libpod: support for cgroup namespace

allow a container to run in a new cgroup namespace.

When running in a new cgroup namespace, the current cgroup appears to
be the root, so that there is no way for the container to access
cgroups outside of its own subtree.

By default it uses --cgroup=host to keep the previous behavior.

To create a new namespace, --cgroup=private must be provided.

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

9 files changed, 140 insertions, 5 deletions

diff --git a/cmd/podman/common.go b/cmd/podman/common.go
index 15f753d55..1e9092bd6 100644
--- a/cmd/podman/common.go
+++ b/cmd/podman/common.go
@@ -130,6 +130,10 @@ func getCreateFlags(c *cliconfig.PodmanCommand) {
 		"Drop capabilities from the container",
 	)
 	createFlags.String(
+		"cgroupns", "host",
+		"cgroup namespace to use",
+	)
+	createFlags.String(
 		"cgroup-parent", "",
 		"Optional parent cgroup for the container",
 	)
diff --git a/cmd/podman/shared/create.go b/cmd/podman/shared/create.go
index 5c37a1875..b14ce431d 100644
--- a/cmd/podman/shared/create.go
+++ b/cmd/podman/shared/create.go
@@ -400,11 +400,12 @@ func ParseCreateOpts(ctx context.Context, c *GenericCLIResults, runtime *libpod.
 		namespaceNet = c.String("net")
 	}
 	namespaces = map[string]string{
-		"pid":  c.String("pid"),
-		"net":  namespaceNet,
-		"ipc":  c.String("ipc"),
-		"user": c.String("userns"),
-		"uts":  c.String("uts"),
+		"cgroup": c.String("cgroupns"),
+		"pid":    c.String("pid"),
+		"net":    namespaceNet,
+		"ipc":    c.String("ipc"),
+		"user":   c.String("userns"),
+		"uts":    c.String("uts"),
 	}
 
 	originalPodName := c.String("pod")
@@ -462,6 +463,11 @@ func ParseCreateOpts(ctx context.Context, c *GenericCLIResults, runtime *libpod.
 		return nil, errors.Errorf("--uts %q is not valid", namespaces["uts"])
 	}
 
+	cgroupMode := ns.CgroupMode(namespaces["cgroup"])
+	if !cgroupMode.Valid() {
+		return nil, errors.Errorf("--cgroup %q is not valid", namespaces["cgroup"])
+	}
+
 	ipcMode := ns.IpcMode(namespaces["ipc"])
 	if !cc.Valid(string(ipcMode), ipcMode) {
 		return nil, errors.Errorf("--ipc %q is not valid", ipcMode)
@@ -652,6 +658,7 @@ func ParseCreateOpts(ctx context.Context, c *GenericCLIResults, runtime *libpod.
 		CapAdd:            c.StringSlice("cap-add"),
 		CapDrop:           c.StringSlice("cap-drop"),
 		CidFile:           c.String("cidfile"),
+		Cgroupns:          c.String("cgroupns"),
 		CgroupParent:      c.String("cgroup-parent"),
 		Command:           command,
 		Detach:            c.Bool("detach"),
@@ -687,6 +694,7 @@ func ParseCreateOpts(ctx context.Context, c *GenericCLIResults, runtime *libpod.
 		NetMode:        netMode,
 		UtsMode:        utsMode,
 		PidMode:        pidMode,
+		CgroupMode:     cgroupMode,
 		Pod:            podName,
 		Privileged:     c.Bool("privileged"),
 		Publish:        c.StringSlice("publish"),
diff --git a/cmd/podman/shared/intermediate.go b/cmd/podman/shared/intermediate.go
index 8337dc647..4062ac48a 100644
--- a/cmd/podman/shared/intermediate.go
+++ b/cmd/podman/shared/intermediate.go
@@ -370,6 +370,7 @@ func NewIntermediateLayer(c *cliconfig.PodmanCommand, remote bool) GenericCLIRes
 	m["blkio-weight-device"] = newCRStringSlice(c, "blkio-weight-device")
 	m["cap-add"] = newCRStringSlice(c, "cap-add")
 	m["cap-drop"] = newCRStringSlice(c, "cap-drop")
+	m["cgroupns"] = newCRString(c, "cgroupns")
 	m["cgroup-parent"] = newCRString(c, "cgroup-parent")
 	m["cidfile"] = newCRString(c, "cidfile")
 	m["conmon-pidfile"] = newCRString(c, "conmon-pidfile")
diff --git a/docs/podman-create.1.md b/docs/podman-create.1.md
index 89f146670..24a78ab93 100644
--- a/docs/podman-create.1.md
+++ b/docs/podman-create.1.md
@@ -63,6 +63,14 @@ Add Linux capabilities
 
