/
derp_server.go
1869 lines (1685 loc) · 54.1 KB
/
derp_server.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package derp
// TODO(crawshaw): with predefined serverKey in clients and HMAC on packets we could skip TLS
import (
"bufio"
"context"
"crypto/ed25519"
crand "crypto/rand"
"crypto/x509"
"crypto/x509/pkix"
"encoding/json"
"errors"
"expvar"
"fmt"
"io"
"io/ioutil"
"log"
"math"
"math/big"
"math/rand"
"net"
"net/http"
"net/netip"
"os/exec"
"runtime"
"strconv"
"strings"
"sync"
"sync/atomic"
"time"
"go4.org/mem"
"golang.org/x/sync/errgroup"
"golang.org/x/time/rate"
"tailscale.com/client/tailscale"
"tailscale.com/disco"
"tailscale.com/envknob"
"tailscale.com/metrics"
"tailscale.com/types/key"
"tailscale.com/types/logger"
"tailscale.com/types/pad32"
"tailscale.com/version"
)
var debug = envknob.Bool("DERP_DEBUG_LOGS")
// verboseDropKeys is the set of destination public keys that should
// verbosely log whenever DERP drops a packet.
var verboseDropKeys = map[key.NodePublic]bool{}
func init() {
keys := envknob.String("TS_DEBUG_VERBOSE_DROPS")
if keys == "" {
return
}
for _, keyStr := range strings.Split(keys, ",") {
k, err := key.ParseNodePublicUntyped(mem.S(keyStr))
if err != nil {
log.Printf("ignoring invalid debug key %q: %v", keyStr, err)
} else {
verboseDropKeys[k] = true
}
}
}
func init() {
rand.Seed(time.Now().UnixNano())
}
const (
perClientSendQueueDepth = 32 // packets buffered for sending
writeTimeout = 2 * time.Second
)
// dupPolicy is a temporary (2021-08-30) mechanism to change the policy
// of how duplicate connection for the same key are handled.
type dupPolicy int8
const (
// lastWriterIsActive is a dupPolicy where the connection
// to send traffic for a peer is the active one.
lastWriterIsActive dupPolicy = iota
// disableFighters is a dupPolicy that detects if peers
// are trying to send interleaved with each other and
// then disables all of them.
disableFighters
)
// Server is a DERP server.
type Server struct {
// WriteTimeout, if non-zero, specifies how long to wait
// before failing when writing to a client.
WriteTimeout time.Duration
privateKey key.NodePrivate
publicKey key.NodePublic
logf logger.Logf
memSys0 uint64 // runtime.MemStats.Sys at start (or early-ish)
meshKey string
limitedLogf logger.Logf
metaCert []byte // the encoded x509 cert to send after LetsEncrypt cert+intermediate
dupPolicy dupPolicy
// Counters:
packetsSent, bytesSent expvar.Int
packetsRecv, bytesRecv expvar.Int
packetsRecvByKind metrics.LabelMap
packetsRecvDisco *expvar.Int
packetsRecvOther *expvar.Int
_ pad32.Four
packetsDropped expvar.Int
packetsDroppedReason metrics.LabelMap
packetsDroppedReasonCounters []*expvar.Int // indexed by dropReason
packetsDroppedType metrics.LabelMap
packetsDroppedTypeDisco *expvar.Int
packetsDroppedTypeOther *expvar.Int
_ pad32.Four
packetsForwardedOut expvar.Int
packetsForwardedIn expvar.Int
peerGoneFrames expvar.Int // number of peer gone frames sent
gotPing expvar.Int // number of ping frames from client
sentPong expvar.Int // number of pong frames enqueued to client
accepts expvar.Int
curClients expvar.Int
curHomeClients expvar.Int // ones with preferred
dupClientKeys expvar.Int // current number of public keys we have 2+ connections for
dupClientConns expvar.Int // current number of connections sharing a public key
dupClientConnTotal expvar.Int // total number of accepted connections when a dup key existed
unknownFrames expvar.Int
homeMovesIn expvar.Int // established clients announce home server moves in
homeMovesOut expvar.Int // established clients announce home server moves out
multiForwarderCreated expvar.Int
multiForwarderDeleted expvar.Int
removePktForwardOther expvar.Int
avgQueueDuration *uint64 // In milliseconds; accessed atomically
// verifyClients only accepts client connections to the DERP server if the clientKey is a
// known peer in the network, as specified by a running tailscaled's client's local api.
