Feature 03 task 5: Trusted proxy handling for forwarded headers (#23)

Co-authored-by: dev_agent <dev_agent@cubecraftlabs.com>
Co-committed-by: dev_agent <dev_agent@cubecraftlabs.com>

internal/forwarded/forwarded.go

@@ -0,0 +1,221 @@
+// Package forwarded resolves where a request really came from. The service
+// runs behind a reverse proxy (Traefik), so the TCP peer is usually the proxy;
+// the X-Forwarded-For and X-Forwarded-Proto headers it adds are believed only
+// when the peer is a trusted proxy. From anyone else they are ignored, so a
+// client cannot spoof its address or scheme.
+package forwarded
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log/slog"
+	"net/http"
+	"net/netip"
+	"strings"
+)
+
+// Header names honoured from trusted proxies.
+const (
+	HeaderFor   = "X-Forwarded-For"
+	HeaderProto = "X-Forwarded-Proto"
+)
+
+// ErrInvalidHeader marks a malformed forwarded header from a trusted peer.
+var ErrInvalidHeader = errors.New("invalid forwarded header")
+
+// Trusted is the set of proxy addresses whose forwarded headers are honoured.
+// The zero value trusts no peer.
+type Trusted []netip.Prefix
+
+// ParseTrusted parses a comma-separated list of IP addresses and CIDR
+// prefixes. A blank list trusts no peer. Each entry must be a plain address
+// or a prefix with no host bits set; zones, IPv4-mapped IPv6 forms and
+// zero-length prefixes, which would trust every peer, are rejected.
+func ParseTrusted(s string) (Trusted, error) {
+	if strings.TrimSpace(s) == "" {
+		return nil, nil
+	}
+	var t Trusted
+	for _, entry := range strings.Split(s, ",") {
+		entry = strings.TrimSpace(entry)
+		if entry == "" {
+			return nil, fmt.Errorf("empty entry in comma-separated list %q", s)
+		}
+		p, err := parseEntry(entry)
+		if err != nil {
+			return nil, err
+		}
+		t = append(t, p)
+	}
+	return t, nil
+}
+
+func parseEntry(entry string) (netip.Prefix, error) {
+	var p netip.Prefix
+	if strings.Contains(entry, "/") {
+		var err error
+		p, err = netip.ParsePrefix(entry)
+		if err != nil {
+			return netip.Prefix{}, fmt.Errorf("invalid CIDR %q", entry)
+		}
+		if p != p.Masked() {
+			return netip.Prefix{}, fmt.Errorf("CIDR %q has host bits set (did you mean %s?)", entry, p.Masked())
+		}
+	} else {
+		a, err := netip.ParseAddr(entry)
+		if err != nil {
+			return netip.Prefix{}, fmt.Errorf("invalid IP address %q", entry)
+		}
+		if a.Zone() != "" {
+			return netip.Prefix{}, fmt.Errorf("IP address %q must not have a zone", entry)
+		}
+		p = netip.PrefixFrom(a, a.BitLen())
+	}
+	switch {
+	case p.Addr().Is4In6():
+		return netip.Prefix{}, fmt.Errorf("%q is IPv4-mapped IPv6; write the IPv4 form", entry)
+	case p.Bits() == 0:
+		return netip.Prefix{}, fmt.Errorf("%q would trust every peer", entry)
+	}
+	return p, nil
+}
+
+// Contains reports whether a is a trusted proxy address. IPv4-mapped IPv6
+// addresses, as a dual-stack listener reports IPv4 peers, match their IPv4
+// form.
+func (t Trusted) Contains(a netip.Addr) bool {
+	a = normalize(a)
+	for _, p := range t {
+		if p.Contains(a) {
+			return true
+		}
+	}
+	return false
+}
+
+func normalize(a netip.Addr) netip.Addr { return a.Unmap().WithZone("") }
+
+// Origin is where a request came from.
+type Origin struct {
+	// ClientIP is the client's address.
+	ClientIP netip.Addr
+	// Scheme is "https" or "http": the scheme the client used.
