Thursday, August 18. 2005
Say goodbye to SHA-1
Comments
Display comments as
(Linear | Threaded)
Couldn't gentoo (and others) use two algorithms (md5 AND sha-1 or an other one) to check for errors and possible modification? Surely fooling one algorithm is becoming easier, but fooling two on the same file has to be way more complicated ... wouldn't it?
Rémi
Rémi
That doesn't help that much. It only improves the security to the better one of the hash-functions and some bits more. It makes no sense to use two broken functions, but it maybe is a good idea to combine several of the new hash-functions, because they're not so well testet and may discover unexpected weaknesses (e.g. using sha-256 + whirlpool + tiger).
I think you didnt understand the initial comment. Trying to modify a file in such a way that would produce a collision with one algo surely cannot also produce a collision with another. The question involves whether there is any gain if integrity is your goal and you take a SHA1 of a given file as well as an MD5 (both of which are 'broken') -- not an MD5 of a SHA1 sum.
-tjs
-tjs
While these results sure are a great advance in cryptography, their effect on practical uses of MD5 or SHA1 is greatly overrated, IMO.
For example, the story on MD5 is here -> http://www.schneier.com/blog/archives/2005/02/cryptanalysis_o.html
It says that you can produce a collision in such and such time and that the algorithm is not mathematically perfect since it produces collisions in the first place. This is very different from being able to break an MD5 digest in such and such time - to do it you need to generate a new message with a given hash, not any two messages with the same hash.
I have not read the entire paper, but the fact that collisions exist does not, by itsefl, imply that they exist for any digest.
In short - even if you were able to create a collision for a given digest, which AFAIK you can not do, there are far simpler and cheaper ways of breaking our precious portage tree :)
For example, the story on MD5 is here -> http://www.schneier.com/blog/archives/2005/02/cryptanalysis_o.html
It says that you can produce a collision in such and such time and that the algorithm is not mathematically perfect since it produces collisions in the first place. This is very different from being able to break an MD5 digest in such and such time - to do it you need to generate a new message with a given hash, not any two messages with the same hash.
I have not read the entire paper, but the fact that collisions exist does not, by itsefl, imply that they exist for any digest.
In short - even if you were able to create a collision for a given digest, which AFAIK you can not do, there are far simpler and cheaper ways of breaking our precious portage tree :)
Hanno Böck explained the current state of SHA1-brokeness.
Tracked: Aug 18, 12:34
SHA-1 ist tod. ...
Tracked: Aug 18, 15:38
As the article some days ago about SHA1 got a lot of interest, I thought I'll write some more background info about this, especially for people thinking that collisions aren't a big problem. Cryptographic hash functions are functions where you can put
Tracked: Aug 22, 01:05