verifyClients bool
mu sync.Mutex
closed bool
netConns map[Conn]chan struct{} // chan is closed when conn closes
clients map[key.NodePublic]clientSet
watchers map[*sclient]bool // mesh peer -> true
// clientsMesh tracks all clients in the cluster, both locally
// and to mesh peers. If the value is nil, that means the
// peer is only local (and thus in the clients Map, but not
// remote). If the value is non-nil, it's remote (+ maybe also
// local).
clientsMesh map[key.NodePublic]PacketForwarder
// sentTo tracks which peers have sent to which other peers,
// and at which connection number. This isn't on sclient
// because it includes intra-region forwarded packets as the
// src.
sentTo map[key.NodePublic]map[key.NodePublic]int64 // src => dst => dst's latest sclient.connNum
// maps from netip.AddrPort to a client's public key
keyOfAddr map[netip.AddrPort]key.NodePublic
}
// clientSet represents 1 or more *sclients.
//
// The two implementations are singleClient and *dupClientSet.
//
// In the common case, client should only have one connection to the
// DERP server for a given key. When they're connected multiple times,
// we record their set of connections in dupClientSet and keep their
// connections open to make them happy (to keep them from spinning,
// etc) and keep track of which is the latest connection. If only the last
// is sending traffic, that last one is the active connection and it
// gets traffic. Otherwise, in the case of a cloned node key, the
// whole set of dups doesn't receive data frames.
//
// All methods should only be called while holding Server.mu.
//
// TODO(bradfitz): Issue 2746: in the future we'll send some sort of
// "health_error" frame to them that'll communicate to the end users
// that they cloned a device key, and we'll also surface it in the
// admin panel, etc.
type clientSet interface {
// ActiveClient returns the most recently added client to
// the set, as long as it hasn't been disabled, in which
// case it returns nil.
ActiveClient() *sclient
// Len returns the number of clients in the set.
Len() int
// ForeachClient calls f for each client in the set.
ForeachClient(f func(*sclient))
}
// singleClient is a clientSet of a single connection.
// This is the common case.
type singleClient struct{ c *sclient }
func (s singleClient) ActiveClient() *sclient { return s.c }
func (s singleClient) Len() int { return 1 }
func (s singleClient) ForeachClient(f func(*sclient)) { f(s.c) }
// A dupClientSet is a clientSet of more than 1 connection.
//
// This can occur in some reasonable cases (temporarily while users
// are changing networks) or in the case of a cloned key. In the
// cloned key case, both peers are speaking and the clients get
// disabled.
//
// All fields are guarded by Server.mu.
type dupClientSet struct {
// set is the set of connected clients for sclient.key.
// The values are all true.
set map[*sclient]bool
// last is the most recent addition to set, or nil if the most
// recent one has since disconnected and nobody else has send
// data since.
last *sclient
// sendHistory is a log of which members of set have sent
// frames to the derp server, with adjacent duplicates
// removed. When a member of set is removed, the same
// element(s) are removed from sendHistory.
sendHistory []*sclient
}
func (s *dupClientSet) ActiveClient() *sclient {
if s.last != nil && !s.last.isDisabled.Load() {
return s.last
}
return nil
}
func (s *dupClientSet) Len() int { return len(s.set) }
func (s *dupClientSet) ForeachClient(f func(*sclient)) {
for c := range s.set {
f(c)
}
}
// removeClient removes c from s and reports whether it was in s
// to begin with.
