forked from juanfont/headscale
/
poll.go
668 lines (553 loc) · 19.3 KB
/
poll.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
package hscontrol
import (
"context"
"fmt"
"net/http"
"strings"
"time"
"github.com/juanfont/headscale/hscontrol/db"
"github.com/juanfont/headscale/hscontrol/mapper"
"github.com/juanfont/headscale/hscontrol/types"
"github.com/rs/zerolog/log"
xslices "golang.org/x/exp/slices"
"gorm.io/gorm"
"tailscale.com/envknob"
"tailscale.com/tailcfg"
)
const (
keepAliveInterval = 60 * time.Second
)
type contextKey string
const nodeNameContextKey = contextKey("nodeName")
type UpdateNode func()
func logPollFunc(
mapRequest tailcfg.MapRequest,
node *types.Node,
) (func(string), func(string), func(error, string)) {
return func(msg string) {
log.Trace().
Caller().
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", node.NodeKey.ShortString()).
Str("node", node.Hostname).
Msg(msg)
},
func(msg string) {
log.Warn().
Caller().
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", node.NodeKey.ShortString()).
Str("node", node.Hostname).
Msg(msg)
},
func(err error, msg string) {
log.Error().
Caller().
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", node.NodeKey.ShortString()).
Str("node", node.Hostname).
Err(err).
Msg(msg)
}
}
// handlePoll ensures the node gets the appropriate updates from either
// polling or immediate responses.
//
//nolint:gocyclo
func (h *Headscale) handlePoll(
writer http.ResponseWriter,
ctx context.Context,
node *types.Node,
mapRequest tailcfg.MapRequest,
) {
logTrace, logWarn, logErr := logPollFunc(mapRequest, node)
// This is the mechanism where the node gives us information about its
// current configuration.
//
// If OmitPeers is true, Stream is false, and ReadOnly is false,
// then te server will let clients update their endpoints without
// breaking existing long-polling (Stream == true) connections.
// In this case, the server can omit the entire response; the client
// only checks the HTTP response status code.
// TODO(kradalby): remove ReadOnly when we only support capVer 68+
if mapRequest.OmitPeers && !mapRequest.Stream && !mapRequest.ReadOnly {
log.Info().
Caller().
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", node.NodeKey.ShortString()).
Str("node", node.Hostname).
Int("cap_ver", int(mapRequest.Version)).
Msg("Received update")
change := node.PeerChangeFromMapRequest(mapRequest)
online := h.nodeNotifier.IsConnected(node.MachineKey)
change.Online = &online
node.ApplyPeerChange(&change)
hostInfoChange := node.Hostinfo.Equal(mapRequest.Hostinfo)
logTracePeerChange(node.Hostname, hostInfoChange, &change)
// Check if the Hostinfo of the node has changed.
// If it has changed, check if there has been a change tod
// the routable IPs of the host and update update them in
// the database. Then send a Changed update
// (containing the whole node object) to peers to inform about
// the route change.
// If the hostinfo has changed, but not the routes, just update
// hostinfo and let the function continue.
if !hostInfoChange {
oldRoutes := node.Hostinfo.RoutableIPs
newRoutes := mapRequest.Hostinfo.RoutableIPs
oldServicesCount := len(node.Hostinfo.Services)
newServicesCount := len(mapRequest.Hostinfo.Services)
node.Hostinfo = mapRequest.Hostinfo
sendUpdate := false
// Route changes come as part of Hostinfo, which means that
// when an update comes, the Node Route logic need to run.
// This will require a "change" in comparison to a "patch",
// which is more costly.
if !xslices.Equal(oldRoutes, newRoutes) {
var err error
sendUpdate, err = h.db.SaveNodeRoutes(node)
if err != nil {
logErr(err, "Error processing node routes")
http.Error(writer, "", http.StatusInternalServerError)
return
}
if h.ACLPolicy != nil {
// update routes with peer information
update, err := h.db.EnableAutoApprovedRoutes(h.ACLPolicy, node)
if err != nil {
logErr(err, "Error running auto approved routes")
}
if update != nil {
sendUpdate = true
}
}
}
// Services is mostly useful for discovery and not critical,
// except for peerapi, which is how nodes talk to eachother.
