jarl/internal/engine/engine.go

// Package engine applies changes to the configuration.
//
// Every change goes through a changeset: a list of operations that is
// validated as a whole, rendered to BIND configuration, checked with
// named-checkconf, written to disk and loaded into named. Only when named
// has loaded the new configuration is the database updated. If anything
// fails, the previous files are restored and nothing is recorded.
package engine

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"io/fs"
	"log/slog"
	"os"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"

	"jarl/internal/auth"
	"jarl/internal/bind"
	"jarl/internal/dnsname"
	"jarl/internal/model"
	"jarl/internal/namedconf"
	"jarl/internal/render"
	"jarl/internal/store"
	"jarl/internal/validate"
)

// Config configures the engine.
type Config struct {
	Paths render.Paths
	// Extras are statements added from jarld's configuration, such as the
	// statistics channel and response log for query statistics.
	Extras render.Extras
	// MainConfig is the configuration file named loads (on Debian
	// /etc/bind/named.conf). Jarl is "adopted" once it includes Jarl's
	// named.conf. If it is Jarl's named.conf itself, Jarl is always
	// adopted.
	MainConfig string
	// StagingDir is where configurations are rendered for named-checkconf.
	StagingDir string
	// FileGroup is the group of generated files (e.g. "bind").
	FileGroup string
	// ReloadTimeout bounds how long Jarl waits for named to load zones.
	ReloadTimeout time.Duration
	// NoReload writes files but never runs rndc (for setups without a
	// running named, and tests).
	NoReload bool
	// SkipCheck disables named-checkconf (only for tests without BIND).
	SkipCheck bool
}

// Engine applies changesets.
type Engine struct {
	store  *store.Store
	tools  *bind.Tools
	cfg    Config
	log    *slog.Logger
	deploy *Deployer
	mu     sync.Mutex // serializes applies
	Now    func() time.Time
}

// New creates an engine.
func New(st *store.Store, tools *bind.Tools, cfg Config, log *slog.Logger) *Engine {
	if cfg.ReloadTimeout == 0 {
		cfg.ReloadTimeout = 15 * time.Second
	}
	if log == nil {
		log = slog.Default()
	}
	return &Engine{
		store: st, tools: tools, cfg: cfg, log: log,
		deploy: &Deployer{Group: cfg.FileGroup, NamedWritable: []string{cfg.Paths.ZoneDir, cfg.Paths.CacheDir}},
		Now:    time.Now,
	}
}

// Config returns the engine configuration.
func (e *Engine) Config() Config { return e.cfg }

// Store returns the engine's store.
func (e *Engine) Store() *store.Store { return e.store }

// Result describes a preview or an applied changeset.
type Result struct {
	ChangesetID int64              `json:"changeset_id,omitempty"`
	Applied     bool               `json:"applied"`
	DryRun      bool               `json:"dry_run,omitempty"`
	Valid       bool               `json:"valid"`
	Problems    []validate.Problem `json:"problems"` // problems this change introduces
	Existing    []validate.Problem `json:"existing_problems,omitempty"`
	Check       []CheckMessage     `json:"check,omitempty"`
	CheckOK     bool               `json:"check_ok"`
	Diff        string             `json:"diff"`
	Files       []string           `json:"files_changed"`
	Serials     []SerialChange     `json:"serials,omitempty"`
	Reloaded    []string           `json:"reloaded,omitempty"`
	Warnings    []string           `json:"warnings,omitempty"`
	Drift       []DriftEntry       `json:"drift,omitempty"`
	Ops         []model.Op         `json:"ops,omitempty"`
	Error       *Error             `json:"error,omitempty"`
}

// stored is what is kept in a changeset's result column.
type stored struct {
	Serials  []SerialChange `json:"serials,omitempty"`
	Warnings []string       `json:"warnings,omitempty"`
	Files    []string       `json:"files_changed,omitempty"`
	Undo     []model.Op     `json:"undo,omitempty"`
}

// ApplyOptions modify Apply and Submit.
type ApplyOptions struct {
	DryRun bool
	// Force overwrites files that were changed outside Jarl.
	Force bool
	// NoReload writes the files without reloading named.
	NoReload bool
}

func (e *Engine) replay(base *model.State, ops []model.Op, p *auth.Principal) (*model.State, []string, error) {
	work := base.Clone()
	c := &applyCtx{p: p, now: e.Now()}
	for _, op := range ops {
		if err := applyOp(work, op, c); err != nil {
			return nil, c.warnings, err
		}
	}
	return work, c.warnings, nil
}

// ValidateOptions returns validation options for st, finding TSIG keys
// defined in files that top-level include statements pull in.
func ValidateOptions(st *model.State) validate.Options {
	var opts validate.Options
	for _, o := range st.ScopedOptions(model.ScopeTopLevel, 0) {
		if o.Name != "include" {
			continue
		}
		path := strings.Trim(strings.TrimSpace(o.Value), `"`)
		b, err := os.ReadFile(path)
		if err != nil {
			opts.ExternalKeysUnknown = true
			continue
		}
		stmts, err := namedconf.Parse(path, string(b))
		if err != nil {
			opts.ExternalKeysUnknown = true
			continue
		}
		for _, s := range namedconf.Find(stmts, "key") {
			opts.ExternalKeys = append(opts.ExternalKeys, s.Arg(0))
		}
	}
	return opts
}

func (e *Engine) validate(st *model.State) []validate.Problem {
	return validate.State(st, ValidateOptions(st))
}

