panel/agent/internal/dispatch/region.go

package dispatch

import (
	"context"
	"fmt"
	"path"
	"sort"
	"strings"
	"time"

	"github.com/dbledeez/panel/pkg/regionmedic"
	panelv1 "github.com/dbledeez/panel/proto/panel/v1"
	"google.golang.org/protobuf/types/known/timestamppb"
)

// Region Medic agent handlers — scan a 7DTD instance's active-world Region
// directory for .7rg corruption, and heal one region from the newest clean
// panel backup. Volume files are reached through the runtime's container-path
// ops (the agent runs unprivileged, so it cannot touch /var/lib/docker/volumes
// directly). The controller orchestrates stop→heal→start; handleRegionHeal
// itself only performs the file swap and refuses to run on a live container.

// activeRegionDir resolves the container-absolute Region directory of the
// instance's active world by reading GameWorld/GameName from the live
// serverconfig.xml. Works on a stopped container (CopyFileFromContainer does).
func (d *Dispatcher) activeRegionDir(ctx context.Context, container string) (string, error) {
	raw, err := d.runtime.CopyFileFromContainer(ctx, container, "/game-saves/serverconfig.xml")
	if err != nil {
		return "", fmt.Errorf("read serverconfig.xml: %w", err)
	}
	world := xmlProp(string(raw), "GameWorld")
	game := xmlProp(string(raw), "GameName")
	if world == "" || game == "" {
		return "", fmt.Errorf("serverconfig.xml missing GameWorld/GameName (got %q/%q)", world, game)
	}
	// HOME is pinned to /game-saves by the 7dtd entrypoint.
	return path.Join("/game-saves/.local/share/7DaysToDie/Saves", world, game, "Region"), nil
}

// xmlProp pulls the value from a 7DTD serverconfig line:
//
//	<property name="GameWorld" value="Navezgane" />
//
// Tolerant of attribute order and single/double quotes; skips commented lines.
func xmlProp(content, name string) string {
	for _, raw := range strings.Split(content, "\n") {
		line := strings.TrimSpace(raw)
		if line == "" || strings.HasPrefix(line, "<!--") {
			continue
		}
		if !strings.Contains(line, `name="`+name+`"`) && !strings.Contains(line, `name='`+name+`'`) {
			continue
		}
		i := strings.Index(line, "value=")
		if i < 0 {
			continue
		}
		rest := line[i+len("value="):]
		if rest == "" {
			continue
		}
		q := rest[0]
		rest = rest[1:]
		if end := strings.IndexByte(rest, q); end >= 0 {
			return strings.TrimSpace(rest[:end])
		}
	}
	return ""
}

// listRegionDirNames returns the file names in the container's Region dir.
// Uses ExecCapture when the container is running; nil + error otherwise (the
// scan path runs while the instance is up).
func (d *Dispatcher) listRegionDirNames(ctx context.Context, container, regionDir string) ([]string, error) {
	stdout, _, _, err := d.runtime.ExecCapture(ctx, container, []string{"ls", "-1", regionDir})
	if err != nil {
		return nil, err
	}
	var names []string
	for _, n := range strings.Split(string(stdout), "\n") {
		if n = strings.TrimSpace(n); n != "" {
			names = append(names, n)
		}
	}
	return names, nil
}

// handleRegionScan reports corruption evidence for an instance's active world:
// error_backup salvage files clustered into regions, plus structural validation
// of each affected region's .7rg file.
func (d *Dispatcher) handleRegionScan(ctx context.Context, corrID string, req *panelv1.RegionScanRequest) {
	rec, err := d.lookupRecord(req.InstanceId)
	if err != nil {
		d.sendRegionScanResult(corrID, &panelv1.RegionScanResult{Error: &panelv1.Error{Code: "not_found", Message: err.Error()}})
		return
	}
	container := "panel-" + rec.InstanceID
	regionDir, err := d.activeRegionDir(ctx, container)
	if err != nil {
		d.sendRegionScanResult(corrID, &panelv1.RegionScanResult{Error: &panelv1.Error{Code: "resolve_region_dir", Message: err.Error()}})
		return
	}
	names, err := d.listRegionDirNames(ctx, container, regionDir)
	if err != nil {
		d.sendRegionScanResult(corrID, &panelv1.RegionScanResult{RegionDir: regionDir, Error: &panelv1.Error{Code: "list_region_dir", Message: err.Error()}})
		return
	}

	// Cluster error_backups by region; note which region files exist.
	type agg struct {
		files   int
		present bool
	}
	clusters := map[regionmedic.RegionCoord]*agg{}
	fileSet := map[regionmedic.RegionCoord]bool{}
	total := 0
	for _, n := range names {
		if cx, cz, ok := regionmedic.ParseErrorBackup(n); ok {
			total++
			rc := regionmedic.RegionForChunk(cx, cz)
			a := clusters[rc]
			if a == nil {
				a = &agg{}
				clusters[rc] = a
			}
			a.files++
			continue
		}
		if rc, ok := regionmedic.ParseRegionFileName(n); ok {
			fileSet[rc] = true
		}
	}