func (s *dupClientSet) removeClient(c *sclient) bool {
n := len(s.set)
delete(s.set, c)
if s.last == c {
s.last = nil
}
if len(s.set) == n {
return false
}
trim := s.sendHistory[:0]
for _, v := range s.sendHistory {
if s.set[v] && (len(trim) == 0 || trim[len(trim)-1] != v) {
trim = append(trim, v)
}
}
for i := len(trim); i < len(s.sendHistory); i++ {
s.sendHistory[i] = nil
}
s.sendHistory = trim
if s.last == nil && len(s.sendHistory) > 0 {
s.last = s.sendHistory[len(s.sendHistory)-1]
}
return true
}
// PacketForwarder is something that can forward packets.
//
// It's mostly an interface for circular dependency reasons; the
// typical implementation is derphttp.Client. The other implementation
// is a multiForwarder, which this package creates as needed if a
// public key gets more than one PacketForwarder registered for it.
type PacketForwarder interface {
ForwardPacket(src, dst key.NodePublic, payload []byte) error
}
// Conn is the subset of the underlying net.Conn the DERP Server needs.
// It is a defined type so that non-net connections can be used.
type Conn interface {
io.WriteCloser
LocalAddr() net.Addr
// The *Deadline methods follow the semantics of net.Conn.
SetDeadline(time.Time) error
SetReadDeadline(time.Time) error
SetWriteDeadline(time.Time) error
}
// NewServer returns a new DERP server. It doesn't listen on its own.
// Connections are given to it via Server.Accept.
func NewServer(privateKey key.NodePrivate, logf logger.Logf) *Server {
var ms runtime.MemStats
runtime.ReadMemStats(&ms)
s := &Server{
privateKey: privateKey,
publicKey: privateKey.Public(),
logf: logf,
limitedLogf: logger.RateLimitedFn(logf, 30*time.Second, 5, 100),
packetsRecvByKind: metrics.LabelMap{Label: "kind"},
packetsDroppedReason: metrics.LabelMap{Label: "reason"},
packetsDroppedType: metrics.LabelMap{Label: "type"},
clients: map[key.NodePublic]clientSet{},
clientsMesh: map[key.NodePublic]PacketForwarder{},
netConns: map[Conn]chan struct{}{},
memSys0: ms.Sys,
watchers: map[*sclient]bool{},
sentTo: map[key.NodePublic]map[key.NodePublic]int64{},
avgQueueDuration: new(uint64),
keyOfAddr: map[netip.AddrPort]key.NodePublic{},
}
s.initMetacert()
s.packetsRecvDisco = s.packetsRecvByKind.Get("disco")
s.packetsRecvOther = s.packetsRecvByKind.Get("other")
s.packetsDroppedReasonCounters = []*expvar.Int{
s.packetsDroppedReason.Get("unknown_dest"),
s.packetsDroppedReason.Get("unknown_dest_on_fwd"),
s.packetsDroppedReason.Get("gone"),
s.packetsDroppedReason.Get("queue_head"),
s.packetsDroppedReason.Get("queue_tail"),
s.packetsDroppedReason.Get("write_error"),
}
s.packetsDroppedTypeDisco = s.packetsDroppedType.Get("disco")
s.packetsDroppedTypeOther = s.packetsDroppedType.Get("other")
return s
}
// SetMesh sets the pre-shared key that regional DERP servers used to mesh
// amongst themselves.
//
// It must be called before serving begins.
func (s *Server) SetMeshKey(v string) {
s.meshKey = v
}
// SetVerifyClients sets whether this DERP server verifies clients through tailscaled.
