rootless: Rearrange setup of rootless containers

In order to run Podman with VM-based runtimes unprivileged, the
network must be set up prior to the container creation. Therefore
this commit modifies Podman to run rootless containers by:
  1. create a network namespace
  2. pass the netns persistent mount path to the slirp4netns
     to create the tap inferface
  3. pass the netns path to the OCI spec, so the runtime can
     enter the netns

Closes #2897

Signed-off-by: Gabi Beyer <gabrielle.n.beyer@intel.com>

6 files changed, 89 insertions, 34 deletions

diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go
index 230b5b888..e7b4b2b22 100644
--- a/libpod/container_internal_linux.go
+++ b/libpod/container_internal_linux.go
@@ -78,15 +78,21 @@ func (c *Container) prepare() (Err error) {
 		// Set up network namespace if not already set up
 		if c.config.CreateNetNS && c.state.NetNS == nil && !c.config.PostConfigureNetNS {
 			netNS, networkStatus, createNetNSErr = c.runtime.createNetNS(c)
+			if createNetNSErr != nil {
+				return
+			}
 
 			tmpStateLock.Lock()
 			defer tmpStateLock.Unlock()
 
 			// Assign NetNS attributes to container
-			if createNetNSErr == nil {
-				c.state.NetNS = netNS
-				c.state.NetworkStatus = networkStatus
-			}
+			c.state.NetNS = netNS
+			c.state.NetworkStatus = networkStatus
+		}
+
+		// handle rootless network namespace setup
+		if c.state.NetNS != nil && c.config.NetMode == "slirp4netns" && !c.config.PostConfigureNetNS {
+			createNetNSErr = c.runtime.setupRootlessNetNS(c)
 		}
 	}()
 	// Mount storage if not mounted
diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go
index 67dd0150b..d854a2de6 100644
--- a/libpod/networking_linux.go
+++ b/libpod/networking_linux.go
@@ -90,9 +90,6 @@ func (r *Runtime) configureNetNS(ctr *Container, ctrNS ns.NetNS) ([]*cnitypes.Re
 
 // Create and configure a new network namespace for a container
 func (r *Runtime) createNetNS(ctr *Container) (n ns.NetNS, q []*cnitypes.Result, err error) {
-	if rootless.IsRootless() {
-		return nil, nil, errors.New("cannot configure a new network namespace in rootless mode, only --network=slirp4netns is supported")
-	}
 	ctrNS, err := netns.NewNS()
 	if err != nil {
 		return nil, nil, errors.Wrapf(err, "error creating network namespace for container %s", ctr.ID())
@@ -110,7 +107,10 @@ func (r *Runtime) createNetNS(ctr *Container) (n ns.NetNS, q []*cnitypes.Result,
 
 	logrus.Debugf("Made network namespace at %s for container %s", ctrNS.Path(), ctr.ID())
 
-	networkStatus, err := r.configureNetNS(ctr, ctrNS)
+	networkStatus := []*cnitypes.Result{}
+	if !rootless.IsRootless() {
+		networkStatus, err = r.configureNetNS(ctr, ctrNS)
+	}
 	return ctrNS, networkStatus, err
 }
 
@@ -138,9 +138,6 @@ func checkSlirpFlags(path string) (bool, bool, bool, error) {
 
 // Configure the network namespace for a rootless container
 func (r *Runtime) setupRootlessNetNS(ctr *Container) (err error) {
-	defer errorhandling.CloseQuiet(ctr.rootlessSlirpSyncR)
-	defer errorhandling.CloseQuiet(ctr.rootlessSlirpSyncW)
-
 	path := r.config.NetworkCmdPath
 
 	if path == "" {
@@ -164,7 +161,7 @@ func (r *Runtime) setupRootlessNetNS(ctr *Container) (err error) {
 
 	cmdArgs := []string{}
 	if havePortMapping {
-		cmdArgs = append(cmdArgs, "--api-socket", apiSocket, fmt.Sprintf("%d", ctr.state.PID))
+		cmdArgs = append(cmdArgs, "--api-socket", apiSocket)
 	}
 	dhp, mtu, sandbox, err := checkSlirpFlags(path)
 	if err != nil {
@@ -179,13 +176,32 @@ func (r *Runtime) setupRootlessNetNS(ctr *Container) (err error) {
 	if sandbox {
 		cmdArgs = append(cmdArgs, "--enable-sandbox")
 	}
-	cmdArgs = append(cmdArgs, "-c", "-e", "3", "-r", "4", fmt.Sprintf("%d", ctr.state.PID), "tap0")
 
