Add pasta-based port forwarding for rootless bridge networks

Add rootless_port_forwarder="pasta" option that uses pesto to update
pasta's forwarding table via UNIX socket, preserving source IPs that
rootlessport's userspace proxy masks.

HostIP is stripped from port mappings in the netavark wrapper when
pasta forwarding is active because pesto handles host-side binding
while pasta's splice changes the destination IP that netavark DNAT
expects. Pesto binds both 0.0.0.0 and [::] for dual-stack support.

Fixes: https://redhat.atlassian.net/browse/RUN-2214
Fixes: #8193
Fixes: https://redhat.atlassian.net/browse/RUN-3587

Signed-off-by: Jan Rodák <hony.com@seznam.cz>

Author: Honny1

libpod/define/info.go

@@ -53,8 +53,11 @@ type HostInfo struct {
 	// RemoteSocket returns the UNIX domain socket the Podman service is listening on
 	RemoteSocket *RemoteSocket `json:"remoteSocket,omitempty"`
 	// RootlessNetworkCmd returns the default rootless network command (pasta)
-	RootlessNetworkCmd string         `json:"rootlessNetworkCmd"`
-	RuntimeInfo        map[string]any `json:"runtimeInfo,omitempty"`
+	RootlessNetworkCmd string `json:"rootlessNetworkCmd"`
+	// RootlessPortForwarder returns the port forwarding mechanism for rootless
+	// bridge networks: "rootlessport" (default) or "pasta" (experimental)
+	RootlessPortForwarder string         `json:"rootlessPortForwarder"`
+	RuntimeInfo           map[string]any `json:"runtimeInfo,omitempty"`
 	// ServiceIsRemote is true when the podman/libpod service is remote to the client
 	ServiceIsRemote bool         `json:"serviceIsRemote"`
 	Security        SecurityInfo `json:"security"`

libpod/info.go

@@ -108,26 +108,27 @@ func (r *Runtime) hostInfo() (*define.HostInfo, error) {
 	}
 
 	info := define.HostInfo{
-		Arch:               runtime.GOARCH,
-		BuildahVersion:     buildah.Version,
-		DatabaseBackend:    r.state.Name(),
-		Linkmode:           linkmode.Linkmode(),
-		CPUs:               runtime.NumCPU(),
-		CPUUtilization:     cpuUtil,
-		Distribution:       hostDistributionInfo,
-		LogDriver:          r.config.Containers.LogDriver,
-		EventLogger:        r.eventer.String(),
-		FreeLocks:          locksFree,
-		Hostname:           host,
-		Kernel:             kv,
-		MemFree:            mi.MemFree,
-		MemTotal:           mi.MemTotal,
-		NetworkBackend:     r.config.Network.NetworkBackend,
-		NetworkBackendInfo: r.network.NetworkInfo(),
-		OS:                 runtime.GOOS,
-		RootlessNetworkCmd: r.config.Network.DefaultRootlessNetworkCmd,
-		SwapFree:           mi.SwapFree,
-		SwapTotal:          mi.SwapTotal,
+		Arch:                  runtime.GOARCH,
+		BuildahVersion:        buildah.Version,
+		DatabaseBackend:       r.state.Name(),
+		Linkmode:              linkmode.Linkmode(),
+		CPUs:                  runtime.NumCPU(),
+		CPUUtilization:        cpuUtil,
+		Distribution:          hostDistributionInfo,
+		LogDriver:             r.config.Containers.LogDriver,
+		EventLogger:           r.eventer.String(),
+		FreeLocks:             locksFree,
+		Hostname:              host,
+		Kernel:                kv,
+		MemFree:               mi.MemFree,
+		MemTotal:              mi.MemTotal,
+		NetworkBackend:        r.config.Network.NetworkBackend,
+		NetworkBackendInfo:    r.network.NetworkInfo(),
+		OS:                    runtime.GOOS,
+		RootlessNetworkCmd:    r.config.Network.DefaultRootlessNetworkCmd,
+		RootlessPortForwarder: r.config.Network.RootlessPortForwarder,
+		SwapFree:              mi.SwapFree,
+		SwapTotal:             mi.SwapTotal,
 	}
 	platform := parse.DefaultPlatform()
 	pArr := strings.Split(platform, "/")

libpod/networking_common.go

@@ -110,11 +110,27 @@ func (r *Runtime) teardownNetwork(ctr *Container) error {
 		return err
 	}
 
