syncthing/lib/connections/connections.go

470 lines
14 KiB
Go
Raw Normal View History

// Copyright (C) 2015 The Syncthing Authors.
//
2015-03-07 13:36:35 -07:00
// This Source Code Form is subject to the terms of the Mozilla Public
// License, v. 2.0. If a copy of the MPL was not distributed with this file,
// You can obtain one at http://mozilla.org/MPL/2.0/.
package connections
import (
"crypto/tls"
"fmt"
"io"
"net"
"net/url"
2015-06-28 08:05:29 -07:00
"sync"
"time"
"github.com/juju/ratelimit"
2015-08-06 02:29:25 -07:00
"github.com/syncthing/syncthing/lib/config"
2015-09-12 12:59:15 -07:00
"github.com/syncthing/syncthing/lib/discover"
2015-08-06 02:29:25 -07:00
"github.com/syncthing/syncthing/lib/events"
"github.com/syncthing/syncthing/lib/model"
2015-09-22 10:38:46 -07:00
"github.com/syncthing/syncthing/lib/protocol"
"github.com/syncthing/syncthing/lib/relay"
2015-09-22 10:38:46 -07:00
"github.com/syncthing/syncthing/lib/relay/client"
2015-06-28 13:09:03 -07:00
"github.com/thejerf/suture"
)
type DialerFactory func(*url.URL, *tls.Config) (*tls.Conn, error)
type ListenerFactory func(*url.URL, *tls.Config, chan<- model.IntermediateConnection)
var (
dialers = make(map[string]DialerFactory, 0)
listeners = make(map[string]ListenerFactory, 0)
)
type Model interface {
protocol.Model
AddConnection(conn model.Connection)
ConnectedTo(remoteID protocol.DeviceID) bool
IsPaused(remoteID protocol.DeviceID) bool
}
// The connection service listens on TLS and dials configured unconnected
2015-04-28 08:34:55 -07:00
// devices. Successful connections are handed to the model.
type connectionSvc struct {
*suture.Supervisor
cfg *config.Wrapper
myID protocol.DeviceID
model Model
tlsCfg *tls.Config
discoverer discover.Finder
conns chan model.IntermediateConnection
relaySvc *relay.Svc
bepProtocolName string
tlsDefaultCommonName string
lans []*net.IPNet
writeRateLimit *ratelimit.Bucket
readRateLimit *ratelimit.Bucket
2015-06-28 08:05:29 -07:00
2015-06-28 13:09:03 -07:00
lastRelayCheck map[protocol.DeviceID]time.Time
2015-07-24 12:07:26 -07:00
mut sync.RWMutex
connType map[protocol.DeviceID]model.ConnectionType
relaysEnabled bool
2015-06-28 08:05:29 -07:00
}
func NewConnectionSvc(cfg *config.Wrapper, myID protocol.DeviceID, mdl Model, tlsCfg *tls.Config, discoverer discover.Finder, relaySvc *relay.Svc,
bepProtocolName string, tlsDefaultCommonName string, lans []*net.IPNet) suture.Service {
svc := &connectionSvc{
Supervisor: suture.NewSimple("connectionSvc"),
cfg: cfg,
myID: myID,
model: mdl,
tlsCfg: tlsCfg,
discoverer: discoverer,
relaySvc: relaySvc,
conns: make(chan model.IntermediateConnection),
bepProtocolName: bepProtocolName,
tlsDefaultCommonName: tlsDefaultCommonName,
lans: lans,
2015-06-28 08:05:29 -07:00
2015-06-28 13:09:03 -07:00
connType: make(map[protocol.DeviceID]model.ConnectionType),
2015-07-24 12:07:26 -07:00
relaysEnabled: cfg.Options().RelaysEnabled,
2015-06-28 13:09:03 -07:00
lastRelayCheck: make(map[protocol.DeviceID]time.Time),
}
2015-06-28 12:09:53 -07:00
cfg.Subscribe(svc)
if svc.cfg.Options().MaxSendKbps > 0 {
svc.writeRateLimit = ratelimit.NewBucketWithRate(float64(1000*svc.cfg.Options().MaxSendKbps), int64(5*1000*svc.cfg.Options().MaxSendKbps))
}
if svc.cfg.Options().MaxRecvKbps > 0 {
svc.readRateLimit = ratelimit.NewBucketWithRate(float64(1000*svc.cfg.Options().MaxRecvKbps), int64(5*1000*svc.cfg.Options().MaxRecvKbps))
}
// There are several moving parts here; one routine per listening address
// to handle incoming connections, one routine to periodically attempt
2015-06-28 12:09:53 -07:00
// outgoing connections, one routine to the the common handling
// regardless of whether the connection was incoming or outgoing.
