cephfs: added support for snapshot-backed volumes

This commit implements most of
docs/design/proposals/cephfs-snapshot-shallow-ro-vol.md design document;
specifically (de-)provisioning of snapshot-backed volumes, mounting such
volumes as well as mounting pre-provisioned snapshot-backed volumes.

Signed-off-by: Robert Vasek <robert.vasek@cern.ch>

internal/cephfs/controllerserver.go

@@ -29,6 +29,7 @@ import (
 	"github.com/ceph/ceph-csi/internal/util"
 	"github.com/ceph/ceph-csi/internal/util/k8s"
 	"github.com/ceph/ceph-csi/internal/util/log"
+	rterrors "github.com/ceph/ceph-csi/internal/util/reftracker/errors"
 
 	"github.com/container-storage-interface/spec/lib/go/csi"
 	"github.com/golang/protobuf/ptypes/timestamp"
@@ -65,47 +66,77 @@ func (cs *ControllerServer) createBackingVolume(
 	volClient := core.NewSubVolume(volOptions.GetConnection(), &volOptions.SubVolume, volOptions.ClusterID)
 
 	if sID != nil {
-		if err = cs.OperationLocks.GetRestoreLock(sID.SnapshotID); err != nil {
+		return cs.createBackingVolumeFromSnapshotSource(ctx, volOptions, parentVolOpt, volClient, sID)
+	}
+
+	if parentVolOpt != nil {
+		return cs.createBackingVolumeFromVolumeSource(ctx, parentVolOpt, volClient, pvID)
+	}
+
+	if err = volClient.CreateVolume(ctx); err != nil {
+		log.ErrorLog(ctx, "failed to create volume %s: %v", volOptions.RequestName, err)
+
+		return status.Error(codes.Internal, err.Error())
+	}
+
+	return nil
+}
+
+func (cs *ControllerServer) createBackingVolumeFromSnapshotSource(
+	ctx context.Context,
+	volOptions *store.VolumeOptions,
+	parentVolOpt *store.VolumeOptions,
+	volClient core.SubVolumeClient,
+	sID *store.SnapshotIdentifier,
+) error {
+	if err := cs.OperationLocks.GetRestoreLock(sID.SnapshotID); err != nil {
 		log.ErrorLog(ctx, err.Error())
 
 		return status.Error(codes.Aborted, err.Error())
 	}
 	defer cs.OperationLocks.ReleaseRestoreLock(sID.SnapshotID)
-		snap := core.Snapshot{
-			SnapshotID: sID.FsSnapshotName,
-			SubVolume:  &parentVolOpt.SubVolume,
-		}
 
-		err = volClient.CreateCloneFromSnapshot(ctx, snap)
-		if err != nil {
-			log.ErrorLog(ctx, "failed to create clone from snapshot %s: %v", sID.FsSnapshotName, err)
-
-			return err
-		}
-
-		return err
-	}
-	if parentVolOpt != nil {
-		if err = cs.OperationLocks.GetCloneLock(pvID.VolumeID); err != nil {
-			log.ErrorLog(ctx, err.Error())
-
-			return status.Error(codes.Aborted, err.Error())
-		}
-		defer cs.OperationLocks.ReleaseCloneLock(pvID.VolumeID)
-		err = volClient.CreateCloneFromSubvolume(
-			ctx, &parentVolOpt.SubVolume)
-		if err != nil {
-			log.ErrorLog(ctx, "failed to create clone from subvolume %s: %v", fsutil.VolumeID(pvID.FsSubvolName), err)
+	if volOptions.BackingSnapshot {
+		if err := store.AddSnapshotBackedVolumeRef(ctx, volOptions); err != nil {
+			log.ErrorLog(ctx, "failed to create snapshot-backed volume from snapshot %s: %v",
+				sID.FsSnapshotName, err)
 
 			return err
 		}
 
 		return nil
 	}
-	if err = volClient.CreateVolume(ctx); err != nil {
-		log.ErrorLog(ctx, "failed to create volume %s: %v", volOptions.RequestName, err)
 
-		return status.Error(codes.Internal, err.Error())
+	err := volClient.CreateCloneFromSnapshot(ctx, core.Snapshot{
+		SnapshotID: sID.FsSnapshotName,
+		SubVolume:  &parentVolOpt.SubVolume,
+	})
+	if err != nil {
+		log.ErrorLog(ctx, "failed to create clone from snapshot %s: %v", sID.FsSnapshotName, err)
+
+		return err
+	}
+
+	return nil
+}
+
+func (cs *ControllerServer) createBackingVolumeFromVolumeSource(
+	ctx context.Context,
+	parentVolOpt *store.VolumeOptions,
+	volClient core.SubVolumeClient,
+	pvID *store.VolumeIdentifier,
+) error {
+	if err := cs.OperationLocks.GetCloneLock(pvID.VolumeID); err != nil {
+		log.ErrorLog(ctx, err.Error())
+
+		return status.Error(codes.Aborted, err.Error())
+	}
+	defer cs.OperationLocks.ReleaseCloneLock(pvID.VolumeID)
+
+	if err := volClient.CreateCloneFromSubvolume(ctx, &parentVolOpt.SubVolume); err != nil {
+		log.ErrorLog(ctx, "failed to create clone from subvolume %s: %v", fsutil.VolumeID(pvID.FsSubvolName), err)
+
+		return err
 	}
 