// If peerapi was not part of the initial mapresponse, we
// need to make sure its sent out later as it is needed for
// Taildrop.
// TODO(kradalby): Length comparison is a bit naive, replace.
if oldServicesCount != newServicesCount {
sendUpdate = true
}
if sendUpdate {
if err := h.db.DB.Save(node).Error; err != nil {
logErr(err, "Failed to persist/update node in the database")
http.Error(writer, "", http.StatusInternalServerError)
return
}
// Send an update to all peers to propagate the new routes
// available.
stateUpdate := types.StateUpdate{
Type: types.StatePeerChanged,
ChangeNodes: types.Nodes{node},
Message: "called from handlePoll -> update -> new hostinfo",
}
if stateUpdate.Valid() {
ctx := types.NotifyCtx(context.Background(), "poll-nodeupdate-peers-hostinfochange", node.Hostname)
h.nodeNotifier.NotifyWithIgnore(
ctx,
stateUpdate,
node.MachineKey.String())
}
// Send an update to the node itself with to ensure it
// has an updated packetfilter allowing the new route
// if it is defined in the ACL.
selfUpdate := types.StateUpdate{
Type: types.StateSelfUpdate,
ChangeNodes: types.Nodes{node},
}
if selfUpdate.Valid() {
ctx := types.NotifyCtx(context.Background(), "poll-nodeupdate-self-hostinfochange", node.Hostname)
h.nodeNotifier.NotifyByMachineKey(
ctx,
selfUpdate,
node.MachineKey)
}
return
}
}
if err := h.db.DB.Save(node).Error; err != nil {
logErr(err, "Failed to persist/update node in the database")
http.Error(writer, "", http.StatusInternalServerError)
return
}
stateUpdate := types.StateUpdate{
Type: types.StatePeerChangedPatch,
ChangePatches: []*tailcfg.PeerChange{&change},
}
if stateUpdate.Valid() {
ctx := types.NotifyCtx(context.Background(), "poll-nodeupdate-peers-patch", node.Hostname)
h.nodeNotifier.NotifyWithIgnore(
ctx,
stateUpdate,
node.MachineKey.String())
}
writer.WriteHeader(http.StatusOK)
if f, ok := writer.(http.Flusher); ok {
f.Flush()
}
return
} else if mapRequest.OmitPeers && !mapRequest.Stream && mapRequest.ReadOnly {
// ReadOnly is whether the client just wants to fetch the
// MapResponse, without updating their Endpoints. The
// Endpoints field will be ignored and LastSeen will not be
// updated and peers will not be notified of changes.
//
// The intended use is for clients to discover the DERP map at
// start-up before their first real endpoint update.
} else if mapRequest.OmitPeers && !mapRequest.Stream && mapRequest.ReadOnly {
h.handleLiteRequest(writer, node, mapRequest)
return
} else if mapRequest.OmitPeers && mapRequest.Stream {
logErr(nil, "Ignoring request, don't know how to handle it")
return
}
change := node.PeerChangeFromMapRequest(mapRequest)
// A stream is being set up, the node is Online
online := true
change.Online = &online
node.ApplyPeerChange(&change)
// Only save HostInfo if changed, update routes if changed
// TODO(kradalby): Remove when capver is over 68
if !node.Hostinfo.Equal(mapRequest.Hostinfo) {
oldRoutes := node.Hostinfo.RoutableIPs
newRoutes := mapRequest.Hostinfo.RoutableIPs
node.Hostinfo = mapRequest.Hostinfo
if !xslices.Equal(oldRoutes, newRoutes) {
_, err := h.db.SaveNodeRoutes(node)
if err != nil {
logErr(err, "Error processing node routes")
http.Error(writer, "", http.StatusInternalServerError)
return
}
}
}
if err := h.db.DB.Save(node).Error; err != nil {
logErr(err, "Failed to persist/update node in the database")
http.Error(writer, "", http.StatusInternalServerError)
return
}
// Set up the client stream
h.pollNetMapStreamWG.Add(1)
defer h.pollNetMapStreamWG.Done()
// Use a buffered channel in case a node is not fully ready
// to receive a message to make sure we dont block the entire
// notifier.
