banger/internal/daemon/capabilities.go

package daemon

import (
	"context"
	"errors"
	"net"
	"os"
	"strings"

	"banger/internal/firecracker"
	"banger/internal/guestconfig"
	"banger/internal/model"
	"banger/internal/system"
	"banger/internal/vmdns"
)

type vmCapability interface {
	Name() string
}

type startPreflightCapability interface {
	AddStartPreflight(context.Context, *Daemon, *system.Preflight, model.VMRecord, model.Image)
}

type guestConfigCapability interface {
	ContributeGuest(*guestconfig.Builder, model.VMRecord, model.Image)
}

type machineConfigCapability interface {
	ContributeMachine(*firecracker.MachineConfig, model.VMRecord, model.Image)
}

type prepareHostCapability interface {
	PrepareHost(context.Context, *Daemon, *model.VMRecord, model.Image) error
}

type postStartCapability interface {
	PostStart(context.Context, *Daemon, model.VMRecord, model.Image) error
}

type cleanupCapability interface {
	Cleanup(context.Context, *Daemon, model.VMRecord) error
}

type configChangeCapability interface {
	ApplyConfigChange(context.Context, *Daemon, model.VMRecord, model.VMRecord) error
}

type doctorCapability interface {
	AddDoctorChecks(context.Context, *Daemon, *system.Report)
}

func (d *Daemon) registeredCapabilities() []vmCapability {
	if len(d.vmCaps) > 0 {
		return d.vmCaps
	}
	return []vmCapability{
		workDiskCapability{},
		dnsCapability{},
		natCapability{},
	}
}

func (d *Daemon) addCapabilityStartPrereqs(ctx context.Context, checks *system.Preflight, vm model.VMRecord, image model.Image) {
	for _, capability := range d.registeredCapabilities() {
		if hook, ok := capability.(startPreflightCapability); ok {
			hook.AddStartPreflight(ctx, d, checks, vm, image)
		}
	}
}

func (d *Daemon) contributeGuestConfig(builder *guestconfig.Builder, vm model.VMRecord, image model.Image) {
	for _, capability := range d.registeredCapabilities() {
		if hook, ok := capability.(guestConfigCapability); ok {
			hook.ContributeGuest(builder, vm, image)
		}
	}
}

func (d *Daemon) contributeMachineConfig(cfg *firecracker.MachineConfig, vm model.VMRecord, image model.Image) {
	for _, capability := range d.registeredCapabilities() {
		if hook, ok := capability.(machineConfigCapability); ok {
			hook.ContributeMachine(cfg, vm, image)
		}
	}
}

func (d *Daemon) prepareCapabilityHosts(ctx context.Context, vm *model.VMRecord, image model.Image) error {
	prepared := make([]vmCapability, 0, len(d.registeredCapabilities()))
	for _, capability := range d.registeredCapabilities() {
		hook, ok := capability.(prepareHostCapability)
		if !ok {
			continue
		}
		if err := hook.PrepareHost(ctx, d, vm, image); err != nil {
			d.cleanupPreparedCapabilities(context.Background(), vm, prepared)
			return err
		}
		prepared = append(prepared, capability)
	}
	return nil
}

func (d *Daemon) postStartCapabilities(ctx context.Context, vm model.VMRecord, image model.Image) error {
	for _, capability := range d.registeredCapabilities() {
		switch capability.Name() {
		case "dns":
			vmCreateStage(ctx, "apply_dns", "publishing vm dns record")
		case "nat":
			if vm.Spec.NATEnabled {
				vmCreateStage(ctx, "apply_nat", "configuring nat")
			}
		}
		if hook, ok := capability.(postStartCapability); ok {
			if err := hook.PostStart(ctx, d, vm, image); err != nil {
				return err
			}
		}
	}
	return nil
}

func (d *Daemon) cleanupCapabilityState(ctx context.Context, vm model.VMRecord) error {
	return d.cleanupPreparedCapabilities(ctx, &vm, d.registeredCapabilities())
}

