jarl/internal/encdns/health.go

package encdns

import (
	"bytes"
	"context"
	"crypto/tls"
	"crypto/x509"
	"errors"
	"fmt"
	"io"
	"net"
	"net/http"
	"strconv"
	"strings"
	"syscall"
	"time"

	"github.com/miekg/dns"
)

// Probe is the outcome of checking one DoT or DoH listener the way a
// client would: connect, check the certificate, ask a question.
type Probe struct {
	Kind    string `json:"kind"` // dot or doh
	Address string `json:"address"`
	Port    int    `json:"port"`
	// OK: it answered over TLS with a certificate for the host name that
	// is current. Problem says what is wrong when not; Warning notes
	// something that doesn't stop clients from working everywhere.
	OK      bool   `json:"ok"`
	Problem string `json:"problem,omitempty"`
	Warning string `json:"warning,omitempty"`
	// What the certificate named served says.
	CertNames    []string   `json:"cert_names,omitempty"`
	CertNotAfter *time.Time `json:"cert_not_after,omitempty"`
	CertSerial   string     `json:"cert_serial,omitempty"`
	// Trusted: it chains to a CA this server trusts (not so for a
	// staging or test CA).
	Trusted bool   `json:"trusted"`
	RCode   string `json:"rcode,omitempty"`
	Millis  int64  `json:"ms"`
}

// Health is the last check of every listener.
type Health struct {
	Checked time.Time `json:"checked"`
	Probes  []Probe   `json:"probes"`
}

// OK reports whether every listener works.
func (h *Health) OK() bool {
	for _, p := range h.Probes {
		if !p.OK {
			return false
		}
	}
	return true
}

// Check probes every DoT and DoH listener of the setup. hostname is what
// clients ask for, serial the certificate Jarl last wrote for named ("" if
// unknown), roots the CAs to trust (nil: the system's).
func Check(ctx context.Context, s Status, hostname, serial string, roots *x509.CertPool) *Health {
	h := &Health{Checked: time.Now(), Probes: []Probe{}}
	for _, e := range s.DoT {
		h.Probes = append(h.Probes, probe(ctx, "dot", e, hostname, serial, roots))
	}
	for _, e := range s.DoH {
		h.Probes = append(h.Probes, probe(ctx, "doh", e, hostname, serial, roots))
	}
	return h
}

const probeTimeout = 5 * time.Second

func probe(ctx context.Context, kind string, e Endpoint, hostname, serial string, roots *x509.CertPool) Probe {
	p := Probe{Kind: kind, Address: e.Address, Port: e.Port}
	addr := net.JoinHostPort(e.Address, strconv.Itoa(e.Port))
	ctx, cancel := context.WithTimeout(ctx, probeTimeout)
	defer cancel()
	start := time.Now()
	q := new(dns.Msg)
	q.SetQuestion(dns.Fqdn(hostname), dns.TypeA)
	q.RecursionDesired = true
	var state tls.ConnectionState
	var answer *dns.Msg
	var err error
	cfg := &tls.Config{ServerName: hostname, InsecureSkipVerify: true, MinVersion: tls.VersionTLS12} // checked below, to say what is wrong
	if kind == "dot" {
		cfg.NextProtos = []string{"dot"}
		state, answer, err = askDoT(ctx, addr, cfg, q)
	} else {
		state, answer, err = askDoH(ctx, addr, hostname, e.Port, cfg, q)
	}
	p.Millis = time.Since(start).Milliseconds()
	if len(state.PeerCertificates) > 0 {
		leaf := state.PeerCertificates[0]
		p.CertNames, p.CertSerial = leaf.DNSNames, fmt.Sprintf("%x", leaf.SerialNumber)
		na := leaf.NotAfter
		p.CertNotAfter = &na
		inter := x509.NewCertPool()
		for _, c := range state.PeerCertificates[1:] {
			inter.AddCert(c)
		}
		_, verr := leaf.Verify(x509.VerifyOptions{Roots: roots, Intermediates: inter, DNSName: hostname})
		p.Trusted = verr == nil
		switch {
		case leaf.VerifyHostname(hostname) != nil:
			p.Problem = fmt.Sprintf("named serves a certificate for %s, not %s: clients reject it", strings.Join(leaf.DNSNames, ", "), hostname)
		case time.Now().After(leaf.NotAfter):
			p.Problem = "named serves an expired certificate (until " + leaf.NotAfter.Format(time.DateOnly) + ")"
		case serial != "" && p.CertSerial != serial:
			p.Problem = "named still serves an older certificate (serial " + p.CertSerial + ", Jarl has " + serial + "): it hasn't reread the files (rndc reconfig)"
		case !p.Trusted:
			p.Warning = "the certificate isn't from a CA this server trusts (a staging or test CA?): clients reject it unless they trust that CA"
		}
	}
	switch {
	case err != nil:
		p.Problem = describe(err, e)
	case p.Problem != "":
	case answer == nil:
		p.Problem = "no DNS answer"
	default:
		p.RCode = dns.RcodeToString[answer.Rcode]
		p.OK = true
		if answer.Rcode == dns.RcodeRefused {
			p.Warning = strings.TrimPrefix(p.Warning+"; ", "; ") + "it answers, but REFUSED: this server's own address gets a view that doesn't answer for " + hostname
		}
	}
	return p
}