+	Scheme string
+}
+
+// Resolve works out r's Origin. When the TCP peer (r.RemoteAddr) is not in
+// trusted, both forwarded headers are ignored: the client is the peer and the
+// scheme is that of the connection.
+//
+// When the peer is trusted:
+//   - X-Forwarded-For entries (across all its header lines, in order) are
+//     walked right to left, skipping trusted proxy addresses; the first
+//     untrusted address is the client. Entries left of it are client-supplied
+//     and not examined. If every entry is trusted the leftmost is the client;
+//     with no X-Forwarded-For the peer is the client.
+//   - X-Forwarded-Proto, if present, must be a single "http" or "https" and
+//     is the scheme; without it the connection's scheme is used.
+//
+// A malformed header from a trusted peer is an error wrapping
+// ErrInvalidHeader; any other error means RemoteAddr is not an IP and port.
+func Resolve(r *http.Request, trusted Trusted) (Origin, error) {
+	peer, err := netip.ParseAddrPort(r.RemoteAddr)
+	if err != nil {
+		return Origin{}, fmt.Errorf("peer address %q is not an IP and port", r.RemoteAddr)
+	}
+	o := Origin{ClientIP: normalize(peer.Addr()), Scheme: "http"}
+	if r.TLS != nil {
+		o.Scheme = "https"
+	}
+	if !trusted.Contains(o.ClientIP) {
+		return o, nil
+	}
+	if values := r.Header.Values(HeaderFor); len(values) > 0 {
+		o.ClientIP, err = clientFromChain(strings.Split(strings.Join(values, ","), ","), trusted)
+		if err != nil {
+			return Origin{}, err
+		}
+	}
+	switch values := r.Header.Values(HeaderProto); len(values) {
+	case 0:
+	case 1:
+		switch scheme := strings.ToLower(strings.TrimSpace(values[0])); scheme {
+		case "http", "https":
+			o.Scheme = scheme
+		default:
+			return Origin{}, fmt.Errorf("%w: %s %q is not http or https", ErrInvalidHeader, HeaderProto, values[0])
+		}
+	default:
+		return Origin{}, fmt.Errorf("%w: %d %s headers, want one", ErrInvalidHeader, len(values), HeaderProto)
+	}
+	return o, nil
+}
+
+// clientFromChain returns the rightmost untrusted address in chain, or the
+// leftmost address when all are trusted.
+func clientFromChain(chain []string, trusted Trusted) (netip.Addr, error) {
+	var client netip.Addr
+	for i := len(chain) - 1; i >= 0; i-- {
+		a, err := netip.ParseAddr(strings.TrimSpace(chain[i]))
+		if err != nil {
+			return netip.Addr{}, fmt.Errorf("%w: %s entry %q is not an IP address", ErrInvalidHeader, HeaderFor, strings.TrimSpace(chain[i]))
+		}
+		client = normalize(a)
+		if !trusted.Contains(client) {
+			break
+		}
+	}
+	return client, nil
+}
+
+type contextKey struct{}
+
+// NewContext returns ctx carrying o.
+func NewContext(ctx context.Context, o Origin) context.Context {
+	return context.WithValue(ctx, contextKey{}, o)
+}
+
+// FromContext returns the Origin ctx carries; ok is false when the request
+// was not resolved.
+func FromContext(ctx context.Context) (o Origin, ok bool) {
+	o, ok = ctx.Value(contextKey{}).(Origin)
+	return o, ok
+}
+
+// ClientIP returns the resolved client address ctx carries.
+func ClientIP(ctx context.Context) (netip.Addr, bool) {
+	o, ok := FromContext(ctx)
+	return o.ClientIP, ok
+}
+
+// Scheme returns the resolved request scheme ctx carries.
+func Scheme(ctx context.Context) (string, bool) {
+	o, ok := FromContext(ctx)
+	return o.Scheme, ok
+}
+
+// LogAttr is the log attribute key for the resolved client address.
+const LogAttr = "client_ip"
+
+// LogHandler wraps h so every record logged with a resolved request's
+// context carries the client address as LogAttr.
