Merge pull request #2515 from devimc/topic/pmem-CSI

Support persistent memory volumes

Gopkg.lock

@@ -412,11 +412,11 @@
   revision = "2f1d1f20f75d5404f53b9edf6b53ed5505508675"
 
 [[projects]]
-  digest = "1:0044fb81f517f480ca3c33675a3af6b4ada77a6faf699a302bc2388c98cacba9"
+  digest = "1:fafdb4aa5b6207f51ec7557818d5f7a534ed44ea4fb31c6f2e8abb01d1627b74"
   name = "github.com/intel/govmm"
   packages = ["qemu"]
   pruneopts = "NUT"
-  revision = "3700c55dd766d37e17af354fb9975dc801619d62"
+  revision = "e969afbec52cf687bbe97b76654c664128cdb04b"
 
 [[projects]]
   digest = "1:d6e9b99fe0150d4c26d81612676e8d59ad045642e4cbc8646e494b50d4f245ef"

Gopkg.toml

@@ -48,7 +48,7 @@
 
 [[constraint]]
   name = "github.com/intel/govmm"
-  revision = "3700c55dd766d37e17af354fb9975dc801619d62"
+  revision = "e969afbec52cf687bbe97b76654c664128cdb04b"
 
 [[constraint]]
   name = "github.com/kata-containers/agent"

vendor/github.com/intel/govmm/qemu/qmp.go

@@ -1428,8 +1428,9 @@ func (q *QMP) ExecHotplugMemory(ctx context.Context, qomtype, id, mempath string
 // a NVDIMM driver with the device_add command.
 // id is the id of the device to add.  It must be a valid QMP identifier.
 // mempath is the path of the device to add, e.g., /dev/rdb0.  size is
-// the data size of the device.
-func (q *QMP) ExecuteNVDIMMDeviceAdd(ctx context.Context, id, mempath string, size int64) error {
+// the data size of the device. pmem is to guarantee the persistence of QEMU writes
+// to the vNVDIMM backend.
+func (q *QMP) ExecuteNVDIMMDeviceAdd(ctx context.Context, id, mempath string, size int64, pmem *bool) error {
 	args := map[string]interface{}{
 		"qom-type": "memory-backend-file",
 		"id":       "nvdimmbackmem" + id,
@@ -1439,6 +1440,14 @@ func (q *QMP) ExecuteNVDIMMDeviceAdd(ctx context.Context, id, mempath string, si
 			"share":    true,
 		},
 	}
+
+	if q.version.Major > 4 || (q.version.Major == 4 && q.version.Minor >= 1) {
+		if pmem != nil {
+			props := args["props"].(map[string]interface{})
+			props["pmem"] = *pmem
+		}
+	}
+
 	err := q.executeCommand(ctx, "object-add", args, nil)
 	if err != nil {
 		return err

virtcontainers/container.go

@@ -643,6 +643,11 @@ func filterDevices(c *Container, devices []ContainerDevice) (ret []ContainerDevi
 }
 
 func (c *Container) createBlockDevices() error {
+	if !c.checkBlockDeviceSupport() {
+		c.Logger().Warn("Block device not supported")
+		return nil
+	}
+
 	// iterate all mounts and create block device if it's block based.
 	for i, m := range c.mounts {
 		if len(m.BlockDeviceID) > 0 || m.Type != "bind" {
@@ -657,18 +662,36 @@ func (c *Container) createBlockDevices() error {
 			return fmt.Errorf("stat %q failed: %v", m.Source, err)
 		}
 
+		var di *config.DeviceInfo
+		var err error
+
 		// Check if mount is a block device file. If it is, the block device will be attached to the host
 		// instead of passing this as a shared mount.
-		if c.checkBlockDeviceSupport() && stat.Mode&unix.S_IFBLK == unix.S_IFBLK {
-			b, err := c.sandbox.devManager.NewDevice(config.DeviceInfo{
+		if stat.Mode&unix.S_IFBLK == unix.S_IFBLK {
+			di = &config.DeviceInfo{
 				HostPath:      m.Source,
 				ContainerPath: m.Destination,
 				DevType:       "b",
 				Major:         int64(unix.Major(stat.Rdev)),
 				Minor:         int64(unix.Minor(stat.Rdev)),
-			})
+			}
+			// check whether source can be used as a pmem device
+		} else if di, err = config.PmemDeviceInfo(m.Source, m.Destination); err != nil {
+			c.Logger().WithError(err).
+				WithField("mount-source", m.Source).
+				Debug("no loop device")
+		}
+
+		if err == nil && di != nil {
+			b, err := c.sandbox.devManager.NewDevice(*di)
+
 			if err != nil {
-				return fmt.Errorf("device manager failed to create new device for %q: %v", m.Source, err)
+				// Do not return an error, try to create
+				// devices for other mounts
+				c.Logger().WithError(err).WithField("mount-source", m.Source).
+					Error("device manager failed to create new device")
+				continue
+
 			}
 
