feat: dual boot disk image

Generate disk image with both grub and sd-boot.

Fixes: #10332

Signed-off-by: Noel Georgi <[email protected]>

.github/workflows/ci.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2025-02-15T06:21:12Z by kres 8a48729.
+# Generated on 2025-02-16T17:00:53Z by kres 8a48729.
 
 name: default
 concurrency:
@@ -2232,6 +2232,17 @@ jobs:
           WITH_DISK_ENCRYPTION: "true"
         run: |
           sudo -E make e2e-qemu
+      - name: e2e-disk-image-bios
+        env:
+          GITHUB_STEP_NAME: ${{ github.job}}-e2e-disk-image-bios
+          IMAGE_REGISTRY: registry.dev.siderolabs.io
+          SHORT_INTEGRATION_TEST: "yes"
+          USE_DISK_IMAGE: "true"
+          VIA_MAINTENANCE_MODE: "true"
+          WITH_DISK_ENCRYPTION: "true"
+          WITH_UEFI: "false"
+        run: |
+          sudo -E make e2e-qemu
       - name: e2e-node-address-v2
         env:
           GITHUB_STEP_NAME: ${{ github.job}}-e2e-disk-image
@@ -2856,8 +2867,10 @@ jobs:
           QEMU_EXTRA_DISKS: "3"
           QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
           QEMU_EXTRA_DISKS_SIZE: "10240"
+          USER_DISKS_MOUNTS: /var/lib/extra,/var/lib/p1,/var/lib/p2
           WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml:@hack/test/patches/dm-raid-module.yaml'
           WITH_JSON_LOGS: "true"
+          WITH_USER_DISK: "true"
         run: |
           sudo -E make e2e-qemu
       - name: save artifacts

.github/workflows/integration-misc-2-cron.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2025-02-15T06:21:12Z by kres 8a48729.
+# Generated on 2025-02-15T07:52:36Z by kres 8a48729.
 
 name: integration-misc-2-cron
 concurrency:
@@ -125,6 +125,17 @@ jobs:
           WITH_DISK_ENCRYPTION: "true"
         run: |
           sudo -E make e2e-qemu
+      - name: e2e-disk-image-bios
+        env:
+          GITHUB_STEP_NAME: ${{ github.job}}-e2e-disk-image-bios
+          IMAGE_REGISTRY: registry.dev.siderolabs.io
+          SHORT_INTEGRATION_TEST: "yes"
+          USE_DISK_IMAGE: "true"
+          VIA_MAINTENANCE_MODE: "true"
+          WITH_DISK_ENCRYPTION: "true"
+          WITH_UEFI: "false"
+        run: |
+          sudo -E make e2e-qemu
       - name: e2e-node-address-v2
         env:
           GITHUB_STEP_NAME: ${{ github.job}}-e2e-disk-image

.github/workflows/integration-qemu-cron.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-11-28T12:57:52Z by kres 232fe63.
+# Generated on 2025-02-16T17:00:53Z by kres 8a48729.
 
 name: integration-qemu-cron
 concurrency:
@@ -84,8 +84,10 @@ jobs:
           QEMU_EXTRA_DISKS: "3"
           QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
           QEMU_EXTRA_DISKS_SIZE: "10240"
+          USER_DISKS_MOUNTS: /var/lib/extra,/var/lib/p1,/var/lib/p2
           WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml:@hack/test/patches/dm-raid-module.yaml'
           WITH_JSON_LOGS: "true"
+          WITH_USER_DISK: "true"
         run: |
           sudo -E make e2e-qemu
       - name: save artifacts