	var affected []*panelv1.AffectedRegionMsg
	for rc, a := range clusters {
		msg := &panelv1.AffectedRegionMsg{
			Region:       rc.String(),
			ErrorBackups: int32(a.files),
			FilePresent:  fileSet[rc],
		}
		// Validate the affected region file (read via container-path op).
		if fileSet[rc] {
			if b, rerr := d.runtime.CopyFileFromContainer(ctx, container, path.Join(regionDir, rc.FileName())); rerr == nil {
				rep := regionmedic.ValidateRegionBytes(b, false)
				msg.FileCorrupt = !rep.Healthy()
				msg.BadChunks = int32(len(rep.BadChunks))
			}
		}
		affected = append(affected, msg)
	}
	sort.Slice(affected, func(i, j int) bool { return affected[i].ErrorBackups > affected[j].ErrorBackups })

	d.sendRegionScanResult(corrID, &panelv1.RegionScanResult{
		ErrorBackups: int32(total),
		Affected:     affected,
		RegionDir:    regionDir,
	})
}

// handleRegionHeal swaps one region file for the newest clean backup copy.
// It REQUIRES the instance to be stopped (the controller stops it first) so the
// running engine can't overwrite the swap. It snapshots the corrupt original
// next to the region file before overwriting. Orchestration (stop/start) is the
// controller's job.
func (d *Dispatcher) handleRegionHeal(ctx context.Context, corrID string, req *panelv1.RegionHealRequest) {
	fail := func(code, msg string) {
		d.sendRegionHealResult(corrID, &panelv1.RegionHealResult{Region: req.Region, Error: &panelv1.Error{Code: code, Message: msg}})
	}
	rec, err := d.lookupRecord(req.InstanceId)
	if err != nil {
		fail("not_found", err.Error())
		return
	}
	region, ok := regionmedic.ParseRegionFileName(regionFileArg(req.Region))
	if !ok {
		fail("bad_region", fmt.Sprintf("invalid region %q", req.Region))
		return
	}
	container := "panel-" + rec.InstanceID

	// Refuse to swap a region on a LIVE container (it would be overwritten on
	// the next save). ExecCapture succeeds only when the container is running.
	if _, _, _, exErr := d.runtime.ExecCapture(ctx, container, []string{"true"}); exErr == nil {
		fail("instance_running", "instance must be stopped before healing a region")
		return
	}

	regionDir, err := d.activeRegionDir(ctx, container)
	if err != nil {
		fail("resolve_region_dir", err.Error())
		return
	}
	liveFile := path.Join(regionDir, region.FileName())
	liveBytes, err := d.runtime.CopyFileFromContainer(ctx, container, liveFile)
	if err != nil {
		fail("read_live_region", err.Error())
		return
	}

	// Find the newest clean backup copy of this region (backups are local to
	// the agent at <backupDir>/<instance>/).
	backups, err := regionmedic.ListPanelBackups(path.Join(d.backupDir, rec.InstanceID))
	if err != nil {
		fail("list_backups", err.Error())
		return
	}
	entry := regionmedic.SaveRelEntry(regionDir, region.FileName())
	best, _, err := regionmedic.FindLastGoodRegion(backups, entry, true, time.Time{})
	if err != nil {
		fail("find_backup", err.Error())
		return
	}
	if !best.Found || !best.Clean() {
		fail("no_clean_backup", "no backup contains a clean copy of this region")
		return
	}

	// Snapshot the corrupt original alongside the region file (no new subdir, so
	// CopyFileToContainer needs no parent-dir creation on a stopped container).
	stamp := time.Now().UTC().Format("20060102-150405")
	snapName := region.FileName() + ".medic-corrupt-" + stamp
	snapPath := path.Join(regionDir, snapName)
	if err := d.runtime.CopyFileToContainer(ctx, container, snapPath, liveBytes); err != nil {
		fail("snapshot", err.Error())
		return
	}
	// Swap in the clean copy (overwrites the live region file in place).
	if err := d.runtime.CopyFileToContainer(ctx, container, liveFile, best.Bytes); err != nil {
		fail("write_region", err.Error())
		return
	}

	d.sendRegionHealResult(corrID, &panelv1.RegionHealResult{
		Region:       region.String(),
		SourceBackup: best.Backup.ID,
		BytesWritten: int64(len(best.Bytes)),
		SnapshotPath: snapPath,
		Healed:       true,
	})
}

// regionFileArg normalizes "r.X.Z" or "r.X.Z.7rg" to a .7rg filename.
func regionFileArg(s string) string {
	if strings.HasSuffix(s, ".7rg") {
		return s
	}
	return s + ".7rg"
}

func (d *Dispatcher) sendRegionScanResult(corrID string, res *panelv1.RegionScanResult) {
	d.sendEnv(&panelv1.AgentEnvelope{
		CorrelationId: corrID,
		SentAt:        timestamppb.Now(),
		Payload:       &panelv1.AgentEnvelope_RegionScanResult{RegionScanResult: res},
	})
}

func (d *Dispatcher) sendRegionHealResult(corrID string, res *panelv1.RegionHealResult) {
	d.sendEnv(&panelv1.AgentEnvelope{
		CorrelationId: corrID,
		SentAt:        timestamppb.Now(),
		Payload:       &panelv1.AgentEnvelope_RegionHealResult{RegionHealResult: res},
	})
}