DebTik/internal/backup/transaction.go

package backup

import (
	"encoding/json"
	"errors"
	"fmt"
	"os"
	"path/filepath"
	"strings"
)

type restoreFile struct {
	path    string
	before  []byte
	after   []byte
	mode    os.FileMode
	existed bool
}

type restoreJournal struct {
	DebtikDir  string        `json:"debtik_dir"`
	NetworkDir string        `json:"network_dir"`
	NewDir     string        `json:"new_dir,omitempty"`
	OldDir     string        `json:"old_dir,omitempty"`
	OldExists  bool          `json:"old_exists"`
	HasDebtik  bool          `json:"has_debtik"`
	Committed  bool          `json:"committed"`
	Files      []journalFile `json:"files"`
}
type journalFile struct {
	Path    string      `json:"path"`
	Data    []byte      `json:"data,omitempty"`
	Mode    os.FileMode `json:"mode"`
	Existed bool        `json:"existed"`
}

func (m *Manager) journalPath() string {
	return filepath.Join(m.backupStore, "restore-transaction.json")
}
func (m *Manager) writeJournal(j *restoreJournal) error {
	if err := os.MkdirAll(m.backupStore, 0700); err != nil {
		return err
	}
	data, err := json.Marshal(j)
	if err != nil {
		return err
	}
	return restoreWrite(m.journalPath(), data, 0600)
}
func syncRestoreDir(path string) error {
	d, err := os.Open(path)
	if err != nil {
		return err
	}
	defer d.Close()
	return d.Sync()
}

// RecoverInterruptedRestore must be called before managers load configuration.
func (m *Manager) RecoverInterruptedRestore() (bool, error) {
	m.mu.Lock()
	defer m.mu.Unlock()
	return m.recoverInterruptedRestoreLocked()
}
func (m *Manager) recoverInterruptedRestoreLocked() (bool, error) {
	data, err := os.ReadFile(m.journalPath())
	if os.IsNotExist(err) {
		return false, nil
	}
	if err != nil {
		return false, err
	}
	var j restoreJournal
	if err = json.Unmarshal(data, &j); err != nil {
		return false, err
	}
	if j.DebtikDir != m.debtikDir || j.NetworkDir != m.networkDir {
		return false, errors.New("restore journal paths do not match configured paths")
	}
	parent := filepath.Dir(m.debtikDir)
	if j.HasDebtik && (j.NewDir == "" || (j.OldExists && j.OldDir == "") || (!j.OldExists && j.OldDir != "")) {
		return false, errors.New("invalid restore journal directory state")
	}
	if !j.HasDebtik && (j.NewDir != "" || j.OldDir != "" || j.OldExists) {
		return false, errors.New("invalid restore journal directory state")
	}
	for _, p := range []string{j.NewDir, j.OldDir} {
		if p != "" && (filepath.Dir(p) != parent || !strings.HasPrefix(filepath.Base(p), ".debtik-restore-")) {
			return false, errors.New("invalid restore journal staging path")
		}
	}
	for _, f := range j.Files {
		if filepath.Dir(f.Path) != m.networkDir || !strings.HasSuffix(f.Path, ".cfg") {
			return false, errors.New("invalid restore journal network path")
		}
	}
	if !j.Committed {
		for _, f := range j.Files {
			if f.Existed {
				if err = os.MkdirAll(m.networkDir, 0755); err == nil {
					err = restoreWrite(f.Path, f.Data, f.Mode)
				}
			} else {
				err = os.Remove(f.Path)
				if os.IsNotExist(err) {
					err = nil
				}
			}
			if err != nil {
				return false, fmt.Errorf("recovering %s: %w", f.Path, err)
			}
		}
		if j.HasDebtik {
			if j.OldExists {
				if _, e := os.Lstat(j.OldDir); e == nil {
					if err = os.RemoveAll(m.debtikDir); err != nil {
						return false, err
					}
					if err = os.Rename(j.OldDir, m.debtikDir); err != nil {
						return false, err
					}
					if err = syncRestoreDir(parent); err != nil {
						return false, err
					}
				} else if !os.IsNotExist(e) {
					return false, e
				} else if _, e = os.Lstat(m.debtikDir); e != nil {
					return false, fmt.Errorf("original configuration unavailable: %w", e)
				}
			} else if _, e := os.Lstat(j.NewDir); os.IsNotExist(e) {
				if err = os.RemoveAll(m.debtikDir); err != nil {
					return false, err
				}
				if err = syncRestoreDir(parent); err != nil {
					return false, err
				}
			} else if e != nil {
				return false, e
			}
		}
	}
	if j.NewDir != "" {
		if err = os.RemoveAll(j.NewDir); err != nil {
			return false, err
		}
	}
	if j.OldDir != "" {
		if err = os.RemoveAll(j.OldDir); err != nil {
			return false, err
		}
	}
	if j.HasDebtik {
		if err = syncRestoreDir(parent); err != nil {
			return false, err
		}
	}
	if err = os.Remove(m.journalPath()); err != nil {
		return false, err
	}
	if err = syncRestoreDir(m.backupStore); err != nil {
		return false, err
	}
	return true, nil
}

