Support Rootless Docker

Requirements:
- Install rootless Docker 20.10 or later, see https://rootlesscontaine.rs/getting-started/docker/
- Enable cgroup v2 delegation, see https://rootlesscontaine.rs/getting-started/common/cgroup2/

Usage: `minikube start --driver=docker --container-runtime=containerd`.
The `--container-runtime` flag needs to be set to "containerd".
CRI-O can be also supported later.

Closes issue 10836 ("add support for rootless Docker").

Support for rootless Podman (issue 8719) is not covered in this commit.

---

Code reading guide:
- `deploy/kicbase/Dockerfile`: updated to install fuse-overlayfs and containerd-fuse-overlayfs, which is used
  instead of `overlayfs` snapshotter

- `deploy/kicbase/entrypoint`: updated to verify cgroup v2 delegation.
  Mostly from https://github.com/kubernetes-sigs/kind/blob/8a83ee46b28a80ccd47a85e24294b3e149361947/images/base/files/usr/local/bin/entrypoint

- `cmd/minikube/cmd/start_flags.go`: updated to set `KubeletInUserNamespace` feature gate when rootless

- `pkg/drivers/kic/oci`: updated to use port forwarding, because rootless container IPs are not reachable from the host

- `pkg/minikube/cruntime`: updated to generate `/etc/containerd/config.toml` with rootless support.

Signed-off-by: Akihiro Suda <[email protected]>

cmd/minikube/cmd/start_flags.go

@@ -492,9 +492,37 @@ func generateNewConfigFromFlags(cmd *cobra.Command, k8sVersion string, drvName s
 		}
 	}
 
+	if driver.IsKIC(drvName) {
+		si, err := oci.CachedDaemonInfo(drvName)
+		if err != nil {
+			exit.Message(reason.Usage, "Ensure your {{.driver_name}} is running and is healthy.", out.V{"driver_name": driver.FullName(drvName)})
+		}
+		if si.Rootless {
+			if cc.KubernetesConfig.ContainerRuntime != "containerd" {
+				exit.Message(reason.Usage, "Container runtime must be set to \"containerd\" for rootless")
+				// TODO: support cri-o (https://kubernetes.io/docs/tasks/administer-cluster/kubelet-in-userns/#configuring-cri)
+			}
+			// KubeletInUserNamespace feature gate is essential for rootless driver.
+			// See https://kubernetes.io/docs/tasks/administer-cluster/kubelet-in-userns/
+			cc.KubernetesConfig.FeatureGates = addFeatureGate(cc.KubernetesConfig.FeatureGates, "KubeletInUserNamespace=true")
+		}
+	}
+
 	return cc
 }
 
+func addFeatureGate(featureGates, s string) string {
+	split := strings.Split(featureGates, ",")
+	m := make(map[string]struct{}, len(split))
+	for _, v := range split {
+		m[v] = struct{}{}
+	}
+	if _, ok := m[s]; !ok {
+		split = append(split, s)
+	}
+	return strings.Join(split, ",")
+}
+
 func checkNumaCount(k8sVersion string) {
 	if viper.GetInt(kvmNUMACount) < 1 || viper.GetInt(kvmNUMACount) > 8 {
 		exit.Message(reason.Usage, "--kvm-numa-count range is 1-8")

deploy/kicbase/Dockerfile

@@ -30,6 +30,8 @@ RUN cd ./cmd/auto-pause/ && go build
 FROM ubuntu:focal-20210401
 
 ARG BUILDKIT_VERSION="v0.9.0"
+ARG FUSE_OVERLAYFS_VERSION="v1.7.1"
+ARG CONTAINERD_FUSE_OVERLAYFS_VERSION="1.0.3"
 
 # copy in static files (configs, scripts)
 COPY deploy/kicbase/10-network-security.conf /etc/sysctl.d/10-network-security.conf
@@ -113,14 +115,29 @@ RUN clean-install \
     openssh-server \
     dnsutils \
     # libglib2.0-0 is required for conmon, which is required for podman
-    libglib2.0-0
+    libglib2.0-0 \
+    # fuse3 is required for fuse-overlayfs
+    fuse3
 
