allow switching of port-forward approaches in rootless/using slirp4netns

As of podman 1.8.0, because of commit da7595a, the default approach of providing
port-forwarding in rootless mode has switched (and been hard-coded) to rootlessport,
for the purpose of providing super performance. The side-effect of this switch is
source within the container to the port-forwarded service always appears to originate
from 127.0.0.1 (see issue #5138).

This commit allows a user to specify if they want to revert to the previous approach
of leveraging slirp4netns add_hostfwd() api which, although not as stellar performance,
restores usefulness of seeing incoming traffic origin IP addresses.

The change should be transparent; when not specified, rootlessport will continue to be
used, however if specifying --net slirp4netns:slirplisten the old approach will be used.

Note: the above may imply the restored port-forwarding via slirp4netns is not as
performant as the new rootlessport approach, however the figures shared in the original
commit that introduced rootlessport are as follows:
slirp4netns: 8.3 Gbps,
RootlessKit: 27.3 Gbps,
which are more than sufficient for many use cases where the origin of traffic is more
important than limits that cannot be reached due to bottlenecks elsewhere.

Signed-off-by: Aleks Mariusz <m.k@alek.cx>
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

4 files changed, 160 insertions, 8 deletions

diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go
index 5a8faa7a4..db8aa219b 100644
--- a/libpod/networking_linux.go
+++ b/libpod/networking_linux.go
@@ -173,6 +173,19 @@ type slirpFeatures struct {
 	HasEnableSeccomp       bool
 }
 
+type slirp4netnsCmdArg struct {
+	Proto     string `json:"proto,omitempty"`
+	HostAddr  string `json:"host_addr"`
+	HostPort  int32  `json:"host_port"`
+	GuestAddr string `json:"guest_addr"`
+	GuestPort int32  `json:"guest_port"`
+}
+
+type slirp4netnsCmd struct {
+	Execute string            `json:"execute"`
+	Args    slirp4netnsCmdArg `json:"arguments"`
+}
+
 func checkSlirpFlags(path string) (*slirpFeatures, error) {
 	cmd := exec.Command(path, "--help")
 	out, err := cmd.CombinedOutput()
@@ -228,6 +241,12 @@ func (r *Runtime) setupRootlessNetNS(ctr *Container) error {
 		cmdArgs = append(cmdArgs, "--enable-seccomp")
 	}
 
+	var apiSocket string
+	if havePortMapping && ctr.config.NetMode.IsPortForwardViaSlirpHostFwd() {
+		apiSocket = filepath.Join(ctr.runtime.config.Engine.TmpDir, fmt.Sprintf("%s.net", ctr.config.ID))
+		cmdArgs = append(cmdArgs, "--api-socket", apiSocket)
+	}
+
 	// 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
@@ -291,7 +310,11 @@ func (r *Runtime) setupRootlessNetNS(ctr *Container) error {
 	}
 
 	if havePortMapping {
-		return r.setupRootlessPortMapping(ctr, netnsPath)
+		if ctr.config.NetMode.IsPortForwardViaSlirpHostFwd() {
+			return r.setupRootlessPortMappingViaSlirp(ctr, cmd, apiSocket)
+		} else {
+			return r.setupRootlessPortMappingViaRLK(ctr, netnsPath)
+		}
 	}
 	return nil
 }
@@ -342,7 +365,7 @@ func waitForSync(syncR *os.File, cmd *exec.Cmd, logFile io.ReadSeeker, timeout t
 	return nil
 }
 
-func (r *Runtime) setupRootlessPortMapping(ctr *Container, netnsPath string) error {
+func (r *Runtime) setupRootlessPortMappingViaRLK(ctr *Container, netnsPath string) error {
 	syncR, syncW, err := os.Pipe()
 	if err != nil {
 		return errors.Wrapf(err, "failed to open pipe")
@@ -419,6 +442,90 @@ func (r *Runtime) setupRootlessPortMapping(ctr *Container, netnsPath string) err
 	return nil
 }
 