//
// It must be called before serving begins.
func (s *Server) SetVerifyClient(v bool) {
s.verifyClients = v
}
// HasMeshKey reports whether the server is configured with a mesh key.
func (s *Server) HasMeshKey() bool { return s.meshKey != "" }
// MeshKey returns the configured mesh key, if any.
func (s *Server) MeshKey() string { return s.meshKey }
// PrivateKey returns the server's private key.
func (s *Server) PrivateKey() key.NodePrivate { return s.privateKey }
// PublicKey returns the server's public key.
func (s *Server) PublicKey() key.NodePublic { return s.publicKey }
// Close closes the server and waits for the connections to disconnect.
func (s *Server) Close() error {
s.mu.Lock()
wasClosed := s.closed
s.closed = true
s.mu.Unlock()
if wasClosed {
return nil
}
var closedChs []chan struct{}
s.mu.Lock()
for nc, closed := range s.netConns {
nc.Close()
closedChs = append(closedChs, closed)
}
s.mu.Unlock()
for _, closed := range closedChs {
<-closed
}
return nil
}
func (s *Server) isClosed() bool {
s.mu.Lock()
defer s.mu.Unlock()
return s.closed
}
// IsClientConnectedForTest reports whether the client with specified key is connected.
// This is used in tests to verify that nodes are connected.
func (s *Server) IsClientConnectedForTest(k key.NodePublic) bool {
s.mu.Lock()
defer s.mu.Unlock()
x, ok := s.clients[k]
if !ok {
return false
}
return x.ActiveClient() != nil
}
// Accept adds a new connection to the server and serves it.
//
// The provided bufio ReadWriter must be already connected to nc.
// Accept blocks until the Server is closed or the connection closes
// on its own.
//
// Accept closes nc.
func (s *Server) Accept(ctx context.Context, nc Conn, brw *bufio.ReadWriter, remoteAddr string) {
closed := make(chan struct{})
s.mu.Lock()
s.accepts.Add(1) // while holding s.mu for connNum read on next line
connNum := s.accepts.Value() // expvar sadly doesn't return new value on Add(1)
s.netConns[nc] = closed
s.mu.Unlock()
defer func() {
nc.Close()
close(closed)
s.mu.Lock()
delete(s.netConns, nc)
s.mu.Unlock()
}()
if err := s.accept(ctx, nc, brw, remoteAddr, connNum); err != nil && !s.isClosed() {
s.logf("derp: %s: %v", remoteAddr, err)
}
}
// initMetacert initialized s.metaCert with a self-signed x509 cert
// encoding this server's public key and protocol version. cmd/derper
// then sends this after the Let's Encrypt leaf + intermediate certs
// after the ServerHello (encrypted in TLS 1.3, not that it matters
// much).
//
// Then the client can save a round trip getting that and can start
// speaking DERP right away. (We don't use ALPN because that's sent in
// the clear and we're being paranoid to not look too weird to any
// middleboxes, given that DERP is an ultimate fallback path). But
// since the post-ServerHello certs are encrypted we can have the
// client also use them as a signal to be able to start speaking DERP
// right away, starting with its identity proof, encrypted to the
// server's public key.
//
// This RTT optimization fails where there's a corp-mandated
// TLS proxy with corp-mandated root certs on employee machines and
// and TLS proxy cleans up unnecessary certs. In that case we just fall
// back to the extra RTT.
func (s *Server) initMetacert() {
pub, priv, err := ed25519.GenerateKey(crand.Reader)
if err != nil {
log.Fatal(err)
}
tmpl := &x509.Certificate{
SerialNumber: big.NewInt(ProtocolVersion),
Subject: pkix.Name{
CommonName: fmt.Sprintf("derpkey%s", s.publicKey.UntypedHexString()),
},
// Windows requires NotAfter and NotBefore set:
NotAfter: time.Now().Add(30 * 24 * time.Hour),
NotBefore: time.Now().Add(-30 * 24 * time.Hour),
// Per https://github.com/golang/go/issues/51759#issuecomment-1071147836,
// macOS requires BasicConstraints when subject == issuer:
BasicConstraintsValid: true,
}
cert, err := x509.CreateCertificate(crand.Reader, tmpl, tmpl, pub, priv)
if err != nil {
log.Fatalf("CreateCertificate: %v", err)
}
s.metaCert = cert
}
// MetaCert returns the server metadata cert that can be sent by the
// TLS server to let the client skip a round trip during start-up.
func (s *Server) MetaCert() []byte { return s.metaCert }
// registerClient notes that client c is now authenticated and ready for packets.
