cmd/derpprobe: check derper TLS certs too

Change-Id: If8c48e012b294570ebbb1a46bacdc58fafbfbcc5
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

cmd/derpprobe/derpprobe.go

@@ -9,7 +9,9 @@ import (
 	"bytes"
 	"context"
 	crand "crypto/rand"
+	"crypto/x509"
 	"encoding/json"
+	"errors"
 	"flag"
 	"fmt"
 	"html"
@@ -33,11 +35,21 @@ var (
 	listen     = flag.String("listen", ":8030", "HTTP listen address")
 )
 
+// certReissueAfter is the time after which we expect all certs to be
+// reissued, at minimum.
+//
+// This is currently set to the date of the LetsEncrypt ALPN revocation event of Jan 2022:
+// https://community.letsencrypt.org/t/questions-about-renewing-before-tls-alpn-01-revocations/170449
+//
+// If there's another revocation event, bump this again.
+var certReissueAfter = time.Unix(1643226768, 0)
+
 var (
 	mu            sync.Mutex
 	state         = map[nodePair]pairStatus{}
 	lastDERPMap   *tailcfg.DERPMap
 	lastDERPMapAt time.Time
+	certs         = map[string]*x509.Certificate{}
 )
 
 func main() {
@@ -46,6 +58,12 @@ func main() {
 	log.Fatal(http.ListenAndServe(*listen, http.HandlerFunc(serve)))
 }
 
+func setCert(name string, cert *x509.Certificate) {
+	mu.Lock()
+	defer mu.Unlock()
+	certs[name] = cert
+}
+
 type overallStatus struct {
 	good, bad []string
 }
@@ -93,6 +111,27 @@ func getOverallStatus() (o overallStatus) {
 			}
 		}
 	}
+
+	var subjs []string
+	for k := range certs {
+		subjs = append(subjs, k)
+	}
+	sort.Strings(subjs)
+
+	soon := time.Now().Add(14 * 24 * time.Hour) // in 2 weeks; autocert does 30 days by default
+	for _, s := range subjs {
+		cert := certs[s]
+		if cert.NotBefore.Before(certReissueAfter) {
+			o.addBadf("cert %q needs reissuing; NotBefore=%v", s, cert.NotBefore.Format(time.RFC3339))
+			continue
+		}
+		if cert.NotAfter.Before(soon) {
+			o.addBadf("cert %q expiring soon (%v); wasn't auto-refreshed", s, cert.NotAfter.Format(time.RFC3339))
+			continue
+		}
+		o.addGoodf("cert %q good %v - %v", s, cert.NotBefore.Format(time.RFC3339), cert.NotAfter.Format(time.RFC3339))
+	}
+
 	return
 }
 
@@ -359,6 +398,21 @@ func newConn(ctx context.Context, dm *tailcfg.DERPMap, n *tailcfg.DERPNode) (*de
 	if err != nil {
 		return nil, err
 	}
+	cs, ok := dc.TLSConnectionState()
+	if !ok {
+		dc.Close()
+		return nil, errors.New("no TLS state")
+	}
+	if len(cs.PeerCertificates) == 0 {
+		dc.Close()
+		return nil, errors.New("no peer certificates")
+	}
+	if cs.ServerName != n.HostName {
+		dc.Close()
+		return nil, fmt.Errorf("TLS server name %q != derp hostname %q", cs.ServerName, n.HostName)
+	}
+	setCert(cs.ServerName, cs.PeerCertificates[0])
+
 	errc := make(chan error, 1)
 	go func() {
 		m, err := dc.Recv()

derp/derphttp/derphttp_client.go

@@ -72,6 +72,7 @@ type Client struct {
 	client       *derp.Client
 	connGen      int // incremented once per new connection; valid values are >0
 	serverPubKey key.NodePublic
+	tlsState     *tls.ConnectionState
 	pingOut      map[derp.PingMessage]chan<- bool // chan to send to on pong
 }
 
@@ -124,6 +125,17 @@ func (c *Client) Connect(ctx context.Context) error {
 	return err
 }
 
+// TLSConnectionState returns the last TLS connection state, if any.
+// The client must already be connected.
+func (c *Client) TLSConnectionState() (_ *tls.ConnectionState, ok bool) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.closed || c.client == nil {
+		return nil, false
+	}
+	return c.tlsState, c.tlsState != nil
+}
+
 // ServerPublicKey returns the server's public key.
 //
 // It only returns a non-zero value once a connection has succeeded
@@ -318,6 +330,7 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 	var httpConn net.Conn        // a TCP conn or a TLS conn; what we speak HTTP to
 	var serverPub key.NodePublic // or zero if unknown (if not using TLS or TLS middlebox eats it)
 	var serverProtoVersion int
+	var tlsState *tls.ConnectionState
 	if c.useHTTPS() {
 		tlsConn := c.tlsClient(tcpConn, node)
 		httpConn = tlsConn
@@ -340,9 +353,10 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 		// Note that we're not specifically concerned about TLS downgrade
 		// attacks. TLS handles that fine:
 		// https://blog.gypsyengineer.com/en/security/how-does-tls-1-3-protect-against-downgrade-attacks.html
-		connState := tlsConn.ConnectionState()
+		cs := tlsConn.ConnectionState()
-		if connState.Version >= tls.VersionTLS13 {
+		tlsState = &cs
-			serverPub, serverProtoVersion = parseMetaCert(connState.PeerCertificates)
+		if cs.Version >= tls.VersionTLS13 {
+			serverPub, serverProtoVersion = parseMetaCert(cs.PeerCertificates)
 		}
 	} else {
 		httpConn = tcpConn
@@ -409,6 +423,7 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 	c.serverPubKey = derpClient.ServerPublicKey()
 	c.client = derpClient
 	c.netConn = tcpConn
+	c.tlsState = tlsState
 	c.connGen++
 	return c.client, c.connGen, nil
 }