 # install docker
 RUN sh -c "echo 'deb https://download.docker.com/linux/ubuntu focal stable' > /etc/apt/sources.list.d/docker.list" && \
     curl -L https://download.docker.com/linux/ubuntu/gpg -o docker.key && \
     apt-key add - < docker.key && \
     clean-install docker-ce docker-ce-cli containerd.io
 
+# install fuse-overlayfs (used by rootless; apt-get version is old)
+RUN curl -sSL --retry 5 --output /usr/local/bin/fuse-overlayfs https://github.com/containers/fuse-overlayfs/releases/download/${FUSE_OVERLAYFS_VERSION}/fuse-overlayfs-$(uname -m) \
+  && chmod +x  /usr/local/bin/fuse-overlayfs
+
+# install containerd-fuse-overlayfs (used by rootless)
+RUN export ARCH=$(dpkg --print-architecture | sed 's/ppc64el/ppc64le/' | sed 's/armhf/arm-v7/') \
+ && echo "Installing containerd-fuse-overlayfs..." \
+    && export CONTAINERD_FUSE_OVERLAYFS_BASE_URL="https://github.com/containerd/fuse-overlayfs-snapshotter/releases/download/v${CONTAINERD_FUSE_OVERLAYFS_VERSION}" \
+    && curl -sSL --retry 5 --output /tmp/containerd-fuse-overlayfs.tgz "${CONTAINERD_FUSE_OVERLAYFS_BASE_URL}/containerd-fuse-overlayfs-${CONTAINERD_FUSE_OVERLAYFS_VERSION}-linux-${ARCH}.tar.gz" \
+    && tar -C /usr/local/bin -xzvf /tmp/containerd-fuse-overlayfs.tgz \
+    && rm -rf /tmp/containerd-fuse-overlayfs.tgz
+COPY deploy/kicbase/containerd-fuse-overlayfs.service /etc/systemd/system/containerd-fuse-overlayfs.service
+
 # install buildkit
 RUN export ARCH=$(dpkg --print-architecture | sed 's/ppc64el/ppc64le/' | sed 's/armhf/arm-v7/') \
  && echo "Installing buildkit ..." \

deploy/kicbase/containerd-fuse-overlayfs.service

@@ -0,0 +1,13 @@
+# From https://github.com/kubernetes-sigs/kind/blob/0d3780371091b2dc9ff6eea1b6054f14ff5d970a/images/base/files/etc/systemd/system/containerd-fuse-overlayfs.service
+[Unit]
+Description=containerd fuse-overlayfs snapshotter
+PartOf=containerd.service
+
+[Service]
+ExecStart=/usr/local/bin/containerd-fuse-overlayfs-grpc /run/containerd-fuse-overlayfs.sock /var/lib/containerd-fuse-overlayfs
+Type=notify
+Restart=always
+RestartSec=1
+
+[Install]
+WantedBy=multi-user.target

deploy/kicbase/entrypoint

@@ -19,6 +19,39 @@ set -o nounset
 set -o pipefail
 set -x
 
+# If /proc/self/uid_map 4294967295 mappings, we are in the initial user namespace, i.e. the host.
+# Otherwise we are in a non-initial user namespace.
+# https://github.com/opencontainers/runc/blob/v1.0.0-rc92/libcontainer/system/linux.go#L109-L118
+userns=""
+if grep -Eqv "0[[:space:]]+0[[:space:]]+4294967295" /proc/self/uid_map; then
+  userns="1"
+  echo 'INFO: running in a user namespace (experimental)'
+fi
+
+validate_userns() {
+  if [[ -z "${userns}" ]]; then
+    return
+  fi
+
+  local nofile_hard
+  nofile_hard="$(ulimit -Hn)"
+  local nofile_hard_expected="64000"
+  if [[ "${nofile_hard}" -lt "${nofile_hard_expected}" ]]; then
+    echo "WARN: UserNS: expected RLIMIT_NOFILE to be at least ${nofile_hard_expected}, got ${nofile_hard}" >&2
+  fi
+
+  if [[ ! -f "/sys/fs/cgroup/cgroup.controllers" ]]; then
+    echo "ERROR: UserNS: cgroup v2 needs to be enabled, see https://rootlesscontaine.rs/getting-started/common/cgroup2/" >&2
+    exit 1
+  fi
+  for f in cpu memory pids; do
+    if ! grep -qw $f /sys/fs/cgroup/cgroup.controllers; then
+      echo "ERROR: UserNS: $f controller needs to be delegated, see https://rootlesscontaine.rs/getting-started/common/cgroup2/" >&2
+    exit 1
+    fi
+  done
+}
+
 configure_containerd() {
   # we need to switch to the 'native' snapshotter on zfs
   if [[ "$(stat -f -c %T /kind)" == 'zfs' ]]; then
@@ -73,12 +106,16 @@ fix_mount() {
     sync
   fi
 
