Bitcoin Forum

UPDATE: 0.5.3 was released (https://bitcointalk.org/index.php?topic=68895.0) with BIP30 support

Hello all,

as some of you already know, a vulnerability was found in the way bitcoin currently handles duplicate transactions. Even though we were able to demonstrate a potential attack on testnet, there is no cause for alarm because such an attack requires significant hashing power, is rather complex, and even if carried out does not allow financial gain for an attacker. Still, it is a vulnerability that can cause problems and we want to fix it as soon as possible.

But before we roll out the fix we wanted to inform you, the community, exactly what the problem is and how we are going to fix it. Essentially we want everyone to be on the same page to make the fix roll-out as smooth and uneventful as possible.

For most problems, this is not required, as we can just release an updated client. However, the nature of the problem dictates a fix that will introduce a new stricter block-validity rule. This means we have to actually change the rules by which blocks are considered to be valid or invalid. This in and of itself poses a risk, because an attacker violating this new rule while only a minority has upgraded their clients could cause a block chain fork. This is a bad thing because it would allow any transaction output that existed before the fork to be spent twice: once in each branch of the fork. However, if a majority enforces the stricter rules, we are sure that any fork will be resolved quickly and not be permanent. This is why we need your support.

The fix, described in BIP 30 (https://en.bitcoin.it/wiki/BIP_0030), is noncontroversial. There was a long discussion on the mailing list (http://sourceforge.net/mailarchive/forum.php?thread_name=CAPg%2BsBhmGHnMResVxPDZdfpmWTb9uqD0RrQD7oSXBQq7oHpm8g%40mail.gmail.com&forum_name=bitcoin-development), and several developers have expressed their support. A majority of mining pools (in terms of hashing power) have also confirmed they will be updating their nodes. BIP 30 is backward-compatible: only what was previously considered a valid block can become invalid and not the other way around. This means that old software will continue to accept blocks "mined" by clients following the new rule, and keep building upon them.

A bit more technical, the actual bug is this: the bitcoin software was written with the assumption that it is impossible to create a transaction with a hash that is identical to that of a previous transaction. This is however not entirely true. It requires buggy software, or malicious intent, but one can create a coinbase transaction that is identical to a previous coinbase, implying it has the same hash. Furthermore, by recreating transactions that use these duplicated coinbase(s), those can be duplicated as well. Now here is how the bitcoin software currently deals with this: it does not check whether that previous hash already exists but simply overwrites it in its transaction index database. Even worse, when a block that contained such a duplicate is reverted (during a reorganisation), the index entry is deleted entirely. If the original transaction was not yet spent, it has now become unspendable.

The way to fix this is simple: simply disallow blocks to contain such duplicated transactions. In order not to make pruning impossible in the future, one exception is added: if the original transaction was already completely spent before a block, it is (for now) still allowed to contain a duplicate. Without this exception, every full node in the network would be required to keep an index of all transactions ever. This was implemented, the change is merged in git master (so it will be included in the 0.6.0 release) as well as in several backports. It was tested by roconnor (who demonstrated the original attack on testnet) to prevent his attack, and it was tested to still accept the current longest block chain on the real network.


To conclude: We want to fix a newly found vulnerability, we need majority of miners to support the fix in order to prevent a potential permanent fork, the fix was thoroughly discussed on the mailing list, the developers support it and it is noncontroversial. With majority support, if all goes well, this change will be activated on march 15th 2012, 0:00 UTC.

PS: thanks to Hazek for proofreading

These pools have confirmed that they have deployed BIP 30 updates:
* deepbit (3400 GH/s)
* btcguild (1300 GH/s)
* mining.bitcoin.cz (1400 GH/s)
* eclipsemc (520 GH/s)
* eligius (290 GH/s)
* ozcoin (600 GH/s)
* bitclockers (175 GH/s)
* bitlc (170 GH/s)
* btcmp.com (120 GH/s)
* btcmine (620 GH/s)

(total over 8 TH/s)

So the strategy is to post updated code on Sourceforge before March 15 and have miners switch at 0:00 UTC?  I assume this version will not include any form of multisig, in order to avoid controversy and resistance to immediate adoption.  

