moby--moby/libnetwork/resolver.go

package libnetwork

import (
	"fmt"
	"math/rand"
	"net"
	"strings"
	"sync"
	"time"

	log "github.com/Sirupsen/logrus"
	"github.com/docker/libnetwork/types"
	"github.com/miekg/dns"
)

// Resolver represents the embedded DNS server in Docker. It operates
// by listening on container's loopback interface for DNS queries.
type Resolver interface {
	// Start starts the name server for the container
	Start() error
	// Stop stops the name server for the container. Stopped resolver
	// can be reused after running the SetupFunc again.
	Stop()
	// SetupFunc() provides the setup function that should be run
	// in the container's network namespace.
	SetupFunc() func()
	// NameServer() returns the IP of the DNS resolver for the
	// containers.
	NameServer() string
	// SetExtServers configures the external nameservers the resolver
	// should use to forward queries
	SetExtServers([]string)
	// FlushExtServers clears the cached UDP connections to external
	// nameservers
	FlushExtServers()
	// ResolverOptions returns resolv.conf options that should be set
	ResolverOptions() []string
}

const (
	resolverIP      = "127.0.0.11"
	dnsPort         = "53"
	ptrIPv4domain   = ".in-addr.arpa."
	ptrIPv6domain   = ".ip6.arpa."
	respTTL         = 600
	maxExtDNS       = 3 //max number of external servers to try
	extIOTimeout    = 4 * time.Second
	defaultRespSize = 512
	maxConcurrent   = 100
	logInterval     = 2 * time.Second
	maxDNSID        = 65536
)

type clientConn struct {
	dnsID      uint16
	respWriter dns.ResponseWriter
}

type extDNSEntry struct {
	ipStr   string
	extConn net.Conn
	extOnce sync.Once
}

type sboxQuery struct {
	sboxID string
	dnsID  uint16
}

type clientConnGC struct {
	toDelete bool
	client   clientConn
}

var (
	queryGCMutex sync.Mutex
	queryGC      map[sboxQuery]*clientConnGC
)

// resolver implements the Resolver interface
type resolver struct {
	sb         *sandbox
	extDNSList [maxExtDNS]extDNSEntry
	server     *dns.Server
	conn       *net.UDPConn
	tcpServer  *dns.Server
	tcpListen  *net.TCPListener
	err        error
	count      int32
	tStamp     time.Time
	queryLock  sync.Mutex
	client     map[uint16]clientConn
}

func init() {
	rand.Seed(time.Now().Unix())
	queryGC = make(map[sboxQuery]*clientConnGC)
	go func() {
		ticker := time.NewTicker(1 * time.Minute)
		for range ticker.C {
			queryGCMutex.Lock()
			for query, conn := range queryGC {
				if !conn.toDelete {
					conn.toDelete = true
					continue
				}
				delete(queryGC, query)
			}
			queryGCMutex.Unlock()
		}
	}()
}

// NewResolver creates a new instance of the Resolver
func NewResolver(sb *sandbox) Resolver {
	return &resolver{
		sb:     sb,
		err:    fmt.Errorf("setup not done yet"),
		client: make(map[uint16]clientConn),
	}
}

func (r *resolver) SetupFunc() func() {
	return (func() {
		var err error

		// DNS operates primarily on UDP
		addr := &net.UDPAddr{
			IP: net.ParseIP(resolverIP),
		}

		r.conn, err = net.ListenUDP("udp", addr)
		if err != nil {
			r.err = fmt.Errorf("error in opening name server socket %v", err)
			return
		}

		// Listen on a TCP as well
		tcpaddr := &net.TCPAddr{
			IP: net.ParseIP(resolverIP),
		}

		r.tcpListen, err = net.ListenTCP("tcp", tcpaddr)
		if err != nil {
			r.err = fmt.Errorf("error in opening name TCP server socket %v", err)
			return
		}
		r.err = nil
	})
}

func (r *resolver) Start() error {
	// make sure the resolver has been setup before starting
	if r.err != nil {
		return r.err
	}

	if err := r.setupIPTable(); err != nil {
		return fmt.Errorf("setting up IP table rules failed: %v", err)
	}

	s := &dns.Server{Handler: r, PacketConn: r.conn}
	r.server = s
	go func() {
		s.ActivateAndServe()
	}()

	tcpServer := &dns.Server{Handler: r, Listener: r.tcpListen}
	r.tcpServer = tcpServer
	go func() {
		tcpServer.ActivateAndServe()
	}()
	return nil
}

