4 ms·
Very good article, but I am not sure why Valerie dedicated so much of the content of the article to the security of hashes. Sure, there may be some CBA systems
by zamalek 13y ago
Very good article, but I am not sure why Valerie dedicated so much of the content of the article to the security of hashes. Sure, there may be some CBA systems that would require secure hashes but they would likely be few and far between (yes, BitTorrent 'cares', but they haven't seen any reason to move off of SHA1).
One could argue that weakening hashes increases the expected probability of two pieces of content having the same hash (which it does); however, the probability of the collision being exploited (as opposed to occurring by mistake) could be argued as much greater (and is what the research deals with).
I would have preferred to have seen two articles, with this one dealing with CBA exclusively and the other the security of hashing algorithms (and how that applies to CBA, for those who care). I would have expected an article on CBA to cover at least Merkle trees, for example. Non-cryptographic hashing functions such as Murmur would have also been a nice addition.
- ajb 13y agoIIRC this article was triggered by concern about use in Git and filesystems; the author is mainly a kernel developer, not a cryptographer, so it's not surprising that she hasn't extended it to cover more general applications.
- zamalek 13y agoThat makes much more sense, thank for the clarification.
- anaphor 13y agoI also would've liked to see a discussion of hash lists and hash trees, because they seem fairly difficult to implement correctly. At least for hash trees my research has shown there isn't much help implementing them correctly, except for maybe reading the bittorrent source, and there are all kinds of subtle things like some sources recommending you use a different hash function for leaves that are on the sides of the tree, etc...
- chubot 13y agoWhen you say hash tree do you mean Merkle trees? What research are you referring to? Git uses the Merkle tree scheme. It just creates a binary listing of a directory, which the SHA1 of each file in it, and then takes the SHA1 of that dir object. Is there something insufficient about that?
- anaphor 13y agoI mean merkle trees more or less yeah. I don't remember all of the sources I researched but here is a source for the use of two different hash functions (it was for leaves not edges, my bad): http://crypto.stackexchange.com/questions/2106/what-is-the-purpose-of-using-different-hash-functions-for-the-leaves-and-interna http://crypto.stackexchange.com/questions/2106/what-is-the-p...
- samatman 13y agoHave you read the Keccak group's Sakura paper? It's fairly accessible as these things go.