// ---- changeset management

func (e *Engine) loadChangeset(ctx context.Context, id int64, p *auth.Principal, write bool) (*model.Changeset, error) {
	cs, err := e.store.GetChangeset(ctx, id)
	if errors.Is(err, store.ErrNotFound) {
		return nil, errf(KindNotFound, "changeset %d does not exist", id)
	}
	if err != nil {
		return nil, err
	}
	if !CanAccessChangeset(p, cs, write) {
		return nil, errf(KindForbidden, "changeset %d belongs to %s", id, cs.Username)
	}
	if write && cs.Status != model.ChangesetOpen {
		return nil, errf(KindConflict, "changeset %d is %s and can't be changed", id, cs.Status)
	}
	return cs, nil
}

// CanAccessChangeset reports whether p may see (or modify) a changeset.
func CanAccessChangeset(p *auth.Principal, cs *model.Changeset, write bool) bool {
	if p.IsAdmin() || (p.User != nil && cs.UserID == p.User.ID) {
		return true
	}
	return !write && p.Can(auth.Action{Resource: auth.ResAudit})
}

// CreateChangeset starts a new open changeset.
func (e *Engine) CreateChangeset(ctx context.Context, p *auth.Principal, via, description string) (*model.Changeset, error) {
	cs := &model.Changeset{Description: description, UserID: p.User.ID, Username: p.Name(), Via: via, Status: model.ChangesetOpen}
	if err := e.store.CreateChangeset(ctx, cs); err != nil {
		return nil, err
	}
	e.audit(ctx, p, "changeset.create", "changeset", cs.ID, cs.Description, cs.ID, "success", "", nil)
	return cs, nil
}

// StageResult is returned when operations are added to a changeset.
type StageResult struct {
	Changeset *model.Changeset   `json:"changeset"`
	Added     []model.Op         `json:"added"`
	Warnings  []string           `json:"warnings,omitempty"`
	Problems  []validate.Problem `json:"problems,omitempty"` // warnings introduced
}

// expandAll expands and applies requests to work, returning the new
// operations.
func (e *Engine) expandAll(ctx context.Context, work *model.State, reqs []Request, p *auth.Principal) ([]model.Op, []string, error) {
	var ops []model.Op
	var warnings []string
	c := &applyCtx{p: p, now: e.Now()}
	for i, req := range reqs {
		x := &expander{ctx: ctx, st: work, alloc: e.store.AllocateIDs, p: p, now: e.Now()}
		more, err := x.expand(req)
		if err != nil {
			if ee, ok := AsError(err); ok && len(reqs) > 1 {
				ee.Message = fmt.Sprintf("request %d (%s): %s", i+1, req.Op, ee.Message)
			}
			return nil, nil, err
		}
		warnings = append(warnings, x.warnings...)
		for _, op := range more {
			if err := applyOp(work, op, c); err != nil {
				if ee, ok := AsError(err); ok {
					ee.OpSeq = 0
					if len(reqs) > 1 {
						ee.Message = fmt.Sprintf("request %d (%s): %s", i+1, req.Op, ee.Message)
					}
				}
				return nil, nil, err
			}
		}
		ops = append(ops, more...)
	}
	return ops, append(warnings, c.warnings...), nil
}

// Stage adds requests to an open changeset. Requests that would introduce
// errors (for example an A record next to a CNAME) are rejected.
func (e *Engine) Stage(ctx context.Context, csID int64, reqs []Request, p *auth.Principal) (*StageResult, error) {
	cs, err := e.loadChangeset(ctx, csID, p, true)
	if err != nil {
		return nil, err
	}
	base, err := e.store.LoadState(ctx)
	if err != nil {
		return nil, err
	}
	work, _, err := e.replay(base, cs.Ops, p)
	if err != nil {
		return nil, errf(KindConflict, "the changeset no longer applies to the current configuration (%v); remove the failing operation first", err)
	}
	before := e.validate(work)
	ops, warnings, err := e.expandAll(ctx, work, reqs, p)
	if err != nil {
		return nil, err
	}
	intro := validate.Introduced(before, e.validate(work))
	if validate.HasErrors(intro) {
		return nil, &Error{Kind: KindValidation, Message: describeErrors(intro), Problems: validate.Errors(intro)}
	}
	now := e.Now()
	for i := range ops {
		ops[i].Seq = nextSeq(cs.Ops) + i
		ops[i].AddedAt = now
	}
	cs.Ops = append(cs.Ops, ops...)
	if err := e.store.UpdateChangeset(ctx, cs); err != nil {
		return nil, err
	}
	for _, op := range ops {
		e.audit(ctx, p, "changeset.stage", "changeset", cs.ID, op.Summary, cs.ID, "success", "", nil)
	}
	var warnProblems []validate.Problem
	for _, pr := range intro {
		if pr.Severity != validate.Error {
			warnProblems = append(warnProblems, pr)
		}
	}
	return &StageResult{Changeset: cs, Added: ops, Warnings: warnings, Problems: warnProblems}, nil
}

func nextSeq(ops []model.Op) int {
	m := 0
	for _, op := range ops {
		if op.Seq > m {
			m = op.Seq
		}
	}
	return m + 1
}

func describeErrors(ps []validate.Problem) string {
	errs := validate.Errors(ps)
	if len(errs) == 0 {
		return "validation failed"
	}
	if len(errs) == 1 {
		return errs[0].Message
	}
	return fmt.Sprintf("%s (and %d more problems)", errs[0].Message, len(errs)-1)
}

