fancy-dns/internal/api/handlers.go

package api

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"net/http"
	"os"
	"sort"
	"strings"
	"time"

	"gopkg.in/yaml.v3"

	"git.k8s-git-k6n.net/mats/fancy-dns/internal/config"
)

// metricsHandler is injected by the app so the api package does not import the
// metrics package directly (keeps the dependency one-way).
var metricsHandler func() http.Handler

// SetMetricsHandler installs the /metrics handler.
func SetMetricsHandler(h http.Handler) {
	metricsHandler = func() http.Handler { return h }
}

// handleHealth is the liveness/readiness probe. It is unauthenticated on
// purpose, but deliberately reveals only a coarse status.
func (s *Server) handleHealth(w http.ResponseWriter, _ *http.Request) {
	h := s.hooks.Health()
	code := http.StatusOK
	if st, ok := h.(map[string]any); ok {
		if v, _ := st["status"].(string); v == "down" {
			code = http.StatusServiceUnavailable
		}
	}
	writeJSON(w, code, h)
}

// handleStatus reports the runtime state of every component.
func (s *Server) handleStatus(w http.ResponseWriter, _ *http.Request) {
	writeJSON(w, http.StatusOK, map[string]any{
		"version":  s.hooks.Version,
		"uptime_s": int64(time.Since(s.hooks.StartedAt).Seconds()),
		"started":  s.hooks.StartedAt,
		"config": map[string]any{
			"version":      s.hooks.Store.Get().Version,
			"generation":   s.hooks.Store.Generation(),
			"path":         s.hooks.Store.Path(),
			"last_error":   s.hooks.Store.LastError(),
			"writable":     isFileWritable(s.hooks.Store.Path()),
			"reload_ready": s.hooks.Store.Path() != "",
		},
		"components": s.hooks.Status(),
	})
}

// handleDashboard serves the aggregated dashboard payload.
func (s *Server) handleDashboard(w http.ResponseWriter, r *http.Request) {
	if s.hooks.Dashboard == nil {
		writeError(w, http.StatusNotImplemented, "dashboard not available")
		return
	}
	v, err := s.hooks.Dashboard(r.Context(), r.URL.Query().Get("range"))
	if err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, v)
}

// handleBlocklists serves the compiled blocklist state.
func (s *Server) handleBlocklists(w http.ResponseWriter, _ *http.Request) {
	out := map[string]any{}
	if s.hooks.BlocklistState != nil {
		if m, ok := s.hooks.BlocklistState().(map[string]any); ok {
			out = m
		}
	}
	writeJSON(w, http.StatusOK, out)
}

// handleBlocklistRefresh recompiles the blocklists.
func (s *Server) handleBlocklistRefresh(w http.ResponseWriter, r *http.Request) {
	if s.hooks.RefreshBlocklists == nil {
		writeError(w, http.StatusNotImplemented, "blocklist refresh not available")
		return
	}
	var in struct {
		Names []string `json:"names"`
	}
	// An empty body is valid and means "refresh everything".
	if r.ContentLength > 0 {
		if err := decodeJSON(r, &in); err != nil {
			writeError(w, http.StatusBadRequest, "malformed request body")
			return
		}
	}
	if in.Names == nil {
		// A nil slice marshals to null, and the console counts the names it
		// gets back to build its toast — so "refresh everything" reported
		// "0 sources". The key set must not depend on the request.
		in.Names = []string{}
	}
	if err := s.hooks.RefreshBlocklists(r.Context(), in.Names); err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	s.log.Info("blocklist refresh requested", "names", in.Names)
	writeJSON(w, http.StatusOK, map[string]any{"started": true, "names": in.Names})
}

// handleBlocklistToggle flips a source's enabled flag.
func (s *Server) handleBlocklistToggle(w http.ResponseWriter, r *http.Request) {
	if s.hooks.ToggleBlocklist == nil {
		writeError(w, http.StatusNotImplemented, "blocklist toggle not available")
		return
	}
	var in struct {
		Name    string `json:"name"`
		Enabled bool   `json:"enabled"`
	}
	if err := decodeJSON(r, &in); err != nil || in.Name == "" {
		writeError(w, http.StatusBadRequest, "name is required")
		return
	}
	if err := s.hooks.ToggleBlocklist(in.Name, in.Enabled); err != nil {
		writeError(w, http.StatusNotFound, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "name": in.Name, "enabled": in.Enabled})
}

