diff options
Diffstat (limited to 'pkg/tcpip')
39 files changed, 1492 insertions, 649 deletions
diff --git a/pkg/tcpip/adapters/gonet/gonet_test.go b/pkg/tcpip/adapters/gonet/gonet_test.go index 672f026b2..8ced960bb 100644 --- a/pkg/tcpip/adapters/gonet/gonet_test.go +++ b/pkg/tcpip/adapters/gonet/gonet_test.go @@ -60,7 +60,10 @@ func TestTimeouts(t *testing.T) { func newLoopbackStack() (*stack.Stack, *tcpip.Error) { // Create the stack and add a NIC. - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{tcp.ProtocolName, udp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol()}, + }) if err := s.CreateNIC(NICID, loopback.New()); err != nil { return nil, err diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 584db710e..7636418b1 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -82,7 +82,19 @@ const ( PacketMMap ) -// An endpoint implements the link-layer using a message-oriented file descriptor. +func (p PacketDispatchMode) String() string { + switch p { + case Readv: + return "Readv" + case RecvMMsg: + return "RecvMMsg" + case PacketMMap: + return "PacketMMap" + default: + return fmt.Sprintf("unknown packet dispatch mode %v", p) + } +} + type endpoint struct { // fds is the set of file descriptors each identifying one inbound/outbound // channel. The endpoint will dispatch from all inbound channels as well as diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index fd6395fc1..26cf1c528 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -16,9 +16,9 @@ // IPv4 addresses into link-local MAC addresses, and advertises IPv4 // addresses of its stack with the local network. // -// To use it in the networking stack, pass arp.ProtocolName as one of the -// network protocols when calling stack.New. Then add an "arp" address to -// every NIC on the stack that should respond to ARP requests. That is: +// To use it in the networking stack, pass arp.NewProtocol() as one of the +// network protocols when calling stack.New. Then add an "arp" address to every +// NIC on the stack that should respond to ARP requests. That is: // // if err := s.AddAddress(1, arp.ProtocolNumber, "arp"); err != nil { // // handle err @@ -33,9 +33,6 @@ import ( ) const ( - // ProtocolName is the string representation of the ARP protocol name. - ProtocolName = "arp" - // ProtocolNumber is the ARP protocol number. ProtocolNumber = header.ARPProtocolNumber @@ -200,8 +197,7 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { var broadcastMAC = tcpip.LinkAddress([]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff}) -func init() { - stack.RegisterNetworkProtocolFactory(ProtocolName, func() stack.NetworkProtocol { - return &protocol{} - }) +// NewProtocol returns an ARP network protocol. +func NewProtocol() stack.NetworkProtocol { + return &protocol{} } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 387fca96e..88b57ec03 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -44,7 +44,10 @@ type testContext struct { } func newTestContext(t *testing.T) *testContext { - s := stack.New([]string{ipv4.ProtocolName, arp.ProtocolName}, []string{icmp.ProtocolName4}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, + }) const defaultMTU = 65536 ep := channel.New(256, defaultMTU, stackLinkAddr) diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 6a40e7ee3..a9741622e 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -172,7 +172,10 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prepen } func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { - s := stack.New([]string{ipv4.ProtocolName}, []string{udp.ProtocolName, tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + }) s.CreateNIC(1, loopback.New()) s.AddAddress(1, ipv4.ProtocolNumber, local) s.SetRouteTable([]tcpip.Route{{ @@ -185,7 +188,10 @@ func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { } func buildIPv6Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { - s := stack.New([]string{ipv6.ProtocolName}, []string{udp.ProtocolName, tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + }) s.CreateNIC(1, loopback.New()) s.AddAddress(1, ipv6.ProtocolNumber, local) s.SetRouteTable([]tcpip.Route{{ diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index b7a06f525..b7b07a6c1 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -14,9 +14,9 @@ // Package ipv4 contains the implementation of the ipv4 network protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv4.ProtocolName (or "ipv4") as one of the -// network protocols when calling stack.New(). Then endpoints can be created -// by passing ipv4.ProtocolNumber as the network protocol number when calling +// activated on the stack by passing ipv4.NewProtocol() as one of the network +// protocols when calling stack.New(). Then endpoints can be created by passing +// ipv4.ProtocolNumber as the network protocol number when calling // Stack.NewEndpoint(). package ipv4 @@ -32,9 +32,6 @@ import ( ) const ( - // ProtocolName is the string representation of the ipv4 protocol name. - ProtocolName = "ipv4" - // ProtocolNumber is the ipv4 protocol number. ProtocolNumber = header.IPv4ProtocolNumber @@ -53,6 +50,7 @@ type endpoint struct { linkEP stack.LinkEndpoint dispatcher stack.TransportDispatcher fragmentation *fragmentation.Fragmentation + protocol *protocol } // NewEndpoint creates a new ipv4 endpoint. @@ -64,6 +62,7 @@ func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWi linkEP: linkEP, dispatcher: dispatcher, fragmentation: fragmentation.NewFragmentation(fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), + protocol: p, } return e, nil @@ -204,7 +203,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen if length > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be // fragmented, so we only assign ids to larger packets. - id = atomic.AddUint32(&ids[hashRoute(r, protocol)%buckets], 1) + id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, protocol, e.protocol.hashIV)%buckets], 1) } ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, @@ -267,7 +266,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect if payload.Size() > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be // fragmented, so we only assign ids to larger packets. - id = atomic.AddUint32(&ids[hashRoute(r, 0 /* protocol */)%buckets], 1) + id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1) } ip.SetID(uint16(id)) } @@ -325,14 +324,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { // Close cleans up resources associated with the endpoint. func (e *endpoint) Close() {} -type protocol struct{} - -// NewProtocol creates a new protocol ipv4 protocol descriptor. This is exported -// only for tests that short-circuit the stack. Regular use of the protocol is -// done via the stack, which gets a protocol descriptor from the init() function -// below. -func