func (d *Daemon) cleanupPreparedCapabilities(ctx context.Context, vm *model.VMRecord, capabilities []vmCapability) error {
	var err error
	for index := len(capabilities) - 1; index >= 0; index-- {
		hook, ok := capabilities[index].(cleanupCapability)
		if !ok {
			continue
		}
		err = joinErr(err, hook.Cleanup(ctx, d, *vm))
	}
	return err
}

func (d *Daemon) applyCapabilityConfigChanges(ctx context.Context, before, after model.VMRecord) error {
	for _, capability := range d.registeredCapabilities() {
		if hook, ok := capability.(configChangeCapability); ok {
			if err := hook.ApplyConfigChange(ctx, d, before, after); err != nil {
				return err
			}
		}
	}
	return nil
}

func (d *Daemon) addCapabilityDoctorChecks(ctx context.Context, report *system.Report) {
	for _, capability := range d.registeredCapabilities() {
		if hook, ok := capability.(doctorCapability); ok {
			hook.AddDoctorChecks(ctx, d, report)
		}
	}
}

type workDiskCapability struct{}

func (workDiskCapability) Name() string { return "work-disk" }

func (workDiskCapability) AddStartPreflight(_ context.Context, _ *Daemon, checks *system.Preflight, vm model.VMRecord, image model.Image) {
	if exists(vm.Runtime.WorkDiskPath) {
		return
	}
	imageSeed := ""
	if image.RootfsPath != "" {
		imageSeed = image.WorkSeedPath
	}
	if exists(imageSeed) {
		if info, err := os.Stat(imageSeed); err == nil && vm.Spec.WorkDiskSizeBytes > info.Size() {
			checks.RequireCommand("e2fsck", toolHint("e2fsck"))
			checks.RequireCommand("resize2fs", toolHint("resize2fs"))
		}
		return
	}
	for _, command := range []string{"mkfs.ext4", "mount", "umount", "cp"} {
		checks.RequireCommand(command, toolHint(command))
	}
}

func (workDiskCapability) ContributeGuest(builder *guestconfig.Builder, _ model.VMRecord, _ model.Image) {
	builder.AddMount(guestconfig.MountSpec{
		Source:  "/dev/vdb",
		Target:  "/root",
		FSType:  "ext4",
		Options: []string{"defaults"},
		Dump:    0,
		Pass:    2,
	})
}

func (workDiskCapability) ContributeMachine(cfg *firecracker.MachineConfig, vm model.VMRecord, _ model.Image) {
	cfg.Drives = append(cfg.Drives, firecracker.DriveConfig{
		ID:       "work",
		Path:     vm.Runtime.WorkDiskPath,
		ReadOnly: false,
	})
}

func (workDiskCapability) PrepareHost(ctx context.Context, d *Daemon, vm *model.VMRecord, image model.Image) error {
	prep, err := d.vm.ensureWorkDisk(ctx, vm, image)
	if err != nil {
		return err
	}
	if err := d.ws.ensureAuthorizedKeyOnWorkDisk(ctx, vm, image, prep); err != nil {
		return err
	}
	if err := d.ws.ensureGitIdentityOnWorkDisk(ctx, vm); err != nil {
		return err
	}
	return d.ws.runFileSync(ctx, vm)
}

func (workDiskCapability) AddDoctorChecks(_ context.Context, d *Daemon, report *system.Report) {
	if d.store != nil && strings.TrimSpace(d.config.DefaultImageName) != "" {
		if image, err := d.store.GetImageByName(context.Background(), d.config.DefaultImageName); err == nil && strings.TrimSpace(image.WorkSeedPath) != "" && exists(image.WorkSeedPath) {
			checks := system.NewPreflight()
			checks.RequireFile(image.WorkSeedPath, "default image work-seed", `rebuild the default image to regenerate the /root seed`)
			report.AddPreflight("feature /root work disk", checks, "seeded /root work disk artifact available")
			return
		}
	}
	checks := system.NewPreflight()
	for _, command := range []string{"mkfs.ext4", "mount", "umount", "cp"} {
		checks.RequireCommand(command, toolHint(command))
	}
	report.AddPreflight("feature /root work disk", checks, "fallback /root work disk tooling available")
	report.AddWarn("feature /root work disk", "default image has no work-seed artifact; new VM creates will be slower until the image is rebuilt")
}