-	if !ctr.config.NetMode.IsPasta() && len(networks) > 0 {
-		netOpts := ctr.getNetworkOptions(networks)
-		return r.teardownNetworkBackend(ctr.state.NetNS, netOpts)
+	if len(networks) == 0 {
+		return nil
 	}
-	return nil
+
+	// --net=pasta: per-container pasta cleans up when it exits, nothing to tear down.
+	if ctr.config.NetMode.IsPasta() {
+		return nil
+	}
+
+	// Pasta forwarding mode: remove port forwarding rules (via pesto) before
+	// netavark tears down bridge/nftables so pasta stops forwarding first.
+	// Rootlessport mode: no explicit teardown needed (exits with conmon).
+	if rootless.IsRootless() && ctr.config.NetMode.IsBridge() && len(ctr.config.PortMappings) > 0 &&
+		r.config.Network.RootlessPortForwarder == config.RootlessPortForwarderPasta {
+		if err := r.teardownRootlessPortMappingViaPesto(ctr); err != nil {
+			logrus.Warnf("pesto port cleanup failed for container %s: %v", ctr.ID(), err)
+		}
+	}
+
+	netOpts := ctr.getNetworkOptions(networks)
+	return r.teardownNetworkBackend(ctr.state.NetNS, netOpts)
 }
 
 // isBridgeNetMode checks if the given network mode is bridge.
@@ -439,7 +455,7 @@ func (c *Container) NetworkDisconnect(nameOrID, netName string, _ bool) error {
 
 	// Reload ports when there are still connected networks, maybe we removed the network interface with the child ip.
 	// Reloading without connected networks does not make sense, so we can skip this step.
-	if rootless.IsRootless() && len(networkStatus) > 0 {
+	if rootless.IsRootless() && c.runtime.config.Network.RootlessPortForwarder == config.RootlessPortForwarderRootlessport && len(networkStatus) > 0 {
 		if err := c.reloadRootlessRLKPortMapping(); err != nil {
 			return err
 		}
@@ -595,7 +611,7 @@ func (c *Container) NetworkConnect(nameOrID, netName string, netOpts types.PerNe
 
 	// The first network needs a port reload to set the correct child ip for the rootlessport process.
 	// Adding a second network does not require a port reload because the child ip is still valid.
-	if rootless.IsRootless() && len(networks) == 0 {
+	if rootless.IsRootless() && c.runtime.config.Network.RootlessPortForwarder == config.RootlessPortForwarderRootlessport && len(networks) == 0 {
 		if err := c.reloadRootlessRLKPortMapping(); err != nil {
 			return err
 		}

libpod/networking_freebsd.go

@@ -224,3 +224,7 @@ func (c *Container) inspectJoinedNetworkNS(_ string) (q types.StatusBlock, retEr
 func (c *Container) reloadRootlessRLKPortMapping() error {
 	return errors.New("unsupported (*Container).reloadRootlessRLKPortMapping")
 }
+
+func (r *Runtime) teardownRootlessPortMappingViaPesto(_ *Container) error {
+	return errors.New("unsupported teardownRootlessPortMappingViaPesto on FreeBSD")
+}

libpod/networking_linux.go

@@ -11,6 +11,7 @@ import (
 	"github.com/sirupsen/logrus"
 	"github.com/vishvananda/netlink"
 	"go.podman.io/common/libnetwork/types"
+	"go.podman.io/common/pkg/config"
 	"go.podman.io/common/pkg/netns"
 	"go.podman.io/podman/v6/libpod/define"
 	"go.podman.io/podman/v6/pkg/rootless"
@@ -59,15 +60,19 @@ func (r *Runtime) configureNetNS(ctr *Container, ctrNS string) (status map[strin
 		}
 	}()
 
-	// set up rootless port forwarder when rootless with ports and the network status is empty,
-	// if this is called from network reload the network status will not be empty and we should
-	// not set up port because they are still active
-	if rootless.IsRootless() && len(ctr.config.PortMappings) > 0 && ctr.getNetworkStatus() == nil {
-		// set up port forwarder for rootless netns
-		// make sure to fix this in container.handleRestartPolicy() as well
-		// Important we have to call this after r.setUpNetwork() so that
-		// we can use the proper netStatus
-		err = r.setupRootlessPortMappingViaRLK(ctr, ctrNS, netStatus)
+	// Set up port forwarding for rootless bridge networks.
+	if rootless.IsRootless() && len(ctr.config.PortMappings) > 0 {
+		switch r.config.Network.RootlessPortForwarder {
+		case config.RootlessPortForwarderPasta:
+			err = r.setupRootlessPortMappingViaPesto(ctr)
+		case config.RootlessPortForwarderRootlessport, "":
+			if ctr.getNetworkStatus() == nil {
+				err = r.setupRootlessPortMappingViaRLK(ctr, ctrNS, netStatus)
+			}
+		default:
+			err = fmt.Errorf("invalid rootless_port_forwarder value %q, must be %q or %q",
+				r.config.Network.RootlessPortForwarder, config.RootlessPortForwarderRootlessport, config.RootlessPortForwarderPasta)
+		}
 	}
 	return netStatus, err
 }