Do we have an ETA on the patch getting merged and included in rc3 or whatever?

If anything will ever go wrong with bitcoin... it'll be now....  the whole idea of a possible fork is somewhat scary.

A technical question - why the patch does not simply discard the block with the duplicate hash, regardless of its contents?

Very unlikely.  Someone would need to plan the attack very carefully, have huge amounts of hashing power hiding offline, and gain essentially nothing from it.

+1

Not really, during this protocol change they only need to mine one valid block.

But as along as the majority of hashing power is supporting BIP30 the difficulty on the forked chain would be left so high that it would probably die fairly quickly and would not be able to produce the 6 valid blocks needed for transaction reversal.

I'm not familiar with how the blockchain is stored in nodes so I have one question.
Can this replacement of a transaction with another one with identical hash break a trace if someone is following the coins? if yes then whoever wants to do this will have to implement a different storing mechanism to allow multiple transactions with identical hash to coexists. Is this true or I'm too noob to understand?

One block forks happen all the time.   Meh.

All of this technical stuff is way over my head, but I'd like to share my ignorant thoughts anyways.

It seems to me that 8 days is not enough time for people to really do their research on the matter. If I were a miner, or the operator of a mining pool, I would want to know EXACTLY what I'm doing before I made any serious changes.

Secondly, even if 8 days is enough time there is something unsettling about this. What you essentially have is a few highly skilled coders in the know. If they can dupe the mining community into something, anything, nefarious it could destroy Bitcoin. Obviously, I'm not saying that they will, but posit this:

Imagine highly skilled feds, or any other criminal organization, infiltrated the upper echelon of Bitcoin coders. They spent long hours making it better and building a reputation for themselves, slowly but surely dominating the entire field of coders. Then, one day, it's time to destroy Bitcoin and they issue some sort of scenario similar to the one we're dealing with now, for the mining community to go to a new blockchain or something, and the mining community takes their word for it, and BAM, Bitcoin is hashed forever.

I understand this is conspiracy theory stuff, but can anyone unsettle my fears about this scenario?

Simple. As long as the process is completely transparent and open there's really nothing anyone could "sneak" past everyone else in order to cause damage. And if you checked the OP, it pretty much doesn't get any more transparent then how they're going about with this fix. I mean you can even read the mailing list how the devs formulated the fix..

My understanding is that there already exists such duplicate coinbase transactions in the production block chain.  (I first learned of this issue when someone came on the forums and pointed out how they "whoops" sent 50 BTC to never never land by mining a duplicate coinbase).  If I find that reference, I'll come back, but I'd like to think this was around block 90000-100000 or so.

EDIT: to include the specific block numbers: 91842 has a duplicate coinbase of 91812.
http://blockexplorer.com/b/91812
http://blockexplorer.com/b/91842

Can the proposal be amended to explain exactly what will happen to these transactions?  Or at the very least acknowledge such transactions exist (assuming I'm not mistaken)?  I assume they must remain valid (so as to not invalidate the entire block chain coming after them).  Anyone reimplementing the Bitcoin protocol will have to account for the fact that these transactions are valid and later ones just like them are not, and there will need to be very clear rules to determine what constitutes a valid transaction.

Two more questions then:

Is 8 days really enough time to vet a potentially nefarious plan?

What if a few people DO see something nefarious in it, will they be listened to, or ignored?

Anyone can read the code and verify that it really does what it claims. It's like 10 lines of code.


They won't be touched. The new rules only apply to blocks with a time after March 15.

It's not just 8 days. The fix was developed for some time now and the process was open so it already was under scrutiny. But even if it was just 8 days it's still enough.


No one is forced to download the updated client. That's the fialsafe in Bitcoin in general.

This is true, no one is forced. But people have a tendency to lose their way.

I just hope Bitcoin doesn't lose its way over time.

