nixos-enter: inherit parent PID namespace

jmbaur · jmbaur · commit 798aa80b7d5e · 2024-12-19T19:58:22.000-08:00
Removes the creation of a new PID namespace during nixos-enter so that systemd doesn't detect the child process as being in a "container". In systemd v257, container detection slightly changed to check if the process is a part of the root PID namespace (see https://github.com/systemd/systemd/blob/96c4d9d94d06c6c0a8b68be376505f8d8b5eba2b/src/basic/virt.c#L735). Tooling such as `bootctl` will not perform certain actions if it detects we are in a container, such as not populating EFI variables. This results in broken systemd-boot VM tests.
diff --git a/pkgs/by-name/ni/nixos-enter/nixos-enter.sh b/pkgs/by-name/ni/nixos-enter/nixos-enter.sh
@@ -12,7 +12,7 @@ if [ -z "$NIXOS_ENTER_REEXEC" ]; then
     if [ "$(id -u)" != 0 ]; then
         extraFlags="-r"
     fi
-    exec unshare --fork --mount --uts --mount-proc --pid $extraFlags -- "$0" "$@"
+    exec unshare --fork --mount --uts --mount-proc $extraFlags -- "$0" "$@"
 else
     mount --make-rprivate /
 fi