// Furthermore, a relay service which handles incoming requests to connect
// via the relays.
//
// TODO: Clean shutdown, and/or handling config changes on the fly. We
// partly do this now - new devices and addresses will be picked up, but
// not new listen addresses and we don't support disconnecting devices
// that are removed and so on...
svc.Add(serviceFunc(svc.connect))
for _, addr := range svc.cfg.Options().ListenAddress {
uri, err := url.Parse(addr)
if err != nil {
l.Infoln("Failed to parse listen address:", addr, err)
continue
}
listener, ok := listeners[uri.Scheme]
if !ok {
l.Infoln("Unknown listen address scheme:", uri.String())
continue
}
Implement facility based logger, debugging via REST API This implements a new debug/trace infrastructure based on a slightly hacked up logger. Instead of the traditional "if debug { ... }" I've rewritten the logger to have no-op Debugln and Debugf, unless debugging has been enabled for a given "facility". The "facility" is just a string, typically a package name. This will be slightly slower than before; but not that much as it's mostly a function call that returns immediately. For the cases where it matters (the Debugln takes a hex.Dump() of something for example, and it's not in a very occasional "if err != nil" branch) there is an l.ShouldDebug(facility) that is fast enough to be used like the old "if debug". The point of all this is that we can now toggle debugging for the various packages on and off at runtime. There's a new method /rest/system/debug that can be POSTed a set of facilities to enable and disable debug for, or GET from to get a list of facilities with descriptions and their current debug status. Similarly a /rest/system/log?since=... can grab the latest log entries, up to 250 of them (hardcoded constant in main.go) plus the initial few. Not implemented in this commit (but planned) is a simple debug GUI available on /debug that shows the current log in an easily pasteable format and has checkboxes to enable the various debug facilities. The debug instructions to a user then becomes "visit this URL, check these boxes, reproduce your problem, copy and paste the log". The actual log viewer on the hypothetical /debug URL can poll regularly for new log entries and this bypass the 250 line limit. The existing STTRACE=foo variable is still obeyed and just sets the start state of the system.
2015-10-03 08:25:21 -07:00
l.Debugln("listening on", uri)
svc.Add(serviceFunc(func() {
listener(uri, svc.tlsCfg, svc.conns)
}))
}
svc.Add(serviceFunc(svc.handle))
if svc.relaySvc != nil {
svc.Add(serviceFunc(svc.acceptRelayConns))
}
return svc
}
func (s *connectionSvc) handle() {
next:
2015-06-28 08:05:29 -07:00
for c := range s.conns {
cs := c.Conn.ConnectionState()
// We should have negotiated the next level protocol "bep/1.0" as part
// of the TLS handshake. Unfortunately this can't be a hard error,
// because there are implementations out there that don't support
// protocol negotiation (iOS for one...).
if !cs.NegotiatedProtocolIsMutual || cs.NegotiatedProtocol != s.bepProtocolName {
l.Infof("Peer %s did not negotiate bep/1.0", c.Conn.RemoteAddr())
2015-03-05 07:58:16 -07:00
}
// We should have received exactly one certificate from the other
// side. If we didn't, they don't have a device ID and we drop the
// connection.