NewProtocol() stack.NetworkProtocol { - return &protocol{} +type protocol struct { + ids []uint32 + hashIV uint32 } // Number returns the ipv4 protocol number. @@ -378,7 +372,7 @@ func calculateMTU(mtu uint32) uint32 { // hashRoute calculates a hash value for the given route. It uses the source & // destination address, the transport protocol number, and a random initial // value (generated once on initialization) to generate the hash. -func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber) uint32 { +func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber, hashIV uint32) uint32 { t := r.LocalAddress a := uint32(t[0]) | uint32(t[1])<<8 | uint32(t[2])<<16 | uint32(t[3])<<24 t = r.RemoteAddress @@ -386,22 +380,16 @@ func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber) uint32 { return hash.Hash3Words(a, b, uint32(protocol), hashIV) } -var ( - ids []uint32 - hashIV uint32 -) - -func init() { - ids = make([]uint32, buckets) +// NewProtocol returns an IPv4 network protocol. +func NewProtocol() stack.NetworkProtocol { + ids := make([]uint32, buckets) // Randomly initialize hashIV and the ids. r := hash.RandN32(1 + buckets) for i := range ids { ids[i] = r[i] } - hashIV = r[buckets] + hashIV := r[buckets] - stack.RegisterNetworkProtocolFactory(ProtocolName, func() stack.NetworkProtocol { - return &protocol{} - }) + return &protocol{ids: ids, hashIV: hashIV} } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index ae827ca27..b6641ccc3 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -33,7 +33,10 @@ import ( ) func TestExcludeBroadcast(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{udp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + }) const defaultMTU = 65536 ep := stack.LinkEndpoint(channel.New(256, defaultMTU, "")) @@ -238,7 +241,9 @@ type context struct { func buildContext(t *testing.T, packetCollectorErrors []*tcpip.Error, mtu uint32) context { // Make the packet and write it. - s := stack.New([]string{ipv4.ProtocolName}, []string{}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + }) ep := newErrorChannel(100 /* Enough for all tests. */, mtu, "", packetCollectorErrors) s.CreateNIC(1, ep) const ( diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 653d984e9..01f5a17ec 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -81,7 +81,10 @@ func (*stubLinkAddressCache) AddLinkAddress(tcpip.NICID, tcpip.Address, tcpip.Li } func TestICMPCounts(t *testing.T) { - s := stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }) { if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -205,8 +208,14 @@ func (e endpointWithResolutionCapability) Capabilities() stack.LinkEndpointCapab func newTestContext(t *testing.T) *testContext { c := &testContext{ - s0: stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}), - s1: stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}), + s0: stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }), + s1: stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }), } const defaultMTU = 65536 diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 331a8bdaa..7de6a4546 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -14,9 +14,9 @@ // Package ipv6 contains the implementation of the ipv6 network protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv6.ProtocolName (or "ipv6") as one of the -// network protocols when calling stack.New(). Then endpoints can be created -// by passing ipv6.ProtocolNumber as the network protocol number when calling +// activated on the stack by passing ipv6.NewProtocol() as one of the network +// protocols when calling stack.New(). Then endpoints can be created by passing +// ipv6.ProtocolNumber as the network protocol number when calling // Stack.NewEndpoint(). package ipv6 @@ -28,9 +28,6 @@ import ( ) const ( - // ProtocolName is the string representation of the ipv6 protocol name. - ProtocolName = "ipv6" - // ProtocolNumber is the ipv6 protocol number. ProtocolNumber = header.IPv6ProtocolNumber @@ -160,14 +157,6 @@ func (*endpoint) Close() {} type protocol struct{} -// NewProtocol creates a new protocol ipv6 protocol descriptor. This is exported -// only for tests that short-circuit the stack. Regular use of the protocol is -// done via the stack, which gets a protocol descriptor from the init() function -// below. -func NewProtocol() stack.NetworkProtocol { - return &protocol{} -} - // Number returns the ipv6 protocol number. func (p *protocol) Number() tcpip.NetworkProtocolNumber { return ProtocolNumber @@ -221,8 +210,7 @@ func calculateMTU(mtu uint32) uint32 { return maxPayloadSize } -func init() { - stack.RegisterNetworkProtocolFactory(ProtocolName, func() stack.NetworkProtocol { - return &protocol{} - }) +// NewProtocol returns an IPv6 network protocol. +func NewProtocol() stack.NetworkProtocol { + return &protocol{} } diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 57bcd5455..78c674c2c 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -124,17 +124,20 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst // UDP packets destined to the IPv6 link-local all-nodes multicast address. func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { tests := []struct { - name string - protocolName string - rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) + name string + protocolFactory stack.TransportProtocol + rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.ProtocolName6, testReceiveICMP}, - {"UDP", udp.ProtocolName, testReceiveUDP}, + {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, + {"UDP", udp.NewProtocol(), testReceiveUDP}, } for _, test := range tests { t.Run(test.name, func(t *testing.T) { - s := stack.New([]string{ProtocolName}, []string{test.protocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -152,19 +155,22 @@ func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { // address. func TestReceiveOnSolicitedNodeAddr(t *testing.T) { tests := []struct { - name string - protocolName string - rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) + name string + protocolFactory stack.TransportProtocol + rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.ProtocolName6, testReceiveICMP}, - {"UDP", udp.ProtocolName, testReceiveUDP}, + {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, + {"UDP", udp.NewProtocol(), testReceiveUDP}, } snmc := header.SolicitedNodeAddr(addr2) for _, test := range tests { t.Run(test.name, func(t *testing.T) { - s := stack.New([]string{ProtocolName}, []string{test.protocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -237,7 +243,9 @@ func TestAddIpv6Address(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - s := stack.New([]string{ProtocolName}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) } diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 571915d3f..e30791fe3 