+func (r *Runtime) setupRootlessPortMappingViaSlirp(ctr *Container, cmd *exec.Cmd, apiSocket string) (err error) {
+	const pidWaitTimeout = 60 * time.Second
+	chWait := make(chan error)
+	go func() {
+		interval := 25 * time.Millisecond
+		for i := time.Duration(0); i < pidWaitTimeout; i += interval {
+			// Check if the process is still running.
+			var status syscall.WaitStatus
+			pid, err := syscall.Wait4(cmd.Process.Pid, &status, syscall.WNOHANG, nil)
+			if err != nil {
+				break
+			}
+			if pid != cmd.Process.Pid {
+				continue
+			}
+			if status.Exited() || status.Signaled() {
+				chWait <- fmt.Errorf("slirp4netns exited with status %d", status.ExitStatus())
+			}
+			time.Sleep(interval)
+		}
+	}()
+	defer close(chWait)
+
+	// wait that API socket file appears before trying to use it.
+	if _, err := WaitForFile(apiSocket, chWait, pidWaitTimeout); err != nil {
+		return errors.Wrapf(err, "waiting for slirp4nets to create the api socket file %s", apiSocket)
+	}
+
+	// for each port we want to add we need to open a connection to the slirp4netns control socket
+	// and send the add_hostfwd command.
+	for _, i := range ctr.config.PortMappings {
+		conn, err := net.Dial("unix", apiSocket)
+		if err != nil {
+			return errors.Wrapf(err, "cannot open connection to %s", apiSocket)
+		}
+		defer func() {
+			if err := conn.Close(); err != nil {
+				logrus.Errorf("unable to close connection: %q", err)
+			}
+		}()
+		hostIP := i.HostIP
+		if hostIP == "" {
+			hostIP = "0.0.0.0"
+		}
+		apiCmd := slirp4netnsCmd{
+			Execute: "add_hostfwd",
+			Args: slirp4netnsCmdArg{
+				Proto:     i.Protocol,
+				HostAddr:  hostIP,
+				HostPort:  i.HostPort,
+				GuestPort: i.ContainerPort,
+			},
+		}
+		// create the JSON payload and send it.  Mark the end of request shutting down writes
+		// to the socket, as requested by slirp4netns.
+		data, err := json.Marshal(&apiCmd)
+		if err != nil {
+			return errors.Wrapf(err, "cannot marshal JSON for slirp4netns")
+		}
+		if _, err := conn.Write([]byte(fmt.Sprintf("%s\n", data))); err != nil {
+			return errors.Wrapf(err, "cannot write to control socket %s", apiSocket)
+		}
+		if err := conn.(*net.UnixConn).CloseWrite(); err != nil {
+			return errors.Wrapf(err, "cannot shutdown the socket %s", apiSocket)
+		}
+		buf := make([]byte, 2048)
+		readLength, err := conn.Read(buf)
+		if err != nil {
+			return errors.Wrapf(err, "cannot read from control socket %s", apiSocket)
+		}
+		// if there is no 'error' key in the received JSON data, then the operation was
+		// successful.
+		var y map[string]interface{}
+		if err := json.Unmarshal(buf[0:readLength], &y); err != nil {
+			return errors.Wrapf(err, "error parsing error status from slirp4netns")
+		}
+		if e, found := y["error"]; found {
+			return errors.Errorf("error from slirp4netns while setting up port redirection: %v", e)
+		}
+	}
+	logrus.Debug("slirp4netns port-forwarding setup via add_hostfwd is ready")
+	return nil
+}
+
 // Configure the network namespace using the container process
 func (r *Runtime) setupNetNS(ctr *Container) error {
 	nsProcess := fmt.Sprintf("/proc/%d/ns/net", ctr.state.PID)
diff --git a/pkg/namespaces/namespaces.go b/pkg/namespaces/namespaces.go
index 2ffbde977..b4ec04699 100644
--- a/pkg/namespaces/namespaces.go
+++ b/pkg/namespaces/namespaces.go
@@ -17,7 +17,9 @@ const (
 	nsType        = "ns"
 	podType       = "pod"
 	privateType   = "private"
+	rlkFwdType    = "port_handler=rootlesskit"
 	shareableType = "shareable"
+	slirpFwdType  = "port_handler=slirp4netns"
 	slirpType     = "slirp4netns"
 )
 
