switch-to-configuration-ng: don't block when the lockfile is already locked (#382797)

Author: r-vdp

nixos/modules/system/activation/switch-to-configuration.pl

@@ -95,7 +95,7 @@
 
 make_path("/run/nixos", { mode => oct(755) });
 open(my $stc_lock, '>>', '/run/nixos/switch-to-configuration.lock') or die "Could not open lock - $!";
-flock($stc_lock, LOCK_EX) or die "Could not acquire lock - $!";
+flock($stc_lock, LOCK_EX|LOCK_NB) or die "Could not acquire lock - $!";
 openlog("nixos", "", LOG_USER);
 
 # run pre-switch checks

nixos/tests/switch-test.nix

@@ -732,6 +732,23 @@ in {
         out = switch_to_specialisation("${machine}", "modifiedSystemConf")
         assert_contains(out, "starting the following units: ${dbusService}\n")
 
+    with subtest("aborts on already locked lock file"):
+        (exitcode, _) = machine.execute(
+            'flock -x --nb /run/nixos/switch-to-configuration.lock -c "${otherSystem}/bin/switch-to-configuration test"',
+            timeout=5
+        )
+        # See man timeout, exit codes above 124 come from the timeout command
+        # We want to make sure that stc actually exited with an error code,
+        # if instead we hit the timeout, then it means that stc hangs, which is
+        # what we don't want
+        # TODO: We cannot match on the exact exit code since it's not consistent between
+        # stc and stc-ng, since errno/last_os_error is not a very stable interface,
+        # we should probably get rid of that in stc-ng once we got rid of the
+        # perl implementation
+        assert exitcode < 124, \
+          "switch-to-configuration did not abort as expected, " + \
+          f"probably it timed out instead (exit code: {exitcode}), 124 means timeout"
+
     with subtest("fstab mounts"):
         switch_to_specialisation("${machine}", "")
         # add a mountpoint

pkgs/by-name/sw/switch-to-configuration-ng/src/src/main.rs

@@ -1037,7 +1037,7 @@ fn do_system_switch(action: Action) -> anyhow::Result<()> {
         die();
     };
 
-    let Ok(_lock) = Flock::lock(lock, FlockArg::LockExclusive) else {
+    let Ok(_lock) = Flock::lock(lock, FlockArg::LockExclusiveNonblock) else {
         eprintln!("Could not acquire lock");
         die();
     };