// handleDomains pages the compiled rules.
func (s *Server) handleDomains(w http.ResponseWriter, r *http.Request) {
	if s.hooks.ListDomains == nil {
		writeError(w, http.StatusNotImplemented, "domain listing not available")
		return
	}
	q := DomainQuery{
		Text:    r.URL.Query().Get("q"),
		Action:  r.URL.Query().Get("action"),
		List:    r.URL.Query().Get("list"),
		Page:    queryInt(r, "page", 1),
		PerPage: queryInt(r, "per_page", 100),
	}
	if q.Page < 1 {
		q.Page = 1
	}
	if q.PerPage < 1 || q.PerPage > 1000 {
		q.PerPage = 100
	}
	v, err := s.hooks.ListDomains(q)
	if err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, v)
}

// handleDomainCheck explains how a name would be treated, which is the fastest
// way for an operator to debug "why is this not blocked".
func (s *Server) handleDomainCheck(w http.ResponseWriter, r *http.Request) {
	if s.hooks.DomainCheck == nil {
		writeError(w, http.StatusNotImplemented, "domain check not available")
		return
	}
	name := strings.TrimSpace(r.URL.Query().Get("name"))
	if name == "" {
		writeError(w, http.StatusBadRequest, "name is required")
		return
	}
	v, err := s.hooks.DomainCheck(name)
	if err != nil {
		writeJSON(w, http.StatusOK, map[string]any{"name": name, "error": err.Error()})
		return
	}
	writeJSON(w, http.StatusOK, v)
}

// handleDNSTest answers one query through the live pipeline. It is the
// console's "is my DNS actually answering" button: the answer comes from the
// same engine the listeners use, so what it shows is what a client gets, and
// a query error is reported inside the payload rather than as a status code —
// a failed lookup is a result here, not a failed request.
func (s *Server) handleDNSTest(w http.ResponseWriter, r *http.Request) {
	if s.hooks.Resolve == nil {
		writeError(w, http.StatusNotImplemented, "dns test not available")
		return
	}
	var in struct {
		Name     string `json:"name"`
		QType    string `json:"type"`
		Upstream string `json:"upstream"`
	}
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body")
		return
	}
	name := strings.TrimSpace(in.Name)
	if name == "" {
		writeError(w, http.StatusBadRequest, "name is required")
		return
	}
	resolve := s.hooks.Resolve
	if in.Upstream != "" {
		resolve = func(ctx context.Context, n, qt string) (any, error) {
			if s.hooks.ResolveUpstream == nil {
				return nil, errors.New("direct upstream lookup unavailable")
			}
			return s.hooks.ResolveUpstream(ctx, n, qt, in.Upstream)
		}
	}
	if resolve == nil {
		writeError(w, http.StatusNotImplemented, "dns test not available")
		return
	}
	v, err := resolve(r.Context(), name, in.QType)
	if err != nil {
		writeJSON(w, http.StatusOK, map[string]any{"name": name, "error": err.Error()})
		return
	}
	writeJSON(w, http.StatusOK, v)
}

func (s *Server) handleAddressOverrides(w http.ResponseWriter, r *http.Request) {
	if s.hooks.AddressOverrides == nil {
		writeError(w, http.StatusNotImplemented, "address overrides unavailable")
		return
	}
	v, err := s.hooks.AddressOverrides(r.Context())
	if err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, v)
}
func (s *Server) handleAddressOverridePut(w http.ResponseWriter, r *http.Request) {
	if s.hooks.SaveAddressOverride == nil {
		writeError(w, http.StatusNotImplemented, "address overrides are read-only")
		return
	}
	var in struct {
		Domain  string `json:"domain"`
		Address string `json:"address"`
	}
	if decodeJSON(r, &in) != nil {
		writeError(w, http.StatusBadRequest, "malformed request body")
		return
	}
	if err := s.hooks.SaveAddressOverride(r.Context(), in.Domain, in.Address); err != nil {
		writeError(w, http.StatusBadRequest, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true})
}
func (s *Server) handleAddressOverrideDelete(w http.ResponseWriter, r *http.Request) {
	if s.hooks.DeleteAddressOverride == nil {
		writeError(w, http.StatusNotImplemented, "address overrides are read-only")
		return
	}
	domain := r.PathValue("domain")
	if err := s.hooks.DeleteAddressOverride(r.Context(), domain); err != nil {
		writeError(w, http.StatusBadRequest, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true})
}

