7 ms·
Ethereum fixes serious “eclipse” flaw
- elmar 9y agothe image in the article is really appropriated, the Turing complete VM of ethereum is a black hole of bugs, on the Genesis, some really bad engineering decisions were made and probably now is too late to change it.
- nissarup 9y agoWell... As it is Turing complete you could just as well say that it is a black hole of awesome stuff. Which, of course, is true of any Turing complete language.
- tree_of_item 9y agoThe image is an eclipse, not a black hole.
- gus_massa 9y agoIIUC it's not a bug in the VM that is Turing complete and run the "smart contracts". The bug is in the surrounding code that tries to connect to the real network of the other Ethereum users and decides if the connection is trustworthy. So a similar attack could be run against other coins. [The article says that Bitcoin is safer, but the article doesn't explain why. Can someone give more details?]
- EthanHeilman 9y ago>The article says that Bitcoin is safer, but the article doesn't explain why. Can someone give more details? We wrote a paper [0] in 2015 performing eclipse attacks against Bitcoin. You can watch me talk about it here [1]. At the time it required an attacker with 400 IP addresses. Many of our countermeasures were adopted by Bitcoin-core raising the difficulty of the attack even further. In fact at this very moment I am still working on adding some of our countermeasures to Bitcoin-core [2]. In our attack on Ethereum, only 2 IP addreses were required. We discuss this in our paper: >"We present new eclipse attacks showing that, prior to the disclosure of this work in January 2018, Ethereum’s peer-to-peer network was significantly less secure than that of Bitcoin. Our eclipse attackers need only control two machines, each with only a single IP address. [..] By contrast, the best known off-path eclipse attacks on Bitcoin [23] require the attacker to control hundreds of host machines, each with a distinct IP address. For most Internet users, it is far from trivial to obtain hundreds (or thousands) of IP addresses. This is why the Bitcoin eclipse attacker envisioned by [23] was a full-fledged botnet or Internet Service Provider, while the BGP-hijacker Bitcoin eclipse attacker envisioned by [17] needed access to a BGP-speaking core Internet router. By contrast, our attacks can be run by any kid with a machine and a script." -Low-Resource Eclipse Attacks on Ethereum's Peer-to-Peer Network [3] [0]: https://eprint.iacr.org/2015/263.pdf https://eprint.iacr.org/2015/263.pdf [1]: https://www.usenix.org/node/190891 https://www.usenix.org/node/190891 [2]: https://github.com/bitcoin/bitcoin/pull/9037 https://github.com/bitcoin/bitcoin/pull/9037 [3]: http://www.cs.bu.edu/~goldbe/projects/eclipseEth http://www.cs.bu.edu/~goldbe/projects/eclipseEth
- heliumcraft 9y agoOut of curiosity, what "really bad engineering decisions" are those?
- foepys 9y agohttps://news.ycombinator.com/item?id=14810008 https://news.ycombinator.com/item?id=14810008 This applies to "Solidity", the most commonly used language to write smart contracts for Ethereum. The VM itself is not so bad but also simpler.
- heliumcraft 9y agoSolidity != EVM & Ethereum.
- elmar 9y agoMy view, of course, every decision as a trade-off, everything i say is controversial - The main ledger and smart contracts should be separated, everyone smart contract is public, why?? the smart contracts should be run on several specific sidechains. Why if I buy a crypto kittie everyone as to know and store the transaction for all eternity. Wait sharding is coming, yes but will reduce the security guarantees.
- heliumcraft 9y agoThere are different solutions for that in the works, from sharding to plasma, etc.. and those have their own challenges & tradeoffs (namely in terms of security, etc..) However engineering decisions 'on genesis' was over 3-4 years ago and at that time doing a blockchain with smart contracts was already challenge on itself. It's not really fair to take the knowledge and lessons from the present day and claim those in the past without that knowledge made horrible decisions. It's a bit like blaming google for not starting angular 1 with the functionality of angular 4, or saying that John Resig should have created Babel and ES7 features instead of creating jQuery.