// 12 is arbitrarily chosen.
chanSize := 3
if size, ok := envknob.LookupInt("HEADSCALE_TUNING_POLL_QUEUE_SIZE"); ok {
chanSize = size
}
updateChan := make(chan types.StateUpdate, chanSize)
defer closeChanWithLog(updateChan, node.Hostname, "updateChan")
// Register the node's update channel
h.nodeNotifier.AddNode(node.MachineKey, updateChan)
defer h.nodeNotifier.RemoveNode(node.MachineKey)
// When a node connects to control, list the peers it has at
// that given point, further updates are kept in memory in
// the Mapper, which lives for the duration of the polling
// session.
peers, err := h.db.ListPeers(node)
if err != nil {
logErr(err, "Failed to list peers when opening poller")
http.Error(writer, "", http.StatusInternalServerError)
return
}
isConnected := h.nodeNotifier.ConnectedMap()
for _, peer := range peers {
online := isConnected[peer.MachineKey]
peer.IsOnline = &online
}
mapp := mapper.NewMapper(
node,
peers,
h.DERPMap,
h.cfg.BaseDomain,
h.cfg.DNSConfig,
h.cfg.LogTail.Enabled,
h.cfg.RandomizeClientPort,
)
// update ACLRules with peer informations (to update server tags if necessary)
if h.ACLPolicy != nil {
// update routes with peer information
// This state update is ignored as it will be sent
// as part of the whole node
// TODO(kradalby): figure out if that is actually correct
_, err = h.db.EnableAutoApprovedRoutes(h.ACLPolicy, node)
if err != nil {
logErr(err, "Error running auto approved routes")
}
}
logTrace("Sending initial map")
mapResp, err := mapp.FullMapResponse(mapRequest, node, h.ACLPolicy)
if err != nil {
logErr(err, "Failed to create MapResponse")
http.Error(writer, "", http.StatusInternalServerError)
return
}
// Send the client an update to make sure we send an initial mapresponse
_, err = writer.Write(mapResp)
if err != nil {
logErr(err, "Could not write the map response")
return
}
if flusher, ok := writer.(http.Flusher); ok {
flusher.Flush()
} else {
return
}
stateUpdate := types.StateUpdate{
Type: types.StatePeerChanged,
ChangeNodes: types.Nodes{node},
Message: "called from handlePoll -> new node added",
}
if stateUpdate.Valid() {
ctx := types.NotifyCtx(context.Background(), "poll-newnode-peers", node.Hostname)
h.nodeNotifier.NotifyWithIgnore(
ctx,
stateUpdate,
node.MachineKey.String())
}
if len(node.Routes) > 0 {
go h.pollFailoverRoutes(logErr, "new node", node)
}
keepAliveTicker := time.NewTicker(keepAliveInterval)
ctx, cancel := context.WithCancel(context.WithValue(ctx, nodeNameContextKey, node.Hostname))
defer cancel()
for {
logTrace("Waiting for update on stream channel")
select {
case <-keepAliveTicker.C:
data, err := mapp.KeepAliveResponse(mapRequest, node)
if err != nil {
logErr(err, "Error generating the keep alive msg")
return
}
_, err = writer.Write(data)
if err != nil {
logErr(err, "Cannot write keep alive message")
return
}
if flusher, ok := writer.(http.Flusher); ok {
flusher.Flush()
} else {
log.Error().Msg("Failed to create http flusher")
return
}
// This goroutine is not ideal, but we have a potential issue here
// where it blocks too long and that holds up updates.
