[bitcoin-dev] Safer NOINPUT with output tagging

[Johnson Lau]

> On 21 Dec 2018, at 1:20 AM, Christian Decker <decker.christian at gmail.com> wrote:
> 
> Johnson Lau <jl2012 at xbt.hk> writes:
>> Correct me if I’m wrong.
>> 
>> For the sake of simplicity, in the following I assume BIP118, 143, and
>> 141-P2WSH are used (i.e. no taproot). Also, I skipped all the possible
>> optimisations.
>> 
>> 1. A and B are going to setup a channel.
>> 
>> 2. They create one setup tx, with a setup output of the following
>> script: <s> CLTV DROP 2 Au Bu 2 CHECKMULTISIG. Do not sign
> 
> If we are using a trigger transaction the output of the setup
> transaction would simply be `2 Au Bu 2 OP_CMS`. If we were to use a CLTV
> in there we would not have an option to later attach a collaborative
> close transaction that is valid immediately. Furthermore the timeout of
> the CLTV would start ticking down the exact moment the setup transaction
> is confirmed, hence whatever effect we are trying to achieve with that
> timelock is limited, and we have a limit to the total lifetime of the
> channel.

CLTV is absolute locktime. Only CSV will have the “time ticking” issue, but that’s not used here. The required locktime <s> is many years in the past. To collaboratively close, you just need to sign with SIGHASH_ALL, with a locktime s+1.

> 
>> 3. They create the update tx 0, spending the setup output with NOINPUT
>> and locktime = s+1, to the update-0 output with the script: IF 2 As0
>> Bs0 2 CHECKMULTISIG ELSE <s+1> CLTV DROP 2 Au Bu 2 CHECKMULTISIG ENDIF
> 
> Update 0 is usually what I call the trigger transaction. It takes the
> 2-of-2 multisig from the setup transaction and translates it into the
> two-branch output that further updates or settlements can be attached
> to. The settlement transaction attached to the trigger / update 0
> reflects the initial state of the channel, i.e., if A added 2 BTC and B
> added 1 BTC then settlement 0 will have 2 outputs with value 2 and 1
> respectively, with the user's keys (this can also be considered the
> refund in case of one party disappearing right away).
> 
> The second branch in the script you posted is the update branch, which is
> not encumbered by a CSV, while the first branch is the one encumbered
> with the CSV and is called the settlement branch since we'll be
> attaching settlement txs to it.
> 
> The CLTV looks correct to me and ensures that we can only attach any
> state >= s+1.
> 
> So just to show the output script for state `i` how I think they are
> correct:
> 
> ```
> OP_IF
>  <timeout> OP_CSV 2 <As_i> <Bs_i> 2 OP_CHECKMULTISIG
> OP_ELSE
>  <s+1> OP_CLTV OP_DROP 2 <Au> <Bu> 2 OP_CHECKMULTISIG 
> ```
> 
> And the input scripts for the update tx and the settlement tx
> respectively would be:
> 
> ```
> OP_FALSE <Sig_Bu> <Sig_Au>
> ```
> 
> and
> 
> ```
> OP_TRUE <Sig_Bs_i> <Sig_As_i>
> ```

I think the use of OP_CSV (BIP112) is not needed here (although it doesn’t really harm except taking a few more bytes). All you need is to sign the settlement tx with a BIP68 relative locktime. Since this is a 2-of-2 branch, both parties need to agree with the relative locktime, so it is not necessary to restrict it through OP_CSV


> 
>> 4. They create the settlement tx 0, spending the update-0 output with
>> As0 and Bs0 using BIP68 relative-locktime, with 2 settlement outputs
> 
> If I'm not mistaken the CSV needs to be in the scriptPubkey (or P2WSH
> equivalent) since segwit witnesses only allow pushes. Hence the script
> in point 3 needs to add that :-)

I believe you confused OP_CSV (BIP112) with BIP68. Relative locktime is enforced by BIP68 (i.e. setting the nSequence). OP_CSV indirectly enforces relative-locktime by checking the value of nSequence. BIP68 could work standalone without OP_CSV, while OP_CSV is dependant on BIP68. In the case of n-of-n eltoo state update, OP_CSV is not needed because all n parties need to agree with the same nSequence value of the settlement tx. This is enough to make sure the settlement tx has delayed settlement.

> 
>> 5. They sign the setup tx and let it confirm
> 
> They also need to sign (but not broadcast) update_0, in order to allow
> either party to initiate the closure if the counterparty become
> unresponsive. The order in which settlement_0 and update_0 are signed is
> not important by the way, so we can just batch these. The important part
> is that signing the setup acts as a commitment.

Sure. This is obvious.

> 
>> 6. To update, they create the update tx 1, spending the setup output
>> with NOINPUT and locktime = s+2, to the update-1 output with the
>> script: IF 2 As1 Bs1 2 CHECKMULTISIG ELSE <s+2> CLTV DROP 2 Au Bu 2
>> CHECKMULTISIG ENDIF and create the settlement tx 1, spending the
>> update-1 output with As1 and Bs1 using relative-locktime, with 2
>> settlement outputs
> 
> The output script of the updates are identical to the ones in the
> trigger or update_0 transaction, so they'd also need a CSV (this is why
> committing to the script structure with masking still works).
> 
>> 7. To close the channel, broadcast update tx 1. Wait for several
>> confirmations. And broadcast settlement-tx-1
> 
> We have to differentiate 2 cases: collaborative close and unilateral
> close. In the collaborative close we come to a mutual agreement that
> we'd like to take this latest state and settle. So we create a new
> transaction that spends the setup output, and add outputs according to
> the state we agreed upon, and we sign it. This transaction is
> immediately valid, and does not need to be signed with NOINPUT. So all
> the chain sees is a setup transaction with some inputs and one multisig
> output (singlesig with Schnorr) and a collaborative close transaction
> that spends the setup (also not signed with NOINPUT). About as normal as
> transactions in Bitcoin can get.

Collaborative close is always simple as I explained in the beginning

> 
> In the unilateral case, one party isn't there anymore, or refuses to
> sign. So we take the trigger transaction (not signed with NOINPUT) and
> the latest update_n transaction (signed with NOINPUT) and broadcast
> them. Then we wait for the CSV timeout to expire, and then send the
> settlement transaction, which gives us the enforcement of the latest
> state that we agreed on. The chain sees a setup transaction and a
> trigger transaction (normal transactions for all intents and purposes,
> except for the output script of the trigger, but we can hide that with
> taproot), followed by two more transactions which are signed with
> NOINPUT. So 4 transactions in the worst case, of which 2 are special,
> and 2 transactions in the good case.
> 
> 
> So all in all I think it's a tradeoff between having a larger on-chain
> footprint (4 txs vs 3 txs in the worst case) and putting a fixed
> lifetime on the channel for the refund case if one party disappears
> right away. We'll probably find out what acceptable parameters are for
> these and where the cutoff points are :-)

If no one is cheating (i.e. only the last update is broadcast), you always need only 3 txs. Think about this: every update tx could be a trigger tx, and you can settle directly on a trigger tx, so effectively you eliminate trigger tx.