node.go

/*
Copyright 2019 The Kubernetes Authors.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package driver

import (
	"context"
	"fmt"
	"os"
	"path/filepath"
	"regexp"
	"strings"

	csi "github.com/container-storage-interface/spec/lib/go/csi"
	"github.com/kubernetes-sigs/aws-ebs-csi-driver/pkg/cloud"
	"github.com/kubernetes-sigs/aws-ebs-csi-driver/pkg/driver/internal"
	"google.golang.org/grpc/codes"
	"google.golang.org/grpc/status"
	"k8s.io/klog"
	"k8s.io/kubernetes/pkg/util/resizefs"
	"k8s.io/utils/exec"
	"k8s.io/utils/mount"
)

const (
	// FSTypeExt2 represents the ext2 filesystem type
	FSTypeExt2 = "ext2"
	// FSTypeExt3 represents the ext3 filesystem type
	FSTypeExt3 = "ext3"
	// FSTypeExt4 represents the ext4 filesystem type
	FSTypeExt4 = "ext4"
	// FSTypeXfs represents te xfs filesystem type
	FSTypeXfs = "xfs"

	// default file system type to be used when it is not provided
	defaultFsType = FSTypeExt4

	// defaultMaxEBSVolumes is the maximum number of volumes that an AWS instance can have attached.
	// More info at https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/volume_limits.html
	defaultMaxEBSVolumes = 39

	// defaultMaxEBSNitroVolumes is the limit of volumes for some smaller instances, like c5 and m5.
	defaultMaxEBSNitroVolumes = 25
)

var (
	ValidFSTypes = []string{FSTypeExt2, FSTypeExt3, FSTypeExt4, FSTypeXfs}
)

var (
	// nodeCaps represents the capability of node service.
	nodeCaps = []csi.NodeServiceCapability_RPC_Type{
		csi.NodeServiceCapability_RPC_STAGE_UNSTAGE_VOLUME,
		csi.NodeServiceCapability_RPC_EXPAND_VOLUME,
	}
)

// nodeService represents the node service of CSI driver
type nodeService struct {
	metadata cloud.MetadataService
	mounter  Mounter
	inFlight *internal.InFlight
}

// newNodeService creates a new node service
// it panics if failed to create the service
func newNodeService() nodeService {
	metadata, err := cloud.NewMetadata()
	if err != nil {
		panic(err)
	}

	return nodeService{
		metadata: metadata,
		mounter:  newNodeMounter(),
		inFlight: internal.NewInFlight(),
	}
}

func (d *nodeService) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRequest) (*csi.NodeStageVolumeResponse, error) {
	klog.V(4).Infof("NodeStageVolume: called with args %+v", *req)

	volumeID := req.GetVolumeId()
	if len(volumeID) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Volume ID not provided")
	}

	target := req.GetStagingTargetPath()
	if len(target) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Staging target not provided")
	}

	volCap := req.GetVolumeCapability()
	if volCap == nil {
		return nil, status.Error(codes.InvalidArgument, "Volume capability not provided")
	}

	if !isValidVolumeCapabilities([]*csi.VolumeCapability{volCap}) {
		return nil, status.Error(codes.InvalidArgument, "Volume capability not supported")
	}

	// If the access type is block, do nothing for stage
	switch volCap.GetAccessType().(type) {
	case *csi.VolumeCapability_Block:
		return &csi.NodeStageVolumeResponse{}, nil
	}

	mount := volCap.GetMount()
	if mount == nil {
		return nil, status.Error(codes.InvalidArgument, "NodeStageVolume: mount is nil within volume capability")
	}

	fsType := mount.GetFsType()
	if len(fsType) == 0 {
		fsType = defaultFsType
	}

	var mountOptions []string
	for _, f := range mount.MountFlags {
		if !hasMountOption(mountOptions, f) {
			mountOptions = append(mountOptions, f)
		}
	}