// RemoveOps removes operations (by sequence number) from an open changeset.
func (e *Engine) RemoveOps(ctx context.Context, csID int64, seqs []int, p *auth.Principal) (*model.Changeset, error) {
	cs, err := e.loadChangeset(ctx, csID, p, true)
	if err != nil {
		return nil, err
	}
	drop := map[int]bool{}
	for _, s := range seqs {
		drop[s] = true
	}
	var kept []model.Op
	found := 0
	for _, op := range cs.Ops {
		if drop[op.Seq] {
			found++
			e.audit(ctx, p, "changeset.unstage", "changeset", cs.ID, op.Summary, cs.ID, "success", "", nil)
			continue
		}
		kept = append(kept, op)
	}
	if found != len(drop) {
		return nil, errf(KindNotFound, "operation not found in changeset %d", csID)
	}
	if kept == nil {
		kept = []model.Op{}
	}
	cs.Ops = kept
	if err := e.store.UpdateChangeset(ctx, cs); err != nil {
		return nil, err
	}
	return cs, nil
}

// UpdateDescription changes an open changeset's description.
func (e *Engine) UpdateDescription(ctx context.Context, csID int64, desc string, p *auth.Principal) (*model.Changeset, error) {
	cs, err := e.loadChangeset(ctx, csID, p, true)
	if err != nil {
		return nil, err
	}
	cs.Description = desc
	return cs, e.store.UpdateChangeset(ctx, cs)
}

// Discard abandons an open changeset.
func (e *Engine) Discard(ctx context.Context, csID int64, p *auth.Principal) error {
	cs, err := e.loadChangeset(ctx, csID, p, true)
	if err != nil {
		return err
	}
	cs.Status = model.ChangesetDiscarded
	if err := e.store.UpdateChangeset(ctx, cs); err != nil {
		return err
	}
	e.audit(ctx, p, "changeset.discard", "changeset", cs.ID, cs.Description, cs.ID, "success", "", nil)
	return nil
}

// ---- planning

type plan struct {
	base, work    *model.State
	introduced    []validate.Problem
	existing      []validate.Problem
	serials       []SerialChange
	out           *render.Output
	changed       []*render.File
	removed       []string
	diff          string
	check         *bind.Result
	checkMsgs     []CheckMessage
	drift         []DriftEntry
	configChanged bool
	reloadZones   []*model.Zone
	verify        map[int64]uint32
	newZones      bool
	backup        *Backup
	reloaded      bool
}

func (pl *plan) files() []string {
	var out []string
	for _, f := range pl.changed {
		out = append(out, f.Path)
	}
	out = append(out, pl.removed...)
	sort.Strings(out)
	return out
}

func (e *Engine) makePlan(ctx context.Context, base, work *model.State, user string) (*plan, error) {
	pl := &plan{base: base, work: work, verify: map[int64]uint32{}}
	before := e.validate(base)
	after := e.validate(work)
	pl.introduced = validate.Introduced(before, after)
	introducedKeys := map[string]bool{}
	for _, p := range pl.introduced {
		introducedKeys[p.Key()] = true
	}
	for _, p := range after {
		if !introducedKeys[p.Key()] {
			pl.existing = append(pl.existing, p)
		}
	}
	if validate.HasErrors(pl.introduced) {
		return pl, nil
	}
	pl.serials = bumpSerials(base, work, e.Now(), user)
	pl.out = render.Render(work, e.cfg.Paths, e.cfg.Extras)
	deployed, err := e.store.DeployedFiles(ctx)
	if err != nil {
		return nil, err
	}
	pl.drift = CheckDrift(deployed)

	var diff strings.Builder
	for _, f := range pl.out.Files {
		old, existed, err := readFile(f.Path)
		if err != nil {
			return nil, err
		}
		if existed && string(old) == string(f.Content) {
			continue
		}
		pl.changed = append(pl.changed, f)
		diff.WriteString(fileDiff(f.Path, old, existed, f.Content, false, f.Secret))
		if f.Kind == render.KindConfig || f.Kind == render.KindKeys {
			pl.configChanged = true
		}
	}
	for path, df := range deployed {
		if pl.out.File(path) != nil || df.Kind != render.KindZone {
			continue
		}
		old, existed, _ := readFile(path)
		if existed {
			pl.removed = append(pl.removed, path)
			diff.WriteString(fileDiff(path, old, true, nil, true, false))
		}
	}
	sort.Strings(pl.removed)
	pl.diff = diff.String()

	// Which zones named must reload, and which serials to verify.
	baseOut := render.Render(base, e.cfg.Paths, e.cfg.Extras)
	changedPaths := map[string]bool{}
	for _, f := range pl.changed {
		changedPaths[f.Path] = true
	}
	for _, z := range work.SortedZones() {
		if z.Type != model.ZonePrimary {
			continue
		}
		path := pl.out.ZoneFiles[z.ID]
		old := base.Zones[z.ID]
		existedInNamed := old != nil && old.Type == model.ZonePrimary && baseOut.ZoneFiles[old.ID] == path
		if !existedInNamed {
			pl.newZones = true
			pl.verify[z.ID] = z.SOA.Serial
			continue
		}
		if changedPaths[path] && render.ZoneContentHash(base, old) != render.ZoneContentHash(work, z) || old.SOA.Serial != z.SOA.Serial {
			pl.reloadZones = append(pl.reloadZones, z)
			pl.verify[z.ID] = z.SOA.Serial
		}
	}
	if len(work.Zones) != len(base.Zones) {
		pl.configChanged = true
	}

	if !e.cfg.SkipCheck {
		if err := e.stagingCheck(ctx, pl); err != nil {
			return nil, err
		}
	}
	return pl, nil
}