+func LogHandler(h slog.Handler) slog.Handler { return logHandler{h} }
+
+type logHandler struct{ slog.Handler }
+
+func (h logHandler) Handle(ctx context.Context, rec slog.Record) error {
+	if ip, ok := ClientIP(ctx); ok {
+		rec = rec.Clone()
+		rec.AddAttrs(slog.String(LogAttr, ip.String()))
+	}
+	return h.Handler.Handle(ctx, rec)
+}
+
+func (h logHandler) WithAttrs(attrs []slog.Attr) slog.Handler {
+	return logHandler{h.Handler.WithAttrs(attrs)}
+}
+
+func (h logHandler) WithGroup(name string) slog.Handler {
+	return logHandler{h.Handler.WithGroup(name)}
+}

internal/forwarded/forwarded_test.go

@@ -0,0 +1,260 @@
+package forwarded
+
+import (
+	"bytes"
+	"context"
+	"crypto/tls"
+	"errors"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"net/netip"
+	"slices"
+	"strings"
+	"testing"
+)
+
+func TestParseTrusted(t *testing.T) {
+	t.Parallel()
+	tests := []struct {
+		in      string
+		want    []string
+		wantErr string
+	}{
+		{in: "", want: nil},
+		{in: "   ", want: nil},
+		{in: "10.60.1.33", want: []string{"10.60.1.33/32"}},
+		{in: "10.60.1.33, 172.16.0.0/12", want: []string{"10.60.1.33/32", "172.16.0.0/12"}},
+		{in: "2001:db8::1,2001:db8::/32", want: []string{"2001:db8::1/128", "2001:db8::/32"}},
+		{in: "traefik", wantErr: "invalid IP address"},
+		{in: "10.60.1.33:443", wantErr: "invalid IP address"},
+		{in: "10.60.1.0/33", wantErr: "invalid CIDR"},
+		{in: "10.60.1.33/24", wantErr: "did you mean 10.60.1.0/24"},
+		{in: "10.60.1.33,,10.60.1.34", wantErr: "empty entry"},
+		{in: "10.60.1.33,", wantErr: "empty entry"},
+		{in: "fe80::1%eth0", wantErr: "zone"},
+		{in: "::ffff:10.60.1.33", wantErr: "IPv4-mapped"},
+		{in: "0.0.0.0/0", wantErr: "every peer"},
+		{in: "::/0", wantErr: "every peer"},
+	}
+	for _, tt := range tests {
+		got, err := ParseTrusted(tt.in)
+		if tt.wantErr != "" {
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Errorf("ParseTrusted(%q) error = %v, want one containing %q", tt.in, err, tt.wantErr)
+			}
+			continue
+		}
+		if err != nil {
+			t.Errorf("ParseTrusted(%q) error = %v", tt.in, err)
+			continue
+		}
+		var gotStr []string
+		for _, p := range got {
+			gotStr = append(gotStr, p.String())
+		}
+		if !slices.Equal(gotStr, tt.want) {
+			t.Errorf("ParseTrusted(%q) = %v, want %v", tt.in, gotStr, tt.want)
+		}
+	}
+}
+
+func TestTrustedContainsMappedIPv4(t *testing.T) {
+	t.Parallel()
+	trusted := mustTrusted(t, "10.60.1.33")
+	if !trusted.Contains(netip.MustParseAddr("::ffff:10.60.1.33")) {
+		t.Error("IPv4-mapped form of a trusted address is not trusted")
+	}
+	if Trusted(nil).Contains(netip.MustParseAddr("10.60.1.33")) {
+		t.Error("an empty set trusts a peer")
+	}
+}
+
+func mustTrusted(t *testing.T, s string) Trusted {
+	t.Helper()
+	tr, err := ParseTrusted(s)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return tr
+}
+
+func TestResolve(t *testing.T) {
+	t.Parallel()
+	// 10.60.1.33 is Traefik; 10.60.2.0/24 is an inner proxy tier.