func (r *resolver) FlushExtServers() {
	for i := 0; i < maxExtDNS; i++ {
		if r.extDNSList[i].extConn != nil {
			r.extDNSList[i].extConn.Close()
		}

		r.extDNSList[i].extConn = nil
		r.extDNSList[i].extOnce = sync.Once{}
	}
}

func (r *resolver) Stop() {
	r.FlushExtServers()

	if r.server != nil {
		r.server.Shutdown()
	}
	if r.tcpServer != nil {
		r.tcpServer.Shutdown()
	}
	r.conn = nil
	r.tcpServer = nil
	r.err = fmt.Errorf("setup not done yet")
	r.tStamp = time.Time{}
	r.count = 0
	r.queryLock = sync.Mutex{}
}

func (r *resolver) SetExtServers(dns []string) {
	l := len(dns)
	if l > maxExtDNS {
		l = maxExtDNS
	}
	for i := 0; i < l; i++ {
		r.extDNSList[i].ipStr = dns[i]
	}
}

func (r *resolver) NameServer() string {
	return resolverIP
}

func (r *resolver) ResolverOptions() []string {
	return []string{"ndots:0"}
}

func setCommonFlags(msg *dns.Msg) {
	msg.RecursionAvailable = true
}

func shuffleAddr(addr []net.IP) []net.IP {
	for i := len(addr) - 1; i > 0; i-- {
		r := rand.Intn(i + 1)
		addr[i], addr[r] = addr[r], addr[i]
	}
	return addr
}

func createRespMsg(query *dns.Msg) *dns.Msg {
	resp := new(dns.Msg)
	resp.SetReply(query)
	setCommonFlags(resp)

	return resp
}

func (r *resolver) handleIPQuery(name string, query *dns.Msg, ipType int) (*dns.Msg, error) {
	addr, ipv6Miss := r.sb.ResolveName(name, ipType)
	if addr == nil && ipv6Miss {
		// Send a reply without any Answer sections
		log.Debugf("Lookup name %s present without IPv6 address", name)
		resp := createRespMsg(query)
		return resp, nil
	}
	if addr == nil {
		return nil, nil
	}

	log.Debugf("Lookup for %s: IP %v", name, addr)

	resp := createRespMsg(query)
	if len(addr) > 1 {
		addr = shuffleAddr(addr)
	}
	if ipType == types.IPv4 {
		for _, ip := range addr {
			rr := new(dns.A)
			rr.Hdr = dns.RR_Header{Name: name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: respTTL}
			rr.A = ip
			resp.Answer = append(resp.Answer, rr)
		}
	} else {
		for _, ip := range addr {
			rr := new(dns.AAAA)
			rr.Hdr = dns.RR_Header{Name: name, Rrtype: dns.TypeAAAA, Class: dns.ClassINET, Ttl: respTTL}
			rr.AAAA = ip
			resp.Answer = append(resp.Answer, rr)
		}
	}
	return resp, nil
}

func (r *resolver) handlePTRQuery(ptr string, query *dns.Msg) (*dns.Msg, error) {
	parts := []string{}

	if strings.HasSuffix(ptr, ptrIPv4domain) {
		parts = strings.Split(ptr, ptrIPv4domain)
	} else if strings.HasSuffix(ptr, ptrIPv6domain) {
		parts = strings.Split(ptr, ptrIPv6domain)
	} else {
		return nil, fmt.Errorf("invalid PTR query, %v", ptr)
	}

	host := r.sb.ResolveIP(parts[0])
	if len(host) == 0 {
		return nil, nil
	}

	log.Debugf("Lookup for IP %s: name %s", parts[0], host)
	fqdn := dns.Fqdn(host)

	resp := new(dns.Msg)
	resp.SetReply(query)
	setCommonFlags(resp)

	rr := new(dns.PTR)
	rr.Hdr = dns.RR_Header{Name: ptr, Rrtype: dns.TypePTR, Class: dns.ClassINET, Ttl: respTTL}
	rr.Ptr = fqdn
	resp.Answer = append(resp.Answer, rr)
	return resp, nil
}

func (r *resolver) handleSRVQuery(svc string, query *dns.Msg) (*dns.Msg, error) {
	srv, ip, err := r.sb.ResolveService(svc)

	if err != nil {
		return nil, err
	}
	if len(srv) != len(ip) {
		return nil, fmt.Errorf("invalid reply for SRV query %s", svc)
	}

