diff options
Diffstat (limited to 'pkg/tcpip/transport')
| -rw-r--r-- | pkg/tcpip/transport/icmp/endpoint.go | 82 | ||||
| -rw-r--r-- | pkg/tcpip/transport/icmp/protocol.go | 38 | ||||
| -rw-r--r-- | pkg/tcpip/transport/raw/endpoint.go | 39 | ||||
| -rw-r--r-- | pkg/tcpip/transport/raw/protocol.go | 9 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/BUILD | 5 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/accept.go | 34 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/connect.go | 39 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/endpoint.go | 298 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/protocol.go | 24 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/snd.go | 9 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/tcp_noracedetector_test.go | 10 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/tcp_test.go | 420 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/testing/context/context.go | 52 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcpconntrack/BUILD | 3 | ||||
| -rw-r--r-- | pkg/tcpip/transport/udp/BUILD | 5 | ||||
| -rw-r--r-- | pkg/tcpip/transport/udp/endpoint.go | 175 | ||||
| -rw-r--r-- | pkg/tcpip/transport/udp/endpoint_state.go | 4 | ||||
| -rw-r--r-- | pkg/tcpip/transport/udp/forwarder.go | 4 | ||||
| -rw-r--r-- | pkg/tcpip/transport/udp/protocol.go | 113 | ||||
| -rw-r--r-- | pkg/tcpip/transport/udp/udp_test.go | 401 |
20 files changed, 1277 insertions, 487 deletions
diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 451d3880e..e17e737e4 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -15,7 +15,6 @@ package icmp import ( - "encoding/binary" "sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -85,6 +84,7 @@ type endpoint struct { // NIC. regNICID tcpip.NICID route stack.Route `state:"manual"` + ttl uint8 } func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { @@ -105,7 +105,7 @@ func (e *endpoint) Close() { e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite switch e.state { case stateBound, stateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e, 0 /* bindToDevice */) } // Close the receive list and drain it. @@ -205,7 +205,7 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // Write writes data to the endpoint's peer. This method does not block // if the data cannot be written. -func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.) if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue @@ -290,17 +290,17 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha } } - v, err := p.Get(p.Size()) + v, err := p.FullPayload() if err != nil { return 0, nil, err } switch e.netProto { case header.IPv4ProtocolNumber: - err = send4(route, e.id.LocalPort, v) + err = send4(route, e.id.LocalPort, v, e.ttl) case header.IPv6ProtocolNumber: - err = send6(route, e.id.LocalPort, v) + err = send6(route, e.id.LocalPort, v, e.ttl) } if err != nil { @@ -315,8 +315,20 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } -// SetSockOpt sets a socket option. Currently not supported. +// SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch o := opt.(type) { + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(o) + e.mu.Unlock() + } + + return nil +} + +// SetSockOptInt sets a socket option. Currently not supported. +func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { return nil } @@ -332,6 +344,18 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { } e.rcvMu.Unlock() return v, nil + case tcpip.SendBufferSizeOption: + e.mu.Lock() + v := e.sndBufSize + e.mu.Unlock() + return v, nil + + case tcpip.ReceiveBufferSizeOption: + e.rcvMu.Lock() + v := e.rcvBufSizeMax + e.rcvMu.Unlock() + return v, nil + } return -1, tcpip.ErrUnknownProtocolOption } @@ -342,40 +366,33 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case tcpip.ErrorOption: return nil - case *tcpip.SendBufferSizeOption: - e.mu.Lock() - *o = tcpip.SendBufferSizeOption(e.sndBufSize) - e.mu.Unlock() + case *tcpip.KeepaliveEnabledOption: + *o = 0 return nil - case *tcpip.ReceiveBufferSizeOption: + case *tcpip.TTLOption: e.rcvMu.Lock() - *o = tcpip.ReceiveBufferSizeOption(e.rcvBufSizeMax) + *o = tcpip.TTLOption(e.ttl) e.rcvMu.Unlock() return nil - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - default: return tcpip.ErrUnknownProtocolOption } } -func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { +func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { if len(data) < header.ICMPv4MinimumSize { return tcpip.ErrInvalidEndpointState } - // Set the ident to the user-specified port. Sequence number should - // already be set by the user. - binary.BigEndian.PutUint16(data[header.ICMPv4PayloadOffset:], ident) - hdr := buffer.NewPrependable(header.ICMPv4MinimumSize + int(r.MaxHeaderLength())) icmpv4 := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) copy(icmpv4, data) + // Set the ident to the user-specified port. Sequence number should + // already be set by the user. + icmpv4.SetIdent(ident) data = data[header.ICMPv4MinimumSize:] // Linux performs these basic checks. @@ -386,22 +403,21 @@ func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { icmpv4.SetChecksum(0) icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0))) - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, r.DefaultTTL()) + return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */) } -func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { +func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { if len(data) < header.ICMPv6EchoMinimumSize { return tcpip.ErrInvalidEndpointState } - // Set the ident. Sequence number is provided by the user. - binary.BigEndian.PutUint16(data[header.ICMPv6MinimumSize:], ident) - - hdr := buffer.NewPrependable(header.ICMPv6EchoMinimumSize + int(r.MaxHeaderLength())) + hdr := buffer.NewPrependable(header.ICMPv6MinimumSize + int(r.MaxHeaderLength())) - icmpv6 := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) + icmpv6 := header.ICMPv6(hdr.Prepend(header.ICMPv6MinimumSize)) copy(icmpv6, data) - data = data[header.ICMPv6EchoMinimumSize:] + // Set the ident. Sequence number is provided by the user. + icmpv6.SetIdent(ident) + data = data[header.ICMPv6MinimumSize:] if icmpv6.Type() != header.ICMPv6EchoRequest || icmpv6.Code() != 0 { return tcpip.ErrInvalidEndpointState @@ -410,7 +426,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { icmpv6.SetChecksum(0) icmpv6.SetChecksum(^header.Checksum(icmpv6, header.Checksum(data, 0))) - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, r.DefaultTTL()) + return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { @@ -541,14 +557,14 @@ func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.Networ if id.LocalPort != 0 { // The endpoint already has a local port, just attempt to // register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false /* reuse */, 0 /* bindToDevice */) return id, err } // We need to find a port for the endpoint. _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { id.LocalPort = p - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false /* reuse */, 0 /* bindtodevice */) switch err { case nil: return true, nil diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index 7fdba5d56..bfb16f7c3 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -14,16 +14,14 @@ // Package icmp contains the implementation of the ICMP and IPv6-ICMP transport // protocols for use in ping. To use it in the networking stack, this package -// must be added to the project, and -// activated on the stack by passing icmp.ProtocolName (or "icmp") and/or -// icmp.ProtocolName6 (or "icmp6") as one of the transport protocols when -// calling stack.New(). Then endpoints can be created by passing +// must be added to the project, and activated on the stack by passing +// icmp.NewProtocol4() and/or icmp.NewProtocol6() as one of the transport +// protocols when calling stack.New(). Then endpoints can be created by passing // icmp.ProtocolNumber or icmp.ProtocolNumber6 as the transport protocol number // when calling Stack.NewEndpoint(). package icmp import ( - "encoding/binary" "fmt" "gvisor.dev/gvisor/pkg/tcpip" @@ -35,15 +33,9 @@ import ( ) const ( - // ProtocolName4 is the string representation of the icmp protocol name. - ProtocolName4 = "icmp4" - // ProtocolNumber4 is the ICMP protocol number. ProtocolNumber4 = header.ICMPv4ProtocolNumber - // ProtocolName6 is the string representation of the icmp protocol name. - ProtocolName6 = "icmp6" - // ProtocolNumber6 is the IPv6-ICMP protocol number. ProtocolNumber6 = header.ICMPv6ProtocolNumber ) @@ -92,7 +84,7 @@ func (p *protocol) MinimumPacketSize() int { case ProtocolNumber4: return header.ICMPv4MinimumSize case ProtocolNumber6: - return header.ICMPv6EchoMinimumSize + return header.ICMPv6MinimumSize } panic(fmt.Sprint("unknown protocol number: ", p.number)) } @@ -101,16 +93,18 @@ func (p *protocol) MinimumPacketSize() int { func (p *protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { switch p.number { case ProtocolNumber4: - return 0, binary.BigEndian.Uint16(v[header.ICMPv4PayloadOffset:]), nil + hdr := header.ICMPv4(v) + return 0, hdr.Ident(), nil case ProtocolNumber6: - return 0, binary.BigEndian.Uint16(v[header.ICMPv6MinimumSize:]), nil + hdr := header.ICMPv6(v) + return 0, hdr.Ident(), nil } panic(fmt.Sprint("unknown protocol number: ", p.number)) } // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, buffer.VectorisedView) bool { +func (p *protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, buffer.View, buffer.VectorisedView) bool { return true } @@ -124,12 +118,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } -func init() { - stack.RegisterTransportProtocolFactory(ProtocolName4, func() stack.TransportProtocol { - return &protocol{ProtocolNumber4} - }) +// NewProtocol4 returns an ICMPv4 transport protocol. +func NewProtocol4() stack.TransportProtocol { + return &protocol{ProtocolNumber4} +} - stack.RegisterTransportProtocolFactory(ProtocolName6, func() stack.TransportProtocol { - return &protocol{ProtocolNumber6} - }) +// NewProtocol6 returns an ICMPv6 transport protocol. +func NewProtocol6() stack.TransportProtocol { + return &protocol{ProtocolNumber6} } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 13e17e2a6..8ae132efa 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -207,7 +207,7 @@ func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMes } // Write implements tcpip.Endpoint.Write. -func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. This also means that MSG_EOR is a no-op. if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue @@ -220,9 +220,8 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (int64 return 0, nil, tcpip.ErrInvalidEndpointState } - payloadBytes, err := payload.Get(payload.Size()) + payloadBytes, err := p.FullPayload() if err != nil { - ep.mu.RUnlock() return 0, nil, err } @@ -230,7 +229,7 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (int64 // destination address, route using that address. if !ep.associated { ip := header.IPv4(payloadBytes) - if !ip.IsValid(payload.Size()) { + if !ip.IsValid(len(payloadBytes)) { ep.mu.RUnlock() return 0, nil, tcpip.ErrInvalidOptionValue } @@ -333,7 +332,7 @@ func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, break } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, route.DefaultTTL()); err != nil { + if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, 0, true /* useDefaultTTL */); err != nil { return 0, nil, err } @@ -493,6 +492,11 @@ func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } +// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. +func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { + return tcpip.ErrUnknownProtocolOption +} + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { switch opt { @@ -505,6 +509,19 @@ func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { } ep.rcvMu.Unlock() return v, nil + + case tcpip.SendBufferSizeOption: + ep.mu.Lock() + v := ep.sndBufSize + ep.mu.Unlock() + return v, nil + + case tcpip.ReceiveBufferSizeOption: + ep.rcvMu.Lock() + v := ep.rcvBufSizeMax + ep.rcvMu.Unlock() + return v, nil + } return -1, tcpip.ErrUnknownProtocolOption @@ -516,18 +533,6 @@ func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case tcpip.ErrorOption: return nil - case *tcpip.SendBufferSizeOption: - ep.mu.Lock() - *o = tcpip.SendBufferSizeOption(ep.sndBufSize) - ep.mu.Unlock() - return nil - - case *tcpip.ReceiveBufferSizeOption: - ep.rcvMu.Lock() - *o = tcpip.ReceiveBufferSizeOption(ep.rcvBufSizeMax) - ep.rcvMu.Unlock() - return nil - case *tcpip.KeepaliveEnabledOption: *o = 0 return nil diff --git a/pkg/tcpip/transport/raw/protocol.go b/pkg/tcpip/transport/raw/protocol.go index 783c21e6b..a2512d666 100644 --- a/pkg/tcpip/transport/raw/protocol.go +++ b/pkg/tcpip/transport/raw/protocol.go @@ -20,13 +20,10 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -type factory struct{} +// EndpointFactory implements stack.UnassociatedEndpointFactory. +type EndpointFactory struct{} // NewUnassociatedRawEndpoint implements stack.UnassociatedEndpointFactory. -func (factory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (EndpointFactory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, transProto, waiterQueue, false /* associated */) } - -func init() { - stack.RegisterUnassociatedFactory(factory{}) -} diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 1ee1a53f8..a42e1f4a2 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -1,7 +1,9 @@ +load("@io_bazel_rules_go//go:def.bzl", "go_test") + package(licenses = ["notice"]) load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") go_template_instance( name = "tcp_segment_list", @@ -47,6 +49,7 @@ go_library( "//pkg/sleep", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/seqnum", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index e9c5099ea..b8b4bcee8 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -143,6 +143,15 @@ func decSynRcvdCount() { synRcvdCount.Unlock() } +// synCookiesInUse() returns true if the synRcvdCount is greater than +// SynRcvdCountThreshold. +func synCookiesInUse() bool { + synRcvdCount.Lock() + v := synRcvdCount.value + synRcvdCount.Unlock() + return v >= SynRcvdCountThreshold +} + // newListenContext creates a new listen context. func newListenContext(stk *stack.Stack, listenEP *endpoint, rcvWnd seqnum.Size, v6only bool, netProto tcpip.NetworkProtocolNumber) *listenContext { l := &listenContext{ @@ -233,7 +242,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort); err != nil { + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort, n.bindToDevice); err != nil { n.Close() return nil, err } @@ -430,7 +439,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { TSEcr: opts.TSVal, MSS: uint16(mss), } - sendSynTCP(&s.route, s.id, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts) + sendSynTCP(&s.route, s.id, e.ttl, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts) e.stack.Stats().TCP.ListenOverflowSynCookieSent.Increment() } @@ -446,6 +455,27 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { return } + if !synCookiesInUse() { + // Send a reset as this is an ACK for which there is no + // half open connections and we are not using cookies + // yet. + // + // The only time we should reach here when a connection + // was opened and closed really quickly and a delayed + // ACK was received from the sender. + replyWithReset(s) + return + } + + // Since SYN cookies are in use this is potentially an ACK to a + // SYN-ACK we sent but don't have a half open connection state + // as cookies are being used to protect against a potential SYN + // flood. In such cases validate the cookie and if valid create + // a fully connected endpoint and deliver to the accept queue. + // + // If not, silently drop the ACK to avoid leaking information + // when under a potential syn flood attack. + // // Validate the cookie. data, ok := ctx.isCookieValid(s.id, s.ackNumber-1, s.sequenceNumber-1) if !ok || int(data) >= len(mssTable) { diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 00d2ae524..1d6e7f5f3 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -238,6 +238,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { h.state = handshakeSynRcvd h.ep.mu.Lock() h.ep.state = StateSynRecv + ttl := h.ep.ttl h.ep.mu.Unlock() synOpts := header.TCPSynOptions{ WS: int(h.effectiveRcvWndScale()), @@ -251,7 +252,10 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { SACKPermitted: rcvSynOpts.SACKPermitted, MSS: h.ep.amss, } - sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + if ttl == 0 { + ttl = s.route.DefaultTTL() + } + sendSynTCP(&s.route, h.ep.id, ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil } @@ -296,7 +300,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { SACKPermitted: h.ep.sackPermitted, MSS: h.ep.amss, } - sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + sendSynTCP(&s.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil } @@ -460,7 +464,7 @@ func (h *handshake) execute() *tcpip.Error { synOpts.WS = -1 } } - sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + sendSynTCP(&h.ep.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) for h.state != handshakeCompleted { switch index, _ := s.Fetch(true); index { case wakerForResend: @@ -469,7 +473,7 @@ func (h *handshake) execute() *tcpip.Error { return tcpip.ErrTimeout } rt.Reset(timeOut) - sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + sendSynTCP(&h.ep.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) case wakerForNotification: n := h.ep.fetchNotifications() @@ -579,9 +583,9 @@ func makeSynOptions(opts header.TCPSynOptions) []byte { return options[:offset] } -func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error { +func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, ttl uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error { options := makeSynOptions(opts) - err := sendTCP(r, id, buffer.VectorisedView{}, r.DefaultTTL(), flags, seq, ack, rcvWnd, options, nil) + err := sendTCP(r, id, buffer.VectorisedView{}, ttl, flags, seq, ack, rcvWnd, options, nil) putOptions(options) return err } @@ -629,7 +633,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise r.Stats().TCP.ResetsSent.Increment() } - return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl) + return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */) } // makeOptions makes an options slice. @@ -678,7 +682,7 @@ func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqn sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) - err := sendTCP(&e.route, e.id, data, e.route.DefaultTTL(), flags, seq, ack, rcvWnd, options, e.gso) + err := sendTCP(&e.route, e.id, data, e.ttl, flags, seq, ack, rcvWnd, options, e.gso) putOptions(options) return err } @@ -720,13 +724,18 @@ func (e *endpoint) handleClose() *tcpip.Error { return nil } -// resetConnectionLocked sends a RST segment and puts the endpoint in an error -// state with the given error code. This method must only be called from the -// protocol goroutine. +// resetConnectionLocked puts the endpoint in an error state with the given +// error code and sends a RST if and only if the error is not ErrConnectionReset +// indicating that the connection is being reset due to receiving a RST. This +// method must only be called from the protocol goroutine. func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { - e.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck|header.TCPFlagRst, e.snd.sndUna, e.rcv.rcvNxt, 0) + // Only send a reset if the connection is being aborted for a reason + // other than receiving a reset. e.state = StateError e.hardError = err + if err != tcpip.ErrConnectionReset { + e.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck|header.TCPFlagRst, e.snd.sndUna, e.rcv.rcvNxt, 0) + } } // completeWorkerLocked is called by the worker goroutine when it's about to @@ -806,7 +815,7 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { if e.keepalive.unacked >= e.keepalive.count { e.keepalive.Unlock() - return tcpip.ErrConnectionReset + return tcpip.ErrTimeout } // RFC1122 4.2.3.6: TCP keepalive is a dataless ACK with @@ -1068,6 +1077,10 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.workMu.Lock() if err := funcs[v].f(); err != nil { e.mu.Lock() + // Ensure we release all endpoint registration and route + // references as the connection is now in an error + // state. + e.workerCleanup = true e.resetConnectionLocked(err) // Lock released below. epilogue() diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index ac927569a..83d92b3e1 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -15,6 +15,7 @@ package tcp import ( + "encoding/binary" "fmt" "math" "strings" @@ -26,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/seqnum" @@ -226,6 +228,7 @@ type endpoint struct { isRegistered bool boundNICID tcpip.NICID `state:"manual"` route stack.Route `state:"manual"` + ttl uint8 v6only bool isConnectNotified bool // TCP should never broadcast but Linux nevertheless supports enabling/ @@ -280,6 +283,9 @@ type endpoint struct { // reusePort is set to true if SO_REUSEPORT is enabled. reusePort bool + // bindToDevice is set to the NIC on which to bind or disabled if 0. + bindToDevice tcpip.NICID + // delay enables Nagle's algorithm. // // delay is a boolean (0 is false) and must be accessed atomically. @@ -564,11 +570,11 @@ func (e *endpoint) Close() { // in Listen() when trying to register. if e.state == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -625,12 +631,12 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.isRegistered = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -806,7 +812,7 @@ func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) { } // Write writes data to the endpoint's peer. -func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // Linux completely ignores any address passed to sendto(2) for TCP sockets // (without the MSG_FASTOPEN flag). Corking is unimplemented, so opts.More // and opts.EndOfRecord are also ignored. @@ -821,47 +827,52 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha return 0, nil, err } - e.sndBufMu.Unlock() - e.mu.RUnlock() - - // Nothing to do if the buffer is empty. - if p.Size() == 0 { - return 0, nil, nil + // We can release locks while copying data. + // + // This is not possible if atomic is set, because we can't allow the + // available buffer space to be consumed by some other caller while we + // are copying data in. + if !opts.Atomic { + e.sndBufMu.Unlock() + e.mu.RUnlock() } - // Copy in memory without holding sndBufMu so that worker goroutine can - // make progress independent of this operation. - v, perr := p.Get(avail) - if perr != nil { + // Fetch data. + v, perr := p.Payload(avail) + if perr != nil || len(v) == 0 { + if opts.Atomic { // See above. + e.sndBufMu.Unlock() + e.mu.RUnlock() + } + // Note that perr may be nil if len(v) == 0. return 0, nil, perr } - e.mu.RLock() - e.sndBufMu.Lock() + if !opts.Atomic { // See above. + e.mu.RLock() + e.sndBufMu.Lock() - // Because we released the lock before copying, check state again - // to make sure the endpoint is still in a valid state for a - // write. - avail, err = e.isEndpointWritableLocked() - if err != nil { - e.sndBufMu.Unlock() - e.mu.RUnlock() - return 0, nil, err - } + // Because we released the lock before copying, check state again + // to make sure the endpoint is still in a valid state for a write. + avail, err = e.isEndpointWritableLocked() + if err != nil { + e.sndBufMu.Unlock() + e.mu.RUnlock() + return 0, nil, err + } - // Discard any excess data copied in due to avail being reduced due to a - // simultaneous write call to the socket. - if avail < len(v) { - v = v[:avail] + // Discard any excess data copied in due to avail being reduced due + // to a simultaneous write call to the socket. + if avail < len(v) { + v = v[:avail] + } } // Add data to the send queue. - l := len(v) s := newSegmentFromView(&e.route, e.id, v) - e.sndBufUsed += l - e.sndBufInQueue += seqnum.Size(l) + e.sndBufUsed += len(v) + e.sndBufInQueue += seqnum.Size(len(v)) e.sndQueue.PushBack(s) - e.sndBufMu.Unlock() // Release the endpoint lock to prevent deadlocks due to lock // order inversion when acquiring workMu. @@ -875,7 +886,8 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha // Let the protocol goroutine do the work. e.sndWaker.Assert() } - return int64(l), nil, nil + + return int64(len(v)), nil, nil } // Peek reads data without consuming it from the endpoint. @@ -946,62 +958,9 @@ func (e *endpoint) zeroReceiveWindow(scale uint8) bool { return ((e.rcvBufSize - e.rcvBufUsed) >> scale) == 0 } -// SetSockOpt sets a socket option. -func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { - case tcpip.DelayOption: - if v == 0 { - atomic.StoreUint32(&e.delay, 0) - - // Handle delayed data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.delay, 1) - } - return nil - - case tcpip.CorkOption: - if v == 0 { - atomic.StoreUint32(&e.cork, 0) - - // Handle the corked data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.cork, 1) - } - return nil - - case tcpip.ReuseAddressOption: - e.mu.Lock() - e.reuseAddr = v != 0 - e.mu.Unlock() - return nil - - case tcpip.ReusePortOption: - e.mu.Lock() - e.reusePort = v != 0 - e.mu.Unlock() - return nil - - case tcpip.QuickAckOption: - if v == 0 { - atomic.StoreUint32(&e.slowAck, 1) - } else { - atomic.StoreUint32(&e.slowAck, 0) - } - return nil - - case tcpip.MaxSegOption: - userMSS := v - if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { - return tcpip.ErrInvalidOptionValue - } - e.mu.Lock() - e.userMSS = int(userMSS) - e.mu.Unlock() - e.notifyProtocolGoroutine(notifyMSSChanged) - return nil - +// SetSockOptInt sets a socket option. +func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { + switch opt { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. @@ -1065,6 +1024,82 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.sndBufMu.Unlock() return nil + default: + return nil + } +} + +// SetSockOpt sets a socket option. +func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch v := opt.(type) { + case tcpip.DelayOption: + if v == 0 { + atomic.StoreUint32(&e.delay, 0) + + // Handle delayed data. + e.sndWaker.Assert() + } else { + atomic.StoreUint32(&e.delay, 1) + } + return nil + + case tcpip.CorkOption: + if v == 0 { + atomic.StoreUint32(&e.cork, 0) + + // Handle the corked data. + e.sndWaker.Assert() + } else { + atomic.StoreUint32(&e.cork, 1) + } + return nil + + case tcpip.ReuseAddressOption: + e.mu.Lock() + e.reuseAddr = v != 0 + e.mu.Unlock() + return nil + + case tcpip.ReusePortOption: + e.mu.Lock() + e.reusePort = v != 0 + e.mu.Unlock() + return nil + + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicid, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicid + return nil + } + } + return tcpip.ErrUnknownDevice + + case tcpip.QuickAckOption: + if v == 0 { + atomic.StoreUint32(&e.slowAck, 1) + } else { + atomic.StoreUint32(&e.slowAck, 0) + } + return nil + + case tcpip.MaxSegOption: + userMSS := v + if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { + return tcpip.ErrInvalidOptionValue + } + e.mu.Lock() + e.userMSS = int(userMSS) + e.mu.Unlock() + e.notifyProtocolGoroutine(notifyMSSChanged) + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.netProto != header.IPv6ProtocolNumber { @@ -1082,6 +1117,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.v6only = v != 0 return nil + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + return nil + case tcpip.KeepaliveEnabledOption: e.keepalive.Lock() e.keepalive.enabled = v != 0 @@ -1176,6 +1217,18 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: return e.readyReceiveSize() + case tcpip.SendBufferSizeOption: + e.sndBufMu.Lock() + v := e.sndBufSize + e.sndBufMu.Unlock() + return v, nil + + case tcpip.ReceiveBufferSizeOption: + e.rcvListMu.Lock() + v := e.rcvBufSize + e.rcvListMu.Unlock() + return v, nil + } return -1, tcpip.ErrUnknownProtocolOption } @@ -1198,18 +1251,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = header.TCPDefaultMSS return nil - case *tcpip.SendBufferSizeOption: - e.sndBufMu.Lock() - *o = tcpip.SendBufferSizeOption(e.sndBufSize) - e.sndBufMu.Unlock() - return nil - - case *tcpip.ReceiveBufferSizeOption: - e.rcvListMu.Lock() - *o = tcpip.ReceiveBufferSizeOption(e.rcvBufSize) - e.rcvListMu.Unlock() - return nil - case *tcpip.DelayOption: *o = 0 if v := atomic.LoadUint32(&e.delay); v != 0 { @@ -1246,6 +1287,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = "" + return nil + case *tcpip.QuickAckOption: *o = 1 if v := atomic.LoadUint32(&e.slowAck); v != 0 { @@ -1269,6 +1320,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.TTLOption: + e.mu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.mu.Unlock() + return nil + case *tcpip.TCPInfoOption: *o = tcpip.TCPInfoOption{} e.mu.RLock() @@ -1452,7 +1509,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er if e.id.LocalPort != 0 { // The endpoint is bound to a port, attempt to register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1462,17 +1519,32 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // address/port for both local and remote (otherwise this // endpoint would be trying to connect to itself). sameAddr := e.id.LocalAddress == e.id.RemoteAddress - if _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { + + // Calculate a port offset based on the destination IP/port and + // src IP to ensure that for a given tuple (srcIP, destIP, + // destPort) the offset used as a starting point is the same to + // ensure that we can cycle through the port space effectively. + h := jenkins.Sum32(e.stack.PortSeed()) + h.Write([]byte(e.id.LocalAddress)) + h.Write([]byte(e.id.RemoteAddress)) + portBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(portBuf, e.id.RemotePort) + h.Write(portBuf) + portOffset := h.Sum32() + + if _, err := e.stack.PickEphemeralPortStable(portOffset, func(p uint16) (bool, *tcpip.Error) { if sameAddr && p == e.id.RemotePort { return false, nil } - if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false) { + // reusePort is false below because connect cannot reuse a port even if + // reusePort was set. + if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false /* reusePort */, e.bindToDevice) { return false, nil } id := e.id id.LocalPort = p - switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) { + switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { case nil: e.id = id return true, nil @@ -1490,7 +1562,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // before Connect: in such a case we don't want to hold on to // reservations anymore. if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -1634,7 +1706,7 @@ func (e *endpoint) Listen(backlog int) (err *tcpip.Error) { } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice); err != nil { return err } @@ -1715,7 +1787,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1725,16 +1797,16 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { e.id.LocalPort = port // Any failures beyond this point must remove the port registration. - defer func() { + defer func(bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice) e.isPortReserved = false e.effectiveNetProtos = nil e.id.LocalPort = 0 e.id.LocalAddress = "" e.boundNICID = 0 } - }() + }(e.bindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index ee04dcfcc..d5d8ab96a 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -14,7 +14,7 @@ // Package tcp contains the implementation of the TCP transport protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing tcp.ProtocolName (or "tcp") as one of the +// activated on the stack by passing tcp.NewProtocol() as one of the // transport protocols when calling stack.New(). Then endpoints can be created // by passing tcp.ProtocolNumber as the transport protocol number when calling // Stack.NewEndpoint(). @@ -34,9 +34,6 @@ import ( ) const ( - // ProtocolName is the string representation of the tcp protocol name. - ProtocolName = "tcp" - // ProtocolNumber is the tcp protocol number. ProtocolNumber = header.TCPProtocolNumber @@ -129,7 +126,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // a reset is sent in response to any incoming segment except another reset. In // particular, SYNs addressed to a non-existent connection are rejected by this // means." -func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, vv buffer.VectorisedView) bool { +func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool { s := newSegment(r, id, vv) defer s.decRef() @@ -254,13 +251,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { } } -func init() { - stack.RegisterTransportProtocolFactory(ProtocolName, func() stack.TransportProtocol { - return &protocol{ - sendBufferSize: SendBufferSizeOption{MinBufferSize, DefaultSendBufferSize, MaxBufferSize}, - recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, - congestionControl: ccReno, - availableCongestionControl: []string{ccReno, ccCubic}, - } - }) +// NewProtocol