[Bitcoin-development] unlinakble static address? & spv-privacy (Re: Stealth Addresses)

[Jeremy Spilman]

On Wed, 15 Jan 2014 15:09:01 -0800, Adam Back <adam at cypherspace.org> wrote:
> I was meaning to comment on the SPV privacy properties.
>
> For full-node use these unlinkable static addresses have quite nice
> properties.  It also nicely solves the problem of having to educate users
> and wallet authors to not reuse addresses.  But for SPV nodes they have  
> no direct-way to find the payments.  So then in Peter Todd's variant  
> (maybe it was suggested by Greg Maxwell?) there is a second address so  
> that the SPV
> client can delegate detection to a full node without giving it the  
> private key allowing it to spend!  (This is something analogous to bloom  
> filtering).

The second pubKey is useful for delegating scanning, or even just being  
able to scan for transactions yourself without keeping bitcoin-encumbered  
private keys decrypted in memory. So even while running your own full node  
there are good reasons to use a second pubKey to derive the shared secret.

> But I think its moderately expensive for the full node because it has to  
> do a DH calculation per transaction and its not precomputable so there  
> is IO
> per query.  (In the P version in fact only payments which are thereby
> reconizable as unlinkable static need to be processed).

And of course, if you have multiple reuseable addresses, then you're doing  
this calculation separately to check each one.

So the load on a popular centralized service would be quite high, which  
you may consider a feature.

>
> Then an artificial prefix is proposed to constrain the query to a subset,
> however that leaks to everyone so in some ways its a worse privacy leak
> than bloom filtering.  It can be used to rule out recipients and could be
> quite a powerful extra lever for statistical analysis.

Choosing how many bits to put in the prefix may be difficult, particularly  
if transaction load changes dramatically over time. 0 or 1 bits may be  
just fine for a single user running their own node, whereas a central  
service might want 4 or 5 bits to keep their computation costs scalable.

But I think it's great people can choose how to trade privacy for  
computation/bandwidth however they want, and services can compete to offer  
monitoring for 0+ bit prefixes.

> (And also there is proposed a version of the prefix computed via
> brute-force to make it somewhat stealthy still).

I think in this case the hash grinding of the prefix would only being used  
if thats how transactions are being indexed. I don't think it adds any  
privacy, it's just added work we're forced to do in order for the prefix  
to work as designed. Peter, please correct me if I'm wrong.


>
> Maybe in the payment address case the service should choose the  
> derivation factor and communicate it and the client with the static
> address, as suggested by Alan Reiner because then it can also serve
> the function of allowing the service to tie the payment to the users
> account.

I think any change which requires more than a single published public  
message (e.g. a posting in a forum, or in a README.me in Github) should be  
seen as solving an entirely different problem.

If you have directed communication from payee->payer, I think there's  
simply no reason to do it this way. (By "this way" I mean ECDH with  
OP_RETURN P).

We could try to define a different reusable address type, for when you can  
make a single directed message from payer->payee, and in that case there's  
probably no need for ECDH or the prefix, like Alan's proposal.

But once you admit having that directed communication, then you are  
swimming very close to the payment protocol.