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -31,7 +31,10 @@ import ( func setupStackAndEndpoint(t *testing.T, llladdr, rlladdr tcpip.Address) (*stack.Stack, stack.NetworkEndpoint) { t.Helper() - s := stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }) if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 315780c0c..30cea8996 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -19,6 +19,7 @@ import ( "math" "math/rand" "sync" + "sync/atomic" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -27,6 +28,10 @@ const ( // FirstEphemeral is the first ephemeral port. FirstEphemeral = 16000 + // numEphemeralPorts it the mnumber of available ephemeral ports to + // Netstack. + numEphemeralPorts = math.MaxUint16 - FirstEphemeral + 1 + anyIPAddress tcpip.Address = "" ) @@ -40,6 +45,13 @@ type portDescriptor struct { type PortManager struct { mu sync.RWMutex allocatedPorts map[portDescriptor]bindAddresses + + // hint is used to pick ports ephemeral ports in a stable order for + // a given port offset. + // + // hint must be accessed using the portHint/incPortHint helpers. + // TODO(gvisor.dev/issue/940): S/R this field. + hint uint32 } type portNode struct { @@ -47,43 +59,76 @@ type portNode struct { refs int } -// bindAddresses is a set of IP addresses. -type bindAddresses map[tcpip.Address]portNode +// deviceNode is never empty. When it has no elements, it is removed from the +// map that references it. +type deviceNode map[tcpip.NICID]portNode -// isAvailable checks whether an IP address is available to bind to. -func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool) bool { - if addr == anyIPAddress { - if len(b) == 0 { - return true - } +// isAvailable checks whether binding is possible by device. If not binding to a +// device, check against all portNodes. If binding to a specific device, check +// against the unspecified device and the provided device. +func (d deviceNode) isAvailable(reuse bool, bindToDevice tcpip.NICID) bool { + if bindToDevice == 0 { + // Trying to binding all devices. if !reuse { + // Can't bind because the (addr,port) is already bound. return false } - for _, n := range b { - if !n.reuse { + for _, p := range d { + if !p.reuse { + // Can't bind because the (addr,port) was previously bound without reuse. return false } } return true } - // If all addresses for this portDescriptor are already bound, no - // address is available. - if n, ok := b[anyIPAddress]; ok { - if !reuse { + if p, ok := d[0]; ok { + if !reuse || !p.reuse { return false } - if !n.reuse { + } + + if p, ok := d[bindToDevice]; ok { + if !reuse || !p.reuse { return false } } - if n, ok := b[addr]; ok { - if !reuse { + return true +} + +// bindAddresses is a set of IP addresses. +type bindAddresses map[tcpip.Address]deviceNode + +// isAvailable checks whether an IP address is available to bind to. If the +// address is the "any" address, check all other addresses. Otherwise, just +// check against the "any" address and the provided address. +func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice tcpip.NICID) bool { + if addr == anyIPAddress { + // If binding to the "any" address then check that there are no conflicts + // with all addresses. + for _, d := range b { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + return true + } + + // Check that there is no conflict with the "any" address. + if d, ok := b[anyIPAddress]; ok { + if !d.isAvailable(reuse, bindToDevice) { return false } - return n.reuse } + + // Check that this is no conflict with the provided address. + if d, ok := b[addr]; ok { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + return true } @@ -97,11 +142,40 @@ func NewPortManager() *PortManager { // is suitable for its needs, and stopping when a port is found or an error // occurs. func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { - count := uint16(math.MaxUint16 - FirstEphemeral + 1) - offset := uint16(rand.Int31n(int32(count))) + offset := uint32(rand.Int31n(numEphemeralPorts)) + return s.pickEphemeralPort(offset, numEphemeralPorts, testPort) +} + +// portHint atomically reads and returns the s.hint value. +func (s *PortManager) portHint() uint32 { + return atomic.LoadUint32(&s.hint) +} + +// incPortHint atomically increments s.hint by 1. +func (s *PortManager) incPortHint() { + atomic.AddUint32(&s.hint, 1) +} - for i := uint16(0); i < count; i++ { - port = FirstEphemeral + (offset+i)%count +// PickEphemeralPortStable starts at the specified offset + s.portHint and +// iterates over all ephemeral ports, allowing the caller to decide whether a +// given port is suitable for its needs and stopping when a port is found or an +// error occurs. +func (s *PortManager) PickEphemeralPortStable(offset uint32, testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { + p, err := s.pickEphemeralPort(s.portHint()+offset, numEphemeralPorts, testPort) + if err == nil { + s.incPortHint() + } + return p, err + +} + +// pickEphemeralPort starts at the offset specified from the FirstEphemeral port +// and iterates over the number of ports specified by count and allows the +// caller to decide whether a given port is suitable for its needs, and stopping +// when a port is found or an error occurs. +func (s *PortManager) pickEphemeralPort(offset, count uint32, testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { + for i := uint32(0); i < count; i++ { + port = uint16(FirstEphemeral + (offset+i)%count) ok, err := testPort(port) if err != nil { return 0, err @@ -116,17 +190,17 @@ func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Er } // IsPortAvailable tests if the given port is available on all given protocols. -func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, reuse) + return s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { for _, network := range networks { desc := portDescriptor{network, transport, port} if addrs, ok := s.allocatedPorts[desc]; ok { - if !addrs.isAvailable(addr, reuse) { + if !addrs.isAvailable(addr, reuse, bindToDevice) { return false } } @@ -138,14 +212,14 @@ func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumb // reserved by another endpoint. If port is zero, ReservePort will search for // an unreserved ephemeral port and reserve it, returning its value in the // "port" return value. -func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() // If a port is specified, just try to reserve it for all network // protocols. if port != 0 { - if !s.reserveSpecificPort(networks, transport, addr, port, reuse) { + if !s.reserveSpecificPort(networks, transport, addr, port, reuse, bindToDevice) { return 0, tcpip.ErrPortInUse } return port, nil @@ -153,13 +227,13 @@ func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transp // A port wasn't specified, so try to find one. return s.