 			c.mounts[i].BlockDeviceID = b.DeviceID()
@@ -1323,7 +1346,7 @@ func (c *Container) hotplugDrive() error {
 			c.rootfsSuffix = ""
 		}
 		// If device mapper device, then fetch the full path of the device
-		devicePath, fsType, err = GetDevicePathAndFsType(dev.mountPoint)
+		devicePath, fsType, err = utils.GetDevicePathAndFsType(dev.mountPoint)
 		if err != nil {
 			return err
 		}

virtcontainers/device/config/config.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"strconv"
+	"strings"
 
 	"github.com/go-ini/ini"
 	"golang.org/x/sys/unix"
@@ -91,6 +92,8 @@ var SysBusPciDevicesPath = "/sys/bus/pci/devices"
 // SysBusPciSlotsPath is static string of /sys/bus/pci/slots
 var SysBusPciSlotsPath = "/sys/bus/pci/slots"
 
+var getSysDevPath = getSysDevPathImpl
+
 // DeviceInfo is an embedded type that contains device data common to all types of devices.
 type DeviceInfo struct {
 	// Hostpath is device path on host
@@ -110,6 +113,10 @@ type DeviceInfo struct {
 	Major int64
 	Minor int64
 
+	// Pmem enabled persistent memory. Use HostPath as backing file
+	// for a nvdimm device in the guest.
+	Pmem bool
+
 	// FileMode permission bits for the device.
 	FileMode os.FileMode
 
@@ -166,6 +173,10 @@ type BlockDrive struct {
 
 	// ReadOnly sets the device file readonly
 	ReadOnly bool
+
+	// Pmem enables persistent memory. Use File as backing file
+	// for a nvdimm device in the guest
+	Pmem bool
 }
 
 // VFIODeviceType indicates VFIO device type
@@ -257,29 +268,14 @@ func GetHostPath(devInfo DeviceInfo, vhostUserStoreEnabled bool, vhostUserStoreP
 		return "", fmt.Errorf("Empty path provided for device")
 	}
 
-	var pathComp string
-
-	switch devInfo.DevType {
-	case "c", "u":
-		pathComp = "char"
-	case "b":
-		pathComp = "block"
-	default:
-		// Unsupported device types. Return nil error to ignore devices
-		// that cannot be handled currently.
-		return "", nil
-	}
-
 	// Filter out vhost-user storage devices by device Major numbers.
 	if vhostUserStoreEnabled && devInfo.DevType == "b" &&
 		(devInfo.Major == VhostUserSCSIMajor || devInfo.Major == VhostUserBlkMajor) {
 		return getVhostUserHostPath(devInfo, vhostUserStorePath)
 	}
 
-	format := strconv.FormatInt(devInfo.Major, 10) + ":" + strconv.FormatInt(devInfo.Minor, 10)
-	sysDevPath := filepath.Join(SysDevPrefix, pathComp, format, "uevent")
-
-	if _, err := os.Stat(sysDevPath); err != nil {
+	ueventPath := filepath.Join(getSysDevPath(devInfo), "uevent")
+	if _, err := os.Stat(ueventPath); err != nil {
 		// Some devices(eg. /dev/fuse, /dev/cuse) do not always implement sysfs interface under /sys/dev
 		// These devices are passed by default by docker.
 		//
@@ -293,7 +289,7 @@ func GetHostPath(devInfo DeviceInfo, vhostUserStoreEnabled bool, vhostUserStoreP
 		return "", err
 	}
 
-	content, err := ini.Load(sysDevPath)
+	content, err := ini.Load(ueventPath)
 	if err != nil {
 		return "", err
 	}
@@ -306,6 +302,35 @@ func GetHostPath(devInfo DeviceInfo, vhostUserStoreEnabled bool, vhostUserStoreP
 	return filepath.Join("/dev", devName.String()), nil
 }
 
