server: harden metric query proxy

server/api/metric.go

@@ -15,42 +15,310 @@
 package api
 
 import (
-	"fmt"
+	"context"
+	"encoding/json"
+	"io"
+	"net"
 	"net/http"
+	"net/netip"
 	"net/url"
+	"strconv"
 	"strings"
+	"time"
+
+	"go.uber.org/zap"
+
+	"github.com/pingcap/errors"
+	"github.com/pingcap/log"
 
-	"github.com/tikv/pd/pkg/utils/apiutil"
 	"github.com/tikv/pd/server"
 )
 
+const (
+	metricQueryTimeout             = 30 * time.Second
+	maxMetricQueryResponseBodySize = int64(32 << 20)
+	metricQueryErrorBody           = `{"status":"error","errorType":"proxy","error":"metric query failed"}`
+)
+
+type metricIPResolver interface {
+	LookupNetIP(context.Context, string, string) ([]netip.Addr, error)
+}
+
 type queryMetric struct {
-	s *server.Server
+	s            *server.Server
+	transport    *http.Transport
+	transportErr error
+	resolver     metricIPResolver
+}
+
+type resolvedMetricTarget struct {
+	url       *url.URL
+	port      string
+	addresses []netip.Addr
 }
 
+type metricTargetContextKey struct{}
+
 func newqueryMetric(s *server.Server) *queryMetric {
-	return &queryMetric{s: s}
+	transport, err := newMetricQueryTransport(s.GetHTTPClient())
+	return &queryMetric{
+		s:            s,
+		transport:    transport,
+		transportErr: err,
+		resolver:     net.DefaultResolver,
+	}
 }
 
 func (h *queryMetric) queryMetric(w http.ResponseWriter, r *http.Request) {
-	metricAddr := h.s.GetConfig().PDServerCfg.MetricStorage
-	if metricAddr == "" {
-		http.Error(w, "metric storage doesn't set", http.StatusInternalServerError)
+	if h.transportErr != nil {
+		writeMetricQueryError(w, h.transportErr)
 		return
 	}
-	u, err := url.Parse(metricAddr)
+	proxyMetricQuery(
+		w,
+		r,
+		h.s.GetConfig().PDServerCfg.MetricStorage,
+		h.transport,
+		h.resolver,
+	)
+}
+
+func proxyMetricQuery(
+	w http.ResponseWriter,
+	r *http.Request,
+	metricStorage string,
+	transport http.RoundTripper,
+	resolver metricIPResolver,
+) {
+	ctx, cancel := context.WithTimeout(r.Context(), metricQueryTimeout)
+	defer cancel()
+
+	target, err := resolveMetricStorageTarget(ctx, metricStorage, resolver)
 	if err != nil {
-		http.Error(w, err.Error(), http.StatusInternalServerError)
+		writeMetricQueryError(w, err)
 		return
 	}
+	ctx = context.WithValue(ctx, metricTargetContextKey{}, target)
+
+	request := r.Clone(ctx)
+	request.RequestURI = ""
+	request.URL.Scheme = target.url.Scheme
+	request.URL.Host = target.url.Host
+	request.URL.Path = strings.Replace(r.URL.Path, "/pd/api/v1/metric", "/api/v1", 1)
+	// Preserve end-to-end headers for authenticated metric storage, matching the
+	// historical proxy contract, but never forward connection-specific headers.
+	request.Header = r.Header.Clone()
+	removeHopByHopHeaders(request.Header)
+	request.Header.Del("Accept-Encoding")
+	request.Header.Set("Accept", "application/json")
+
+	// RoundTrip does not follow redirects. The target is resolved and validated before dialing.
+	response, err := transport.RoundTrip(request) //nolint:gosec
+	if err != nil {
+		writeMetricQueryError(w, errors.Annotate(err, "metric storage request failed"))
+		return
+	}
+	defer response.Body.Close()
+	if response.StatusCode < http.StatusOK || response.StatusCode >= http.StatusMultipleChoices {
+		writeMetricQueryError(w, errors.Errorf("metric storage returned status %d", response.StatusCode))
+		return
+	}
+
+	body, err := readMetricQueryResponse(response.Body, maxMetricQueryResponseBodySize)
+	if err != nil {
+		writeMetricQueryError(w, err)
+		return
+	}
+	if err := validatePrometheusQueryResponse(body); err != nil {
+		writeMetricQueryError(w, err)
+		return
+	}
+
+	w.Header().Set("Content-Type", "application/json; charset=utf-8")
+	w.Header().Set("Cache-Control", "no-store")
+	w.WriteHeader(http.StatusOK)
+	if _, err := w.Write(body); err != nil {
+		log.Warn("failed to write metric query response", zap.Error(err))
+	}
+}
+
+func resolveMetricStorageTarget(
+	ctx context.Context,
+	rawURL string,
+	resolver metricIPResolver,
+) (*resolvedMetricTarget, error) {
+	targetURL, err := url.Parse(rawURL)
+	if err != nil || targetURL.Opaque != "" || targetURL.Hostname() == "" || targetURL.User != nil || targetURL.Fragment != "" {
+		return nil, errors.New("invalid metric-storage URL")
+	}
+	targetURL.Scheme = strings.ToLower(targetURL.Scheme)
+	if targetURL.Scheme != "http" && targetURL.Scheme != "https" {
+		return nil, errors.New("metric-storage must use HTTP or HTTPS")
+	}
+	hostname := targetURL.Hostname()
+	port := targetURL.Port()
+	if port == "" {
+		if strings.HasSuffix(targetURL.Host, ":") {
+			return nil, errors.New("metric-storage URL has an empty port")
+		}
+		if targetURL.Scheme == "http" {
+			port = "80"
+		} else {
+			port = "443"
+		}
+	} else if portNumber, err := strconv.Atoi(port); err != nil || portNumber < 1 || portNumber > 65535 {
+		return nil, errors.New("metric-storage URL has an invalid port")
+	}
 
