Received-SPF: pass (sog-mx-2.v43.ch3.sourceforge.com: domain of gmail.com
	designates 74.125.82.53 as permitted sender)
	client-ip=74.125.82.53; envelope-from=pieter.wuille@gmail.com;
	helo=mail-ww0-f53.google.com; 
Received-SPF: pass (google.com: domain of pieter.wuille@gmail.com designates
	10.180.83.42 as permitted sender) client-ip=10.180.83.42; 
MIME-Version: 1.0
In-Reply-To: <CAPg+sBhb+gYMwp1OJuCHYt5=BU63=YBWOFaLLthHBkN_U-scaA@mail.gmail.com>
References: <CAPg+sBhb+gYMwp1OJuCHYt5=BU63=YBWOFaLLthHBkN_U-scaA@mail.gmail.com>
Date: Fri, 2 Mar 2012 02:56:34 +0100
Message-ID: <CAPg+sBj_zOvNS91BeAzJ8bFWmyYxk1CUhA=Y2_J8vFt4-fruyg@mail.gmail.com>
From: Pieter Wuille <pieter.wuille@gmail.com>
To: Bitcoin Dev <bitcoin-development@lists.sourceforge.net>
Content-Type: text/plain; charset=ISO-8859-1
Subject: Re: [Bitcoin-development] Duplicate transactions vulnerability
Precedence: list

On Tue, Feb 28, 2012 at 17:48, Pieter Wuille <pieter.wuille@gmail.com> wrote:
> I've written about it in BIP30[2]. There is a patch for the reference
> client, which has been tested and verified to make the attack
> impossible. The change is backward compatible in the same way BIP16
> is: if a supermajority of mining power implements it, old clients can
> continue to function without risk.

After some private discussion, Ben Reeves pointed out two potential
small weaknesses in the proposed patch, which seem viable to me.

First: disconnecting the same coinbase transaction twice would fail,
as EraseTxIndex will not find anything the second time. This is
extremely hard to pull off, as it requires reverting a chain of at
least 120 blocks long. Still, the fix is very easy imho: allow
EraseTxIndex to fail.

Second: assume the following order of events: block with coinbase A is
created, 120 blocks later, A:0 is spent in transaction B. Then, a dupe
of A is created, and another 120 blocks are waited. At this point, A:0
and B:0 are still spendable. Now a block is created with two
transactions: first C which spends B:0, followed by a dupe of B. This
dupe is accepted, as its former instance is completely spent now.
However, if this last block is disconnected again, B:0 is not
spendable anymore, causing a risk for chain split. Ben suggested
moving the check for dupes up, turning the new network rule into:

  Blocks are not allowed to contain transactions whose hash matches
that of an earlier transaction in the same chain, unless that
transaction was already completely spent before said block.

I've updated the patch, and will update the BIP soon.

What do you all think? Can we still move forward with deploying this?

-- 
Pieter