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) 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
}
func (r *resolver) ServeDNS(w dns.ResponseWriter, query *dns.Msg) {
var (
resp *dns.Msg
err error
)
if query == nil || len(query.Question) == 0 {
return
}
name := query.Question[0].Name
if query.Question[0].Qtype == dns.TypeA {
resp, err = r.handleIPv4Query(name, query)
} else if query.Question[0].Qtype == dns.TypePTR {
resp, err = r.handlePTRQuery(name, query)
}
if err != nil {
log.Error(err)
return
}
if resp == nil {
if len(r.extDNS) == 0 {
return
}
num := maxExtDNS
if len(r.extDNS) < maxExtDNS {
num = len(r.extDNS)
}
for i := 0; i < num; i++ {
log.Debugf("Querying ext dns %s:%s for %s[%d]", w.LocalAddr().Network(), r.extDNS[i], name, query.Question[0].Qtype)
c := &dns.Client{Net: w.LocalAddr().Network()}
addr := fmt.Sprintf("%s:%d", r.extDNS[i], 53)
resp, _, err = c.Exchange(query, addr)
if err == nil {
resp.Compress = true
break
}
log.Errorf("external resolution failed, %s", err)
}
if resp == nil {
return
}
}
err = w.WriteMsg(resp)
if err != nil {
log.Errorf("error writing resolver resp, %s", err)
}
}
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) 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) ServeDNS(w dns.ResponseWriter, query *dns.Msg) {
var (
extConn net.Conn
resp *dns.Msg
err error
)
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 {
logrus.Error(err)
return
}
if resp == nil {
// If the backend doesn't support proxying dns request
// fail the response
if !r.proxyDNS {
resp = new(dns.Msg)
resp.SetRcode(query, dns.RcodeServerFailure)
w.WriteMsg(resp)
return
}
// If the user sets ndots > 0 explicitly and the query is
// in the root domain don't forward it out. We will return
// failure and let the client retry with the search domain
// attached
switch query.Question[0].Qtype {
case dns.TypeA:
fallthrough
case dns.TypeAAAA:
if r.backend.NdotsSet() && !strings.Contains(strings.TrimSuffix(name, "."), ".") {
resp = createRespMsg(query)
}
}
}
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")
}
} else {
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)
}
if extDNS.hostLoopback {
extConnect()
} else {
execErr := r.backend.ExecFunc(extConnect)
if execErr != nil {
logrus.Warn(execErr)
continue
}
}
if err != nil {
logrus.Warnf("Connect failed: %s", err)
continue
}
logrus.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,
UDPSize: uint16(maxSize),
}
defer co.Close()
// limits the number of outstanding concurrent queries.
if r.forwardQueryStart() == false {
old := r.tStamp
r.tStamp = time.Now()
if r.tStamp.Sub(old) > logInterval {
logrus.Errorf("More than %v concurrent queries from %s", maxConcurrent, extConn.LocalAddr().String())
}
continue
}
err = co.WriteMsg(query)
if err != nil {
r.forwardQueryEnd()
logrus.Debugf("Send to DNS server failed, %s", err)
continue
}
resp, err = co.ReadMsg()
// Truncated DNS replies should be sent to the client so that the
// client can retry over TCP
if err != nil && err != dns.ErrTruncated {
r.forwardQueryEnd()
logrus.Debugf("Read from DNS server failed, %s", err)
continue
}
r.forwardQueryEnd()
if resp != nil {
for _, rr := range resp.Answer {
h := rr.Header()
switch h.Rrtype {
case dns.TypeA:
ip := rr.(*dns.A).A
r.backend.HandleQueryResp(h.Name, ip)
case dns.TypeAAAA:
ip := rr.(*dns.AAAA).AAAA
r.backend.HandleQueryResp(h.Name, ip)
}
}
}
resp.Compress = true
break
}
if resp == nil {
return
}
}
if err = w.WriteMsg(resp); err != nil {
logrus.Errorf("error writing resolver resp, %s", err)
}
}
func (r *resolver) ServeDNS(w dns.ResponseWriter, query *dns.Msg) {
var (
extConn net.Conn
resp *dns.Msg
err error
)
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")
}
} else {
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)
}
r.sb.execFunc(extConnect)
if err != nil {
log.Debugf("Connect failed, %s", err)
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}
defer co.Close()
// limits the number of outstanding concurrent queries.