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - return s.reserveSpecificPort(networks, transport, addr, p, reuse), nil + return s.reserveSpecificPort(networks, transport, addr, p, reuse, bindToDevice), nil }) } // reserveSpecificPort tries to reserve the given port on all given protocols. -func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, reuse) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) { return false } @@ -171,11 +245,16 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber m = make(bindAddresses) s.allocatedPorts[desc] = m } - if n, ok := m[addr]; ok { + d, ok := m[addr] + if !ok { + d = make(deviceNode) + m[addr] = d + } + if n, ok := d[bindToDevice]; ok { n.refs++ - m[addr] = n + d[bindToDevice] = n } else { - m[addr] = portNode{reuse: reuse, refs: 1} + d[bindToDevice] = portNode{reuse: reuse, refs: 1} } } @@ -184,22 +263,28 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber // ReleasePort releases the reservation on a port/IP combination so that it can // be reserved by other endpoints. -func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, bindToDevice tcpip.NICID) { s.mu.Lock() defer s.mu.Unlock() for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { - n, ok := m[addr] + d, ok := m[addr] + if !ok { + continue + } + n, ok := d[bindToDevice] if !ok { continue } n.refs-- + d[bindToDevice] = n if n.refs == 0 { + delete(d, bindToDevice) + } + if len(d) == 0 { delete(m, addr) - } else { - m[addr] = n } if len(m) == 0 { delete(s.allocatedPorts, desc) diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index 689401661..19f4833fc 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -15,6 +15,7 @@ package ports import ( + "math/rand" "testing" "gvisor.dev/gvisor/pkg/tcpip" @@ -34,6 +35,7 @@ type portReserveTestAction struct { want *tcpip.Error reuse bool release bool + device tcpip.NICID } func TestPortReservation(t *testing.T) { @@ -100,6 +102,112 @@ func TestPortReservation(t *testing.T) { {port: 24, ip: anyIPAddress, release: true}, {port: 24, ip: anyIPAddress, reuse: false, want: nil}, }, + }, { + tname: "bind twice with device fails", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 3, want: nil}, + {port: 24, ip: fakeIPAddress, device: 3, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 1, want: nil}, + {port: 24, ip: fakeIPAddress, device: 2, want: nil}, + }, + }, { + tname: "bind to device and then without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, want: nil}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + }, + }, { + tname: "binding with reuse and device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "mixing reuse and not reuse by binding to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: nil}, + }, + }, { + tname: "can't bind to 0 after mixing reuse and not reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind and release", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + + // Release the bind to device 0 and try again. + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil}, + }, + }, { + tname: "bind twice with reuse once", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "release an unreserved device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil}, + // The below don't exist. + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil, release: true}, + {port: 9999, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + // Release all. + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil, release: true}, + }, }, } { t.Run(test.tname, func(t *testing.T) { @@ -108,12 +216,12 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.device) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse, test.device) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %t) = %v, want %v", test.ip, test.port, test.release, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %t, %d) = %v, want %v", test.ip, test.port, test.reuse, test.device, err, test.want) } if test.port == 0 && (gotPort == 0 || gotPort < FirstEphemeral) { t.Fatalf("ReservePort(.., .., .., 0) = %d, want port number >= %d to be picked", gotPort, FirstEphemeral) @@ -125,7 +233,6 @@ func TestPortReservation(t *testing.T) { } func TestPickEphemeralPort(t *testing.T) { - pm := NewPortManager() customErr := &tcpip.Error{} for _, test := range []struct { name string @@ -169,9 +276,63 @@ func TestPickEphemeralPort(t *testing.T) { }, } { t.Run(test.name, func(t *testing.T) { + pm := NewPortManager() if port, err := pm.PickEphemeralPort(test.f); port != test.wantPort || err != test.wantErr { t.Errorf("PickEphemeralPort(..) = (port %d, err %v); want (port %d, err %v)", port, err, test.wantPort, test.wantErr) } }) } } + +func TestPickEphemeralPortStable(t *testing.T) { + customErr := &tcpip.Error{} + for _, test := range []struct { + name string + f func(port uint16) (bool, *tcpip.Error) + wantErr *tcpip.Error + wantPort uint16 + }{ + { + name: "no-port-available", + f: func(port uint16) (bool, *tcpip.Error) { + return false, nil + }, + wantErr: tcpip.ErrNoPortAvailable, + }, + { + name: "port-tester-error", + f: func(port uint16) (bool, *tcpip.Error) { + return false, customErr + }, + wantErr: customErr, + }, + { + name: "only-port-16042-available", + f: func(port uint16) (bool, *tcpip.Error) { + if port == FirstEphemeral+42 { + return true, nil + } + return false, nil + }, + wantPort: FirstEphemeral + 42, + }, + { + name: "only-port-under-16000-available", + f: func(port uint16) (bool, *tcpip.Error) { + if port < FirstEphemeral { + return true, nil + } + return false, nil + }, + wantErr: tcpip.ErrNoPortAvailable, + }, + } { + t.Run(test.name, func(t *testing.T) { + pm := NewPortManager() + portOffset := uint32(rand.Int31n(int32(numEphemeralPorts))) + if port, err := pm.PickEphemeralPortStable(portOffset, test.f); port != test.wantPort || err != test.wantErr { + t.Errorf("PickEphemeralPort(..) = (port %d, err %v); want (port %d, err %v)", port, err, test.wantPort, test.wantErr) + } + }) + } +} diff --git a/pkg/tcpip/sample/tun_tcp_connect/main.go b/pkg/tcpip/sample/tun_tcp_connect/main.go index f12189580..2239c1e66 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/main.go +++ b/pkg/tcpip/sample/tun_tcp_connect/main.go @@ -126,7 +126,10 @@ func main() { // Create the stack with ipv4 and tcp protocols, then add a tun-based // NIC and ipv4 address. - 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()}, + }) mtu, err := rawfile.GetMTU(tunName) if err != nil { diff --git a/pkg/tcpip/sample/tun_tcp_echo/main.go b/pkg/tcpip/sample/tun_tcp_echo/main.go index 329941775..bca73cbb1 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/main.go +++ b/pkg/tcpip/sample/tun_tcp_echo/main.go @@ -111,7 +111,10 @@ func main() { // Create the stack with ip and tcp protocols, then add a tun-based // NIC and address. - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName, arp.