[Bitcoin-development] Eliminating double-spends with two-party self-escrow for high value transactions

Peter Todd pete at petertodd.org
Sat Apr 26 11:23:12 UTC 2014

In the majority of high-value transactions the fact that funds will be
sent is known prior to when they actually are. For instance, if Alice is
to meet Bob in person to buy a car or sell some Bitcoins, both parties
know the transaction will probably happen in the near future some time
before it actually does. Existing escrow solutions already take
advantage of this fact; for instance Localbitcoins provides sellers the
ability to escrow their funds with Localbitcoins prior to when the funds
are released to the buyer.

However with nLockTime a third-party escrow agent is *not* required.
Instead prior to Alice sending the funds to the escrow address, she has
Bob sign a refund transaction that unlocks at some time in the future.
Generally the transaction does go through, and Alice and Bob sign a
second transaction sending the funds to Bob. Sometimes it doesn't, and
Alice either gets Bob to sign a transaction sending the funds back to
her, or in the worst-case, just waits for the timeout to elapse.

Note that this technique can be used in addition to a third-party escrow
agent - the third-party then only plays a role in exceptional

Implementation sketch: Mycelium Local Trader

The Mycelium Local Trader(1) is functionality built into the Mycelium
Android wallet that helps users trade cash for bitcoins by finding
traders in their local area. To attempt to prevent double-spends it uses
"transaction confidence", a technique that attempts to determine how
many nodes on the network a given transaction has propagated too. Of
course this technique is fragile and vulnerable to many attacks.

Local Trader already has pre-meetup buyer to seller communication built
in. Within the application the buyers and sellers communicate and
negotiate the amount and price of the Bitcoins prior to arranging a
meeting place. We can extend this to self-escrow as follows:

1) Alice publishes her offer to sell Bitcoins for cash.

2) Bob replies to the offer with an unused pubkey, B.

3) Alice creates tx1 paying a CHECKMULTISIG scriptPubKey spendable by
   the co-operation of her pubkey, A, and Bob's pubkey, B. She signs tx1
   but does not publish it.

4) Alice creates tx_refund which is nLockTime'd until some point in the
   future and returns the output of tx1 to an address she controls. Note
   how tx_refund depends on the signed tx1.

5) Alice sends Bob the hash of tx_refund for him to sign. As Bob does
   not have the actual transaction Bob can't selectivly target tx1 for a
   transaction mutability attack. Bob is free to sign the hash as the
   pubkey has never been used before. Note that the tx_refund hash
   Bob signs should be calculated with SIGHASH_SINGLE|SIGHASH_ANYONECANPAY
   to allow Alice to, for instance, add fees.

6) Bob replies with the signature.

7) Alice checks the validity of the signature, and if satisfied,
   publishes tx1 to the network.

8) Alice and/or Bob travel to actually meet in person. If this takes
   time t the probability of a block being generated during that
   time is P=1-e^(1/10mins*t) For instance, with a travel time of 30
   minutes we get 95% success, 1 hour 99.75%, and 2 hours 99.999%

9a) If the cash is handed over successfully Alice signs a
    SIGHASH_SINGLE|SIGHASH_ANYONECANPAY signature for the tx1 output
    spending the funds to a scriptPubKey specified by Bob and gives that
    signature to him.

10a) Bob creates a transaction spending the output, adds Alice's
     signature to it, and finally signs it himself. He broadcasts this
     transaction to the network, completing the transfer.

9b) If the cash is NOT handed over successfully Alice and Bob can either
    co-operate to cancel the transaction immediately, sending the
    escrowed funds back to Alice, or Alice can wait until the timeout to
    use the signature she had Bob prepare in advance.

While the above is relatively complex, from the user's point of view the
process is quite similar to how Mycelium already works:

Alice: Publish offer -> Accept offer  -> Travel -> Release funds
Bob:   Browse ads    -> Reserve funds -> Travel -> Accept

The chief difference being that the funds for the transaction have been
reserved, and if the transaction does not go through, will not be
unlocked without the co-operation of the other party, or the expiration
of the timeout.

Transaction Malleability

While the above is fairly secure if transactions aren't being mutated
en-mass, better protections would be desirable. First of all adding a
third-party escrow to the two-party escrow is a prudent last ditch
measure to ensure that if malleability is an issue the third-party can
release locked funds manually; note how SIGHASH_SINGLE is used as
opposed to SIGHASH_NONE to prevent that third-party from having access
to the funds. Secondly a future soft-fork such as Pieter Wuille's
BIP62(2) can eliminate malleability. In particular, a soft-fork that
enabled the creation of signatures that did *not* include the txin txid
would be particularly valuable; in step 4 above Bob's refund signature
signed over the scriptPubKey and nLockTime only would cover all possible
cases whre a refund would be needed, such as accidental multiple
payments and previously unknown sources of malleability.

1) http://www.reddit.com/r/Bitcoin/comments/236k5d/mycelium_local_trader_is_now_available/

2) https://github.com/bitcoin/bips/blob/master/bip-0062.mediawiki

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 685 bytes
Desc: Digital signature
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20140426/97b21175/attachment.sig>

More information about the bitcoin-dev mailing list