disk: call sfdisk directly to expand partition

build/multi/Dockerfile.multi

@@ -32,7 +32,7 @@ LABEL maintainers="Alibaba Cloud Authors" description="Alibaba Cloud CSI Plugin"
 
 ARG TARGETARCH
 RUN --mount=type=cache,target=/var/cache/dnf,sharing=locked,id=dnf-cache-$TARGETARCH \
-    dnf install -y ca-certificates file tzdata nfs-utils xfsprogs e4fsprogs pciutils iputils strace cloud-utils-growpart gdisk nc telnet tar cpio lsof && \
+    dnf install -y ca-certificates file tzdata nfs-utils xfsprogs e4fsprogs pciutils iputils strace util-linux nc telnet tar cpio lsof && \
     ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && echo 'Asia/Shanghai' >/etc/timezone
 
 FROM base as build-util-linux

build/multi/Dockerfile.multi.asi

@@ -27,7 +27,7 @@ RUN --mount=type=cache,target=/root/.cache/go-build \
 FROM registry.eu-west-1.aliyuncs.com/acs/alinux:3-update as base
 LABEL maintainers="Alibaba Cloud Authors" description="Alibaba Cloud CSI Plugin"
 
-RUN yum install -y ca-certificates file tzdata nfs-utils xfsprogs e4fsprogs pciutils iputils strace cloud-utils-growpart gdisk nc telnet tar cpio lsof && \
+RUN yum install -y ca-certificates file tzdata nfs-utils xfsprogs e4fsprogs pciutils iputils strace util-linux nc telnet tar cpio lsof && \
     yum clean all
 RUN ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && echo 'Asia/Shanghai' >/etc/timezone
 

pkg/disk/nodeserver.go

@@ -17,14 +17,12 @@ limitations under the License.
 package disk
 
 import (
-	"bytes"
 	"context"
 	"crypto/sha256"
 	"errors"
 	"fmt"
 	"io/fs"
 	"os"
-	"os/exec"
 	"path/filepath"
 	"regexp"
 	"strconv"
@@ -34,6 +32,7 @@ import (
 	"github.com/container-storage-interface/spec/lib/go/csi"
 	"github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/cloud/metadata"
 	"github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/common"
+	"github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/disk/sfdisk"
 	"github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/features"
 	"github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/utils"
 	utilsio "github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/utils/io"
@@ -974,29 +973,22 @@ func localExpandVolume(ctx context.Context, req *csi.NodeExpandVolumeRequest) (*
 		}
 		return nil, status.Errorf(codes.Internal, "NodeExpandVolume: VolumeId: %s, get device name error: %s", req.VolumeId, err.Error())
 	}
+	logger := klog.FromContext(ctx).WithValues("device", devicePath)
+	ctx = klog.NewContext(ctx, logger)
 
-	klog.Infof("NodeExpandVolume:: volumeId: %s, devicePath: %s, volumePath: %s", diskID, devicePath, volumePath)
 	rootPath, index, err := DefaultDeviceManager.GetDeviceRootAndPartitionIndex(devicePath)
 	if err != nil {
 		return nil, status.Errorf(codes.Internal, "GetDeviceRootAndIndex(%s) failed: %v", diskID, err)
 	}
 	if index != "" {
-		output, err := exec.Command("growpart", rootPath, index).CombinedOutput()
+		err := sfdisk.ExpandPartition(ctx, rootPath, index)
 		if err != nil {
-			if bytes.Contains(output, []byte("NOCHANGE")) {
-				if bytes.Contains(output, []byte("it cannot be grown")) || bytes.Contains(output, []byte("could only be grown by")) {
-					deviceCapacity := getBlockDeviceCapacity(devicePath)
-					rootCapacity := getBlockDeviceCapacity(rootPath)
-					klog.Infof("NodeExpandVolume: Volume %s with Device Partition %s no need to grown, with request: %v, root: %v, partition: %v",
-						diskID, devicePath, DiskSize{requestBytes}, DiskSize{rootCapacity}, DiskSize{deviceCapacity})
-					return &csi.NodeExpandVolumeResponse{}, nil
-				}
-			}
-			return nil, status.Errorf(codes.InvalidArgument, "NodeExpandVolume: expand volume %s at %s %s failed: %s, with output %s", diskID, rootPath, index, err.Error(), string(output))
+			return nil, status.Error(codes.Internal, err.Error())
 		}
-		klog.Infof("NodeExpandVolume: Successful expand partition for volume: %s device: %s partition: %s", diskID, rootPath, index)
+		logger.V(2).Info("Successful expand partition", "root", rootPath, "partition", index)
 	}
 