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) mtu, err := rawfile.GetMTU(tunName) if err != nil { diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 28c49e8ff..baf88bfab 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -36,6 +36,7 @@ go_library( ], deps = [ "//pkg/ilist", + "//pkg/rand", "//pkg/sleep", "//pkg/tcpip", "//pkg/tcpip/buffer", @@ -54,6 +55,7 @@ go_test( size = "small", srcs = [ "stack_test.go", + "transport_demuxer_test.go", "transport_test.go", ], deps = [ @@ -64,6 +66,9 @@ go_test( "//pkg/tcpip/iptables", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", + "//pkg/tcpip/network/ipv4", + "//pkg/tcpip/network/ipv6", + "//pkg/tcpip/transport/udp", "//pkg/waiter", ], ) diff --git a/pkg/tcpip/stack/icmp_rate_limit.go b/pkg/tcpip/stack/icmp_rate_limit.go index f8156be47..3a20839da 100644 --- a/pkg/tcpip/stack/icmp_rate_limit.go +++ b/pkg/tcpip/stack/icmp_rate_limit.go @@ -15,8 +15,6 @@ package stack import ( - "sync" - "golang.org/x/time/rate" ) @@ -33,54 +31,11 @@ const ( // ICMPRateLimiter is a global rate limiter that controls the generation of // ICMP messages generated by the stack. type ICMPRateLimiter struct { - mu sync.RWMutex - l *rate.Limiter + *rate.Limiter } // NewICMPRateLimiter returns a global rate limiter for controlling the rate // at which ICMP messages are generated by the stack. func NewICMPRateLimiter() *ICMPRateLimiter { - return &ICMPRateLimiter{l: rate.NewLimiter(icmpLimit, icmpBurst)} -} - -// Allow returns true if we are allowed to send at least 1 message at the -// moment. -func (i *ICMPRateLimiter) Allow() bool { - i.mu.RLock() - allow := i.l.Allow() - i.mu.RUnlock() - return allow -} - -// Limit returns the maximum number of ICMP messages that can be sent in one -// second. -func (i *ICMPRateLimiter) Limit() rate.Limit { - i.mu.RLock() - defer i.mu.RUnlock() - return i.l.Limit() -} - -// SetLimit sets the maximum number of ICMP messages that can be sent in one -// second. -func (i *ICMPRateLimiter) SetLimit(newLimit rate.Limit) { - i.mu.RLock() - defer i.mu.RUnlock() - i.l.SetLimit(newLimit) -} - -// Burst returns how many ICMP messages can be sent at any single instant. -func (i *ICMPRateLimiter) Burst() int { - i.mu.RLock() - defer i.mu.RUnlock() - return i.l.Burst() -} - -// SetBurst sets the maximum number of ICMP messages allowed at any single -// instant. -// -// NOTE: Changing Burst causes the underlying rate limiter to be recreated. -func (i *ICMPRateLimiter) SetBurst(burst int) { - i.mu.Lock() - i.l = rate.NewLimiter(i.l.Limit(), burst) - i.mu.Unlock() + return &ICMPRateLimiter{Limiter: rate.NewLimiter(icmpLimit, icmpBurst)} } diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index a719058b4..f6106f762 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -34,8 +34,6 @@ type NIC struct { linkEP LinkEndpoint loopback bool - demux *transportDemuxer - mu sync.RWMutex spoofing bool promiscuous bool @@ -85,7 +83,6 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback name: name, linkEP: ep, loopback: loopback, - demux: newTransportDemuxer(stack), primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List), endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint), mcastJoins: make(map[NetworkEndpointID]int32), @@ -148,37 +145,6 @@ func (n *NIC) setSpoofing(enable bool) { n.mu.Unlock() } -func (n *NIC) getMainNICAddress(protocol tcpip.NetworkProtocolNumber) (tcpip.AddressWithPrefix, *tcpip.Error) { - n.mu.RLock() - defer n.mu.RUnlock() - - var r *referencedNetworkEndpoint - - // Check for a primary endpoint. - if list, ok := n.primary[protocol]; ok { - for e := list.Front(); e != nil; e = e.Next() { - ref := e.(*referencedNetworkEndpoint) - if ref.getKind() == permanent && ref.tryIncRef() { - r = ref - break - } - } - - } - - if r == nil { - return tcpip.AddressWithPrefix{}, tcpip.ErrNoLinkAddress - } - - addressWithPrefix := tcpip.AddressWithPrefix{ - Address: r.ep.ID().LocalAddress, - PrefixLen: r.ep.PrefixLen(), - } - r.decRef() - - return addressWithPrefix, nil -} - // primaryEndpoint returns the primary endpoint of n for the given network // protocol. func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedNetworkEndpoint { @@ -398,10 +364,12 @@ func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpo return err } -// Addresses returns the addresses associated with this NIC. -func (n *NIC) Addresses() []tcpip.ProtocolAddress { +// AllAddresses returns all addresses (primary and non-primary) associated with +// this NIC. +func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { n.mu.RLock() defer n.mu.RUnlock() + addrs := make([]tcpip.ProtocolAddress, 0, len(n.endpoints)) for nid, ref := range n.endpoints { // Don't include expired or tempory endpoints to avoid confusion and @@ -421,6 +389,34 @@ func (n *NIC) Addresses() []tcpip.ProtocolAddress { return addrs } +// PrimaryAddresses returns the primary addresses associated with this NIC. +func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { + n.mu.RLock() + defer n.mu.RUnlock() + + var addrs []tcpip.ProtocolAddress + for proto, list := range n.primary { + for e := list.Front(); e != nil; e = e.Next() { + ref := e.(*referencedNetworkEndpoint) + // Don't include expired or tempory endpoints to avoid confusion and + // prevent the caller from using those. + switch ref.getKind() { + case permanentExpired, temporary: + continue + } + + addrs = append(addrs, tcpip.ProtocolAddress{ + Protocol: proto, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: ref.ep.ID().LocalAddress, + PrefixLen: ref.ep.PrefixLen(), + }, + }) + } + } + return addrs +} + // AddAddressRange adds a range of addresses to n, so that it starts accepting // packets targeted at the given addresses and network protocol. The range is // given by a subnet address, and all addresses contained in the subnet are @@ -708,9 +704,7 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // Raw socket packets are delivered based solely on the transport // protocol number. We do not inspect the payload to ensure it's // validly formed. - if !n.demux.deliverRawPacket(r, protocol, netHeader, vv) { - n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) - } + n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) if len(vv.First()) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() @@ -724,9 +718,6 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} - if n.demux.deliverPacket(r, protocol, netHeader, vv, id) { - return - } if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) { return } @@ -768,10 +759,7 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp } id := TransportEndpointID{srcPort, local, dstPort, remote} - if n.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { - return - } - if n.stack.