[Lightning-dev] Deriving channel keys deterministically from seed, musig, and channel establishment v2
somber.night at protonmail.com
Fri Sep 17 16:14:22 UTC 2021
TL;DR: an approach is described how to derive channel keys
deterministically that allows certain forms of recovery from just
a seed, that works today. This approach however will no longer work
with e.g. MuSig key aggregation in the future. An idea for a proposal
is given how the channel-open flow (e.g. as part of channel v2) could be
changed to make a similar approach work independent of key aggregation.
While implementing anchor output support in Electrum, we have realised
one difficulty is to do with the remote-force-close case where the
to_remote output is no longer a simple p2wpkh.
Currently, pre-anchor-support, Electrum sets option_static_remotekey
to required (feature bit 12), and we restrict lightning usage to wallets
that derive p2wpkh addresses, and payment_basepoint is set
to a bip32-derived pubkey that corresponds to a wallet address.
Hence, if a user were to restore from seed words, and their channel
counterparty force closed their channel, the to_remote output of the
broadcast commitment tx would directly pay to the user's wallet.
That is, in many typical scenarios, funds are not lost when restoring
(Also, if we are the channel-funder/opener, it is easy to find the
funding transaction, just by testing txs in the wallet history.
Further, for the cases we are the channel-funder/opener,
there is a setting to put an OP_RETURN in the funding tx, which stores
the nodeid of the counterparty, allowing us to identify who to contact
to get the channel closed.
Also, we are (ab)using dataloss_protect to ask the remote
to force-close when restoring from seed, so the user does not even have
to wait for an arbitrarily long time.)
With anchors, the to_remote is now a p2wsh that involves a CSV,
and we cannot easily make this script correspond to a wallet address,
i.e. we lose the property that the remote-force-close pays directly
to a wallet address.
So, the problem we would like to solve, is:
- having seed words
- having access to blockchain data
- somehow having identified our channel counterparties (node IDs),
and our channels with them (funding outpoints)
- and assuming we can get the remote to do a force-close
--> we would like to be able to spend the to_remote output
1) Naively, we could just derive a static key to be used as
payment_basepoint, reused between all our channels, and watch the
single resulting p2wsh script on-chain.
Clearly this has terrible privacy implications.
2) Or, we could derive a new bip32 chain/sequence of pubkeys
used as payment_basepoint for channels, and watch these p2wsh scripts,
with a gap limit.
Particularly the gap limit part makes this undesirable though
(just consider having more than "gap limit" channels open and restoring
Instead, it seems desirable to see whether we can extract some entropy
from the blockchain, and use that as a nonce to be combined with a
static private secret derived from our seed.
We could extract data either from the funding tx, or from the
remote-commitment-transaction that spent the funding output.
3) We exploit the fact that the funding output uses a
2of2 OP_CHECKMULTISIG script composed of the funding pubkeys of
each party. The funding pubkey itself can be used as a nonce, and
it can be recovered from the witness of the commitment tx.
The privkey for payment_basepoint can then be derived as e.g.
hash(bip32_derive(seed, HARDCODED_PATH) + funding_pubkey).
In fact (3) is not novel at all: eclair has been deriving
all their channel keys like this  for some time, from
a static seed-based secret combined with the funding_pubkey as nonce,
and generating the funding_privkey from ~os.urandom.
Electrum will likely use (3) at least for the payment_basepoint,
as part of adapting to anchors.
Note that the idea (3) relies on recovering the funding_pubkey from
the witness of the spending transaction, which will break in the future
if the funding script is changed to e.g. a p2tr that uses musig.
Crucially, note that all the approach needs is some blockchain-visible
nonce that is already known at the time we need to construct the
open_channel message (as we want to be able to derive some of the keys
that are sent as part of the open_channel message
(e.g. payment_basepoint) from it).
As long as the funding output uses a 2of2 OP_CHECKMULTISIG,
the local funding_pubkey fits the bill.
Note that irrespective of any restrictions on the script used in
the funding output, we could use the funding scriptPubKey/address as
the nonce, if the open_channel/accept_channel messages were
split into two.
For example, instead of the single round of open_channel/accept_channel,
there could be two rounds:
- an open_channel_part1, where the peers exchange only
the funding_pubkey (and the other non-pubkey fields), and
- an open_channel_part2, where the rest of the pubkeys are sent
This way the peers would learn what the funding address would be after
the first round finishes, and could potentially use that to derive
their other channel keys (needed for round 2).
So, to arrive at my point, I would like to ask whether people think
such a change - or something similar - might be useful, and if so,
whether it could/should be incorporated to the current
channel establishment v2 proposal .
If there is agreement that this would be useful, a spec change
would be most useful before changing the 2of2 multisig script.
More information about the Lightning-dev