2015-03-05 07:58:16 -07:00
certs := cs.PeerCertificates
if cl := len(certs); cl != 1 {
l.Infof("Got peer certificate list of length %d != 1 from %s; protocol error", cl, c.Conn.RemoteAddr())
c.Conn.Close()
continue
}
remoteCert := certs[0]
remoteID := protocol.NewDeviceID(remoteCert.Raw)
// The device ID should not be that of ourselves. It can happen
2015-04-28 08:34:55 -07:00
// though, especially in the presence of NAT hairpinning, multiple
// clients between the same NAT gateway, and global discovery.
if remoteID == s.myID {
l.Infof("Connected to myself (%s) - should not happen", remoteID)
c.Conn.Close()
continue
}
2015-06-28 13:09:03 -07:00
// If we have a relay connection, and the new incoming connection is
// not a relay connection, we should drop that, and prefer the this one.
s.mut.RLock()
ct, ok := s.connType[remoteID]
s.mut.RUnlock()
if ok && !ct.IsDirect() && c.Type.IsDirect() {
Implement facility based logger, debugging via REST API This implements a new debug/trace infrastructure based on a slightly hacked up logger. Instead of the traditional "if debug { ... }" I've rewritten the logger to have no-op Debugln and Debugf, unless debugging has been enabled for a given "facility". The "facility" is just a string, typically a package name. This will be slightly slower than before; but not that much as it's mostly a function call that returns immediately. For the cases where it matters (the Debugln takes a hex.Dump() of something for example, and it's not in a very occasional "if err != nil" branch) there is an l.ShouldDebug(facility) that is fast enough to be used like the old "if debug". The point of all this is that we can now toggle debugging for the various packages on and off at runtime. There's a new method /rest/system/debug that can be POSTed a set of facilities to enable and disable debug for, or GET from to get a list of facilities with descriptions and their current debug status. Similarly a /rest/system/log?since=... can grab the latest log entries, up to 250 of them (hardcoded constant in main.go) plus the initial few. Not implemented in this commit (but planned) is a simple debug GUI available on /debug that shows the current log in an easily pasteable format and has checkboxes to enable the various debug facilities. The debug instructions to a user then becomes "visit this URL, check these boxes, reproduce your problem, copy and paste the log". The actual log viewer on the hypothetical /debug URL can poll regularly for new log entries and this bypass the 250 line limit. The existing STTRACE=foo variable is still obeyed and just sets the start state of the system.
2015-10-03 08:25:21 -07:00
l.Debugln("Switching connections", remoteID)
2015-06-28 13:09:03 -07:00
s.model.Close(remoteID, fmt.Errorf("switching connections"))
} else if s.model.ConnectedTo(remoteID) {
// We should not already be connected to the other party. TODO: This
// could use some better handling. If the old connection is dead but
// hasn't timed out yet we may want to drop *that* connection and keep
// this one. But in case we are two devices connecting to each other
// in parallel we don't want to do that or we end up with no
// connections still established...
l.Infof("Connected to already connected device (%s)", remoteID)
c.Conn.Close()
continue
2015-08-23 12:56:10 -07:00
} else if s.model.IsPaused(remoteID) {
l.Infof("Connection from paused device (%s)", remoteID)
c.Conn.Close()
continue
}
for deviceID, deviceCfg := range s.cfg.Devices() {
if deviceID == remoteID {
// Verify the name on the certificate. By default we set it to
// "syncthing" when generating, but the user may have replaced
// the certificate and used another name.
certName := deviceCfg.CertName
if certName == "" {
certName = s.tlsDefaultCommonName
}
err := remoteCert.VerifyHostname(certName)
if err != nil {
// Incorrect certificate name is something the user most
// likely wants to know about, since it's an advanced
// config. Warn instead of Info.
l.Warnf("Bad certificate from %s (%v): %v", remoteID, c.Conn.RemoteAddr(), err)
c.Conn.Close()
continue next
}
// If rate limiting is set, and based on the address we should
// limit the connection, then we wrap it in a limiter.
