Merge pull request #14787 from giuseppe/move-systemd-service-to-subcgroup

service: do not run under the root cgroup

cmd/podman/system/service_abi.go

@@ -11,17 +11,39 @@ import (
 	"os"
 	"path/filepath"
 
+	"github.com/containers/common/pkg/cgroups"
 	"github.com/containers/podman/v4/cmd/podman/registry"
 	api "github.com/containers/podman/v4/pkg/api/server"
 	"github.com/containers/podman/v4/pkg/domain/entities"
 	"github.com/containers/podman/v4/pkg/domain/infra"
 	"github.com/containers/podman/v4/pkg/servicereaper"
+	"github.com/containers/podman/v4/utils"
 	"github.com/coreos/go-systemd/v22/activation"
 	"github.com/sirupsen/logrus"
 	"github.com/spf13/pflag"
 	"golang.org/x/sys/unix"
 )
 
+// maybeMoveToSubCgroup moves the current process in a sub cgroup when
+// it is running in the root cgroup on a system that uses cgroupv2.
+func maybeMoveToSubCgroup() error {
+	unifiedMode, err := cgroups.IsCgroup2UnifiedMode()
+	if err != nil {
+		return err
+	}
+	if !unifiedMode {
+		return nil
+	}
+	cgroup, err := utils.GetOwnCgroup()
+	if err != nil {
+		return err
+	}
+	if cgroup == "/" {
+		return utils.MoveUnderCgroupSubtree("init")
+	}
+	return nil
+}
+
 func restService(flags *pflag.FlagSet, cfg *entities.PodmanConfig, opts entities.ServiceOptions) error {
 	var (
 		listener net.Listener
@@ -103,6 +125,10 @@ func restService(flags *pflag.FlagSet, cfg *entities.PodmanConfig, opts entities
 		return err
 	}
 
+	if err := maybeMoveToSubCgroup(); err != nil {
+		return err
+	}
+
 	servicereaper.Start()
 	infra.StartWatcher(libpodRuntime)
 	server, err := api.NewServerWithSettings(libpodRuntime, listener, opts)

utils/utils_supported.go

@@ -137,7 +137,7 @@ func moveUnderCgroup(cgroup, subtree string, processes []uint32) error {
 		}
 
 		// root cgroup, skip it
-		if parts[2] == "/" {
+		if parts[2] == "/" && !(unifiedMode && parts[1] == "") {
 			continue
 		}