 Drop Linux capabilities
 
+**--cgroupns**=*mode*
+
+Set the cgroup namespace mode for the container, by default **host** is used.
+    **host**: use the host's cgroup namespace inside the container.
+    **container:<NAME|ID>**: join the namespace of the specified container.
+    **private**: create a new cgroup namespace.
+    **ns:<PATH>**: join the namespace at the specified path.
+
 **--cgroup-parent**=*path*
 
 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.
diff --git a/docs/podman-run.1.md b/docs/podman-run.1.md
index ebf774b24..6315afea6 100644
--- a/docs/podman-run.1.md
+++ b/docs/podman-run.1.md
@@ -77,6 +77,14 @@ Add Linux capabilities
 
 Drop Linux capabilities
 
+**--cgroupns**=*mode*
+
+Set the cgroup namespace mode for the container, by default **host** is used.
+    **host**: use the host's cgroup namespace inside the container.
+    **container:<NAME|ID>**: join the namespace of the specified container.
+    **private**: create a new cgroup namespace.
+    **ns:<PATH>**: join the namespace at the specified path.
+
 **--cgroup-parent**=*cgroup*
 
 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.
diff --git a/pkg/namespaces/namespaces.go b/pkg/namespaces/namespaces.go
index ec9276344..7ed95bd0f 100644
--- a/pkg/namespaces/namespaces.go
+++ b/pkg/namespaces/namespaces.go
@@ -4,6 +4,63 @@ import (
 	"strings"
 )
 
+// CgroupMode represents cgroup mode in the container.
+type CgroupMode string
+
+// IsHost indicates whether the container uses the host's cgroup.
+func (n CgroupMode) IsHost() bool {
+	return n == "host"
+}
+
+// IsNS indicates a cgroup namespace passed in by path (ns:<path>)
+func (n CgroupMode) IsNS() bool {
+	return strings.HasPrefix(string(n), "ns:")
+}
+
+// NS gets the path associated with a ns:<path> cgroup ns
+func (n CgroupMode) NS() string {
+	parts := strings.SplitN(string(n), ":", 2)
+	if len(parts) > 1 {
+		return parts[1]
+	}
+	return ""
+}
+
+// IsContainer indicates whether the container uses a new cgroup namespace.
+func (n CgroupMode) IsContainer() bool {
+	parts := strings.SplitN(string(n), ":", 2)
+	return len(parts) > 1 && parts[0] == "container"
+}
+
+// Container returns the name of the container whose cgroup namespace is going to be used.
+func (n CgroupMode) Container() string {
+	parts := strings.SplitN(string(n), ":", 2)
+	if len(parts) > 1 {
+		return parts[1]
+	}
+	return ""
+}
+
+// IsPrivate indicates whether the container uses the a private cgroup.
+func (n CgroupMode) IsPrivate() bool {
+	return n == "private"
+}
+
+// Valid indicates whether the Cgroup namespace is valid.
+func (n CgroupMode) Valid() bool {
+	parts := strings.Split(string(n), ":")
+	switch mode := parts[0]; mode {
+	case "", "host", "private", "ns":
+	case "container":
+		if len(parts) != 2 || parts[1] == "" {
+			return false
+		}
+	default:
+		return false
+	}
+	return true
+}
+
 // UsernsMode represents userns mode in the container.
 type UsernsMode string
 