-	cmd := exec.Command(path, cmdArgs...)
+	// the slirp4netns arguments being passed are describes as follows:
+	// from the slirp4netns documentation: https://github.com/rootless-containers/slirp4netns
+	// -c, --configure Brings up the tap interface
+	// -e, --exit-fd=FD specify the FD for terminating slirp4netns
+	// -r, --ready-fd=FD specify the FD to write to when the initialization steps are finished
+	cmdArgs = append(cmdArgs, "-c", "-e", "3", "-r", "4")
+	if !ctr.config.PostConfigureNetNS {
+		ctr.rootlessSlirpSyncR, ctr.rootlessSlirpSyncW, err = os.Pipe()
+		if err != nil {
+			return errors.Wrapf(err, "failed to create rootless network sync pipe")
+		}
+		cmdArgs = append(cmdArgs, "--netns-type=path", ctr.state.NetNS.Path(), "tap0")
+	} else {
+		defer errorhandling.CloseQuiet(ctr.rootlessSlirpSyncR)
+		defer errorhandling.CloseQuiet(ctr.rootlessSlirpSyncW)
+		cmdArgs = append(cmdArgs, fmt.Sprintf("%d", ctr.state.PID), "tap0")
+	}
 
+	cmd := exec.Command(path, cmdArgs...)
+	logrus.Debugf("slirp4netns command: %s", strings.Join(cmd.Args, " "))
 	cmd.SysProcAttr = &syscall.SysProcAttr{
 		Setpgid: true,
 	}
+
+	// Leak one end of the pipe in slirp4netns, the other will be sent to conmon
 	cmd.ExtraFiles = append(cmd.ExtraFiles, ctr.rootlessSlirpSyncR, syncW)
 
 	if err := cmd.Start(); err != nil {
@@ -388,20 +404,22 @@ func (r *Runtime) teardownNetNS(ctr *Container) error {
 
 	logrus.Debugf("Tearing down network namespace at %s for container %s", ctr.state.NetNS.Path(), ctr.ID())
 
-	var requestedIP net.IP
-	if ctr.requestedIP != nil {
-		requestedIP = ctr.requestedIP
-		// cancel request for a specific IP in case the container is reused later
-		ctr.requestedIP = nil
-	} else {
-		requestedIP = ctr.config.StaticIP
-	}
+	// rootless containers do not use the CNI plugin
+	if !rootless.IsRootless() {
+		var requestedIP net.IP
+		if ctr.requestedIP != nil {
+			requestedIP = ctr.requestedIP
+			// cancel request for a specific IP in case the container is reused later
+			ctr.requestedIP = nil
+		} else {
+			requestedIP = ctr.config.StaticIP
+		}
 
-	podNetwork := r.getPodNetwork(ctr.ID(), ctr.Name(), ctr.state.NetNS.Path(), ctr.config.Networks, ctr.config.PortMappings, requestedIP)
+		podNetwork := r.getPodNetwork(ctr.ID(), ctr.Name(), ctr.state.NetNS.Path(), ctr.config.Networks, ctr.config.PortMappings, requestedIP)
 
-	// The network may have already been torn down, so don't fail here, just log
-	if err := r.netPlugin.TearDownPod(podNetwork); err != nil {
-		return errors.Wrapf(err, "error tearing down CNI namespace configuration for container %s", ctr.ID())
+		if err := r.netPlugin.TearDownPod(podNetwork); err != nil {
+			return errors.Wrapf(err, "error tearing down CNI namespace configuration for container %s", ctr.ID())
+		}
 	}
 
 	// First unmount the namespace
diff --git a/libpod/oci_internal_linux.go b/libpod/oci_internal_linux.go
index a90af32ed..a5cce795b 100644
--- a/libpod/oci_internal_linux.go
+++ b/libpod/oci_internal_linux.go
@@ -131,9 +131,14 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, restoreOptions *Containe
 	}
 