returns a TCP transport protocol. +func NewProtocol() stack.TransportProtocol { + return &protocol{ + sendBufferSize: SendBufferSizeOption{MinBufferSize, DefaultSendBufferSize, MaxBufferSize}, + recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, + congestionControl: ccReno, + availableCongestionControl: []string{ccReno, ccCubic}, + } } diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 1f9b1e0ef..735edfe55 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -664,7 +664,14 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se segEnd = seg.sequenceNumber.Add(1) // Transition to FIN-WAIT1 state since we're initiating an active close. s.ep.mu.Lock() - s.ep.state = StateFinWait1 + switch s.ep.state { + case StateCloseWait: + // We've already received a FIN and are now sending our own. The + // sender is now awaiting a final ACK for this FIN. + s.ep.state = StateLastAck + default: + s.ep.state = StateFinWait1 + } s.ep.mu.Unlock() } else { // We're sending a non-FIN segment. diff --git a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go index 272bbcdbd..9fa97528b 100644 --- a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go +++ b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go @@ -38,7 +38,7 @@ func TestFastRecovery(t *testing.T) { c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) const iterations = 7 data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) @@ -190,7 +190,7 @@ func TestExponentialIncreaseDuringSlowStart(t *testing.T) { c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) const iterations = 7 data := buffer.NewView(maxPayload * (tcp.InitialCwnd << (iterations + 1))) @@ -232,7 +232,7 @@ func TestCongestionAvoidance(t *testing.T) { c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) const iterations = 7 data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) @@ -336,7 +336,7 @@ func TestCubicCongestionAvoidance(t *testing.T) { enableCUBIC(t, c) - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) const iterations = 7 data := buffer.NewView(2 * maxPayload * (tcp.InitialCwnd << (iterations + 1))) @@ -445,7 +445,7 @@ func TestRetransmit(t *testing.T) { c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) const iterations = 7 data := buffer.NewView(maxPayload * (tcp.InitialCwnd << (iterations + 1))) diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index f79b8ec5f..a86123829 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -84,7 +84,7 @@ func TestConnectIncrementActiveConnection(t *testing.T) { stats := c.Stack().Stats() want := stats.TCP.ActiveConnectionOpenings.Value() + 1 - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) if got := stats.TCP.ActiveConnectionOpenings.Value(); got != want { t.Errorf("got stats.TCP.ActtiveConnectionOpenings.Value() = %v, want = %v", got, want) } @@ -97,7 +97,7 @@ func TestConnectDoesNotIncrementFailedConnectionAttempts(t *testing.T) { stats := c.Stack().Stats() want := stats.TCP.FailedConnectionAttempts.Value() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) if got := stats.TCP.FailedConnectionAttempts.Value(); got != want { t.Errorf("got stats.TCP.FailedConnectionOpenings.Value() = %v, want = %v", got, want) } @@ -131,7 +131,7 @@ func TestTCPSegmentsSentIncrement(t *testing.T) { stats := c.Stack().Stats() // SYN and ACK want := stats.TCP.SegmentsSent.Value() + 2 - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) if got := stats.TCP.SegmentsSent.Value(); got != want { t.Errorf("got stats.TCP.SegmentsSent.Value() = %v, want = %v", got, want) @@ -190,21 +190,122 @@ func TestTCPResetsSentIncrement(t *testing.T) { } } +// TestTCPResetSentForACKWhenNotUsingSynCookies checks that the stack generates +// a RST if an ACK is received on the listening socket for which there is no +// active handshake in progress and we are not using SYN cookies. +func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %v", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %v", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %v", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %v", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + c.GetPacket() + + // Now resend the same ACK, this ACK should generate a RST as there + // should be no endpoint in SYN-RCVD state and we are not using + // syn-cookies yet. The reason we send the same ACK is we need a valid + // cookie(IRS) generated by the netstack without which the ACK will be + // rejected. + c.SendPacket(nil, ackHeaders) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) +} + func TestTCPResetsReceivedIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() stats := c.Stack().Stats() want := stats.TCP.ResetsReceived.Value() + 1 - ackNum := seqnum.Value(789) + iss := seqnum.Value(789) rcvWnd := seqnum.Size(30000) - c.CreateConnected(ackNum, rcvWnd, nil) + c.CreateConnected(iss, rcvWnd, -1 /* epRcvBuf */) c.SendPacket(nil, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, - SeqNum: c.IRS.Add(2), - AckNum: ackNum.Add(2), + SeqNum: iss.Add(1), + AckNum: c.IRS.Add(1), RcvWnd: rcvWnd, Flags: header.TCPFlagRst, }) @@ -214,18 +315,43 @@ func TestTCPResetsReceivedIncrement(t *testing.T) { } } +func TestTCPResetsDoNotGenerateResets(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + stats := c.Stack().Stats() + want := stats.TCP.ResetsReceived.Value() + 1 + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.CreateConnected(iss, rcvWnd, -1 /* epRcvBuf */) + + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + SeqNum: iss.Add(1), + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + Flags: header.TCPFlagRst, + }) + + if got := stats.TCP.ResetsReceived.Value(); got != want { + t.Errorf("got stats.TCP.ResetsReceived.Value() = %v, want = %v", got, want) + } + c.CheckNoPacketTimeout("got an unexpected packet", 100*time.Millisecond) +} + func TestActiveHandshake(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) } func TestNonBlockingClose(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) ep := c.EP c.EP = nil @@ -241,7 +367,7 @@ func TestConnectResetAfterClose(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) ep := c.EP c.EP = nil @@ -291,7 +417,7 @@ func TestSimpleReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -339,11 +465,71 @@ func TestSimpleReceive(t *testing.T) { ) } +func TestConnectBindToDevice(t *testing.T) { + for _, test := range []struct { + name string + device string + want tcp.EndpointState + }{ + {"RightDevice", "nic1", tcp.StateEstablished}, + {"WrongDevice", "nic2", tcp.StateSynSent}, + {"AnyDevice", "", tcp.StateEstablished}, + } { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + bindToDevice := tcpip.BindToDeviceOption(test.device) + c.EP.SetSockOpt(bindToDevice) + // Start connection attempt. + waitEntry, _ := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagSyn | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + + c.GetPacket() + if got, want := tcp.EndpointState(c.EP.State()), test.want; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + }) + } +} + func TestOutOfOrderReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -431,8 +617,7 @@ func TestOutOfOrderFlood(t *testing.T) { defer c.Cleanup() // Create a new connection with initial window size of 10. - opt := tcpip.ReceiveBufferSizeOption(10) - c.CreateConnected(789, 30000, &opt) + c.CreateConnected(789, 30000, 10) if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) @@ -505,7 +690,7 @@ func TestRstOnCloseWithUnreadData(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -574,7 +759,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -659,7 +844,7 @@ func TestShutdownRead(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) @@ -678,8 +863,7 @@ func TestFullWindowReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - opt := tcpip.ReceiveBufferSizeOption(10) - c.CreateConnected(789, 30000, &opt) + c.CreateConnected(789, 30000, 10) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -746,11 +930,9 @@ func TestNoWindowShrinking(t *testing.T) { defer c.Cleanup() // Start off with a window size of 10, then shrink it to 5. - opt := tcpip.ReceiveBufferSizeOption(10) - c.CreateConnected(789, 30000, &opt) + c.CreateConnected(789, 30000, 10) - opt = 5 - if err := c.EP.SetSockOpt(opt); err != nil { + if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 5); err != nil { t.Fatalf("SetSockOpt failed: %v", err) } @@ -850,7 +1032,7 @@ func TestSimpleSend(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) data := []byte{1, 2, 3} view := buffer.NewView(len(data)) @@ -891,7 +1073,7 @@ func TestZeroWindowSend(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 0, nil) + c.CreateConnected(789, 0, -1 /* epRcvBuf */) data := []byte{1, 2, 3} view := buffer.NewView(len(data)) @@ -949,8 +1131,7 @@ func TestScaledWindowConnect(t *testing.T) { defer c.Cleanup() // Set the window size greater than the maximum non-scaled window. - opt := tcpip.ReceiveBufferSizeOption(65535 * 3) - c.CreateConnectedWithRawOptions(789, 30000, &opt, []byte{ + c.CreateConnectedWithRawOptions(789, 30000, 65535*3, []byte{ header.TCPOptionWS, 3, 0, header.TCPOptionNOP, }) @@ -984,8 +1165,7 @@ func TestNonScaledWindowConnect(t *testing.T) { defer c.Cleanup() // Set the window size greater than the maximum non-scaled window. - opt := tcpip.ReceiveBufferSizeOption(65535 * 3) - c.CreateConnected(789, 30000, &opt) + c.CreateConnected(789, 30000, 65535*3) data := []byte{1, 2, 3} view := buffer.NewView(len(data)) @@ -1025,7 +1205,7 @@ func TestScaledWindowAccept(t *testing.T) { defer ep.Close() // Set the window size greater than the maximum non-scaled window. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(65535 * 3)); err != nil { + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 65535*3); err != nil { t.Fatalf("SetSockOpt failed failed: %v", err) } @@ -1098,7 +1278,7 @@ func TestNonScaledWindowAccept(t *testing.T) { defer ep.Close() // Set the window size greater than the maximum non-scaled window. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(65535 * 3)); err != nil { + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 65535*3); err != nil { t.Fatalf("SetSockOpt failed failed: %v", err) } @@ -1167,8 +1347,7 @@ func TestZeroScaledWindowReceive(t *testing.T) { // Set the window size such that a window scale of 4 will be used. const wnd = 65535 * 10 const ws = uint32(4) - opt := tcpip.ReceiveBufferSizeOption(wnd) - c.CreateConnectedWithRawOptions(789, 30000, &opt, []byte{ + c.CreateConnectedWithRawOptions(789, 30000, wnd, []byte{ header.TCPOptionWS, 3, 0, header.TCPOptionNOP, }) @@ -1273,7 +1452,7 @@ func TestSegmentMerging(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Prevent the endpoint from processing packets. test.stop(c.EP) @@ -1323,7 +1502,7 @@ func TestDelay(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) c.EP.SetSockOpt(tcpip.DelayOption(1)) @@ -1371,7 +1550,7 @@ func TestUndelay(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) c.EP.SetSockOpt(tcpip.DelayOption(1)) @@ -1453,7 +1632,7 @@ func TestMSSNotDelayed(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnectedWithRawOptions(789, 30000, nil, []byte{ + c.CreateConnectedWithRawOptions(789, 30000, -1 /* epRcvBuf */, []byte{ header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), }) @@ -1569,16 +1748,44 @@ func TestSendGreaterThanMTU(t *testing.T) { c := context.New(t, uint32(header.TCPMinimumSize+header.IPv4MinimumSize+maxPayload)) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) testBrokenUpWrite(t, c, maxPayload) } +func TestSetTTL(t *testing.T) { + for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} { + t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) { + c := context.New(t, 65535) + defer c.Cleanup() + + var err *tcpip.Error + c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed: %v", err) + } + + if err := c.EP.