[manila-csi-plugin] CephFS adapter missing staticVolume parameter for ceph-csi compatibility

[gouthampacha]

What happened:

The Manila CSI CephFS share adapter sends provisionVolume: "false" in the volume context to indicate a statically provisioned volume to the ceph-csi node plugin. However, ceph-csi introduced a volume ID format validator in e2a1f6e that rejects volume IDs not matching the ceph-csi internal format (^[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[a-zA-Z0-9\-\_]+$).

Manila CSI passes Manila share UUIDs as volume IDs (e.g. fee6cec9-988a-45fd-b01d-dc1f5e562104), which don't match this format. The format check is skipped when IsStaticVol() returns true, but that function checks for staticVolume: "true" in the volume attributes — not provisionVolume: "false".

The result is that NodeStageVolume fails with: volumeID has an unexpected format.

What you expected to happen:

Manila CSI volumes should mount successfully with ceph-csi regardless of the volume ID format, since they are statically provisioned volumes.

How to reproduce:

Deploy Manila CSI with CephFS protocol using ceph-csi >= v3.14.0 as the forwarding node plugin. Create a CephFS PVC — the pod will fail to start with the format validation error.

Fix:

Add staticVolume: "true" to the CephFS volume context alongside the existing provisionVolume: "false". The staticVolume parameter is simply ignored by older ceph-csi versions that don't recognize it, making this change backwards compatible.

Environment:

• cloud-provider-openstack version: master
• ceph-csi version: v3.14.0+

[gouthampacha]

Analysis update: staticVolume: "true" alone is insufficient

After tracing through the ceph-csi code, adding staticVolume: "true" without clusterID actually makes things worse on v3.14+. Here's the full code path:

1. Format validation (internal/util/validate.go)

var validator = regexp.MustCompile(`^[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[a-zA-Z0-9\-\_]+$`)

func ValidateVolumeID(volumeID string, skipFormatCheck bool) error {
    // ...
    if matches := validator.MatchString(volumeID); !skipFormatCheck && !matches {
        return fmt.Errorf("the volumeID has an unexpected format: %q", volumeID)
    }
    return nil
}

skipFormatCheck is set by IsStaticVol(ctx), which only checks for staticVolume: "true" — not provisionVolume: "false".

2. Volume options fallback (internal/cephfs/nodeserver.go)

func (ns *cephfsNodeServer) getVolumeOptions(...) (*store.VolumeOptions, error) {
    volOptions, _, err := store.NewVolumeOptionsFromVolID(...)
    if err != nil {
        if !errors.Is(err, cerrors.ErrInvalidVolID) {
            return nil, status.Error(codes.Internal, err.Error())
        }
        volOptions, _, err = store.NewVolumeOptionsFromStaticVolume(...)
        if err != nil {
            if !errors.Is(err, cerrors.ErrNonStaticVolume) {
                return nil, status.Error(codes.Internal, err.Error()) // HARD STOP
            }
            volOptions, _, err = store.NewVolumeOptionsFromMonitorList(...)
        }
    }
    return volOptions, nil
}

3. Static volume path (internal/cephfs/store/volumeoptions.go)

NewVolumeOptionsFromStaticVolume checks for staticVolume: "true" early. If found and true, it calls GetClusterInformation() which requires clusterID:

clusterID, ok := options["clusterID"]
if !ok {
    err := errors.New("clusterID must be set")
    return nil, err  // NOT ErrNonStaticVolume
}

This error is NOT ErrNonStaticVolume, so NewVolumeOptionsFromMonitorList is never reached.

Result

With staticVolume: "true" but no clusterID, Manila CSI gets "clusterID must be set" instead of the format validation error. Without staticVolume: "true", the format validation rejects Manila's UUID volume IDs. Either way, it fails on ceph-csi v3.14+.

Root cause

IsStaticVol() only recognizes staticVolume: "true" but not provisionVolume: "false". Manila CSI's volumes are pre-provisioned static volumes — ceph-csi just mounts them. The existing NewVolumeOptionsFromMonitorList path handles them correctly using monitors + provisionVolume: "false" from the volume context. But the format validation blocks this path on v3.14+ because IsStaticVol() doesn't recognize provisionVolume: "false".

Filing an upstream ceph-csi issue to fix this.

[gouthampacha]

Upstream ceph-csi issue filed: ceph/ceph-csi#6347

Requesting that IsStaticVol() recognize provisionVolume: "false" as indicating a pre-provisioned volume, which would skip the format validation and let the existing NewVolumeOptionsFromMonitorList fallback handle Manila CSI volumes correctly.