
more useful and usable the number of voluntary participants becomes much
greater, and we will provide the necessary tools to enable these volunteers.

So in a world where tens of thousands of Bitcoiners are using an aliasing
standard (which, in and of itself, is a convenience service anyway), and
hundreds of individuals and companies are hosting DNSCrypt resolvers, is it even
a valid argument to harp on the number of "proxies"? Thus it is not worth
talking about today. It is definitely worth discussing in future if the number
of DNSCrypt resolvers doesn't increase, but that is a different discussion for a
different time.

> For the highly privacy conscious they can, today do their DNS lookups over a
> non logging VPN connection without forcing everyone else through a handful of
> centralized servers.  Or they can use DNSCrypt optionally themselves.  All of
> our tools have always been open source and folks can modify them for their own
> desired uses, or submit pull requests with their own ideas.

Everyone should be highly privacy conscious when it comes to financial
transactions, and it would be unconscionable of both you and I not to defend
end-user privacy.

> We'd love to hear others thoughts on this.  While I believe that for now the
> centralization trade offs required to use DNSCrypt today (via a limited number
> of proxies) outweigh any perceived privacy benefits it provides, we are always
> open to what others in the community believe and have made modifications to how
> things work before as a result of feedback from industry participants.

It's important to remember that the "paranoid" won't use an aliasing service, or
at best will use a local Namecoin blockchain for that purpose. This is a
convenience service to provide general and broad appeal for the non-technical,
and those are the very people that need to be protected from nosey neighbours /
workmates / ISPs. Privacy is not only (or even at all) about protecting people
buying drugs on a darknet market, it is about defending personal liberties.

> I think DANE is a great idea.  We were just discussing that with Andreas S.,
> and are currently looking at whether we want to add this as optional versus
> mandatory, based on how widely available DANE is for folks using services like
> Cloudflare, Akamai, etc for their DNS, which many providers in the space today
> are.
>
> Of course, the security conscious could setup DANE on the URL we use AS IS.
> There is no need to create a special kv pair for this as is done in OpenAlias.
> As you know, DNSSEC and HTTPS support this today out of the box.

Embedding the TLSA record in a KV pair just means that pinning takes one less
step.

> The CA validation, in our case, is an ADDITIONAL signature based validation to
> the DNSSEC chain, not a replacement for it.

Without DANE it's a weakness. It's trusting an additional CA (over and above the
domain owner), when we know that this is - and has been - an issue in the past.
Were it not an issue DANE (or certificate pinning in general) would not have to
exist.

> We looked at doing this in a single lookup as you did.  With one or two
> currencies this can be potentially more efficient.  As the number of supported
> currencies and addresses under a single name grows, however, this solution
> becomes potentially more problematic.  Please follow the use cases below:

(snipped quote for brevity)

> Many currencies and colored coin addresses are supported under the same name,
> lets say 100.  When you count different currencies and colored coin types, it
> could easily be hundreds, or over a thousand.

Coinmarketcap lists 643 currencies and assets, of which only 131 have had more
than $500 in trade volume in the last 24 hours (and only 8 have done over $100
000 in volume). ShapeShift only lists 44 of those. I seriously doubt that a
convenience service such as aliasing will find great use amongst every
fly-by-night scamcoin that crops up, but that is an aside.

> While you may end doing "less lookups" under Open Alias, as it scales, you end
> up causing a significant amount of extra, unnecessary traffic.

"Scale" is a misnomer. Someone trying to collect every single active
cryptocurrency and house them all under a single sub-domain is an outlier, not a
problem to be faced at scale. I do not think we will see a large scale movement
to "collect" all the various cryptocurrency tokens, no matter how worthless they
are, and then subsequently setup aliases for them.

> In addition to the obvious impact of being orders of magnitude more wasteful
> than necessary, it also creates privacy "leakage" by returning someone 100
> different addresses when they only asked for one.

I'm not sure how this is any greater leakage than 100 individual requests for
the openly accessible data, especially since it would be encrypted if made via
DNSCrypt?

> Finally, because a single packet UDP transaction for a DNS lookup can create
> possibly hundreds of packets in response, the service can essentially become an
> amplifier for DDoS attacks.  (If I spoof the source address of my target with a
> query to a lookup that issues hundreds of packets in response to one packet,
> and I can have a real impact :( )

Naughty naughty, you're doing that thing again where you're using a smattering
of expertise to appear knowledgeable about a subject.

So let's hypothetically say that an individual was crazy enough to have all 643
of the Coinmarketcap currencies/assets aliased on a single sub-domain. The
OpenAlias example of a Monero address (with a recipient name) is 157 bytes long,
due to there being two public keys serialised in the address, plus the ~12 bytes
of overhead per RR (the DNS wire format uses label compression, so the FQDN
wouldn't be repeated for each returned record). Let's call it 170 bytes. That
makes the returned data just over 100kb.

Now let's first address a couple of things, assuming that someone would be nuts
enough to do this:

1. This is way larger than the UDP packet maximum, and this would never come
back as a "regular" ol' DNS request (512 bytes max). This may seem bad, until
you consider that DNSSEC responses are almost assured to exceed 512 bytes (eg.
an NXDOMAIN with NSEC3). The size of the response is big, but that's hardly
something to write home about.

2. If the DNS server supports RFC2671 (EDNS) then it would try and send it via
UDP, and as long as the client says it can receive such a huge response over UDP
it'll come over the wire.

3. However, because RFC2671 can result in a DNS amplification attack, it's been
obsoleted by RFC6891 (EDNS0), which is pretty much ubiquitous for all resolvers
that support DNSSEC (because of the very large DNSSEC responses, and the fact
that DNSSEC resolvers want to avoid participation in an amplification attack).
EDNS0 mitigates amplification attacks.