 	return nil
@@ -158,6 +189,7 @@ func checkValidCreateVolumeRequest(
 	parentVol *store.VolumeOptions,
 	pvID *store.VolumeIdentifier,
 	sID *store.SnapshotIdentifier,
+	req *csi.CreateVolumeRequest,
 ) error {
 	switch {
 	case pvID != nil:
@@ -168,6 +200,10 @@ func checkValidCreateVolumeRequest(
 				parentVol.Size,
 				vol.Size)
 		}
+
+		if vol.BackingSnapshot {
+			return errors.New("cloning snapshot-backed volumes is currently not supported")
+		}
 	case sID != nil:
 		if vol.Size < parentVol.Size {
 			return fmt.Errorf(
@@ -176,6 +212,25 @@ func checkValidCreateVolumeRequest(
 				parentVol.Size,
 				vol.Size)
 		}
+
+		if vol.BackingSnapshot {
+			if vol.Size != parentVol.Size {
+				return fmt.Errorf(
+					"cannot create snapshot-backed volume of different size: expected %d bytes, got %d bytes",
+					parentVol.Size,
+					vol.Size,
+				)
+			}
+
+			volCaps := req.GetVolumeCapabilities()
+			for _, volCap := range volCaps {
+				mode := volCap.AccessMode.Mode
+				if mode != csi.VolumeCapability_AccessMode_MULTI_NODE_READER_ONLY &&
+					mode != csi.VolumeCapability_AccessMode_SINGLE_NODE_READER_ONLY {
+					return errors.New("backingSnapshot may be used only with read-only access modes")
+				}
+			}
+		}
 	}
 
 	return nil
@@ -233,7 +288,7 @@ func (cs *ControllerServer) CreateVolume(
 		defer parentVol.Destroy()
 	}
 
-	err = checkValidCreateVolumeRequest(volOptions, parentVol, pvID, sID)
+	err = checkValidCreateVolumeRequest(volOptions, parentVol, pvID, sID, req)
 	if err != nil {
 		return nil, status.Error(codes.InvalidArgument, err.Error())
 	}
@@ -249,7 +304,7 @@ func (cs *ControllerServer) CreateVolume(
 	// TODO return error message if requested vol size greater than found volume return error
 
 	if vID != nil {
-		if sID != nil || pvID != nil {
+		if sID != nil || pvID != nil && !volOptions.BackingSnapshot {
 			volClient := core.NewSubVolume(volOptions.GetConnection(), &volOptions.SubVolume, volOptions.ClusterID)
 			err = volClient.ExpandVolume(ctx, volOptions.Size)
 			if err != nil {
@@ -321,6 +376,11 @@ func (cs *ControllerServer) CreateVolume(
 		return nil, err
 	}
 
+	if !volOptions.BackingSnapshot {
+		// Get root path for the created subvolume.
+		// Note that root path for snapshot-backed volumes has been already set when
+		// building VolumeOptions.
+
 		volClient := core.NewSubVolume(volOptions.GetConnection(), &volOptions.SubVolume, volOptions.ClusterID)
 		volOptions.RootPath, err = volClient.GetVolumeRootPathCeph(ctx)
 		if err != nil {
@@ -342,6 +402,7 @@ func (cs *ControllerServer) CreateVolume(
 
 			return nil, status.Error(codes.Internal, err.Error())
 		}
+	}
 
 	log.DebugLog(ctx, "cephfs: successfully created backing volume named %s for request name %s",
 		vID.FsSubvolName, requestName)
@@ -452,16 +513,8 @@ func (cs *ControllerServer) DeleteVolume(
 	}
 	defer cr.DeleteCredentials()
 
-	volClient := core.NewSubVolume(volOptions.GetConnection(), &volOptions.SubVolume, volOptions.ClusterID)
-	if err = volClient.PurgeVolume(ctx, false); err != nil {
-		log.ErrorLog(ctx, "failed to delete volume %s: %v", volID, err)
-		if errors.Is(err, cerrors.ErrVolumeHasSnapshots) {
-			return nil, status.Error(codes.FailedPrecondition, err.Error())
-		}
-
-		if !errors.Is(err, cerrors.ErrVolumeNotFound) {
-			return nil, status.Error(codes.Internal, err.Error())
-		}
+	if err := cleanUpBackingVolume(ctx, volOptions, vID, cr); err != nil {
+		return nil, err
 	}
 
 	if err := store.UndoVolReservation(ctx, volOptions, *vID, secrets); err != nil {
@@ -473,6 +526,84 @@ func (cs *ControllerServer) DeleteVolume(
 	return &csi.DeleteVolumeResponse{}, nil
 }
 