.kres.yaml

@@ -337,6 +337,8 @@ spec:
             QEMU_EXTRA_DISKS_DRIVERS: "ide,nvme"
             WITH_CONFIG_PATCH_WORKER: "@hack/test/patches/ephemeral-nvme.yaml:@hack/test/patches/dm-raid-module.yaml"
             WITH_JSON_LOGS: "true"
+            WITH_USER_DISK: "true"
+            USER_DISKS_MOUNTS: "/var/lib/extra,/var/lib/p1,/var/lib/p2"
         - name: save-talos-logs
           conditions:
             - always
@@ -870,6 +872,17 @@ spec:
             VIA_MAINTENANCE_MODE: true
             WITH_DISK_ENCRYPTION: true
             IMAGE_REGISTRY: registry.dev.siderolabs.io
+        - name: e2e-disk-image-bios
+          command: e2e-qemu
+          withSudo: true
+          environment:
+            GITHUB_STEP_NAME: ${{ github.job}}-e2e-disk-image-bios
+            SHORT_INTEGRATION_TEST: yes
+            USE_DISK_IMAGE: true
+            VIA_MAINTENANCE_MODE: true
+            WITH_DISK_ENCRYPTION: true
+            WITH_UEFI: false
+            IMAGE_REGISTRY: registry.dev.siderolabs.io
         - name: e2e-node-address-v2
           command: e2e-qemu
           withSudo: true

cmd/installer/pkg/install/install.go

@@ -61,11 +61,12 @@ type Options struct {
 	ImageCachePath string
 
 	// Options specific for the image creation mode.
-	ImageSecureboot bool
-	Version         string
-	BootAssets      bootloaderoptions.BootAssets
-	Printf          func(string, ...any)
-	MountPrefix     string
+	ImageSecureboot     bool
+	DiskImageBootloader string
+	Version             string
+	BootAssets          bootloaderoptions.BootAssets
+	Printf              func(string, ...any)
+	MountPrefix         string
 }
 
 // Mode is the install mode.
@@ -285,7 +286,10 @@ func (i *Installer) Install(ctx context.Context, mode Mode) (err error) {
 	switch mode {
 	case ModeImage:
 		// on image creation, we don't care about disk contents
-		bootlder = bootloader.New(i.options.ImageSecureboot, i.options.Version)
+		bootlder, err = bootloader.New(i.options.DiskImageBootloader, i.options.Version, i.options.Arch)
+		if err != nil {
+			return fmt.Errorf("failed to create bootloader: %w", err)
+		}
 	case ModeInstall:
 		if !i.options.Zero && !i.options.Force {
 			// verify that the disk is either empty or has an empty GPT partition table, otherwise fail the install

go.mod

@@ -145,7 +145,7 @@ require (
 	github.com/siderolabs/gen v0.8.0
 	github.com/siderolabs/go-api-signature v0.3.6
 	github.com/siderolabs/go-blockdevice v0.4.8
-	github.com/siderolabs/go-blockdevice/v2 v2.0.14
+	github.com/siderolabs/go-blockdevice/v2 v2.0.15
 	github.com/siderolabs/go-circular v0.2.1
 	github.com/siderolabs/go-cmd v0.1.3
 	github.com/siderolabs/go-copy v0.1.0

go.sum

@@ -647,8 +647,8 @@ github.com/siderolabs/go-api-signature v0.3.6 h1:wDIsXbpl7Oa/FXvxB6uz4VL9INA9fmr
 github.com/siderolabs/go-api-signature v0.3.6/go.mod h1:hoH13AfunHflxbXfh+NoploqV13ZTDfQ1mQJWNVSW9U=
 github.com/siderolabs/go-blockdevice v0.4.8 h1:KfdWvIx0Jft5YVuCsFIJFwjWEF1oqtzkgX9PeU9cX4c=
 github.com/siderolabs/go-blockdevice v0.4.8/go.mod h1:4PeOuk71pReJj1JQEXDE7kIIQJPVe8a+HZQa+qjxSEA=
-github.com/siderolabs/go-blockdevice/v2 v2.0.14 h1:9Nu4ceeKpCSUhSub6RbxU2eat5IwAOR11Vdb5mPVASo=
-github.com/siderolabs/go-blockdevice/v2 v2.0.14/go.mod h1:74htzCV913UzaLZ4H+NBXkwWlYnBJIq5m/379ZEcu8w=
+github.com/siderolabs/go-blockdevice/v2 v2.0.15 h1:22EZ3w125OdPBobckwGZOILhAGo4xUdAn9/OQ5tR/Ys=
+github.com/siderolabs/go-blockdevice/v2 v2.0.15/go.mod h1:74htzCV913UzaLZ4H+NBXkwWlYnBJIq5m/379ZEcu8w=
 github.com/siderolabs/go-circular v0.2.1 h1:a++iVCn9jyhICX3POQZZX8n72p2h5JGdGU6w1ulmpcA=
 github.com/siderolabs/go-circular v0.2.1/go.mod h1:ZDItzVyXK+B/XuqTBV5MtQtSv06VI+oCmWGRnNCATo8=
 github.com/siderolabs/go-cmd v0.1.3 h1:JrgZwqhJQeoec3QRON0LK+fv+0y7d0DyY7zsfkO6ciw=