 	if ctr.config.NetMode.IsSlirp4netns() {
-		ctr.rootlessSlirpSyncR, ctr.rootlessSlirpSyncW, err = os.Pipe()
-		if err != nil {
-			return errors.Wrapf(err, "failed to create rootless network sync pipe")
+		if ctr.config.PostConfigureNetNS {
+			ctr.rootlessSlirpSyncR, ctr.rootlessSlirpSyncW, err = os.Pipe()
+			if err != nil {
+				return errors.Wrapf(err, "failed to create rootless network sync pipe")
+			}
+		} else {
+			defer errorhandling.CloseQuiet(ctr.rootlessSlirpSyncR)
+			defer errorhandling.CloseQuiet(ctr.rootlessSlirpSyncW)
 		}
 		// Leak one end in conmon, the other one will be leaked into slirp4netns
 		cmd.ExtraFiles = append(cmd.ExtraFiles, ctr.rootlessSlirpSyncW)
diff --git a/libpod/runtime_pod_infra_linux.go b/libpod/runtime_pod_infra_linux.go
index ad6662f03..6a27c2800 100644
--- a/libpod/runtime_pod_infra_linux.go
+++ b/libpod/runtime_pod_infra_linux.go
@@ -99,7 +99,9 @@ func (r *Runtime) makeInfraContainer(ctx context.Context, p *Pod, imgName, imgID
 	if isRootless {
 		netmode = "slirp4netns"
 	}
-	options = append(options, WithNetNS(p.config.InfraContainer.PortBindings, isRootless, netmode, networks))
+	// PostConfigureNetNS should not be set since user namespace sharing is not implemented
+	// and rootless networking no longer supports post configuration setup
+	options = append(options, WithNetNS(p.config.InfraContainer.PortBindings, false, netmode, networks))
 
 	return r.newContainer(ctx, g.Config, options...)
 }
diff --git a/pkg/netns/netns_linux.go b/pkg/netns/netns_linux.go
index 1d6fb873c..18d638809 100644
--- a/pkg/netns/netns_linux.go
+++ b/pkg/netns/netns_linux.go
@@ -23,23 +23,42 @@ import (
 	"fmt"
 	"os"
 	"path"
+	"path/filepath"
 	"runtime"
 	"strings"
 	"sync"
 
 	"github.com/containernetworking/plugins/pkg/ns"
+	"github.com/containers/libpod/pkg/rootless"
+	"github.com/containers/libpod/pkg/util"
 	"github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
 )
 
-const nsRunDir = "/var/run/netns"
+// get NSRunDir returns the dir of where to create the netNS. When running
+// rootless, it needs to be at a location writable by user.
+func getNSRunDir() (string, error) {
+	if rootless.IsRootless() {
+		rootlessDir, err := util.GetRuntimeDir()
+		if err != nil {
+			return "", err
+		}
+		return filepath.Join(rootlessDir, "netns"), nil
+	}
+	return "/var/run/netns", nil
+}
 
 // NewNS creates a new persistent (bind-mounted) network namespace and returns
 // an object representing that namespace, without switching to it.
 func NewNS() (ns.NetNS, error) {
 
+	nsRunDir, err := getNSRunDir()
+	if err != nil {
+		return nil, err
+	}
+
 	b := make([]byte, 16)
-	_, err := rand.Reader.Read(b)
+	_, err = rand.Reader.Read(b)
 	if err != nil {
 		return nil, fmt.Errorf("failed to generate random netns name: %v", err)
 	}
@@ -127,7 +146,7 @@ func NewNS() (ns.NetNS, error) {
 		// Put this thread back to the orig ns, since it might get reused (pre go1.10)
 		defer func() {
 			if err := origNS.Set(); err != nil {
-				logrus.Errorf("unable to set namespace: %q", err)
+				logrus.Warnf("unable to set namespace: %q", err)
 			}
 		}()
 
@@ -150,6 +169,11 @@ func NewNS() (ns.NetNS, error) {
 
 // UnmountNS unmounts the NS held by the netns object
 func UnmountNS(ns ns.NetNS) error {
+	nsRunDir, err := getNSRunDir()
+	if err != nil {
+		return err
+	}
+
 	nsPath := ns.Path()
 	// Only unmount if it's been bind-mounted (don't touch namespaces in /proc...)
 	if strings.HasPrefix(nsPath, nsRunDir) {
diff --git a/pkg/spec/createconfig.go b/pkg/spec/createconfig.go
index c17172016..7c3195be4 100644
--- a/pkg/spec/createconfig.go
+++ b/pkg/spec/createconfig.go
@@ -275,7 +275,7 @@ func (c *CreateConfig) getContainerCreateOptions(runtime *libpod.Runtime, pod *l
 		options = append(options, libpod.WithNetNSFrom(connectedCtr))
 	} else if !c.NetMode.IsHost() && !c.NetMode.IsNone() {
 		hasUserns := c.UsernsMode.IsContainer() || c.UsernsMode.IsNS() || len(c.IDMappings.UIDMap) > 0 || len(c.IDMappings.GIDMap) > 0
-		postConfigureNetNS := c.NetMode.IsSlirp4netns() || (hasUserns && !c.UsernsMode.IsHost())
+		postConfigureNetNS := hasUserns && !c.UsernsMode.IsHost()
 		options = append(options, libpod.WithNetNS(portBindings, postConfigureNetNS, string(c.NetMode), networks))
 	}