- elmar 9y ago
- heliumcraft 9y agoThis wasn't a flaw with Ethereum itself or its protocol, but rather an issue with one of the its client implementations (go-ethereum), other clients like Parity, Harmony, etc.. are unaffected.
- UncleMeat 9y agoIs there any other field of computer security where this argument is acceptable? Usable security matters. Ecosystems matter. You cannot put a boundary somewhere in the middle of what the end user sees and call your security job done. When people majorly botch x509 cert validation because the spec is so monstrously complex that's still a problem with x509.
- dahdum 9y agoEthereum specifically chose to develop around multiple clients to mitigate the risks of implementation errors such as this. It would be ideal if all were perfect from the start, but no software is, security or otherwise.
- heliumcraft 9y agoOne can't claim the Web is broken just because a bug was found in one of the many browsers.
- WiseWeasel 9y agoThe article states it's only been discovered and mitigated in geth, allegedly the most popular ethereum client; whether other clients are vulnerable is unknown. This statement from ethereum/geth developer Felix Lange even hedges on how completely the vulnerability has been mitigated, which may not bode well for ethereum in general: >We have done our best to mitigate the attacks within the limits of the protocol. The paper is concerned with 'low-resource' eclipse attacks. As far as we know, the bar has been raised high enough that eclipse attacks are not feasible without more substantial resources... He did go on to mention his belief alternative ethereum client Parity isn't vulnerable, so there's that at least.
- josephagoss 9y agoA thing I've noticed about cryptocurrency news. The people that are pro cryptocurrency are often far too optimistic and sometimes are not the most academic people. However I've also noticed that those that constantly critique cryptocurrency are often so wrong about their assumptions it's just pointless reading on. It's so hard to actually get a reasonable discussion about cryptocurrency.
- drcode 9y agoThat's how you can tell that cryptocurrency really is a from of money... only money could lead to this many strawman arguments and online mudslinging :-)
- dvh 9y agoMaybe we should create some discussion system where in order to post you need proof of expertise (PoE)
- Tinyyy 9y agoWell it’s hard to be negatively exposed to crypto so the discussion will be biased.
- osteele 9y agoYou can short crypto on Cboe and CME. A few minutes on the internet will reveal other means for shorting crypto.
- douglaswlance 9y agoWith a methodology for generating consensus...
- desireco42 9y agoExcellent, and we need to release the token that you will get based on PoE.
- ty_a 9y agoMaybe you could count number of Ph.D contributors? I bet Bitcoin has as many or more than every other project combined.
- rargulati 9y agoFor those interested, here's the paper: https://www.cs.bu.edu/~goldbe/projects/eclipseEth.pdf https://www.cs.bu.edu/~goldbe/projects/eclipseEth.pdf Here's the partial-implementation of Countermeasure 2 in geth: https://github.com/ethereum/go-ethereum/pull/16069 https://github.com/ethereum/go-ethereum/pull/16069 Of note, this fix is still susceptible to attacks, though still an improvement: > Since geth v1.8.0 still allows the attacker to freely craft node IDs that land in specific buckets, the partial implementation of Countermeasure 2 in geth v1.8.0 means that for a given victim’s table, there can be at most 10 attacker node IDs associated with each attacker IP address. While this improves on the situation prior to geth v1.8.0 (where we could eclipse our victim using just one or two IP addresses), it does not raise the bar for attackers quite as high as we had hoped. Still making my way through the paper and the PR to identify other things fixed. Here's the entire 1.8 release: https://github.com/ethereum/go-ethereum/releases/tag/v1.8.0 https://github.com/ethereum/go-ethereum/releases/tag/v1.8.0
- swang 9y ago> The fix for the second attack involved limiting the number of outgoing connections a target can make to the same /24 chunk of IP address to 10. The changes are designed to make it significantly harder to completely isolate a user from other legitimate users. Maybe it is the way the article is written, or how I've read it, but the original flaw involved just 2 nodes w/ 2 IP address. This fix sounds like you need 2 nodes w/ 2 IP addresses that don't have the same /24 octal? Is it that much more difficult? > When even a single node presents users with a different version of the blockchain, they will be warned of an error that effectively defeats the attack. Doesn't this just mean the attacker has to make sure to control all 13 connections before starting the attack?