hack/release.toml

@@ -101,6 +101,29 @@ The NQN can be read by `talosctl read /etc/nvme/hostnqn`
         title = "ISO"
         description = """\
 Talos starting with 1.10 will have ISO's that will use GRUB only for legacy BIOS and systemd-boot for modern UEFI systems.
+"""
+
+    [notes.disk-image]
+        title = "Disk Image"
+        description = """\
+Talos starting with 1.10 will have disk images that will use GRUB only for legacy BIOS and systemd-boot for modern UEFI systems.
+On first boot Talos determines the boot method and will wipe the unused bootloader.
+
+Secureboot disk-images will be sd-boot only.
+
+For ARM64 imager will still generate GRUB bootloader for Talos < 1.10 and for Talos >= 1.10 all ARM64 boot assets will use systemd-boot.
+
+Imager supports overwriting bootloader when generating a disk image via the Imager profile `output` option.
+
+Eg:
+
+```yaml
+output:
+  kind: image
+  imageOptions:
+    bootloader: sd-boot # supported options are sd-boot, grub, dual-boot
+```
+
 """
 
     [notes.ethernet]

hack/test/e2e-qemu.sh

@@ -2,8 +2,6 @@
 
 set -eou pipefail
 
-export USER_DISKS_MOUNTS="/var/lib/extra,/var/lib/p1,/var/lib/p2"
-
 # shellcheck source=/dev/null
 source ./hack/test/e2e.sh
 
@@ -248,6 +246,15 @@ case "${WITH_UKI_BOOT:-false}" in
     ;;
 esac
 
+case "${WITH_USER_DISK:-false}" in
+  false)
+    ;;
+  *)
+    QEMU_FLAGS+=("--user-disk=/var/lib/extra:350MB")
+    QEMU_FLAGS+=("--user-disk=/var/lib/p1:350MB:/var/lib/p2:350MB")
+    ;;
+esac
+
 
 function create_cluster {
   build_registry_mirrors
@@ -268,8 +275,6 @@ function create_cluster {
     --cpus="${QEMU_CPUS:-2}" \
     --cpus-workers="${QEMU_CPUS_WORKERS:-2}" \
     --cidr=172.20.1.0/24 \
-    --user-disk=/var/lib/extra:350MB \
-    --user-disk=/var/lib/p1:350MB:/var/lib/p2:350MB \
     --install-image="${INSTALLER_IMAGE}" \
     --with-init-node=false \
     --cni-bundle-url="${ARTIFACTS}/talosctl-cni-bundle-\${ARCH}.tar.gz" \

internal/app/machined/pkg/runtime/v1alpha1/bootloader/bootloader.go

@@ -6,13 +6,20 @@
 package bootloader
 
 import (
+	"fmt"
 	"os"
 
