5 ms·
Git Security Vulnerability Announced
- rosndo 5y ago> This vulnerability affects users working on multi-user machines where a malicious actor could create a .git directory in a shared location above a victim’s current working directory. On Windows, for example, an attacker could create C:\.git\config, which would cause all git invocations that occur outside of a repository to read its configured values. lol. “vulnerability” Neither of these were worth an announcement blog post.
- nijave 5y agoOn Windows, you can just put whatever you want in `git.cmd` to achieve the same thing since it adds the current directory to PATH implicitly
- csense 5y agoSo wait a minute. The situation here is that a directory Alice owns is a subdirectory of a directory Eve owns. I find it highly surprising that anyone would configure a multi-user machine where a parent directory is owned by an untrusted user different from the current user. Usually a parent directory would be owned by the same user as the current directory, root, or possibly a system service account. Although now that I think about it, /tmp might be an example -- if Eve is a local user, she can create /tmp/.git. Then if Alice tries to use the git command in a subdirectory of /tmp that's not already a Git repository, she might be bitten by this. It's still a pretty far-fetched scenario, but I suppose it's possible a non-zero number of users are affected. Are multi-user machines even that common? I was under the impression that most people don't really use OS level user accounts, and instead use VM's or containers if they want to have multiple people using the same physical box (especially if those people are untrusted). AFAIK traditional Unix shared hosting is mostly in the past. I use Git regularly but I'm not going to lose any sleep over this one.
- nijave 5y agoMaybe (abused) jump servers or "admin" servers? (Those abused catchall servers will scripts and crons that don't have their own systems get thrown)
- remram 5y ago/tmp seems like a huge target here. I definitely run `git clone` in /tmp, sometimes on shared machines (HPC). I even run `git log` or similar in there, when I'm distracted I forget to `cd` down first.
- junon 5y agoYet another case where the CVE system is absolutely broken. This has no reason to be a "medium" vulnerability, just like regular expression complexity DOS (redos) vulnerabilities have no business being classified as "high risk" in the majority of cases they're found. CVE scoring is antiquated and obsolete IMO. It means nothing to me anymore.
- icegreentea2 5y agoDo you think the scores in this case is too high or too low?
- junon 5y agoToo high. But it's not exactly about the score. The components of each rating don't make sense for how software is written these days. For example, since libraries aren't always used in networking contexts, but usually can be, CVE scores against them almost always include "network vector" as a means of exploitation, which immediately puts the score into the high 5 or 6 (medium high, if memory serves) - even if the library has nothing to do with networking. It makes little sense these days as opposed to back in the earlier days of the internet where many network-relayed applications were self contained and didn't use a lot of disparate libraries like we do today. Stuff like that - it's antiquated. It needs to be redone in my opinion.
- remram 5y agoI disagree. DOS should probably not be "high" but this has a non-imaginary arbitrary code execution. If this is not "medium", what would be a "medium" for Git? Any remote code execution should definitely be "high".
- junon 5y agoAnd by definition these days, almost every exploit is remotely exploitable under the right conditions. We put everything in front of a network gateway nowadays.
- 5y ago
- mdaniel 5y agoI'd guess when dang changed the URL for the other submission it evaded the dupe check but the currently #1 with 61 comments version: https://news.ycombinator.com/item?id=31009675 https://news.ycombinator.com/item?id=31009675