func (s *Server) handleLocalZones(w http.ResponseWriter, r *http.Request) {
	if s.hooks.LocalZones == nil {
		writeError(w, http.StatusNotImplemented, "local zone editor is unavailable")
		return
	}
	zones, err := s.hooks.LocalZones(r.Context())
	if err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"zones": zones})
}

func (s *Server) handleLocalZonePut(w http.ResponseWriter, r *http.Request) {
	if s.hooks.SaveLocalZone == nil {
		writeError(w, http.StatusNotImplemented, "local zone editor is unavailable")
		return
	}
	var in struct {
		Content string `json:"content"`
	}
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body")
		return
	}
	name := r.PathValue("name")
	if err := s.hooks.SaveLocalZone(r.Context(), name, in.Content); err != nil {
		writeError(w, http.StatusBadRequest, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "name": name})
}

func (s *Server) handleLocalZoneDelete(w http.ResponseWriter, r *http.Request) {
	if s.hooks.DeleteLocalZone == nil {
		writeError(w, http.StatusNotImplemented, "local zone editor is unavailable")
		return
	}
	name := r.PathValue("name")
	if err := s.hooks.DeleteLocalZone(r.Context(), name); err != nil {
		writeError(w, http.StatusBadRequest, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "name": name})
}

// handleDomainAdd adds an inline rule.
func (s *Server) handleDomainAdd(w http.ResponseWriter, r *http.Request) {
	if s.hooks.AddDomain == nil {
		writeError(w, http.StatusNotImplemented, "domain add not available")
		return
	}
	var in DomainInput
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body")
		return
	}
	if strings.TrimSpace(in.Name) == "" {
		writeError(w, http.StatusBadRequest, "name is required")
		return
	}
	v, err := s.hooks.AddDomain(r.Context(), in)
	if err != nil {
		writeError(w, http.StatusBadRequest, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, v)
}

// handleDomainDelete removes a rule.
func (s *Server) handleDomainDelete(w http.ResponseWriter, r *http.Request) {
	if s.hooks.DeleteDomain == nil {
		writeError(w, http.StatusNotImplemented, "domain delete not available")
		return
	}
	// The route is {name...} so a rule name containing dots or a wildcard
	// prefix survives without percent-encoding gymnastics.
	name := r.PathValue("name")
	if name == "" {
		writeError(w, http.StatusBadRequest, "name is required")
		return
	}
	if err := s.hooks.DeleteDomain(r.Context(), name, r.URL.Query().Get("action"), r.URL.Query().Get("list")); err != nil {
		writeError(w, http.StatusNotFound, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "name": name})
}

// handleClients returns the ACL groups.
func (s *Server) handleClients(w http.ResponseWriter, _ *http.Request) {
	if s.hooks.GetACL == nil {
		writeError(w, http.StatusNotImplemented, "acl not available")
		return
	}
	writeJSON(w, http.StatusOK, s.hooks.GetACL())
}

// aclPayload is the PUT body for the clients endpoint.
type aclPayload struct {
	Groups        []config.Group `json:"groups"`
	DefaultPolicy string         `json:"default_policy"`
}

// handleClientsPut validates and persists new ACL groups.
func (s *Server) handleClientsPut(w http.ResponseWriter, r *http.Request) {
	if s.hooks.PutACL == nil {
		writeError(w, http.StatusNotImplemented, "acl write not available")
		return
	}
	var in aclPayload
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body: "+err.Error())
		return
	}
	version, err := s.hooks.PutACL(r.Context(), in.Groups, in.DefaultPolicy)
	if err != nil {
		writeErrorFields(w, http.StatusBadRequest, err.Error(), configErrorFields(err))
		return
	}
	s.log.Info("acl updated", "groups", len(in.Groups), "version", version)
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "version": version})
}

// handleUpstreams returns the upstream pool state.
func (s *Server) handleUpstreams(w http.ResponseWriter, _ *http.Request) {
	if s.hooks.GetUpstream == nil {
		writeError(w, http.StatusNotImplemented, "upstreams not available")
		return
	}
	writeJSON(w, http.StatusOK, s.hooks.GetUpstream())
}