+	klog.V(2).Info("Expand filesystem start", "volumePath", volumePath)
 	// use resizer to expand volume filesystem
 	r := k8smount.NewResizeFs(utilexec.New())
 	ok, err := r.Resize(devicePath, volumePath)

pkg/disk/sfdisk/expand.go

@@ -0,0 +1,53 @@
+package sfdisk
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"os/exec"
+
+	utilsos "github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/utils/os"
+
+	"golang.org/x/sys/unix"
+	"k8s.io/klog/v2"
+)
+
+func ExpandPartition(ctx context.Context, disk, partition string) error {
+	logger := klog.FromContext(ctx)
+	fd, err := unix.Open(disk, unix.O_RDONLY, 0)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if err := unix.Close(fd); err != nil {
+			logger.Error(err, "failed to close", "fd", fd)
+		}
+	}()
+
+	err = unix.Flock(fd, unix.LOCK_EX) // as suggested in the man sfdisk(8)
+	if err != nil {
+		return fmt.Errorf("failed to lock %s exclusively: %v", disk, err)
+	}
+	defer func() {
+		if err := unix.Flock(fd, unix.LOCK_UN); err != nil {
+			logger.Error(err, "failed to unlock", "fd", fd)
+		}
+	}()
+
+	dump, err := exec.CommandContext(ctx, "sfdisk", "--dump", disk).Output()
+	if err != nil {
+		return fmt.Errorf("failed to dump current partition table of %s: %v", disk, utilsos.ErrWithStderr(err))
+	}
+	dumpStr := string(dump)
+	logger.V(4).Info("sfdisk dump before expansion", "dump", dumpStr)
+
+	// Don't cancel this, we don't want to corrupt the partition table
+	cmd := exec.Command("sfdisk", disk, "-N", partition)
+	cmd.Stdin = bytes.NewReader([]byte(",+")) // enlarge the partition as much as possible
+	result, err := cmd.Output()
+	if err != nil {
+		return fmt.Errorf("failed to expand partition %s on %s: %v\noriginal table looked like:\n%s", partition, disk, utilsos.ErrWithStderr(err), dumpStr)
+	}
+	logger.V(3).Info("sfdisk success", "output", string(result))
+	return nil
+}

pkg/disk/sfdisk/expand_test.go

@@ -0,0 +1,41 @@
+package sfdisk
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"os"
+	"os/exec"
+	"strings"
+	"testing"
+
+	utilsos "github.com/kubernetes-sigs/alibaba-cloud-csi-driver/pkg/utils/os"
+	"github.com/stretchr/testify/assert"
+)
+
+func TestExpandPartition(t *testing.T) {
+	path, err := exec.LookPath("sfdisk")
+	if errors.Is(err, exec.ErrNotFound) {
+		t.Skip("sfdisk not found")
+	}
+	assert.NoError(t, err)
+	t.Logf("sfdisk found at: %s", path)
+
+	testImage := t.TempDir() + "/test.img"
+	_, err = os.Create(testImage)
+	assert.NoError(t, err)
+	assert.NoError(t, os.Truncate(testImage, 1<<23)) // 8MB
+
+	cmd := exec.Command("sfdisk", testImage)
+	cmd.Stdin = bytes.NewReader([]byte("label: gpt\n,\n")) // create a single partition
+	result, err := cmd.Output()
+	assert.NoError(t, utilsos.ErrWithStderr(err))
+	t.Logf("create partition success: %s", string(result))
+
+	assert.NoError(t, os.Truncate(testImage, 1<<24)) // expand to 16MB
+	assert.NoError(t, ExpandPartition(context.Background(), testImage, "1"))
+
+	dump, err := exec.Command("sfdisk", "--dump", testImage).Output()
+	assert.NoError(t, utilsos.ErrWithStderr(err))
+	assert.Contains(t, strings.ReplaceAll(string(dump), " ", ""), "test.img1:start=2048,size=30687")
+}

pkg/utils/os/exec.go

@@ -0,0 +1,23 @@
+package os
+
+import (
+	"bytes"
+	"errors"
+	"os/exec"
+)
+
+type ExitErrorWithStderr struct {
+	*exec.ExitError
+}
+
+func (err ExitErrorWithStderr) Error() string {
+	return err.ExitError.Error() + ", with stderr: " + string(bytes.TrimSpace(err.Stderr))
+}
+
+func ErrWithStderr(err error) error {
+	var exitErr *exec.ExitError
+	if errors.As(err, &exitErr) && len(exitErr.Stderr) > 0 {
+		return ExitErrorWithStderr{exitErr}
+	}
+	return err
+}

pkg/utils/os/exec_test.go

@@ -0,0 +1,19 @@
+package os
+
+import (
+	"os/exec"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestErrWithStderr(t *testing.T) {
+	_, err := exec.Command("sh", "-c", "echo 'test error' 1>&2; exit 2").Output()
+	assert.Error(t, err)
+	assert.EqualError(t, ErrWithStderr(err), "exit status 2, with stderr: test error")
+}
+
+func TestErrWithStderrNoChange(t *testing.T) {
+	err := exec.ErrNotFound
+	assert.Equal(t, ErrWithStderr(err), err)
+}