// One alternative is to split these different channels into
// goroutines, but then you might have a problem without a lock
// if a keepalive is written at the same time as an update.
go h.updateNodeOnlineStatus(true, node)
case update := <-updateChan:
logTrace("Received update")
now := time.Now()
var data []byte
var err error
// Ensure the node object is updated, for example, there
// might have been a hostinfo update in a sidechannel
// which contains data needed to generate a map response.
node, err = h.db.GetNodeByMachineKey(node.MachineKey)
if err != nil {
logErr(err, "Could not get machine from db")
return
}
startMapResp := time.Now()
switch update.Type {
case types.StateFullUpdate:
logTrace("Sending Full MapResponse")
data, err = mapp.FullMapResponse(mapRequest, node, h.ACLPolicy)
case types.StatePeerChanged:
logTrace(fmt.Sprintf("Sending Changed MapResponse: %s", update.Message))
isConnectedMap := h.nodeNotifier.ConnectedMap()
for _, node := range update.ChangeNodes {
// If a node is not reported to be online, it might be
// because the value is outdated, check with the notifier.
// However, if it is set to Online, and not in the notifier,
// this might be because it has announced itself, but not
// reached the stage to actually create the notifier channel.
if node.IsOnline != nil && !*node.IsOnline {
isOnline := isConnectedMap[node.MachineKey]
node.IsOnline = &isOnline
}
}
data, err = mapp.PeerChangedResponse(mapRequest, node, update.ChangeNodes, h.ACLPolicy, update.Message)
case types.StatePeerChangedPatch:
logTrace("Sending PeerChangedPatch MapResponse")
data, err = mapp.PeerChangedPatchResponse(mapRequest, node, update.ChangePatches, h.ACLPolicy)
case types.StatePeerRemoved:
logTrace("Sending PeerRemoved MapResponse")
data, err = mapp.PeerRemovedResponse(mapRequest, node, update.Removed)
case types.StateSelfUpdate:
if len(update.ChangeNodes) == 1 {
logTrace("Sending SelfUpdate MapResponse")
node = update.ChangeNodes[0]
data, err = mapp.LiteMapResponse(mapRequest, node, h.ACLPolicy, types.SelfUpdateIdentifier)
} else {
logWarn("SelfUpdate contained too many nodes, this is likely a bug in the code, please report.")
}
case types.StateDERPUpdated:
logTrace("Sending DERPUpdate MapResponse")
data, err = mapp.DERPMapResponse(mapRequest, node, update.DERPMap)
}
if err != nil {
logErr(err, "Could not get the create map update")
return
}
log.Trace().Str("node", node.Hostname).TimeDiff("timeSpent", time.Now(), startMapResp).Str("mkey", node.MachineKey.String()).Int("type", int(update.Type)).Msg("finished making map response")
// Only send update if there is change
if data != nil {
startWrite := time.Now()
_, err = writer.Write(data)
if err != nil {
logErr(err, "Could not write the map response")
updateRequestsSentToNode.WithLabelValues(node.User.Name, node.Hostname, "failed").
Inc()
return
}
if flusher, ok := writer.(http.Flusher); ok {
flusher.Flush()
} else {
log.Error().Msg("Failed to create http flusher")
return
}
log.Trace().Str("node", node.Hostname).TimeDiff("timeSpent", time.Now(), startWrite).Str("mkey", node.MachineKey.String()).Int("type", int(update.Type)).Msg("finished writing mapresp to node")
log.Debug().
Caller().
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", node.NodeKey.ShortString()).
Str("machine_key", node.MachineKey.ShortString()).
Str("node", node.Hostname).
TimeDiff("timeSpent", time.Now(), now).