-	switch u.Scheme {
-	case "http", "https":
-		// Replace the pd path with the prometheus http API path.
-		r.URL.Path = strings.Replace(r.URL.Path, "pd/api/v1/metric", "api/v1", 1)
-		apiutil.NewCustomReverseProxies(h.s.GetHTTPClient(), []url.URL{*u}).ServeHTTP(w, r)
+	var addresses []netip.Addr
+	if address, err := netip.ParseAddr(hostname); err == nil {
+		addresses = []netip.Addr{address}
+	} else {
+		addresses, err = resolver.LookupNetIP(ctx, "ip", hostname)
+		if err != nil {
+			return nil, errors.Annotate(err, "failed to resolve metric-storage hostname")
+		}
+	}
+	if len(addresses) == 0 {
+		return nil, errors.New("metric-storage hostname resolved to no addresses")
+	}
+
+	for i, address := range addresses {
+		addresses[i] = address.Unmap()
+		if !isSafeMetricTargetIP(addresses[i]) {
+			return nil, errors.New("metric-storage resolved to an unsafe address")
+		}
+	}
+
+	return &resolvedMetricTarget{
+		url:       targetURL,
+		port:      port,
+		addresses: addresses,
+	}, nil
+}
+
+func isSafeMetricTargetIP(address netip.Addr) bool {
+	address = address.Unmap()
+	unsafe := !address.IsValid() || address.Zone() != "" || address.IsLoopback() ||
+		address.IsLinkLocalUnicast() || address.IsLinkLocalMulticast() ||
+		address.IsUnspecified() || address.IsMulticast() || !address.IsGlobalUnicast()
+	if address.Is4() {
+		octets := address.As4()
+		unsafe = unsafe || octets[0] == 0 || octets[0] >= 240
+	}
+	switch address.String() {
+	case "169.254.169.254", "100.100.100.200", "fd00:ec2::254":
+		unsafe = true
+	}
+	return !unsafe
+}
+
+func newMetricQueryTransport(baseClient *http.Client) (*http.Transport, error) {
+	baseTransport := http.DefaultTransport
+	if baseClient != nil && baseClient.Transport != nil {
+		baseTransport = baseClient.Transport
+	}
+	transport, ok := baseTransport.(*http.Transport)
+	if !ok {
+		return nil, errors.New("metric query requires an HTTP transport")
+	}
+	transport = transport.Clone()
+	dialContext := transport.DialContext
+	if dialContext == nil {
+		dialContext = (&net.Dialer{}).DialContext
+	}
+	// A proxy would resolve and dial the target outside this transport, bypassing
+	// the destination validation below.
+	transport.Proxy = nil
+	transport.DialTLS = nil //nolint:staticcheck // Clear the deprecated hook as well to prevent a validation bypass.
+	transport.DialTLSContext = nil
+	transport.DialContext = func(ctx context.Context, network, _ string) (net.Conn, error) {
+		target, ok := ctx.Value(metricTargetContextKey{}).(*resolvedMetricTarget)
+		if !ok {
+			return nil, errors.New("metric query target is not validated")
+		}
+		var lastErr error
+		for _, address := range target.addresses {
+			connection, err := dialContext(ctx, network, net.JoinHostPort(address.String(), target.port))
