docker run: specify cgroup namespace mode with --cgroupns

Signed-off-by: Rob Gulewich <rgulewich@netflix.com>

cli/command/container/opts.go

@@ -70,6 +70,7 @@ type containerOptions struct {
 	pidMode            string
 	utsMode            string
 	usernsMode         string
+	cgroupnsMode       string
 	publishAll         bool
 	stdin              bool
 	tty                bool
@@ -198,6 +199,12 @@ func addFlags(flags *pflag.FlagSet) *containerOptions {
 	flags.BoolVar(&copts.privileged, "privileged", false, "Give extended privileges to this container")
 	flags.Var(&copts.securityOpt, "security-opt", "Security Options")
 	flags.StringVar(&copts.usernsMode, "userns", "", "User namespace to use")
+	flags.StringVar(&copts.cgroupnsMode, "cgroupns", "", `Cgroup namespace to use (host|private)
+'host':    Run the container in the Docker host's cgroup namespace
+'private': Run the container in its own private cgroup namespace
+'':        Use the cgroup namespace as configured by the
+           default-cgroupns-mode option on the daemon (default)`)
+	flags.SetAnnotation("cgroupns", "version", []string{"1.41"})
 
 	// Network and port publishing flag
 	flags.Var(&copts.extraHosts, "add-host", "Add a custom host-to-IP mapping (host:ip)")
@@ -469,6 +476,11 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 		return nil, errors.Errorf("--userns: invalid USER mode")
 	}
 
+	cgroupnsMode := container.CgroupnsMode(copts.cgroupnsMode)
+	if !cgroupnsMode.Valid() {
+		return nil, errors.Errorf("--cgroupns: invalid CGROUP mode")
+	}
+
 	restartPolicy, err := opts.ParseRestartPolicy(copts.restartPolicy)
 	if err != nil {
 		return nil, err
@@ -620,6 +632,7 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 		PidMode:        pidMode,
 		UTSMode:        utsMode,
 		UsernsMode:     usernsMode,
+		CgroupnsMode:   cgroupnsMode,
 		CapAdd:         strslice.StrSlice(copts.capAdd.GetAll()),
 		CapDrop:        strslice.StrSlice(copts.capDrop.GetAll()),
 		GroupAdd:       copts.groupAdd.GetAll(),

cli/command/system/info_test.go

@@ -65,6 +65,7 @@ var sampleInfoNoSwarm = types.Info{
 	NEventsListener:    0,
 	KernelVersion:      "4.4.0-87-generic",
 	OperatingSystem:    "Ubuntu 16.04.3 LTS",
+	OSVersion:          "",
 	OSType:             "linux",
 	Architecture:       "x86_64",
 	IndexServerAddress: "https://index.docker.io/v1/",

cli/compose/schema/data/config_schema_v3.9.json

@@ -96,6 +96,7 @@
         },
         "cap_add": {"type": "array", "items": {"type": "string"}, "uniqueItems": true},
         "cap_drop": {"type": "array", "items": {"type": "string"}, "uniqueItems": true},
+        "cgroupns_mode": {"type": "string"},
         "cgroup_parent": {"type": "string"},
         "command": {
           "oneOf": [

cli/compose/types/types.go

@@ -11,6 +11,7 @@ var UnsupportedProperties = []string{
 	"build",
 	"cap_add",
 	"cap_drop",
+	"cgroupns_mode",
 	"cgroup_parent",
 	"devices",
 	"domainname",
@@ -159,6 +160,7 @@ type ServiceConfig struct {
 	Build           BuildConfig                      `yaml:",omitempty" json:"build,omitempty"`
 	CapAdd          []string                         `mapstructure:"cap_add" yaml:"cap_add,omitempty" json:"cap_add,omitempty"`
 	CapDrop         []string                         `mapstructure:"cap_drop" yaml:"cap_drop,omitempty" json:"cap_drop,omitempty"`
+	CgroupNSMode    string                           `mapstructure:"cgroupns_mode" yaml:"cgroupns_mode,omitempty" json:"cgroupns_mode,omitempty"`
 	CgroupParent    string                           `mapstructure:"cgroup_parent" yaml:"cgroup_parent,omitempty" json:"cgroup_parent,omitempty"`
 	Command         ShellCommand                     `yaml:",omitempty" json:"command,omitempty"`
 	Configs         []ServiceConfigObjConfig         `yaml:",omitempty" json:"configs,omitempty"`

contrib/completion/bash/docker

@@ -1865,6 +1865,7 @@ _docker_container_run_and_create() {
 		--blkio-weight-device
 		--cap-add
 		--cap-drop
+		--cgroupns
 		--cgroup-parent
 		--cidfile
 		--cpu-period
@@ -2022,6 +2023,10 @@ _docker_container_run_and_create() {
 			_filedir
 			return
 			;;
+		--cgroupns)
+			COMPREPLY=( $( compgen -W "host private" -- "$cur" ) )
+			return
+			;;
 		--device|--tmpfs|--volume|-v)
 			case "$cur" in
 				*:*)

