Almost every program that we run has access to the environment, so nothing stops them from curling our credentials to some nefarious server.
Why don’t we put credentials in files and then pass them to the programs that need them? Maybe coupled with some mechanism that prevents executables from reading any random file except those approved.
Environments are per-process. Every program can have its own environment, so don’t inject secrets where they’re not needed.
I’m using bubblewrap to restrict access to FS.
The environment of other processes is readable in procfs.
/proc/PID/environ
Thanks to the permissions it’s read-only, and only by the user with which the process runs, but it’s still bad, I think
Don’t all programs run as the user anyways? That changes nothing on a single user machine
A proper server should have one user per service.
yay password rotation month
Service users generally don’t have passwords
You don’t login as service users, they’re just a means of taking advantage of the user separation features. They have the login shell set to /bin/false typically.
Not
/bin/nologin/
?From a quick search I’ve just done, the major difference is that
/bin/false
can’t return any text, the only thing it can do as specified via POSIX standards is returnfalse
.So if you set someone’s shell to
/bin/nologin
there can be some text that says “You’re not allowed shell access”, similar to what happens if you try to SSH into say GitHub.Of course, for a service account that won’t be operated by a person, that doesn’t matter - so whichever one you use is just whichever the operator thought of first, most likely.
true dat; false trends to CVE vs nologin
Some have their own users, like gitlab