+	trusted := mustTrusted(t, "10.60.1.33,10.60.2.0/24")
+	tests := []struct {
+		name       string
+		peer       string
+		tls        bool
+		xff        []string
+		xfp        []string
+		wantIP     string
+		wantScheme string
+	}{
+		{
+			name: "untrusted peer, no headers",
+			peer: "198.51.100.9:5000", wantIP: "198.51.100.9", wantScheme: "http",
+		},
+		{
+			name: "untrusted peer over TLS",
+			peer: "198.51.100.9:5000", tls: true, wantIP: "198.51.100.9", wantScheme: "https",
+		},
+		{
+			name: "untrusted peer spoofing both headers is ignored",
+			peer: "198.51.100.9:5000", xff: []string{"203.0.113.7"}, xfp: []string{"https"},
+			wantIP: "198.51.100.9", wantScheme: "http",
+		},
+		{
+			name: "untrusted peer's malformed headers are ignored, not rejected",
+			peer: "198.51.100.9:5000", xff: []string{"not-an-ip"}, xfp: []string{"gopher"},
+			wantIP: "198.51.100.9", wantScheme: "http",
+		},
+		{
+			name: "trusted peer: client from X-Forwarded-For, scheme from X-Forwarded-Proto",
+			peer: "10.60.1.33:41000", xff: []string{"203.0.113.7"}, xfp: []string{"https"},
+			wantIP: "203.0.113.7", wantScheme: "https",
+		},
+		{
+			name: "trusted peer seen as IPv4-mapped IPv6",
+			peer: "[::ffff:10.60.1.33]:41000", xff: []string{"203.0.113.7"}, xfp: []string{"https"},
+			wantIP: "203.0.113.7", wantScheme: "https",
+		},
+		{
+			name: "trusted peer without headers is itself the client",
+			peer: "10.60.1.33:41000", wantIP: "10.60.1.33", wantScheme: "http",
+		},
+		{
+			name: "rightmost untrusted address wins over a client-supplied prefix",
+			peer: "10.60.1.33:41000", xff: []string{"1.2.3.4, 203.0.113.7"}, xfp: []string{"https"},
+			wantIP: "203.0.113.7", wantScheme: "https",
+		},
+		{
+			name: "trusted proxies in the chain are skipped",
+			peer: "10.60.1.33:41000", xff: []string{"1.2.3.4, 203.0.113.7, 10.60.2.5"}, xfp: []string{"https"},
+			wantIP: "203.0.113.7", wantScheme: "https",
+		},
+		{
+			name: "several header lines form one chain",
+			peer: "10.60.1.33:41000", xff: []string{"1.2.3.4", "203.0.113.7, 10.60.2.5"},
+			wantIP: "203.0.113.7", wantScheme: "http",
+		},
+		{
+			name: "garbage left of the client is not examined",
+			peer: "10.60.1.33:41000", xff: []string{"spoofed, 203.0.113.7"},
+			wantIP: "203.0.113.7", wantScheme: "http",
+		},
+		{
+			name: "all trusted: leftmost is the client",
+			peer: "10.60.1.33:41000", xff: []string{"10.60.2.5, 10.60.2.6"},
+			wantIP: "10.60.2.5", wantScheme: "http",
+		},
+		{
+			name: "IPv6 client",
+			peer: "10.60.1.33:41000", xff: []string{"2001:db8::7"}, xfp: []string{"HTTPS"},
+			wantIP: "2001:db8::7", wantScheme: "https",
+		},
+		{
+			name: "trusted peer may downgrade to http",
+			peer: "10.60.1.33:41000", tls: true, xfp: []string{"http"},
+			wantIP: "10.60.1.33", wantScheme: "http",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			r := request(tt.peer, tt.xff, tt.xfp)
+			if tt.tls {
+				r.TLS = &tls.ConnectionState{}
+			}
+			o, err := Resolve(r, trusted)
+			if err != nil {
+				t.Fatalf("Resolve error = %v", err)