+func cleanUpBackingVolume(
+	ctx context.Context,
+	volOptions *store.VolumeOptions,
+	volID *store.VolumeIdentifier,
+	cr *util.Credentials,
+) error {
+	if !volOptions.BackingSnapshot {
+		// Regular volumes need to be purged.
+
+		volClient := core.NewSubVolume(volOptions.GetConnection(), &volOptions.SubVolume, volOptions.ClusterID)
+		if err := volClient.PurgeVolume(ctx, false); err != nil {
+			log.ErrorLog(ctx, "failed to delete volume %s: %v", volID, err)
+			if errors.Is(err, cerrors.ErrVolumeHasSnapshots) {
+				return status.Error(codes.FailedPrecondition, err.Error())
+			}
+
+			if !errors.Is(err, cerrors.ErrVolumeNotFound) {
+				return status.Error(codes.Internal, err.Error())
+			}
+		}
+
+		return nil
+	}
+
+	// Snapshot-backed volumes need to un-reference the backing snapshot, and
+	// the snapshot itself may need to be deleted if its reftracker doesn't
+	// hold any references anymore.
+
+	backingSnapNeedsDelete, err := store.UnrefSnapshotBackedVolume(ctx, volOptions)
+	if err != nil {
+		if errors.Is(err, rterrors.ErrObjectOutOfDate) {
+			return status.Error(codes.Aborted, err.Error())
+		}
+
+		return status.Error(codes.Internal, err.Error())
+	}
+
+	if !backingSnapNeedsDelete {
+		return nil
+	}
+
+	snapParentVolOptions, _, snapID, err := store.NewSnapshotOptionsFromID(ctx, volOptions.BackingSnapshotID, cr)
+	if err != nil {
+		absorbErrs := []error{
+			util.ErrPoolNotFound,
+			util.ErrKeyNotFound,
+			cerrors.ErrSnapNotFound,
+			cerrors.ErrVolumeNotFound,
+		}
+
+		fatalErr := true
+		for i := range absorbErrs {
+			if errors.Is(err, absorbErrs[i]) {
+				fatalErr = false
+
+				break
+			}
+		}
+
+		if fatalErr {
+			return status.Error(codes.Internal, err.Error())
+		}
+	} else {
+		snapClient := core.NewSnapshot(
+			snapParentVolOptions.GetConnection(),
+			snapID.FsSnapshotName,
+			&snapParentVolOptions.SubVolume,
+		)
+
+		err = deleteSnapshotAndUndoReservation(ctx, snapClient, snapParentVolOptions, snapID, cr)
+		if err != nil {
+			return status.Error(codes.Internal, err.Error())
+		}
+	}
+
+	return nil
+}
+
 // ValidateVolumeCapabilities checks whether the volume capabilities requested
 // are supported.
 func (cs *ControllerServer) ValidateVolumeCapabilities(
@@ -537,6 +668,10 @@ func (cs *ControllerServer) ControllerExpandVolume(
 	}
 	defer volOptions.Destroy()
 
+	if volOptions.BackingSnapshot {
+		return nil, status.Error(codes.InvalidArgument, "cannot expand snapshot-backed volume")
+	}
+
 	RoundOffSize := util.RoundOffBytes(req.GetCapacityRange().GetRequiredBytes())
 	volClient := core.NewSubVolume(volOptions.GetConnection(), &volOptions.SubVolume, volOptions.ClusterID)
 	if err = volClient.ResizeVolume(ctx, RoundOffSize); err != nil {
@@ -615,6 +750,10 @@ func (cs *ControllerServer) CreateSnapshot(
 			parentVolOptions.ClusterID)
 	}
 
+	if parentVolOptions.BackingSnapshot {
+		return nil, status.Error(codes.InvalidArgument, "cannot snapshot a snapshot-backed volume")
+	}
+
 	cephfsSnap, genSnapErr := store.GenSnapFromOptions(ctx, req)
 	if genSnapErr != nil {
 		return nil, status.Error(codes.Internal, genSnapErr.Error())
@@ -780,6 +919,7 @@ func (cs *ControllerServer) validateSnapshotReq(ctx context.Context, req *csi.Cr
 
 // DeleteSnapshot deletes the snapshot in backend and removes the
 // snapshot metadata from store.
+// nolint:gocyclo,cyclop // TODO: reduce complexity
 func (cs *ControllerServer) DeleteSnapshot(
 	ctx context.Context,
 	req *csi.DeleteSnapshotRequest,
@@ -878,17 +1018,51 @@ func (cs *ControllerServer) DeleteSnapshot(
 			return nil, status.Error(codes.Internal, err.Error())
 		}
 	}
-	err = snapClient.DeleteSnapshot(ctx)
+
+	needsDelete, err := store.UnrefSelfInSnapshotBackedVolumes(ctx, volOpt, sid.SnapshotID)
+	if err != nil {
+		if errors.Is(err, rterrors.ErrObjectOutOfDate) {
+			return nil, status.Error(codes.Aborted, err.Error())
+		}
+
+		return nil, status.Error(codes.Internal, err.Error())
+	}
+
+	if needsDelete {
+		err = deleteSnapshotAndUndoReservation(
+			ctx,
+			snapClient,
+			volOpt,
+			sid,
+			cr,
+		)
 		if err != nil {
 			return nil, status.Error(codes.Internal, err.Error())
 		}
-	err = store.UndoSnapReservation(ctx, volOpt, *sid, sid.RequestName, cr)
-	if err != nil {
-		log.ErrorLog(ctx, "failed to remove reservation for snapname (%s) with backing snap (%s) (%s)",
-			sid.RequestName, sid.FsSnapshotName, err)
-
-		return nil, status.Error(codes.Internal, err.Error())
 	}
 
 	return &csi.DeleteSnapshotResponse{}, nil
 }
+
+func deleteSnapshotAndUndoReservation(
+	ctx context.Context,
+	snapClient core.SnapshotClient,
+	parentVolOptions *store.VolumeOptions,
+	snapID *store.SnapshotIdentifier,
+	cr *util.Credentials,
+) error {
+	err := snapClient.DeleteSnapshot(ctx)
+	if err != nil {
+		return err
+	}
+
+	err = store.UndoSnapReservation(ctx, parentVolOptions, *snapID, snapID.RequestName, cr)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to remove reservation for snapname (%s) with backing snap (%s) (%s)",
+			snapID.RequestName, snapID.RequestName, err)
+
+		return err
+	}
+
+	return nil
+}

internal/cephfs/fuserecovery.go

@@ -192,7 +192,7 @@ func (ns *NodeServer) tryRestoreFuseMountsInNodePublish(
 	// Unmount and mount the volume.
 