// stagingCheck renders work into a staging directory and runs
// named-checkconf -z on it.
func (e *Engine) stagingCheck(ctx context.Context, pl *plan) error {
	if err := os.MkdirAll(e.cfg.StagingDir, 0o750); err != nil {
		return fmt.Errorf("creating staging directory: %w", err)
	}
	dir, err := os.MkdirTemp(e.cfg.StagingDir, "check-")
	if err != nil {
		return err
	}
	defer os.RemoveAll(dir)
	sp := render.Paths{ConfigDir: filepath.Join(dir, "config"), ZoneDir: filepath.Join(dir, "zones"), CacheDir: filepath.Join(dir, "cache")}
	sout := render.Render(pl.work, sp, e.cfg.Extras)
	for _, f := range sout.Files {
		if err := os.MkdirAll(filepath.Dir(f.Path), 0o750); err != nil {
			return err
		}
		if err := os.WriteFile(f.Path, f.Content, 0o640); err != nil {
			return err
		}
	}
	main, err := e.stagingMain(dir, sp.MainConfig())
	if err != nil {
		return err
	}
	res, err := e.tools.CheckConfig(ctx, main, true)
	if err != nil {
		return &Error{Kind: KindBIND, Message: "running named-checkconf failed: " + err.Error()}
	}
	pl.check = res
	// Map staging paths back to objects and production paths.
	zoneByFile := map[string]int64{}
	for id, path := range sout.ZoneFiles {
		zoneByFile[path] = id
	}
	replacer := strings.NewReplacer(sp.ConfigDir, e.cfg.Paths.ConfigDir, sp.ZoneDir, e.cfg.Paths.ZoneDir, sp.CacheDir, e.cfg.Paths.CacheDir, dir, "(staging)")
	for _, m := range res.Messages {
		cm := CheckMessage{Message: m}
		if span, ok := sout.Lookup(m.File, m.Line); ok {
			cm.ObjectType, cm.ObjectID, cm.Object = span.ObjectType, span.ObjectID, span.Label
		} else if id, ok := zoneByFile[m.File]; ok {
			z := pl.work.Zones[id]
			cm.ObjectType, cm.ObjectID, cm.Object = "zone", id, "zone "+zoneDesc(pl.work, z)
		}
		cm.File = replacer.Replace(cm.File)
		cm.Text = replacer.Replace(cm.Text)
		pl.checkMsgs = append(pl.checkMsgs, cm)
	}
	return nil
}

// includesPath reports whether config text includes path.
func includesPath(text, path string) bool {
	stmts, err := namedconf.Parse("", text)
	if err != nil {
		return strings.Contains(text, `"`+path+`"`)
	}
	for _, s := range namedconf.Find(stmts, "include") {
		if filepath.Clean(s.Arg(0)) == filepath.Clean(path) {
			return true
		}
	}
	return false
}

// stagingMain returns the main configuration file to check: named's own
// main file with Jarl's include pointed at the staging copy, so anything
// else in it is checked too.
func (e *Engine) stagingMain(dir, stagingJarl string) (string, error) {
	main := e.cfg.MainConfig
	if main == "" || filepath.Clean(main) == filepath.Clean(e.cfg.Paths.MainConfig()) {
		return stagingJarl, nil
	}
	b, err := os.ReadFile(main)
	if err != nil {
		if errors.Is(err, fs.ErrNotExist) {
			return stagingJarl, nil
		}
		// Not readable by Jarl: check Jarl's configuration alone.
		e.log.Warn("cannot read main named configuration; checking Jarl's configuration alone", "path", main, "err", err)
		return stagingJarl, nil
	}
	text := string(b)
	if !includesPath(text, e.cfg.Paths.MainConfig()) {
		return stagingJarl, nil
	}
	text = strings.ReplaceAll(text, `"`+e.cfg.Paths.MainConfig()+`"`, `"`+stagingJarl+`"`)
	path := filepath.Join(dir, "main.conf")
	return path, os.WriteFile(path, []byte(text), 0o640)
}

// Adopted reports whether named loads Jarl's configuration.
func (e *Engine) Adopted() (bool, error) {
	main := e.cfg.MainConfig
	if main == "" || filepath.Clean(main) == filepath.Clean(e.cfg.Paths.MainConfig()) {
		return true, nil
	}
	b, err := os.ReadFile(main)
	if err != nil {
		return false, err
	}
	return includesPath(string(b), e.cfg.Paths.MainConfig()), nil
}

func (pl *plan) result() *Result {
	r := &Result{
		Valid:    !validate.HasErrors(pl.introduced),
		Problems: nonNilProblems(pl.introduced),
		Existing: pl.existing,
		Check:    pl.checkMsgs,
		CheckOK:  pl.check == nil || pl.check.OK,
		Diff:     pl.diff,
		Files:    pl.files(),
		Serials:  pl.serials,
		Drift:    pl.drift,
	}
	r.Valid = r.Valid && r.CheckOK
	if r.Files == nil {
		r.Files = []string{}
	}
	return r
}

func nonNilProblems(ps []validate.Problem) []validate.Problem {
	if ps == nil {
		return []validate.Problem{}
	}
	return ps
}

// Preview validates a changeset and shows what applying it would do.
func (e *Engine) Preview(ctx context.Context, csID int64, p *auth.Principal) (*Result, error) {
	cs, err := e.loadChangeset(ctx, csID, p, false)
	if err != nil {
		return nil, err
	}
	base, err := e.store.LoadState(ctx)
	if err != nil {
		return nil, err
	}
	res := &Result{ChangesetID: cs.ID, DryRun: true, Ops: RedactOps(cs.Ops), Problems: []validate.Problem{}, Files: []string{}}
	work, warnings, err := e.replay(base, cs.Ops, p)
	if err != nil {
		ee, ok := AsError(err)
		if !ok {
			return nil, err
		}
		res.Error = ee
		return res, nil
	}
	pl, err := e.makePlan(ctx, base, work, p.Name())
	if err != nil {
		return nil, err
	}
	r := pl.result()
	r.ChangesetID, r.DryRun, r.Ops, r.Warnings = cs.ID, true, RedactOps(cs.Ops), warnings
	return r, nil
}

