Github co-creator & Pro Git author skewers Git 3.0's SHA-256 move
In a well-fleshed-out post, Scott Chacon shows how unneecessary Git 3.0's move to replace SHA-1 with SHA-256 is.
https://blog.gitbutler.com/git-3-sha-256Open linkView original on leminal.space
22 replies
It's sad to see this community only listening to people because they're rich: there are a lot of better engineers who knows more than this guy but they're not "co-creator of GitHub". I read this title as boot licking silicon valley.
That being said you don't hash git commits for security reason : YOU SIGN YOUR COMMITS FOR SECURITY. Sorry for the caps but let's make this visible.
But signing keys can be stolen, have to be updated, revoked etc. A secure hash is an elegant way to say "this repo contains what I want"
There's people who seriously think commit hashes are there for security purposes? Furthermore, there are git MAINTAINERS that think this?
How absurd. Author is completely right, making SHA-256 the default implementation serves no purpose. SHA-1 is a completely valid algorithm for non-security use cases, which is precisely how git uses it. This is a solution looking for a problem.
I think he's maybe overblowing the impact of this change. Realistically, the only ones who are going to be impacted are the folks who maintain git-related tools and forges, as he mentions. The rest of us probably won't even notice. But that's still a ton of pointless work for those folks.
Wait, they don't affect security? Wouldn't a hash collision mean pulling that hash from GitHub would pull wrong code? Or maybe delete code? I'd assume the hash is used as a lookup key in a database somewhere.
You also pin dependencies to specific hashes for security reasons.
If you need security, you must sign the commits.
The blog post lays it out pretty well.
The vulnerability of SHA-1 is that it's possible for an attacker to find colliding hashes. But that's not really an attack vector, because it's not like they can find aatching hash for just ANY input. In order to pull off an attack, using a hash collision, an attacker would have to already be trusted within the controlling system, in which case they can do a hell of a lot worse than swapping out a file.
Like you say, the hash is used as a lookup key within the git repo's database. The requirement for that is uniqueness, which SHA-1 still fulfills just fine. The trust isn't in the database lookup, the trust is in the systems that controls access to and distribution of that database.
As the article points out, Linus Torvalds himself said as much in 2005, when he first wrote git:
Maybe... But I don't look forward to working with the dozens of IT folks at my company who all installed git once and have never upgraded. This will be multiple meetings, show up on slide decks, break builds...
It's going to be a right pain for nothing.
This post misses, uh, a few important things that get mentioned in the lobsters comments: https://lobste.rs/s/bytzgl/git_3_0_s_upcoming_sha_256_default_will_be
Thanks for sharing.
Why didn't I know lobsters before? This look 1000 times better than HN... oh you can't join. Makes sense it isn't that known.
His defense of SHA-1:
For the most part the article mostly talks about collision attacks, which quite frankly i think is silly.
He also talks about hashes pointing to other (older) projects using a different hashing algorithm, but can't the software just detect if its a SHA-1 hash or SHA-256, and fetch it accordingly? He acts like its the end of the world when in my mind most everything can (and probably will) be compensated for pretty easily. It's reminiscent of the IPv6 fear-mongering.
The thing that gets me is he doesn't say if the current hashing algorithm or the new one has any support for when hashes do collide. Since that can theoretically happen with both, to me that sounds like a problem worth tackling.
Collision concerns make sense. If you manage to get two different files with the same hash, that will cause issues for your repo - the concerns over a collision justify the move to SHA-256 though.
I would have thought they’d prefix the hash with “sha256-“ or something, or even just 256 prefix the hash.
They have different fixed lengths, no prefix needed to tell them apart.
Like how y2k wasn't a worry.
This statement requires nuance. It was a nothingburger because people patched and upgraded ahead of time.
Madness
We had a similar stupid issue at my last job. Someone made an edict that all Sha-1 and md5 references in the code must be removed for security reasons.
Why did they skip the 255 versions?!
/s
Don't be insane, they only skipped 254 versions
Filler episodes.
Every time I use the shitty command-line flags of git, there is a laugh track.
I'm not qualified to talk about it but Hacker News has arguments against his post: https://news.ycombinator.com/item?id=49924179
You're probably more qualified than most HNers, just less confident, less inclined to indulge in performative expertise and pseudo-intellectual posturing, and more resistant to joining the circle-jerks with those who do indulge.