// upstreamPayload is the PUT body for the upstreams endpoint. The health
// fields the console sends (health, latency_ms, fail_rate, breaker_open and
// the counters) are accepted and ignored: the server owns them, and refusing
// the object the console just read is what made every save answer 400. They are
// carried on config.UpstreamServer with `json:"-"` on the yaml side, so
// settings.yaml does not gain runtime state.
//
// HealthCheckInterval accepts both spellings the console has used: the
// duration string ("30s") and the older seconds number under its old name.
type upstreamPayload struct {
	Policy             string                     `json:"policy"`
	Servers            []config.UpstreamServer    `json:"servers"`
	Bootstraps         []config.UpstreamBootstrap `json:"bootstraps"`
	HealthCheck        config.Duration            `json:"healthcheck_interval"`
	HealthCheckSeconds float64                    `json:"healthcheck_interval_s"`
	MaxAttempts        int                        `json:"max_attempts"`
	BreakerThreshold   int                        `json:"breaker_threshold"`
	// BreakerCooldown was read-only in the console; accepting it means a
	// round-trip of the object the console read does not fail.
	BreakerCooldown config.Duration `json:"breaker_cooldown"`
}

// handleUpstreamsPut validates and persists a new upstream config.
func (s *Server) handleUpstreamsPut(w http.ResponseWriter, r *http.Request) {
	if s.hooks.PutUpstream == nil {
		writeError(w, http.StatusNotImplemented, "upstream write not available")
		return
	}
	var in upstreamPayload
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body: "+err.Error())
		return
	}
	// Preserve fields the UI does not manage, so a UI write never silently
	// resets a breaker cooldown someone set by hand.
	cur := s.hooks.Store.Get().Upstream
	healthCheck := in.HealthCheck
	if healthCheck <= 0 {
		// The older seconds-number spelling; zero also means "not supplied".
		healthCheck = config.Duration(in.HealthCheckSeconds * float64(time.Second))
	}
	cooldown := cur.BreakerCooldown
	if in.BreakerCooldown > 0 {
		cooldown = in.BreakerCooldown
	}
	up := config.Upstream{
		Policy:           in.Policy,
		Servers:          in.Servers,
		Bootstraps:       in.Bootstraps,
		MaxAttempts:      in.MaxAttempts,
		HealthCheck:      healthCheck,
		BreakerThreshold: in.BreakerThreshold,
		BreakerCooldown:  cooldown,
	}
	if up.HealthCheck <= 0 {
		up.HealthCheck = cur.HealthCheck
	}
	if up.BreakerThreshold <= 0 {
		up.BreakerThreshold = cur.BreakerThreshold
	}
	version, err := s.hooks.PutUpstream(r.Context(), up)
	if err != nil {
		writeErrorFields(w, http.StatusBadRequest, err.Error(), configErrorFields(err))
		return
	}
	s.log.Info("upstreams updated", "servers", len(up.Servers), "version", version)
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "version": version})
}

// handleLogs pages the query log.
func (s *Server) handleLogs(w http.ResponseWriter, r *http.Request) {
	if s.hooks.SearchLog == nil {
		writeError(w, http.StatusNotImplemented, "query log not available")
		return
	}
	q := LogQuery{
		Limit:       queryInt(r, "limit", 200),
		Offset:      queryInt(r, "offset", 0),
		Text:        r.URL.Query().Get("q"),
		Action:      r.URL.Query().Get("action"),
		Client:      r.URL.Query().Get("client"),
		RCode:       queryInt(r, "rcode", 0),
		BlockedOnly: queryBool(r, "blocked"),
		Since:       queryTime(r, "since"),
	}
	v, err := s.hooks.SearchLog(r.Context(), q)
	if err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, v)
}