- goldenkey 9y agoI was downvoted when I mentioned this issue 52 days ago: https://news.ycombinator.com/item?id=16114295 https://news.ycombinator.com/item?id=16114295
- panarky 9y agoThis is an attack on the Kademlia DHT. Why exactly does Ethereum use Kademlia anyway? Kademlia's main purpose is to find content distributed among a small percentage of nodes. Ethereum doesn't need this, since the entire blockchain is stored on every node. And does the attack also work against other applications that use Kademlia, such as IPFS and I2P?
- GistNoesis 9y agoThere are some fun systemic probabilistic attacks based on peer discovery. One such principle which doesn't need many attacking nodes, rely on peers self-clustering themselves over time is the following : You divide the peer network in 2 groups. Using a few number of attacking nodes, when a peer from one group ask you for a node you know, you answer with a peer from the same group. Then each peer from each group is more likely to recommend a peer from the same group. Over time bias accumulate, and cluster forms, with very few bridges between the two groups which you control. Obviously it can be mitigated easily by adding some hard-coded "central" nodes or monitoring the network. Afaik robust peer discovery in p2p network is still not solved.
- zebraflask 9y agoAfter CryptoKitties, it's difficult to take Ethereum seriously.
- bufferoverflow 9y agoThat's like saying "after the fart app, it's hard to take iOS seriously".
- zebraflask 9y agoI don't really take iOS all that seriously, either. Of all the things anyone would trade for money Cryptokitties is really riding the line between smart and so dumb it devalues the concept of Ethereum, which might have been the point.
- davesque 9y agoSo...because we got hamsterdance.com at one point I guess that means the internet is hard to take seriously?
- zebraflask 9y agoHamsterdance.com was a masterpiece of early Internet adoption. Try to get anything that viral. Ethereum is a weak attempt at money laundering.
- dlubarov 9y agoIsn't relying on IP diversity as a measure of attack difficulty a bad assumption in the first place? If an attacker has access to the cable outside my home, then they can pretend to control as many IPs as they want. It seems to me that the most reliable defense against eclipse attacks is to just compare the block rate of the longest chain we know about to the historical average. If the block rate is roughly normal, then either we're seeing the real longest chain, or an attacker has acquired nearly 50% of hash power, which is unlikely. The same idea works with proof of stake -- if the participation rate drops substantially below the historical average, we're probably not seeing the main chain, and we shouldn't consider anything final until the participation rate return to normal.
- EthanHeilman 9y ago> Isn't relying on IP diversity as a measure of attack difficulty a bad assumption in the first place? If an attacker has access to the cable outside my home, then they can pretend to control as many IPs as they want. If an attacker has access to the cable outside your home they have already won, it's game over. >It seems to me that the most reliable defense against eclipse attacks is to just compare the block rate of the longest chain we know about to the historical average. If the block rate is roughly normal, then either we're seeing the real longest chain, or an attacker has acquired nearly 50% of hash power, which is unlikely. Block creation have high variance. Over short periods of time, say a few hours no blocks being announced is fairly common. Over long periods of time how do you know the mining power has not decreased or increased at a slow rate than your person prediction. >The same idea works with proof of stake -- if the participation rate drops substantially below the historical average, we're probably not seeing the main chain, and we shouldn't consider anything final until the participation rate return to normal. Interesting you can design PoS systems such that if not a quorum of stake is online, no new blocks are created. Algorand has some very neat properties in this regard.
- dlubarov 9y ago> If an attacker has access to the cable outside your home they have already won, it's game over. Not if I'm looking at the block times or PoS participation rates. The attacker can feed me their own fork while preventing me from seeing the actual highest-difficulty chain, but their fork will be suspiciously weak, so I'll know not to trust it. > Block creation have high variance. Over short periods of time, say a few hours no blocks being announced is fairly common. Over long periods of time how do you know the mining power has not decreased or increased at a slow rate than your person prediction. Yeah; with Bitcoin you would need to wait several hours to get a good estimate of hash power. With more granular blockchains like Ethereum you could get a good estimate more quickly.