networking, userNS: configure the network namespace after create

so that the OCI runtime creates the network namespace from the correct
userNS.

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>

Closes: #690
Approved by: mheon

5 files changed, 106 insertions, 27 deletions

diff --git a/cmd/podman/spec.go b/cmd/podman/spec.go
index 15dab6c4d..747d76359 100644
--- a/cmd/podman/spec.go
+++ b/cmd/podman/spec.go
@@ -167,6 +167,7 @@ func createConfigToOCISpec(config *createConfig) (*spec.Spec, error) {
 	cgroupPerm := "ro"
 	g := generate.New()
 	g.HostSpecific = true
+	addCgroup := true
 	if config.Privileged {
 		cgroupPerm = "rw"
 		g.RemoveMount("/sys")
@@ -177,14 +178,27 @@ func createConfigToOCISpec(config *createConfig) (*spec.Spec, error) {
 			Options:     []string{"nosuid", "noexec", "nodev", "rw"},
 		}
 		g.AddMount(sysMnt)
+	} else if !config.UsernsMode.IsHost() && config.NetMode.IsHost() {
+		addCgroup = false
+		g.RemoveMount("/sys")
+		sysMnt := spec.Mount{
+			Destination: "/sys",
+			Type:        "bind",
+			Source:      "/sys",
+			Options:     []string{"nosuid", "noexec", "nodev", "ro", "rbind"},
+		}
+		g.AddMount(sysMnt)
 	}
-	cgroupMnt := spec.Mount{
-		Destination: "/sys/fs/cgroup",
-		Type:        "cgroup",
-		Source:      "cgroup",
-		Options:     []string{"nosuid", "noexec", "nodev", "relatime", cgroupPerm},
+
+	if addCgroup {
+		cgroupMnt := spec.Mount{
+			Destination: "/sys/fs/cgroup",
+			Type:        "cgroup",
+			Source:      "cgroup",
+			Options:     []string{"nosuid", "noexec", "nodev", "relatime", cgroupPerm},
+		}
+		g.AddMount(cgroupMnt)
 	}
-	g.AddMount(cgroupMnt)
 	g.SetProcessCwd(config.WorkDir)
 	g.SetProcessArgs(config.Command)
 	g.SetProcessTerminal(config.Tty)
@@ -697,8 +711,9 @@ func (c *createConfig) GetContainerCreateOptions() ([]libpod.CtrCreateOption, er
 		}
 		options = append(options, libpod.WithNetNSFrom(connectedCtr))
 	} else if !c.NetMode.IsHost() && !c.NetMode.IsNone() {
-		options = append(options, libpod.WithNetNS([]ocicni.PortMapping{}))
-		options = append(options, libpod.WithNetNS(portBindings))
+		postConfigureNetNS := (len(c.IDMappings.UIDMap) > 0 || len(c.IDMappings.GIDMap) > 0) && !c.UsernsMode.IsHost()
+		options = append(options, libpod.WithNetNS([]ocicni.PortMapping{}, postConfigureNetNS))
+		options = append(options, libpod.WithNetNS(portBindings, postConfigureNetNS))
 	}
 
 	if c.PidMode.IsContainer() {
diff --git a/libpod/container.go b/libpod/container.go
index fb1f83c29..5fa27af58 100644
--- a/libpod/container.go
+++ b/libpod/container.go
@@ -285,6 +285,8 @@ type ContainerConfig struct {
 	// File containing the conmon PID
 	ConmonPidFile string `json:"conmonPidFile,omitempty"`
 	// TODO log options for log drivers
+
+	PostConfigureNetNS bool `json:"postConfigureNetNS"`
 }
 
 // ContainerStatus returns a string representation for users
diff --git a/libpod/container_internal.go b/libpod/container_internal.go
index 73095316e..18b56e23c 100644
--- a/libpod/container_internal.go
+++ b/libpod/container_internal.go
@@ -416,6 +416,16 @@ func (c *Container) checkDependenciesRunningLocked(depCtrs map[string]*Container
 	return notRunning, nil
 }
 
+func (c *Container) completeNetworkSetup() error {
+	if !c.config.PostConfigureNetNS {
+		return nil
+	}
+	if err := c.syncContainer(); err != nil {
+		return err
+	}
+	return c.runtime.setupNetNS(c)
+}
+
 // Initialize a container, creating it in the runtime
 func (c *Container) init(ctx context.Context) error {
 	if err := c.makeBindMounts(); err != nil {
@@ -442,7 +452,11 @@ func (c *Container) init(ctx context.Context) error {
 
 	c.state.State = ContainerStateCreated
 
-	return c.save()
+	if err := c.save(); err != nil {
+		return err
+	}
+
+	return c.completeNetworkSetup()
 }
 
 // Reinitialize a container
@@ -626,7 +640,7 @@ func (c *Container) prepare() (err error) {
 	}
 
 	// Set up network namespace if not already set up
-	if c.config.CreateNetNS && c.state.NetNS == nil {
+	if c.config.CreateNetNS && c.state.NetNS == nil && !c.config.PostConfigureNetNS {
 		if err := c.runtime.createNetNS(c); err != nil {
 			// Tear down storage before exiting to make sure we
 			// don't leak mounts
@@ -913,7 +927,11 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) {
 
 	// If network namespace was requested, add it now
 	if c.config.CreateNetNS {
-		g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, c.state.NetNS.Path())
+		if c.config.PostConfigureNetNS {
+			g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, "")
+		} else {
+			g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, c.state.NetNS.Path())
+		}
 	}
 
