Re: [bitcoin-dev] bustapay BIP :: a practical sender/receiver coinjoin protocol

From: Adam Gibson <ekaggata@gmail•com>
To: bitcoin-dev@lists•linuxfoundation.org
Subject: Re: [bitcoin-dev] bustapay BIP :: a practical sender/receiver coinjoin protocol
Date: Fri, 25 Jan 2019 15:47:34 +0100	[thread overview]
Message-ID: <e15c5dd7-6fe1-b253-e129-aeae6493acd1@gmail.com> (raw)
In-Reply-To: <TtjH2zicjKr8PBVCMOvA7ryt2z_XXvtrpC4y1wuWSxexNwMdbPGE7vPmu6UnzmfYqYBMxZ8NNoz4VUnODdIcjR4j-E1sYz_FA6ZZMjKHtuM=@protonmail.com>

Ryan and list,
I want to add some commentary to this (BIP79) to see if we can get
further in standardizing this idea.

When I first mulled it over I thought it too impractical, but its virtue
of steganographic hiding means only minimal uptake is still enormously
interesting and worth pursuing; that's my current feeling. I've offered
more detailed thoughts in my blog post[1] (def not required reading here).

Both Joinmarket and Samourai have started implementing this kind of
transaction. And while that's interesting experimentally, some kind of
cross-wallet standard would be helpful, albeit there some differences
between that and the merchant/centralized service use-case.

We might imagine as a concrete goal for this BIP to create something
that would be acceptable for inclusion into a project like BTCPayServer,
so that it could be used in a realistic use case by smaller bitcoin
accepting merchants.

Comments to the BIP[2] as follows, with generic comments first, and then
specific comments for existing points in the BIP:

[1] https://joinmarket.me/blog/blog/payjoin
[2] https://github.com/bitcoin/bips/blob/master/bip-0079.mediawiki

Generic comments
============
* Protocol versioning. Since inevitably (even if only merchants), this
must be implemented by multiple wallets to be useful, the communication
protocol will need versioning (for example i have in my
simple/experimental Joinmarket PayJoin that sender sends min and max
supported version and receiver responds with a chosen protocol version
so we can update). I do understand that as a client-server model can
apply here, we can ditch a lot of the complexities around network/p2p
interaction, but this much at least seems necessary.
* Although it has its logic, I don't think "Bustapay" is a good name for
this protocol. I prefer "PayJoin" which is neutral sounding and
self-descriptive. Needless to say this is not a hill I intend to die on.
* PSBT/BIP174. I realise this has already been discussed, but this is a
good example of what this standardisation was designed for, so I'd be
against not including it, even given the reality that, as you correctly
observe, it is not yet implemented in the majority of wallets and
libraries. One way round that is to make it optional (possibly combined
with above point about versioning). Note that for example you were
observing the necessity to check the sequence number was unchanged; that
would be encapsulated by checking equality of PSBT Input objects/fields.
While one can make such software architecture arguments, the really
fundamental point is the need for standards for x-wallet compatibility.
* Version, Locktime: Perhaps this is not needed; in a peer to peer
wallet scenario I think there might be logic in trying to get cover
traffic of (Core, Electrum, others), say, by using
last-block-locktime-mostly, as they do. Version should be 2 and sequence
is a function of your suggestion to use BIP125. Worth noting that BIP125
is *not* currently widely used on the network, though (see
https://p2sh.info/dashboard/db/replace-by-fee?orgId=1). For this reason
it should perhaps be explicitly only optional.
* Avoidance of non-payment "Unnecessary Input Heuristic" (1, 2). For
reference, see the definition here
https://gist.github.com/AdamISZ/4551b947789d3216bacfcb7af25e029e#gistcomment-2796539
and some data here
https://gist.github.com/AdamISZ/4551b947789d3216bacfcb7af25e029e#gistcomment-2800791
(whole comment thread may be of interest) - note this UIH name is afaik
Chris Belcher's invention, it seems useful as a categorisation.
 So, it seems that UIH2 is more important to avoid; while some more
sophisticated wallet coin selection algorithms *may* occasionally pick
an input set where one input is larger than any output, most won't, and
some in particular never will. So I think the text here should indicate
that *the receiver's contributed input(s) SHOULD be chosen to avoid
triggering the UIH2 heuristic where possible, so that the final payjoin
transaction is maximally plausible as an ordinary payment" or similar.
UIH1 is a nice-to-have (meaning the plausibility extends to two
different (both wrong) payment amounts, but it may not be necessary to
mention it in the BIP.

