Bitcoin Optech Newsletter #95

This week’s newsletter summarizes the disclosure of an issue affecting the safety of routed LN payments and announces a new presigned vault proposal. Also included are our regular sections with popular questions and answers from the Bitcoin StackExchange, announcements of releases and release candidates, and descriptions of notable code changes in popular Bitcoin infrastructure projects.

Action items

  • Review the disclosure of a potential LN issue: as explained in the news section below, a public disclosure this week describes a new method for stealing money from LN nodes. The issue partly overlaps with an existing well-known fee management issue that has not been exploited (to our knowledge) because almost all onchain transactions relayed in the past two years confirmed relatively quickly even if they only paid the default minimum relay feerate. If feerates increase significantly for an extended period of time, these issues will become more critical. See our explanation in the news section for details and contact your LN software vendor if you have concerns about how the this issue might affect your channels.

News

  • New attack against LN payment atomicity: Matt Corallo started a thread across both the Lightning-Dev and Bitcoin-Dev mailing lists disclosing an attack discovered during a discussion about allowing LN commitment transactions to be CPFP fee bumped via anchor outputs. We’ll describe the attack using an extended example: Alice uses an LN channel to send Bob a Hash Time Lock Contract (HTLC) which is designed to be settled in either of the following ways:

    • If Bob discloses the preimage for <hash>, he can spend 1 BTC of Alice’s money

    • Otherwise, after 80 blocks, Alice can refund that 1 BTC back to herself

    Alice also told Bob that the goal of her payment is to pay Mallory, so Bob uses a channel he has with Mallory to send her a related HTLC:

    • If Mallory discloses the preimage for <hash>, she can spend 1 BTC of Bob’s money (we’re ignoring routing fees in this example)

    • Otherwise, after 40 blocks, Bob can refund that 1 BTC back to himself

    Although the above HTLCs are usually created and settled offchain, each party also has a commitment transaction they can use to put the HTLC commitment onchain. A separate onchain settlement transaction can fulfill either condition of the HTLC.

    For example, Mallory can publish the commitment transaction and then create a settlement transaction that provides the preimage and claims Bob’s 1 BTC. If Bob sees Mallory’s preimage settlement transaction before the 80 block timeout from the Alice-Bob contract, Bob can extract that preimage and use it to claim the 1 BTC from Alice (either onchain or offchain). Or, if Bob doesn’t see a preimage settlement transaction, Bob can create his own refund settlement transaction after 40 blocks that takes back his 1 BTC, allowing him to also initiate the refund of Alice’s 1 BTC (again, either onchain or offchain). In either case, this leaves everyone in compliance with the intent of their contracts.

    Unfortunately, as disclosed this week, there appears to be a way for Mallory to circumvent the process by both preventing Bob from learning the preimage while also preventing him from sending his refund settlement transaction.

    • Preimage denial: Mallory can prevent Bob from learning the preimage by giving her preimage settlement transaction a low feerate that keeps it from being confirmed quickly. If Bob is only looking for preimages in the block chain, he won’t see Mallory’s transaction while it remains unconfirmed.

    • Refund denial: Mallory’s prior broadcast of the preimage settlement transaction can prevent miners and Bitcoin relay nodes from accepting Bob’s later broadcast of the refund settlement transaction because the two transactions conflict, meaning they both spend the same input (a UTXO created in the commitment transaction). In theory, Bob’s refund settlement transaction will pay a higher feerate and so can replace Mallory’s preimage settlement but, in practice, Mallory can use various transaction pinning techniques to prevent that replacement from happening.

    Because Bob is prevented from either learning about the preimage settlement transaction or getting his refund settlement transaction confirmed, Alice is able to reclaim the 1 BTC she offered Bob in the Alice-Bob HTLC once its 80 block timeout expires. When Mallory’s preimage settlement transaction does eventually confirm, Mallory gets the 1 BTC that Bob offered her in the Bob-Mallory HTLC. This leaves Bob 1 BTC poorer than when he started.

    Several solutions were considered in the thread, but all had problems or involved significant tradeoffs:

    • Require a mempool: Bob could use a Bitcoin full node to monitor the Bitcoin P2P relay network and learn about Mallory’s settlement transaction. Some LN nodes such as Eclair already do this and it seems like a reasonable amount of extra burden since the problem only directly affects routing nodes (like Bob). Nodes that just send or receive payments on behalf of themselves are only indirectly affected,1 so everyday users could still run lightweight LN clients on mobile devices. Unfortunately, not all full nodes receive the same transactions as other nodes even when everything is working perfectly. Worse, there are techniques attackers like Mallory can use to send different conflicting transactions to different peers (for example, sending the pinned preimage settlement transaction to known miners but sending a different non-settlement transaction with at least one of the same inputs to non-miner relay nodes).

    • Beg or pay for preimages: The relay network could provide information about conflicts to transaction submitters such as Bob so they wouldn’t need to continuously monitor relay themselves. This still suffers from the problem of bad actors such as Mallory using targeted relay to send different transactions to miners and non-miners. Additionally Bob might be able to pay miners or other third party nodes for the preimage he needs, although this requires some people run additional software and might not be as easy to do after the deployment of some proposed upgrades to the LN protocol.

