Thanks Mike.  The BIPS process is ideally an implementation & draft BIP, like IETF RFCs.  Thanks for being a model citizen.  :)  Having an idea is good; having an implementation is better.

Reviewing the code at the pull request, it appears OK, and I did give it a quick test.  I have a few minor nits that I'll put in the PR, that are not worth mentioning here.

Being able to query UTXOs is obviously useful.  Many existing applications have been built on top of bitcoind (gettxout RPC), insight and other existing tools that make this query available.  This is not a new feature.  That effectively reduces our evaluation to

     1. Why implement in P2P protocol, versus RPC or external API?

Neither your pull request nor email addresses this much.  I do understand your app uses "getutxos"  But it is entirely fair and reasonable to ask why all bitcoind users should carry this feature.

Turning to the protocol itself, "getutxos" does what is expected:  Return that node's view of the UTXO set.  This bring us to the main issue I and others raised in the pull request,

     2. This view of UTXO is entirely untrusted, may be malicious or wrong.  Why export potentially dangerous information to victims?  What are the consequences to the victims of receiving targeted, maliciously wrong returned data?

Let us assume for the sake of progress that #2 is answered to satisfaction.  In my view, the BIP (and implementation? haven't looked at lighthouse) is missing

     3. An explicit solution to #2.

If one implements your BIP in a naive manner -- simply find a node, and issue a single query -- they are dangerously exposed to malicious information.  The BIP should describe this major security issue, and describe at least one method of solving it (ditto implementation, if lighthouse has not already solved this).

Comparison between this and BIP 35 (mempool command) are not apt, as miners and full nodes treat "mempool" returned data just like any other randomly solicited "tx" command on the network.  Unlike "mempool" cmd, this "getutxos" cmd proffers post-verification trusted data.

I fear that this addition will lead to people building insecure apps, when they could have just as easily queried a slightly-more-trusted-than-just-a-random-P2P-peer network of N bitcoind's or N Insight servers running somewhere (akin to Electrum servers).








On Thu, Jul 10, 2014 at 10:29 AM, Mike Hearn <mike@plan99.net> wrote:
I opened up a pull req for a draft BIP for getutxo.

   https://github.com/bitcoin/bips/pull/88

I include a rendering below for your reading convenience. If you'd like to comment on design/security/etc then please first familiarise yourself with the long discussions that were already had here:

   https://github.com/bitcoin/bitcoin/pull/4351



  BIP: 45
  Title: getutxo message
  Author: Mike Hearn <hearn@vinumeris.com>
  Status: Draft
  Type: Standards Track
  Created: 2014-06-10

Table of Contents

Abstract

This document describes a small P2P protocol extension that performs UTXO lookups given a set of outpoints.

Motivation

All full Bitcoin nodes maintain a database called the unspent transaction output set. This set is how double spending is checked for: to be valid a transaction must identify unspent outputs in this set using an identifier called an "outpoint", which is merely the hash of the output's containing transaction plus an index.

The ability to query this can sometimes be useful for a lightweight/SPV client which does not have the full UTXO set at hand. For example, it can be useful in applications implementing assurance contracts to do a quick check when a new pledge becomes visible to test whether that pledge was already revoked via a double spend. Although this message is not strictly necessary because e.g. such an app could be implemented by fully downloading and storing the block chain, it is useful for obtaining acceptable performance and resolving various UI cases.

Another example of when this data can be useful is for performing floating fee calculations in an SPV wallet. This use case requires some other changes to the Bitcoin protocol however, so we will not dwell on it here.

Specification

Two new messages are defined. The "getutxos" message has the following structure:

Field Size DescriptionData typeComments
1check mempoolbool Whether to apply mempool transactions during the calculation, thus exposing their UTXOs and removing outputs that they spend.
?outpointsvector The list of outpoints to be queried. Each outpoint is serialized in the same way it is in a tx message.

The response message "utxos" has the following structure:

Field Size DescriptionData typeComments
4chain heightuint32 The height of the chain at the moment the result was calculated.
32chain tip hashuint256 Block hash of the top of the chain at the moment the result was calculated.
?hit bitmapbyte[] An array of bytes encoding one bit for each outpoint queried. Each bit indicates whether the queried outpoint was found in the UTXO set or not.
?result utxosresult[] A list of result objects (defined below), one for each outpoint that is unspent (i.e. has a bit set in the bitmap).

The result object is defined as:

Field Size DescriptionData typeComments
4tx versionuint32 The version number of the transaction the UTXO was found in.
4heightuint256 The height of the block containing the defining transaction, or 0x7FFFFFFF if the tx is in the mempool.
?outputCTxOut The output itself, serialized in the same way as in a tx message.

Backward compatibility

Nodes indicate support by advertising a protocol version above 70003 and by setting a new NODE_GETUTXO flag in their nServices field, which has a value of 2 (1

Authentication

The UTXO set is not currently authenticated by anything. There are proposals to resolve this by introducing a new consensus rule that commits to a root hash of the UTXO set in blocks, however this feature is not presently available in the Bitcoin protocol. Once it is, the utxos message could be upgraded to include Merkle branches showing inclusion of the UTXOs in the committed sets.

If the requesting client is looking up outputs for a signed transaction that they have locally, the client can partly verify the returned output by running the input scripts with it. Currently this verifies only that the script is correct. A future version of the Bitcoin protocol is likely to also allow the value to be checked in this way. It does not show that the output is really unspent or was ever actually created in the block chain however.

If the requesting client has a mapping of chain heights to block hashes in the best chain e.g. obtained via getheaders, then they can obtain a proof that the output did at one point exist by requesting the block and searching for the output within it. When combined with Bloom filtering this can be reasonably efficient.

Note that even when the outputs are being checked against something this protocol has the same security model as Bloom filtering: a remote node can lie through omission by claiming the requested UTXO does not exist / was already spent (they are the same, from the perspective of a full node). Querying multiple nodes and combining their answers can be a partial solution to this, although as nothing authenticates the Bitcoin P2P network a man in the middle could still yield incorrect results.

Implementation

https://github.com/bitcoin/bitcoin/pull/4351/files


------------------------------------------------------------------------------
Open source business process management suite built on Java and Eclipse
Turn processes into business applications with Bonita BPM Community Edition
Quickly connect people, data, and systems into organized workflows
Winner of BOSSIE, CODIE, OW2 and Gartner awards
http://p.sf.net/sfu/Bonitasoft
_______________________________________________
Bitcoin-development mailing list
Bitcoin-development@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bitcoin-development




--
Jeff Garzik
Bitcoin core developer and open source evangelist
BitPay, Inc.      https://bitpay.com/