 	if stagingTargetMs != msMounted {
-		if err := mounter.UnmountVolume(ctx, stagingTargetPath); err != nil {
+		if err := mounter.UnmountAll(ctx, stagingTargetPath); err != nil {
 			return err
 		}
 
@@ -269,5 +269,5 @@ func (ns *NodeServer) tryRestoreFuseMountInNodeStage(
 
 	// Restoration here means only unmounting the volume.
 	// NodeStageVolume should take care of the rest.
-	return mounter.UnmountVolume(ctx, stagingTargetPath)
+	return mounter.UnmountAll(ctx, stagingTargetPath)
 }

internal/cephfs/mounter/fuse.go

@@ -118,8 +118,8 @@ func (m *FuseMounter) Mount(
 
 func (m *FuseMounter) Name() string { return "Ceph FUSE driver" }
 
-func UnmountVolume(ctx context.Context, mountPoint string) error {
-	if _, stderr, err := util.ExecCommand(ctx, "umount", mountPoint); err != nil {
+func UnmountVolume(ctx context.Context, mountPoint string, opts ...string) error {
+	if _, stderr, err := util.ExecCommand(ctx, "umount", append([]string{mountPoint}, opts...)...); err != nil {
 		err = fmt.Errorf("%w stderr: %s", err, stderr)
 		if strings.Contains(err.Error(), fmt.Sprintf("umount: %s: not mounted", mountPoint)) ||
 			strings.Contains(err.Error(), "No such file or directory") {
@@ -149,3 +149,7 @@ func UnmountVolume(ctx context.Context, mountPoint string) error {
 
 	return nil
 }
+
+func UnmountAll(ctx context.Context, mountPoint string) error {
+	return UnmountVolume(ctx, mountPoint, "--all-targets")
+}

internal/cephfs/nodeserver.go

@@ -21,6 +21,7 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"path"
 	"strings"
 
 	cerrors "github.com/ceph/ceph-csi/internal/cephfs/errors"
@@ -101,6 +102,20 @@ func (ns *NodeServer) getVolumeOptions(
 	return volOptions, nil
 }
 
+func validateSnapshotBackedVolCapability(volCap *csi.VolumeCapability) error {
+	// Snapshot-backed volumes may be used with read-only volume access modes only.
+
+	mode := volCap.AccessMode.Mode
+
+	if mode != csi.VolumeCapability_AccessMode_MULTI_NODE_READER_ONLY &&
+		mode != csi.VolumeCapability_AccessMode_SINGLE_NODE_READER_ONLY {
+		return status.Error(codes.InvalidArgument,
+			"snapshot-backed volume supports only read-only access mode")
+	}
+
+	return nil
+}
+
 // NodeStageVolume mounts the volume to a staging path on the node.
 func (ns *NodeServer) NodeStageVolume(
 	ctx context.Context,
@@ -140,6 +155,12 @@ func (ns *NodeServer) NodeStageVolume(
 		}
 	}
 
+	if volOptions.BackingSnapshot {
+		if err = validateSnapshotBackedVolCapability(req.GetVolumeCapability()); err != nil {
+			return nil, err
+		}
+	}
+
 	mnt, err := mounter.New(volOptions)
 	if err != nil {
 		log.ErrorLog(ctx, "failed to create mounter for volume %s: %v", volID, err)
@@ -192,7 +213,7 @@ func (ns *NodeServer) NodeStageVolume(
 			log.ErrorLog(ctx, "cephfs: failed to write NodeStageMountinfo for volume %s: %v", volID, err)
 
 			// Try to clean node stage mount.
-			if unmountErr := mounter.UnmountVolume(ctx, stagingTargetPath); unmountErr != nil {
+			if unmountErr := mounter.UnmountAll(ctx, stagingTargetPath); unmountErr != nil {
 				log.ErrorLog(ctx, "cephfs: failed to unmount %s in WriteNodeStageMountinfo clean up: %v",
 					stagingTargetPath, unmountErr)
 			}
@@ -223,7 +244,7 @@ func (*NodeServer) mount(
 
 	log.DebugLog(ctx, "cephfs: mounting volume %s with %s", volID, mnt.Name())
 
-	readOnly := "ro"
+	const readOnly = "ro"
 
 	if volCap.AccessMode.Mode == csi.VolumeCapability_AccessMode_MULTI_NODE_READER_ONLY ||
 		volCap.AccessMode.Mode == csi.VolumeCapability_AccessMode_SINGLE_NODE_READER_ONLY {
@@ -248,9 +269,112 @@ func (*NodeServer) mount(
 		return status.Error(codes.Internal, err.Error())
 	}
 