+	"github.com/siderolabs/go-blockdevice/v2/block"
+	"github.com/siderolabs/go-blockdevice/v2/partitioning/gpt"
+
 	"github.com/siderolabs/talos/internal/app/machined/pkg/runtime"
+	"github.com/siderolabs/talos/internal/app/machined/pkg/runtime/v1alpha1/bootloader/dual"
 	"github.com/siderolabs/talos/internal/app/machined/pkg/runtime/v1alpha1/bootloader/grub"
 	"github.com/siderolabs/talos/internal/app/machined/pkg/runtime/v1alpha1/bootloader/options"
 	"github.com/siderolabs/talos/internal/app/machined/pkg/runtime/v1alpha1/bootloader/sdboot"
 	"github.com/siderolabs/talos/internal/pkg/partition"
+	"github.com/siderolabs/talos/pkg/imager/profile"
+	"github.com/siderolabs/talos/pkg/machinery/constants"
 	"github.com/siderolabs/talos/pkg/machinery/imager/quirks"
 )
 
@@ -67,14 +74,85 @@ func NewAuto() Bootloader {
 	return grub.NewConfig()
 }
 
-// New returns a new bootloader based on the secureboot flag.
-func New(secureboot bool, talosVersion string) Bootloader {
-	if secureboot {
-		return sdboot.New()
+// New returns a new bootloader based on the secureboot flag and architecture.
+func New(bootloader, talosVersion, arch string) (Bootloader, error) {
+	switch bootloader {
+	case profile.DiskImageBootloaderGrub.String():
+		g := grub.NewConfig()
+		g.AddResetOption = quirks.New(talosVersion).SupportsResetGRUBOption()
+
+		return g, nil
+	case profile.DiskImageBootloaderSDBoot.String():
+		return sdboot.New(), nil
+	case profile.DiskImageBootloaderDualBoot.String():
+		return dual.New(), nil
+	default:
+		return nil, fmt.Errorf("unsupported bootloader %q", bootloader)
+	}
+}
+
+// CleanupBootloader cleans up the alternate bootloader when booting off via BIOS or UEFI.
+func CleanupBootloader(disk string, sdboot bool) error {
+	dev, err := block.NewFromPath(disk, block.OpenForWrite())
+	if err != nil {
+		return err
+	}
+
+	defer dev.Close() //nolint:errcheck
+
+	if err := dev.Lock(true); err != nil {
+		return fmt.Errorf("failed to lock device: %w", err)
+	}
+
+	defer dev.Unlock() //nolint:errcheck
+
+	gptDev, err := gpt.DeviceFromBlockDevice(dev)
+	if err != nil {
+		return fmt.Errorf("failed to get GPT device: %w", err)
+	}
+
+	gptTable, err := gpt.Read(gptDev)
+	if err != nil {
+		return fmt.Errorf("failed to read GPT: %w", err)
+	}
+
+	if sdboot {
+		// we wipe upto 446 bytes where the protective MBR is located
+		if _, err := dev.WipeRange(0, 446); err != nil {
+			return fmt.Errorf("failed to wipe MBR: %w", err)
+		}
+
+		if err := deletePartitions(gptTable, constants.BIOSGrubPartitionLabel, constants.BootPartitionLabel); err != nil {
+			return err
+		}
+	} else {
+		// means we are using GRUB
+		if err := deletePartitions(gptTable, constants.EFIPartitionLabel); err != nil {
+			return err
+		}
 	}
 
-	g := grub.NewConfig()
-	g.AddResetOption = quirks.New(talosVersion).SupportsResetGRUBOption()
+	if err := gptTable.Write(); err != nil {
+		return fmt.Errorf("failed to write GPT: %w", err)
+	}
+
+	return nil
+}
+
+func deletePartitions(gptTable *gpt.Table, labels ...string) error {
+	for i, part := range gptTable.Partitions() {
+		if part == nil {
+			continue
+		}
+
+		for _, label := range labels {
+			if part.Name == label {
+				if err := gptTable.DeletePartition(i); err != nil {
+					return fmt.Errorf("failed to delete partition %s %d: %w", part.Name, i, err)
+				}
+			}
+		}
+	}
 
-	return g
+	return nil
 }