machine/linux: Support virtiofs mounts (retain 9p default)

I'm hitting a bug with 9p when trying to transfer large files.
In RHEL at least 9p isn't supported because it's known to have a
lot of design flaws; virtiofsd is the supported and recommended
way to share files between a host and guest.

Add a new hidden `PODMAN_MACHINE_VIRTFS` environment
variable  that can be set to `virtiofs` to switch to virtiofsd.

Signed-off-by: Colin Walters <walters@verbum.org>

pkg/machine/qemu/command/qemu_command_test.go

@@ -53,6 +53,8 @@ func TestQemuCmd(t *testing.T) {
 
 	expected := []string{
 		"/usr/bin/qemu-system-x86_64",
+		"-object",
+		"memory-backend-memfd,id=mem,size=2048M,share=on",
 		"-m", "2048",
 		"-smp", "4",
 		"-fw_cfg", fmt.Sprintf("name=opt/com.coreos/config,file=%s", ignPath),

pkg/machine/qemu/machine.go

@@ -25,9 +25,28 @@ import (
 )
 
 const (
-	MountType9p = "9p"
+	MountType9p       = "9p"
+	MountTypeVirtiofs = "virtiofs"
 )
 
+func NewStubber() (*QEMUStubber, error) {
+	var mountType string
+	if v, ok := os.LookupEnv("PODMAN_MACHINE_VIRTFS"); ok {
+		logrus.Debugf("using virtiofs %q", v)
+		switch v {
+		case MountType9p, MountTypeVirtiofs:
+			mountType = v
+		default:
+			return nil, fmt.Errorf("failed to parse PODMAN_MACHINE_VIRTFS=%s", v)
+		}
+	} else {
+		mountType = MountType9p
+	}
+	return &QEMUStubber{
+		mountType: mountType,
+	}, nil
+}
+
 // qemuPid returns -1 or the PID of the running QEMU instance.
 func qemuPid(pidFile *define.VMFile) (int, error) {
 	pidData, err := os.ReadFile(pidFile.GetPath())

pkg/machine/qemu/stubber.go

@@ -31,6 +31,11 @@ type QEMUStubber struct {
 	vmconfigs.QEMUConfig
 	// Command describes the final QEMU command line
 	Command command.QemuCmd
+
+	// mountType is one of virtiofs or 9p
+	mountType string
+	// virtiofsHelpers are virtiofsd child processes
+	virtiofsHelpers []virtiofsdHelperCmd
 }
 
 var (
@@ -83,13 +88,6 @@ func (q *QEMUStubber) setQEMUCommandLine(mc *vmconfigs.MachineConfig) error {
 	}
 	q.Command.SetSerialPort(*readySocket, *mc.QEMUHypervisor.QEMUPidPath, mc.Name)
 
-	// Add volumes to qemu command line
-	for _, mount := range mc.Mounts {
-		// the index provided in this case is thrown away
-		_, _, _, _, securityModel := vmconfigs.SplitVolume(0, mount.OriginalInput)
-		q.Command.SetVirtfsMount(mount.Source, mount.Tag, securityModel, mount.ReadOnly)
-	}
-
 	q.Command.SetUSBHostPassthrough(mc.Resources.USBs)
 
 	return nil
@@ -168,6 +166,33 @@ func (q *QEMUStubber) StartVM(mc *vmconfigs.MachineConfig) (func() error, func()
 	defer dnr.Close()
 	defer dnw.Close()
 
+	if q.mountType == MountTypeVirtiofs {
+		runtime, err := mc.RuntimeDir()
+		if err != nil {
+			return nil, nil, err
+		}
+		spawner, err := newVirtiofsdSpawner(runtime)
+		if err != nil {
+			return nil, nil, err
+		}
+
+		for _, hostmnt := range mc.Mounts {
+			qemuArgs, virtiofsdHelper, err := spawner.spawnForMount(hostmnt)
+			if err != nil {
+				return nil, nil, fmt.Errorf("failed to init virtiofsd for mount %s: %w", hostmnt.Source, err)
+			}
+			q.Command = append(q.Command, qemuArgs...)
+			q.virtiofsHelpers = append(q.virtiofsHelpers, *virtiofsdHelper)
+		}
+	} else {
+		// Add volumes to qemu command line
+		for _, mount := range mc.Mounts {
+			// the index provided in this case is thrown away
+			_, _, _, _, securityModel := vmconfigs.SplitVolume(0, mount.OriginalInput)
+			q.Command.SetVirtfsMount(mount.Source, mount.Tag, securityModel, mount.ReadOnly)
+		}
+	}
+
 	cmdLine := q.Command
 