contrib/completion/fish/docker.fish

@@ -183,6 +183,7 @@ complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l blkio-weight
 complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l blkio-weight-device -d 'Block IO weight (relative device weight)'
 complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l cap-add -d 'Add Linux capabilities'
 complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l cap-drop -d 'Drop Linux capabilities'
+complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l cgroupns -d 'Cgroup namespace mode to use'
 complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l cgroup-parent -d 'Optional parent cgroup for the container'
 complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l cidfile -d 'Write the container ID to the file'
 complete -c docker -A -f -n '__fish_seen_subcommand_from create' -l cpu-count -d 'CPU count (Windows only)'
@@ -458,6 +459,7 @@ complete -c docker -A -f -n '__fish_seen_subcommand_from run' -s c -l cpu-shares
 complete -c docker -A -f -n '__fish_seen_subcommand_from run' -l cap-add -d 'Add Linux capabilities'
 complete -c docker -A -f -n '__fish_seen_subcommand_from run' -l cap-drop -d 'Drop Linux capabilities'
 complete -c docker -A -f -n '__fish_seen_subcommand_from run' -l cidfile -d 'Write the container ID to the file'
+complete -c docker -A -f -n '__fish_seen_subcommand_from run' -l cgroupns -d 'Cgroup namespace mode to use'
 complete -c docker -A -f -n '__fish_seen_subcommand_from run' -l cpuset -d 'CPUs in which to allow execution (0-3, 0,1)'
 complete -c docker -A -f -n '__fish_seen_subcommand_from run' -s d -l detach -d 'Detached mode: run the container in the background and print the new container ID'
 complete -c docker -A -f -n '__fish_seen_subcommand_from run' -l device -d 'Add a host device to the container (e.g. --device=/dev/sdc:/dev/xvdc:rwm)'

contrib/completion/zsh/_docker

@@ -605,6 +605,7 @@ __docker_container_subcommand() {
         "($help)*--blkio-weight-device=[Block IO (relative device weight)]:device:Block IO weight: "
         "($help)*--cap-add=[Add Linux capabilities]:capability: "
         "($help)*--cap-drop=[Drop Linux capabilities]:capability: "
+        "($help)--cgroupns=[Cgroup namespace mode to use]:cgroup namespace mode: "
         "($help)--cgroup-parent=[Parent cgroup for the container]:cgroup: "
         "($help)--cidfile=[Write the container ID to the file]:CID file:_files"
         "($help)--cpus=[Number of CPUs (default 0.000)]:cpus: "

docs/reference/commandline/create.md

@@ -29,6 +29,10 @@ Options:
       --blkio-weight-device value     Block IO weight (relative device weight) (default [])
       --cap-add value                 Add Linux capabilities (default [])
       --cap-drop value                Drop Linux capabilities (default [])
+      --cgroupns string               Cgroup namespace to use
+                                      'host':    Run the container in the Docker host's cgroup namespace
+                                      'private': Run the container in its own private cgroup namespace
+                                      '':        Use the default Docker daemon cgroup namespace specified by the `--default-cgroupns-mode` option
       --cgroup-parent string          Optional parent cgroup for the container
       --cidfile string                Write the container ID to the file
       --cpu-count int                 The number of CPUs available for execution by the container.

docs/reference/commandline/dockerd.md

@@ -38,6 +38,7 @@ Options:
       --cpu-rt-runtime int                    Limit the CPU real-time runtime in microseconds
       --data-root string                      Root directory of persistent Docker state (default "/var/lib/docker")
   -D, --debug                                 Enable debug mode
+      --default-cgroupns-mode string          Container default cgroup namespace mode (default "host")
       --default-gateway ip                    Container default gateway IPv4 address
       --default-gateway-v6 ip                 Container default gateway IPv6 address
       --default-address-pool                  Set the default address pool for local node networks
@@ -1299,6 +1300,7 @@ This is a full example of the allowed configuration options on Linux:
 {
 	"authorization-plugins": [],
 	"data-root": "",
+	"default-cgroupns-mode": "private",
 	"dns": [],
 	"dns-opts": [],
 	"dns-search": [],

docs/reference/commandline/run.md

@@ -27,6 +27,10 @@ Options:
       --blkio-weight-device value     Block IO weight (relative device weight) (default [])
       --cap-add value                 Add Linux capabilities (default [])
       --cap-drop value                Drop Linux capabilities (default [])
+      --cgroupns string               Cgroup namespace to use
+                                      'host':    Run the container in the Docker host's cgroup namespace
+                                      'private': Run the container in its own private cgroup namespace
+                                      '':        Use the default Docker daemon cgroup namespace specified by the `--default-cgroupns-mode` option
       --cgroup-parent string          Optional parent cgroup for the container
       --cidfile string                Write the container ID to the file
       --cpu-count int                 The number of CPUs available for execution by the container.