The bitcoins in block 91812 coinbase have not been spent.  My understanding is the current client will only allow them to be spent once.  If the block chain spec is being amended to say that duplicate coinbases are valid prior to March 15 2012, it should also explicitly address the bitcoins in the duplicate block 91842 and formally declare them as a valid block containing an invalid/unspendable transaction, otherwise there is an ambiguity that may be implemented by future clients/utilities the wrong way.

Also (all?) of the coders are known by real name. If they f*ckup bitcoin, they will be known by everyone as the person who screwed things up.
That would be the same as destroying many peoples money, and time invested and many companies and that is not like something you would like to do, ofcourse there is allways the possibility that they are threatened by jail or whatever unless they do this. But in this case its far from very likely.

I don't see very much as unlikely when you're talking about challenging the global banking cartel.

My guess is because of duplicate transactions already on the block chain.  Discarding the old blocks would discard half the block chain and half of the bitcoins in existence with it.

For all the paranoid people, here is the patch for everyone to read. The red lines came out and the green lines were added.

https://github.com/sipa/bitcoin/commit/a206b0ea12eb4606b93323268fc81a4f1f952531

That is basically what happens: a block with a duplicate transaction will become invalid, and thus will be ignored.

EDIT: obviously, this does not happen to blocks with existing duplicate transactions, as that would require invalidating the entire chain since then.

@casascius: you are right that technically a specification is needed for what to do with duplicate transactions in blocks before the switchover point. However, there is no formal specification of bitcoin's network rules currently either (only the code...), and the BIP only specifies the changes to the current rules. The current rules are: the transaction index is overwritten for a duplicate transaction, and a reorganisation removes the relevant transaction index entries.

I understand the "IF NEWER THAN SOMEDATE" in code, this is to prevent invalidating the entire chain since when first duplicate transaction appeared.

But, with this date switch implemented, why to check for spent/unspent transactions, instead of simply comparing the hash against all previous (just for blocks after the switch)?

It has been known for quite a while that a new coinbase with the same hash as an old coinbase was possible and not handled at all.  But since they are referenced by their (identical) hashes, there was no way to specify which one you meant to redeem, so redeeming one would redeem all of them.  This is not in any way a security problem, and can only be used to lose money.  There are faster and easier ways to lose money.

What was only recently discovered was that if you spend a coinbase, and then generate another identical one, you can then spend the second one too.  Spending a transaction doesn't mark that hash invalid for all time, it just removes it from the list of unspent transactions.  Another one can show up and be accepted.  This is, again, not a security problem, and doesn't, by itself, make it possible for people to lose coins unintentionally.

The list of all used transactions isn't readily available, and once pruning shows up, it might not even exist at all.  So, it only makes sense to compare the new coinbase to the list of transaction hashes that are unspent at the time, which aren't subject to pruning.  This also preserves the old behavior of being able to generate and spend multiple identical coinbases, so long as they are spent before the next one shows up.  And finally, it gives us a really short and simple patch, that anyone can read and understand without having to worry that a new side effect has snuck in.

Nice explained. Thanks!

This thread really needs to be more visible:  Perhaps a giant bold red link at the top of the forum?

BTC Guild will be pushing the patch tomorrow, and my understanding is both Slush & Deepbit will be shortly as well, which means the chain will have majority support.  That doesn't solve the problem of the other 30-40% of bitcoin miners that may not even know about the issue due to lack of visibility.

I have updated the second post with a list of pools from whom I have received confirmation they will update. I will keep that list up-to-date as I receive more confirmations.

Maybe the person who owns http://bitcoinwatch.com/ could post this thread in the "Latest news" section of their website?

IMO, you haven't made clear whether or not ALL users, or just miners, NEED to upgrade to 0.6 to facilitate/support this transition!

edit: by "support" you mean upgrade to 0.6?

Support was meant in a general sense: such a network rule change can only happen when the community agrees to it.

If you are a miner or pool operator: yes, please upgrade quickly, as the higher the percentage enforcing the new rules, the less chance for (temporary and short) block chain splits one could try to create.