 	// Disable graphic window when not in debug mode
@@ -198,8 +223,20 @@ func (q *QEMUStubber) StartVM(mc *vmconfigs.MachineConfig) (func() error, func()
 		return waitForReady(readySocket, cmd.Process.Pid, stderrBuf)
 	}
 
+	releaseFunc := func() error {
+		if err := cmd.Process.Release(); err != nil {
+			return err
+		}
+		for _, virtiofsdCmd := range q.virtiofsHelpers {
+			if err := virtiofsdCmd.command.Process.Release(); err != nil {
+				return err
+			}
+		}
+		return nil
+	}
+
 	// if this is not the last line in the func, make it a defer
-	return cmd.Process.Release, readyFunc, nil
+	return releaseFunc, readyFunc, nil
 }
 
 func waitForReady(readySocket *define.VMFile, pid int, stdErrBuffer *bytes.Buffer) error {
@@ -325,7 +362,9 @@ func (q *QEMUStubber) MountVolumesToVM(mc *vmconfigs.MachineConfig, quiet bool)
 		if err != nil {
 			return err
 		}
-		switch mount.Type {
+		// NOTE: We ignore q.Type here (and it should eventually be dropped from being serialized); we
+		// want the mount type to be dynamic, not static.
+		switch q.mountType {
 		case MountType9p:
 			mountOptions := []string{"-t", "9p"}
 			mountOptions = append(mountOptions, []string{"-o", "trans=virtio", mount.Tag, mount.Target}...)
@@ -337,6 +376,18 @@ func (q *QEMUStubber) MountVolumesToVM(mc *vmconfigs.MachineConfig, quiet bool)
 			if err != nil {
 				return err
 			}
+		case MountTypeVirtiofs:
+			mountOptions := []string{"-t", "virtiofs"}
+			mountOptions = append(mountOptions, []string{mount.Tag, mount.Target}...)
+			mountFlags := fmt.Sprintf("context=\"%s\"", machine.NFSSELinuxContext)
+			if mount.ReadOnly {
+				mountFlags += ",ro"
+			}
+			mountOptions = append(mountOptions, "-o", mountFlags)
+			err = machine.CommonSSH(mc.SSH.RemoteUsername, mc.SSH.IdentityPath, mc.Name, mc.SSH.Port, append([]string{"sudo", "mount"}, mountOptions...))
+			if err != nil {
+				return err
+			}
 		default:
 			return fmt.Errorf("unknown mount type: %s", mount.Type)
 		}
@@ -345,7 +396,14 @@ func (q *QEMUStubber) MountVolumesToVM(mc *vmconfigs.MachineConfig, quiet bool)
 }
 
 func (q *QEMUStubber) MountType() vmconfigs.VolumeMountType {
-	return vmconfigs.NineP
+	switch q.mountType {
+	case MountType9p:
+		return vmconfigs.NineP
+	case MountTypeVirtiofs:
+		return vmconfigs.VirtIOFS
+	default:
+		panic(fmt.Errorf("unexpected mount type %q", q.mountType))
+	}
 }
 
 func (q *QEMUStubber) PostStartNetworking(mc *vmconfigs.MachineConfig, noInfo bool) error {