	resp := createRespMsg(query)

	for i, r := range srv {
		rr := new(dns.SRV)
		rr.Hdr = dns.RR_Header{Name: svc, Rrtype: dns.TypePTR, Class: dns.ClassINET, Ttl: respTTL}
		rr.Port = r.Port
		rr.Target = r.Target
		resp.Answer = append(resp.Answer, rr)

		rr1 := new(dns.A)
		rr1.Hdr = dns.RR_Header{Name: r.Target, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: respTTL}
		rr1.A = ip[i]
		resp.Extra = append(resp.Extra, rr1)
	}
	return resp, nil

}

func truncateResp(resp *dns.Msg, maxSize int, isTCP bool) {
	if !isTCP {
		resp.Truncated = true
	}

	srv := resp.Question[0].Qtype == dns.TypeSRV
	// trim the Answer RRs one by one till the whole message fits
	// within the reply size
	for resp.Len() > maxSize {
		resp.Answer = resp.Answer[:len(resp.Answer)-1]

		if srv && len(resp.Extra) > 0 {
			resp.Extra = resp.Extra[:len(resp.Extra)-1]
		}
	}
}

func (r *resolver) ServeDNS(w dns.ResponseWriter, query *dns.Msg) {
	var (
		extConn net.Conn
		resp    *dns.Msg
		err     error
		writer  dns.ResponseWriter
	)

	if query == nil || len(query.Question) == 0 {
		return
	}
	name := query.Question[0].Name

	switch query.Question[0].Qtype {
	case dns.TypeA:
		resp, err = r.handleIPQuery(name, query, types.IPv4)
	case dns.TypeAAAA:
		resp, err = r.handleIPQuery(name, query, types.IPv6)
	case dns.TypePTR:
		resp, err = r.handlePTRQuery(name, query)
	case dns.TypeSRV:
		resp, err = r.handleSRVQuery(name, query)
	}

	if err != nil {
		log.Error(err)
		return
	}

	proto := w.LocalAddr().Network()
	maxSize := 0
	if proto == "tcp" {
		maxSize = dns.MaxMsgSize - 1
	} else if proto == "udp" {
		optRR := query.IsEdns0()
		if optRR != nil {
			maxSize = int(optRR.UDPSize())
		}
		if maxSize < defaultRespSize {
			maxSize = defaultRespSize
		}
	}

	if resp != nil {
		if resp.Len() > maxSize {
			truncateResp(resp, maxSize, proto == "tcp")
		}
		writer = w
	} else {
		queryID := query.Id
	extQueryLoop:
		for i := 0; i < maxExtDNS; i++ {
			extDNS := &r.extDNSList[i]
			if extDNS.ipStr == "" {
				break
			}
			extConnect := func() {
				addr := fmt.Sprintf("%s:%d", extDNS.ipStr, 53)
				extConn, err = net.DialTimeout(proto, addr, extIOTimeout)
			}

			// For udp clients connection is persisted to reuse for further queries.
			// Accessing extDNS.extConn be a race here between go rouines. Hence the
			// connection setup is done in a Once block and fetch the extConn again
			extConn = extDNS.extConn
			if extConn == nil || proto == "tcp" {
				if proto == "udp" {
					extDNS.extOnce.Do(func() {
						r.sb.execFunc(extConnect)
						extDNS.extConn = extConn
					})
					extConn = extDNS.extConn
				} else {
					r.sb.execFunc(extConnect)
				}
				if err != nil {
					log.Debugf("Connect failed, %s", err)
					continue
				}
			}
			// If two go routines are executing in parralel one will
			// block on the Once.Do and in case of error connecting
			// to the external server it will end up with a nil err
			// but extConn also being nil.
			if extConn == nil {
				continue
			}
			log.Debugf("Query %s[%d] from %s, forwarding to %s:%s", name, query.Question[0].Qtype,
				extConn.LocalAddr().String(), proto, extDNS.ipStr)

			// Timeout has to be set for every IO operation.
			extConn.SetDeadline(time.Now().Add(extIOTimeout))
			co := &dns.Conn{Conn: extConn}

