[bitcoin-dev] BIP sighash_noinput

[Johnson Lau]

In BIP143, the nSequence of the same input is always signed, with any hashtype. Why do you need to sign the sequence of other inputs?

> On 26 Sep 2018, at 5:36 PM, Jonas Nick via bitcoin-dev <bitcoin-dev at lists.linuxfoundation.org> wrote:
> 
>> At the risk of bikeshedding, shouldn't NOINPUT also zero out the
>> hashSequence so that its behaviour is consistent with ANYONECANPAY?
> 
> There is a good reason for not doing that. If NOINPUT would sign the
> hashSequence then it would be possible to get rid of OP_CSV in eltoo update
> scripts. As a result update scripts could be taprootified because the more
> common branch (settlement) would be just a 2-of-2 multisig. Applying taproot
> would then make unilateral settlement look like a single pubkey spend and avoid
> having to reveal the unexecuted (update) branch.
> 
> Eltoo update transaction outputs consist of two branches, update and
> settlement, where the update branch can be spend by a more recent update
> transaction if an obsolete update transaction ends up spending the funding
> output. The settlement branch is a 2-of-2 multisig with a relative timelock
> using OP_CSV. Removing OP_CSV is possible because both parties signature is
> required to spend the update transaction. They will only sign if the input has
> the right sequence numbers which is sufficient to enforce the timeout (BIP68) -
> assuming they are covered by the signature.
> 
> There's a catch: hashSequence includes the sequence numbers of all transaction
> inputs. That's not a problem for eltoo because settlement transactions only
> have one input. The update mechanism with update transactions relies on being
> able to bump the fee by unilaterally adding inputs and and change outputs to
> the transaction. That's also not a problem because update spends do not use
> relative timelocks and they are signed with SINGLE. So whenever NOINPUT is
> combined SINGLE the hashSequence should be zeroed. This is in fact what a
> minimal change to the current NOINPUT implementation would naturally do (see
> below). However, that's error-prone when using NOINPUT in other contexts so in
> general it would be better if NOINPUT would only sign the sequence number of
> the corresponding input.
> 
> Another downside of this approach is that you can never rebind to an output
> with an OP_CSV that requires a larger sequence number, unless you also sign
> with SIGHASH_SINGLE. It's difficult to imagine application where this would be
> an issue.
> 
> This is the modification to the NOINPUT implementation
> (https://github.com/cdecker/bitcoin/commits/noinput) which makes eltoo
> unilateral closes taprootifiable:
> +++ b/src/script/interpreter.cpp
> @@ -1223,7 +1223,7 @@ uint256 SignatureHash(const CScript& scriptCode, const CTransaction& txTo, unsig
>             hashPrevouts = cacheready ? cache->hashPrevouts : GetPrevoutHash(txTo);
>         }
> 
> -        if (!(nHashType & SIGHASH_ANYONECANPAY) && (nHashType & 0x1f) != SIGHASH_SINGLE && (nHashType & 0x1f) != SIGHASH_NONE && !noinput) {
> +        if (!(nHashType & SIGHASH_ANYONECANPAY) && (nHashType & 0x1f) != SIGHASH_SINGLE && (nHashType & 0x1f) != SIGHASH_NONE) {
>             hashSequence = cacheready ? cache->hashSequence : GetSequenceHash(txTo);
>         }
> 
> On 5/1/18 4:58 PM, Russell O'Connor via bitcoin-dev wrote:
>> At the risk of bikeshedding, shouldn't NOINPUT also zero out the
>> hashSequence so that its behaviour is consistent with ANYONECANPAY?
>> 
>