pkg/machine/qemu/virtiofsd.go

@@ -0,0 +1,99 @@
+//go:build linux || freebsd
+
+package qemu
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"time"
+
+	"github.com/containers/common/pkg/config"
+	"github.com/containers/podman/v5/pkg/machine/define"
+	"github.com/containers/podman/v5/pkg/machine/vmconfigs"
+	"github.com/containers/storage/pkg/fileutils"
+	"github.com/sirupsen/logrus"
+)
+
+// VirtiofsdSpawner spawns an instance of virtiofsd
+type virtiofsdSpawner struct {
+	runtimeDir *define.VMFile
+	binaryPath string
+	mountCount uint
+}
+
+func newVirtiofsdSpawner(runtimeDir *define.VMFile) (*virtiofsdSpawner, error) {
+	var binaryPath string
+	cfg, err := config.Default()
+	if err != nil {
+		return nil, err
+	}
+	binaryPath, err = cfg.FindHelperBinary("virtiofsd", true)
+	if err != nil {
+		return nil, fmt.Errorf("failed to find virtiofsd: %w", err)
+	}
+	return &virtiofsdSpawner{binaryPath: binaryPath, runtimeDir: runtimeDir}, nil
+}
+
+// createVirtiofsCmd returns a new command instance configured to launch virtiofsd.
+func (v *virtiofsdSpawner) createVirtiofsCmd(directory, socketPath string) *exec.Cmd {
+	args := []string{"--sandbox", "none", "--socket-path", socketPath, "--shared-dir", "."}
+	// We don't need seccomp filtering; we trust our workloads. This incidentally
+	// works around issues like https://gitlab.com/virtio-fs/virtiofsd/-/merge_requests/200.
+	args = append(args, "--seccomp=none")
+	cmd := exec.Command(v.binaryPath, args...)
+	// This sets things up so that the `.` we passed in the arguments is the target directory
+	cmd.Dir = directory
+	// Quiet the daemon by default
+	cmd.Env = append(cmd.Environ(), "RUST_LOG=ERROR")
+	cmd.Stdin = nil
+	cmd.Stdout = nil
+	if logrus.IsLevelEnabled(logrus.DebugLevel) {
+		cmd.Stderr = os.Stderr
+	}
+	return cmd
+}
+
+type virtiofsdHelperCmd struct {
+	command exec.Cmd
+	socket  *define.VMFile
+}
+
+// spawnForMount returns on success a combination of qemu commandline and child process for virtiofsd
+func (v *virtiofsdSpawner) spawnForMount(hostmnt *vmconfigs.Mount) ([]string, *virtiofsdHelperCmd, error) {
+	logrus.Debugf("Initializing virtiofsd mount for %s", hostmnt.Source)
+	// By far the most common failure to spawn virtiofsd will be a typo'd source directory,
+	// so let's synchronously check that ourselves here.
+	if err := fileutils.Exists(hostmnt.Source); err != nil {
+		return nil, nil, fmt.Errorf("failed to access virtiofs source directory %s", hostmnt.Source)
+	}
+	virtiofsChar := fmt.Sprintf("virtiofschar%d", v.mountCount)
+	virtiofsCharPath, err := v.runtimeDir.AppendToNewVMFile(virtiofsChar, nil)
+	if err != nil {
+		return nil, nil, err
+	}
+
+	qemuCommand := []string{}
+
+	qemuCommand = append(qemuCommand, "-chardev", fmt.Sprintf("socket,id=%s,path=%s", virtiofsChar, virtiofsCharPath.Path))
+	qemuCommand = append(qemuCommand, "-device", fmt.Sprintf("vhost-user-fs-pci,queue-size=1024,chardev=%s,tag=%s", virtiofsChar, hostmnt.Tag))
+	// TODO: Honor hostmnt.readonly somehow here (add an option to virtiofsd)
+	virtiofsdCmd := v.createVirtiofsCmd(hostmnt.Source, virtiofsCharPath.Path)
+	if err := virtiofsdCmd.Start(); err != nil {
+		return nil, nil, fmt.Errorf("failed to start virtiofsd")
+	}
+	// Wait for the socket
+	for {
+		if err := fileutils.Exists(virtiofsCharPath.Path); err == nil {
+			break
+		}
+		logrus.Debugf("waiting for virtiofsd socket %q", virtiofsCharPath.Path)
+		time.Sleep(time.Millisecond * 100)
+	}
+	// Increment our count of mounts which are used to create unique names for the devices
+	v.mountCount += 1
+	return qemuCommand, &virtiofsdHelperCmd{
+		command: *virtiofsdCmd,
+		socket:  virtiofsCharPath,
+	}, nil
+}