Merge pull request #104094 from flokli/systemd-unified-cgroup-hierarchy

systemd: switch to unified cgroup hierarchy by default

7 files changed, 71 insertions, 16 deletions

diff --git a/nixos/doc/manual/release-notes/rl-2103.xml b/nixos/doc/manual/release-notes/rl-2103.xml
index 10d5cda77464..a099bbb76384 100644
--- a/nixos/doc/manual/release-notes/rl-2103.xml
+++ b/nixos/doc/manual/release-notes/rl-2103.xml
@@ -277,6 +277,19 @@
      <literal>unbound-control</literal> without passing a custom configuration location.
     </para>
    </listitem>
+   <listitem>
+    <para>
+     NixOS now defaults to the unified cgroup hierarchy (cgroupsv2).
+     See the <link xlink:href="https://www.redhat.com/sysadmin/fedora-31-control-group-v2">Fedora Article for 31</link>
+     for details on why this is desirable, and how it impacts containers.
+    </para>
+    <para>
+     If you want to run containers with a runtime that does not yet support cgroupsv2,
+     you can switch back to the old behaviour by setting
+     <xref linkend="opt-systemd.enableUnifiedCgroupHierarchy"/> = <literal>false</literal>;
+     and rebooting.
+    </para>
+   </listitem>
   </itemizedlist>
  </section>
 </section>
diff --git a/nixos/modules/security/hidepid.nix b/nixos/modules/security/hidepid.nix
index 55a48ea3c9c6..4953f517e93b 100644
--- a/nixos/modules/security/hidepid.nix
+++ b/nixos/modules/security/hidepid.nix
@@ -23,5 +23,9 @@ with lib;
 
     boot.specialFileSystems."/proc".options = [ "hidepid=2" "gid=${toString config.ids.gids.proc}" ];
     systemd.services.systemd-logind.serviceConfig.SupplementaryGroups = [ "proc" ];
+
+    # Disable cgroupsv2, which doesn't work with hidepid.
+    # https://github.com/NixOS/nixpkgs/pull/104094#issuecomment-729996203
+    systemd.enableUnifiedCgroupHierarchy = false;
   };
 }
diff --git a/nixos/modules/services/cluster/k3s/default.nix b/nixos/modules/services/cluster/k3s/default.nix
index 2e8bf20a68fc..f0317fdbd160 100644
--- a/nixos/modules/services/cluster/k3s/default.nix
+++ b/nixos/modules/services/cluster/k3s/default.nix
@@ -76,6 +76,10 @@ in
       enable = mkDefault true;
     };
 
+    # TODO: disable this once k3s supports cgroupsv2, either by docker
+    # supporting it, or their bundled containerd
+    systemd.enableUnifiedCgroupHierarchy = false;
+
     systemd.services.k3s = {
       description = "k3s service";
       after = mkIf cfg.docker [ "docker.service" ];
diff --git a/nixos/modules/system/boot/systemd.nix b/nixos/modules/system/boot/systemd.nix
index c22264b3e92d..cbf9e7b49d36 100644
--- a/nixos/modules/system/boot/systemd.nix
+++ b/nixos/modules/system/boot/systemd.nix
@@ -550,6 +550,14 @@ in
       '';
     };
 
+    systemd.enableUnifiedCgroupHierarchy = mkOption {
+      default = true;
+      type = types.bool;
+      description = ''
+        Whether to enable the unified cgroup hierarchy (cgroupsv2).
+      '';
+    };
+
     systemd.coredump.enable = mkOption {
       default = true;
       type = types.bool;
@@ -1178,6 +1186,7 @@ in
     boot.kernel.sysctl = mkIf (!cfg.coredump.enable) {
       "kernel.core_pattern" = "core";
     };
+    boot.kernelParams = optional (!cfg.enableUnifiedCgroupHierarchy) "systemd.unified_cgroup_hierarchy=0";
   };
 
   # FIXME: Remove these eventually.
diff --git a/nixos/modules/virtualisation/docker.nix b/nixos/modules/virtualisation/docker.nix
index d87ada35a0ae..ec257801b330 100644
--- a/nixos/modules/virtualisation/docker.nix
+++ b/nixos/modules/virtualisation/docker.nix
@@ -155,6 +155,9 @@ in
       users.groups.docker.gid = config.ids.gids.docker;
       systemd.packages = [ cfg.package ];
 
+      # TODO: remove once docker 20.10 is released
+      systemd.enableUnifiedCgroupHierarchy = false;
+
       systemd.services.docker = {
         wantedBy = optional cfg.enableOnBoot "multi-user.target";
         environment = proxy_env;
diff --git a/nixos/tests/podman.nix b/nixos/tests/podman.nix
index cd8c2b4308c8..bccd2de7c9b9 100644
--- a/nixos/tests/podman.nix
+++ b/nixos/tests/podman.nix
@@ -34,7 +34,6 @@ import ./make-test-python.nix (
       podman.wait_for_unit("sockets.target")
       start_all()
 