limit := s.shouldLimit(c.Conn.RemoteAddr())
wr := io.Writer(c.Conn)
if limit && s.writeRateLimit != nil {
wr = NewWriteLimiter(c.Conn, s.writeRateLimit)
}
rd := io.Reader(c.Conn)
if limit && s.readRateLimit != nil {
rd = NewReadLimiter(c.Conn, s.readRateLimit)
}
name := fmt.Sprintf("%s-%s (%s)", c.Conn.LocalAddr(), c.Conn.RemoteAddr(), c.Type)
protoConn := protocol.NewConnection(remoteID, rd, wr, s.model, name, deviceCfg.Compression)
l.Infof("Established secure connection to %s at %s", remoteID, name)
Implement facility based logger, debugging via REST API This implements a new debug/trace infrastructure based on a slightly hacked up logger. Instead of the traditional "if debug { ... }" I've rewritten the logger to have no-op Debugln and Debugf, unless debugging has been enabled for a given "facility". The "facility" is just a string, typically a package name. This will be slightly slower than before; but not that much as it's mostly a function call that returns immediately. For the cases where it matters (the Debugln takes a hex.Dump() of something for example, and it's not in a very occasional "if err != nil" branch) there is an l.ShouldDebug(facility) that is fast enough to be used like the old "if debug". The point of all this is that we can now toggle debugging for the various packages on and off at runtime. There's a new method /rest/system/debug that can be POSTed a set of facilities to enable and disable debug for, or GET from to get a list of facilities with descriptions and their current debug status. Similarly a /rest/system/log?since=... can grab the latest log entries, up to 250 of them (hardcoded constant in main.go) plus the initial few. Not implemented in this commit (but planned) is a simple debug GUI available on /debug that shows the current log in an easily pasteable format and has checkboxes to enable the various debug facilities. The debug instructions to a user then becomes "visit this URL, check these boxes, reproduce your problem, copy and paste the log". The actual log viewer on the hypothetical /debug URL can poll regularly for new log entries and this bypass the 250 line limit. The existing STTRACE=foo variable is still obeyed and just sets the start state of the system.
2015-10-03 08:25:21 -07:00
l.Debugf("cipher suite: %04X in lan: %t", c.Conn.ConnectionState().CipherSuite, !limit)
2015-06-28 08:05:29 -07:00
s.model.AddConnection(model.Connection{
c.Conn,
2015-06-28 08:05:29 -07:00
protoConn,
c.Type,
2015-06-28 08:05:29 -07:00
})
s.mut.Lock()
s.connType[remoteID] = c.Type
2015-06-28 08:05:29 -07:00
s.mut.Unlock()
continue next
}
}
if !s.cfg.IgnoredDevice(remoteID) {
events.Default.Log(events.DeviceRejected, map[string]string{
"device": remoteID.String(),
"address": c.Conn.RemoteAddr().String(),
})
}
l.Infof("Connection from %s (%s) with ignored device ID %s", c.Conn.RemoteAddr(), c.Type, remoteID)
c.Conn.Close()
}
}
func (s *connectionSvc) connect() {
delay := time.Second
for {
nextDevice:
for deviceID, deviceCfg := range s.cfg.Devices() {
if deviceID == s.myID {
continue
}
2015-08-23 12:56:10 -07:00
if s.model.IsPaused(deviceID) {
continue
}
2015-06-28 13:09:03 -07:00
connected := s.model.ConnectedTo(deviceID)
s.mut.RLock()
ct, ok := s.connType[deviceID]
2015-07-24 12:07:26 -07:00
relaysEnabled := s.relaysEnabled
2015-06-28 13:09:03 -07:00
s.mut.RUnlock()
if connected && ok && ct.IsDirect() {
continue
}
addrs, relays := s.resolveAddresses(deviceID, deviceCfg.Addresses)
for _, addr := range addrs {
if conn := s.connectDirect(deviceID, addr); conn != nil {
if connected {
s.model.Close(deviceID, fmt.Errorf("switching connections"))
}
s.conns <- model.IntermediateConnection{
conn, model.ConnectionTypeDirectDial,
}
continue nextDevice
2015-06-28 08:05:29 -07:00
}
}
2015-06-28 13:09:03 -07:00
// Only connect via relays if not already connected
// Also, do not set lastRelayCheck time if we have no relays,
// as otherwise when we do discover relays, we might have to
// wait up to RelayReconnectIntervalM to connect again.