// Apply applies an open changeset.
func (e *Engine) Apply(ctx context.Context, csID int64, p *auth.Principal, opts ApplyOptions) (*Result, error) {
	cs, err := e.loadChangeset(ctx, csID, p, true)
	if err != nil {
		return nil, err
	}
	if len(cs.Ops) == 0 {
		return nil, errf(KindBadRequest, "changeset %d is empty", csID)
	}
	return e.applyOps(ctx, cs, nil, p, opts)
}

// Submit expands requests into a new changeset and applies it at once.
// Nothing is stored if it fails.
func (e *Engine) Submit(ctx context.Context, reqs []Request, p *auth.Principal, via, description string, opts ApplyOptions) (*Result, error) {
	cs := &model.Changeset{Description: description, UserID: p.User.ID, Username: p.Name(), Via: via, CreatedAt: e.Now()}
	return e.applyOps(ctx, cs, reqs, p, opts)
}

func (e *Engine) applyOps(ctx context.Context, cs *model.Changeset, reqs []Request, p *auth.Principal, opts ApplyOptions) (res *Result, err error) {
	e.mu.Lock()
	defer e.mu.Unlock()
	defer func() {
		if err != nil {
			outcome := "failure"
			if ee, ok := AsError(err); ok && ee.Kind == KindForbidden {
				outcome = "denied"
			}
			summary := cs.Description
			if summary == "" && len(cs.Ops) > 0 {
				summary = cs.Ops[0].Summary
			}
			if !opts.DryRun {
				e.audit(ctx, p, "changeset.apply", "changeset", cs.ID, summary, cs.ID, outcome, err.Error(), nil)
			}
		}
	}()
	base, err := e.store.LoadState(ctx)
	if err != nil {
		return nil, err
	}
	var work *model.State
	var warnings []string
	if reqs != nil {
		work = base.Clone()
		var ops []model.Op
		ops, warnings, err = e.expandAll(ctx, work, reqs, p)
		if err != nil {
			return nil, err
		}
		now := e.Now()
		for i := range ops {
			ops[i].Seq, ops[i].AddedAt = i+1, now
		}
		cs.Ops = ops
		if len(ops) == 0 {
			return nil, errf(KindBadRequest, "nothing to do")
		}
	} else {
		work, warnings, err = e.replay(base, cs.Ops, p)
		if err != nil {
			return nil, err
		}
	}
	if cs.Description == "" && len(cs.Ops) > 0 {
		cs.Description = cs.Ops[0].Summary
		if len(cs.Ops) > 1 {
			cs.Description += fmt.Sprintf(" (+%d more)", len(cs.Ops)-1)
		}
	}
	pl, err := e.makePlan(ctx, base, work, p.Name())
	if err != nil {
		return nil, err
	}
	res = pl.result()
	res.ChangesetID, res.Warnings, res.Ops, res.DryRun = cs.ID, warnings, RedactOps(cs.Ops), opts.DryRun
	if validate.HasErrors(pl.introduced) {
		return res, &Error{Kind: KindValidation, Message: describeErrors(pl.introduced), Problems: validate.Errors(pl.introduced)}
	}
	if pl.check != nil && !pl.check.OK {
		return res, &Error{Kind: KindValidation, Message: "named-checkconf rejected the new configuration: " + firstCheckError(pl.checkMsgs), Check: pl.checkMsgs}
	}
	if opts.DryRun {
		return res, nil
	}
	if len(pl.drift) > 0 && !opts.Force {
		var paths []string
		for _, d := range pl.drift {
			paths = append(paths, d.Path+" ("+d.Reason+")")
		}
		return res, &Error{Kind: KindDrift, Message: "files managed by Jarl were changed outside Jarl: " + strings.Join(paths, ", ") + "; review them and apply with force to overwrite"}
	}
	reload := !e.cfg.NoReload && !opts.NoReload
	if reload {
		if adopted, aerr := e.Adopted(); aerr != nil || !adopted {
			// named doesn't read Jarl's files yet, so there is nothing
			// to reload; the files are ready for 'jarld adopt'.
			reload = false
			res.Warnings = append(res.Warnings, fmt.Sprintf("named does not load Jarl's configuration yet (%s does not include %s): the files were written but named was not reloaded; run 'jarld adopt' to switch named over", e.cfg.MainConfig, e.cfg.Paths.MainConfig()))
		}
	}
	if err := e.deployAndReload(ctx, pl, res, reload); err != nil {
		return res, err
	}
	// Record everything.
	undo := DiffOps(work, base)
	st := stored{Serials: pl.serials, Warnings: warnings, Files: res.Files, Undo: undo}
	cs.Result = mustJSON(st)
	cs.Diff = pl.diff
	cs.AppliedBy = p.Name()
	deployed := deployedList(pl.out, cs.ID, e.Now())
	err = e.store.InTx(ctx, func(tx *store.Tx) error {
		if err := tx.SaveState(base, work); err != nil {
			return err
		}
		if cs.ID == 0 {
			if err := tx.InsertAppliedChangeset(cs); err != nil {
				return err
			}
		} else if err := tx.MarkApplied(cs); err != nil {
			return err
		}
		for _, d := range deployed {
			d.ChangesetID = cs.ID
		}
		if err := tx.SetDeployedFiles(deployed); err != nil {
			return err
		}
		details, _ := json.Marshal(map[string]any{"ops": RedactOps(cs.Ops), "serials": pl.serials, "files": res.Files})
		return tx.AddAudit(e.auditEvent(p, "changeset.apply", "changeset", cs.ID, cs.Description, cs.ID, "success", "", details))
	})
	if err != nil {
		e.log.Error("recording applied changeset failed; restoring previous configuration", "err", err)
		e.rollback(ctx, pl)
		return res, fmt.Errorf("recording the change failed (the previous configuration was restored): %w", err)
	}
	res.ChangesetID = cs.ID
	res.Applied = true
	e.log.Info("changeset applied", "changeset", cs.ID, "user", p.Name(), "ops", len(cs.Ops), "files", len(res.Files), "description", cs.Description)
	return res, nil
}