+// getBackingFile is used to fetch the backing file for the device.
+func getBackingFile(devInfo DeviceInfo) (string, error) {
+	backingFilePath := filepath.Join(getSysDevPath(devInfo), "loop", "backing_file")
+	data, err := ioutil.ReadFile(backingFilePath)
+	if err != nil {
+		return "", err
+	}
+
+	return strings.TrimSpace(string(data)), nil
+}
+
+func getSysDevPathImpl(devInfo DeviceInfo) string {
+	var pathComp string
+
+	switch devInfo.DevType {
+	case "c", "u":
+		pathComp = "char"
+	case "b":
+		pathComp = "block"
+	default:
+		// Unsupported device types. Return nil error to ignore devices
+		// that cannot be handled currently.
+		return ""
+	}
+
+	format := strconv.FormatInt(devInfo.Major, 10) + ":" + strconv.FormatInt(devInfo.Minor, 10)
+	return filepath.Join(SysDevPrefix, pathComp, format)
+}
+
 // getVhostUserHostPath is used to fetch host path for the vhost-user device.
 // For vhost-user block device like vhost-user-blk or vhost-user-scsi, its
 // socket should be under directory "<vhostUserStorePath>/block/sockets/";

virtcontainers/device/config/config_test.go

@@ -0,0 +1,73 @@
+// Copyright (c) 2020 Intel Corporation
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+package config
+
+import (
+	"fmt"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestGetBackingFile(t *testing.T) {
+	assert := assert.New(t)
+
+	dir, err := ioutil.TempDir("", "backing")
+	assert.NoError(err)
+	defer os.RemoveAll(dir)
+
+	orgGetSysDevPath := getSysDevPath
+	getSysDevPath = func(info DeviceInfo) string {
+		return dir
+	}
+	defer func() { getSysDevPath = orgGetSysDevPath }()
+
+	info := DeviceInfo{}
+	path, err := getBackingFile(info)
+	assert.Error(err)
+	assert.Empty(path)
+
+	loopDir := filepath.Join(dir, "loop")
+	err = os.Mkdir(loopDir, os.FileMode(0755))
+	assert.NoError(err)
+
+	backingFile := "/fake-img"
+
+	err = ioutil.WriteFile(filepath.Join(loopDir, "backing_file"), []byte(backingFile), os.FileMode(0755))
+	assert.NoError(err)
+
+	path, err = getBackingFile(info)
+	assert.NoError(err)
+	assert.Equal(backingFile, path)
+}
+
+func TestGetSysDevPathImpl(t *testing.T) {
+	assert := assert.New(t)
+
+	info := DeviceInfo{
+		DevType: "",
+		Major:   127,
+		Minor:   0,
+	}
+
+	path := getSysDevPathImpl(info)
+	assert.Empty(path)
+
+	expectedFormat := fmt.Sprintf("%d:%d", info.Major, info.Minor)
+
+	info.DevType = "c"
+	path = getSysDevPathImpl(info)
+	assert.Contains(path, expectedFormat)
+	assert.Contains(path, "char")
+
+	info.DevType = "b"
+	path = getSysDevPathImpl(info)
+	assert.Contains(path, expectedFormat)
+	assert.Contains(path, "block")
+}