+	defer func() {
+		if err == nil {
+			return
+		}
+
+		unmountErr := mounter.UnmountAll(ctx, stagingTargetPath)
+		if unmountErr != nil {
+			log.ErrorLog(ctx, "failed to clean up mounts in rollback procedure: %v", unmountErr)
+		}
+	}()
+
+	if volOptions.BackingSnapshot {
+		snapshotRoot, err := getBackingSnapshotRoot(ctx, volOptions, stagingTargetPath)
+		if err != nil {
+			return err
+		}
+
+		absoluteSnapshotRoot := path.Join(stagingTargetPath, snapshotRoot)
+		err = mounter.BindMount(
+			ctx,
+			absoluteSnapshotRoot,
+			stagingTargetPath,
+			true,
+			[]string{"bind", "_netdev"},
+		)
+
+		if err != nil {
+			log.ErrorLog(ctx,
+				"failed to bind mount snapshot root %s: %v", absoluteSnapshotRoot, err)
+
+			return status.Error(codes.Internal, err.Error())
+		}
+	}
+
 	return nil
 }
 
+func getBackingSnapshotRoot(
+	ctx context.Context,
+	volOptions *store.VolumeOptions,
+	stagingTargetPath string,
+) (string, error) {
+	if volOptions.ProvisionVolume {
+		// Provisioned snapshot-backed volumes should have their BackingSnapshotRoot
+		// already populated.
+		return volOptions.BackingSnapshotRoot, nil
+	}
+
+	// Pre-provisioned snapshot-backed volumes are more involved:
+	//
+	// Snapshots created with `ceph fs subvolume snapshot create` have following
+	// snap directory name format inside <root path>/.snap:
+	//
+	//   _<snapshot>_<snapshot inode number>
+	//
+	// We don't know what <snapshot inode number> is, and so <root path>/.snap
+	// needs to be traversed in order to determine the full snapshot directory name.
+
+	snapshotsBase := path.Join(stagingTargetPath, ".snap")
+
+	dir, err := os.Open(snapshotsBase)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to open %s when searching for snapshot root: %v", snapshotsBase, err)
+
+		return "", status.Errorf(codes.Internal, err.Error())
+	}
+
+	// Read the contents of <root path>/.snap directory into a string slice.
+
+	contents, err := dir.Readdirnames(0)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to read %s when searching for snapshot root: %v", snapshotsBase, err)
+
+		return "", status.Errorf(codes.Internal, err.Error())
+	}
+
+	var (
+		found           bool
+		snapshotDirName string
+	)
+
+	// Look through the directory's contents and try to find the correct snapshot
+	// dir name. The search must be exhaustive to catch possible ambiguous results.
+
+	for i := range contents {
+		if !strings.Contains(contents[i], volOptions.BackingSnapshotID) {
+			continue
+		}
+
+		if !found {
+			found = true
+			snapshotDirName = contents[i]
+		} else {
+			return "", status.Errorf(codes.InvalidArgument, "ambiguous backingSnapshotID %s in %s",
+				volOptions.BackingSnapshotID, snapshotsBase)
+		}
+	}
+
+	if !found {
+		return "", status.Errorf(codes.InvalidArgument, "no snapshot with backingSnapshotID %s found in %s",
+			volOptions.BackingSnapshotID, snapshotsBase)
+	}
+
+	return path.Join(".snap", snapshotDirName), nil
+}
+
 // NodePublishVolume mounts the volume mounted to the staging path to the target
 // path.
 func (ns *NodeServer) NodePublishVolume(
@@ -444,7 +568,7 @@ func (ns *NodeServer) NodeUnstageVolume(
 		return &csi.NodeUnstageVolumeResponse{}, nil
 	}
 	// Unmount the volume
-	if err = mounter.UnmountVolume(ctx, stagingTargetPath); err != nil {
+	if err = mounter.UnmountAll(ctx, stagingTargetPath); err != nil {
 		return nil, status.Error(codes.Internal, err.Error())
 	}
 

