banger/internal/daemon/capabilities.go

package daemon

import (
	"context"
	"errors"
	"log/slog"
	"net"
	"os"
	"strings"

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

// vmCapability is the base capability tag. Actual behaviour lives on
// optional sub-interfaces (startPreflight / guestConfig / machineConfig
// / prepareHost / postStart / cleanup / configChange / doctor); a
// capability implements whichever subset it cares about. None of them
// take *Daemon — each capability is a struct constructed with its
// explicit service-pointer dependencies at wireServices time.
type vmCapability interface {
	Name() string
}

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

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

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

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

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

// defaultCapabilities builds the production capability list from
// already-constructed services. Called from wireServices once d.vm /
// d.ws / d.net are populated, so every capability ships with the
// concrete service pointers it needs and none of them reach through
// *Daemon at dispatch time.
func (d *Daemon) defaultCapabilities() []vmCapability {
	return []vmCapability{
		newWorkDiskCapability(d.vm, d.ws, d.store, d.config.DefaultImageName),
		newDNSCapability(d.net),
		newNATCapability(d.vm, d.net, d.logger),
	}
}

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

func (d *Daemon) contributeGuestConfig(builder *guestconfig.Builder, vm model.VMRecord, image model.Image) {
	for _, capability := range d.vmCaps {
		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.vmCaps {
		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.vmCaps))
	for _, capability := range d.vmCaps {
		hook, ok := capability.(prepareHostCapability)
		if !ok {
			continue
		}
		if err := hook.PrepareHost(ctx, 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.vmCaps {
		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, 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.vmCaps)
}

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, *vm))
	}
	return err
}

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

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

// workDiskCapability provisions a per-VM work disk (image-seeded or
// freshly formatted) and syncs host-side authorised keys + git
// identity + file_sync entries onto it. Holds pointers to the VM and
// workspace services because PrepareHost orchestrates across both,
// plus the store + default image name for its doctor check.
type workDiskCapability struct {
	vm               *VMService
	ws               *WorkspaceService
	store            *store.Store
	defaultImageName string
}

func newWorkDiskCapability(vm *VMService, ws *WorkspaceService, st *store.Store, defaultImageName string) workDiskCapability {
	return workDiskCapability{
		vm:               vm,
		ws:               ws,
		store:            st,
		defaultImageName: defaultImageName,
	}
}

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

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

func (c workDiskCapability) AddDoctorChecks(_ context.Context, report *system.Report) {
	if c.store != nil && strings.TrimSpace(c.defaultImageName) != "" {
		if image, err := c.store.GetImageByName(context.Background(), c.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")
}

// dnsCapability publishes + removes <vm>.vm records on the in-process
// DNS server. Only needs HostNetwork.
type dnsCapability struct {
	net *HostNetwork
}

func newDNSCapability(net *HostNetwork) dnsCapability {
	return dnsCapability{net: net}
}

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

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

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

func (dnsCapability) AddDoctorChecks(_ context.Context, 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)
}

// natCapability sets up host-side NAT so guest traffic can reach the
// outside world. Needs VMService (tap lookup + aliveness) and
// HostNetwork (NAT rules), plus the daemon logger for the cleanup
// short-circuit note.
type natCapability struct {
	vm     *VMService
	net    *HostNetwork
	logger *slog.Logger
}

func newNATCapability(vm *VMService, net *HostNetwork, logger *slog.Logger) natCapability {
	return natCapability{vm: vm, net: net, logger: logger}
}

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

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

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

func (c natCapability) Cleanup(ctx context.Context, vm model.VMRecord) error {
	if !vm.Spec.NATEnabled {
		return nil
	}
	// Handle cache is volatile across daemon restarts; Runtime is
	// the persisted DB-backed copy. Fall back so a crash / corrupt
	// handles.json doesn't leak iptables rules keyed off the tap.
	tap := strings.TrimSpace(c.vm.vmHandles(vm.ID).TapDevice)
	if tap == "" {
		tap = strings.TrimSpace(vm.Runtime.TapDevice)
	}
	if strings.TrimSpace(vm.Runtime.GuestIP) == "" || tap == "" {
		if c.logger != nil {
			c.logger.Debug("skipping nat cleanup without runtime network handles", append(vmLogAttrs(vm), "guest_ip", vm.Runtime.GuestIP, "tap_device", tap)...)
		}
		return nil
	}
	return c.net.ensureNAT(ctx, vm.Runtime.GuestIP, tap, false)
}

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

func (c natCapability) AddDoctorChecks(ctx context.Context, report *system.Report) {
	checks := system.NewPreflight()
	checks.RequireCommand("ip", toolHint("ip"))
	c.net.addNATPrereqs(ctx, checks)
	if len(checks.Problems()) > 0 {
		report.Add(system.CheckStatusFail, "feature nat", checks.Problems()...)
		return
	}
	uplink, err := c.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)
}