libpod/networking_pesto_linux.go

@@ -0,0 +1,52 @@
+//go:build !remote
+
+// Pesto integration for rootless bridge network port forwarding.
+//
+// A shared pasta instance in the rootless netns (-c pasta.sock) handles
+// host-side port forwarding. On container start/stop, pesto incrementally
+// adds or deletes port forwarding rules for that container. Pasta forwards
+// via kernel splice (localhost) or TAP (external), preserving source IPs.
+// The container sees the real client's address instead of a proxy or bridge
+// gateway address.
+//
+// Container start:
+// - netavark sets up bridge + DNAT
+// - pesto --add: adds this container's ports to pasta
+//
+// Container stop:
+// - pesto --delete: removes this container's ports from pasta
+// - netavark tears down bridge/DNAT
+
+package libpod
+
+import (
+	"go.podman.io/common/libnetwork/pasta"
+)
+
+func (r *Runtime) pestoSocketPath() string {
+	info, err := r.network.RootlessNetnsInfo()
+	if err != nil || info == nil {
+		return ""
+	}
+	return info.PestoSocketPath
+}
+
+// setupRootlessPortMappingViaPesto adds this container's port forwarding
+// rules to the shared pasta instance.
+func (r *Runtime) setupRootlessPortMappingViaPesto(ctr *Container) error {
+	ports := ctr.convertPortMappings()
+	if len(ports) == 0 {
+		return nil
+	}
+	return pasta.PestoAddPorts(r.config, r.pestoSocketPath(), ports)
+}
+
+// teardownRootlessPortMappingViaPesto removes this container's port
+// forwarding rules from the shared pasta instance.
+func (r *Runtime) teardownRootlessPortMappingViaPesto(ctr *Container) error {
+	ports := ctr.convertPortMappings()
+	if len(ports) == 0 {
+		return nil
+	}
+	return pasta.PestoDeletePorts(r.config, r.pestoSocketPath(), ports)
+}

libpod/oci_conmon_common.go

@@ -1188,8 +1188,11 @@ func (r *ConmonOCIRuntime) createOCIContainer(ctr *Container, restoreOptions *Co
 		// process cannot use them.
 		cmd.ExtraFiles = append(cmd.ExtraFiles, ports...)
 
-		// For rootless port forwarding, create sync pipe and leak write end to conmon
-		if rootless.IsRootless() && len(ctr.config.PortMappings) > 0 {
+		// For rootless port forwarding via rootlessport, create sync pipe and
+		// leak write end to conmon. Pasta forwarding mode does not use
+		// rootlessport, so no pipe is needed.
+		if rootless.IsRootless() && len(ctr.config.PortMappings) > 0 &&
+			ctr.runtime.config.Network.RootlessPortForwarder == config.RootlessPortForwarderRootlessport {
 			ctr.rootlessPortSyncR, ctr.rootlessPortSyncW, err = os.Pipe()
 			if err != nil {
 				return 0, fmt.Errorf("failed to create rootless port sync pipe: %w", err)

test/e2e/common_test.go

@@ -1629,6 +1629,35 @@ func testPortConnection(port int) {
 	Expect(err).ToNot(HaveOccurred())
 }
 
+// startNCContainer starts a detached container running nc (netcat) listening
+// on the given port, waits for it to be ready, and returns the container name.
+//
+//nolint:unused,nolintlint // only called from linux-only test files, unused on freebsd
+func (p *PodmanTestIntegration) startNCContainer(name string, listenPort int, extraArgs ...string) string {
+	GinkgoHelper()
+	portStr := strconv.Itoa(listenPort)
+	args := append([]string{"run", "-d", "--name", name}, extraArgs...)
+	args = append(args, ALPINE, "sh", "-c", "nc -l -n -v -p "+portStr+" 2>&1")
+	p.PodmanExitCleanly(args...)
+	p.WaitForContainerLog(name, "listening")
+	return name
+}
+
+// WaitForContainerLog polls container logs until the given substring appears
+// in either stdout or stderr. Fails the test if not found within the timeout.
+func (p *PodmanTestIntegration) WaitForContainerLog(ctrName string, substr string) {
+	GinkgoHelper()
+	for range 10 {
+		logs := p.Podman([]string{"logs", ctrName})
+		logs.WaitWithDefaultTimeout()
+		if strings.Contains(logs.ErrorToString(), substr) || strings.Contains(logs.OutputToString(), substr) {
+			return
+		}
+		time.Sleep(500 * time.Millisecond)
+	}
+	Fail(fmt.Sprintf("timed out waiting for %q in logs of container %s", substr, ctrName))
+}
+
 func createNetworkName(name string) string {
 	return name + stringid.GenerateRandomID()[:10]
 }