virtcontainers/device/config/pmem.go

@@ -0,0 +1,116 @@
+// Copyright (c) 2020 Intel Corporation
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+package config
+
+import (
+	"fmt"
+	"os"
+	"syscall"
+
+	"github.com/kata-containers/runtime/virtcontainers/utils"
+	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
+)
+
+const (
+	// This signature is defined in the linux NVDIMM driver.
+	// devices or backing files with this signature can be used
+	// as pmem (persistent memory) devices in the guest.
+	pfnSignature = "NVDIMM_PFN_INFO"
+
+	// offset in the backing file where the signature should be
+	pfnSignatureOffset = int64(4 * 1024)
+)
+
+var (
+	pmemLog = logrus.WithField("source", "virtcontainers/device/config")
+)
+
+// PmemDeviceInfo returns a DeviceInfo if a loop device
+// is mounted on source, and the backing file of the loop device
+// has the PFN signature.
+func PmemDeviceInfo(source, destination string) (*DeviceInfo, error) {
+	stat := syscall.Stat_t{}
+	err := syscall.Stat(source, &stat)
+	if err != nil {
+		return nil, err
+	}
+
+	// device object is still incomplete,
+	// but it can be used to fetch the backing file
+	device := &DeviceInfo{
+		ContainerPath: destination,
+		DevType:       "b",
+		Major:         int64(unix.Major(stat.Dev)),
+		Minor:         int64(unix.Minor(stat.Dev)),
+		Pmem:          true,
+		DriverOptions: make(map[string]string),
+	}
+
+	pmemLog.WithFields(
+		logrus.Fields{
+			"major": device.Major,
+			"minor": device.Minor,
+		}).Debug("looking for backing file")
+
+	device.HostPath, err = getBackingFile(*device)
+	if err != nil {
+		return nil, err
+	}
+
+	pmemLog.WithField("backing-file", device.HostPath).
+		Debug("backing file found: looking for PFN signature")
+
+	if !hasPFNSignature(device.HostPath) {
+		return nil, fmt.Errorf("backing file %v has not PFN signature", device.HostPath)
+	}
+
+	_, fstype, err := utils.GetDevicePathAndFsType(source)
+	if err != nil {
+		pmemLog.WithError(err).WithField("mount-point", source).Warn("failed to get fstype: using ext4")
+		fstype = "ext4"
+	}
+
+	pmemLog.WithField("fstype", fstype).Debug("filesystem for mount point")
+	device.DriverOptions["fstype"] = fstype
+
+	return device, nil
+}
+
+// returns true if the file/device path has the PFN signature
+// required to use it as PMEM device and enable DAX.
+// See [1] to know more about the PFN signature.
+//
+// [1] - https://github.com/kata-containers/osbuilder/blob/master/image-builder/nsdax.gpl.c
+func hasPFNSignature(path string) bool {
+	f, err := os.Open(path)
+	if err != nil {
+		pmemLog.WithError(err).Error("Could not get PFN signature")
+		return false
+	}
+	defer f.Close()
+
+	signatureLen := len(pfnSignature)
+	signature := make([]byte, signatureLen)
+
+	l, err := f.ReadAt(signature, pfnSignatureOffset)
+	if err != nil {
+		pmemLog.WithError(err).Debug("Could not read pfn signature")
+		return false
+	}
+
+	pmemLog.WithFields(logrus.Fields{
+		"path":      path,
+		"signature": string(signature),
+	}).Debug("got signature")
+
+	if l != signatureLen {
+		pmemLog.WithField("read-bytes", l).Debug("Incomplete signature")
+		return false
+	}
+
+	return pfnSignature == string(signature)
+}

virtcontainers/device/config/pmem_test.go

@@ -0,0 +1,49 @@
+// Copyright (c) 2020 Intel Corporation
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+package config
+
+import (
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func createPFNFile(assert *assert.Assertions, dir string) string {
+	pfnPath := filepath.Join(dir, "pfn")
+	file, err := os.Create(pfnPath)
+	assert.NoError(err)
+	defer file.Close()
+
+	l, err := file.WriteAt([]byte(pfnSignature), pfnSignatureOffset)
+	assert.NoError(err)
+	assert.Equal(len(pfnSignature), l)
+
+	return pfnPath
+}
+
+func TestHasPFNSignature(t *testing.T) {
+	assert := assert.New(t)
+
+	b := hasPFNSignature("/abc/xyz/123/sw")
+	assert.False(b)
+
+	f, err := ioutil.TempFile("", "pfn")
+	assert.NoError(err)
+	f.Close()
+	defer os.Remove(f.Name())
+
+	b = hasPFNSignature(f.Name())
+	assert.False(b)
+
+	pfnFile := createPFNFile(assert, os.TempDir())
+	defer os.Remove(pfnFile)
+
+	b = hasPFNSignature(pfnFile)
+	assert.True(b)
+}

virtcontainers/device/drivers/block.go

@@ -65,6 +65,11 @@ func (device *BlockDevice) Attach(devReceiver api.DeviceReceiver) (err error) {
 		Format: "raw",
 		ID:     utils.MakeNameID("drive", device.DeviceInfo.ID, maxDevIDSize),
 		Index:  index,
+		Pmem:   device.DeviceInfo.Pmem,
+	}
+
+	if fs, ok := device.DeviceInfo.DriverOptions["fstype"]; ok {
+		drive.Format = fs
 	}
 