	if ok := d.inFlight.Insert(req); !ok {
		msg := fmt.Sprintf("request to stage volume=%q is already in progress", volumeID)
		return nil, status.Error(codes.Internal, msg)
	}
	defer func() {
		klog.V(4).Infof("NodeStageVolume: volume=%q operation finished", req.GetVolumeId())
		d.inFlight.Delete(req)
		klog.V(4).Info("donedone")
	}()

	devicePath, ok := req.PublishContext[DevicePathKey]
	if !ok {
		return nil, status.Error(codes.InvalidArgument, "Device path not provided")
	}

	source, err := d.findDevicePath(devicePath, volumeID)
	if err != nil {
		return nil, status.Errorf(codes.Internal, "Failed to find device path %s. %v", devicePath, err)
	}

	klog.V(4).Infof("NodeStageVolume: find device path %s -> %s", devicePath, source)

	exists, err := d.mounter.ExistsPath(target)
	if err != nil {
		msg := fmt.Sprintf("failed to check if target %q exists: %v", target, err)
		return nil, status.Error(codes.Internal, msg)
	}
	// When exists is true it means target path was created but device isn't mounted.
	// We don't want to do anything in that case and let the operation proceed.
	// Otherwise we need to create the target directory.
	if !exists {
		// If target path does not exist we need to create the directory where volume will be staged
		klog.V(4).Infof("NodeStageVolume: creating target dir %q", target)
		if err = d.mounter.MakeDir(target); err != nil {
			msg := fmt.Sprintf("could not create target dir %q: %v", target, err)
			return nil, status.Error(codes.Internal, msg)
		}
	}

	// Check if a device is mounted in target directory
	device, _, err := d.mounter.GetDeviceName(target)
	if err != nil {
		msg := fmt.Sprintf("failed to check if volume is already mounted: %v", err)
		return nil, status.Error(codes.Internal, msg)
	}

	// This operation (NodeStageVolume) MUST be idempotent.
	// If the volume corresponding to the volume_id is already staged to the staging_target_path,
	// and is identical to the specified volume_capability the Plugin MUST reply 0 OK.
	if device == source {
		klog.V(4).Infof("NodeStageVolume: volume=%q already staged", volumeID)
		return &csi.NodeStageVolumeResponse{}, nil
	}

	// FormatAndMount will format only if needed
	klog.V(5).Infof("NodeStageVolume: formatting %s and mounting at %s with fstype %s", source, target, fsType)
	err = d.mounter.FormatAndMount(source, target, fsType, mountOptions)
	if err != nil {
		msg := fmt.Sprintf("could not format %q and mount it at %q", source, target)
		return nil, status.Error(codes.Internal, msg)
	}

	return &csi.NodeStageVolumeResponse{}, nil
}

func (d *nodeService) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstageVolumeRequest) (*csi.NodeUnstageVolumeResponse, error) {
	klog.V(4).Infof("NodeUnstageVolume: called with args %+v", *req)
	volumeID := req.GetVolumeId()
	if len(volumeID) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Volume ID not provided")
	}

	target := req.GetStagingTargetPath()
	if len(target) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Staging target not provided")
	}

	// Check if target directory is a mount point. GetDeviceNameFromMount
	// given a mnt point, finds the device from /proc/mounts
	// returns the device name, reference count, and error code
	dev, refCount, err := d.mounter.GetDeviceName(target)
	if err != nil {
		msg := fmt.Sprintf("failed to check if volume is mounted: %v", err)
		return nil, status.Error(codes.Internal, msg)
	}

	// From the spec: If the volume corresponding to the volume_id
	// is not staged to the staging_target_path, the Plugin MUST
	// reply 0 OK.
	if refCount == 0 {
		klog.V(5).Infof("NodeUnstageVolume: %s target not mounted", target)
		return &csi.NodeUnstageVolumeResponse{}, nil
	}

	if refCount > 1 {
		klog.Warningf("NodeUnstageVolume: found %d references to device %s mounted at target path %s", refCount, dev, target)
	}

	klog.V(5).Infof("NodeUnstageVolume: unmounting %s", target)
	err = d.mounter.Unmount(target)
	if err != nil {
		return nil, status.Errorf(codes.Internal, "Could not unmount target %q: %v", target, err)
	}