// applyStagedRestore restores the original on-disk files if an apply step fails.
// The optional step hook enables deterministic failure tests.
func (m *Manager) applyStagedRestore(stage string, step func(string) error) (err error) {
	runStep := func(name string) error {
		if step != nil {
			return step(name)
		}
		return nil
	}
	stagedDebtik := filepath.Join(stage, "debtik")
	stagedNetwork := filepath.Join(stage, "network")
	var files []restoreFile
	entries, readErr := os.ReadDir(stagedNetwork)
	if readErr != nil && !os.IsNotExist(readErr) {
		return fmt.Errorf("failed listing staged network files: %w", readErr)
	}
	for _, entry := range entries {
		if entry.IsDir() || !strings.HasSuffix(entry.Name(), ".cfg") {
			continue
		}
		target := filepath.Join(m.networkDir, entry.Name())
		f := restoreFile{path: target, mode: 0644}
		f.after, err = os.ReadFile(filepath.Join(stagedNetwork, entry.Name()))
		if err != nil {
			return fmt.Errorf("failed reading staged network file %s: %w", entry.Name(), err)
		}
		info, statErr := os.Lstat(target)
		if statErr == nil {
			if !info.Mode().IsRegular() {
				return fmt.Errorf("failed restoring network file %s: destination is not a regular file", entry.Name())
			}
			f.existed = true
			f.mode = info.Mode().Perm()
			f.before, err = os.ReadFile(target)
			if err != nil {
				return fmt.Errorf("failed reading current network file %s: %w", entry.Name(), err)
			}
		} else if !os.IsNotExist(statErr) {
			return statErr
		}
		files = append(files, f)
	}
	hasDebtik := false
	if info, e := os.Stat(stagedDebtik); e == nil {
		if !info.IsDir() {
			return errors.New("invalid staged configuration")
		}
		hasDebtik = true
	} else if !os.IsNotExist(e) {
		return e
	}
	var newDir, oldDir string
	oldExists := false

	if hasDebtik {
		parent := filepath.Dir(m.debtikDir)
		if err = os.MkdirAll(parent, 0755); err != nil {
			return err
		}
		newDir, err = os.MkdirTemp(parent, ".debtik-restore-new-*")
		if err != nil {
			return err
		}
		defer func() {
			if _, e := os.Stat(m.journalPath()); os.IsNotExist(e) {
				_ = os.RemoveAll(newDir)
			}
		}()
		if err = m.syncDirectory(stagedDebtik, newDir); err != nil {
			return fmt.Errorf("failed staging debtik configuration: %w", err)
		}
		info, e := os.Lstat(m.debtikDir)
		if e == nil {
			if !info.IsDir() {
				return errors.New("debtik configuration path is not a directory")
			}
			oldExists = true
		} else if !os.IsNotExist(e) {
			return e
		}
		if oldExists {
			oldDir, err = os.MkdirTemp(parent, ".debtik-restore-old-*")
			if err != nil {
				return err
			}
			if err = os.Remove(oldDir); err != nil {
				return err
			}
		}
	}
	if _, err := os.Stat(m.journalPath()); err == nil {
		return errors.New("an interrupted restore must be recovered first")
	} else if !os.IsNotExist(err) {
		return err
	}
	journal := &restoreJournal{DebtikDir: m.debtikDir, NetworkDir: m.networkDir, NewDir: newDir, OldDir: oldDir, OldExists: oldExists, HasDebtik: hasDebtik}
	for _, f := range files {
		journal.Files = append(journal.Files, journalFile{Path: f.path, Data: f.before, Mode: f.mode, Existed: f.existed})
	}
	if err = m.writeJournal(journal); err != nil {
		return fmt.Errorf("failed recording restore transaction: %w", err)
	}
	rollback := func(cause error) error {
		if _, err := m.recoverInterruptedRestoreLocked(); err != nil {
			return fmt.Errorf("restore failed and rollback incomplete: %w", errors.Join(cause, err))
		}
		return fmt.Errorf("restore failed; previous files restored: %w", cause)
	}

	if hasDebtik {
		if oldExists {
			if err = os.Rename(m.debtikDir, oldDir); err != nil {
				return rollback(err)
			}
			if err = syncRestoreDir(filepath.Dir(m.debtikDir)); err != nil {
				return rollback(err)
			}
			if err = runStep("old-renamed"); err != nil {
				return rollback(err)
			}
		}
		if err = os.Rename(newDir, m.debtikDir); err != nil {
			return rollback(err)
		}
		if err = syncRestoreDir(filepath.Dir(m.debtikDir)); err != nil {
			return rollback(err)
		}

		if err = runStep("debtik"); err != nil {
			return rollback(err)
		}
	}
	if len(files) > 0 {
		if err = os.MkdirAll(m.networkDir, 0755); err != nil {
			return rollback(err)
		}
	}
	for _, f := range files {
		if err = restoreWrite(f.path, f.after, 0644); err != nil {
			return rollback(fmt.Errorf("failed restoring network file %s: %w", filepath.Base(f.path), err))
		}
		if err = runStep("network"); err != nil {
			return rollback(err)
		}
	}
	journal.Committed = true
	if err = m.writeJournal(journal); err != nil {
		return fmt.Errorf("could not commit restore journal; recovery required: %w", err)
	}
	if err = runStep("committed"); err != nil {
		return fmt.Errorf("restore committed; cleanup pending: %w", err)
	}
	if _, err = m.recoverInterruptedRestoreLocked(); err != nil {
		return fmt.Errorf("configuration restored but transaction cleanup failed: %w", err)
	}

	return nil
}
func restoreWrite(path string, data []byte, mode os.FileMode) error {
	f, err := os.CreateTemp(filepath.Dir(path), ".debtik-restore-*")
	if err != nil {
		return err
	}
	name := f.Name()
	defer os.Remove(name)
	if err = f.Chmod(mode); err == nil {
		_, err = f.Write(data)
	}
	if err == nil {
		err = f.Sync()
	}
	if closeErr := f.Close(); err == nil {
		err = closeErr
	}
	if err != nil {
		return err
	}
	if err = os.Rename(name, path); err != nil {
		return err
	}
	return syncRestoreDir(filepath.Dir(path))
}