-  echo 'INFO: remounting /sys read-only'
-  # systemd-in-a-container should have read only /sys
-  # https://systemd.io/CONTAINER_INTERFACE/
-  # however, we need other things from `docker run --privileged` ...
-  # and this flag also happens to make /sys rw, amongst other things
-  mount -o remount,ro /sys
+  if [[ -z "${userns}" ]]; then
+    echo 'INFO: remounting /sys read-only'
+    # systemd-in-a-container should have read only /sys
+    # https://systemd.io/CONTAINER_INTERFACE/
+    # however, we need other things from `docker run --privileged` ...
+    # and this flag also happens to make /sys rw, amongst other things
+    #
+    # This step is skipped when running inside UserNS, because it fails with EACCES.
+    mount -o remount,ro /sys
+  fi
 
   echo 'INFO: making mounts shared' >&2
   # for mount propagation
@@ -334,6 +371,9 @@ enable_network_magic(){
   fi
 }
 
+# validate state
+validate_userns
+
 # run pre-init fixups
 # NOTE: it's important that we do configure* first in this order to avoid races
 configure_containerd

hack/preload-images/generate.go

@@ -93,7 +93,7 @@ func generateTarball(kubernetesVersion, containerRuntime, tarballFilename string
 	if err != nil {
 		return errors.Wrap(err, "failed create new runtime")
 	}
-	if err := cr.Enable(true, false); err != nil {
+	if err := cr.Enable(true, false, false); err != nil {
 		return errors.Wrap(err, "enable container runtime")
 	}
 

pkg/drivers/kic/oci/network.go

@@ -34,6 +34,33 @@ import (
 // RoutableHostIPFromInside returns the ip/dns of the host that container lives on
 // is routable from inside the container
 func RoutableHostIPFromInside(ociBin string, clusterName string, containerName string) (net.IP, error) {
+	si, err := CachedDaemonInfo(ociBin)
+	if err != nil {
+		return nil, err
+	}
+	if si.Rootless {
+		if IsExternalDaemonHost(ociBin) {
+			return nil, fmt.Errorf("function RoutableHostIPFromInside is not implemented for external rootless daemons")
+			// TODO: parse DaemonHost()
+		}
+		addrs, err := net.InterfaceAddrs()
+		if err != nil {
+			return nil, err
+		}
+		for _, addr := range addrs {
+			var ip net.IP
+			switch v := addr.(type) {
+			case *net.IPAddr:
+				ip = v.IP
+			case *net.IPNet:
+				ip = v.IP
+			}
+			if ip != nil && !ip.IsLoopback() {
+				return ip, nil
+			}
+		}
+		return nil, fmt.Errorf("could not detect host IP, tried %v", addrs)
+	}
 	if ociBin == Docker {
 		if runtime.GOOS == "linux" {
 			info, err := containerNetworkInspect(ociBin, clusterName)

pkg/drivers/kic/oci/oci.go

@@ -162,6 +162,9 @@ func CreateContainerNode(p CreateParams) error {
 		// including some ones docker would otherwise do by default.
 		// for now this is what we want. in the future we may revisit this.
 		"--privileged",
+		// enable /dev/fuse explicitly for fuse-overlayfs
+		// (Rootless Docker does not automatically mount /dev/fuse with --privileged)
+		"--device", "/dev/fuse",
 		"--security-opt", "seccomp=unconfined", //  ignore seccomp
 		"--tmpfs", "/tmp", // various things depend on working /tmp
 		"--tmpfs", "/run", // systemd wants a writable /run

pkg/minikube/cruntime/containerd.go

@@ -69,6 +69,12 @@ oom_score = 0
 [cgroup]
   path = ""
 