    • Settlement transaction anchor outputs: Onchain settlement transactions could be redesigned to spend their value to anchor outputs that could be CPFP fee bumped using CPFP carve-out. This would require those transactions to be larger (increasing onchain fees) and presigned (reducing flexibility). This would only directly affect channels which are unilaterally closed while payments are pending, which is already a situation that can significantly increase onchain costs and so is something users try to avoid. However, raising the cost of onchain enforcement also raises the minimum practical value of payments that can be sent trustlessly through LN. Despite these challenges, as of this writing, this appears to be the most preferred solution.

    Corallo labeled this a severe issue but noted its similar consequences to another known issue related to fee management in onchain LN transactions. The existing issue (described in Newsletter #78) is that commitment transactions have their feerate set at the time the transaction is signed, which might be days or weeks before they’re broadcast to Bitcoin relay nodes. If the minimum feerate necessary to get a transaction included in the next few blocks has increased significantly since the transaction was last signed, then the commitment transaction might not confirm until after Alice is able to reclaim her funds from Bob, again creating the opportunity for Bob to end up both paying Mallory and giving a refund to Alice. (This existing issue is what developers were working on fixing when the new issue was discovered.2)

    So far we’re unaware of any real-world losses due to onchain fee management problems in LN, possibly in part because the past two years has seen few large fee spikes that lasted long enough to significantly delay the confirmation of transactions with previously acceptable feerates. That good luck is unlikely to continue indefinitely, so this new problem gives LN developers an additional reason to prioritize the implementation of improved onchain fee management. In the interim, node operators concerned about the attack may wish to increase their cltv_expiry_delta to give preimage settlement transactions more time to confirm. Current defaults in popular LN nodes are 14 for C-Lightning, 40 for LND, 72 for Rust-Lightning, and 144 for Eclair. Note that increasing the value will make your channels less desirable to spenders, as higher values increases the normal worst case amount of time a payment could be stuck waiting to be settled.

  • Multiparty vault architecture: Antoine “Darosior” Poinsot announced a demo implementation of a vaults covenant prototype based on the same basic presigned transaction concept mentioned in last week’s newsletter. This new implementation, named Revault, focuses on storing funds shared between multiple parties with multisig security. The protocol allows a subset of the parties to initiate a withdrawal process by getting a beacon transaction confirmed; if the other parties to the vault object to the withdrawal, they have the opportunity to broadcast a second transaction that returns the funds to an emergency address in the vault. If there’s no objection within a certain amount of time, another transaction can complete the withdrawal of the funds. Poinsot is seeking feedback on the proposal.

Selected Q&A from Bitcoin StackExchange

Bitcoin StackExchange is one of the first places Optech contributors look for answers to their questions—or when we have a few spare moments to help curious or confused users. In this monthly feature, we highlight some of the top-voted questions and answers posted since our last update.

Releases and release candidates

New releases and release candidates for popular Bitcoin infrastructure projects. Please consider upgrading to new releases or helping to test release candidates.

Notable code and documentation changes

Notable changes this week in Bitcoin Core, C-Lightning, Eclair, LND, libsecp256k1, Rust-Lightning, Bitcoin Improvement Proposals (BIPs), and Lightning BOLTs.

Note: the commits to Bitcoin Core mentioned below apply to its master development branch and so those changes will likely not be released until version 0.21, about six months after the release of the upcoming version 0.20.

  • Bitcoin Core #15761 adds an upgradewallet RPC, doing away with the old method of upgrading upon startup, thereby allowing users to unlock and upgrade their wallets to Hierarchical Deterministic (HD) while it is loaded. This addition is also compatible with multi-wallet as it works on the individual wallet that is specified by the RPC.

  • Bitcoin Core #17509 allows the wallet GUI to save a Partially Signed Bitcoin Transactions (PSBT) to a file, as well as load a PSBT from a file. Saving is available in private-key-disabled wallets where Bitcoin Core previously automatically copied a PSBT to the clipboard (see the PRs described in Newsletters #74 and #82). Loading a PSBT will offer to finalize and broadcast the transaction if all signatures are available; otherwise, the PSBT will be copied to the clipboard for signing by a separate user action (for example using an RPC from the GUI console or using a separate tool such as HWI). A follow-up PRs is expected to add the ability to sign PSBTs in the GUI.

Special thanks

We thank Antoine Riard, ZmnSCPxj, and Matt Corallo for reviewing drafts of this newsletter and helping us understand the details of the LN atomicity issue. Any remaining errors are the fault of the newsletter author.

Footnotes

  1. Although the only known way to use this attack to steal money directly is by abusing a routing node (such as Bob in Alice→Bob→Mallory), when the same attack of delaying the preimage settlement and blocking the refund settlement is executed against a spender (such as Alice in Alice→Mallory), it may produce a “payment failed” error that causes the user to initiate a second payment without realizing the first payment hasn’t been revoked. This indirect attack can perhaps be dealt with by warning the user that the payment is stuck—not failed—and that sending additional payments could result in losses. 

  2. The ultimate solution to deal with arbitrary feerates in LN also depends on Bitcoin full nodes being able to perform package relay, a feature that’s long been discussed but never fully implemented. For now, LN commitment transactions can usually just pay slightly higher feerates than strictly necessary to avoid the need for package relay.