Thanks for the reply.

The way we would do it is:

Let's say we have 3 cold keys for multisig: A B and C

Whose xpubs are: xA xB and xC

We all sign each other's xpubs, whose signatures are:
sAxB
sAxC

sBxA
sBxC

sCxA
sCxB

We can then create a wallet that says "when verifying change with 0x01 global type proposed by Andrew Chow, if the change is multisig, we MUST require the other pubkeys to have signatures via my 0x02 proposal"

This way, all my PSBTs for my cold will have:
1. an 0x01 entry to tell me how to get my change.
2. All 6 of the signatures above.

And the signer will then look at the change, check my pubkey by deriving the xpub and checking equality to the BIP_DERIVATION of the output... it will then check the OTHER pubkeys via BIP32_DERIVATION to master fingerprint, then link that fingerprint to a 0x02 sig from MY key, verifying all pubkeys.

So this proposal of mine would not only fix the "send to address verification" problem for HD, but also the multisig change problem with 0x01.

Cool.

Only thing that is kind of sad is having to include n! (of m-of-n) signatures in every PSBT... but tbh, the PSBT size is not of much concern.

Thanks for the reply.

- Jonathan

2019年6月27日(木) 13:49 Dmitry Petukhov <dp@simplexum.com>:

Hi!

I wonder how your scheme handles multisig ?

As I understand, you sign individual xpubs with cold keys, so that cold
keys can check destination addresses are trusted.

I seems to me that if you sign individual xpubs of a multisig warm
wallet, and one key from that multisig is compromized, attackers can
then create a single-sig destination address that they control, and
move the coins in a chain of two transactions, first to this single-sig
address, and then to an address that they independently control.

My idea to prevent this [1] is to sign the whole 'xpub package' of the
multisig wallet, but there is also an issue of 'partial compromize',
where some of the keys in a multisig warm wallet is compromized, and
you do not want to regard a particular 'xpub package' as trusted. My
idea was [2] to use an auxiliary message that would be signed along with
the 'xpub package', and that message can include specific 'epoch' word
that hardware wallet can show prominently before signing, or have
'serial number' for xpub packages (but that will require to store last
known serial inside hw wallet, making it stateful).

I like the idea to extend PSBT to accomodate these schemes, but given
that the huge number of possible schemes that each may probably
require its own PSBT field type, I think that this is better dealt with
outside of PSBT, as 'PSBT metainformation', or using some form of
'vendor-specific', or 'metainformation-specific' PSBT field. This way
each usecase can be independently described in its own documentation,
that would include the particulars of the format for the
metainformation. This would also make it easier to implement PSBT for
simple cases, because the 'core specification' would not grow that big.

[1]
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2019-May/016917.html

[2]
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2019-May/016926.html

В Thu, 27 Jun 2019 11:11:23 +0900 Jonathan Underwood via bitcoin-dev
<bitcoin-dev@lists.linuxfoundation.org> wrote:

> Hello all,
>
> Just wanted to pick your brains about an idea for PSBT extension.
>
> One problem we try to solve with cold -> warm and warm -> hot sends
> for our exchange wallet is "How do I know that the address I am
> sending to is not a hacker's address that was swapped in between
> unsigned tx creation and first signature?"
>
> We have a proprietary JSON based encoding system which we are looking
> to move towards PSBT, but PSBT is missing this key functionality.
>
> BIP32_DERIVATION does allow us to verify the address is from a certain
> XPUB, but, for example, it can not allow us to verify a signature of
> that xpub.
>
> I have made a rough draft of the proposed key value specification.
> https://github.com/junderw/bips/blob/addXpubSig/bip-0174.mediawiki#specification
>
> The signing key path used in the spec is just randomly chosen 31 x 4
> bits shown as numbers with hardened paths.
>
> Since this issue seems similar to the change address issue, I started
> from that as a base. With the HW wallet case, I can verify the xpub
> by just deriving it locally and comparing equality, however, in our
> case, we need to verify an xpub that we do not have access to via
> derivation from our cold key(s) (since we don't want to import our
> warm private key into our cold signer)
>
> So the flow would be:
> 1. Securely verify the xpub of the warm / hot wallet.
> 2. Using the airgap signing tool, sign the xpub with all cold keys.
> 3. Upload the signature/xpub pairs to the online unsigned transaction
> generator.
> 4. Include one keyval pair per coldkey/xpub pairing.
> 5. When offline signing, if the wallet detects there is a global
> keyval XPUB_SIGNATURE with its pubkey in the key, it must verify that
> all outputs have BIP32_DERIVATION and that it can verify the outputs
> through the derivation, to the xpub, and to the signature.
>
> In my attempt to fitting this into PSBT, I am slightly altering our
> current system, so don't take this as an indication 100% of how we
> work in the backend.
>
> However, I would like to hear any feedback on this proposal.
>
> Thanks,
> Jonathan
>