@@ -385,7 +387,29 @@ func (n NetworkMode) IsBridge() bool {
 
 // IsSlirp4netns indicates if we are running a rootless network stack
 func (n NetworkMode) IsSlirp4netns() bool {
-	return n == slirpType
+	return n == slirpType || strings.HasPrefix(string(n), slirpType+":")
+}
+
+// IsPortForwardViaRootlessKit indicates if we are doing rootless port-forwarding via rootlesskit/rootlessport
+func (n NetworkMode) IsPortForwardViaRootlessKit() bool {
+	if !n.IsSlirp4netns() {
+		return false
+	}
+	parts := strings.SplitN(string(n), ":", 2)
+	if len(parts) == 2 {
+		return parts[1] == rlkFwdType
+	}
+	return true
+}
+
+// IsPortForwardViaSlirpHostFwd indicates if we are doing rootless port-forwarding via slirp4netns add_hostfwd()
+func (n NetworkMode) IsPortForwardViaSlirpHostFwd() bool {
+	if !n.IsSlirp4netns() {
+		return false
+	}
+	// below here, implied IsSlirp4netns() == true
+	parts := strings.SplitN(string(n), ":", 2)
+	return len(parts) > 1 && parts[1] == slirpFwdType
 }
 
 // IsNS indicates a network namespace passed in by path (ns:<path>)
diff --git a/pkg/specgen/generate/namespaces.go b/pkg/specgen/generate/namespaces.go
index 09d6ba445..e038fc84e 100644
--- a/pkg/specgen/generate/namespaces.go
+++ b/pkg/specgen/generate/namespaces.go
@@ -2,6 +2,7 @@ package generate
 
 import (
 	"context"
+	"fmt"
 	"os"
 	"strings"
 
@@ -226,7 +227,11 @@ func namespaceOptions(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.
 		if err != nil {
 			return nil, err
 		}
-		toReturn = append(toReturn, libpod.WithNetNS(portMappings, postConfigureNetNS, "slirp4netns", nil))
+		val := "slirp4netns"
+		if s.NetNS.Value != "" {
+			val = fmt.Sprintf("slirp4netns:%s", s.NetNS.Value)
+		}
+		toReturn = append(toReturn, libpod.WithNetNS(portMappings, postConfigureNetNS, val, nil))
 	case specgen.Bridge:
 		portMappings, err := createPortMappings(ctx, s, img)
 		if err != nil {
diff --git a/pkg/specgen/namespaces.go b/pkg/specgen/namespaces.go
index 5f56b242b..e62c8b709 100644
--- a/pkg/specgen/namespaces.go
+++ b/pkg/specgen/namespaces.go
@@ -108,7 +108,19 @@ func validateNetNS(n *Namespace) error {
 		return nil
 	}
 	switch n.NSMode {
-	case "", Default, Host, Path, FromContainer, FromPod, Private, NoNetwork, Bridge, Slirp:
+	case Slirp:
+		if n.Value != "" {
+			parts := strings.Split(n.Value, ",")
+			for _, p := range parts {
+				switch p {
+				case "port_handler=slirp4netns", "port_handler=rootlesskit":
+				default:
+					return errors.Errorf("invalid value for slirp %q", n.Value)
+				}
+			}
+		}
+		break
+	case "", Default, Host, Path, FromContainer, FromPod, Private, NoNetwork, Bridge:
 		break
 	default:
 		return errors.Errorf("invalid network %q", n.NSMode)
@@ -119,8 +131,8 @@ func validateNetNS(n *Namespace) error {
 		if len(n.Value) < 1 {
 			return errors.Errorf("namespace mode %s requires a value", n.NSMode)
 		}
-	} else {
-		// All others must NOT set a string value
+	} else if n.NSMode != Slirp {
+		// All others except must NOT set a string value
 		if len(n.Value) > 0 {
 			return errors.Errorf("namespace value %s cannot be provided with namespace mode %s", n.Value, n.NSMode)
 		}
@@ -250,8 +262,12 @@ func ParseNetworkNamespace(ns string) (Namespace, []string, error) {
 	var cniNetworks []string
 	// Net defaults to Slirp on rootless
 	switch {
-	case ns == "slirp4netns":
+	case ns == "slirp4netns", strings.HasPrefix(ns, "slirp4netns:"):
+		split := strings.SplitN(ns, ":", 2)
 		toReturn.NSMode = Slirp
+		if len(split) > 1 {
+			toReturn.Value = split[1]
+		}
 	case ns == "pod":
 		toReturn.NSMode = FromPod
 	case ns == "":