 	customOptions := device.DeviceInfo.DriverOptions
@@ -169,6 +174,7 @@ func (device *BlockDevice) Save() persistapi.DeviceState {
 			NvdimmID: drive.NvdimmID,
 			VirtPath: drive.VirtPath,
 			DevNo:    drive.DevNo,
+			Pmem:     drive.Pmem,
 		}
 	}
 	return ds
@@ -194,6 +200,7 @@ func (device *BlockDevice) Load(ds persistapi.DeviceState) {
 		NvdimmID: bd.NvdimmID,
 		VirtPath: bd.VirtPath,
 		DevNo:    bd.DevNo,
+		Pmem:     bd.Pmem,
 	}
 }
 

virtcontainers/device/manager/manager.go

@@ -98,11 +98,15 @@ func (dm *deviceManager) findDeviceByMajorMinor(major, minor int64) api.Device {
 
 // createDevice creates one device based on DeviceInfo
 func (dm *deviceManager) createDevice(devInfo config.DeviceInfo) (dev api.Device, err error) {
-	path, err := config.GetHostPathFunc(devInfo, dm.vhostUserStoreEnabled, dm.vhostUserStorePath)
-	if err != nil {
-		return nil, err
+	// pmem device may points to block devices or raw files,
+	// do not change its HostPath.
+	if !devInfo.Pmem {
+		path, err := config.GetHostPathFunc(devInfo, dm.vhostUserStoreEnabled, dm.vhostUserStorePath)
+		if err != nil {
+			return nil, err
+		}
+		devInfo.HostPath = path
 	}
-	devInfo.HostPath = path
 
 	defer func() {
 		if err == nil {
@@ -119,7 +123,7 @@ func (dm *deviceManager) createDevice(devInfo config.DeviceInfo) (dev api.Device
 	if devInfo.ID, err = dm.newDeviceID(); err != nil {
 		return nil, err
 	}
-	if isVFIO(path) {
+	if isVFIO(devInfo.HostPath) {
 		return drivers.NewVFIODevice(&devInfo), nil
 	} else if isVhostUserBlk(devInfo) {
 		if devInfo.DriverOptions == nil {
@@ -134,7 +138,7 @@ func (dm *deviceManager) createDevice(devInfo config.DeviceInfo) (dev api.Device
 		devInfo.DriverOptions["block-driver"] = dm.blockDriver
 		return drivers.NewBlockDevice(&devInfo), nil
 	} else {
-		deviceLogger().WithField("device", path).Info("Device has not been passed to the container")
+		deviceLogger().WithField("device", devInfo.HostPath).Info("Device has not been passed to the container")
 		return drivers.NewGenericDevice(&devInfo), nil
 	}
 }

virtcontainers/kata_agent.go

@@ -1114,6 +1114,13 @@ func (k *kataAgent) appendBlockDevice(dev ContainerDevice, c *Container) *grpc.D
 		return nil
 	}
 
+	if d.Pmem {
+		// block drive is a persistent memory device that
+		// was passed as volume (-v) not as device (--device).
+		// It shouldn't be visible in the container
+		return nil
+	}
+
 	kataDevice := &grpc.Device{
 		ContainerPath: dev.ContainerPath,
 	}
@@ -1476,6 +1483,12 @@ func (k *kataAgent) handleDeviceBlockVolume(c *Container, device api.Device) (*g
 		return nil, fmt.Errorf("malformed block drive")
 	}
 	switch {
+	// pmem volumes case
+	case blockDrive.Pmem:
+		vol.Driver = kataNvdimmDevType
+		vol.Source = fmt.Sprintf("/dev/pmem%s", blockDrive.NvdimmID)
+		vol.Fstype = blockDrive.Format
+		vol.Options = []string{"dax"}
 	case c.sandbox.config.HypervisorConfig.BlockDeviceDriver == config.VirtioBlockCCW:
 		vol.Driver = kataBlkCCWDevType
 		vol.Source = blockDrive.DevNo
@@ -1553,8 +1566,12 @@ func (k *kataAgent) handleBlockVolumes(c *Container) ([]*grpc.Storage, error) {
 		}
 
 		vol.MountPoint = m.Destination
-		vol.Fstype = "bind"
-		vol.Options = []string{"bind"}
+		if vol.Fstype == "" {
+			vol.Fstype = "bind"
+		}
+		if len(vol.Options) == 0 {
+			vol.Options = []string{"bind"}
+		}
 
 		volumeStorages = append(volumeStorages, vol)
 	}

virtcontainers/mount.go

@@ -6,17 +6,16 @@
 package virtcontainers
 
 import (
-	"bufio"
 	"context"
 	"errors"
 	"fmt"
-	"io"
 	"os"
 	"path/filepath"
 	"strings"
 	"syscall"
 
 	merr "github.com/hashicorp/go-multierror"
+	"github.com/kata-containers/runtime/virtcontainers/utils"
 	"github.com/sirupsen/logrus"
 )
 