+[proxy_plugins]
+# fuse-overlayfs is used for rootless
+[proxy_plugins."fuse-overlayfs"]
+  type = "snapshot"
+  address = "/run/containerd-fuse-overlayfs.sock"
+
 [plugins]
   [plugins.cgroups]
     no_prometheus = false
@@ -80,6 +86,7 @@ oom_score = 0
     stats_collect_period = 10
     enable_tls_streaming = false
     max_container_log_line_size = 16384
+    restrict_oom_score_adj = {{ .RestrictOOMScoreAdj }}
 
 	[plugins."io.containerd.grpc.v1.cri"]
       [plugins."io.containerd.grpc.v1.cri".containerd]
@@ -90,7 +97,7 @@ oom_score = 0
               SystemdCgroup = {{ .SystemdCgroup }}
 
     [plugins.cri.containerd]
-      snapshotter = "overlayfs"
+      snapshotter = "{{ .Snapshotter }}"
       [plugins.cri.containerd.default_runtime]
         runtime_type = "io.containerd.runc.v2"
       [plugins.cri.containerd.untrusted_workload_runtime]
@@ -193,23 +200,31 @@ func (r *Containerd) Available() error {
 }
 
 // generateContainerdConfig sets up /etc/containerd/config.toml
-func generateContainerdConfig(cr CommandRunner, imageRepository string, kv semver.Version, forceSystemd bool, insecureRegistry []string) error {
+func generateContainerdConfig(cr CommandRunner, imageRepository string, kv semver.Version, forceSystemd bool, insecureRegistry []string, inUserNamespace bool) error {
 	cPath := containerdConfigFile
 	t, err := template.New("containerd.config.toml").Parse(containerdConfigTemplate)
 	if err != nil {
 		return err
 	}
 	pauseImage := images.Pause(kv, imageRepository)
+	snapshotter := "overlayfs"
+	if inUserNamespace {
+		snapshotter = "fuse-overlayfs"
+	}
 	opts := struct {
 		PodInfraContainerImage string
 		SystemdCgroup          bool
 		InsecureRegistry       []string
 		CNIConfDir             string
+		RestrictOOMScoreAdj    bool
+		Snapshotter            string
 	}{
 		PodInfraContainerImage: pauseImage,
 		SystemdCgroup:          forceSystemd,
 		InsecureRegistry:       insecureRegistry,
 		CNIConfDir:             cni.ConfDir,
+		RestrictOOMScoreAdj:    inUserNamespace,
+		Snapshotter:            snapshotter,
 	}
 	var b bytes.Buffer
 	if err := t.Execute(&b, opts); err != nil {
@@ -223,7 +238,7 @@ func generateContainerdConfig(cr CommandRunner, imageRepository string, kv semve
 }
 
 // Enable idempotently enables containerd on a host
-func (r *Containerd) Enable(disOthers, forceSystemd bool) error {
+func (r *Containerd) Enable(disOthers, forceSystemd, inUserNamespace bool) error {
 	if disOthers {
 		if err := disableOthers(r, r.Runner); err != nil {
 			klog.Warningf("disableOthers: %v", err)
@@ -232,13 +247,19 @@ func (r *Containerd) Enable(disOthers, forceSystemd bool) error {
 	if err := populateCRIConfig(r.Runner, r.SocketPath()); err != nil {
 		return err
 	}
-	if err := generateContainerdConfig(r.Runner, r.ImageRepository, r.KubernetesVersion, forceSystemd, r.InsecureRegistry); err != nil {
+	if err := generateContainerdConfig(r.Runner, r.ImageRepository, r.KubernetesVersion, forceSystemd, r.InsecureRegistry, inUserNamespace); err != nil {
 		return err
 	}
 	if err := enableIPForwarding(r.Runner); err != nil {
 		return err
 	}
 