//
// If c.key is connected more than once, the earlier connection(s) are
// placed in a non-active state where we read from them (primarily to
// observe EOFs/timeouts) but won't send them frames on the assumption
// that they're dead.
func (s *Server) registerClient(c *sclient) {
s.mu.Lock()
defer s.mu.Unlock()
set := s.clients[c.key]
switch set := set.(type) {
case nil:
s.clients[c.key] = singleClient{c}
case singleClient:
s.dupClientKeys.Add(1)
s.dupClientConns.Add(2) // both old and new count
s.dupClientConnTotal.Add(1)
old := set.ActiveClient()
old.isDup.Store(true)
c.isDup.Store(true)
s.clients[c.key] = &dupClientSet{
last: c,
set: map[*sclient]bool{
old: true,
c: true,
},
sendHistory: []*sclient{old},
}
case *dupClientSet:
s.dupClientConns.Add(1) // the gauge
s.dupClientConnTotal.Add(1) // the counter
c.isDup.Store(true)
set.set[c] = true
set.last = c
set.sendHistory = append(set.sendHistory, c)
}
if _, ok := s.clientsMesh[c.key]; !ok {
s.clientsMesh[c.key] = nil // just for varz of total users in cluster
}
s.keyOfAddr[c.remoteIPPort] = c.key
s.curClients.Add(1)
s.broadcastPeerStateChangeLocked(c.key, true)
}
// broadcastPeerStateChangeLocked enqueues a message to all watchers
// (other DERP nodes in the region, or trusted clients) that peer's
// presence changed.
//
// s.mu must be held.
func (s *Server) broadcastPeerStateChangeLocked(peer key.NodePublic, present bool) {
for w := range s.watchers {
w.peerStateChange = append(w.peerStateChange, peerConnState{peer: peer, present: present})
go w.requestMeshUpdate()
}
}
// unregisterClient removes a client from the server.
func (s *Server) unregisterClient(c *sclient) {
s.mu.Lock()
defer s.mu.Unlock()
set := s.clients[c.key]
switch set := set.(type) {
case nil:
c.logf("[unexpected]; clients map is empty")
case singleClient:
c.logf("removing connection")
delete(s.clients, c.key)
if v, ok := s.clientsMesh[c.key]; ok && v == nil {
delete(s.clientsMesh, c.key)
s.notePeerGoneFromRegionLocked(c.key)
}
s.broadcastPeerStateChangeLocked(c.key, false)
case *dupClientSet:
if set.removeClient(c) {
s.dupClientConns.Add(-1)
} else {
c.logf("[unexpected]; dup client set didn't shrink")
}
if set.Len() == 1 {
s.dupClientConns.Add(-1) // again; for the original one's
s.dupClientKeys.Add(-1)
var remain *sclient
for remain = range set.set {
break
}
if remain == nil {
panic("unexpected nil remain from single element dup set")
}
remain.isDisabled.Store(false)
remain.isDup.Store(false)
s.clients[c.key] = singleClient{remain}
}
}
if c.canMesh {
delete(s.watchers, c)
}
delete(s.keyOfAddr, c.remoteIPPort)
s.curClients.Add(-1)
if c.preferred {
s.curHomeClients.Add(-1)
}
}
// notePeerGoneFromRegionLocked sends peerGone frames to parties that
// key has sent to previously (whether those sends were from a local
// client or forwarded). It must only be called after the key has
// been removed from clientsMesh.
func (s *Server) notePeerGoneFromRegionLocked(key key.NodePublic) {
if _, ok := s.clientsMesh[key]; ok {
panic("usage")
}
// Find still-connected peers and either notify that we've gone away
// so they can drop their route entries to us (issue 150)
// or move them over to the active client (in case a replaced client
// connection is being unregistered).
