[bitcoin-dev] Proposal: Full-RBF in Bitcoin Core 24.0

* [bitcoin-dev] Proposal: Full-RBF in Bitcoin Core 24.0
@ 2021-06-15 16:55 Antoine Riard
  2021-06-17  0:58 ` Billy Tetrud
  2021-12-18 16:51 ` Jeremy
  0 siblings, 2 replies; 12+ messages in thread
From: Antoine Riard @ 2021-06-15 16:55 UTC (permalink / raw)
  To: Bitcoin Protocol Discussion

[-- Attachment #1: Type: text/plain, Size: 7434 bytes --]

Hi,

I'm writing to propose deprecation of opt-in RBF in favor of full-RBF as
the Bitcoin Core's default replacement policy in version 24.0. As a
reminder, the next release is 22.0, aimed for August 1st, assuming
agreement is reached, this policy change would enter into deployment phase
a year from now.

Even if this replacement policy has been deemed as highly controversial a
few years ago, ongoing and anticipated changes in the Bitcoin ecosystem are
motivating this proposal.

# RBF opt-out as a DoS Vector against Multi-Party Funded Transactions

As explained in "On Mempool Funny Games against Multi-Party Funded
Transactions'', 2nd issue [0], an attacker can easily DoS a multi-party
funded transactions by propagating an RBF opt-out double-spend of its
contributed input before the honest transaction is broadcasted by the
protocol orchester. DoSes are qualified in the sense of either an attacker
wasting timevalue of victim's inputs or forcing exhaustion of the
fee-bumping  reserve.

This affects a series of Bitcoin protocols such as Coinjoin, onchain DLCs
and dual-funded LN channels. As those protocols are still in the early
phase of deployment, it doesn't seem to have been executed in the wild for
now.  That said, considering that dual-funded are more efficient from a
liquidity standpoint, we can expect them to be widely relied on, once
Lightning enters in a more mature phase. At that point, it should become
economically rational for liquidity service providers to launch those DoS
attacks against their competitors to hijack user traffic.

Beyond that, presence of those DoSes will complicate the design and
deployment of multi-party Bitcoin protocols such as payment
pools/multi-party channels. Note, Lightning Pool isn't affected as there is
a preliminary stage where batch participants are locked-in their funds
within an account witnessScript shared with the orchestrer.

Of course, even assuming full-rbf, propagation of the multi-party funded
transactions can still be interfered with by an attacker, simply
broadcasting a double-spend with a feerate equivalent to the honest
transaction. However, it tightens the attack scenario to a scorched earth
approach, where the attacker has to commit equivalent fee-bumping reserve
to maintain the pinning and might lose the "competing" fees to miners.

# RBF opt-out as a Mempools Partitions Vector

A longer-term issue is the risk of mempools malicious partitions, where an
attacker exploits network topology or divergence in mempools policies to
partition network mempools in different subsets. From then a wide range of
attacks can be envisioned such as package pinning [1], artificial
congestion to provoke LN channels closure or manipulation of
fee-estimator's feerate (the Core's one wouldn't be affected as it relies
on block confirmation, though other fee estimators designs deployed across
the ecosystem are likely going to be affected).

Traditionally, mempools partitions have been gauged as a spontaneous
outcome of a distributed systems like Bitcoin p2p network and I'm not aware
it has been studied in-depth for adversarial purposes. Though, deployment
of second-layer
protocols, heavily relying on sanity of a local mempool for fee-estimation
and robust propagation of their time-sensitive transactions might lead to
reconsider this position. Acknowledging this, RBF opt-out is a low-cost
partitioning tool, of which the existence nullifies most of potential
progresses to mitigate malicious partitioning.

To resume, opt-in RBF doesn't suit well deployment of robust second-layers
protocol, even if those issues are still early and deserve more research.
At the same time, I believe a meaningful subset of the ecosystem  are still
relying
on 0-confs transactions, even if their security is relying on far weaker
assumptions (opt-in RBF rule is a policy rule, not a consensus one) [2] A
rapid change of Core's mempool rules would be harming their quality of
services and should be
weighed carefully. On the other hand, it would be great to nudge them
towards more secure handling of their 0-confs flows [3]

Let's examine what could be deployed ecosystem-wise as enhancements to the
0-confs security model.

# Proactive security models : Double-spend Monitoring/Receiver-side
Fee-Topping with Package Relay

From an attacker viewpoint, opt-in RBF isn't a big blocker to successful
double-spends. Any motivated attacker can modify Core to mass-connect to a
wide portion of the network, announce txA to this subset, announce txA' to
the
merchant. TxA' propagation will be encumbered by the privacy-preserving
inventory timers (`OUTBOUND_INVENTORY_BROADCAST_INTERVAL`), of which an
attacker has no care to respect.

To detect a successful double-spend attempt, a Bitcoin service should run
few full-nodes with well-spread connection graphs and unlinkable between
them, to avoid being identified then maliciously partitioned from the rest
of the network.

I believe this tactic is already deployed by few Bitcoin services, and even
one can throw flame at it because it over consumes network resources
(bandwidth, connection slots, ...), it does procure a security advantage to
the ones doing it.

One further improvement on top of this protection could be to react after
the double-spend detection by attaching a CPFP to the merchant transaction,
with a higher package feerate than the double-spend. Expected deployment of
package-relay as a p2p mechanism/mempool policy in Bitcoin Core should
enable it to do so.

# Reactive security models : EconomicReputation-based Compensations

Another approach could be to react after the fact if a double-spend has
been qualified. If the sender is already known to the service provider, the
service account can be slashed.  If the sender is a low-trusted
counterparty to the merchant, "side-trust" models could be relied on. For
e.g a LN pubkey with a stacked reputation from your autopilot, LSATs, stake
certificates, a HTLC-as-a-fidelity-bond, ... The space is quite wide there
but I foresee those trust-minimized, decentralized solutions being adopted
by the LN ecosystem to patch the risks when you enter in a channel/HTLC
operation with an anonymous counterparty.

What other cool new tools could be considered to enhance 0-confs security ?

To conclude, let's avoid replaying the contentious threads of a few years
ago. What this new thread highlights is the fact that a transaction
relay/mempool acceptance policy might be beneficial to some class of
already-deployed
Bitcoin applications while being detrimental to newer ones. How do we
preserve the current interests of 0-confs users while enabling upcoming
interests of fancy L2s to flourish is a good conversation to have. I think.

If there is ecosystem agreement on switching to full-RBF, but 0.24 sounds
too early, let's defer it to 0.25 or 0.26. I don't think Core has a
consistent deprecation process w.r.t to policy rules heavily relied-on by
Bitcoin users, if we do so let sets a precedent satisfying as many folks as
we can.

Cheers,
Antoine

[0]
https://lists.linuxfoundation.org/pipermail/lightning-dev/2021-May/003033.html

[1] See scenario 3 :
https://lists.linuxfoundation.org/pipermail/lightning-dev/2020-June/002758.html

[2] https://github.com/bitcoin/bitcoin/pull/10823#issuecomment-466485121

[3] And the LN ecosystem does have an interest to fix zero-confs security,
if "turbo-channels"-like become normalized for mobile nodes

[-- Attachment #2: Type: text/html, Size: 8068 bytes --]

^ permalink raw reply	[flat|nested] 12+ messages in thread