internal/cephfs/store/backingsnapshot.go

@@ -0,0 +1,168 @@
+/*
+Copyright 2022 The Ceph-CSI 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 store
+
+import (
+	"context"
+	"fmt"
+
+	fsutil "github.com/ceph/ceph-csi/internal/cephfs/util"
+	"github.com/ceph/ceph-csi/internal/util/log"
+	"github.com/ceph/ceph-csi/internal/util/reftracker"
+	"github.com/ceph/ceph-csi/internal/util/reftracker/radoswrapper"
+	"github.com/ceph/ceph-csi/internal/util/reftracker/reftype"
+)
+
+func fmtBackingSnapshotReftrackerName(backingSnapID string) string {
+	return fmt.Sprintf("rt-backingsnapshot-%s", backingSnapID)
+}
+
+func AddSnapshotBackedVolumeRef(
+	ctx context.Context,
+	volOptions *VolumeOptions,
+) error {
+	ioctx, err := volOptions.conn.GetIoctx(volOptions.MetadataPool)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to create RADOS ioctx: %s", err)
+
+		return err
+	}
+	defer ioctx.Destroy()
+
+	ioctx.SetNamespace(fsutil.RadosNamespace)
+
+	var (
+		backingSnapID = volOptions.BackingSnapshotID
+		ioctxW        = radoswrapper.NewIOContext(ioctx)
+	)
+
+	created, err := reftracker.Add(
+		ioctxW,
+		fmtBackingSnapshotReftrackerName(backingSnapID),
+		map[string]struct{}{
+			backingSnapID:    {},
+			volOptions.VolID: {},
+		},
+	)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to add refs for backing snapshot %s: %v",
+			backingSnapID, err)
+
+		return err
+	}
+
+	defer func() {
+		if err == nil {
+			return
+		}
+
+		// Clean up after failure.
+
+		var deleted bool
+		deleted, err = reftracker.Remove(
+			ioctxW,
+			fmtBackingSnapshotReftrackerName(backingSnapID),
+			map[string]reftype.RefType{
+				backingSnapID:    reftype.Normal,
+				volOptions.VolID: reftype.Normal,
+			},
+		)
+		if err != nil {
+			log.ErrorLog(ctx, "failed to remove refs in cleanup procedure for backing snapshot %s: %v",
+				backingSnapID, err)
+		}
+
+		if created && !deleted {
+			log.ErrorLog(ctx, "orphaned reftracker object %s (pool %s, namespace %s)",
+				backingSnapID, volOptions.MetadataPool, fsutil.RadosNamespace)
+		}
+	}()
+
+	// There may have been a race between adding a ref to the reftracker and
+	// deleting the backing snapshot. Make sure the snapshot still exists by
+	// trying to retrieve it again.
+	_, _, _, err = NewSnapshotOptionsFromID(ctx, volOptions.BackingSnapshotID, volOptions.conn.Creds)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to get backing snapshot %s: %v", volOptions.BackingSnapshotID, err)
+	}
+
+	return err
+}
+
+func UnrefSnapshotBackedVolume(
+	ctx context.Context,
+	volOptions *VolumeOptions,
+) (bool, error) {
+	ioctx, err := volOptions.conn.GetIoctx(volOptions.MetadataPool)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to create RADOS ioctx: %s", err)
+
+		return false, err
+	}
+	defer ioctx.Destroy()
+
+	ioctx.SetNamespace(fsutil.RadosNamespace)
+
+	var (
+		backingSnapID = volOptions.BackingSnapshotID
+		ioctxW        = radoswrapper.NewIOContext(ioctx)
+	)
+
+	deleted, err := reftracker.Remove(
+		ioctxW,
+		fmtBackingSnapshotReftrackerName(backingSnapID),
+		map[string]reftype.RefType{
+			volOptions.VolID: reftype.Normal,
+		},
+	)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to remove refs for backing snapshot %s: %v",
+			backingSnapID, err)
+
+		return false, err
+	}
+
+	return deleted, err
+}
+
+// UnrefSelfInSnapshotBackedVolumes removes (masks) snapshot ID in the
+// reftracker for volumes backed by this snapshot. The returned boolean
+// value signals whether the snapshot is not referenced by any such volumes
+// and needs to be removed.
+func UnrefSelfInSnapshotBackedVolumes(
+	ctx context.Context,
+	snapParentVolOptions *VolumeOptions,
+	snapshotID string,
+) (bool, error) {
+	ioctx, err := snapParentVolOptions.conn.GetIoctx(snapParentVolOptions.MetadataPool)
+	if err != nil {
+		log.ErrorLog(ctx, "failed to create RADOS ioctx: %s", err)
+
+		return false, err
+	}
+	defer ioctx.Destroy()
+
+	ioctx.SetNamespace(fsutil.RadosNamespace)
+
+	return reftracker.Remove(
+		radoswrapper.NewIOContext(ioctx),
+		fmtBackingSnapshotReftrackerName(snapshotID),
+		map[string]reftype.RefType{
+			snapshotID: reftype.Mask,
+		},
+	)
+}

internal/cephfs/store/volumeoptions.go

@@ -20,6 +20,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"path"
 	"strconv"
 	"strings"
 
@@ -34,27 +35,31 @@ import (
 
 type VolumeOptions struct {
 	core.SubVolume
-	TopologyPools       *[]util.TopologyConstrainedPool
-	TopologyRequirement *csi.TopologyRequirement
-	Topology            map[string]string
+
 	RequestName  string
 	NamePrefix   string
 	ClusterID    string
-	FscID               int64
 	MetadataPool string
 	// ReservedID represents the ID reserved for a subvolume
 	ReservedID           string
 	Monitors             string `json:"monitors"`
 	RootPath             string `json:"rootPath"`
 	Mounter              string `json:"mounter"`
-	ProvisionVolume    bool   `json:"provisionVolume"`
+	BackingSnapshotRoot  string // Snapshot root relative to RootPath.
+	BackingSnapshotID    string
 	KernelMountOptions   string `json:"kernelMountOptions"`
 	FuseMountOptions     string `json:"fuseMountOptions"`
-	// Network namespace file path to execute nsenter command
 	NetNamespaceFilePath string
+	TopologyPools        *[]util.TopologyConstrainedPool
+	TopologyRequirement  *csi.TopologyRequirement
+	Topology             map[string]string
+	FscID                int64
 
 	// conn is a connection to the Ceph cluster obtained from a ConnPool
 	conn *util.ClusterConnection
+
+	ProvisionVolume bool `json:"provisionVolume"`
+	BackingSnapshot bool `json:"backingSnapshot"`
 }
 
 // Connect a CephFS volume to the Ceph cluster.
@@ -184,6 +189,11 @@ func (vo *VolumeOptions) GetConnection() *util.ClusterConnection {
 	return vo.conn
 }
 