for pubKey, connNum := range s.sentTo[key] {
set, ok := s.clients[pubKey]
if !ok {
continue
}
set.ForeachClient(func(peer *sclient) {
if peer.connNum == connNum {
go peer.requestPeerGoneWrite(key)
}
})
}
delete(s.sentTo, key)
}
func (s *Server) addWatcher(c *sclient) {
if !c.canMesh {
panic("invariant: addWatcher called without permissions")
}
if c.key == s.publicKey {
// We're connecting to ourself. Do nothing.
return
}
s.mu.Lock()
defer s.mu.Unlock()
// Queue messages for each already-connected client.
for peer := range s.clients {
c.peerStateChange = append(c.peerStateChange, peerConnState{peer: peer, present: true})
}
// And enroll the watcher in future updates (of both
// connections & disconnections).
s.watchers[c] = true
go c.requestMeshUpdate()
}
func (s *Server) accept(ctx context.Context, nc Conn, brw *bufio.ReadWriter, remoteAddr string, connNum int64) error {
br := brw.Reader
nc.SetDeadline(time.Now().Add(10 * time.Second))
bw := &lazyBufioWriter{w: nc, lbw: brw.Writer}
if err := s.sendServerKey(bw); err != nil {
return fmt.Errorf("send server key: %v", err)
}
nc.SetDeadline(time.Now().Add(10 * time.Second))
clientKey, clientInfo, err := s.recvClientKey(br)
if err != nil {
return fmt.Errorf("receive client key: %v", err)
}
if err := s.verifyClient(clientKey, clientInfo); err != nil {
return fmt.Errorf("client %x rejected: %v", clientKey, err)
}
// At this point we trust the client so we don't time out.
nc.SetDeadline(time.Time{})
ctx, cancel := context.WithCancel(ctx)
defer cancel()
remoteIPPort, _ := netip.ParseAddrPort(remoteAddr)
c := &sclient{
connNum: connNum,
s: s,
key: clientKey,
nc: nc,
br: br,
bw: bw,
logf: logger.WithPrefix(s.logf, fmt.Sprintf("derp client %v/%x: ", remoteAddr, clientKey)),
done: ctx.Done(),
remoteAddr: remoteAddr,
remoteIPPort: remoteIPPort,
connectedAt: time.Now(),
sendQueue: make(chan pkt, perClientSendQueueDepth),
discoSendQueue: make(chan pkt, perClientSendQueueDepth),
sendPongCh: make(chan [8]byte, 1),
peerGone: make(chan key.NodePublic),
canMesh: clientInfo.MeshKey != "" && clientInfo.MeshKey == s.meshKey,
}
if c.canMesh {
c.meshUpdate = make(chan struct{})
}
if clientInfo != nil {
c.info = *clientInfo
}
s.registerClient(c)
defer s.unregisterClient(c)
err = s.sendServerInfo(c.bw, clientKey)
if err != nil {
return fmt.Errorf("send server info: %v", err)
}
return c.run(ctx)
}
// for testing
var (
timeSleep = time.Sleep
timeNow = time.Now
)
// run serves the client until there's an error.
// If the client hangs up or the server is closed, run returns nil, otherwise run returns an error.
func (c *sclient) run(ctx context.Context) error {
// Launch sender, but don't return from run until sender goroutine is done.