func firstCheckError(msgs []CheckMessage) string {
	for _, m := range msgs {
		if m.Error {
			s := m.Text
			if m.Object != "" {
				s = m.Object + ": " + s
			}
			return s
		}
	}
	return "see the check output"
}

func deployedList(out *render.Output, csID int64, now time.Time) []*model.DeployedFile {
	var list []*model.DeployedFile
	for _, f := range out.Files {
		list = append(list, &model.DeployedFile{Path: f.Path, SHA256: sha(f.Content), Size: int64(len(f.Content)), Kind: f.Kind, WrittenAt: now, ChangesetID: csID})
	}
	return list
}

// deployAndReload writes the files and (if reload) makes named load them.
// On failure the previous files are restored.
func (e *Engine) deployAndReload(ctx context.Context, pl *plan, res *Result, reload bool) error {
	backup, err := e.deploy.Install(pl.changed, pl.removed, pl.out.Dirs)
	if err != nil {
		return &Error{Kind: KindBIND, Message: "writing configuration files failed (nothing was changed): " + err.Error()}
	}
	pl.backup = backup
	pl.reloaded = reload
	if !reload {
		return nil
	}
	reloaded, err := e.reload(ctx, pl)
	res.Reloaded = reloaded
	if err != nil {
		e.log.Error("reloading named failed; restoring previous configuration", "err", err)
		e.rollback(ctx, pl)
		return &Error{Kind: KindBIND, Message: err.Error() + " (the previous configuration was restored)"}
	}
	return nil
}

func (e *Engine) rollback(ctx context.Context, pl *plan) {
	if err := pl.backup.Restore(); err != nil {
		e.log.Error("restoring files failed", "err", err)
	}
	if !pl.reloaded {
		return
	}
	if pl.configChanged || pl.newZones {
		if err := e.tools.Reconfig(ctx); err != nil {
			e.log.Error("rndc reconfig after restore failed", "err", err)
		}
	}
	for _, z := range pl.reloadZones {
		if err := e.tools.ReloadZone(ctx, z.Name, e.namedView(pl.base, pl.base.Zones[z.ID])); err != nil {
			e.log.Error("rndc reload after restore failed", "zone", z.Name, "err", err)
		}
	}
}

// namedView returns the view name to pass to rndc for a zone ("" without
// views).
func (e *Engine) namedView(st *model.State, z *model.Zone) string {
	if z == nil {
		return ""
	}
	return st.ViewName(z)
}

func (e *Engine) reload(ctx context.Context, pl *plan) ([]string, error) {
	var done []string
	if pl.configChanged || pl.newZones {
		if err := e.tools.Reconfig(ctx); err != nil {
			return done, fmt.Errorf("named rejected the new configuration: %v", err)
		}
		done = append(done, "reconfig")
	}
	for _, z := range pl.reloadZones {
		if err := e.tools.ReloadZone(ctx, z.Name, e.namedView(pl.work, z)); err != nil {
			return done, fmt.Errorf("reloading zone %s failed: %v", zoneDesc(pl.work, z), err)
		}
		done = append(done, "reload "+zoneDesc(pl.work, z))
	}
	// named reloads zones asynchronously and reports load errors only in
	// its log, so wait until it serves the new serials.
	deadline := time.Now().Add(e.cfg.ReloadTimeout)
	pending := map[int64]uint32{}
	for id, s := range pl.verify {
		pending[id] = s
	}
	var lastErr error
	for len(pending) > 0 {
		for id, want := range pending {
			z := pl.work.Zones[id]
			st, err := e.tools.ZoneStatus(ctx, z.Name, e.namedView(pl.work, z))
			if err == nil && st.HasSerial && st.Serial == want {
				delete(pending, id)
				continue
			}
			if err != nil {
				lastErr = err
			} else {
				lastErr = fmt.Errorf("named serves serial %d", st.Serial)
			}
		}
		if len(pending) == 0 {
			break
		}
		if time.Now().After(deadline) {
			var names []string
			for id, want := range pending {
				names = append(names, fmt.Sprintf("%s (expected serial %d)", zoneDesc(pl.work, pl.work.Zones[id]), want))
			}
			sort.Strings(names)
			return done, fmt.Errorf("named did not load %s within %s: %v; check named's log (journalctl -u named)", strings.Join(names, ", "), e.cfg.ReloadTimeout, lastErr)
		}
		select {
		case <-ctx.Done():
			return done, ctx.Err()
		case <-time.After(100 * time.Millisecond):
		}
	}
	return done, nil
}