-
       with subtest("Run container as root with runc"):
           podman.succeed("tar cv --files-from /dev/null | podman import - scratchimg")
           podman.succeed(
@@ -53,22 +52,32 @@ import ./make-test-python.nix (
           podman.succeed("podman stop sleeping")
           podman.succeed("podman rm sleeping")
 
-      with subtest("Run container rootless with runc"):
+      with subtest("Run container as root with the default backend"):
+          podman.succeed("tar cv --files-from /dev/null | podman import - scratchimg")
+          podman.succeed(
+              "podman run -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
+          )
+          podman.succeed("podman ps | grep sleeping")
+          podman.succeed("podman stop sleeping")
+          podman.succeed("podman rm sleeping")
+
+      with subtest("Run container rootless with crun"):
           podman.succeed(su_cmd("tar cv --files-from /dev/null | podman import - scratchimg"))
           podman.succeed(
               su_cmd(
-                  "podman run --runtime=runc -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
+                  "podman run --runtime=crun -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
               )
           )
           podman.succeed(su_cmd("podman ps | grep sleeping"))
           podman.succeed(su_cmd("podman stop sleeping"))
           podman.succeed(su_cmd("podman rm sleeping"))
+      # As of 2020-11-20, the runc backend doesn't work with cgroupsv2 yet, so we don't run that test.
 
-      with subtest("Run container rootless with crun"):
+      with subtest("Run container rootless with the default backend"):
           podman.succeed(su_cmd("tar cv --files-from /dev/null | podman import - scratchimg"))
           podman.succeed(
               su_cmd(
-                  "podman run --runtime=crun -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
+                  "podman run -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
               )
           )
           podman.succeed(su_cmd("podman ps | grep sleeping"))
diff --git a/nixos/tests/systemd.nix b/nixos/tests/systemd.nix
index dfa16eecfad2..390a1bd30f90 100644
--- a/nixos/tests/systemd.nix
+++ b/nixos/tests/systemd.nix
@@ -82,6 +82,10 @@ import ./make-test-python.nix ({ pkgs, ... }: {
             "systemd-run --pty --property=Type=oneshot --property=DynamicUser=yes --property=User=iamatest whoami"
         )
 
+    with subtest("regression test for https://bugs.freedesktop.org/show_bug.cgi?id=77507"):
+        retcode, output = machine.execute("systemctl status testservice1.service")
+        assert retcode in [0, 3]  # https://bugs.freedesktop.org/show_bug.cgi?id=77507
+
     # Regression test for https://github.com/NixOS/nixpkgs/issues/35268
     with subtest("file system with x-initrd.mount is not unmounted"):
         machine.succeed("mountpoint -q /test-x-initrd-mount")
@@ -122,17 +126,6 @@ import ./make-test-python.nix ({ pkgs, ... }: {
         machine.wait_for_unit("multi-user.target")
         assert "fq_codel" in machine.succeed("sysctl net.core.default_qdisc")
 
-    # Test cgroup accounting is enabled
-    with subtest("systemd cgroup accounting is enabled"):
-        machine.wait_for_unit("multi-user.target")
-        assert "yes" in machine.succeed(
-            "systemctl show testservice1.service -p IOAccounting"
-        )
-
-        retcode, output = machine.execute("systemctl status testservice1.service")
-        assert retcode in [0, 3]  # https://bugs.freedesktop.org/show_bug.cgi?id=77507
-        assert "CPU:" in output
-
     # Test systemd is configured to manage a watchdog
     with subtest("systemd manages hardware watchdog"):
         machine.wait_for_unit("multi-user.target")
@@ -168,5 +161,25 @@ import ./make-test-python.nix ({ pkgs, ... }: {
         machine.succeed("systemctl status systemd-cryptsetup@luks1.service")
         machine.succeed("mkdir -p /tmp/luks1")
         machine.succeed("mount /dev/mapper/luks1 /tmp/luks1")
+
+    # Do some IP traffic
+    output_ping = machine.succeed(
+        "systemd-run --wait -- /run/wrappers/bin/ping -c 1 127.0.0.1 2>&1"
+    )
+
+    with subtest("systemd reports accounting data on system.slice"):
+        output = machine.succeed("systemctl status system.slice")
+        assert "CPU:" in output
+        assert "Memory:" in output
+
+        assert "IP:" in output
+        assert "0B in, 0B out" not in output
+
+        assert "IO:" in output
+        assert "0B read, 0B written" not in output
+
+    with subtest("systemd per-unit accounting works"):
+        assert "IP traffic received: 84B" in output_ping
+        assert "IP traffic sent: 84B" in output_ping
   '';
 })