[bitcoin-dev] Revault: a multi-party vault architecture

[darosior <darosior@protonmail.com>]

Hi all,

Kevin Loaec and I have been working on a new multiparty vault architecture and I think it reached the point where we’d welcome some feedback.


Intended usage and limitations
==============================

The aim is to secure the shared storage of coins without relying on a trusted third party and by disincentivizing theft attempts, while not restricting the usage of the funds for day-to-day operations.

Revault uses N-of-N multisigs and thus does not protect against intentional locking of funds (such as refusal to sign, or key erasure). Therefore it assumes its users (likely companies with already on-going agreements between shareholders) to be able to solve intentional blockage outside the Bitcoin network (such as through legal contracts).


The actual architecture
=======================

We called it revault as it relies on pre-signed and revocable (revaultable) transactions.
The users pre-sign a transaction chain as the only used way to spend from a vault output.
They would have signed a set of transactions to either cancel a spend attempt or lock the funds for some time beforehand. The funds are always better locked for a long time than stolen.


The transactions
----------------

The system is composed of mainly 6 transaction types (with N the number of stakeholders) :

- The “vault” transaction which pays to a N-of-N, by which funds are received.
- The “emergency” transaction, which spends the vault output and pays to a [here goes a
high value]-days timelocked N-of-N (with N differents but statics keys, assumed to be physically stored in hard(/long) to access locations).
- The “unvault” transaction, which spends the vault output and pays to [either the vault’s N-of-N, or *after X blocks* to a subset of the stakeholders AND a co-signing server].
- The “unvault emergency” transaction, which spends the unvault output and pays to the
same script as the first emergency transaction.
- The “cancel” transaction, which spends the unvault output and pays back to a new vault utxo.
- The “spend” transaction, which spends the unvault output and pays to an external address (potentially contained in a list of destinations previously agreed-upon by all the stakeholders).


The process
-----------

The stakeholders would exchange the signatures of all the revaulting transactions after the reception of a new vault utxo, and then exchange the signatures of the unvaulting transaction. Before doing so, the coins are not available to be spent.

In order to spend a vault, the subset of the stakeholders who manages the funds (for example, the traders of an investment fund) would make the cosigning server (which only signs a transaction once) sign the spend transaction.
They would then present it to the other watchers which would ACK the spend (if paying to an authorized address), and broadcast the "unvault" transaction. Finally, and after X blocks have passed they would be able to broadcast the spend transaction.
If a stakeholder's watcher detects an unvaulting transaction without knowing about its child “spend” transaction, it triggers an automatic “cancel” transaction (not encumbered by the timelock).

At any point -even in the middle of a spend- any of the stakeholder can trigger an emergency transaction if anything nasty is happening.
Any network watcher noticing the broadcast of an emergency transaction would also broadcast all other vaults’ emergency transactions.

This network watching and revaulting power can be replicated (watchtowers) to further decrease the reliance on a single machine or internet access.


Pre-signed transactions fun
---------------------------

In order to avoid our security assumptions to be as weak as betting on the value of the feerate in the future, stakeholders exchange SINGLE | ANYONECANPAY signatures for the revaulting transactions and append their own as SIGHASH_ALL before broadcasting.
They can add another input (and potentially output) in order to bump the fees before doing so.

We protect ourselves from the bug by leveraging the fact the revaulting (namely the "emergency", "unvault emergency", and "cancel" transactions) only have *strictly* one input and one output. The change being part of the spend transaction.

In addition, revaulting transactions may signal for RBF to cover a feerate increase after the broadcast. Anyhow, a significant breathing room can be added to the feerate as these transactions are not intended to be used under normal circumstances.


Worth mentioning
================

The original draft of this architecture was first designed by Kevin Loaec who was hired by NOIA to do so. It was inspired by Bryan Bishop’s single-party vault architecture (https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2019-August/017229.html), who published a demo implementation of it last week (https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2020-April/017755.html, https://github.com/kanzure/python-vaults).
Kevin and I since detailed and reworked our new architecture together.

A WIP draft / demo / PoC / [enter adjective with “insecure” meaning] implementation is available at https://github.com/re-vault/revault-demo, which uses 4 stakeholders, 2 or 3 traders (doing the day-to-day moves) a CSV of 6 blocks for the unvault script and a CSV of ~1 month for the emergency scripts.
The transactions used are detailed in the doc/ directory of the same repo, and are coded in the revault/transactions/ module.

The “revault” name was coined by Lea Thiebaut (Lexyon).


Thanks for reading,
Antoine / Darosior