// handleLogStream is the Server-Sent Events endpoint feeding the live view.
func (s *Server) handleLogStream(w http.ResponseWriter, r *http.Request) {
	flusher, ok := w.(http.Flusher)
	if !ok {
		writeError(w, http.StatusInternalServerError, "streaming unsupported")
		return
	}
	w.Header().Set("Content-Type", "text/event-stream")
	w.Header().Set("Cache-Control", "no-store")
	w.Header().Set("Connection", "keep-alive")
	// Without this, a reverse proxy buffers the stream and the UI shows
	// nothing until the buffer fills.
	w.Header().Set("X-Accel-Buffering", "no")
	w.WriteHeader(http.StatusOK)
	flusher.Flush()

	send := func(event string, v any) bool {
		b, err := json.Marshal(v)
		if err != nil {
			return true
		}
		// A single-line data field is what the SSE grammar requires, so the
		// payload is compact JSON with no newlines (json.Marshal never emits
		// a raw newline).
		if _, err := fmt.Fprintf(w, "event: %s\ndata: %s\n\n", event, b); err != nil {
			return false
		}
		flusher.Flush()
		return true
	}
	// An initial event tells the client the stream is live, so the UI can show
	// a "connected" state before any query happens.
	if !send("ready", map[string]any{"at": time.Now(), "version": s.hooks.Version}) {
		return
	}

	var (
		entries <-chan LogEntry
		cancel  func()
	)
	if s.hooks.SubscribeLog != nil {
		entries, cancel = s.hooks.SubscribeLog()
		defer cancel()
	}

	stats := time.NewTicker(time.Second)
	defer stats.Stop()
	ping := time.NewTicker(20 * time.Second)
	defer ping.Stop()

	ctx := r.Context()
	for {
		select {
		case <-ctx.Done():
			return
		case e, ok := <-entries:
			if !ok {
				return
			}
			if !send("entry", e) {
				return
			}
		case <-stats.C:
			if s.hooks.LogStats == nil {
				continue
			}
			if !send("stats", s.hooks.LogStats()) {
				return
			}
		case <-ping.C:
			// A comment frame keeps proxies from timing the stream out.
			if _, err := fmt.Fprint(w, ": ping\n\n"); err != nil {
				return
			}
			flusher.Flush()
		}
	}
}

// handleLogsClear empties the query log.
func (s *Server) handleLogsClear(w http.ResponseWriter, _ *http.Request) {
	if s.hooks.ClearLog == nil {
		writeError(w, http.StatusNotImplemented, "query log not available")
		return
	}
	if err := s.hooks.ClearLog(); err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{"ok": true})
}

// handleCache returns the cache snapshot.
func (s *Server) handleCache(w http.ResponseWriter, _ *http.Request) {
	if s.hooks.CacheStats == nil {
		writeError(w, http.StatusNotImplemented, "cache not available")
		return
	}
	writeJSON(w, http.StatusOK, s.hooks.CacheStats())
}

// handleCacheTop returns the hottest cached names.
func (s *Server) handleCacheTop(w http.ResponseWriter, r *http.Request) {
	if s.hooks.CacheTop == nil {
		writeError(w, http.StatusNotImplemented, "cache not available")
		return
	}
	writeJSON(w, http.StatusOK, map[string]any{
		"items": s.hooks.CacheTop(queryInt(r, "limit", 50)),
	})
}

// handleCacheFlush empties the cache.
func (s *Server) handleCacheFlush(w http.ResponseWriter, _ *http.Request) {
	if s.hooks.FlushCache == nil {
		writeError(w, http.StatusNotImplemented, "cache not available")
		return
	}
	n := s.hooks.FlushCache()
	s.log.Info("cache flushed", "entries", n)
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "entries": n})
}

// handleSettingsGet returns the config document plus what is knowable about it.
func (s *Server) handleSettingsGet(w http.ResponseWriter, _ *http.Request) {
	path := s.hooks.Store.Path()
	var fileYAML string
	if path != "" {
		if b, err := os.ReadFile(path); err == nil {
			fileYAML = string(b)
		}
	}
	// The base, not the effective config: the settings editor writes what it
	// shows back to the file, and cluster objects must not end up in the YAML.
	cur := s.hooks.Store.Base()
	effective, err := config.Marshal(cur)
	if err != nil {
		writeError(w, http.StatusInternalServerError, "cannot render config: "+err.Error())
		return
	}
	defaults, _ := config.Marshal(config.Default())
	if fileYAML == "" {
		// Running without a file: show the effective document as the editable
		// text so the UI still works.
		fileYAML = string(effective)
	}
	writeJSON(w, http.StatusOK, map[string]any{
		"version":          cur.Version,
		"path":             path,
		"config_yaml":      fileYAML,
		"effective_yaml":   string(effective),
		"defaults_yaml":    string(defaults),
		"last_error":       s.hooks.Store.LastError(),
		"reload_available": path != "",
		"writable":         isFileWritable(path),
	})
}