var grp errgroup.Group
sendCtx, cancelSender := context.WithCancel(ctx)
grp.Go(func() error { return c.sendLoop(sendCtx) })
defer func() {
cancelSender()
if err := grp.Wait(); err != nil && !c.s.isClosed() {
c.logf("sender failed: %v", err)
}
}()
for {
ft, fl, err := readFrameHeader(c.br)
if err != nil {
if errors.Is(err, io.EOF) {
c.logf("read EOF")
return nil
}
if c.s.isClosed() {
c.logf("closing; server closed")
return nil
}
return fmt.Errorf("client %x: readFrameHeader: %w", c.key, err)
}
c.s.noteClientActivity(c)
switch ft {
case frameNotePreferred:
err = c.handleFrameNotePreferred(ft, fl)
case frameSendPacket:
err = c.handleFrameSendPacket(ft, fl)
case frameForwardPacket:
err = c.handleFrameForwardPacket(ft, fl)
case frameWatchConns:
err = c.handleFrameWatchConns(ft, fl)
case frameClosePeer:
err = c.handleFrameClosePeer(ft, fl)
case framePing:
err = c.handleFramePing(ft, fl)
default:
err = c.handleUnknownFrame(ft, fl)
}
if err != nil {
return err
}
}
}
func (c *sclient) handleUnknownFrame(ft frameType, fl uint32) error {
_, err := io.CopyN(ioutil.Discard, c.br, int64(fl))
return err
}
func (c *sclient) handleFrameNotePreferred(ft frameType, fl uint32) error {
if fl != 1 {
return fmt.Errorf("frameNotePreferred wrong size")
}
v, err := c.br.ReadByte()
if err != nil {
return fmt.Errorf("frameNotePreferred ReadByte: %v", err)
}
c.setPreferred(v != 0)
return nil
}
func (c *sclient) handleFrameWatchConns(ft frameType, fl uint32) error {
if fl != 0 {
return fmt.Errorf("handleFrameWatchConns wrong size")
}
if !c.canMesh {
return fmt.Errorf("insufficient permissions")
}
c.s.addWatcher(c)
return nil
}
func (c *sclient) handleFramePing(ft frameType, fl uint32) error {
c.s.gotPing.Add(1)
var m PingMessage
if fl < uint32(len(m)) {
return fmt.Errorf("short ping: %v", fl)
}
if fl > 1000 {
// unreasonably extra large. We leave some extra
// space for future extensibility, but not too much.
return fmt.Errorf("ping body too large: %v", fl)
}
_, err := io.ReadFull(c.br, m[:])
if err != nil {
return err
}
if extra := int64(fl) - int64(len(m)); extra > 0 {
_, err = io.CopyN(ioutil.Discard, c.br, extra)
}
select {
case c.sendPongCh <- [8]byte(m):
default:
// They're pinging too fast. Ignore.
// TODO(bradfitz): add a rate limiter too.
}
return err
}
func (c *sclient) handleFrameClosePeer(ft frameType, fl uint32) error {
if fl != keyLen {
return fmt.Errorf("handleFrameClosePeer wrong size")
}
if !c.canMesh {
return fmt.Errorf("insufficient permissions")
}
var targetKey key.NodePublic
if err := targetKey.ReadRawWithoutAllocating(c.br); err != nil {
return err
}
s := c.s
s.mu.Lock()
defer s.mu.Unlock()
if set, ok := s.clients[targetKey]; ok {
if set.Len() == 1 {
c.logf("frameClosePeer closing peer %x", targetKey)
} else {
c.logf("frameClosePeer closing peer %x (%d connections)", targetKey, set.Len())
}
set.ForeachClient(func(target *sclient) {
go target.nc.Close()
})
} else {
c.logf("frameClosePeer failed to find peer %x", targetKey)
}
return nil
}
// handleFrameForwardPacket reads a "forward packet" frame from the client
// (which must be a trusted client, a peer in our mesh).
func (c *sclient) handleFrameForwardPacket(ft frameType, fl uint32) error {
if !c.canMesh {
return fmt.Errorf("insufficient permissions")
}
s := c.s
srcKey, dstKey, contents, err := s.recvForwardPacket(c.br, fl)
if err != nil {
return fmt.Errorf("client %x: recvForwardPacket: %v", c.key, err)
}
s.packetsForwardedIn.Add(1)
var dstLen int
var dst *sclient
s.mu.Lock()
if set, ok := s.clients[dstKey]; ok {
dstLen = set.Len()
dst = set.ActiveClient()
}
if dst != nil {
s.notePeerSendLocked(srcKey, dst)
}
s.mu.Unlock()
if dst == nil {
reason := dropReasonUnknownDestOnFwd
if dstLen > 1 {
reason = dropReasonDupClient
}
s.recordDrop(contents, srcKey, dstKey, reason)
return nil
}
return c.sendPkt(dst, pkt{
bs: contents,
enqueuedAt: time.Now(),
src: srcKey,
})
}
// notePeerSendLocked records that src sent to dst. We keep track of
// that so when src disconnects, we can tell dst (if it's still
// around) that src is gone (a peerGone frame).
