headscale/hscontrol/poll.go

358 lines
9.2 KiB
Go
Raw Normal View History

package hscontrol
import (
"context"
"fmt"
"net/http"
"time"
"github.com/juanfont/headscale/hscontrol/mapper"
"github.com/juanfont/headscale/hscontrol/types"
"github.com/juanfont/headscale/hscontrol/util"
"github.com/rs/zerolog/log"
"tailscale.com/tailcfg"
)
const (
keepAliveInterval = 60 * time.Second
)
2022-05-16 06:59:46 -06:00
type contextKey string
const machineNameContextKey = contextKey("machineName")
type UpdateNode func()
func logPollFunc(
mapRequest tailcfg.MapRequest,
machine *types.Machine,
isNoise bool,
) (func(string), func(error, string)) {
return func(msg string) {
log.Info().
Caller().
Bool("noise", isNoise).
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", machine.NodeKey).
Str("machine", machine.Hostname).
Msg(msg)
},
func(err error, msg string) {
log.Error().
Caller().
Bool("noise", isNoise).
Bool("readOnly", mapRequest.ReadOnly).
Bool("omitPeers", mapRequest.OmitPeers).
Bool("stream", mapRequest.Stream).
Str("node_key", machine.NodeKey).
Str("machine", machine.Hostname).
Err(err).
Msg(msg)
}
}
// handlePoll is the common code for the legacy and Noise protocols to
2022-08-15 02:43:39 -06:00
// managed the poll loop.
func (h *Headscale) handlePoll(
2022-06-26 04:06:25 -06:00
writer http.ResponseWriter,
ctx context.Context,
machine *types.Machine,
mapRequest tailcfg.MapRequest,
isNoise bool,
2022-06-20 04:30:51 -06:00
) {
// Immediate open the channel and register it if the client wants
// a stream of MapResponses to prevent initial map response and
// following updates missing
var updateChan chan types.StateUpdate
if mapRequest.Stream {
h.pollNetMapStreamWG.Add(1)
defer h.pollNetMapStreamWG.Done()
updateChan = make(chan types.StateUpdate)
defer closeChanWithLog(updateChan, machine.Hostname, "updateChan")
// Register the node's update channel
h.nodeNotifier.AddNode(machine.MachineKey, updateChan)
defer h.nodeNotifier.RemoveNode(machine.MachineKey)
}
logInfo, logErr := logPollFunc(mapRequest, machine, isNoise)
mapp := mapper.NewMapper(
machine,
h.db,
h.privateKey2019,
isNoise,
h.DERPMap,
h.cfg.BaseDomain,
h.cfg.DNSConfig,
h.cfg.LogTail.Enabled,
h.cfg.RandomizeClientPort,
)
2022-06-26 04:06:25 -06:00
machine.Hostname = mapRequest.Hostinfo.Hostname
machine.HostInfo = types.HostInfo(*mapRequest.Hostinfo)
machine.DiscoKey = util.DiscoPublicKeyStripPrefix(mapRequest.DiscoKey)
now := time.Now().UTC()
err := h.db.ProcessMachineRoutes(machine)
if err != nil {
logErr(err, "Error processing machine routes")
}
// update ACLRules with peer informations (to update server tags if necessary)
if h.ACLPolicy != nil {
// update routes with peer information
err = h.db.EnableAutoApprovedRoutes(h.ACLPolicy, machine)
if err != nil {
logErr(err, "Error running auto approved routes")
}
}
// From Tailscale client:
//
// 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.
2022-06-26 04:06:25 -06:00
if !mapRequest.ReadOnly {
machine.Endpoints = mapRequest.Endpoints
2021-11-15 09:15:50 -07:00
machine.LastSeen = &now
}
2022-05-30 07:39:24 -06:00
// TODO(kradalby): Save specific stuff, not whole object.
if err := h.db.MachineSave(machine); err != nil {
logErr(err, "Failed to persist/update machine in the database")
http.Error(writer, "", http.StatusInternalServerError)
2022-05-30 07:39:24 -06:00
return
2022-05-30 07:39:24 -06:00
}
if !mapRequest.ReadOnly {
// It sounds like we should update the nodes when we have received a endpoint update
// even tho the comments in the tailscale code dont explicitly say so.
updateRequestsFromNode.WithLabelValues(machine.User.Name, machine.Hostname, "endpoint-update").
Inc()
// Tell all the other nodes about the new endpoint, but dont update ourselves.
