Especially happy with the faster map implementation and also with text/template now supporting range-over-func. Go 1.23 introduced range over function types but the latter didn't work in Go templates until now.
What would be an use case for `os.Root`? Based on my understanding ( https://github.com/golang/go/issues/67002 ), it is related to security. However, under the hood, it doesn't use `Chroot`, so I could imagine, that eventually someone finds a way to escape from the Root.
chroot only makes sense for applications which can commit to exclusively operating out of a single directory, ever. (It also requires the process to have superuser privileges, so it can't be used by applications which are run as users.)
os.Root() is more about putting a "seatbelt" on filesystem operations - like restricting operations related to an application's cache to its cache directory, or restricting a file server to serving files from the appropriate shared directory. It's not the same kind of ironclad guarantee as chroot, but it'll still protect an application from simple directory traversals.
After this dance, you can call chroot from within the new namespace. It's often also possible to use unprivileged bind-mount /dev, /sys, /proc, for a more regular execution environment (although some container runtimes block this unfortunately).
Yeah, I like your examples. In such scenarios, it makes sense when we're just trying to protect against our own bugs rather than a user deliberately sending a path that leads to the password.txt file.
Why would it use `chroot`? Combined with a sandboxing facility, like Capsicum, you can open a directory before entering capability mode and later, you use `os.Root` to open files in the file system tree under the opened directory.
I am not sure, is this custom Os.Root implementation good enough to relay on it? I see that it is based on openat, and validation of paths/symlinks. But should we expect CVEs, which will break this protection layer?
> When GODEBUG=fips140=only is used, in addition to the above, cryptographic algorithms that are not FIPS 140-3 compliant will return an error or panic.
Not sure I love the idea of stdlib panicking on purpose here. I haven't looked at the code, but I wonder if it's just in functions that don't currently return an error for backwards compat...
This is a feature that’s required in Government environments. You need a check at runtime to ensure that FIPS is set or you run the risk of breaking compliance. Which leads to inevitable audits and endless meetings. I would much prefer a panic causing an issue for 30 minutes vs. endless days of meetings to set up new controls and validations that will make your life more miserable.
The only real alternative is erroring, but those can be caught and ignored - is that something that is desireable when encryption is at stake?
Warning log levels should be avoided; it's either important and actionable (error or fatal), or it isn't, in which case it's 'info' log level. I had a blog post about it (appeal to authority) but I can't find it at the moment.
As sibling comments mentioned, someone's you want to enforce things.
Also consider it a way to avoid hidden bugs, similar to "The server chose violence" [1] [2] as well as how much Postel's law held back interoperability (which is forgotten by many aspect of FIPS certifications)
