banger/internal/daemon/vm.go

package daemon

import (
	"context"
	"errors"
	"fmt"
	"os"
	"strconv"
	"strings"
	"time"

	"banger/internal/daemon/fcproc"
	"banger/internal/model"
	"banger/internal/namegen"
	"banger/internal/system"
	"banger/internal/vmdns"
)

// Cross-service constants. Kept in vm.go because both lifecycle
// (VMService) and networking (HostNetwork) reference them; moving
// them to either owner would read as a layering violation.
var (
	errWaitForExitTimeout = fcproc.ErrWaitForExitTimeout
	gracefulShutdownWait  = 10 * time.Second
	vsockReadyWait        = 30 * time.Second
	vsockReadyPoll        = 200 * time.Millisecond
)

// rebuildDNS enumerates live VMs and republishes the DNS record set.
// Lives on VMService because "alive" is a VM-state concern that
// HostNetwork shouldn't need to reach into. VMService orchestrates:
// VM list from the store, alive filter, hand the resulting map to
// HostNetwork.replaceDNS.
func (s *VMService) rebuildDNS(ctx context.Context) error {
	if s.net == nil {
		return nil
	}
	vms, err := s.store.ListVMs(ctx)
	if err != nil {
		return err
	}
	records := make(map[string]string)
	for _, vm := range vms {
		if !s.vmAlive(vm) {
			continue
		}
		if strings.TrimSpace(vm.Runtime.GuestIP) == "" {
			continue
		}
		records[vmdns.RecordName(vm.Name)] = vm.Runtime.GuestIP
	}
	return s.net.replaceDNS(records)
}

func persistRuntimeTeardownState(vm *model.VMRecord, h model.VMHandles) {
	if vm == nil {
		return
	}
	vm.Runtime.TapDevice = h.TapDevice
	vm.Runtime.BaseLoop = h.BaseLoop
	vm.Runtime.COWLoop = h.COWLoop
	vm.Runtime.DMName = h.DMName
	vm.Runtime.DMDev = h.DMDev
}

func clearRuntimeTeardownState(vm *model.VMRecord) {
	if vm == nil {
		return
	}
	vm.Runtime.TapDevice = ""
	vm.Runtime.BaseLoop = ""
	vm.Runtime.COWLoop = ""
	vm.Runtime.DMName = ""
	vm.Runtime.DMDev = ""
}

func teardownHandlesForCleanup(vm model.VMRecord, live model.VMHandles) model.VMHandles {
	recovered := live
	if strings.TrimSpace(recovered.TapDevice) == "" {
		recovered.TapDevice = strings.TrimSpace(vm.Runtime.TapDevice)
	}
	if strings.TrimSpace(recovered.BaseLoop) == "" {
		recovered.BaseLoop = strings.TrimSpace(vm.Runtime.BaseLoop)
	}
	if strings.TrimSpace(recovered.COWLoop) == "" {
		recovered.COWLoop = strings.TrimSpace(vm.Runtime.COWLoop)
	}
	if strings.TrimSpace(recovered.DMName) == "" {
		recovered.DMName = strings.TrimSpace(vm.Runtime.DMName)
	}
	if strings.TrimSpace(recovered.DMDev) == "" {
		recovered.DMDev = strings.TrimSpace(vm.Runtime.DMDev)
	}
	return recovered
}

// cleanupRuntime tears down the host-side state for a VM: firecracker
// process, DM snapshot, capabilities, tap, sockets. Lives on VMService
// because it reaches into handles (VMService-owned); the capability
// teardown goes through the capHooks seam to keep Daemon out of the
// dependency chain.
func (s *VMService) cleanupRuntime(ctx context.Context, vm model.VMRecord, preserveDisks bool) error {
	if s.logger != nil {
		s.logger.Debug("cleanup runtime", append(vmLogAttrs(vm), "preserve_disks", preserveDisks)...)
	}
	h := s.vmHandles(vm.ID)
	cleanupPID := h.PID
	if vm.Runtime.APISockPath != "" {
		if pid, err := s.net.findFirecrackerPID(ctx, vm.Runtime.APISockPath); err == nil && pid > 0 {
			cleanupPID = pid
		}
	}
	if cleanupPID > 0 && system.ProcessRunning(cleanupPID, vm.Runtime.APISockPath) {
		_ = s.net.killVMProcess(ctx, cleanupPID)
		if err := s.net.waitForExit(ctx, cleanupPID, vm.Runtime.APISockPath, 30*time.Second); err != nil {
			return err
		}
	}
	handles := teardownHandlesForCleanup(vm, h)
	snapshotErr := s.net.cleanupDMSnapshot(ctx, dmSnapshotHandles{
		BaseLoop: handles.BaseLoop,
		COWLoop:  handles.COWLoop,
		DMName:   handles.DMName,
		DMDev:    handles.DMDev,
	})
	featureErr := s.capHooks.cleanupState(ctx, vm)
	var tapErr error
	// Prefer the handle cache (fresh from startVMLocked), but fall
	// back to the VMRuntime mirrors so restart-time cleanup still works
	// when handles.json is missing or corrupt.
	tap := handles.TapDevice
	if tap != "" {
		tapErr = s.net.releaseTap(ctx, tap)
	}
	if vm.Runtime.APISockPath != "" {
		_ = os.Remove(vm.Runtime.APISockPath)
	}
	if vm.Runtime.VSockPath != "" {
		_ = os.Remove(vm.Runtime.VSockPath)
	}
	// The handles are only meaningful while the kernel objects exist;
	// dropping them here keeps the cache in sync with reality even
	// when the caller forgets to call clearVMHandles explicitly.
	s.clearVMHandles(vm)
	if !preserveDisks && vm.Runtime.VMDir != "" {
		return errors.Join(snapshotErr, featureErr, tapErr, os.RemoveAll(vm.Runtime.VMDir))
	}
	return errors.Join(snapshotErr, featureErr, tapErr)
}

func (s *VMService) generateName(ctx context.Context) (string, error) {
	_ = ctx
	if name := strings.TrimSpace(namegen.Generate()); name != "" {
		return name, nil
	}
	return "vm-" + strconv.FormatInt(time.Now().Unix(), 10), nil
}

func bridgePrefix(bridgeIP string) string {
	parts := strings.Split(bridgeIP, ".")
	if len(parts) < 3 {
		return bridgeIP
	}
	return strings.Join(parts[:3], ".")
}

func optionalIntOrDefault(value *int, fallback int) int {
	if value != nil {
		return *value
	}
	return fallback
}

func validateOptionalPositiveSetting(label string, value *int) error {
	if value == nil {
		return nil
	}
	if *value <= 0 {
		return fmt.Errorf("%s must be a positive integer", label)
	}
	return nil
}