[bitcoin-dev] Ephemeral Anchors: Fixing V3 Package RBF against package limit pinning

[Greg Sanders]

Hello Everyone,

Following up on the "V3 Transaction" discussion here
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2022-September/020937.html
, I would like to elaborate a bit further on some potential follow-on work
that would make pinning severely constrained in many setups].

V3 transactions may solve bip125 rule#3 and rule#5 pinning attacks under
some constraints[0]. This means that when a replacement is to be made and
propagated, it costs the expected amount of fees to do so. This is a great
start. What's left in this subset of pinning is *package limit* pinning. In
other words, a fee-paying transaction cannot enter the mempool due to the
existing mempool package it is being added to already being too large in
count or vsize.

Zooming into the V3 simplified scenario for sake of discussion, though this
problem exists in general today:

V3 transactions restrict the package limit of a V3 package to one parent
and one child. If the parent transaction includes two outputs which can be
immediately spent by separate parties, this allows one party to disallow a
spend from the other. In Gloria's proposal for ln-penalty, this is worked
around by reducing the number of anchors per commitment transaction to 1,
and each version of the commitment transaction has a unique party's key on
it. The honest participant can spend their version with their anchor and
package RBF the other commitment transaction safely.

What if there's only one version of the commitment transaction, such as in
other protocols like duplex payment channels, eltoo? What about multi party
payments?

In the package RBF proposal, if the parent transaction is identical to an
existing transaction in the mempool, the parent will be detected and
removed from the package proposal. You are then left with a single V3 child
transaction, which is then proposed for entry into the mempool. In the case
of another parent output already being spent, this is simply rejected,
regardless of feerate of the new child.

I have two proposed solutions, of which I strongly prefer the latter:

1) Expand a carveout for "sibling eviction", where if the new child is
paying "enough" to bump spends from the same parent, it knocks its sibling
out of the mempool and takes the one child slot. This would solve it, but
is a new eviction paradigm that would need to be carefully worked through.

2) Ephemeral Anchors (my real policy-only proposal)

Ephemeral Anchors is a term which means an output is watermarked as an
output that MUST be spent in a V3 package. We mark this anchor by being the
bare script `OP_TRUE` and of course make these outputs standard to relay
and spend with empty witness data.

Also as a simplifying assumption, we require the parent transaction with
such an output to be 0-fee. This makes mempool reasoning simpler in case
the child-spend is somehow evicted, guaranteeing the parent will be as well.

Implications:

a) If the ephemeral anchor MUST be spent, we can allow *any* value, even
dust, even 0, without worrying about bloating the utxo set. We relax this
policy for maximum smart contract flexibility and specification simplicity..

b) Since this anchor MUST be spent, any spending of other outputs in the
same parent transaction MUST directly double-spend prior spends of the
ephemeral anchor. This causes the 1 block CSV timelock on outputs to be
removed in these situations. This greatly magnifies composability of smart
contracts, as now we can do things like safely splice directly into new
channels, into statechains, your custodial wallet account, your cold
wallet, wherever, without requiring other wallets to support arbitrary
scripts. Also it hurts that 1 CSV time locked scripts may not be miniscript
compatible to begin with...

c) *Anyone* can bump the transaction, without any transaction key material.
This is essentially achieving Jeremy's Transaction Sponsors (
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2020-September/018168.html)
proposal without consensus changes. As long as someone gets a fully signed
parent, they can execute a bump with minimal wallet tooling. If a
transaction author doesn’t want a “sponsor”, do not include the output.

d) Lightning Carve-out(
https://lists.linuxfoundation.org/pipermail/lightning-dev/2019-October/002240.html)
is superseded by this logic, as we are not restricted to two immediately
spendable output scenarios. In its place, robust multi-party fee bumping is
possible.

e) This also benefits more traditional wallet scenarios, as change outputs
can no longer be pinned, and RBF/CPFP becomes robust. Payees in simple
spends cannot pin you. Batched payouts become a lot less painful. This was
one of the motivating use cases that created the term “pinning” in the
first place(
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2018-February/015717.html),
even if LN/L2 discussion has largely overtaken it due to HTLC theft risks.

Open Question(s):


   1.

   If we allow non-zero value in ephemeral outputs, does this open up a MEV
   we are worried about? Wallets should toss all the value directly to fees,
   and add their own additional fees on top, otherwise miners have incentive
   to make the smallest utxo burn transaction to claim those funds. They just
   confirmed your parent transaction anyways, so do we care?
   2.

   SIGHASH_GROUP like constructs would allow uncommitted ephemeral anchors
   to be added at spend time, depending on spending requirements.
   SIGHASH_SINGLE already allows this.




Hopefully this gives people something to consider as we move forward in
thinking about mempool design within the constraints we have today.


Greg

0: With V3 transactions where you have "veto power" over all the inputs in
that transaction. Therefore something like ANYONECANPAY is still broken. We
need a more complex solution, which I’m punting for the sake of progress.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20221018/76bfb957/attachment.html>