+func fmtBackingSnapshotOptionMismatch(optName, expected, actual string) error {
+	return fmt.Errorf("%s option mismatch with backing snapshot: got %s, expected %s",
+		optName, actual, expected)
+}
+
 // NewVolumeOptions generates a new instance of volumeOptions from the provided
 // CSI request parameters.
 func NewVolumeOptions(ctx context.Context, requestName string, req *csi.CreateVolumeRequest,
@@ -191,6 +201,7 @@ func NewVolumeOptions(ctx context.Context, requestName string, req *csi.CreateVo
 ) (*VolumeOptions, error) {
 	var (
 		opts                VolumeOptions
+		backingSnapshotBool string
 		err                 error
 	)
 
@@ -228,6 +239,16 @@ func NewVolumeOptions(ctx context.Context, requestName string, req *csi.CreateVo
 		return nil, err
 	}
 
+	if err = extractOptionalOption(&backingSnapshotBool, "backingSnapshot", volOptions); err != nil {
+		return nil, err
+	}
+
+	if backingSnapshotBool != "" {
+		if opts.BackingSnapshot, err = strconv.ParseBool(backingSnapshotBool); err != nil {
+			return nil, fmt.Errorf("failed to parse backingSnapshot: %w", err)
+		}
+	}
+
 	opts.RequestName = requestName
 
 	err = opts.Connect(cr)
@@ -261,6 +282,19 @@ func NewVolumeOptions(ctx context.Context, requestName string, req *csi.CreateVo
 
 	opts.ProvisionVolume = true
 
+	if opts.BackingSnapshot {
+		if req.GetVolumeContentSource() == nil || req.GetVolumeContentSource().GetSnapshot() == nil {
+			return nil, errors.New("backingSnapshot option requires snapshot volume source")
+		}
+
+		opts.BackingSnapshotID = req.GetVolumeContentSource().GetSnapshot().GetSnapshotId()
+
+		err = opts.populateVolumeOptionsFromBackingSnapshot(ctx, cr)
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	return &opts, nil
 }
 
@@ -364,23 +398,109 @@ func NewVolumeOptionsFromVolID(
 		}
 	}
 
-	volOptions.ProvisionVolume = true
-	volOptions.SubVolume.VolID = vid.FsSubvolName
-	vol := core.NewSubVolume(volOptions.conn, &volOptions.SubVolume, volOptions.ClusterID)
-	info, err := vol.GetSubVolumeInfo(ctx)
-	if err == nil {
-		volOptions.RootPath = info.Path
-		volOptions.Features = info.Features
-		volOptions.Size = info.BytesQuota
+	if imageAttributes.BackingSnapshotID != "" || volOptions.BackingSnapshotID != "" {
+		volOptions.BackingSnapshot = true
+		volOptions.BackingSnapshotID = imageAttributes.BackingSnapshotID
 	}
 
-	if errors.Is(err, cerrors.ErrInvalidCommand) {
-		volOptions.RootPath, err = vol.GetVolumeRootPathCeph(ctx)
+	volOptions.ProvisionVolume = true
+	volOptions.SubVolume.VolID = vid.FsSubvolName
+
+	if volOptions.BackingSnapshot {
+		err = volOptions.populateVolumeOptionsFromBackingSnapshot(ctx, cr)
+	} else {
+		err = volOptions.populateVolumeOptionsFromSubvolume(ctx)
 	}
 
 	return &volOptions, &vid, err
 }
 
+func (vo *VolumeOptions) populateVolumeOptionsFromSubvolume(ctx context.Context) error {
+	vol := core.NewSubVolume(vo.conn, &vo.SubVolume, vo.ClusterID)
+
+	var info *core.Subvolume
+	info, err := vol.GetSubVolumeInfo(ctx)
+	if err == nil {
+		vo.RootPath = info.Path
+		vo.Features = info.Features
+		vo.Size = info.BytesQuota
+	}
+
+	if errors.Is(err, cerrors.ErrInvalidCommand) {
+		vo.RootPath, err = vol.GetVolumeRootPathCeph(ctx)
+	}
+
+	return err
+}
+
+func (vo *VolumeOptions) populateVolumeOptionsFromBackingSnapshot(
+	ctx context.Context,
+	cr *util.Credentials,
+) error {
+	// As of CephFS snapshot v2 API, snapshots may be found in two locations:
+	//
+	// (a) /volumes/<volume group>/<subvolume>/.snap/<snapshot>/<UUID>
+	// (b) /volumes/<volume group>/<subvolume>/<UUID>/.snap/_<snapshot>_<snapshot inode number>
+
+	if !vo.ProvisionVolume {
+		// Case (b)
+		//
+		// If the volume is not provisioned by us, we assume that we have access only
+		// to snapshot's parent volume root. In this case, o.RootPath is expected to
+		// be already set in the volume context.
+
+		// BackingSnapshotRoot cannot be determined at this stage, because the
+		// full directory name is not known (see snapshot path format for case
+		// (b) above). RootPath/.snap must be traversed in order to find out
+		// the snapshot directory name.
+
+		return nil
+	}
+
+	parentBackingSnapVolOpts, _, snapID, err := NewSnapshotOptionsFromID(ctx, vo.BackingSnapshotID, cr)
+	if err != nil {
+		return fmt.Errorf("failed to retrieve backing snapshot %s: %w", vo.BackingSnapshotID, err)
+	}
+
+	// Ensure that backing snapshot parent's volume options match the context.
+	// Snapshot-backed volume inherits all its parent's (parent of the snapshot) options.
+
+	if vo.ClusterID != parentBackingSnapVolOpts.ClusterID {
+		return fmtBackingSnapshotOptionMismatch("clusterID", vo.ClusterID, parentBackingSnapVolOpts.ClusterID)
+	}
+
+	if vo.Pool != "" {
+		return errors.New("cannot set pool for snapshot-backed volume")
+	}
+
+	if vo.MetadataPool != parentBackingSnapVolOpts.MetadataPool {
+		return fmtBackingSnapshotOptionMismatch("MetadataPool", vo.MetadataPool, parentBackingSnapVolOpts.MetadataPool)
+	}
+
+	if vo.FsName != parentBackingSnapVolOpts.FsName {
+		return fmtBackingSnapshotOptionMismatch("fsName", vo.FsName, parentBackingSnapVolOpts.FsName)
+	}
+
+	if vo.SubvolumeGroup != parentBackingSnapVolOpts.SubvolumeGroup {
+		return fmtBackingSnapshotOptionMismatch("SubvolumeGroup", vo.SubvolumeGroup, parentBackingSnapVolOpts.SubvolumeGroup)
+	}
+
+	vo.Features = parentBackingSnapVolOpts.Features
+	vo.Size = parentBackingSnapVolOpts.Size
+
+	// For case (a) (o.ProvisionVolume==true is assumed), snapshot root path
+	// can be built out of subvolume root path, which is in following format:
+	//
+	//   /volumes/<volume group>/<subvolume>/<subvolume UUID>
+
+	subvolRoot, subvolUUID := path.Split(parentBackingSnapVolOpts.RootPath)
+
+	vo.RootPath = subvolRoot
+	vo.BackingSnapshotRoot = path.Join(".snap", snapID.FsSnapshotName, subvolUUID)
+
+	return nil
+}
+
 // NewVolumeOptionsFromMonitorList generates a new instance of VolumeOptions and
 // VolumeIdentifier from the provided CSI volume context.
 func NewVolumeOptionsFromMonitorList(
@@ -438,9 +558,17 @@ func NewVolumeOptionsFromMonitorList(
 		return nil, nil, err
 	}
 