Specific comments
=================

>> ====Step 4. Receiver validates, re-signs, and propagates on the
bitcoin network====

I believe this should say "Sender" not Receiver. Also for the next
sentence, s/receiver/sender/:

>> The receiver MUST validate the ''partial transaction'' was changed
correctly and non-maliciously (to allow using potentially untrusted
communication channels), re-sign its original inputs and propagate the
final transaction over the bitcoin network.

Your very correct highlighting of the attack vector of "receiver gives
sender other inputs belonging to sender to unwittingly sign (described
below), should be highlighted here, perhaps with the phrase "re-sign its
ORIGINAL inputs" (only!)".

>> When the sender is creating a "template transaction" it is done
almost identically to creating a normal send, with the exception that
*only* segwit inputs may be used. The sender is also encouraged to use a
slightly more aggressive feerate than usual as well as BIP125 (Opt-in
Full Replace-by-Fee Signaling), but neither is strictly required.

"slightly more aggressive feerate than usual" - this I understand is to
make up for receiver contributed utxo, OK.

"*only* segwit inputs" - it certainly makes things simpler. One can work
with non-segwit inputs but especially considering (as mentioned below)
we really ought to "MUST" the part about matching input types, I tend to
agree that non-segwit should be disallowed.

>> The receiver must add at least one input to the transaction (the
"contributed inputs"). If the receiver has no inputs, it should use a
500 internal server error, so the client can send the transaction as per
normal (or try again later).

Would it not be much simpler for the server to return a different
(non-error) response indicating that it will broadcast the template tx
in this case?

>> Its generally advised to only add a single contributed input, however
they are circumstances where adding more than a single input can be useful.

I don't see a good reason to advise the use of only 1 input? (but this
will also connect with the above generic comment about "UIH"). I guess
it's because of your approach to fees. I'd prefer not to create a
limitation here.

>> To prevent an attack where a receiver is continually sent variations
of the same transaction to enumerate the receivers utxo set, it is
essential that the receiver always returns the same contributed inputs
when it's seen the same inputs.

This is an approach to avoiding this problem which has the virtue of
simplicity, but it seems a little problematic. (1) You must keep a
mapping of proposed payment utxos to one's proposed contributed input
utxos, but (2) how should this be updated if you need to spend the
contribution mentioned in (1)? Ironically use of payjoin exacerbates
this issue, because it results in a smaller number of utxos being held
by the receiver at any one time :) All this considered, I still see the
value in your approach, but it might end up with a re-attempted payment
being rejected. Certainly the more complex suggested solutions coming
out of the summer 2018 coinjoin workshop aren't as practical as this,
and may be overkill for small merchants/receivers.

>> It is strongly preferable that the receiver makes an effort to pick a
contributed input of the same type as the other transaction inputs if
possible.

I have also thought about this and you could reasonably argue this
should be a MUST section in the BIP, that is, if the receiver cannot use
inputs of the same type, he should fall back to the template
transaction. A mixed-input payjoin/coinjoin is essentially
near-perfectly identifiable as such (there is almost zero other usage of
multi-type-input transactions), which is a very different thing than a
non-identifiable payjoin transaction. That may or may not be OK to the
sender. This is debatable though, for sure.

>> After adding inputs to the transaction, the receiver generally will
want to adjust the output that pays himself by increasing it by the sum
of the contributed input amounts (minus any fees he wants to
contribute). However the only strict requirement is that the receiver
*must never* add or remove inputs, and *must not* ever decrease any
output amount.

"*must never* add or remove inputs" - did you mean "must never remove
inputs"? he surely has to add one! Or, perhaps you mean he must not
alter the list of inputs provided by the sender (in which case it should
be clarified).

"must not decrease any output amount" - I initally disagreed with this
but it is a better option than the one I currently chose in Joinmarket
payjoin (sender pays all fee as long as receiver utxos are not too
much). So this means that the receiver either consciously chooses to not
increase the fee, meaning the fee rate may be a bit low (hence your
earlier comment about being generous, got it), or contributes via the
payout amount. I guess the latter might break merchant software
expecting to have amount output fixed and fees determined by change.