func askDoT(ctx context.Context, addr string, cfg *tls.Config, q *dns.Msg) (tls.ConnectionState, *dns.Msg, error) {
	d := &tls.Dialer{NetDialer: &net.Dialer{Timeout: probeTimeout}, Config: cfg}
	conn, err := d.DialContext(ctx, "tcp", addr)
	if err != nil {
		return tls.ConnectionState{}, nil, err
	}
	defer conn.Close()
	tc := conn.(*tls.Conn)
	state := tc.ConnectionState()
	conn.SetDeadline(time.Now().Add(probeTimeout))
	co := &dns.Conn{Conn: conn}
	if err := co.WriteMsg(q); err != nil {
		return state, nil, err
	}
	r, err := co.ReadMsg()
	return state, r, err
}

func askDoH(ctx context.Context, addr, hostname string, port int, cfg *tls.Config, q *dns.Msg) (tls.ConnectionState, *dns.Msg, error) {
	cfg = cfg.Clone()
	cfg.NextProtos = []string{"h2", "http/1.1"}
	tr := &http.Transport{
		TLSClientConfig:   cfg,
		ForceAttemptHTTP2: true,
		// Whatever the host name resolves to here: the listener's address.
		DialContext: func(ctx context.Context, network, _ string) (net.Conn, error) {
			return (&net.Dialer{Timeout: probeTimeout}).DialContext(ctx, network, addr)
		},
	}
	defer tr.CloseIdleConnections()
	wire, err := q.Pack()
	if err != nil {
		return tls.ConnectionState{}, nil, err
	}
	url := "https://" + net.JoinHostPort(hostname, strconv.Itoa(port)) + "/dns-query"
	req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(wire))
	if err != nil {
		return tls.ConnectionState{}, nil, err
	}
	req.Header.Set("Content-Type", "application/dns-message")
	req.Header.Set("Accept", "application/dns-message")
	resp, err := (&http.Client{Transport: tr, Timeout: probeTimeout}).Do(req)
	if err != nil {
		return tls.ConnectionState{}, nil, err
	}
	defer resp.Body.Close()
	var state tls.ConnectionState
	if resp.TLS != nil {
		state = *resp.TLS
	}
	body, err := io.ReadAll(io.LimitReader(resp.Body, 64<<10))
	if err != nil {
		return state, nil, err
	}
	if resp.StatusCode != http.StatusOK {
		return state, nil, fmt.Errorf("HTTP %d at /dns-query", resp.StatusCode)
	}
	r := new(dns.Msg)
	if err := r.Unpack(body); err != nil {
		return state, nil, fmt.Errorf("not a DNS answer: %w", err)
	}
	return state, r, nil
}

// describe says what a failed probe means.
func describe(err error, e Endpoint) string {
	var ne net.Error
	switch {
	case errors.Is(err, syscall.ECONNREFUSED):
		return fmt.Sprintf("nothing listens on %s port %d: named may not have been able to bind it (look in named's log; ports below 1024 need named to start as root)", e.Address, e.Port)
	case errors.As(err, &ne) && ne.Timeout(), errors.Is(err, context.DeadlineExceeded):
		return fmt.Sprintf("no answer from %s port %d within %s (a firewall on the server?)", e.Address, e.Port, probeTimeout)
	}
	return err.Error()
}