WIP: Query PCIe ports for correct amount of devices that can be attached to the node

Signed-off-by: Niclas Schad <niclas.schad@stackit.cloud>

Author: nschad

pkg/csi/blockstorage/controllerserver.go

@@ -370,6 +370,11 @@ func (cs *controllerServer) ControllerPublishVolume(ctx context.Context, req *cs
 
 	_, err = cloud.AttachVolume(ctx, instanceID, volumeID)
 	if err != nil {
+		// Trigger's an immediate `NodeGetInfo` RPC call when MutableCSINodeAllocatableCount is enabled
+		// TODO: Finish Implementation of IsTooManyDevicesError
+		//if stackiterrors.IsTooManyDevicesError(err) {
+		//	return nil, status.Errorf(codes.ResourceExhausted, "[ControllerPublishVolume] Node can't accept any more volumes %v. All PCIe lanes are exhausted!", err)
+		//}
 		klog.Errorf("Failed to AttachVolume: %v", err)
 		return nil, status.Errorf(codes.Internal, "[ControllerPublishVolume] Attach Volume failed with error %v", err)
 	}

pkg/csi/blockstorage/nodeserver.go

@@ -308,8 +308,20 @@ func (ns *nodeServer) NodeGetInfo(ctx context.Context, _ *csi.NodeGetInfoRequest
 	}
 
 	maxVolumesPerNode := DetermineMaxVolumesByFlavor(flavor)
-	// Subtract 1 for root disk and another for configDrive/spare
-	maxVolumesPerNode -= 2
+
+	// Subtract already mounted Volumes
+	emptyPCIeRootPorts, err := mount.GetEmptyPCIeRootPorts()
+	if err != nil {
+		klog.Errorf("[NodeGetInfo] unable to retrieve PCIe root ports %v", err)
+	} else {
+		if emptyPCIeRootPorts <= 0 {
+			klog.Warningf("[NodeGetInfo] node does not have any free PCIe ports")
+			maxVolumesPerNode = 0
+		} else {
+			maxVolumesPerNode = emptyPCIeRootPorts
+		}
+	}
+
 	klog.V(4).Infof("Determined node to support %d volumes", maxVolumesPerNode)
 
 	nodeInfo := &csi.NodeGetInfoResponse{

pkg/csi/blockstorage/utils.go

@@ -96,7 +96,7 @@ func DetermineMaxVolumesByFlavor(flavor string) int64 {
 		// AMD 2nd Gen
 		return 159
 	default:
-		// All other flavors can mount 28 volumes
+		// All other flavors can mount 25 volumes
 		return 25
 	}
 }

pkg/csi/util/mount/mount.go

@@ -20,6 +20,7 @@ import (
 	"fmt"
 	"os"
 	"path"
+	"path/filepath"
 	"slices"
 	"strings"
 	"time"
@@ -119,6 +120,57 @@ func probeVolume() error {
 	return nil
 }
 
+// GetEmptyPCIeRootPorts returns the number of PCIe Root ports who currently
+// do not have ANY downstream devices, therefore considered "empty". This
+// function is used dynamically how many more devices (in this case "volumes")
+// can be mounted to the given node.
+func GetEmptyPCIeRootPorts() (int64, error) {
+	const pciPath = "/sys/bus/pci/devices"
+
+	// Get all PCI devices
+	devices, err := os.ReadDir(pciPath)
+	if err != nil {
+		return 0, fmt.Errorf("failed to read PCI bus: %w", err)
+	}
+
+	emptyCount := 0
+
+	for _, dev := range devices {
+		devPath := filepath.Join(pciPath, dev.Name())
+
+		// 1. Identify if it's a Root Port / Bridge
+		// We check the 'class' file. PCI Bridge class code starts with 0x0604
+		classBuf, err := os.ReadFile(filepath.Join(devPath, "class"))
+		if err != nil {
+			continue
+		}
+		class := strings.TrimSpace(string(classBuf))
+
+		// Class 0x060400 is a PCI-to-PCI bridge (standard for Root Ports)
+		if strings.HasPrefix(class, "0x0604") {
+
+			// 2. Check if the port has downstream devices
+			// If the bridge has children, they appear as subdirectories
+			// matching the PCI address format (e.g., 0000:01:00.0)
+			hasChild := false
+			files, _ := os.ReadDir(devPath)
+			for _, file := range files {
+				// PCI addresses always contain a colon
+				if strings.Contains(file.Name(), ":") {
+					hasChild = true
+					break
+				}
+			}
+
+			if !hasChild {
+				emptyCount++
+			}
+		}
+	}
+
+	return int64(emptyCount), nil
+}
+
 // GetDevicePath returns the path of an attached block storage volume, specified by its id.
 func (m *Mount) GetDevicePath(volumeID string) (string, error) {
 	backoff := wait.Backoff{

pkg/stackit/stackiterrors/errors.go

@@ -4,9 +4,10 @@ import (
 	"errors"
 	"fmt"
 	"net/http"
+	"strings"
 
 	oapiError "github.com/stackitcloud/stackit-sdk-go/core/oapierror"
-	wait "github.com/stackitcloud/stackit-sdk-go/services/iaas/v2api/wait"
+	"github.com/stackitcloud/stackit-sdk-go/services/iaas/v2api/wait"
 )
 
 var ErrNotFound = errors.New("failed to find object")
@@ -20,6 +21,17 @@ func IsNotFound(err error) bool {
 	return oAPIError.StatusCode == http.StatusNotFound
 }
 
+func IsTooManyDevicesError(err error) bool {
+	var oAPIError *oapiError.GenericOpenAPIError
+	if ok := errors.As(err, &oAPIError); !ok {
+		return false
+	}
+
+	// TODO: This is just a placeholder. Implement this correctly
+	return oAPIError.StatusCode == http.StatusInternalServerError &&
+		strings.Contains(oAPIError.ErrorMessage, "devices")
+}
+
 func IgnoreNotFound(err error) error {
 	if IsNotFound(err) {
 		return nil