Regards,
Adam Gibson/waxwing

On 30. 08. 18 22:24, Ryan Havar via bitcoin-dev wrote:
> I've just finished writing an implementing of this, and extremely happy
> with how it turned out. So I'd like to go and try go down the path of
> more formally describing it and getting some comments and ultimately
> encourage its wide-spread use.
> 
> 
> ==Abstract==
> 
> The way bitcoin transactions are overwhelming used is known to leak more
> information than desirable. This has lead to fungibility concerns in bitcoin
> and a raise of unreasonably effective blockchain analysis.
> 
> Bustapay proposes a simple, practical way to bust these assumptions to
> immediate
> benefit of the sender and recievers. Furthermore it does so in such a
> way that
> helps recievers avoid utxo bloat, a constant problem for bitcoin merchants.
> 
> ==Copyright==
> 
> This BIP is in the public domain.
> 
> ==Motivation==
> 
> One of the most powerful heuristic's employed by those whose goal is to
> undermine
> bitcoin's fungiblity has been to assume all inputs of a transaction are
> signed by
> a single party. In the few cases this assumption does not hold, it is
> generally
> readibly recognizable (e.g. traditional coinjoins have a very obvious
> structure,
> or multisig outputs are most frequently validated onchain).
> 
> Bustapay requires no changes to bitcoin and creates bitcoin transactions
> that are
> indistinguishable from normal ones.
> 
> It is worth noting that this specification has been intentionally kept
> as simple
> as possible to encourage adoption. There are almost an endless amount of
> extensions
> possible but the harder the implementation of clients/server the less
> likely it
> will ever be done. Should bustapay enjoy widespread adoption, a "v2"
> specification
> will be created with desired extensions.
> 
> ==Specification==
> 
> A bustapay payment is made from a sender to a receiver.
> 
> Step 1. Sender creates a bitcoin transaction paying the receiver
> 
> This transaction must be fully valid, signed and all inputs must use
> segwit. This transaction is known as the "template transaction". This
> transaction must not be propagated on the bitcoin network.
> 
> Step 2. Sender gives the "template transaction" to the receiver
> 
> This would generally be done as an HTTP POST. The exact URL to submit it
> to could be specified with a bip21 encoded address. Such as
> bitcoin:2NABbUr9yeRCp1oUCtVmgJF8HGRCo3ifpTT?bustapay=https://bp.bustabit.com/submit
> and the HTTP body should be the raw transaction hex encoded as text.
> 
> Step 3. Receiver processes the transaction and returns a partially
> signed coinjoin
> 
> The receiver validates the transaction is valid, pays himself and is
> eligible for propation. The receiver then adds one of his own inputs
> (known as the "contributed input") and increase the output that pays
> himself by the contributed input amount. Doing so will invalidate the
> "template transaction"'s original input signatures, so the sender needs
> to return this "partial transaction" back to the receiver to sign. This
> is returned as a hex-encoded raw transaction a response to the original
> HTTP POST request.
> 
> Step 4. Receiver validates, re-signs, and propagates on the bitcoin network
> 
> The receiver is responsible in making sure the "partial transaction"
> returned by the sender was changed correctly (it should assume the
> connection has been MITM'd and act accordingly), resign its original
> inputs and propagates this transaction over the bitcoin network. The
> client must be aware that the server can reorder inputs and outputs.
> 
> Step 5. Receiver observes the finalized transaction on the bitcoin network
> 
> Once the receiver has seen the finalized transactions on the network
> (and has enough confirmations) it can process it like a normal payment
> for the sent amount (as opposed to the amount that it looks like on the
> network). If the receiver does not see the finalized transaction after a
> timeout will propagate the original "template transaction" to ensure the
> payment happens and function a strong anti-DoS mechanism.
> 
> 
> === Implementation Notes ===
> For anyone wanting to implement bustapay payments, here are some notes
> for receivers:
> 
> * A transaction can easily be checked if it's suitable for the mempool
> with testmempoolaccept in bitcoin core 0.17
> * Tracking transactions by txid is precarious. To keep your sanity make
> sure all inputs are segwit. But remember segwit does not prevent txid
> malleability unless you validate the transaction. So really make sure
> you're using testmempoolaccept at the very least
> * Bustapay could be abused by a malicious party to query if you own a
> deposit address or not. So never accept a bustapay transaction that pays
> an already used deposit address
> * You will need to keep a mapping of which utxos people have showed you
> and which you revealed. So if you see them again, you can reveal the
> same one of your own
> * Check if the transaction was already sorted according to BIP69, if so
> ensure the result stays that way. Otherwise probably just shuffle the
> inputs/outpus
> 
> Notes for sending applications:
> 
> * The HTTP response must *not* be trusted. It should be fully validated
> that no unexpected changes have been made to the transaction.
> * The sender should be aware the original "template transaction" may be
> propagated at any time, and in fact can intentionally be
>   done so for the purpose of RBF as it should have a slightly higher fee
> rate.
> 
> == Credits ==
> The idea is obviously based upon Dr. Maxwell's seminal CoinJoin
> proposal, and reduced scope inspired by a simplification of the "pay 2
> endpoint" (now offline) blog post by blockstream.
> 
> 
> -Ryan
> 
> 
> 
> _______________________________________________
> bitcoin-dev mailing list
> bitcoin-dev@lists•linuxfoundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev
> 