// ---- revert, resync, check

// Revert creates an open changeset that undoes an applied changeset.
func (e *Engine) Revert(ctx context.Context, csID int64, p *auth.Principal) (*model.Changeset, error) {
	cs, err := e.loadChangeset(ctx, csID, p, false)
	if err != nil {
		return nil, err
	}
	if cs.Status != model.ChangesetApplied {
		return nil, errf(KindConflict, "only applied changesets can be reverted")
	}
	var st stored
	if len(cs.Result) > 0 {
		if err := json.Unmarshal(cs.Result, &st); err != nil {
			return nil, err
		}
	}
	if len(st.Undo) == 0 {
		return nil, errf(KindConflict, "changeset %d has nothing to undo", csID)
	}
	rev := &model.Changeset{Description: fmt.Sprintf("revert changeset #%d: %s", cs.ID, cs.Description), UserID: p.User.ID, Username: p.Name(), Via: "revert", Status: model.ChangesetOpen, Ops: st.Undo}
	now := e.Now()
	for i := range rev.Ops {
		rev.Ops[i].Seq, rev.Ops[i].AddedAt = i+1, now
	}
	if err := e.store.CreateChangeset(ctx, rev); err != nil {
		return nil, err
	}
	e.audit(ctx, p, "changeset.revert", "changeset", rev.ID, rev.Description, cs.ID, "success", "", nil)
	return rev, nil
}

// Resync renders the current configuration and deploys it, overwriting
// whatever is on disk (used for the first deployment and to repair
// drift).
func (e *Engine) Resync(ctx context.Context, p *auth.Principal, opts ApplyOptions) (*Result, error) {
	e.mu.Lock()
	defer e.mu.Unlock()
	base, err := e.store.LoadState(ctx)
	if err != nil {
		return nil, err
	}
	work := base.Clone()
	pl, err := e.makePlan(ctx, base, work, p.Name())
	if err != nil {
		return nil, err
	}
	res := pl.result()
	res.DryRun = opts.DryRun
	if validate.HasErrors(pl.existing) {
		// Resync deploys the stored state as is; it must be valid.
		return res, &Error{Kind: KindValidation, Message: "the stored configuration has errors: " + describeErrors(pl.existing), Problems: validate.Errors(pl.existing)}
	}
	if pl.check != nil && !pl.check.OK {
		return res, &Error{Kind: KindValidation, Message: "named-checkconf rejected the configuration: " + firstCheckError(pl.checkMsgs), Check: pl.checkMsgs}
	}
	if opts.DryRun {
		return res, nil
	}
	if len(pl.drift) > 0 && !opts.Force {
		var paths []string
		for _, d := range pl.drift {
			paths = append(paths, d.Path+" ("+d.Reason+")")
		}
		return res, &Error{Kind: KindDrift, Message: "files managed by Jarl were changed outside Jarl: " + strings.Join(paths, ", ") + "; resync with force to overwrite them"}
	}
	// Verify every primary zone and reload them all.
	pl.configChanged = true
	pl.reloadZones = nil
	for _, z := range work.SortedZones() {
		if z.Type == model.ZonePrimary {
			pl.reloadZones = append(pl.reloadZones, z)
			pl.verify[z.ID] = z.SOA.Serial
		}
	}
	reload := !e.cfg.NoReload && !opts.NoReload
	if adopted, aerr := e.Adopted(); reload && (aerr != nil || !adopted) {
		reload = false
		res.Warnings = append(res.Warnings, "named does not load Jarl's configuration yet; files written without reloading named")
	}
	if err := e.deployAndReload(ctx, pl, res, reload); err != nil {
		return res, err
	}
	deployed := deployedList(pl.out, 0, e.Now())
	err = e.store.InTx(ctx, func(tx *store.Tx) error {
		if err := tx.SetDeployedFiles(deployed); err != nil {
			return err
		}
		return tx.AddAudit(e.auditEvent(p, "system.resync", "", 0, "", 0, "success", fmt.Sprintf("%d files written", len(res.Files)), nil))
	})
	if err != nil {
		return res, err
	}
	res.Applied = true
	return res, nil
}

// SyncConfig rewrites named's configuration when it differs from what the
// database and jarld's own configuration produce, for example after query
// statistics were switched on or off in jarl.yaml. Files changed by hand
// are never overwritten. It reports whether anything was written.
func (e *Engine) SyncConfig(ctx context.Context) (bool, *Result, error) {
	dry, err := e.Resync(ctx, auth.System, ApplyOptions{DryRun: true})
	if err != nil || len(dry.Files) == 0 {
		return false, dry, err
	}
	res, err := e.Resync(ctx, auth.System, ApplyOptions{})
	return err == nil, res, err
}

// SetResponseLogging turns named's response logging on or off with rndc.
// The responselog option only takes effect when named starts, so query
// statistics switch it at runtime. It reports whether it changed.
func (e *Engine) SetResponseLogging(ctx context.Context, on bool) (bool, error) {
	if e.cfg.NoReload || e.tools == nil {
		return false, nil
	}
	out, err := e.tools.RNDCRun(ctx, "status")
	if err != nil {
		return false, err
	}
	want := "response logging is OFF"
	if !on {
		want = "response logging is ON"
	}
	if !strings.Contains(out, want) {
		return false, nil // already as wanted (or too old a named to say)
	}
	arg := "on"
	if !on {
		arg = "off"
	}
	if _, err := e.tools.RNDCRun(ctx, "responselog", arg); err != nil {
		return false, err
	}
	return true, nil
}