	return &csi.NodeUnstageVolumeResponse{}, nil
}

func (d *nodeService) NodeExpandVolume(ctx context.Context, req *csi.NodeExpandVolumeRequest) (*csi.NodeExpandVolumeResponse, error) {
	klog.V(4).Infof("NodeExpandVolume: called with args %+v", *req)
	volumeID := req.GetVolumeId()
	if len(volumeID) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Volume ID not provided")
	}

	args := []string{"-o", "source", "--noheadings", "--target", req.GetVolumePath()}
	output, err := d.mounter.Command("findmnt", args...).Output()
	if err != nil {
		return nil, status.Errorf(codes.Internal, "Could not determine device path: %v", err)

	}

	devicePath := strings.TrimSpace(string(output))
	if len(devicePath) == 0 {
		return nil, status.Errorf(codes.Internal, "Could not get valid device for mount path: %q", req.GetVolumePath())
	}

	// TODO: refactor Mounter to expose a mount.SafeFormatAndMount object
	r := resizefs.NewResizeFs(&mount.SafeFormatAndMount{
		Interface: mount.New(""),
		Exec:      exec.New(),
	})

	// TODO: lock per volume ID to have some idempotency
	if _, err := r.Resize(devicePath, req.GetVolumePath()); err != nil {
		return nil, status.Errorf(codes.Internal, "Could not resize volume %q (%q):  %v", volumeID, devicePath, err)
	}

	return &csi.NodeExpandVolumeResponse{}, nil
}

func (d *nodeService) NodePublishVolume(ctx context.Context, req *csi.NodePublishVolumeRequest) (*csi.NodePublishVolumeResponse, error) {
	klog.V(4).Infof("NodePublishVolume: called with args %+v", *req)
	volumeID := req.GetVolumeId()
	if len(volumeID) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Volume ID not provided")
	}

	source := req.GetStagingTargetPath()
	if len(source) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Staging target not provided")
	}

	target := req.GetTargetPath()
	if len(target) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Target path not provided")
	}

	volCap := req.GetVolumeCapability()
	if volCap == nil {
		return nil, status.Error(codes.InvalidArgument, "Volume capability not provided")
	}

	if !isValidVolumeCapabilities([]*csi.VolumeCapability{volCap}) {
		return nil, status.Error(codes.InvalidArgument, "Volume capability not supported")
	}

	mountOptions := []string{"bind"}
	if req.GetReadonly() {
		mountOptions = append(mountOptions, "ro")
	}

	switch mode := volCap.GetAccessType().(type) {
	case *csi.VolumeCapability_Block:
		if err := d.nodePublishVolumeForBlock(req, mountOptions); err != nil {
			return nil, err
		}
	case *csi.VolumeCapability_Mount:
		if err := d.nodePublishVolumeForFileSystem(req, mountOptions, mode); err != nil {
			return nil, err
		}
	}

	return &csi.NodePublishVolumeResponse{}, nil
}

func (d *nodeService) NodeUnpublishVolume(ctx context.Context, req *csi.NodeUnpublishVolumeRequest) (*csi.NodeUnpublishVolumeResponse, error) {
	klog.V(4).Infof("NodeUnpublishVolume: called with args %+v", *req)
	volumeID := req.GetVolumeId()
	if len(volumeID) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Volume ID not provided")
	}

	target := req.GetTargetPath()
	if len(target) == 0 {
		return nil, status.Error(codes.InvalidArgument, "Target path not provided")
	}

	klog.V(5).Infof("NodeUnpublishVolume: unmounting %s", target)
	err := d.mounter.Unmount(target)
	if err != nil {
		return nil, status.Errorf(codes.Internal, "Could not unmount %q: %v", target, err)
	}

	return &csi.NodeUnpublishVolumeResponse{}, nil
}

func (d *nodeService) NodeGetVolumeStats(ctx context.Context, req *csi.NodeGetVolumeStatsRequest) (*csi.NodeGetVolumeStatsResponse, error) {
	return nil, status.Error(codes.Unimplemented, "NodeGetVolumeStats is not implemented yet")
}