func (s *Server) notePeerSendLocked(src key.NodePublic, dst *sclient) {
m, ok := s.sentTo[src]
if !ok {
m = map[key.NodePublic]int64{}
s.sentTo[src] = m
}
m[dst.key] = dst.connNum
}
// handleFrameSendPacket reads a "send packet" frame from the client.
func (c *sclient) handleFrameSendPacket(ft frameType, fl uint32) error {
s := c.s
dstKey, contents, err := s.recvPacket(c.br, fl)
if err != nil {
return fmt.Errorf("client %x: recvPacket: %v", c.key, err)
}
var fwd PacketForwarder
var dstLen int
var dst *sclient
s.mu.Lock()
if set, ok := s.clients[dstKey]; ok {
dstLen = set.Len()
dst = set.ActiveClient()
}
if dst != nil {
s.notePeerSendLocked(c.key, dst)
} else if dstLen < 1 {
fwd = s.clientsMesh[dstKey]
}
s.mu.Unlock()
if dst == nil {
if fwd != nil {
s.packetsForwardedOut.Add(1)
if err := fwd.ForwardPacket(c.key, dstKey, contents); err != nil {
// TODO:
return nil
}
return nil
}
reason := dropReasonUnknownDest
if dstLen > 1 {
reason = dropReasonDupClient
}
s.recordDrop(contents, c.key, dstKey, reason)
return nil
}
p := pkt{
bs: contents,
enqueuedAt: time.Now(),
src: c.key,
}
return c.sendPkt(dst, p)
}
// dropReason is why we dropped a DERP frame.
type dropReason int
//go:generate go run tailscale.com/cmd/addlicense -year 2021 -file dropreason_string.go go run golang.org/x/tools/cmd/stringer -type=dropReason -trimprefix=dropReason
const (
dropReasonUnknownDest dropReason = iota // unknown destination pubkey
dropReasonUnknownDestOnFwd // unknown destination pubkey on a derp-forwarded packet
dropReasonGone // destination tailscaled disconnected before we could send
dropReasonQueueHead // destination queue is full, dropped packet at queue head
dropReasonQueueTail // destination queue is full, dropped packet at queue tail
dropReasonWriteError // OS write() failed
dropReasonDupClient // the public key is connected 2+ times (active/active, fighting)
)
func (s *Server) recordDrop(packetBytes []byte, srcKey, dstKey key.NodePublic, reason dropReason) {
s.packetsDropped.Add(1)
s.packetsDroppedReasonCounters[reason].Add(1)
if disco.LooksLikeDiscoWrapper(packetBytes) {
s.packetsDroppedTypeDisco.Add(1)
} else {
s.packetsDroppedTypeOther.Add(1)
}
if verboseDropKeys[dstKey] {
// Preformat the log string prior to calling limitedLogf. The
// limiter acts based on the format string, and we want to
// rate-limit per src/dst keys, not on the generic "dropped
// stuff" message.
msg := fmt.Sprintf("drop (%s) %s -> %s", srcKey.ShortString(), reason, dstKey.ShortString())
s.limitedLogf(msg)
}
if debug {
s.logf("dropping packet reason=%s dst=%s disco=%v", reason, dstKey, disco.LooksLikeDiscoWrapper(packetBytes))
}
}
func (c *sclient) sendPkt(dst *sclient, p pkt) error {
s := c.s
dstKey := dst.key
// Attempt to queue for sending up to 3 times. On each attempt, if
// the queue is full, try to drop from queue head to prioritize
// fresher packets.
sendQueue := dst.sendQueue
if disco.LooksLikeDiscoWrapper(p.bs) {
sendQueue = dst.discoSendQueue
}
for attempt := 0; attempt < 3; attempt++ {
select {