+	if inUserNamespace {
+		if err := r.Init.EnableNow("containerd-fuse-overlayfs"); err != nil {
+			return err
+		}
+	}
+
 	// Otherwise, containerd will fail API requests with 'Unimplemented'
 	return r.Init.Restart("containerd")
 }

pkg/minikube/cruntime/crio.go

@@ -139,7 +139,10 @@ func enableIPForwarding(cr CommandRunner) error {
 }
 
 // Enable idempotently enables CRIO on a host
-func (r *CRIO) Enable(disOthers, _ bool) error {
+func (r *CRIO) Enable(disOthers, _, inUserNamespace bool) error {
+	if inUserNamespace {
+		return errors.New("inUserNamespace must not be true for cri-o (yet)")
+	}
 	if disOthers {
 		if err := disableOthers(r, r.Runner); err != nil {
 			klog.Warningf("disableOthers: %v", err)

pkg/minikube/cruntime/cruntime.go

@@ -78,7 +78,7 @@ type Manager interface {
 	// Version retrieves the current version of this runtime
 	Version() (string, error)
 	// Enable idempotently enables this runtime on a host
-	Enable(bool, bool) error
+	Enable(bool, bool, bool) error
 	// Disable idempotently disables this runtime on a host
 	Disable() error
 	// Active returns whether or not a runtime is active on a host

pkg/minikube/cruntime/cruntime_test.go

@@ -668,7 +668,7 @@ func TestEnable(t *testing.T) {
 			if err != nil {
 				t.Fatalf("New(%s): %v", tc.runtime, err)
 			}
-			err = cr.Enable(true, false)
+			err = cr.Enable(true, false, false)
 			if err != nil {
 				t.Errorf("%s disable unexpected error: %v", tc.runtime, err)
 			}

pkg/minikube/cruntime/docker.go

@@ -107,7 +107,10 @@ func (r *Docker) Active() bool {
 }
 
 // Enable idempotently enables Docker on a host
-func (r *Docker) Enable(disOthers, forceSystemd bool) error {
+func (r *Docker) Enable(disOthers, forceSystemd, inUserNamespace bool) error {
+	if inUserNamespace {
+		return errors.New("inUserNamespace must not be true for docker")
+	}
 	containerdWasActive := r.Init.Active("containerd")
 
 	if disOthers {

pkg/minikube/driver/driver.go

@@ -185,7 +185,15 @@ func NeedsPortForward(name string) bool {
 		return true
 	}
 	// Docker for Desktop
-	return runtime.GOOS == "darwin" || runtime.GOOS == "windows" || detect.IsMicrosoftWSL()
+	if runtime.GOOS == "darwin" || runtime.GOOS == "windows" || detect.IsMicrosoftWSL() {
+		return true
+	}
+
+	si, err := oci.CachedDaemonInfo(name)
+	if err != nil {
+		panic(err)
+	}
+	return si.Rootless
 }
 
 // HasResourceLimits returns true if driver can set resource limits such as memory size or CPU count.

pkg/minikube/node/start.go

@@ -349,7 +349,8 @@ func configureRuntimes(runner cruntime.CommandRunner, cc config.ClusterConfig, k
 		}
 	}
 
-	err = cr.Enable(disableOthers, forceSystemd())
+	inUserNamespace := strings.Contains(cc.KubernetesConfig.FeatureGates, "KubeletInUserNamespace=true")
+	err = cr.Enable(disableOthers, forceSystemd(), inUserNamespace)
 	if err != nil {
 		exit.Error(reason.RuntimeEnable, "Failed to enable container runtime", err)
 	}

pkg/minikube/registry/drvs/docker/docker.go

@@ -146,14 +146,7 @@ func status() (retState registry.State) {
 		return suggestFix("info", -1, serr, fmt.Errorf("docker info error: %s", serr))
 	}
 
-	if si.Rootless {
-		return registry.State{
-			Reason:    "PROVIDER_DOCKER_ROOTLESS",
-			Error:     errors.New("rootless Docker not supported yet"),
-			Installed: true,
-			Healthy:   false,
-			Doc:       "https://github.com/kubernetes/minikube/issues/10836"}
-	}
+	// TODO: validate cgroup v2 delegation when si.Rootless is true
 
 	return checkNeedsImprovement()
 }