+	if err = extractOptionalOption(&opts.BackingSnapshotID, "backingSnapshotID", options); err != nil {
+		return nil, nil, err
+	}
+
 	vid.FsSubvolName = volID
 	vid.VolumeID = volID
 
+	if opts.BackingSnapshotID != "" {
+		opts.BackingSnapshot = true
+	}
+
 	return &opts, &vid, nil
 }
 
@@ -511,6 +639,10 @@ func NewVolumeOptionsFromStaticVolume(
 	vid.FsSubvolName = opts.RootPath
 	vid.VolumeID = volID
 
+	if opts.BackingSnapshotID != "" {
+		opts.BackingSnapshot = true
+	}
+
 	return &opts, &vid, nil
 }
 
@@ -599,6 +731,7 @@ func NewSnapshotOptionsFromID(
 	}
 	volOptions.Features = subvolInfo.Features
 	volOptions.Size = subvolInfo.BytesQuota
+	volOptions.RootPath = subvolInfo.Path
 	snap := core.NewSnapshot(volOptions.conn, sid.FsSnapshotName, &volOptions.SubVolume)
 	info, err := snap.GetSnapshotInfo(ctx)
 	if err != nil {

internal/journal/voljournal.go

@@ -152,7 +152,7 @@ type Config struct {
 	// ownerKey is used to identify the owner of the volume, can be used with some KMS configurations
 	ownerKey string
 
-	// backingSnapshotIDKey is snapshot ID on which a shallow CephFS volume is based.
+	// backingSnapshotIDKey ID of the snapshot on which the CephFS snapshot-backed volume is based
 	backingSnapshotIDKey string
 
 	// commonPrefix is the prefix common to all omap keys for this Config
@@ -532,7 +532,7 @@ Input arguments:
 	- kmsConf: Name of the key management service used to encrypt the image (optional)
 	- volUUID: UUID need to be reserved instead of auto-generating one (this is useful for mirroring and metro-DR)
 	- owner: the owner of the volume (optional)
-	- backingSnapshotID:  (optional)
+	- backingSnapshotID: ID of the snapshot on which the CephFS snapshot-backed volume is based (optional)
 
 Return values:
 	- string: Contains the UUID that was reserved for the passed in reqName
@@ -645,7 +645,7 @@ func (conn *Connection) ReserveName(ctx context.Context,
 		omapValues[cj.cephSnapSourceKey] = parentName
 	}
 
-	// Update backing snapshot ID for shallow CephFS volume
+	// Update backing snapshot ID for snapshot-backed CephFS volume
 	if backingSnapshotID != "" {
 		omapValues[cj.backingSnapshotIDKey] = backingSnapshotID
 	}
@@ -667,7 +667,7 @@ type ImageAttributes struct {
 	Owner             string // Contains the owner to be used in combination with KmsID (for some KMS)
 	ImageID           string // Contains the image id
 	JournalPoolID     int64  // Pool ID of the CSI journal pool, stored in big endian format (on-disk data)
-	BackingSnapshotID string // ID of the backing snapshot of a shallow CephFS volume
+	BackingSnapshotID string // ID of the snapshot on which the CephFS snapshot-backed volume is based
 }
 
 // GetImageAttributes fetches all keys and their values, from a UUID directory, returning ImageAttributes structure.