// settingsPayload is the PUT body for the settings endpoint.
type settingsPayload struct {
	ConfigYAML string `json:"config_yaml"`
}

// handleSettingsPut validates, writes and swaps a new config.
func (s *Server) handleSettingsPut(w http.ResponseWriter, r *http.Request) {
	cfg := s.hooks.Store.Get()
	if !cfg.API.AllowConfigWrite {
		writeError(w, http.StatusForbidden, "config writes are disabled")
		return
	}
	var in settingsPayload
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body")
		return
	}
	if strings.TrimSpace(in.ConfigYAML) == "" {
		writeError(w, http.StatusBadRequest, "config_yaml is empty")
		return
	}
	next, err := parseConfigYAML(in.ConfigYAML)
	if err != nil {
		writeError(w, http.StatusBadRequest, "cannot parse config: "+err.Error())
		return
	}
	// Relative paths in the document resolve against the config's directory,
	// so a file written from a different cwd still finds its zone files.
	next.WorkingDirOrSet(cfg.Server.WorkingDir)
	if err := next.Validate(); err != nil {
		writeErrorFields(w, http.StatusBadRequest, "invalid config: "+err.Error(), configErrorFields(err))
		return
	}
	if err := s.commit(next); err != nil {
		writeError(w, http.StatusInternalServerError, err.Error())
		return
	}
	s.log.Info("config updated via api", "version", next.Version)
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "version": next.Version, "warnings": []string{}})
}

// handleSettingsValidate checks a document without applying it.
func (s *Server) handleSettingsValidate(w http.ResponseWriter, r *http.Request) {
	var in settingsPayload
	if err := decodeJSON(r, &in); err != nil {
		writeError(w, http.StatusBadRequest, "malformed request body")
		return
	}
	next, err := parseConfigYAML(in.ConfigYAML)
	if err != nil {
		writeJSON(w, http.StatusOK, map[string]any{
			"ok":       false,
			"errors":   []map[string]string{{"field": "config", "message": err.Error()}},
			"warnings": []map[string]string{},
		})
		return
	}
	next.WorkingDirOrSet(s.hooks.Store.Get().Server.WorkingDir)
	verr := next.Validate()
	out := map[string]any{"ok": verr == nil, "errors": []map[string]string{}, "warnings": []map[string]string{}}
	if verr != nil {
		errs := make([]map[string]string, 0)
		for _, f := range configErrorFields(verr) {
			errs = append(errs, map[string]string{"field": fieldOf(f), "message": f})
		}
		out["errors"] = errs
	}
	if warnings := configWarnings(next); len(warnings) > 0 {
		ws := make([]map[string]string, 0, len(warnings))
		for _, warn := range warnings {
			ws = append(ws, map[string]string{"field": fieldOf(warn), "message": warn})
		}
		out["warnings"] = ws
	}
	writeJSON(w, http.StatusOK, out)
}

// handleSettingsReload re-reads the config file from disk.
func (s *Server) handleSettingsReload(w http.ResponseWriter, _ *http.Request) {
	if s.hooks.Store.Path() == "" {
		writeError(w, http.StatusBadRequest, "no config file to reload from")
		return
	}
	if err := s.hooks.Store.Reload(); err != nil {
		writeError(w, http.StatusInternalServerError, "reload failed: "+err.Error())
		return
	}
	s.log.Info("config reloaded from disk", "version", s.hooks.Store.Get().Version)
	writeJSON(w, http.StatusOK, map[string]any{"ok": true, "version": s.hooks.Store.Get().Version})
}

// commit writes the config to disk, swaps it in and notifies the cluster.
func (s *Server) commit(next *config.Config) error {
	path := s.hooks.Store.Path()
	if path == "" {
		return fmt.Errorf("cannot persist: no config file path is known")
	}
	if !isFileWritable(path) {
		return fmt.Errorf("cannot persist: %s is not writable", path)
	}
	next.Version = s.hooks.Store.Get().Version + 1
	if err := next.Save(path); err != nil {
		return fmt.Errorf("write config: %w", err)
	}
	return s.hooks.Store.Swap(next)
}