2015-07-24 12:07:26 -07:00
// Also, do not try relays if we are explicitly told not to.
if connected || len(relays) == 0 || !relaysEnabled {
2015-06-28 13:09:03 -07:00
continue nextDevice
}
reconIntv := time.Duration(s.cfg.Options().RelayReconnectIntervalM) * time.Minute
if last, ok := s.lastRelayCheck[deviceID]; ok && time.Since(last) < reconIntv {
Implement facility based logger, debugging via REST API This implements a new debug/trace infrastructure based on a slightly hacked up logger. Instead of the traditional "if debug { ... }" I've rewritten the logger to have no-op Debugln and Debugf, unless debugging has been enabled for a given "facility". The "facility" is just a string, typically a package name. This will be slightly slower than before; but not that much as it's mostly a function call that returns immediately. For the cases where it matters (the Debugln takes a hex.Dump() of something for example, and it's not in a very occasional "if err != nil" branch) there is an l.ShouldDebug(facility) that is fast enough to be used like the old "if debug". The point of all this is that we can now toggle debugging for the various packages on and off at runtime. There's a new method /rest/system/debug that can be POSTed a set of facilities to enable and disable debug for, or GET from to get a list of facilities with descriptions and their current debug status. Similarly a /rest/system/log?since=... can grab the latest log entries, up to 250 of them (hardcoded constant in main.go) plus the initial few. Not implemented in this commit (but planned) is a simple debug GUI available on /debug that shows the current log in an easily pasteable format and has checkboxes to enable the various debug facilities. The debug instructions to a user then becomes "visit this URL, check these boxes, reproduce your problem, copy and paste the log". The actual log viewer on the hypothetical /debug URL can poll regularly for new log entries and this bypass the 250 line limit. The existing STTRACE=foo variable is still obeyed and just sets the start state of the system.
2015-10-03 08:25:21 -07:00
l.Debugln("Skipping connecting via relay to", deviceID, "last checked at", last)
2015-06-28 13:09:03 -07:00
continue nextDevice
Implement facility based logger, debugging via REST API This implements a new debug/trace infrastructure based on a slightly hacked up logger. Instead of the traditional "if debug { ... }" I've rewritten the logger to have no-op Debugln and Debugf, unless debugging has been enabled for a given "facility". The "facility" is just a string, typically a package name. This will be slightly slower than before; but not that much as it's mostly a function call that returns immediately. For the cases where it matters (the Debugln takes a hex.Dump() of something for example, and it's not in a very occasional "if err != nil" branch) there is an l.ShouldDebug(facility) that is fast enough to be used like the old "if debug". The point of all this is that we can now toggle debugging for the various packages on and off at runtime. There's a new method /rest/system/debug that can be POSTed a set of facilities to enable and disable debug for, or GET from to get a list of facilities with descriptions and their current debug status. Similarly a /rest/system/log?since=... can grab the latest log entries, up to 250 of them (hardcoded constant in main.go) plus the initial few. Not implemented in this commit (but planned) is a simple debug GUI available on /debug that shows the current log in an easily pasteable format and has checkboxes to enable the various debug facilities. The debug instructions to a user then becomes "visit this URL, check these boxes, reproduce your problem, copy and paste the log". The actual log viewer on the hypothetical /debug URL can poll regularly for new log entries and this bypass the 250 line limit. The existing STTRACE=foo variable is still obeyed and just sets the start state of the system.
