Use chroot over pivot_root

[jD91mZM2]

We only used pivot_root in order to temporarily bind /nix to the host and lift in different files. Now, we instead bind everything beforehand and finally do a simple chroot.

---

As discussed in #22, I kept the container.sh script.

You suggested only using the CLONE_NEWUSER flag in case uid != 0, however, that did not work for me. For me, that seemed to not clone the namespace at all (I know, right, wtf?) because mount listed the temporary directories we created. I got me a real scare when I realised that tempdir would probably try to remove the files recursively, although I rebooted and my system doesn't appear to be harmed.

Thus, this PR does not tackle the issue you perceived when the user was root. Sorry.

[Mic92]

I mean only using CloneFlags::CLONE_NEWNS instead of CLONE_USER. Also if the tempdir is only deleted by the process not in the chroot than it would not see the bind mounts when cleanup up everything.

[jD91mZM2]

I was unclear, I meant "only" CLONE_NEWUSER when X as if "only in this condition". Of course CLONE_NEWNS should always be used, which is why it's so strange and scary honestly that the binds seemed to show up anyway when running mount to list all mounts

[Mic92]

At least with docker I get:

root@cdcb9eb19dc3:/app# run bash
    Finished dev [unoptimized + debuginfo] target(s) in 0.01s
     Running `target/debug/nix-user-chroot .nix bash`
thread 'main' panicked at 'unshare failed: Sys(EPERM)', src/main.rs:108:70
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
thread 'main' panicked at 'failed to remove temporary directory: /tmp/.tmpL9ifcs', src/main.rs:186:21
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

adding --privileged makes it work.

[jD91mZM2]

Yay, thanks 😄