diff --git a/pkg/spec/createconfig.go b/pkg/spec/createconfig.go
index 0042ed401..1fb1f829b 100644
--- a/pkg/spec/createconfig.go
+++ b/pkg/spec/createconfig.go
@@ -63,6 +63,7 @@ type CreateConfig struct {
 	CapDrop            []string // cap-drop
 	CidFile            string
 	ConmonPidFile      string
+	Cgroupns           string
 	CgroupParent       string // cgroup-parent
 	Command            []string
 	Detach             bool              // detach
@@ -101,6 +102,7 @@ type CreateConfig struct {
 	NetworkAlias       []string               //network-alias
 	PidMode            namespaces.PidMode     //pid
 	Pod                string                 //pod
+	CgroupMode         namespaces.CgroupMode  //cgroup
 	PortBindings       nat.PortMap
 	Privileged         bool     //privileged
 	Publish            []string //publish
@@ -268,6 +270,23 @@ func (c *CreateConfig) getContainerCreateOptions(runtime *libpod.Runtime, pod *l
 		options = append(options, libpod.WithNetNS(portBindings, postConfigureNetNS, string(c.NetMode), networks))
 	}
 
+	if c.CgroupMode.IsNS() {
+		ns := c.CgroupMode.NS()
+		if ns == "" {
+			return nil, errors.Errorf("invalid empty user-defined network namespace")
+		}
+		_, err := os.Stat(ns)
+		if err != nil {
+			return nil, err
+		}
+	} else if c.CgroupMode.IsContainer() {
+		connectedCtr, err := runtime.LookupContainer(c.CgroupMode.Container())
+		if err != nil {
+			return nil, errors.Wrapf(err, "container %q not found", c.CgroupMode.Container())
+		}
+		options = append(options, libpod.WithCgroupNSFrom(connectedCtr))
+	}
+
 	if c.PidMode.IsContainer() {
 		connectedCtr, err := runtime.LookupContainer(c.PidMode.Container())
 		if err != nil {
diff --git a/pkg/spec/spec.go b/pkg/spec/spec.go
index 41054633f..a8ab4911a 100644
--- a/pkg/spec/spec.go
+++ b/pkg/spec/spec.go
@@ -325,6 +325,10 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM
 	if err := addIpcNS(config, &g); err != nil {
 		return nil, err
 	}
+
+	if err := addCgroupNS(config, &g); err != nil {
+		return nil, err
+	}
 	configSpec := g.Config
 
 	// HANDLE CAPABILITIES
@@ -622,6 +626,23 @@ func addIpcNS(config *CreateConfig, g *generate.Generator) error {
 	return nil
 }
 
+func addCgroupNS(config *CreateConfig, g *generate.Generator) error {
+	cgroupMode := config.CgroupMode
+	if cgroupMode.IsNS() {
+		return g.AddOrReplaceLinuxNamespace(string(spec.CgroupNamespace), NS(string(cgroupMode)))
+	}
+	if cgroupMode.IsHost() {
+		return g.RemoveLinuxNamespace(spec.CgroupNamespace)
+	}
+	if cgroupMode.IsPrivate() {
+		return g.AddOrReplaceLinuxNamespace(spec.CgroupNamespace, "")
+	}
+	if cgroupMode.IsContainer() {
+		logrus.Debug("Using container cgroup mode")
+	}
+	return nil
+}
+
 func addRlimits(config *CreateConfig, g *generate.Generator) error {
 	var (
 		kernelMax  uint64 = 1048576
diff --git a/test/e2e/run_ns_test.go b/test/e2e/run_ns_test.go
index 6ba0d1aba..e3e86fc66 100644
--- a/test/e2e/run_ns_test.go
+++ b/test/e2e/run_ns_test.go
@@ -51,6 +51,15 @@ var _ = Describe("Podman run ns", func() {
 		Expect(session.ExitCode()).To(Not(Equal(0)))
 	})
 
+	It("podman run --cgroup private test", func() {
+		session := podmanTest.Podman([]string{"run", "--cgroupns=private", fedoraMinimal, "cat", "/proc/self/cgroup"})
+		session.WaitWithDefaultTimeout()
+		Expect(session.ExitCode()).To(Equal(0))
+
+		output := session.OutputToString()
+		Expect(output).ToNot(ContainSubstring("slice"))
+	})
+
 	It("podman run ipcns test", func() {
 		setup := SystemExec("ls", []string{"--inode", "-d", "/dev/shm"})
 		Expect(setup.ExitCode()).To(Equal(0))