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { + t.Fatalf("SetSockOpt failed: %v", err) + } + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + + checker.IPv4(t, b, checker.TTL(wantTTL)) + }) + } +} + func TestActiveSendMSSLessThanMTU(t *testing.T) { const maxPayload = 100 c := context.New(t, 65535) defer c.Cleanup() - c.CreateConnectedWithRawOptions(789, 30000, nil, []byte{ + c.CreateConnectedWithRawOptions(789, 30000, -1 /* epRcvBuf */, []byte{ header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), }) testBrokenUpWrite(t, c, maxPayload) @@ -1601,7 +1808,7 @@ func TestPassiveSendMSSLessThanMTU(t *testing.T) { // Set the buffer size to a deterministic size so that we can check the // window scaling option. const rcvBufferSize = 0x20000 - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(rcvBufferSize)); err != nil { + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { t.Fatalf("SetSockOpt failed failed: %v", err) } @@ -1745,7 +1952,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { // window scaling option. const rcvBufferSize = 0x20000 const wndScale = 2 - if err := c.EP.SetSockOpt(tcpip.ReceiveBufferSizeOption(rcvBufferSize)); err != nil { + if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { t.Fatalf("SetSockOpt failed failed: %v", err) } @@ -1847,7 +2054,7 @@ func TestReceiveOnResetConnection(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Send RST segment. c.SendPacket(nil, &context.Headers{ @@ -1884,7 +2091,7 @@ func TestSendOnResetConnection(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Send RST segment. c.SendPacket(nil, &context.Headers{ @@ -1909,7 +2116,7 @@ func TestFinImmediately(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Shutdown immediately, check that we get a FIN. if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { @@ -1952,7 +2159,7 @@ func TestFinRetransmit(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Shutdown immediately, check that we get a FIN. if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { @@ -2006,7 +2213,7 @@ func TestFinWithNoPendingData(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Write something out, and have it acknowledged. view := buffer.NewView(10) @@ -2077,7 +2284,7 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Write enough segments to fill the congestion window before ACK'ing // any of them. @@ -2165,7 +2372,7 @@ func TestFinWithPendingData(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Write something out, and acknowledge it to get cwnd to 2. view := buffer.NewView(10) @@ -2251,7 +2458,7 @@ func TestFinWithPartialAck(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Write something out, and acknowledge it to get cwnd to 2. Also send // FIN from the test side. @@ -2383,7 +2590,7 @@ func scaledSendWindow(t *testing.T, scale uint8) { defer c.Cleanup() maxPayload := defaultMTU - header.IPv4MinimumSize - header.TCPMinimumSize - c.CreateConnectedWithRawOptions(789, 0, nil, []byte{ + c.CreateConnectedWithRawOptions(789, 0, -1 /* epRcvBuf */, []byte{ header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), header.TCPOptionWS, 3, scale, header.TCPOptionNOP, }) @@ -2433,7 +2640,7 @@ func TestScaledSendWindow(t *testing.T) { func TestReceivedValidSegmentCountIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) stats := c.Stack().Stats() want := stats.TCP.ValidSegmentsReceived.Value() + 1 @@ -2454,7 +2661,7 @@ func TestReceivedValidSegmentCountIncrement(t *testing.T) { func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) stats := c.Stack().Stats() want := stats.TCP.InvalidSegmentsReceived.Value() + 1 vv := c.BuildSegment(nil, &context.Headers{ @@ -2478,7 +2685,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { func TestReceivedIncorrectChecksumIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) stats := c.Stack().Stats() want := stats.TCP.ChecksumErrors.Value() + 1 vv := c.BuildSegment([]byte{0x1, 0x2, 0x3}, &context.Headers{ @@ -2509,7 +2716,7 @@ func TestReceivedSegmentQueuing(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) // Send 200 segments. data := []byte{1, 2, 3} @@ -2555,7 +2762,7 @@ func TestReadAfterClosedState(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -2730,8 +2937,8 @@ func TestReusePort(t *testing.T) { func checkRecvBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { t.Helper() - var s tcpip.ReceiveBufferSizeOption - if err := ep.GetSockOpt(&s); err != nil { + s, err := ep.GetSockOptInt(tcpip.ReceiveBufferSizeOption) + if err != nil { t.Fatalf("GetSockOpt failed: %v", err) } @@ -2743,8 +2950,8 @@ func checkRecvBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { func checkSendBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { t.Helper() - var s tcpip.SendBufferSizeOption - if err := ep.GetSockOpt(&s); err != nil { + s, err := ep.GetSockOptInt(tcpip.SendBufferSizeOption) + if err != nil { t.Fatalf("GetSockOpt failed: %v", err) } @@ -2754,7 +2961,10 @@ func checkSendBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { } func TestDefaultBufferSizes(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) // Check the default values. ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) @@ -2800,7 +3010,10 @@ func TestDefaultBufferSizes(t *testing.T) { } func TestMinMaxBufferSizes(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) // Check the default values. ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) @@ -2819,37 +3032,96 @@ func TestMinMaxBufferSizes(t *testing.T) { } // Set values below the min. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(199)); err != nil { + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 199); err != nil { t.Fatalf("GetSockOpt failed: %v", err) } checkRecvBufferSize(t, ep, 200) - if err := ep.SetSockOpt(tcpip.SendBufferSizeOption(299)); err != nil { + if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 299); err != nil { t.Fatalf("GetSockOpt failed: %v", err) } checkSendBufferSize(t, ep, 300) // Set values above the max. - if err := ep.SetSockOpt(tcpip.ReceiveBufferSizeOption(1 + tcp.DefaultReceiveBufferSize*20)); err != nil { + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 1+tcp.DefaultReceiveBufferSize*20); err != nil { t.Fatalf("GetSockOpt failed: %v", err) } checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20) - if err := ep.SetSockOpt(tcpip.SendBufferSizeOption(1 + tcp.DefaultSendBufferSize*30)); err != nil { + if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 1+tcp.DefaultSendBufferSize*30); err != nil { t.Fatalf("GetSockOpt failed: %v", err) } checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) } +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}}) + + ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) + } + defer ep.Close() + + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } + + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) + } + + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s + } + + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) + } +} + func makeStack() (*stack.Stack, *tcpip.Error) { - s := stack.New([]string{ - ipv4.ProtocolName, - ipv6.ProtocolName, - }, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ + ipv4.NewProtocol(), + ipv6.NewProtocol(), + }, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) id := loopback.New() if testing.Verbose() { @@ -3105,7 +3377,7 @@ func TestPathMTUDiscovery(t *testing.T) { // Create new connection with MSS of 1460. const maxPayload = 1500 - header.TCPMinimumSize - header.IPv4MinimumSize - c.CreateConnectedWithRawOptions(789, 30000, nil, []byte{ + c.CreateConnectedWithRawOptions(789, 30000, -1 /* epRcvBuf */, []byte{ header.TCPOptionMSS, 4, byte(maxPayload / 256), byte(maxPayload % 256), }) @@ -3182,7 +3454,7 @@ func TestTCPEndpointProbe(t *testing.T) { invoked <- struct{}{} }) - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) data := []byte{1, 2, 3} c.SendPacket(data, &context.Headers{ @@ -3356,7 +3628,7 @@ func TestKeepalive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, nil) + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(10 * time.Millisecond)) @@ -3459,8 +3731,8 @@ func TestKeepalive(t *testing.T) { ), ) - if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionReset { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrConnectionReset) + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) } } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 272481aa0..ef823e4ae 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -137,7 +137,10 @@ type Context struct { // New allocates and initializes a test context containing a new // stack and a link-layer endpoint. func New(t *testing.T, mtu uint32) *Context { - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) // Allow minimum send/receive buffer sizes to be 1 during tests. if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{1, tcp.DefaultSendBufferSize, 10 * tcp.DefaultSendBufferSize}); err != nil { @@ -150,11 +153,19 @@ func New(t *testing.T, mtu uint32) *Context { // Some of the congestion control tests send up to 640 packets, we so // set the channel size to 1000. - id, linkEP := channel.New(1000, mtu, "") + ep := channel.New(1000, mtu, "") + wep := stack.LinkEndpoint(ep) + if testing.Verbose() { + wep = sniffer.New(ep) + } + if err := s.CreateNamedNIC(1, "nic1", wep); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + wep2 := stack.LinkEndpoint(channel.New(1000, mtu, "")) if testing.Verbose() { - id = sniffer.New(id) + wep2 = sniffer.New(channel.New(1000, mtu, "")) } - if err := s.CreateNIC(1, id); err != nil { + if err := s.CreateNamedNIC(2, "nic2", wep2); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -180,7 +191,7 @@ func New(t *testing.T, mtu uint32) *Context { return &Context{ t: t, s: s, - linkEP: linkEP, + linkEP: ep, WindowScale: uint8(tcp.FindWndScale(tcp.DefaultReceiveBufferSize)), } } @@ -267,7 +278,7 @@ func (c *Context) GetPacketNonBlocking() []byte { // SendICMPPacket builds and sends an ICMPv4 packet via the link layer endpoint. func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byte, maxTotalSize int) { // Allocate a buffer data and headers. - buf := buffer.NewView(header.IPv4MinimumSize + header.ICMPv4PayloadOffset + len(p1) + len(p2)) + buf := buffer.NewView(header.IPv4MinimumSize + header.ICMPv4PayloadOffset + len(p2)) if len(buf) > maxTotalSize { buf = buf[:maxTotalSize] } @@ -286,9 +297,9 @@ func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byt icmp := header.ICMPv4(buf[header.IPv4MinimumSize:]) icmp.SetType(typ) icmp.SetCode(code) - - copy(icmp[header.ICMPv4PayloadOffset:], p1) - copy(icmp[header.ICMPv4PayloadOffset+len(p1):], p2) + const icmpv4VariableHeaderOffset = 4 + copy(icmp[icmpv4VariableHeaderOffset:], p1) + copy(icmp[header.ICMPv4PayloadOffset:], p2) // Inject packet. c.linkEP.Inject(ipv4.ProtocolNumber, buf.ToVectorisedView()) @@ -511,7 +522,7 @@ func (c *Context) SendV6Packet(payload []byte, h *Headers) { } // CreateConnected creates a connected TCP endpoint. -func (c *Context) CreateConnected(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf *tcpip.ReceiveBufferSizeOption) { +func (c *Context) CreateConnected(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int) { c.CreateConnectedWithRawOptions(iss, rcvWnd, epRcvBuf, nil) } @@ -584,12 +595,8 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) c.Port = tcpHdr.SourcePort() } -// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends -// the specified option bytes as the Option field in the initial SYN packet. -// -// It also sets the receive buffer for the endpoint to the specified -// value in epRcvBuf. -func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf *tcpip.ReceiveBufferSizeOption, options []byte) { +// Create creates a TCP endpoint. +func (c *Context) Create(epRcvBuf int) { // Create TCP endpoint. var err *tcpip.Error c.EP, err = c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) @@ -597,11 +604,20 @@ func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum. c.t.Fatalf("NewEndpoint failed: %v", err) } - if epRcvBuf != nil { - if err := c.EP.SetSockOpt(*epRcvBuf); err != nil { + if epRcvBuf != -1 { + if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, epRcvBuf); err != nil { c.t.Fatalf("SetSockOpt failed failed: %v", err) } } +} + +// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends +// the specified option bytes as the Option field in the initial SYN packet. +// +// It also sets the receive buffer for the endpoint to the specified +// value in epRcvBuf. +func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { + c.Create(epRcvBuf) c.Connect(iss, rcvWnd, options) } diff --git a/pkg/tcpip/transport/tcpconntrack/BUILD b/pkg/tcpip/transport/tcpconntrack/BUILD index 4bec48c0f..43fcc27f0 100644 --- a/pkg/tcpip/transport/tcpconntrack/BUILD +++ b/pkg/tcpip/transport/tcpconntrack/BUILD @@ -1,4 +1,5 @@ -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") package(licenses = ["notice"]) diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index ac2666f69..7a635ab8d 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -1,7 +1,9 @@ +load("@io_bazel_rules_go//go:def.bzl", "go_test") + package(licenses = ["notice"]) load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") go_template_instance( name = "udp_packet_list", @@ -50,6 +52,7 @@ go_test( "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", + "//pkg/tcpip/link/loopback", "//pkg/tcpip/link/sniffer", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index ac5905772..8ae83437e 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -15,7 +15,6 @@ package udp import ( - "math" "sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -37,13 +36,17 @@ type udpPacket struct { views [8]buffer.View `state:"nosave"` } -type endpointState int +// EndpointState represents the state of a UDP endpoint. +type EndpointState uint32 +// Endpoint states. Note that are represented in a netstack-specific manner and +// may not be meaningful externally. Specifically, they need to be translated to +// Linux's representation for these states if presented to userspace. const ( - stateInitial endpointState = iota - stateBound - stateConnected - stateClosed + StateInitial EndpointState = iota + StateBound + StateConnected + StateClosed ) // endpoint represents a UDP endpoint. This struct serves as the interface @@ -74,17 +77,19 @@ type endpoint struct { mu sync.RWMutex `state:"nosave"` sndBufSize int id stack.TransportEndpointID - state endpointState + state EndpointState bindNICID tcpip.NICID regNICID tcpip.NICID route stack.Route `state:"manual"` dstPort uint16 v6only bool + ttl uint8 multicastTTL uint8 multicastAddr tcpip.Address multicastNICID tcpip.NICID multicastLoop bool reusePort bool + bindToDevice tcpip.NICID broadcast bool // shutdownFlags represent the current shutdown state of the endpoint. @@ -140,9 +145,9 @@ func (e *endpoint) Close() { e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite switch e.state { - case stateBound, stateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + case StateBound, StateConnected: + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) } for _, mem := range e.multicastMemberships { @@ -163,7 +168,7 @@ func (e *endpoint) Close() { e.route.Release() // Update the state. - e.state = stateClosed + e.state = StateClosed e.mu.Unlock() @@ -211,11 +216,11 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess // Returns true for retry if preparation should be retried. func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpip.Error) { switch e.state { - case stateInitial: - case stateConnected: + case StateInitial: + case StateConnected: return false, nil - case stateBound: + case StateBound: if to == nil { return false, tcpip.ErrDestinationRequired } @@ -232,7 +237,7 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // The state changed when we released the shared locked and re-acquired // it in exclusive mode. Try again. - if e.state != stateInitial { + if e.state != StateInitial { return true, nil } @@ -273,17 +278,12 @@ func (e *endpoint) connectRoute(nicid tcpip.NICID, addr tcpip.FullAddress, netPr // Write writes data to the endpoint's peer. This method does not block // if the data cannot be written. -func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { +func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // MSG_MORE is unimplemented. (This also means that MSG_EOR is a no-op.) if opts.More { return 0, nil, tcpip.ErrInvalidOptionValue } - if p.Size() > math.MaxUint16 { - // Payload can't possibly fit in a packet. - return 0, nil, tcpip.ErrMessageTooLong - } - to := opts.To e.mu.RLock() @@ -322,7 +322,7 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha defer e.mu.Unlock() // Recheck state after lock was re-acquired. - if e.state != stateConnected { + if e.state != StateConnected { return 0, nil, tcpip.ErrInvalidEndpointState } } @@ -366,17 +366,25 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha } } - v, err := p.Get(p.Size()) + v, err := p.FullPayload() if err != nil { return 0, nil, err } + if len(v) > header.UDPMaximumPacketSize { + // Payload can't possibly fit in a packet. + return 0, nil, tcpip.ErrMessageTooLong + } + + ttl := e.ttl + useDefaultTTL := ttl == 0 - ttl := route.DefaultTTL() if header.IsV4MulticastAddress(route.RemoteAddress) || header.IsV6MulticastAddress(route.RemoteAddress) { ttl = e.multicastTTL + // Multicast allows a 0 TTL. + useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl); err != nil { + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl, useDefaultTTL); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -387,7 +395,12 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } -// SetSockOpt sets a socket option. Currently not supported. +// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. +func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { + return nil +} + +// SetSockOpt implements tcpip.Endpoint.SetSockOpt. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { switch v := opt.(type) { case tcpip.V6OnlyOption: @@ -400,12 +413,17 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { defer e.mu.Unlock() // We only allow this to be set when we're in the initial state. - if e.state != stateInitial { + if e.state != StateInitial { return tcpip.ErrInvalidEndpointState } e.v6only = v != 0 + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + case tcpip.MulticastTTLOption: e.mu.Lock() e.multicastTTL = uint8(v) @@ -544,6 +562,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.reusePort = v != 0 e.mu.Unlock() + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicid, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicid + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 @@ -566,7 +599,20 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { } e.rcvMu.Unlock() return v, nil + + case tcpip.SendBufferSizeOption: + e.mu.Lock() + v := e.sndBufSize + e.mu.Unlock() + return v, nil + + case tcpip.ReceiveBufferSizeOption: + e.rcvMu.Lock() + v := e.rcvBufSizeMax + e.rcvMu.Unlock() + return v, nil } + return -1, tcpip.ErrUnknownProtocolOption } @@ -576,18 +622,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case tcpip.ErrorOption: return nil - case *tcpip.SendBufferSizeOption: - e.mu.Lock() - *o = tcpip.SendBufferSizeOption(e.sndBufSize) - e.mu.Unlock() - return nil - - case *tcpip.ReceiveBufferSizeOption: - e.rcvMu.Lock() - *o = tcpip.ReceiveBufferSizeOption(e.rcvBufSizeMax) - e.rcvMu.Unlock() - return nil - case *tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.netProto != header.IPv6ProtocolNumber { @@ -604,6 +638,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.TTLOption: + e.mu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.mu.Unlock() + return nil + case *tcpip.MulticastTTLOption: e.mu.Lock() *o = tcpip.MulticastTTLOption(e.multicastTTL) @@ -638,6 +678,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = tcpip.BindToDeviceOption("") + return nil + case *tcpip.KeepaliveEnabledOption: *o = 0 return nil @@ -660,7 +710,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. -func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8) *tcpip.Error { +func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool) *tcpip.Error { // Allocate a buffer for the UDP header. hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) @@ -686,7 +736,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u // Track count of packets sent. r.Stats().UDP.PacketsSent.Increment() - return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl) + return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl, useDefaultTTL) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { @@ -726,7 +776,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - if e.state != stateConnected { + if e.state != StateConnected { return nil } id := stack.TransportEndpointID{} @@ -741,12 +791,16 @@ func (e *endpoint) Disconnect() *tcpip.Error { if err != nil { return err } - e.state = stateBound + e.state = StateBound } else { - e.state = stateInitial + if e.id.LocalPort != 0 { + // Release the ephemeral port. + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) + } + e.state = StateInitial } - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.id = id e.route.Release() e.route = stack.Route{} @@ -772,8 +826,8 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { nicid := addr.NIC var localPort uint16 switch e.state { - case stateInitial: - case stateBound, stateConnected: + case StateInitial: + case StateBound, StateConnected: localPort = e.id.LocalPort if e.bindNICID == 0 { break @@ -801,7 +855,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { RemoteAddress: r.RemoteAddress, } - if e.state == stateInitial { + if e.state == StateInitial { id.LocalAddress = r.LocalAddress } @@ -823,7 +877,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Remove the old registration. if e.id.LocalPort != 0 { - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) } e.id = id @@ -832,7 +886,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { e.regNICID = nicid e.effectiveNetProtos = netProtos - e.state = stateConnected + e.state = StateConnected e.rcvMu.Lock() e.rcvReady = true @@ -854,7 +908,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // A socket in the bound state can still receive multicast messages, // so we need to notify waiters on shutdown. - if e.state != stateBound && e.state != stateConnected { + if e.state != StateBound && e.state != StateConnected { return tcpip.ErrNotConnected } @@ -886,16 +940,16 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { if e.id.LocalPort == 0 { - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) if err != nil { return id, err } id.LocalPort = port } - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.bindToDevice) } return id, err } @@ -903,7 +957,7 @@ func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.Networ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { // Don't allow binding once endpoint is not in the initial state // anymore. - if e.state != stateInitial { + if e.state != StateInitial { return tcpip.ErrInvalidEndpointState } @@ -946,7 +1000,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { e.effectiveNetProtos = netProtos // Mark endpoint as bound. - e.state = stateBound + e.state = StateBound e.rcvMu.Lock() e.rcvReady = true @@ -989,7 +1043,7 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { e.mu.RLock() defer e.mu.RUnlock() - if e.state != stateConnected { + if e.state != StateConnected { return tcpip.FullAddress{}, tcpip.ErrNotConnected } @@ -1069,10 +1123,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, vv buffer.VectorisedView) { } -// State implements socket.Socket.State. +// State implements tcpip.Endpoint.State. func (e *endpoint) State() uint32 { - // TODO(b/112063468): Translate internal state to values returned by Linux. - return 0 + e.mu.Lock() + defer e.mu.Unlock() + return uint32(e.state) } func isBroadcastOrMulticast(a tcpip.Address) bool { diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go index 5cbb56120..be46e6d4e 100644 --- a/pkg/tcpip/transport/udp/endpoint_state.go +++ b/pkg/tcpip/transport/udp/endpoint_state.go @@ -77,7 +77,7 @@ func (e *endpoint) Resume(s *stack.Stack) { } } - if e.state != stateBound && e.state != stateConnected { + if e.state != StateBound && e.state != StateConnected { return } @@ -92,7 +92,7 @@ func (e *endpoint) Resume(s *stack.Stack) { } var err *tcpip.Error - if e.state == stateConnected { + if e.state == StateConnected { e.route, err = e.stack.FindRoute(e.regNICID, e.id.LocalAddress, e.id.RemoteAddress, netProto, e.multicastLoop) if err != nil { panic(err) diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a874fc9fd..2d0bc5221 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -74,7 +74,7 @@ func (r *ForwarderRequest) ID() stack.TransportEndpointID { // CreateEndpoint creates a connected UDP endpoint for the session request. func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { ep := newEndpoint(r.stack, r.route.NetProto, queue) - if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort); err != nil { + if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort, ep.bindToDevice); err != nil { ep.Close() return nil, err } @@ -84,7 +84,7 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, ep.dstPort = r.id.RemotePort ep.regNICID = r.route.NICID() - ep.state = stateConnected + ep.state = StateConnected ep.rcvMu.Lock() ep.rcvReady = true diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index f76e7fbe1..0b24bbc85 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -14,7 +14,7 @@ // Package udp contains the implementation of the UDP transport protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing udp.ProtocolName (or "udp") as one of the +// activated on the stack by passing udp.NewProtocol() as one of the // transport protocols when calling stack.New(). Then endpoints can be created // by passing udp.ProtocolNumber as the transport protocol number when calling // Stack.NewEndpoint(). @@ -30,9 +30,6 @@ import ( ) const ( - // ProtocolName is the string representation of the udp protocol name. - ProtocolName = "udp" - // ProtocolNumber is the udp protocol number. ProtocolNumber = header.UDPProtocolNumber ) @@ -69,7 +66,106 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, buffer.VectorisedView) bool { +func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, netHeader buffer.View, vv buffer.VectorisedView) bool { + // Get the header then trim it from the view. + hdr := header.UDP(vv.First()) + if int(hdr.Length()) > vv.Size() { + // Malformed packet. + r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() + return true + } + // TODO(b/129426613): only send an ICMP message if UDP checksum is valid. + + // Only send ICMP error if the address is not a multicast/broadcast + // v4/v6 address or the source is not the unspecified address. + // + // See: point e) in https://tools.ietf.org/html/rfc4443#section-2.4 + if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) || id.RemoteAddress == header.IPv6Any || id.RemoteAddress == header.IPv4Any { + return true + } + + // As per RFC: 1122 Section 3.2.2.1 A host SHOULD generate Destination + // Unreachable messages with code: + // + // 2 (Protocol Unreachable), when the designated transport protocol + // is not supported; or + // + // 3 (Port Unreachable), when the designated transport protocol + // (e.g., UDP) is unable to demultiplex the datagram but has no + // protocol mechanism to inform the sender. + switch len(id.LocalAddress) { + case header.IPv4AddressSize: + if !r.Stack().AllowICMPMessage() { + r.Stack().Stats().ICMP.V4PacketsSent.RateLimited.Increment() + return true + } + // As per RFC 1812 Section 4.3.2.3 + // + // ICMP datagram SHOULD contain as much of the original + // datagram as possible without the length of the ICMP + // datagram exceeding 576 bytes + // + // NOTE: The above RFC referenced is different from the original + // recommendation in RFC 1122 where it mentioned that at least 8 + // bytes of the payload must be included. Today linux and other + // systems implement the] RFC1812 definition and not the original + // RFC 1122 requirement. + mtu := int(r.MTU()) + if mtu > header.IPv4MinimumProcessableDatagramSize { + mtu = header.IPv4MinimumProcessableDatagramSize + } + headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize + available := int(mtu) - headerLen + payloadLen := len(netHeader) + vv.Size() + if payloadLen > available { + payloadLen = available + } + + payload := buffer.NewVectorisedView(len(netHeader), []buffer.View{netHeader}) + payload.Append(vv) + payload.CapLength(payloadLen) + + hdr := buffer.NewPrependable(headerLen) + pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) + pkt.SetType(header.ICMPv4DstUnreachable) + pkt.SetCode(header.ICMPv4PortUnreachable) + pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) + r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv4ProtocolNumber, 0, true /* useDefaultTTL */) + + case header.IPv6AddressSize: + if !r.Stack().AllowICMPMessage() { + r.Stack().Stats().ICMP.V6PacketsSent.RateLimited.Increment() + return true + } + + // As per RFC 4443 section 2.4 + // + // (c) Every ICMPv6 error message (type < 128) MUST include + // as much of the IPv6 offending (invoking) packet (the + // packet that caused the error) as possible without making + // the error message packet exceed the minimum IPv6 MTU + // [IPv6]. + mtu := int(r.MTU()) + if mtu > header.IPv6MinimumMTU { + mtu = header.IPv6MinimumMTU + } + headerLen := int(r.MaxHeaderLength()) + header.ICMPv6DstUnreachableMinimumSize + available := int(mtu) - headerLen + payloadLen := len(netHeader) + vv.Size() + if payloadLen > available { + payloadLen = available + } + payload := buffer.NewVectorisedView(len(netHeader), []buffer.View{netHeader}) + payload.Append(vv) + payload.CapLength(payloadLen) + + hdr := buffer.NewPrependable(headerLen) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6DstUnreachableMinimumSize)) + pkt.SetType(header.ICMPv6DstUnreachable) + pkt.SetCode(header.ICMPv6PortUnreachable) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) + r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv6ProtocolNumber, 0, true /* useDefaultTTL */) + } return true } @@ -83,8 +179,7 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } -func init() { - stack.RegisterTransportProtocolFactory(ProtocolName, func() stack.TransportProtocol { - return &protocol{} - }) +// NewProtocol returns a UDP transport protocol. +func NewProtocol() stack.TransportProtocol { + return &protocol{} } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 9da6edce2..faa728b68 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -17,7 +17,6 @@ package udp_test import ( "bytes" "fmt" - "math" "math/rand" "testing" "time" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/checker" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" @@ -274,13 +274,17 @@ type testContext struct { func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{udp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + }) + ep := channel.New(256, mtu, "") + wep := stack.LinkEndpoint(ep) - id, linkEP := channel.New(256, mtu, "") if testing.Verbose() { - id = sniffer.New(id) + wep = sniffer.New(ep) } - if err := s.CreateNIC(1, id); err != nil { + if err := s.CreateNIC(1, wep); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -306,7 +310,7 @@ func newDualTestContext(t *testing.T, mtu uint32) *testContext { return &testContext{ t: t, s: s, - linkEP: linkEP, + linkEP: ep, } } @@ -461,101 +465,78 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple) { } func newPayload() []byte { - b := make([]byte, 30+rand.Intn(100)) + return newMinPayload(30) +} + +func newMinPayload(minSize int) []byte { + b := make([]byte, minSize+rand.Intn(100)) for i := range b { b[i] = byte(rand.Intn(256)) } return b } -func TestBindPortReuse(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - var eps [5]tcpip.Endpoint - reusePortOpt := tcpip.ReusePortOption(1) +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) - pollChannel := make(chan tcpip.Endpoint) - for i := 0; i < len(eps); i++ { - // Try to receive the data. - wq := waiter.Queue{} - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - defer close(ch) - - var err *tcpip.Error - eps[i], err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) - if err != nil { - c.t.Fatalf("NewEndpoint failed: %v", err) - } - - go func(ep tcpip.Endpoint) { - for range ch { - pollChannel <- ep - } - }(eps[i]) - - defer eps[i].Close() - if err := eps[i].SetSockOpt(reusePortOpt); err != nil { - c.t.Fatalf("SetSockOpt failed failed: %v", err) - } - if err := eps[i].Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } + ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) } + defer ep.Close() - npackets := 100000 - nports := 10000 - ports := make(map[uint16]tcpip.Endpoint) - stats := make(map[tcpip.Endpoint]int) - for i := 0; i < npackets; i++ { - // Send a packet. - port := uint16(i % nports) - payload := newPayload() - c.injectV6Packet(payload, &header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: testV6Addr, Port: testPort + port}, - dstAddr: tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}, - }) + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } - var addr tcpip.FullAddress - ep := <-pollChannel - _, _, err := ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - stats[ep]++ - if i < nports { - ports[uint16(i)] = ep - } else { - // Check that all packets from one client are handled - // by the same socket. - if ports[port] != ep { - t.Fatalf("Port mismatch") - } - } + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) } - if len(stats) != len(eps) { - t.Fatalf("Only %d(expected %d) sockets received packets", len(stats), len(eps)) + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s } - // Check that a packet distribution is fair between sockets. - for _, c := range stats { - n := float64(npackets) / float64(len(eps)) - // The deviation is less than 10%. - if math.Abs(float64(c)-n) > n/10 { - t.Fatal(c, n) - } + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) } } -// testRead sends a packet of the given test flow into the stack by injecting it -// into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { +// testReadInternal sends a packet of the given test flow into the stack by +// injecting it into the link endpoint. It then attempts to read it from the +// UDP endpoint and depending on if this was expected to succeed verifies its +// correctness. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped bool) { c.t.Helper() payload := newPayload() @@ -573,27 +554,51 @@ func testRead(c *testContext, flow testFlow) { select { case <-ch: v, _, err = c.ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for data") + case <-time.After(300 * time.Millisecond): + if packetShouldBeDropped { + return // expected to time out + } + c.t.Fatal("timed out waiting for data") } } + if err != nil { + c.t.Fatal("Read failed:", err) + } + + if packetShouldBeDropped { + c.t.Fatalf("Read unexpectedly received data from %s", addr.Addr) + } + // Check the peer address. h := flow.header4Tuple(incoming) if addr.Addr != h.srcAddr.Addr { - c.t.Fatalf("Unexpected remote address: got %v, want %v", addr.Addr, h.srcAddr) + c.t.Fatalf("unexpected remote address: got %s, want %s", addr.Addr, h.srcAddr) } // Check the payload. if !bytes.Equal(payload, v) { - c.t.Fatalf("Bad payload: got %x, want %x", v, payload) + c.t.Fatalf("bad payload: got %x, want %x", v, payload) } } +// testRead sends a packet of the given test flow into the stack by injecting it +// into the link endpoint. It then reads it from the UDP endpoint and verifies +// its correctness. +func testRead(c *testContext, flow testFlow) { + c.t.Helper() + testReadInternal(c, flow, false /* packetShouldBeDropped */) +} + +// testFailingRead sends a packet of the given test flow into the stack by +// injecting it into the link endpoint. It then tries to read it from the UDP +// endpoint and expects this to fail. +func testFailingRead(c *testContext, flow testFlow) { + c.t.Helper() + testReadInternal(c, flow, true /* packetShouldBeDropped */) +} + func TestBindEphemeralPort(t *testing.T) { c := newDualTestContext(t, defaultMTU) defer c.cleanup() @@ -763,13 +768,17 @@ func TestReadOnBoundToMulticast(t *testing.T) { c.t.Fatal("SetSockOpt failed:", err) } + // Check that we receive multicast packets but not unicast or broadcast + // ones. testRead(c, flow) + testFailingRead(c, broadcast) + testFailingRead(c, unicastV4) }) } } // TestV4ReadOnBoundToBroadcast checks that an endpoint can bind to a broadcast -// address and receive broadcast data on it. +// address and can receive only broadcast data. func TestV4ReadOnBoundToBroadcast(t *testing.T) { for _, flow := range []testFlow{broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -784,8 +793,31 @@ func TestV4ReadOnBoundToBroadcast(t *testing.T) { c.t.Fatalf("Bind failed: %s", err) } - // Test acceptance. + // Check that we receive broadcast packets but not unicast ones. + testRead(c, flow) + testFailingRead(c, unicastV4) + }) + } +} + +// TestV4ReadBroadcastOnBoundToWildcard checks that an endpoint can bind to ANY +// and receive broadcast and unicast data. +func TestV4ReadBroadcastOnBoundToWildcard(t *testing.T) { + for _, flow := range []testFlow{broadcast, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s (", err) + } + + // Check that we receive both broadcast and unicast packets. testRead(c, flow) + testRead(c, unicastV4) }) } } @@ -1175,6 +1207,39 @@ func TestTTL(t *testing.T) { } } +func TestSetTTL(t *testing.T) { + for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} { + t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } + + var p stack.NetworkProtocol + if flow.isV4() { + p = ipv4.NewProtocol() + } else { + p = ipv6.NewProtocol() + } + ep, err := p.NewEndpoint(0, tcpip.AddressWithPrefix{}, nil, nil, nil) + if err != nil { + t.Fatal(err) + } + ep.Close() + + testWrite(c, flow, checker.TTL(wantTTL)) + }) + } + }) + } +} + func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -1238,3 +1303,153 @@ func TestMulticastInterfaceOption(t *testing.T) { }) } } + +// TestV4UnknownDestination verifies that we generate an ICMPv4 Destination +// Unreachable message when a udp datagram is received on ports for which there +// is no bound udp socket. +func TestV4UnknownDestination(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + testCases := []struct { + flow testFlow + icmpRequired bool + // largePayload if true, will result in a payload large enough + // so that the final generated IPv4 packet is larger than + // header.IPv4MinimumProcessableDatagramSize. + largePayload bool + }{ + {unicastV4, true, false}, + {unicastV4, true, true}, + {multicastV4, false, false}, + {multicastV4, false, true}, + {broadcast, false, false}, + {broadcast, false, true}, + } + for _, tc := range testCases { + t.Run(fmt.Sprintf("flow:%s icmpRequired:%t largePayload:%t", tc.flow, tc.icmpRequired, tc.largePayload), func(t *testing.T) { + payload := newPayload() + if tc.largePayload { + payload = newMinPayload(576) + } + c.injectPacket(tc.flow, payload) + if !tc.icmpRequired { + select { + case p := <-c.linkEP.C: + t.Fatalf("unexpected packet received: %+v", p) + case <-time.After(1 * time.Second): + return + } + } + + select { + case p := <-c.linkEP.C: + var pkt []byte + pkt = append(pkt, p.Header...) + pkt = append(pkt, p.Payload...) + if got, want := len(pkt), header.IPv4MinimumProcessableDatagramSize; got > want { + t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) + } + + hdr := header.IPv4(pkt) + checker.IPv4(t, hdr, checker.ICMPv4( + checker.ICMPv4Type(header.ICMPv4DstUnreachable), + checker.ICMPv4Code(header.ICMPv4PortUnreachable))) + + icmpPkt := header.ICMPv4(hdr.Payload()) + payloadIPHeader := header.IPv4(icmpPkt.Payload()) + wantLen := len(payload) + if tc.largePayload { + wantLen = header.IPv4MinimumProcessableDatagramSize - header.IPv4MinimumSize*2 - header.ICMPv4MinimumSize - header.UDPMinimumSize + } + + // In case of large payloads the IP packet may be truncated. Update + // the length field before retrieving the udp datagram payload. + payloadIPHeader.SetTotalLength(uint16(wantLen + header.UDPMinimumSize + header.IPv4MinimumSize)) + + origDgram := header.UDP(payloadIPHeader.Payload()) + if got, want := len(origDgram.Payload()), wantLen; got != want { + t.Fatalf("unexpected payload length got: %d, want: %d", got, want) + } + if got, want := origDgram.Payload(), payload[:wantLen]; !bytes.Equal(got, want) { + t.Fatalf("unexpected payload got: %d, want: %d", got, want) + } + case <-time.After(1 * time.Second): + t.Fatalf("packet wasn't written out") + } + }) + } +} + +// TestV6UnknownDestination verifies that we generate an ICMPv6 Destination +// Unreachable message when a udp datagram is received on ports for which there +// is no bound udp socket. +func TestV6UnknownDestination(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + testCases := []struct { + flow testFlow + icmpRequired bool + // largePayload if true will result in a payload large enough to + // create an IPv6 packet > header.IPv6MinimumMTU bytes. + largePayload bool + }{ + {unicastV6, true, false}, + {unicastV6, true, true}, + {multicastV6, false, false}, + {multicastV6, false, true}, + } + for _, tc := range testCases { + t.Run(fmt.Sprintf("flow:%s icmpRequired:%t largePayload:%t", tc.flow, tc.icmpRequired, tc.largePayload), func(t *testing.T) { + payload := newPayload() + if tc.largePayload { + payload = newMinPayload(1280) + } + c.injectPacket(tc.flow, payload) + if !tc.icmpRequired { + select { + case p := <-c.linkEP.C: + t.Fatalf("unexpected packet received: %+v", p) + case <-time.After(1 * time.Second): + return + } + } + + select { + case p := <-c.linkEP.C: + var pkt []byte + pkt = append(pkt, p.Header...) + pkt = append(pkt, p.Payload...) + if got, want := len(pkt), header.IPv6MinimumMTU; got > want { + t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) + } + + hdr := header.IPv6(pkt) + checker.IPv6(t, hdr, checker.ICMPv6( + checker.ICMPv6Type(header.ICMPv6DstUnreachable), + checker.ICMPv6Code(header.ICMPv6PortUnreachable))) + + icmpPkt := header.ICMPv6(hdr.Payload()) + payloadIPHeader := header.IPv6(icmpPkt.Payload()) + wantLen := len(payload) + if tc.largePayload { + wantLen = header.IPv6MinimumMTU - header.IPv6MinimumSize*2 - header.ICMPv6MinimumSize - header.UDPMinimumSize + } + // In case of large payloads the IP packet may be truncated. Update + // the length field before retrieving the udp datagram payload. + payloadIPHeader.SetPayloadLength(uint16(wantLen + header.UDPMinimumSize)) + + origDgram := header.UDP(payloadIPHeader.Payload()) + if got, want := len(origDgram.Payload()), wantLen; got != want { + t.Fatalf("unexpected payload length got: %d, want: %d", got, want) + } + if got, want := origDgram.Payload(), payload[:wantLen]; !bytes.Equal(got, want) { + t.Fatalf("unexpected payload got: %v, want: %v", got, want) + } + case <-time.After(1 * time.Second): + t.Fatalf("packet wasn't written out") + } + }) + } +} |