4. In the event that an EDNS0 response fails (eg. the client says it can't
accept anything over, say, 4kb, which is quite common) then there's an automatic
and silent switch to DNS-over-TCP (RFC5966). DNS-over-TCP uses TFO (TCP Fast
Open) to do an extremely fast handshake and passing a cookie to the client in
the SYN-ACK, which can then be used for subsequent requests, but data is still
carried in the SYN. TFO mitigates amplification attacks.

You can't both be overly concerned about amplification attacks *and* use DNSSEC,
which necessitates large records. And, at any rate, the issue with amplification
attacks *isn't* the size of the records (there are tons of records just under
4kb, like an ANY request against isc.org, that are far better suited to
amplification attacks), it is the number of recursive open resolvers. There is
improvement in this space, though, and many open recursive resolvers have been
fixed in recent years.

> It is important to note, that ICANN has "required" for some years that
> registrars and registries support DNSSEC on the domains they register.  I
> personally believe we shouldn't delay use of DNSSEC until their registries had
> come up to current required Internet standards.  (Here are ICANN's registrar
> requirements showing the DNSSEC requirement, btw:
> https://www.icann.org/resources/pages/approved-with-specs-2013-09-17-en#
> operation)

Doesn't really matter what they require as long as there are zones that remain
unsigned. Plus it's not like new .za / .sg / .to registrations magically get
DNSSEC, they're also out of luck.

> That said, what do others in the industry think?  We are basing our current
> standard on our believed best practices, and defaulted to "first, lose no
> money", given the irreversibility of bitcoin.

Oh, ok. "First, lose no money, but it's ok if your ISP / neighbour / colleague
reports you to the cops because you sent a donation somewhere you shouldn't."

> I think "DNSSEC is hard" is a bit of a boogey man that's not really true.  We
> are working on developing registrar by registrar instructions of how to do
> this, and we have typically found that if you are setting up DNS by yourself,
> adding DNSSEC doesn't take a lot of additional time, maybe an hour or so
> depending on your registrar.

Adding the DS record to a domain is trivial, but to use DNSSEC with Gandi or
GoDaddy (if you don't have their PremiumDNS product) you have to host your own
DNS server. Sorry, but that is a non-trivial task. Even worse: you need to
secure your private KSK and not keep it on the server, and if Bitcoiners are
anything to go by this won't happen.

Oh, and incidentally, ENOM/Namecheap doesn't have DNSSEC support yet.

You're literally layering complexity on top of a convenience service, and to
what end?

> This known concern, however, is why when we launched our product (based on our
> standard record formats) that we wanted to launch it with a variety of options
> for people.

That's completely, 100% centralised. You're creating decentralisation theatre by
providing "options" that no ordinary person will use.

> That's some interesting data, and runs counter to the research of the IETF DNS
> working group.  If you are willing to share your data, I can put you in touch
> with the appropriate folks there to share your research.  I'd also love to see
> it!

I doubt that very much. See:
http://stats.labs.apnic.net/dnssec/XA?c=XA&x=1&g=1&r=1&w=30

As can be seen, only ~14% of all DNS queries request DNSSEC validation. That's
very far from ubiquitous, and completely matches what Thomas and I found in
Berlin. Unsurprisingly, this stat is particularly bad given that it also shows
that ~15% of all queries are being handled by Google's Public DNS, without which
the stat would be much lower.

> I'd argue that we aren't locking "huge portions" of the Internet.  You are
> correct that about 15% of TLD's are not yet signed, even though they were
> required to be by ICANN.

Fine, so we're just cutting Africa out, then?
http://www.internetsociety.org/deploy360/wp-content/uploads/2015/06/cctlddnssec-
2015-06-19.pdf

Even beyond that, ICANN's page listing DNSSEC-capable registrars (last updated
December 2014) only lists only a handful:
https://www.icann.org/resources/pages/deployment-2012-02-25-en

> As I said above, I believe the requirement to not lose money and the fact that
> other options are available for those running on TLD's that are out of
> compliance, is worth the trade off that some existing names won't work until
> their TLD's come into compliance with current Internet standards.

A soft fail doesn't magically let the money go. It warns users of the risk and
asks them to verify the address by site. This could even be built out so that
higher value transactions (say, anything over $1000) hard fails in the absence
of DNSSEC, and anything particularly high value (say, anything over $50 000)
refuses to use an alias at all and requires an actual cryptocurrency address.

> I'm a little confused by these closing statements.  Our system has, from the
> beginning been open in terms of the fact that anyone could both serve names or
> do lookups without ever touching our servers, talking to us, or us even knowing
> that they did it or that they even exist.  Our system has NEVER been one where
> folks were required to use us for any portion of the service, and from our
> first beta product launch our open source tools did all lookups against DNS
> records and the blockchain, never any proprietary servers or interfaces on our
> side.

Now you're just trolling.

https://github.com/bitpay/copay/pull/2431/files

Which has this lovely line in it:
https://github.com/wdawg33/copay/blob/be6c3e80ab7601d245b186f7802d7050992eb1f0/
config.js#L98

So you provide an open standard that uses DNS...but then you wanted to force
CoPay users to use your centralised API?

> I'd love to know where you got information that we were in some way a closed
> and centralized system so that we can have an opportunity to clarify that
> misconception.

In December 2014 your website had no "developer" section (curl -s
https://web.archive.org/web/20141221141023/https://netki.com/ | grep
"Developers")

The first time that section got scraped was the end of April:
https://web.archive.org/web/20150428231016/https://www.netki.com/partials/
developers.html

Even in its current form your website makes no mention of alternatives or
options for those wishing to secure an alias. End users are undoubtedly left
with the distinct impression that they can only get one by paying you.

Riccardo