// Commit exposes the config commit path so a component (for example the ACL or
// upstream editor) goes through the same validate-write-swap flow.
func (s *Server) Commit(next *config.Config) error { return s.commit(next) }

// parseConfigYAML decodes a config document on top of the defaults, which is
// what makes a partial document from the UI editor work.
func parseConfigYAML(text string) (*config.Config, error) {
	next := config.Default()
	if err := yaml.Unmarshal([]byte(text), next); err != nil {
		return nil, err
	}
	return next, nil
}

// configErrorFields splits a joined validation error into individual messages.
func configErrorFields(err error) []string {
	if err == nil {
		return nil
	}
	msg := err.Error()
	parts := strings.Split(msg, "\n")
	out := make([]string, 0, len(parts))
	for _, p := range parts {
		if p = strings.TrimSpace(p); p != "" {
			out = append(out, p)
		}
	}
	if len(out) == 0 {
		out = append(out, msg)
	}
	sort.Strings(out)
	return out
}

// fieldOf extracts the leading dotted path from a validation message, e.g.
// "blocklist.lists[2]: name is required" becomes "blocklist.lists[2]".
func fieldOf(msg string) string {
	i := strings.Index(msg, ":")
	if i <= 0 {
		return "config"
	}
	if f := strings.TrimSpace(msg[:i]); f != "" {
		return f
	}
	return "config"
}

// configWarnings reports non-fatal configuration smells. A config can be valid
// and still a bad idea, and this is where a validator earns its keep.
func configWarnings(cfg *config.Config) []string {
	var out []string
	add := func(f string, a ...any) { out = append(out, fmt.Sprintf(f, a...)) }

	if cfg.API.Auth.Mode == "none" {
		add("api.auth.mode: the admin API is unauthenticated; anyone who can reach it can rewrite your config")
	}
	if !cfg.DNS.DoT.Enabled && !cfg.DNS.DoH.Enabled {
		add("dns: neither DoT nor DoH is enabled, so clients have no encrypted transport")
	}
	if cfg.Cache.Enabled && !cfg.Cache.Persist.Enabled {
		add("cache.persist.enabled: the cache will be cold after every restart")
	}
	if cfg.Cache.MaxTTL.D() > 24*time.Hour {
		add("cache.max_ttl: caching for more than a day serves records upstream has already forgotten")
	}
	if !cfg.Block.Enabled {
		add("blocklist.enabled: blocklists are off, so this instance is an open resolver for everything")
	} else if len(cfg.Block.Lists) == 0 && len(cfg.Block.LocalZones) == 0 &&
		len(cfg.Block.BlockEntries) == 0 {
		add("blocklist: enabled but no source is configured")
	}
	for _, l := range cfg.Block.Lists {
		if strings.HasPrefix(l.URL, "http://") {
			add("blocklist.lists[%s]: the source is fetched over plain http and can be tampered with in transit", l.Name)
		}
		if l.Refresh.D() > 0 && l.Refresh.D() < time.Hour {
			add("blocklist.lists[%s]: a refresh interval under an hour is likely to get you rate-limited", l.Name)
		}
		if strings.EqualFold(l.Format, "rpz") && strings.ContainsAny(l.Path, "*?") {
			add("blocklist.lists[%s]: an rpz source should point at a zone file, not a pattern", l.Name)
		}
	}
	if cfg.ACL.DefaultPolicy == "allow" {
		for _, g := range cfg.ACL.Groups {
			if g.Match.Any {
				add("acl.groups[%s]: a catch-all group with policy allow makes the ACL a no-op", g.Name)
			}
		}
	}
	if len(cfg.Upstream.Servers) == 1 {
		add("upstream.servers: only one upstream is configured, so a failure of that resolver is an outage")
	}
	for _, u := range cfg.Upstream.Servers {
		if u.InsecureSkipVerify {
			add("upstream.servers[%s]: insecure_skip_verify disables TLS validation", u.Addr)
		}
	}
	if cfg.QueryLog.Enabled && cfg.QueryLog.SampleRate == 1 && cfg.QueryLog.LogCached {
		add("qlog: logging every cached query on a busy instance is a lot of disk for little value; consider lowering sample_rate")
	}
	if cfg.Metrics.Enabled && !cfg.Metrics.Protect {
		add("metrics.protect: /metrics is unauthenticated and exposes client and domain statistics")
	}
	return out
}