type dnsCapability struct{}

func (dnsCapability) Name() string { return "dns" }

func (dnsCapability) PostStart(ctx context.Context, d *Daemon, vm model.VMRecord, _ model.Image) error {
	return d.net.setDNS(ctx, vm.Name, vm.Runtime.GuestIP)
}

func (dnsCapability) Cleanup(_ context.Context, d *Daemon, vm model.VMRecord) error {
	return d.net.removeDNS(vm.Runtime.DNSName)
}

func (dnsCapability) AddDoctorChecks(_ context.Context, _ *Daemon, report *system.Report) {
	conn, err := net.ListenPacket("udp", vmdns.DefaultListenAddr)
	if err != nil {
		if strings.Contains(strings.ToLower(err.Error()), "address already in use") {
			report.AddWarn("feature vm dns", "listener address "+vmdns.DefaultListenAddr+" is already in use")
			return
		}
		report.AddFail("feature vm dns", "cannot bind "+vmdns.DefaultListenAddr+": "+err.Error())
		return
	}
	_ = conn.Close()
	report.AddPass("feature vm dns", "listener can bind "+vmdns.DefaultListenAddr)
}

type natCapability struct{}

func (natCapability) Name() string { return "nat" }

func (natCapability) AddStartPreflight(ctx context.Context, d *Daemon, checks *system.Preflight, vm model.VMRecord, _ model.Image) {
	if !vm.Spec.NATEnabled {
		return
	}
	d.net.addNATPrereqs(ctx, checks)
}

func (natCapability) PostStart(ctx context.Context, d *Daemon, vm model.VMRecord, _ model.Image) error {
	if !vm.Spec.NATEnabled {
		return nil
	}
	return d.net.ensureNAT(ctx, vm.Runtime.GuestIP, d.vm.vmHandles(vm.ID).TapDevice, true)
}

func (natCapability) Cleanup(ctx context.Context, d *Daemon, vm model.VMRecord) error {
	if !vm.Spec.NATEnabled {
		return nil
	}
	tap := d.vm.vmHandles(vm.ID).TapDevice
	if strings.TrimSpace(vm.Runtime.GuestIP) == "" || strings.TrimSpace(tap) == "" {
		if d.logger != nil {
			d.logger.Debug("skipping nat cleanup without runtime network handles", append(vmLogAttrs(vm), "guest_ip", vm.Runtime.GuestIP, "tap_device", tap)...)
		}
		return nil
	}
	return d.net.ensureNAT(ctx, vm.Runtime.GuestIP, tap, false)
}

func (natCapability) ApplyConfigChange(ctx context.Context, d *Daemon, before, after model.VMRecord) error {
	if before.Spec.NATEnabled == after.Spec.NATEnabled {
		return nil
	}
	if !d.vm.vmAlive(after) {
		return nil
	}
	return d.net.ensureNAT(ctx, after.Runtime.GuestIP, d.vm.vmHandles(after.ID).TapDevice, after.Spec.NATEnabled)
}

func (natCapability) AddDoctorChecks(ctx context.Context, d *Daemon, report *system.Report) {
	checks := system.NewPreflight()
	checks.RequireCommand("ip", toolHint("ip"))
	d.net.addNATPrereqs(ctx, checks)
	if len(checks.Problems()) > 0 {
		report.Add(system.CheckStatusFail, "feature nat", checks.Problems()...)
		return
	}
	uplink, err := d.net.defaultUplink(ctx)
	if err != nil {
		report.AddFail("feature nat", err.Error())
		return
	}
	report.AddPass("feature nat", "iptables/sysctl available, uplink "+uplink)
}

func joinErr(current, next error) error {
	return errors.Join(current, next)
}