Arch Linux

The md5sums are only used for a simple and quick integrity check to see if the file was downloaded correctly (e.g. if there were no network error). There is no need to add better hash sums as they wont improve security. You'll need signed packages for this. (see pacman-dev list for more details)

@Pierre
The MD5 sums are only used for that *by pacman*. My script paccheck uses the MD5 sums to try to verify packages more comprehensively using several mirrors, but the MD5 hash is a weakness. Users need to (and do) download full packages from multiple mirrors as a result. Having the SHA256 sum would thus help many users and reduce load on the mirrors, regardless of pacman's lack of use for it. Also, I have submitted a feature request to devtools for them to sign the database with a server key. Adding SHA256 sums is an integral part of that request as well, since if the database is signed, the SHA256 sum can be used to authenicate the packages reliably. It is very simple to do and doesn't hurt anything, so please add SHA256 sums to repo-add. Thank you.

> My script paccheck uses the MD5 sums to try to verify packages more comprehensively using several mirrors, but the MD5 > hash is a weakness. Users need to (and do) download full packages from multiple mirrors as a result.

I was playing with a similar idea about 4 months ago. The point is that several mirrors are unlikely to get hacked simultaneously, right? Although I added my vote for this ticket, I have to say that for paccheck (is it available, btw?) sha256 is irrelevant, since you are using statistical unlikeness of a compromise, not trustability of any particular .db.tar.gz package...

@Leonid Isaev
Yes paccheck is available:
http://igurublog.wordpress.com/downloads/script-paccheck/
https://bbs.archlinux.org/viewtopic.php?id=113917&p=1

> The point is that several mirrors are unlikely to get hacked simultaneously, right?

Yes, and paccheck can test an unlimited number of mirrors, so it improves the situation considerably through polling. Pretty much to the user's desired level vs time and bandwidth.

> sha256 is irrelevant, since you are using statistical unlikeness of a compromise, not trustability of any particular .db.tar.gz package...

With the MD5 sums, paccheck can download the database file from multiple mirrors. The MD5 sum provides some degree of authentication, aside from an MD5 collision attack. If a user is concerned about this vulnerability (which is reasonable), then the user must use paccheck's 'compare' function to download full packages from multiple mirrors and do a byte-for-byte comparison. Having the SHA256 hash would change this requirement - comparing the relatively small database files from multiple mirrors would assure that an MD5 collision is not present. Then the only vulnerability is multiple compromised mirrors, which is relatively unlikely if enough tier 1 mirrors are polled.

So in short, I would find the SHA256 sums very valuable in making paccheck more efficient and secure, and of course they are a component of the server signature scheme in the devtools request.

I find it hilarious you call md5 a compromised hash, but then try to throw percentages out saying it is unlikely someone would hack multiple mirrors. The chance of someone owning all our mirrors is a hell of a lot higher than someone falsifying a package with a clashing md5sum.

Either way, I see no actual patch here. We take git format-patch style patches for pacman, on the ML, inline, as is documented here: http://www.archlinux.org/pacman/submitting-patches.html. The rules are quite easy to follow.

Are you willing to add it if I take the time to submit a patch, or are you just wasting my time? I ask because thus far I have met nothing but unwillingness, so please don't waste my time. I don't really see why a patch is necessary as it is a trivial addition, but if you want one I will be happy to provide it. Thank you.