func (d *nodeService) NodeGetCapabilities(ctx context.Context, req *csi.NodeGetCapabilitiesRequest) (*csi.NodeGetCapabilitiesResponse, error) {
	klog.V(4).Infof("NodeGetCapabilities: called with args %+v", *req)
	var caps []*csi.NodeServiceCapability
	for _, cap := range nodeCaps {
		c := &csi.NodeServiceCapability{
			Type: &csi.NodeServiceCapability_Rpc{
				Rpc: &csi.NodeServiceCapability_RPC{
					Type: cap,
				},
			},
		}
		caps = append(caps, c)
	}
	return &csi.NodeGetCapabilitiesResponse{Capabilities: caps}, nil
}

func (d *nodeService) NodeGetInfo(ctx context.Context, req *csi.NodeGetInfoRequest) (*csi.NodeGetInfoResponse, error) {
	klog.V(4).Infof("NodeGetInfo: called with args %+v", *req)

	topology := &csi.Topology{
		Segments: map[string]string{TopologyKey: d.metadata.GetAvailabilityZone()},
	}

	return &csi.NodeGetInfoResponse{
		NodeId:             d.metadata.GetInstanceID(),
		MaxVolumesPerNode:  d.getVolumesLimit(),
		AccessibleTopology: topology,
	}, nil
}

func (d *nodeService) nodePublishVolumeForBlock(req *csi.NodePublishVolumeRequest, mountOptions []string) error {
	target := req.GetTargetPath()
	volumeID := req.GetVolumeId()

	devicePath, exists := req.PublishContext[DevicePathKey]
	if !exists {
		return status.Error(codes.InvalidArgument, "Device path not provided")
	}
	source, err := d.findDevicePath(devicePath, volumeID)
	if err != nil {
		return status.Errorf(codes.Internal, "Failed to find device path %s. %v", devicePath, err)
	}

	klog.V(4).Infof("NodePublishVolume [block]: find device path %s -> %s", devicePath, source)

	globalMountPath := filepath.Dir(target)

	// create the global mount path if it is missing
	// Path in the form of /var/lib/kubelet/plugins/kubernetes.io/csi/volumeDevices/publish/{volumeName}
	exists, err = d.mounter.ExistsPath(globalMountPath)
	if err != nil {
		return status.Errorf(codes.Internal, "Could not check if path exists %q: %v", globalMountPath, err)
	}

	if !exists {
		if err := d.mounter.MakeDir(globalMountPath); err != nil {
			return status.Errorf(codes.Internal, "Could not create dir %q: %v", globalMountPath, err)
		}
	}

	// Create the mount point as a file since bind mount device node requires it to be a file
	klog.V(5).Infof("NodePublishVolume [block]: making target file %s", target)
	err = d.mounter.MakeFile(target)
	if err != nil {
		if removeErr := os.Remove(target); removeErr != nil {
			return status.Errorf(codes.Internal, "Could not remove mount target %q: %v", target, removeErr)
		}
		return status.Errorf(codes.Internal, "Could not create file %q: %v", target, err)
	}

	klog.V(5).Infof("NodePublishVolume [block]: mounting %s at %s", source, target)
	if err := d.mounter.Mount(source, target, "", mountOptions); err != nil {
		if removeErr := os.Remove(target); removeErr != nil {
			return status.Errorf(codes.Internal, "Could not remove mount target %q: %v", target, removeErr)
		}
		return status.Errorf(codes.Internal, "Could not mount %q at %q: %v", source, target, err)
	}

	return nil
}

func (d *nodeService) nodePublishVolumeForFileSystem(req *csi.NodePublishVolumeRequest, mountOptions []string, mode *csi.VolumeCapability_Mount) error {
	target := req.GetTargetPath()
	source := req.GetStagingTargetPath()
	if m := mode.Mount; m != nil {
		for _, f := range m.MountFlags {
			if !hasMountOption(mountOptions, f) {
				mountOptions = append(mountOptions, f)
			}
		}
	}