if r.forwardQueryStart() == 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
}
err = co.WriteMsg(query)
if err != nil {
r.forwardQueryEnd()
log.Debugf("Send to DNS server failed, %s", err)
continue
}
resp, err = co.ReadMsg()
// Truncated DNS replies should be sent to the client so that the
// client can retry over TCP
if err != nil && err != dns.ErrTruncated {
r.forwardQueryEnd()
log.Debugf("Read from DNS server failed, %s", err)
continue
}
r.forwardQueryEnd()
resp.Compress = true
break
}
if resp == nil {
return
}
}
if err = w.WriteMsg(resp); err != nil {
log.Errorf("error writing resolver resp, %s", err)
}
}
func (r *resolver) ServeDNS(w dns.ResponseWriter, query *dns.Msg) {
var (
extConn net.Conn
resp *dns.Msg
err error
)
if query == nil || len(query.Question) == 0 {
return
}
name := query.Question[0].Name
if query.Question[0].Qtype == dns.TypeA {
resp, err = r.handleIPv4Query(name, query)
} else if query.Question[0].Qtype == dns.TypePTR {
resp, err = r.handlePTRQuery(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")
}
} else {
for i := 0; i < maxExtDNS; i++ {
extDNS := &r.extDNSList[i]
if extDNS.ipStr == "" {
break
}
log.Debugf("Querying ext dns %s:%s for %s[%d]", proto, extDNS.ipStr, name, query.Question[0].Qtype)
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
}
// Timeout has to be set for every IO operation.
extConn.SetDeadline(time.Now().Add(extIOTimeout))
co := &dns.Conn{Conn: extConn}
defer func() {
if proto == "tcp" {
co.Close()
}
}()
err = co.WriteMsg(query)
if err != nil {
log.Debugf("Send to DNS server failed, %s", err)
continue
}
resp, err = co.ReadMsg()
if err != nil {
log.Debugf("Read from DNS server failed, %s", err)
continue
}
resp.Compress = true
break
}
if resp == nil {
return
}
}
err = w.WriteMsg(resp)
if err != nil {
log.Errorf("error writing resolver resp, %s", err)
}
}
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
if query.Question[0].Qtype == dns.TypeA {
resp, err = r.handleIPQuery(name, query, netutils.IPv4)
} else if query.Question[0].Qtype == dns.TypeAAAA {
resp, err = r.handleIPQuery(name, query, netutils.IPv6)
} else if query.Question[0].Qtype == dns.TypePTR {
resp, err = r.handlePTRQuery(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
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
}
resp, err = co.ReadMsg()
if err != nil {
r.forwardQueryEnd(w, query)
log.Debugf("Read from DNS server failed, %s", err)
continue
}
// 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)
}
}
// HandleVIC returns a response to a container name/id request
func (s *Server) HandleVIC(w mdns.ResponseWriter, r *mdns.Msg) (bool, error) {
defer trace.End(trace.Begin(r.String()))
question := r.Question[0]
ctx := network.DefaultContext
if ctx == nil {
log.Errorf("DefaultContext is not initialized")
return false, fmt.Errorf("DefaultContext is not initialized")
}
clientIP, _, err := net.SplitHostPort(w.RemoteAddr().String())
if err != nil {
log.Errorf("SplitHostPort failed: %q", err)
return false, err
}
log.Debugf("RemoteAddr: %s", clientIP)
ip := net.ParseIP(clientIP)
var name string
var domain string
var scopename string
name = strings.TrimSuffix(question.Name, ".")
// Do we have a domain?
i := strings.IndexRune(name, '.')
if i >= 0 {
name, domain = name[:i], name[i+1:]
}
// first look for the network alias
c := ctx.Container(uid.UID(name))
if c == nil {
// find the scope of the request
scopes, _ := ctx.Scopes(nil)
// FIXME: We are doing linear search here
for i := range scopes {
s := scopes[i].Subnet()
if s.Contains(ip) {
scopename = scopes[i].Name()
log.Debugf("Found the scope, using %s", scopename)
// convert for or foo.bar to bar:foo
if domain == "" || domain == scopename {
name = scopename + ":" + name
break
}
}
}
c = ctx.Container(uid.UID(name))
if c == nil {
log.Debugf("Can't find the container: %q", name)
return false, fmt.Errorf("Can't find the container: %q", name)
}
}
var answer []mdns.RR
for _, e := range c.Endpoints() {
log.Debugf("Working on %s", e.Scope().Name())
if scopename != "" && e.Scope().Name() != scopename {
log.Debugf("Skipping non-matching scope %s", e.Scope().Name())
continue
}
if e.IP().IsUnspecified() {
log.Debugf("Skipping unspecified IP for %s", e.Scope().Name())
continue
}
// FIXME: Add AAAA when we support it
rr := &mdns.A{
Hdr: mdns.RR_Header{
Name: question.Name,
Rrtype: mdns.TypeA,
Class: mdns.ClassINET,
Ttl: uint32(DefaultTTL.Seconds()),
},
A: e.IP(),
}
answer = append(answer, rr)
}
// Start crafting reply msg
m := &mdns.Msg{
MsgHdr: mdns.MsgHdr{
Authoritative: true,
RecursionAvailable: true,
},
Compress: true,
}
m.SetReply(r)
m.Answer = append(m.Answer, answer...)