// CheckCurrent validates the stored configuration and runs named-checkconf
// on it without changing anything.
func (e *Engine) CheckCurrent(ctx context.Context) (*Result, error) {
	base, err := e.store.LoadState(ctx)
	if err != nil {
		return nil, err
	}
	pl, err := e.makePlan(ctx, base, base.Clone(), "")
	if err != nil {
		return nil, err
	}
	res := pl.result()
	res.DryRun = true
	res.Problems = nonNilProblems(pl.existing)
	res.Existing = nil
	res.Valid = !validate.HasErrors(pl.existing) && res.CheckOK
	return res, nil
}

// Status summarizes the state of Jarl and named.
type Status struct {
	Adopted        bool               `json:"adopted"`
	AdoptError     string             `json:"adopt_error,omitempty"`
	Named          *bind.ServerStatus `json:"named,omitempty"`
	NamedError     string             `json:"named_error,omitempty"`
	BINDVersion    string             `json:"bind_version,omitempty"`
	Drift          []DriftEntry       `json:"drift"`
	Deployed       bool               `json:"deployed"`
	LastApplied    *model.Changeset   `json:"last_applied,omitempty"`
	OpenChangesets int                `json:"open_changesets"`
	Problems       []validate.Problem `json:"problems"`
	Counts         map[string]int     `json:"counts"`
	Paths          render.Paths       `json:"paths"`
	MainConfig     string             `json:"main_config"`
}

// Status returns the current status.
func (e *Engine) Status(ctx context.Context) (*Status, error) {
	st, err := e.store.LoadState(ctx)
	if err != nil {
		return nil, err
	}
	s := &Status{Paths: e.cfg.Paths, MainConfig: e.cfg.MainConfig, Drift: []DriftEntry{}}
	s.Adopted, err = e.Adopted()
	if err != nil {
		s.AdoptError = err.Error()
	}
	if !e.cfg.NoReload && e.tools != nil {
		ns, err := e.tools.Status(ctx)
		if err != nil {
			s.NamedError = err.Error()
		} else {
			s.Named = ns
		}
	}
	if e.tools != nil {
		if v, err := e.tools.Version(ctx); err == nil {
			s.BINDVersion = v
		}
	}
	deployed, err := e.store.DeployedFiles(ctx)
	if err != nil {
		return nil, err
	}
	s.Deployed = len(deployed) > 0
	s.Drift = append(s.Drift, CheckDrift(deployed)...)
	if list, err := e.store.ListChangesets(ctx, store.ChangesetFilter{Status: model.ChangesetApplied, Limit: 1}); err == nil && len(list) > 0 {
		s.LastApplied = list[0]
		s.LastApplied.Ops = RedactOps(s.LastApplied.Ops)
	}
	if list, err := e.store.ListChangesets(ctx, store.ChangesetFilter{Status: model.ChangesetOpen, Limit: 1000}); err == nil {
		s.OpenChangesets = len(list)
	}
	s.Problems = nonNilProblems(e.validate(st))
	s.Counts = map[string]int{"views": len(st.Views), "zones": len(st.Zones), "records": len(st.Records), "acls": len(st.ACLs), "tsig_keys": len(st.Keys), "options": len(st.Options)}
	return s, nil
}

// ---- audit

func (e *Engine) auditEvent(p *auth.Principal, action, objType string, objID int64, objName string, csID int64, outcome, msg string, details json.RawMessage) *model.AuditEvent {
	ev := &model.AuditEvent{Time: e.Now(), Action: action, ObjectType: objType, ObjectID: objID, ObjectName: objName, ChangesetID: csID, Outcome: outcome, Message: msg, Details: details}
	if p != nil {
		if p.User != nil {
			ev.UserID, ev.Username = p.User.ID, p.User.Username
		}
		ev.Auth, ev.IP, ev.UserAgent = p.Method, p.IP, p.Agent
		if p.Token != nil {
			ev.TokenID, ev.TokenName = p.Token.ID, p.Token.Name
		}
	}
	return ev
}

func (e *Engine) audit(ctx context.Context, p *auth.Principal, action, objType string, objID int64, objName string, csID int64, outcome, msg string, details json.RawMessage) {
	ev := e.auditEvent(p, action, objType, objID, objName, csID, outcome, msg, details)
	if err := e.store.AddAudit(ctx, ev); err != nil {
		e.log.Error("writing audit event failed", "err", err, "action", action)
	}
	e.log.Info("audit", "action", action, "user", ev.Username, "auth", ev.Auth, "token", ev.TokenName, "ip", ev.IP, "object", objName, "changeset", csID, "outcome", outcome, "message", msg)
}

// Audit records an audit event (used by other packages).
func (e *Engine) Audit(ctx context.Context, p *auth.Principal, action, objType string, objID int64, objName, outcome, msg string) {
	e.audit(ctx, p, action, objType, objID, objName, 0, outcome, msg, nil)
}

// ZoneLabel is exported for other packages.
func ZoneLabel(st *model.State, z *model.Zone) string { return zoneDesc(st, z) }

// Display is a helper for names.
func Display(n string) string { return dnsname.Display(n) }

// WorkingState returns the configuration as it would be after applying an
// open changeset (work), and the current configuration (base).
func (e *Engine) WorkingState(ctx context.Context, csID int64, p *auth.Principal) (work, base *model.State, err error) {
	cs, err := e.loadChangeset(ctx, csID, p, false)
	if err != nil {
		return nil, nil, err
	}
	base, err = e.store.LoadState(ctx)
	if err != nil {
		return nil, nil, err
	}
	if cs.Status != model.ChangesetOpen {
		return base, base, nil
	}
	work, _, err = e.replay(base, cs.Ops, p)
	if err != nil {
		return nil, nil, err
	}
	return work, base, nil
}