e2e/container/run_test.go

@@ -135,3 +135,12 @@ func createRemoteImage(t *testing.T) string {
 	icmd.RunCommand("docker", "rmi", image).Assert(t, icmd.Success)
 	return image
 }
+
+func TestRunWithCgroupNamespace(t *testing.T) {
+	environment.SkipIfDaemonNotLinux(t)
+	environment.SkipIfCgroupNamespacesNotSupported(t)
+
+	result := icmd.RunCommand("docker", "run", "--cgroupns=private", "--rm", fixtures.AlpineImage,
+		"/bin/grep", "-q", "':memory:/$'", "/proc/1/cgroup")
+	result.Assert(t, icmd.Success)
+}

internal/test/environment/testenv.go

@@ -1,6 +1,7 @@
 package environment
 
 import (
+	"fmt"
 	"os"
 	"strings"
 	"testing"
@@ -88,3 +89,22 @@ func SkipIfNotExperimentalDaemon(t *testing.T) {
 	experimentalBuild := strings.TrimSpace(result.Stdout()) == "true"
 	skip.If(t, !experimentalBuild, "running against a non-experimental daemon")
 }
+
+// SkipIfDaemonNotLinux skips the test unless the running docker daemon is on Linux
+func SkipIfDaemonNotLinux(t *testing.T) {
+	t.Helper()
+	result := icmd.RunCmd(icmd.Command("docker", "info", "--format", "{{.OSType}}"))
+	result.Assert(t, icmd.Expected{Err: icmd.None})
+	isLinux := strings.TrimSpace(result.Stdout()) == "linux"
+	skip.If(t, !isLinux, "running against a Linux daemon")
+}
+
+// SkipIfCgroupNamespacesNotSupported skips the test if the running docker daemon doesn't support cgroup namespaces
+func SkipIfCgroupNamespacesNotSupported(t *testing.T) {
+	t.Helper()
+	result := icmd.RunCmd(icmd.Command("docker", "info", "--format", "{{.SecurityOptions}}"))
+	result.Assert(t, icmd.Expected{Err: icmd.None})
+	cgroupNsFound := strings.Contains(result.Stdout(), "name=cgroupns")
+
+	skip.If(t, !cgroupNsFound, fmt.Sprintf("running against a daemon that doesn't support cgroup namespaces (security options: %s)", result.Stdout()))
+}

man/docker-run.1.md

@@ -13,6 +13,7 @@ docker-run - Run a command in a new container
 [**--cpu-shares**[=*0*]]
 [**--cap-add**[=*[]*]]
 [**--cap-drop**[=*[]*]]
+[**--cgroupns**[=*[]*]]
 [**--cgroup-parent**[=*CGROUP-PATH*]]
 [**--cidfile**[=*CIDFILE*]]
 [**--cpu-count**[=*0*]]
@@ -173,6 +174,12 @@ division of CPU shares:
 **--cap-drop**=[]
    Drop Linux capabilities
 
+**--cgroupns**=""
+   Set the cgroup namespace mode for the container.
+     **host**:    run the container in the host's cgroup namespace
+     **private**: run the container in its own private cgroup namespace
+     **""**:      (unset) run the container in the host's cgroup namespace
+
 **--cgroup-parent**=""
    Path to cgroups under which the cgroup for the container will be created. If the path is not absolute, the path is considered to be relative to the cgroups path of the init process. Cgroups will be created if they do not already exist.
 

man/dockerd.8.md

@@ -20,6 +20,7 @@ dockerd - Enable daemon mode
 [**--containerd**[=*SOCKET-PATH*]]
 [**--data-root**[=*/var/lib/docker*]]
 [**-D**|**--debug**]
+[**--default-cgroupns-mode**[=*host*]]
 [**--default-gateway**[=*DEFAULT-GATEWAY*]]
 [**--default-gateway-v6**[=*DEFAULT-GATEWAY-V6*]]
 [**--default-address-pool**[=*DEFAULT-ADDRESS-POOL*]]
@@ -178,6 +179,10 @@ $ sudo dockerd --add-runtime runc=runc --add-runtime custom=/usr/local/bin/my-ru
 **-D**, **--debug**=*true*|*false*
   Enable debug mode. Default is false.
 
+**--default-cgroupns-mode**="**host**|**private**"
+  Set the default cgroup namespace mode for newly created containers. The argument
+  can either be **host** or **private**. If unset, this defaults to `host`.
+
 **--default-gateway**=""
   IPv4 address of the container default gateway; this address must be part of
   the bridge subnet (which is defined by \-b or \--bip)