Is it time to deprecate md5?

[copumpkin]

It's terrible hash function under any assumption of malicious actors. It would be nice in an upcoming Nix release to print out a warning that a fixed-output derivation is using md5 and instruct people to move to something more sensible. Then perhaps in a couple of releases we could break it altogether except after someone opts in.

SHA1 isn't great, but isn't as bad. It would be nice to set some precedent on what to do about that, too.

[edolstra]

👍

[domenkozar]

👍 for deprecation (and noting in what version the support will be removed).

[copumpkin]

I'm not one for release schedules, but perhaps add it as a warning to 1.12, make it an error that can be overridden with e.g., --allow-insecure-md5 in 1.13, and make it unoverridable in 1.14. The error message in 1.14 onwards would tell you what to do instead and will stay for foreseeable future, so it won't just say something unfriendly like "unknown hash: md5".

[vcunat]

We should start by phasing md5 out in nixpkgs. After that is done, we can remove support in nix but I don't think that part is really important as it's a kind of opt-in.

[copumpkin]

It seemed easier to start from Nix since there are a dozen fixed-output
derivation functions that could each warn. Perhaps we can jump in at the
stdenv/generic.nix level, now that I think about it?
On Mon, Feb 15, 2016 at 12:54 Vladimír Čunát [email protected]
wrote:

We should start by phasing md5 out in nixpkgs. After that is done, we
can remove support in nix but I don't think that part is really important
as it's a kind of opt-in.

—
Reply to this email directly or view it on GitHub
#802 (comment).

[vcunat]

I see, changing nix is probably the easiest way to produce warnings in all these cases.

[davidak]

👍

@vcunat some people use private packages.

[ivan]

chromium/update.nix appears to use an md5 collision to achieve something.

[copumpkin]

@edolstra would you support adding an annoying warning message for md5 for the next release or two, and then removing support for it after that?

[edolstra]

For backward compatibility, it would be better to deprecate/remove it in Nixpkgs. Otherwise we would lose the ability to build packages from old Nixpkgs versions.

[FRidh]

md5 support was removed in NixOS 17.03 so we're 4 releases without by now.

[domenkozar]

It would still make sense to deprecate md5 in Nix itself?

[edolstra]

No, we need to keep it for backward compatibility. We don't want to lose the ability to build old Nix expressions.

[domenkozar]

I wasn't clear enough. I'd only add a warning when md5 is used like so:

filename:line:col: md5 fixed-output derivations are deprecated

[vcunat]

I suppose we're finally ready for that: NixOS/nixpkgs@46cf3a51269 ;-)