h.nodeNotifier.NotifyWithIgnore(
types.StateUpdate{
Type: types.StatePeerChanged,
Changed: []uint64{machine.ID},
},
machine.MachineKey)
}
// We update our peers if the client is not sending ReadOnly in the MapRequest
// so we don't distribute its initial request (it comes with
// empty endpoints to peers)
// Details on the protocol can be found in https://github.com/tailscale/tailscale/blob/main/tailcfg/tailcfg.go#L696
logInfo("Client map request processed")
2022-06-26 04:06:25 -06:00
if mapRequest.ReadOnly {
logInfo("Client is starting up. Probably interested in a DERP map")
2022-06-20 04:30:51 -06:00
mapResp, err := mapp.FullMapResponse(mapRequest, machine, h.ACLPolicy)
if err != nil {
logErr(err, "Failed to create MapResponse")
http.Error(writer, "", http.StatusInternalServerError)
return
}
2022-06-26 04:06:25 -06:00
writer.Header().Set("Content-Type", "application/json; charset=utf-8")
writer.WriteHeader(http.StatusOK)
_, err = writer.Write(mapResp)
2022-06-26 04:21:35 -06:00
if err != nil {
logErr(err, "Failed to write response")
2022-06-26 04:21:35 -06:00
}
2021-11-14 08:46:09 -07:00
if f, ok := writer.(http.Flusher); ok {
f.Flush()
}
return
}
2022-06-26 04:06:25 -06:00
if mapRequest.OmitPeers && !mapRequest.Stream {
logInfo("Client sent endpoint update and is ok with a response without peer list")
mapResp, err := mapp.LiteMapResponse(mapRequest, machine, h.ACLPolicy)
if err != nil {
logErr(err, "Failed to create MapResponse")
http.Error(writer, "", http.StatusInternalServerError)
return
}
2022-06-26 04:06:25 -06:00
writer.Header().Set("Content-Type", "application/json; charset=utf-8")
writer.WriteHeader(http.StatusOK)
_, err = writer.Write(mapResp)
2022-06-26 04:21:35 -06:00
if err != nil {
logErr(err, "Failed to write response")
2022-06-26 04:21:35 -06:00
}
2021-11-14 08:46:09 -07:00
return
2022-06-26 04:06:25 -06:00
} else if mapRequest.OmitPeers && mapRequest.Stream {
log.Warn().
Str("handler", "PollNetMap").
2022-08-14 15:11:33 -06:00
Bool("noise", isNoise).
2022-04-24 13:55:54 -06:00
Str("machine", machine.Hostname).
Msg("Ignoring request, don't know how to handle it")
2021-11-14 08:46:09 -07:00
return
}
logInfo("Sending initial map")
mapResp, err := mapp.FullMapResponse(mapRequest, machine, 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
}
keepAliveTicker := time.NewTicker(keepAliveInterval)
ctx = context.WithValue(ctx, machineNameContextKey, machine.Hostname)
2022-06-20 04:30:51 -06:00
ctx, cancel := context.WithCancel(ctx)
defer cancel()
2022-06-20 13:40:28 -06:00
for {
logInfo("Waiting for update on stream channel")
select {
case <-keepAliveTicker.C:
data, err := mapp.KeepAliveResponse(mapRequest, machine)
if err != nil {
logErr(err, "Error generating the keep alive msg")
2021-11-14 08:46:09 -07:00
2022-06-20 13:40:28 -06:00
return
}
_, err = writer.Write(data)
if err != nil {
logErr(err, "Cannot write keep alive message")
2022-06-26 04:25:26 -06:00
return
2022-06-26 04:25:26 -06:00
}
if flusher, ok := writer.(http.Flusher); ok {
flusher.Flush()
} else {
log.Error().Msg("Failed to create http flusher")
2022-06-20 13:40:28 -06:00
return
}
err = h.db.TouchMachine(machine)
if err != nil {
logErr(err, "Cannot update machine LastSeen")
2022-06-20 13:40:28 -06:00
return
}
2021-11-14 08:46:09 -07:00
case update := <-updateChan:
logInfo("Received update")
var data []byte
var err error
switch update.Type {
case types.StatePeerChanged:
logInfo("Sending PeerChanged MapResponse")
data, err = mapp.PeerChangedResponse(mapRequest, machine, update.Changed, h.ACLPolicy)
case types.StatePeerRemoved:
logInfo("Sending PeerRemoved MapResponse")
data, err = mapp.PeerRemovedResponse(mapRequest, machine, update.Removed)
case types.StateDERPUpdated:
logInfo("Sending DERPUpdate MapResponse")
data, err = mapp.DERPMapResponse(mapRequest, machine, update.DERPMap)
case types.StateFullUpdate:
logInfo("Sending Full MapResponse")
data, err = mapp.FullMapResponse(mapRequest, machine, h.ACLPolicy)
}
if err != nil {
logErr(err, "Could not get the create map update")
2021-11-14 08:46:09 -07:00
2022-06-20 13:40:28 -06:00
return
}
2022-06-20 13:40:28 -06:00
_, err = writer.Write(data)
if err != nil {
logErr(err, "Could not write the map response")
updateRequestsSentToNode.WithLabelValues(machine.User.Name, machine.Hostname, "failed").
Inc()
2022-06-20 13:40:28 -06:00
return
}
2021-11-14 08:46:09 -07:00
if flusher, ok := writer.(http.Flusher); ok {
flusher.Flush()
} else {
log.Error().Msg("Failed to create http flusher")
return
}
2021-11-14 08:46:09 -07:00
// Keep track of the last successful update,
// we sometimes end in a state were the update
// is not picked up by a client and we use this
// to determine if we should "force" an update.
err = h.db.TouchMachine(machine)
if err != nil {
logErr(err, "Cannot update machine LastSuccessfulUpdate")
2022-06-20 13:40:28 -06:00
return
}
logInfo("Update sent")
case <-ctx.Done():
logInfo("The client has closed the connection")
err := h.db.TouchMachine(machine)
if err != nil {
logErr(err, "Cannot update machine LastSeen")
}
2022-06-20 13:40:28 -06:00
// The connection has been closed, so we can stop polling.
return
case <-h.shutdownChan:
logInfo("The long-poll handler is shutting down")
2022-06-26 04:06:25 -06:00
return
}
2022-06-20 04:30:51 -06:00
}
}
func closeChanWithLog[C chan []byte | chan struct{} | chan types.StateUpdate](channel C, machine, name string) {
log.Trace().
Str("handler", "PollNetMap").
Str("machine", machine).
Str("channel", "Done").
Msg(fmt.Sprintf("Closing %s channel", name))
close(channel)
}