@@ -187,59 +186,6 @@ func getDeviceForPath(path string) (device, error) {
 	return dev, nil
 }
 
-const (
-	procMountsFile = "/proc/mounts"
-
-	fieldsPerLine = 6
-)
-
-const (
-	procDeviceIndex = iota
-	procPathIndex
-	procTypeIndex
-)
-
-// GetDevicePathAndFsType gets the device for the mount point and the file system type
-// of the mount.
-func GetDevicePathAndFsType(mountPoint string) (devicePath, fsType string, err error) {
-	if mountPoint == "" {
-		err = fmt.Errorf("Mount point cannot be empty")
-		return
-	}
-
-	var file *os.File
-
-	file, err = os.Open(procMountsFile)
-	if err != nil {
-		return
-	}
-
-	defer file.Close()
-
-	reader := bufio.NewReader(file)
-	for {
-		var line string
-
-		line, err = reader.ReadString('\n')
-		if err == io.EOF {
-			err = fmt.Errorf("Mount %s not found", mountPoint)
-			return
-		}
-
-		fields := strings.Fields(line)
-		if len(fields) != fieldsPerLine {
-			err = fmt.Errorf("Incorrect no of fields (expected %d, got %d)) :%s", fieldsPerLine, len(fields), line)
-			return
-		}
-
-		if mountPoint == fields[procPathIndex] {
-			devicePath = fields[procDeviceIndex]
-			fsType = fields[procTypeIndex]
-			return
-		}
-	}
-}
-
 var blockFormatTemplate = "/sys/dev/block/%d:%d/dm"
 
 var checkStorageDriver = isDeviceMapper
@@ -445,7 +391,7 @@ func IsEphemeralStorage(path string) bool {
 		return false
 	}
 
-	if _, fsType, _ := GetDevicePathAndFsType(path); fsType == "tmpfs" {
+	if _, fsType, _ := utils.GetDevicePathAndFsType(path); fsType == "tmpfs" {
 		return true
 	}
 
@@ -460,7 +406,7 @@ func Isk8sHostEmptyDir(path string) bool {
 		return false
 	}
 
-	if _, fsType, _ := GetDevicePathAndFsType(path); fsType != "tmpfs" {
+	if _, fsType, _ := utils.GetDevicePathAndFsType(path); fsType != "tmpfs" {
 		return true
 	}
 	return false

virtcontainers/mount_test.go

@@ -206,22 +206,6 @@ func TestGetDeviceForPathBindMount(t *testing.T) {
 	assert.Equal(sourceDev, destDev)
 }
 
-func TestGetDevicePathAndFsTypeEmptyMount(t *testing.T) {
-	assert := assert.New(t)
-	_, _, err := GetDevicePathAndFsType("")
-	assert.Error(err)
-}
-
-func TestGetDevicePathAndFsTypeSuccessful(t *testing.T) {
-	assert := assert.New(t)
-
-	path, fstype, err := GetDevicePathAndFsType("/proc")
-	assert.NoError(err)
-
-	assert.Equal(path, "proc")
-	assert.Equal(fstype, "proc")
-}
-
 func TestIsDeviceMapper(t *testing.T) {
 	assert := assert.New(t)
 

virtcontainers/persist/api/device.go

@@ -41,6 +41,10 @@ type BlockDrive struct {
 
 	// DevNo
 	DevNo string
+
+	// Pmem enabled persistent memory. Use File as backing file
+	// for a nvdimm device in the guest.
+	Pmem bool
 }
 