2015-10-03 08:25:21 -07:00
} else {
2015-06-28 13:09:03 -07:00
l.Debugln("Trying relay connections to", deviceID, relays)
}
s.lastRelayCheck[deviceID] = time.Now()
for _, addr := range relays {
if conn := s.connectViaRelay(deviceID, addr); conn != nil {
s.conns <- model.IntermediateConnection{
conn, model.ConnectionTypeRelayDial,
}
continue nextDevice
2015-06-28 13:09:03 -07:00
}
}
}
time.Sleep(delay)
delay *= 2
if maxD := time.Duration(s.cfg.Options().ReconnectIntervalS) * time.Second; delay > maxD {
delay = maxD
}
}
}
func (s *connectionSvc) resolveAddresses(deviceID protocol.DeviceID, inAddrs []string) (addrs []string, relays []discover.Relay) {
for _, addr := range inAddrs {
if addr == "dynamic" {
if s.discoverer != nil {
if t, r, err := s.discoverer.Lookup(deviceID); err == nil {
addrs = append(addrs, t...)
relays = append(relays, r...)
}
}
} else {
addrs = append(addrs, addr)
}
}
return
}
func (s *connectionSvc) connectDirect(deviceID protocol.DeviceID, addr string) *tls.Conn {
uri, err := url.Parse(addr)
if err != nil {
l.Infoln("Failed to parse connection url:", addr, err)
return nil
}
dialer, ok := dialers[uri.Scheme]
if !ok {
l.Infoln("Unknown address schema", uri)
return nil
}
l.Debugln("dial", deviceID, uri)
conn, err := dialer(uri, s.tlsCfg)
if err != nil {
l.Debugln("dial failed", deviceID, uri, err)
return nil
}
return conn
}
func (s *connectionSvc) connectViaRelay(deviceID protocol.DeviceID, addr discover.Relay) *tls.Conn {
uri, err := url.Parse(addr.URL)
if err != nil {
l.Infoln("Failed to parse relay connection url:", addr, err)
return nil
}
2015-11-23 14:14:46 -07:00
inv, err := client.GetInvitationFromRelay(uri, deviceID, s.tlsCfg.Certificates, 10*time.Second)
if err != nil {
l.Debugf("Failed to get invitation for %s from %s: %v", deviceID, uri, err)
return nil
}
l.Debugln("Succesfully retrieved relay invitation", inv, "from", uri)
conn, err := client.JoinSession(inv)
if err != nil {
l.Debugf("Failed to join relay session %s: %v", inv, err)
return nil
}
2015-11-11 19:20:34 -07:00
l.Debugln("Successfully joined relay session", inv)
var tc *tls.Conn
if inv.ServerSocket {
tc = tls.Server(conn, s.tlsCfg)
} else {
tc = tls.Client(conn, s.tlsCfg)
}
err = tc.Handshake()
if err != nil {
l.Infof("TLS handshake (BEP/relay %s): %v", inv, err)
tc.Close()
return nil
}
return tc
}
func (s *connectionSvc) acceptRelayConns() {
for {
conn := s.relaySvc.Accept()
s.conns <- model.IntermediateConnection{
Conn: conn,
Type: model.ConnectionTypeRelayAccept,
}
}
}
func (s *connectionSvc) shouldLimit(addr net.Addr) bool {
if s.cfg.Options().LimitBandwidthInLan {
return true
}
tcpaddr, ok := addr.(*net.TCPAddr)
if !ok {
return true
}
for _, lan := range s.lans {
if lan.Contains(tcpaddr.IP) {
return false
}
}
return !tcpaddr.IP.IsLoopback()
}
func (s *connectionSvc) VerifyConfiguration(from, to config.Configuration) error {
return nil
}
func (s *connectionSvc) CommitConfiguration(from, to config.Configuration) bool {
2015-07-24 12:07:26 -07:00
s.mut.Lock()
s.relaysEnabled = to.Options.RelaysEnabled
s.mut.Unlock()
// We require a restart if a device as been removed.
newDevices := make(map[protocol.DeviceID]bool, len(to.Devices))
for _, dev := range to.Devices {
newDevices[dev.DeviceID] = true
}
for _, dev := range from.Devices {
if !newDevices[dev.DeviceID] {
return false
}
}
return true
}
// serviceFunc wraps a function to create a suture.Service without stop
// functionality.
type serviceFunc func()
func (f serviceFunc) Serve() { f() }
func (f serviceFunc) Stop() {}