+			}
+			if o.ClientIP.String() != tt.wantIP || o.Scheme != tt.wantScheme {
+				t.Errorf("Resolve = %s %s, want %s %s", o.ClientIP, o.Scheme, tt.wantIP, tt.wantScheme)
+			}
+		})
+	}
+}
+
+func TestResolveRejectsMalformedHeadersFromTrustedPeer(t *testing.T) {
+	t.Parallel()
+	trusted := mustTrusted(t, "10.60.1.33")
+	for _, tt := range []struct {
+		name     string
+		xff, xfp []string
+	}{
+		{"X-Forwarded-For not an IP", []string{"not-an-ip"}, nil},
+		{"X-Forwarded-For with a port", []string{"203.0.113.7:5000"}, nil},
+		{"X-Forwarded-For empty entry", []string{"203.0.113.7, "}, nil},
+		{"X-Forwarded-For empty", []string{""}, nil},
+		{"X-Forwarded-Proto unknown scheme", nil, []string{"ftp"}},
+		{"X-Forwarded-Proto list", nil, []string{"https, http"}},
+		{"X-Forwarded-Proto repeated", nil, []string{"https", "https"}},
+	} {
+		_, err := Resolve(request("10.60.1.33:41000", tt.xff, tt.xfp), trusted)
+		if !errors.Is(err, ErrInvalidHeader) {
+			t.Errorf("%s: Resolve error = %v, want ErrInvalidHeader", tt.name, err)
+		}
+	}
+}
+
+func TestResolveBadPeerAddress(t *testing.T) {
+	t.Parallel()
+	r := request("@", nil, nil)
+	_, err := Resolve(r, nil)
+	if err == nil || errors.Is(err, ErrInvalidHeader) {
+		t.Errorf("Resolve error = %v, want a peer address error", err)
+	}
+}
+
+func request(peer string, xff, xfp []string) *http.Request {
+	r := httptest.NewRequest(http.MethodGet, "/", nil)
+	r.RemoteAddr = peer
+	for _, v := range xff {
+		r.Header.Add(HeaderFor, v)
+	}
+	for _, v := range xfp {
+		r.Header.Add(HeaderProto, v)
+	}
+	return r
+}
+
+func TestContextHelpers(t *testing.T) {
+	t.Parallel()
+	if _, ok := ClientIP(context.Background()); ok {
+		t.Error("ClientIP ok on an unresolved context")
+	}
+	if _, ok := Scheme(context.Background()); ok {
+		t.Error("Scheme ok on an unresolved context")
+	}
+	ctx := NewContext(context.Background(), Origin{ClientIP: netip.MustParseAddr("203.0.113.7"), Scheme: "https"})
+	if ip, ok := ClientIP(ctx); !ok || ip.String() != "203.0.113.7" {
+		t.Errorf("ClientIP = %s %v, want 203.0.113.7 true", ip, ok)
+	}
+	if s, ok := Scheme(ctx); !ok || s != "https" {
+		t.Errorf("Scheme = %q %v, want https true", s, ok)
+	}
+}
+
+func TestLogHandler(t *testing.T) {
+	t.Parallel()
+	var buf bytes.Buffer
+	logger := slog.New(LogHandler(slog.NewTextHandler(&buf, nil))).With("component", "test")
+	ctx := NewContext(context.Background(), Origin{ClientIP: netip.MustParseAddr("203.0.113.7"), Scheme: "https"})
+
+	logger.InfoContext(ctx, "resolved")
+	logger.InfoContext(context.Background(), "unresolved")
+
+	lines := strings.Split(strings.TrimSpace(buf.String()), "\n")
+	if len(lines) != 2 {
+		t.Fatalf("logged %d lines, want 2: %q", len(lines), buf.String())
+	}
+	if !strings.Contains(lines[0], "client_ip=203.0.113.7") || !strings.Contains(lines[0], "component=test") {
+		t.Errorf("resolved line = %q, want client_ip and the With attribute", lines[0])
+	}
+	if strings.Contains(lines[1], "client_ip") {
+		t.Errorf("unresolved line = %q, want no client_ip", lines[1])
+	}
+}