 // VFIODev represents a VFIO drive used for hotplugging

virtcontainers/qemu.go

@@ -1051,16 +1051,28 @@ func (q *qemu) qmpShutdown() {
 }
 
 func (q *qemu) hotplugAddBlockDevice(drive *config.BlockDrive, op operation, devID string) (err error) {
-	if q.config.BlockDeviceDriver == config.Nvdimm {
+	// drive can be a pmem device, in which case it's used as backing file for a nvdimm device
+	if q.config.BlockDeviceDriver == config.Nvdimm || drive.Pmem {
 		var blocksize int64
 		file, err := os.Open(drive.File)
 		if err != nil {
 			return err
 		}
-		if _, _, err := syscall.Syscall(syscall.SYS_IOCTL, file.Fd(), unix.BLKGETSIZE64, uintptr(unsafe.Pointer(&blocksize))); err != 0 {
+		defer file.Close()
+
+		st, err := file.Stat()
+		if err != nil {
+			return fmt.Errorf("failed to get information from nvdimm device %v: %v", drive.File, err)
+		}
+
+		// regular files do not support syscall BLKGETSIZE64
+		if st.Mode().IsRegular() {
+			blocksize = st.Size()
+		} else if _, _, err := syscall.Syscall(syscall.SYS_IOCTL, file.Fd(), unix.BLKGETSIZE64, uintptr(unsafe.Pointer(&blocksize))); err != 0 {
 			return err
 		}
-		if err = q.qmpMonitorCh.qmp.ExecuteNVDIMMDeviceAdd(q.qmpMonitorCh.ctx, drive.ID, drive.File, blocksize); err != nil {
+
+		if err = q.qmpMonitorCh.qmp.ExecuteNVDIMMDeviceAdd(q.qmpMonitorCh.ctx, drive.ID, drive.File, blocksize, &drive.Pmem); err != nil {
 			q.Logger().WithError(err).Errorf("Failed to add NVDIMM device %s", drive.File)
 			return err
 		}

virtcontainers/utils/utils_linux.go

@@ -6,10 +6,13 @@
 package utils
 
 import (
+	"bufio"
 	"crypto/rand"
 	"fmt"
+	"io"
 	"math/big"
 	"os"
+	"strings"
 	"syscall"
 	"unsafe"
 
@@ -85,3 +88,56 @@ func FindContextID() (*os.File, uint64, error) {
 	vsockFd.Close()
 	return nil, 0, fmt.Errorf("Could not get a unique context ID for the vsock : %s", err)
 }
+
+const (
+	procMountsFile = "/proc/mounts"
+
+	fieldsPerLine = 6
+)
+
+const (
+	procDeviceIndex = iota
+	procPathIndex
+	procTypeIndex
+)
+
+// GetDevicePathAndFsType gets the device for the mount point and the file system type
+// of the mount.
+func GetDevicePathAndFsType(mountPoint string) (devicePath, fsType string, err error) {
+	if mountPoint == "" {
+		err = fmt.Errorf("Mount point cannot be empty")
+		return
+	}
+
+	var file *os.File
+
+	file, err = os.Open(procMountsFile)
+	if err != nil {
+		return
+	}
+
+	defer file.Close()
+
+	reader := bufio.NewReader(file)
+	for {
+		var line string
+
+		line, err = reader.ReadString('\n')
+		if err == io.EOF {
+			err = fmt.Errorf("Mount %s not found", mountPoint)
+			return
+		}
+
+		fields := strings.Fields(line)
+		if len(fields) != fieldsPerLine {
+			err = fmt.Errorf("Incorrect no of fields (expected %d, got %d)) :%s", fieldsPerLine, len(fields), line)
+			return
+		}
+
+		if mountPoint == fields[procPathIndex] {
+			devicePath = fields[procDeviceIndex]
+			fsType = fields[procTypeIndex]
+			return
+		}
+	}
+}

virtcontainers/utils/utils_linux_test.go

@@ -33,3 +33,19 @@ func TestFindContextID(t *testing.T) {
 	assert.Zero(cid)
 	assert.Error(err)
 }
+
+func TestGetDevicePathAndFsTypeEmptyMount(t *testing.T) {
+	assert := assert.New(t)
+	_, _, err := GetDevicePathAndFsType("")
+	assert.Error(err)
+}
+
+func TestGetDevicePathAndFsTypeSuccessful(t *testing.T) {
+	assert := assert.New(t)
+
+	path, fstype, err := GetDevicePathAndFsType("/proc")
+	assert.NoError(err)
+
+	assert.Equal(path, "proc")
+	assert.Equal(fstype, "proc")
+}