	klog.V(5).Infof("NodePublishVolume: creating dir %s", target)
	if err := d.mounter.MakeDir(target); err != nil {
		return status.Errorf(codes.Internal, "Could not create dir %q: %v", target, err)
	}

	fsType := mode.Mount.GetFsType()
	if len(fsType) == 0 {
		fsType = defaultFsType
	}

	klog.V(5).Infof("NodePublishVolume: mounting %s at %s with option %s as fstype %s", source, target, mountOptions, fsType)
	if err := d.mounter.Mount(source, target, fsType, mountOptions); err != nil {
		if removeErr := os.Remove(target); removeErr != nil {
			return status.Errorf(codes.Internal, "Could not remove mount target %q: %v", target, err)
		}
		return status.Errorf(codes.Internal, "Could not mount %q at %q: %v", source, target, err)
	}

	return nil
}

// findDevicePath finds path of device and verifies its existence
// if the device is not nvme, return the path directly
// if the device is nvme, finds and returns the nvme device path eg. /dev/nvme1n1
func (d *nodeService) findDevicePath(devicePath, volumeID string) (string, error) {
	exists, err := d.mounter.ExistsPath(devicePath)
	if err != nil {
		return "", err
	}

	// If the path exists, assume it is not nvme device
	if exists {
		return devicePath, nil
	}

	// Else find the nvme device path using volume ID
	// This is the magic name on which AWS presents NVME devices under /dev/disk/by-id/
	// For example, vol-0fab1d5e3f72a5e23 creates a symlink at
	// /dev/disk/by-id/nvme-Amazon_Elastic_Block_Store_vol0fab1d5e3f72a5e23
	nvmeName := "nvme-Amazon_Elastic_Block_Store_" + strings.Replace(volumeID, "-", "", -1)

	return findNvmeVolume(nvmeName)
}

// findNvmeVolume looks for the nvme volume with the specified name
// It follows the symlink (if it exists) and returns the absolute path to the device
func findNvmeVolume(findName string) (device string, err error) {
	p := filepath.Join("/dev/disk/by-id/", findName)
	stat, err := os.Lstat(p)
	if err != nil {
		if os.IsNotExist(err) {
			klog.V(5).Infof("nvme path %q not found", p)
			return "", fmt.Errorf("nvme path %q not found", p)
		}
		return "", fmt.Errorf("error getting stat of %q: %v", p, err)
	}

	if stat.Mode()&os.ModeSymlink != os.ModeSymlink {
		klog.Warningf("nvme file %q found, but was not a symlink", p)
		return "", fmt.Errorf("nvme file %q found, but was not a symlink", p)
	}
	// Find the target, resolving to an absolute path
	// For example, /dev/disk/by-id/nvme-Amazon_Elastic_Block_Store_vol0fab1d5e3f72a5e23 -> ../../nvme2n1
	resolved, err := filepath.EvalSymlinks(p)
	if err != nil {
		return "", fmt.Errorf("error reading target of symlink %q: %v", p, err)
	}

	if !strings.HasPrefix(resolved, "/dev") {
		return "", fmt.Errorf("resolved symlink for %q was unexpected: %q", p, resolved)
	}

	return resolved, nil
}

// getVolumesLimit returns the limit of volumes that the node supports
func (d *nodeService) getVolumesLimit() int64 {
	ebsNitroInstanceTypeRegex := "^[cmr]5.*|t3|z1d"
	instanceType := d.metadata.GetInstanceType()
	if ok, _ := regexp.MatchString(ebsNitroInstanceTypeRegex, instanceType); ok {
		return defaultMaxEBSNitroVolumes
	}
	return defaultMaxEBSVolumes
}

// hasMountOption returns a boolean indicating whether the given
// slice already contains a mount option. This is used to prevent
// passing duplicate option to the mount command.
func hasMountOption(options []string, opt string) bool {
	for _, o := range options {
		if o == opt {
			return true
		}
	}
	return false
}