Msg("update sent")
}
case <-ctx.Done():
logTrace("The client has closed the connection")
go h.updateNodeOnlineStatus(false, node)
// Failover the node's routes if any.
go h.pollFailoverRoutes(logErr, "node closing connection", node)
// The connection has been closed, so we can stop polling.
return
case <-h.shutdownChan:
logTrace("The long-poll handler is shutting down")
return
}
}
}
func (h *Headscale) pollFailoverRoutes(logErr func(error, string), where string, node *types.Node) {
update, err := db.Write(h.db.DB, func(tx *gorm.DB) (*types.StateUpdate, error) {
return db.EnsureFailoverRouteIsAvailable(tx, h.nodeNotifier.ConnectedMap(), node)
})
if err != nil {
logErr(err, fmt.Sprintf("failed to ensure failover routes, %s", where))
return
}
if update != nil && !update.Empty() && update.Valid() {
ctx := types.NotifyCtx(context.Background(), fmt.Sprintf("poll-%s-routes-ensurefailover", strings.ReplaceAll(where, " ", "-")), node.Hostname)
h.nodeNotifier.NotifyWithIgnore(ctx, *update, node.MachineKey.String())
}
}
// updateNodeOnlineStatus records the last seen status of a node and notifies peers
// about change in their online/offline status.
// It takes a StateUpdateType of either StatePeerOnlineChanged or StatePeerOfflineChanged.
func (h *Headscale) updateNodeOnlineStatus(online bool, node *types.Node) {
now := time.Now()
node.LastSeen = &now
statusUpdate := types.StateUpdate{
Type: types.StatePeerChangedPatch,
ChangePatches: []*tailcfg.PeerChange{
{
NodeID: tailcfg.NodeID(node.ID),
Online: &online,
LastSeen: &now,
},
},
}
if statusUpdate.Valid() {
ctx := types.NotifyCtx(context.Background(), "poll-nodeupdate-onlinestatus", node.Hostname)
h.nodeNotifier.NotifyWithIgnore(ctx, statusUpdate, node.MachineKey.String())
}
err := h.db.DB.Transaction(func(tx *gorm.DB) error {
return db.UpdateLastSeen(tx, node.ID, *node.LastSeen)
})
if err != nil {
log.Error().Err(err).Msg("Cannot update node LastSeen")
return
}
}
func closeChanWithLog[C chan []byte | chan struct{} | chan types.StateUpdate](channel C, node, name string) {
log.Trace().
Str("handler", "PollNetMap").
Str("node", node).
Str("channel", "Done").
Msg(fmt.Sprintf("Closing %s channel", name))
close(channel)
}
func (h *Headscale) handleLiteRequest(
writer http.ResponseWriter,
node *types.Node,
mapRequest tailcfg.MapRequest,
) {
logTrace, _, logErr := logPollFunc(mapRequest, node)
mapp := mapper.NewMapper(
node,
types.Nodes{},
h.DERPMap,
h.cfg.BaseDomain,
h.cfg.DNSConfig,
h.cfg.LogTail.Enabled,
h.cfg.RandomizeClientPort,
)
logTrace("Client asked for a lite update, responding without peers")
mapResp, err := mapp.LiteMapResponse(mapRequest, node, h.ACLPolicy)
if err != nil {
logErr(err, "Failed to create MapResponse")
http.Error(writer, "", http.StatusInternalServerError)
return
}
writer.Header().Set("Content-Type", "application/json; charset=utf-8")
writer.WriteHeader(http.StatusOK)
_, err = writer.Write(mapResp)
if err != nil {
logErr(err, "Failed to write response")
}
}
func logTracePeerChange(hostname string, hostinfoChange bool, change *tailcfg.PeerChange) {
trace := log.Trace().Str("node_id", change.NodeID.String()).Str("hostname", hostname)
if change.Key != nil {
trace = trace.Str("node_key", change.Key.ShortString())
}
if change.DiscoKey != nil {
trace = trace.Str("disco_key", change.DiscoKey.ShortString())
}
if change.Online != nil {
trace = trace.Bool("online", *change.Online)
}
if change.Endpoints != nil {
eps := make([]string, len(change.Endpoints))
for idx, ep := range change.Endpoints {
eps[idx] = ep.String()
}
trace = trace.Strs("endpoints", eps)
}
if hostinfoChange {
trace = trace.Bool("hostinfo_changed", hostinfoChange)
}
if change.DERPRegion != 0 {
trace = trace.Int("derp_region", change.DERPRegion)
}
trace.Time("last_seen", *change.LastSeen).Msg("PeerChange received")
}