Re:Update on Meltdown and Spectre
by liamzebedee on Thursday January 04, 2018 @11:23PM
(#16076510)
[related] Has anyone considered the possibility of a Spectre-style attack in Ethereum's Turing-complete EVM? Not that the state would be unique for all contracts, but there's a possibility of communicating to an external contract with the output.
[ Parent
| on HN ]
From my understanding there are a couple things that probably make this a non-issue:
1) Worst case scenario, it only allow you to read the memory of systems running validator nodes. There aren't very many of these (tens of thousands perhaps) and they generally don't store particularly valuable secrets. A small number of them may store private keys to Eth accounts with a small amount of Eth, but that's not typical operating procedure.
2) The EVM doesn't have any internal mechanism for measuring time (beyond existing blocks), so any timing attack within the EVM would require some very clever way of measuring time.
There's a twitter thread where some cryptographers speculate on the possibility. The upshot is that it doesn't seem like it will be possible:
In the Spectre paper they note that while Chrome degrades the resolution of `performance.now()`, they were able to get a timer with sufficient resolution by using a Web Worker (thread) which repeatedly decrements a value in shared memory. As far as I know the EVM intentionally doesn't provide any concurrency because execution must be deterministic, and it seems doubtful that any kind of message-passing from outside the contract would be fast enough to provide the resolution needed.
However, the block-lattice cryptos like RaiBlocks that find a way to build in concurrency and shared memory might be different.
Does the Ethereum VM have branch prediction or speculative execution. I'm not sure how you'd construct a side channel of this nature against it if it doesn't.
Re:Update on Meltdown and Spectre
by TazeTSchnitzel on Friday January 05, 2018 @06:51AM
(#16078058)
The VM just needs branches, it's the host CPU's prediction you're messing with.