 	// Remove the default /dev/shm mount to ensure we overwrite it
diff --git a/libpod/networking.go b/libpod/networking.go
index cceeb18d6..54a1c78de 100644
--- a/libpod/networking.go
+++ b/libpod/networking.go
@@ -1,7 +1,12 @@
 package libpod
 
 import (
+	"crypto/rand"
+	"fmt"
 	"net"
+	"os"
+	"path/filepath"
+	"strings"
 
 	cnitypes "github.com/containernetworking/cni/pkg/types/current"
 	"github.com/containernetworking/plugins/pkg/ns"
@@ -9,7 +14,7 @@ import (
 	"github.com/pkg/errors"
 	"github.com/projectatomic/libpod/utils"
 	"github.com/sirupsen/logrus"
-	"strings"
+	"golang.org/x/sys/unix"
 )
 
 // Get an OCICNI network config
@@ -24,20 +29,7 @@ func getPodNetwork(id, name, nsPath string, ports []ocicni.PortMapping) ocicni.P
 }
 
 // Create and configure a new network namespace for a container
-func (r *Runtime) createNetNS(ctr *Container) (err error) {
-	ctrNS, err := ns.NewNS()
-	if err != nil {
-		return errors.Wrapf(err, "error creating network namespace for container %s", ctr.ID())
-	}
-	defer func() {
-		if err != nil {
-			if err2 := ctrNS.Close(); err2 != nil {
-				logrus.Errorf("Error closing partially created network namespace for container %s: %v", ctr.ID(), err2)
-			}
-		}
-	}()
-
-	logrus.Debugf("Made network namespace at %s for container %s", ctrNS.Path(), ctr.ID())
+func (r *Runtime) configureNetNS(ctr *Container, ctrNS ns.NetNS) (err error) {
 	podNetwork := getPodNetwork(ctr.ID(), ctr.Name(), ctrNS.Path(), ctr.config.PortMappings)
 
 	result, err := r.netPlugin.SetUpPod(podNetwork)
@@ -74,6 +66,57 @@ func (r *Runtime) createNetNS(ctr *Container) (err error) {
 	return nil
 }
 
+// Create and configure a new network namespace for a container
+func (r *Runtime) createNetNS(ctr *Container) (err error) {
+	ctrNS, err := ns.NewNS()
+	if err != nil {
+		return errors.Wrapf(err, "error creating network namespace for container %s", ctr.ID())
+	}
+	defer func() {
+		if err != nil {
+			if err2 := ctrNS.Close(); err2 != nil {
+				logrus.Errorf("Error closing partially created network namespace for container %s: %v", ctr.ID(), err2)
+			}
+		}
+	}()
+
+	logrus.Debugf("Made network namespace at %s for container %s", ctrNS.Path(), ctr.ID())
+	return r.configureNetNS(ctr, ctrNS)
+}
+
+// Configure the network namespace using the container process
+func (r *Runtime) setupNetNS(ctr *Container) (err error) {
+	nsProcess := fmt.Sprintf("/proc/%d/ns/net", ctr.state.PID)
+
+	b := make([]byte, 16)
+
+	if _, err := rand.Reader.Read(b); err != nil {
+		return errors.Wrapf(err, "failed to generate random netns name")
+	}
+
+	nsPath := fmt.Sprintf("/var/run/netns/cni-%x-%x-%x-%x-%x", b[0:4], b[4:6], b[6:8], b[8:10], b[10:])
+
+	if err := os.MkdirAll(filepath.Dir(nsPath), 0711); err != nil {
+		return errors.Wrapf(err, "cannot create %s", filepath.Dir(nsPath))
+	}
+
+	mountPointFd, err := os.Create(nsPath)
+	if err != nil {
+		return errors.Wrapf(err, "cannot open %s", nsPath)
+	}
+	mountPointFd.Close()
+
+	if err := unix.Mount(nsProcess, nsPath, "none", unix.MS_BIND, ""); err != nil {
+		return errors.Wrapf(err, "cannot mount %s", nsPath)
+	}
+
+	netNS, err := ns.GetNS(nsPath)
+	if err != nil {
+		return err
+	}
+	return r.configureNetNS(ctr, netNS)
+}
+
 // iptablesDNS accepts an arg (-I|-D) and IP address of the container and then
 // generates an iptables command to either add or subtract the needed rule
 func iptablesDNS(arg, ip string) error {
diff --git a/libpod/options.go b/libpod/options.go
index eaca70afc..2b3ad2eb4 100644
--- a/libpod/options.go
+++ b/libpod/options.go
@@ -707,7 +707,7 @@ func WithDependencyCtrs(ctrs []*Container) CtrCreateOption {
 // namespace with a minimal configuration.
 // An optional array of port mappings can be provided.
 // Conflicts with WithNetNSFrom().
-func WithNetNS(portMappings []ocicni.PortMapping) CtrCreateOption {
+func WithNetNS(portMappings []ocicni.PortMapping, postConfigureNetNS bool) CtrCreateOption {
 	return func(ctr *Container) error {
 		if ctr.valid {
 			return ErrCtrFinalized
@@ -717,6 +717,7 @@ func WithNetNS(portMappings []ocicni.PortMapping) CtrCreateOption {
 			return errors.Wrapf(ErrInvalidArg, "container is already set to join another container's net ns, cannot create a new net ns")
 		}
 
+		ctr.config.PostConfigureNetNS = postConfigureNetNS
 		ctr.config.CreateNetNS = true
 		ctr.config.PortMappings = portMappings