Apart from that, if all goes well, eventually everyone will upgrade, but as a normal user, there is relatively little risk in keeping an old version. The more people running a version that enforces the new rule, the smaller the effects of an attacker can be. 0.6.0 will still take some time to be released, but there will be a bugfix-only 0.5.3 soon.

thats a little better.  let me repeat and if i'm wrong please correct:

Anyone who mines needs to upgrade to 0.5.3 as soon as it comes out to minimize the chances of a blockchain fork.

i'm not trolling; seriously it needs to be spelled out.

Is there an official Bitcoin.org Twitter account? This would be great place to post notifications, warning, new version release, and so on.

Glad to see you guys killing this annoying bug.

You might as well throw out the label  "conspiracy theory".  History, long past, and recent, has shown that discerning minds often reveal the truths behind events. So don't let lesser minds diminish accurate perception.
What inspired BTC? Bored geeks?  Current world events?

Recall the debian ssl "issue". It can happen here too. What did the spooks talk about with Andresen when he went to visit them?
Oh and have a look at github notification about recent security breach.  Isn't bitcoin code stored there?
The question is not if but when. The spooks always have a control point. They don't rest until they do and even then they hedge because that's the game.

Did you all choose sf.net for mailing list because no one likes it? I'm not clear on the technical details of this but from a higher-level view it seems that the devs should, if not already, think about how to push updates out.  i.e. decentralize the software

I'm not sure how to do what I'm getting at but having everyone update, in the way that's needed to implement this change, isn't going to scale into the future.
Has this been discussed already? Some kind of enforcement that would prevent unsupported software access to the network? Am I making any sense at all?

I realize pushing updates out automatically would be unpopular so another model would be needed. Not sure what model would work.

The people that need to update to make the change stick network-wide are already aware of this, and have committed to updating their nodes.  See the second post.

The way this has been handled is appropriate for the problem at hand and the network of today.  When either of those are different, different methods will be (have been) used to update things.

Getting more than half of the hashing power onboard is all that is needed for this fix, and that is already done.  The fix does open a tiny sliver of a window of vulnerability where people who are doing things that they know they should not be doing might be at a slight risk if an attacker expends great effort for no gain.

The people that want to eliminate that tiny risk and keep being irresponsible have the option to upgrade.

Also, from your other (crazier) post, github has a copy of the bitcoin source, but the developers collectively have the bitcoin source.

+1 LOL!

Anyone who can't satisfy themselves about this change in an hour or two isn't trying.

No one is going to jail for destroying someone else's bitcoins until a government considers bitcoin a currency.

I'm running a CentOS 5.6 build that I had to compile myself. Major alterations had to be made to the make file and tons of libraries had to be prepared. And I really, really don't want to do that again.

Any chance of a CentOS binary anytime soon? I know a lot of merchant sites could really use one. And if not - what happens if we don't upgrade?

Wow, I'm glad you guys are so smart.  I posted this on BitcoinTrading.com to spread the word.  Keep up the good work.

Exactly. When you see these (http://bitcointalk.org/index.php?topic=67634.0) kind of guys rolling out of blue, and security fix is coming up, it makes you think twice before you do anything. Bankers can gain from this one thing, the only thing they want is total control. If BTC is destroyed, then there is no alternative. Or, it would have to be done all over again, from scratch, taking into account all previous mistakes.
Cheers,
May the BitForce (http://www.butterflylabs.com/) be with you  ;D

This is already part of 0.4.4rc3 and 0.5.3rc3.

I guess the question was, when will an 0.6 RC3 be released, that includes this fix :). I use RC2 and don't want to switch back to 0.5.x.

Dia

 is there an ebuild and / or a USE flag for bip30 in the gentoo ebuild ?

0.5.3 was just released (https://bitcointalk.org/index.php?topic=68895.0), with BIP30 support.

There are 0.4.4 and 0.5.3 ebuilds. It's not optional, so no USE flag.

FWIW, this issue has been assigned CVE-2012-1909