// Which protocol we are talking
tcp := false
if _, ok := w.LocalAddr().(*net.TCPAddr); ok {
tcp = true
}
// 512 byte payload guarantees that DNS packets can be reassembled if fragmented in transit.
bufsize := 512
// With EDNS0 in use a larger payload size can be specified.
if o := r.IsEdns0(); o != nil {
bufsize = int(o.UDPSize())
}
// Make sure we are not smaller than 512
if bufsize < 512 {
bufsize = 512
}
// With TCP we can send up to 64K
if tcp {
bufsize = mdns.MaxMsgSize - 1
}
// Trim the answer RRs one by one till the whole message fits within the reply size
if m.Len() > bufsize {
if tcp {
m.Truncated = true
}
for m.Len() > bufsize {
m.Answer = m.Answer[:len(m.Answer)-1]
}
}
if err := w.WriteMsg(m); err != nil {
log.Errorf("Error writing response, %s", err)
return true, err
}
w.Close()
return true, nil
}
// HandleVIC returns a response to a container name/id request
func (s *Server) HandleVIC(w mdns.ResponseWriter, r *mdns.Msg) (bool, error) {
defer trace.End(trace.Begin(r.String()))
question := r.Question[0]
ctx := network.DefaultContext
if ctx == nil {
log.Errorf("DefaultContext is not initialized")
return false, fmt.Errorf("DefaultContext is not initialized")
}
clientIP, _, err := net.SplitHostPort(w.RemoteAddr().String())
if err != nil {
log.Errorf("SplitHostPort failed: %q", err)
return false, err
}
log.Debugf("RemoteAddr: %s", clientIP)
ip := net.ParseIP(clientIP)
var name string
var domain string
name = strings.TrimSuffix(question.Name, ".")
// Do we have a domain?
i := strings.IndexRune(name, '.')
if i >= 0 {
name, domain = name[:i], name[i+1:]
}
// get the requesting container's endpoint
e := ctx.ContainerByAddr(ip)
if e == nil {
return false, fmt.Errorf("Could not find requesting container with ip %s", ip)
}
scope := e.Scope()
if domain != "" && scope.Name() != domain {
return false, fmt.Errorf("Inter-scope request for container %s in %s from %s", name, domain, scope.Name())
}
// container specific alias search
c := ctx.Container(fmt.Sprintf("%s:%s:%s", scope.Name(), e.Container().Name(), name))
if c == nil {
// scope-wide search
c = ctx.Container(fmt.Sprintf("%s:%s", scope.Name(), name))
}
if c == nil {
log.Debugf("Can't find the container: %q", name)
return false, fmt.Errorf("Can't find the container: %q", name)
}
e = c.Endpoint(scope)
if e.IP().IsUnspecified() {
return false, fmt.Errorf("No ip for container %q", name)
}
// FIXME: Add AAAA when we support it
answer := []mdns.RR{
&mdns.A{
Hdr: mdns.RR_Header{
Name: question.Name,
Rrtype: mdns.TypeA,
Class: mdns.ClassINET,
Ttl: uint32(DefaultTTL.Seconds()),
},
A: e.IP(),
},
}
// Start crafting reply msg
m := &mdns.Msg{
MsgHdr: mdns.MsgHdr{
Authoritative: true,
RecursionAvailable: true,
},
Compress: true,
}
m.SetReply(r)
m.Answer = append(m.Answer, answer...)
// Which protocol we are talking
tcp := false
if _, ok := w.LocalAddr().(*net.TCPAddr); ok {
tcp = true
}
// 512 byte payload guarantees that DNS packets can be reassembled if fragmented in transit.
bufsize := 512
// With EDNS0 in use a larger payload size can be specified.
if o := r.IsEdns0(); o != nil {
bufsize = int(o.UDPSize())
}
// Make sure we are not smaller than 512
if bufsize < 512 {
bufsize = 512
}
// With TCP we can send up to 64K
if tcp {
bufsize = mdns.MaxMsgSize - 1
}
// Trim the answer RRs one by one till the whole message fits within the reply size
if m.Len() > bufsize {
if tcp {
m.Truncated = true
}
for m.Len() > bufsize {
m.Answer = m.Answer[:len(m.Answer)-1]
}
}
if err := w.WriteMsg(m); err != nil {
log.Errorf("Error writing response, %s", err)
return true, err
}
w.Close()
return true, nil
}