fix: lock provisioning order of user disk partitions

Fixes #9877 As a side-effect, fix alignment of user disks for newer QEMU versions. Signed-off-by: Andrey Smirnov <[email protected]>
siderolabs · Dec 5, 2024 · dd61ad8 · dd61ad8
1 parent d0773ff
commit dd61ad8
Show file tree

Hide file tree

Showing 5 changed files with 19 additions and 8 deletions.
diff --git a/cmd/talosctl/cmd/mgmt/cluster/create.go b/cmd/talosctl/cmd/mgmt/cluster/create.go
@@ -1162,6 +1162,8 @@ func parseCPUShare(cpus string) (int64, error) {
 }
 
 func getDisks() ([]*provision.Disk, error) {
+	const GPTAlignment = 2 * 1024 * 1024 // 2 MB
+
 	// should have at least a single primary disk
 	disks := []*provision.Disk{
 		{
@@ -1211,8 +1213,8 @@ func getDisks() ([]*provision.Disk, error) {
 		}
 
 		disks = append(disks, &provision.Disk{
-			// add 1 MB to make extra room for GPT and alignment
-			Size:            diskSize + 2*1024*1024,
+			// add 2 MB per partition to make extra room for GPT and alignment
+			Size:            diskSize + GPTAlignment*uint64(len(diskPartitions)+1),
 			Partitions:      diskPartitions,
 			SkipPreallocate: !clusterDiskPreallocate,
 			Driver:          "ide",

diff --git a/internal/app/machined/pkg/controllers/block/user_disk_config.go b/internal/app/machined/pkg/controllers/block/user_disk_config.go
@@ -105,7 +105,11 @@ func (ctrl *UserDiskConfigController) Run(ctx context.Context, r controller.Runt
 							vc.TypedSpec().Type = block.VolumeTypePartition
 
 							vc.TypedSpec().Provisioning = block.ProvisioningSpec{
-								Wave: block.WaveUserDisks,
+								// it's crucial to keep the order of provisioning locked within each disk, otherwise
+								// provisioning might order them different way, and create partitions in wrong order
+								// the matcher on partition idx would then discover partitions in wrong order, and mount them
+								// in wrong order
+								Wave: block.WaveLegacyUserDisks + idx,
 								DiskSelector: block.DiskSelector{
 									Match: diskPathMatch(resolvedDevicePath),
 								},

diff --git a/internal/app/machined/pkg/controllers/block/user_disk_config_test.go b/internal/app/machined/pkg/controllers/block/user_disk_config_test.go
@@ -132,6 +132,7 @@ func (suite *UserDiskConfigSuite) TestReconcileUserDisk() {
 		ctest.AssertResource(suite, id, func(r *block.VolumeConfig, asrt *assert.Assertions) {
 			asrt.NotEmpty(r.TypedSpec().Provisioning)
 			asrt.Contains(r.Metadata().Labels().Raw(), block.UserDiskLabel)
+			asrt.GreaterOrEqual(r.TypedSpec().Provisioning.Wave, block.WaveLegacyUserDisks)
 		})
 	}
 

diff --git a/pkg/machinery/resources/block/volume_config.go b/pkg/machinery/resources/block/volume_config.go
@@ -45,8 +45,9 @@ type VolumeConfigSpec struct {
 
 // Wave constants.
 const (
-	WaveSystemDisk = -1
-	WaveUserDisks  = 0
+	WaveSystemDisk      = -1
+	WaveUserDisks       = 0
+	WaveLegacyUserDisks = 1000000 // legacy user disks rely on specific order of provisioning
 )
 
 // ProvisioningSpec is the spec for volume provisioning.

diff --git a/pkg/provision/providers/vm/disk.go b/pkg/provision/providers/vm/disk.go
@@ -24,10 +24,13 @@ func (p *Provisioner) UserDiskName(index int) string {
 
 // CreateDisks creates empty disk files for each disk.
 func (p *Provisioner) CreateDisks(state *State, nodeReq provision.NodeRequest) (diskPaths []string, err error) {
+	const QEMUAlignment = 4 * 1024 * 1024 // 4 MiB, required by QEMU
+
 	diskPaths = make([]string, len(nodeReq.Disks))
 
 	for i, disk := range nodeReq.Disks {
 		diskPath := state.GetRelativePath(fmt.Sprintf("%s-%d.disk", nodeReq.Name, i))
+		diskSize := (disk.Size + QEMUAlignment - 1) / QEMUAlignment * QEMUAlignment
 
 		var diskF *os.File
 
@@ -38,13 +41,13 @@ func (p *Provisioner) CreateDisks(state *State, nodeReq provision.NodeRequest) (
 
 		defer diskF.Close() //nolint:errcheck
 
-		if err = diskF.Truncate(int64(disk.Size)); err != nil {
+		if err = diskF.Truncate(int64(diskSize)); err != nil {
 			return nil, err
 		}
 
 		if !disk.SkipPreallocate {
-			if err = syscall.Fallocate(int(diskF.Fd()), 0, 0, int64(disk.Size)); err != nil {
-				fmt.Fprintf(os.Stderr, "WARNING: failed to preallocate disk space for %q (size %d): %s", diskPath, disk.Size, err)
+			if err = syscall.Fallocate(int(diskF.Fd()), 0, 0, int64(diskSize)); err != nil {
+				fmt.Fprintf(os.Stderr, "WARNING: failed to preallocate disk space for %q (size %d): %s", diskPath, diskSize, err)
 			}
 		}