			// forwardQueryStart stores required context to mux multiple client queries over
			// one connection; and limits the number of outstanding concurrent queries.
			if r.forwardQueryStart(w, query, queryID) == false {
				old := r.tStamp
				r.tStamp = time.Now()
				if r.tStamp.Sub(old) > logInterval {
					log.Errorf("More than %v concurrent queries from %s", maxConcurrent, extConn.LocalAddr().String())
				}
				continue
			}

			defer func() {
				if proto == "tcp" {
					co.Close()
				}
			}()
			err = co.WriteMsg(query)
			if err != nil {
				r.forwardQueryEnd(w, query)
				log.Debugf("Send to DNS server failed, %s", err)
				continue
			}
			for {
				// If a reply comes after a read timeout it will remain in the socket buffer
				// and will be read after sending next query. To ignore such stale replies
				// save the query context in a GC queue when read timesout. On the next reply
				// if the context is present in the GC queue its a old reply. Ignore it and
				// read again
				resp, err = co.ReadMsg()
				if err != nil {
					if nerr, ok := err.(net.Error); ok && nerr.Timeout() {
						r.addQueryToGC(w, query)
					}
					r.forwardQueryEnd(w, query)
					log.Debugf("Read from DNS server failed, %s", err)
					continue extQueryLoop
				}

				if !r.checkRespInGC(w, resp) {
					break
				}
			}
			// Retrieves the context for the forwarded query and returns the client connection
			// to send the reply to
			writer = r.forwardQueryEnd(w, resp)
			if writer == nil {
				continue
			}

			resp.Compress = true
			break
		}
		if resp == nil || writer == nil {
			return
		}
	}

	if writer == nil {
		return
	}
	if err = writer.WriteMsg(resp); err != nil {
		log.Errorf("error writing resolver resp, %s", err)
	}
}

func (r *resolver) forwardQueryStart(w dns.ResponseWriter, msg *dns.Msg, queryID uint16) bool {
	proto := w.LocalAddr().Network()
	dnsID := uint16(rand.Intn(maxDNSID))

	cc := clientConn{
		dnsID:      queryID,
		respWriter: w,
	}

	r.queryLock.Lock()
	defer r.queryLock.Unlock()

	if r.count == maxConcurrent {
		return false
	}
	r.count++

	switch proto {
	case "tcp":
		break
	case "udp":
		for ok := true; ok == true; dnsID = uint16(rand.Intn(maxDNSID)) {
			_, ok = r.client[dnsID]
		}
		log.Debugf("client dns id %v, changed id %v", queryID, dnsID)
		r.client[dnsID] = cc
		msg.Id = dnsID
	default:
		log.Errorf("Invalid protocol..")
		return false
	}

	return true
}

func (r *resolver) addQueryToGC(w dns.ResponseWriter, msg *dns.Msg) {
	if w.LocalAddr().Network() != "udp" {
		return
	}

	r.queryLock.Lock()
	cc, ok := r.client[msg.Id]
	r.queryLock.Unlock()
	if !ok {
		return
	}

	query := sboxQuery{
		sboxID: r.sb.ID(),
		dnsID:  msg.Id,
	}
	clientGC := &clientConnGC{
		client: cc,
	}
	queryGCMutex.Lock()
	queryGC[query] = clientGC
	queryGCMutex.Unlock()
}

func (r *resolver) checkRespInGC(w dns.ResponseWriter, msg *dns.Msg) bool {
	if w.LocalAddr().Network() != "udp" {
		return false
	}

	query := sboxQuery{
		sboxID: r.sb.ID(),
		dnsID:  msg.Id,
	}

	queryGCMutex.Lock()
	defer queryGCMutex.Unlock()
	if _, ok := queryGC[query]; ok {
		delete(queryGC, query)
		return true
	}
	return false
}

func (r *resolver) forwardQueryEnd(w dns.ResponseWriter, msg *dns.Msg) dns.ResponseWriter {
	var (
		cc clientConn
		ok bool
	)
	proto := w.LocalAddr().Network()

	r.queryLock.Lock()
	defer r.queryLock.Unlock()

	if r.count == 0 {
		log.Errorf("Invalid concurrent query count")
	} else {
		r.count--
	}

	switch proto {
	case "tcp":
		break
	case "udp":
		if cc, ok = r.client[msg.Id]; ok == false {
			log.Debugf("Can't retrieve client context for dns id %v", msg.Id)
			return nil
		}
		log.Debugf("dns msg id %v, client id %v", msg.Id, cc.dnsID)
		delete(r.client, msg.Id)
		msg.Id = cc.dnsID
		w = cc.respWriter
	default:
		log.Errorf("Invalid protocol")
		return nil
	}
	return w
}