[bitcoin-dev] SegWit GBT updates

Cory Fields lists at coryfields.com
Tue Feb 2 01:40:49 UTC 2016


On Mon, Feb 1, 2016 at 6:08 PM, Luke Dashjr <luke at dashjr.org> wrote:
> On Monday, February 01, 2016 9:43:33 PM Cory Fields wrote:
>> On Mon, Feb 1, 2016 at 2:46 PM, Luke Dashjr <luke at dashjr.org> wrote:
>> > Allowing for simpler cases both encourages the lazy case, and enables
>> > pools to require miners use it. It also complicates the server-side
>> > implementation somewhat, and could in some cases make it more vulnerable
>> > to DoS attacks. Keep in mind that GBT is not merely a bitcoind protocol,
>> > but is used between pool<->miner as well... For now, it makes sense to
>> > leave
>> > "default_witness_commitment" as a bitcoind-specific extension to
>> > encourage adoption, but it seems better to leave it out of the standard
>> > protocol. Let me know if this makes sense or if I'm overlooking
>> > something.
>>
>> I think that's a bit of a loaded answer. What's to keep a pool from
>> building its own commitment and requiring miners to use that? I don't
>> see how providing the known-working commitment for the
>> passed-in-hashes allows the pool/miner to do anything they couldn't
>> already, with the exception of skipping some complexity. Please don't
>> confuse encouraging with enabling.
>
> Making it simpler to do a centralised implementation than a decentralised one,
> is both enabling and encouraging. GBT has always been designed to make it
> difficult to do in a centralised manner.
>

But your suggestion is "use libblkmaker" which will build the trees
for me. By that logic, isn't libblkmaker making a centralized
implementation easier? Shouldn't that usage be discouraged as well?
And along those lines, shouldn't the fact that it's used as a pool <->
miner protocol be discouraged rather than touted as a feature?

I don't wish to sound hostile, I'm just trying follow the logic. I
can't rationalize why GBT shouldn't expose the commitment that it
knows to be correct (when paired with the transactions it provides),
purely to make things difficult.

>> What's the DoS vector here?
>
> It's more work for the pool to provide it, similar to the "midstate" field was
> with getwork. Someone performing a DoS needs to do less work to force the pool
> to do complex calculations (unless the same transaction tree / commitment is
> used for all miners, which would be an unfortunate limitation).

It's being provided to them. And if they're using a modified set of
tx's, they'll need to re-calculate it in order to verify the result
anyway. I suspect I'm not understanding this argument.

>
>> >> The issue in particular here is that a non-trivial burden is thrust
>> >> upon mining software, increasing the odds of bugs in the process.
>> >
>> > It can always use libblkmaker to handle the "heavy lifting"... In any
>> > case, the calculation for the commitment isn't significantly more than
>> > what it must already do for the stripped merkle tree.
>>
>> Agreed. However for the sake of initial adoption, it's much easier to
>> have a known-correct value to use. Even if it's just for the sake of
>> checking against.
>
> Sure, I'm not suggesting we remove this from bitcoind (probably the only place
> that makes initial adoption easier).
>

How about exposing it as a feature/capability, then? That way pools
can expect it from bitcoind, but won't be required to expose it
downstream.

>> >> [4]:
>> >> https://github.com/theuni/ckpool/commit/7d84b1d76b39591cc1c1ef495ebec513
>> >> cb 19a08e
>> >
>> > I'm pretty sure this commit is actually /introducing/ a bug in working
>> > (albeit ugly) code. The height, while always positive, is serialised as
>> > a signed number, so 0x80 needs to be two bytes: 80 00.
>>
>> You're right, thanks. The current code breaks on heights of (for ex)
>> 16513. I'll fix up my changes to take the sign bit into account.
>
> I'm curious what bug it was fixing? Was it overwriting data beyond the number?

Using 16513 as an example:

serialized by bitcoind: 0x028140
serialized by ckpool: 0x03814000

ckpool works because blocks after 32768 end up looking the same, but
it will break again at 2113664.

>
> Luke


More information about the bitcoin-dev mailing list