+			if err == nil {
+				return connection, nil
+			}
+			lastErr = err
+			if ctx.Err() != nil {
+				break
+			}
+		}
+		return nil, errors.Annotate(lastErr, "failed to dial metric-storage target")
+	}
+
+	return transport, nil
+}
+
+func removeHopByHopHeaders(header http.Header) {
+	for _, connectionHeader := range header.Values("Connection") {
+		for _, name := range strings.Split(connectionHeader, ",") {
+			header.Del(strings.TrimSpace(name))
+		}
+	}
+	for _, name := range []string{
+		"Connection",
+		"Proxy-Connection",
+		"Keep-Alive",
+		"Proxy-Authenticate",
+		"Proxy-Authorization",
+		"Te",
+		"Trailer",
+		"Transfer-Encoding",
+		"Upgrade",
+	} {
+		header.Del(name)
+	}
+}
+
+func readMetricQueryResponse(body io.Reader, limit int64) ([]byte, error) {
+	limitedBody := io.LimitReader(body, limit+1)
+	data, err := io.ReadAll(limitedBody)
+	if err != nil {
+		return nil, errors.Annotate(err, "failed to read metric query response")
+	}
+	if int64(len(data)) > limit {
+		return nil, errors.New("metric query response exceeds the size limit")
+	}
+	return data, nil
+}
+
+func validatePrometheusQueryResponse(body []byte) error {
+	response := struct {
+		Status string `json:"status"`
+		Data   *struct {
+			ResultType string          `json:"resultType"`
+			Result     json.RawMessage `json:"result"`
+		} `json:"data"`
+	}{}
+	if err := json.Unmarshal(body, &response); err != nil {
+		return errors.Annotate(err, "metric storage returned invalid JSON")
+	}
+	if response.Status != "success" || response.Data == nil {
+		return errors.New("metric storage returned a non-success Prometheus response")
+	}
+	result := strings.TrimSpace(string(response.Data.Result))
+	if len(result) < 2 || result[0] != '[' || result[len(result)-1] != ']' {
+		return errors.New("metric storage returned an invalid Prometheus result")
+	}
+	switch response.Data.ResultType {
+	case "matrix", "vector", "scalar", "string":
+		return nil
 	default:
-		// TODO: Support read data by self after support store metric data in PD/TiKV.
-		http.Error(w, fmt.Sprintf("schema of metric storage address is no supported, address: %v", metricAddr), http.StatusInternalServerError)
+		return errors.New("metric storage returned an invalid Prometheus result type")
+	}
+}
+
+func writeMetricQueryError(w http.ResponseWriter, err error) {
+	log.Warn("failed to query metric storage", zap.Error(err))
+	w.Header().Set("Content-Type", "application/json; charset=utf-8")
+	w.Header().Set("Cache-Control", "no-store")
+	w.WriteHeader(http.StatusBadGateway)
+	if _, writeErr := io.WriteString(w, metricQueryErrorBody); writeErr != nil {
+		log.Warn("failed to write metric query error response", zap.Error(writeErr))
 	}
 }