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { + if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, vv, id) { return } } diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 07e4c770d..80101d4bb 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -366,14 +366,6 @@ type LinkAddressCache interface { RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) } -// TransportProtocolFactory functions are used by the stack to instantiate -// transport protocols. -type TransportProtocolFactory func() TransportProtocol - -// NetworkProtocolFactory provides methods to be used by the stack to -// instantiate network protocols. -type NetworkProtocolFactory func() NetworkProtocol - // UnassociatedEndpointFactory produces endpoints for writing packets not // associated with a particular transport protocol. Such endpoints can be used // to write arbitrary packets that include the IP header. @@ -381,34 +373,6 @@ type UnassociatedEndpointFactory interface { NewUnassociatedRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) } -var ( - transportProtocols = make(map[string]TransportProtocolFactory) - networkProtocols = make(map[string]NetworkProtocolFactory) - - unassociatedFactory UnassociatedEndpointFactory -) - -// RegisterTransportProtocolFactory registers a new transport protocol factory -// with the stack so that it becomes available to users of the stack. This -// function is intended to be called by init() functions of the protocols. -func RegisterTransportProtocolFactory(name string, p TransportProtocolFactory) { - transportProtocols[name] = p -} - -// RegisterNetworkProtocolFactory registers a new network protocol factory with -// the stack so that it becomes available to users of the stack. This function -// is intended to be called by init() functions of the protocols. -func RegisterNetworkProtocolFactory(name string, p NetworkProtocolFactory) { - networkProtocols[name] = p -} - -// RegisterUnassociatedFactory registers a factory to produce endpoints not -// associated with any particular transport protocol. This function is intended -// to be called by init() functions of the protocols. -func RegisterUnassociatedFactory(f UnassociatedEndpointFactory) { - unassociatedFactory = f -} - // GSOType is the type of GSO segments. // // +stateify savable diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 1fe21b68e..90c2cf1be 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -17,18 +17,15 @@ // // For consumers, the only function of interest is New(), everything else is // provided by the tcpip/public package. -// -// For protocol implementers, RegisterTransportProtocolFactory() and -// RegisterNetworkProtocolFactory() are used to register protocol factories with -// the stack, which will then be used to instantiate protocol objects when -// consumers interact with the stack. package stack import ( + "encoding/binary" "sync" "time" "golang.org/x/time/rate" + "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -351,6 +348,9 @@ type Stack struct { networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver + // unassociatedFactory creates unassociated endpoints. If nil, raw + // endpoints are disabled. It is set during Stack creation and is + // immutable. unassociatedFactory UnassociatedEndpointFactory demux *transportDemuxer @@ -359,10 +359,6 @@ type Stack struct { linkAddrCache *linkAddrCache - // raw indicates whether raw sockets may be created. It is set during - // Stack creation and is immutable. - raw bool - mu sync.RWMutex nics map[tcpip.NICID]*NIC forwarding bool @@ -394,10 +390,22 @@ type Stack struct { // icmpRateLimiter is a global rate limiter for all ICMP messages generated // by the stack. icmpRateLimiter *ICMPRateLimiter + + // portSeed is a one-time random value initialized at stack startup + // and is used to seed the TCP port picking on active connections + // + // TODO(gvisor.dev/issues/940): S/R this field. + portSeed uint32 } // Options contains optional Stack configuration. type Options struct { + // NetworkProtocols lists the network protocols to enable. + NetworkProtocols []NetworkProtocol + + // TransportProtocols lists the transport protocols to enable. + TransportProtocols []TransportProtocol + // Clock is an optional clock source used for timestampping packets. // // If no Clock is specified, the clock source will be time.Now. @@ -411,8 +419,9 @@ type Options struct { // stack (false). HandleLocal bool - // Raw indicates whether raw sockets may be created. - Raw bool + // UnassociatedFactory produces unassociated endpoints raw endpoints. + // Raw endpoints are enabled only if this is non-nil. + UnassociatedFactory UnassociatedEndpointFactory } // New allocates a new networking stack with only the requested networking and @@ -422,7 +431,7 @@ type Options struct { // SetNetworkProtocolOption/SetTransportProtocolOption methods provided by the // stack. Please refer to individual protocol implementations as to what options // are supported. -func New(network []string, transport []string, opts Options) *Stack { +func New(opts Options) *Stack { clock := opts.Clock if clock == nil { clock = &tcpip.StdClock{} @@ -438,17 +447,12 @@ func New(network []string, transport []string, opts Options) *Stack { clock: clock, stats: opts.Stats.FillIn(), handleLocal: opts.HandleLocal, - raw: opts.Raw, icmpRateLimiter: NewICMPRateLimiter(), + portSeed: generateRandUint32(), } // Add specified network protocols. - for _, name := range network { - netProtoFactory, ok := networkProtocols[name] - if !ok { - continue - } - netProto := netProtoFactory() + for _, netProto := range opts.NetworkProtocols { s.networkProtocols[netProto.Number()] = netProto if r, ok := netProto.(LinkAddressResolver); ok { s.linkAddrResolvers[r.LinkAddressProtocol()] = r @@ -456,18 +460,14 @@ func New(network []string, transport []string, opts Options) *Stack { } // Add specified transport protocols. - for _, name := range transport { - transProtoFactory, ok := transportProtocols[name] - if !ok { - continue - } - transProto := transProtoFactory() + for _, transProto := range opts.TransportProtocols { s.transportProtocols[transProto.Number()] = &transportProtocolState{ proto: transProto, } } - s.unassociatedFactory = unassociatedFactory + // Add the factory for unassociated endpoints, if present. + s.unassociatedFactory = opts.UnassociatedFactory // Create the global transport demuxer. s.demux = newTransportDemuxer(s) @@ -602,7 +602,7 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp // protocol. Raw endpoints receive all traffic for a given protocol regardless // of address. func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) { - if !s.raw { + if s.unassociatedFactory == nil { return nil, tcpip.ErrNotPermitted } @@ -738,7 +738,7 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { nics[id] = NICInfo{ Name: nic.name, LinkAddress: nic.linkEP.LinkAddress(), - ProtocolAddresses: nic.Addresses(), + ProtocolAddresses: nic.PrimaryAddresses(), Flags: flags, MTU: nic.linkEP.MTU(), Stats: nic.stats, @@ -845,19 +845,37 @@ func (s *Stack) RemoveAddress(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { return tcpip.ErrUnknownNICID } -// GetMainNICAddress returns the first primary address (and the subnet that -// contains it) for the given NIC and protocol. Returns an arbitrary endpoint's -// address if no primary addresses exist. Returns an error if the NIC doesn't -// exist or has no endpoints. +// AllAddresses returns a map of NICIDs to their protocol addresses (primary +// and non-primary). +func (s *Stack) AllAddresses() map[tcpip.NICID][]tcpip.ProtocolAddress { + s.mu.RLock() + defer s.mu.RUnlock() + + nics := make(map[tcpip.NICID][]tcpip.ProtocolAddress) + for id, nic := range s.nics { + nics[id] = nic.AllAddresses() + } + return nics +} + +// GetMainNICAddress returns the first primary address and prefix for the given +// NIC and protocol. Returns an error if the NIC doesn't exist and an empty +// value if the NIC doesn't have a primary address for the given protocol. func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocolNumber) (tcpip.AddressWithPrefix, *tcpip.Error) { s.mu.RLock() defer s.mu.RUnlock() - if nic, ok := s.nics[id]; ok { - return nic.getMainNICAddress(protocol) + nic, ok := s.nics[id] + if !ok { + return tcpip.AddressWithPrefix{}, tcpip.ErrUnknownNICID } - return tcpip.AddressWithPrefix{}, tcpip.ErrUnknownNICID + for _, a := range nic.PrimaryAddresses() { + if a.Protocol == protocol { + return a.AddressWithPrefix, nil + } + } + return tcpip.AddressWithPrefix{}, nil } func (s *Stack) getRefEP(nic *NIC, localAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) { @@ -1024,73 +1042,27 @@ func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep. // transport dispatcher. Received packets that match the provided id will be // delivered to the given endpoint; specifying a nic is optional, but // nic-specific IDs have precedence over global ones. -func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { - if nicID == 0 { - return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) +func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort, bindToDevice) } // UnregisterTransportEndpoint removes the endpoint with the given id from the // stack transport dispatcher. -func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) { - if nicID == 0 { - s.demux.unregisterEndpoint(netProtos, protocol, id, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterEndpoint(netProtos, protocol, id, ep) - } +func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) } // RegisterRawTransportEndpoint registers the given endpoint with the stack // transport dispatcher. Received packets that match the provided transport // protocol will be delivered to the given endpoint. func (s *Stack) RegisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error { - if nicID == 0 { - return s.demux.registerRawEndpoint(netProto, transProto, ep) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerRawEndpoint(netProto, transProto, ep) + return s.demux.registerRawEndpoint(netProto, transProto, ep) } // UnregisterRawTransportEndpoint removes the endpoint for the transport // protocol from the stack transport dispatcher. func (s *Stack) UnregisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) { - if nicID == 0 { - s.demux.unregisterRawEndpoint(netProto, transProto, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterRawEndpoint(netProto, transProto, ep) - } + s.demux.unregisterRawEndpoint(netProto, transProto, ep) } // RegisterRestoredEndpoint records e as an endpoint that has been restored on @@ -1234,3 +1206,19 @@ func (s *Stack) SetICMPBurst(burst int) { func (s *Stack) AllowICMPMessage() bool { return s.icmpRateLimiter.Allow() } + +// PortSeed returns a 32 bit value that can be used as a seed value for port +// picking. +// +// NOTE: The seed is generated once during stack initialization only. +func (s *Stack) PortSeed() uint32 { + return s.portSeed +} + +func generateRandUint32() uint32 { + b := make([]byte, 4) + if _, err := rand.Read(b); err != nil { + panic(err) + } + return binary.LittleEndian.Uint32(b) +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 0c26c9911..d2dede8a9 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -222,11 +222,17 @@ func (f *fakeNetworkProtocol) Option(option interface{}) *tcpip.Error { } } +func fakeNetFactory() stack.NetworkProtocol { + return &fakeNetworkProtocol{} +} + func TestNetworkReceive(t *testing.T) { // Create a stack with the fake network protocol, one nic, and two // addresses attached to it: 1 & 2. ep := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -370,7 +376,9 @@ func TestNetworkSend(t *testing.T) { // address: 1. The route table sends all packets through the only // existing nic. ep := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("NewNIC failed:", err) } @@ -395,7 +403,9 @@ func TestNetworkSendMultiRoute(t *testing.T) { // Create a stack with the fake network protocol, two nics, and two // addresses per nic, the first nic has odd address, the second one has // even addresses. - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -476,7 +486,9 @@ func TestRoutes(t *testing.T) { // Create a stack with the fake network protocol, two nics, and two // addresses per nic, the first nic has odd address, the second one has // even addresses. - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -554,7 +566,9 @@ func TestAddressRemoval(t *testing.T) { localAddr := tcpip.Address([]byte{localAddrByte}) remoteAddr := tcpip.Address("\x02") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -599,7 +613,9 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) { localAddr := tcpip.Address([]byte{localAddrByte}) remoteAddr := tcpip.Address("\x02") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -688,7 +704,9 @@ func TestEndpointExpiration(t *testing.T) { for _, promiscuous := range []bool{true, false} { for _, spoofing := range []bool{true, false} { t.Run(fmt.Sprintf("promiscuous=%t spoofing=%t", promiscuous, spoofing), func(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { @@ -844,7 +862,9 @@ func TestEndpointExpiration(t *testing.T) { } func TestPromiscuousMode(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -894,7 +914,9 @@ func TestSpoofingWithAddress(t *testing.T) { nonExistentLocalAddr := tcpip.Address("\x02") dstAddr := tcpip.Address("\x03") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -930,10 +952,10 @@ func TestSpoofingWithAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != nonExistentLocalAddr { - t.Errorf("Route has wrong local address: got %v, wanted %v", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("Route has wrong local address: got %s, want %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %v, wanted %v", r.RemoteAddress, dstAddr) + t.Errorf("Route has wrong remote address: got %s, want %s", r.RemoteAddress, dstAddr) } // Sending a packet works. testSendTo(t, s, dstAddr, ep, nil) @@ -945,10 +967,10 @@ func TestSpoofingWithAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != localAddr { - t.Errorf("Route has wrong local address: got %v, wanted %v", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("Route has wrong local address: got %s, want %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %v, wanted %v", r.RemoteAddress, dstAddr) + t.Errorf("Route has wrong remote address: got %s, want %s", r.RemoteAddress, dstAddr) } // Sending a packet using the route works. testSend(t, r, ep, nil) @@ -958,7 +980,9 @@ func TestSpoofingNoAddress(t *testing.T) { nonExistentLocalAddr := tcpip.Address("\x01") dstAddr := tcpip.Address("\x02") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -992,10 +1016,10 @@ func TestSpoofingNoAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != nonExistentLocalAddr { - t.Errorf("Route has wrong local address: got %v, wanted %v", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("Route has wrong local address: got %s, want %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %v, wanted %v", r.RemoteAddress, dstAddr) + t.Errorf("Route has wrong remote address: got %s, want %s", r.RemoteAddress, dstAddr) } // Sending a packet works. // FIXME(b/139841518):Spoofing doesn't work if there is no primary address. @@ -1003,7 +1027,9 @@ func TestSpoofingNoAddress(t *testing.T) { } func TestBroadcastNeedsNoRoute(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1074,7 +1100,9 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { {"IPv6 Unicast Not Link-Local 7", true, "\xf0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"}, } { t.Run(tc.name, func(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1130,7 +1158,9 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { // Add a range of addresses, then check that a packet is delivered. func TestAddressRangeAcceptsMatchingPacket(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1196,7 +1226,9 @@ func testNicForAddressRange(t *testing.T, nicID tcpip.NICID, s *stack.Stack, sub // existent. func TestCheckLocalAddressForSubnet(t *testing.T) { const nicID tcpip.NICID = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1234,7 +1266,9 @@ func TestCheckLocalAddressForSubnet(t *testing.T) { // Set a range of addresses, then send a packet to a destination outside the // range and then check it doesn't get delivered. func TestAddressRangeRejectsNonmatchingPacket(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1266,7 +1300,10 @@ func TestAddressRangeRejectsNonmatchingPacket(t *testing.T) { } func TestNetworkOptions(t *testing.T) { - s := stack.New([]string{"fakeNet"}, []string{}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{}, + }) // Try an unsupported network protocol. if err := s.SetNetworkProtocolOption(tcpip.NetworkProtocolNumber(99999), fakeNetGoodOption(false)); err != tcpip.ErrUnknownProtocol { @@ -1319,7 +1356,9 @@ func stackContainsAddressRange(s *stack.Stack, id tcpip.NICID, addrRange tcpip.S } func TestAddresRangeAddRemove(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1360,7 +1399,9 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { t.Run(fmt.Sprintf("canBe=%d", canBe), func(t *testing.T) { for never := 0; never < 3; never++ { t.Run(fmt.Sprintf("never=%d", never), func(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1400,20 +1441,20 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { // Check that GetMainNICAddress returns an address if at least // one primary address was added. In that case make sure the // address/prefixLen matches what we added. + gotAddr, err := s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("GetMainNICAddress failed:", err) + } if len(primaryAddrAdded) == 0 { - // No primary addresses present, expect an error. - if _, err := s.GetMainNICAddress(1, fakeNetNumber); err != tcpip.ErrNoLinkAddress { - t.Fatalf("got s.GetMainNICAddress(...) = %v, wanted = %s", err, tcpip.ErrNoLinkAddress) + // No primary addresses present. + if wantAddr := (tcpip.AddressWithPrefix{}); gotAddr != wantAddr { + t.Fatalf("GetMainNICAddress: got addr = %s, want = %s", gotAddr, wantAddr) } } else { - // At least one primary address was added, expect a valid - // address and prefixLen. - gotAddressWithPefix, err := s.GetMainNICAddress(1, fakeNetNumber) - if err != nil { - t.Fatal("GetMainNICAddress failed:", err) - } - if _, ok := primaryAddrAdded[gotAddressWithPefix]; !ok { - t.Fatalf("GetMainNICAddress: got addressWithPrefix = %v, wanted any in {%v}", gotAddressWithPefix, primaryAddrAdded) + // At least one primary address was added, verify the returned + // address is in the list of primary addresses we added. + if _, ok := primaryAddrAdded[gotAddr]; !ok { + t.Fatalf("GetMainNICAddress: got = %s, want any in {%v}", gotAddr, primaryAddrAdded) } } }) @@ -1425,7 +1466,9 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { } func TestGetMainNICAddressAddRemove(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1452,19 +1495,25 @@ func TestGetMainNICAddressAddRemove(t *testing.T) { } // Check that we get the right initial address and prefix length. - if gotAddressWithPrefix, err := s.GetMainNICAddress(1, fakeNetNumber); err != nil { + gotAddr, err := s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { t.Fatal("GetMainNICAddress failed:", err) - } else if gotAddressWithPrefix != protocolAddress.AddressWithPrefix { - t.Fatalf("got GetMainNICAddress = %+v, want = %+v", gotAddressWithPrefix, protocolAddress.AddressWithPrefix) + } + if wantAddr := protocolAddress.AddressWithPrefix; gotAddr != wantAddr { + t.Fatalf("got s.GetMainNICAddress(...) = %s, want = %s", gotAddr, wantAddr) } if err := s.RemoveAddress(1, protocolAddress.AddressWithPrefix.Address); err != nil { t.Fatal("RemoveAddress failed:", err) } - // Check that we get an error after removal. - if _, err := s.GetMainNICAddress(1, fakeNetNumber); err != tcpip.ErrNoLinkAddress { - t.Fatalf("got s.GetMainNICAddress(...) = %v, want = %s", err, tcpip.ErrNoLinkAddress) + // Check that we get no address after removal. + gotAddr, err = s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("GetMainNICAddress failed:", err) + } + if wantAddr := (tcpip.AddressWithPrefix{}); gotAddr != wantAddr { + t.Fatalf("got GetMainNICAddress(...) = %s, want = %s", gotAddr, wantAddr) } }) } @@ -1479,8 +1528,10 @@ func (g *addressGenerator) next(addrLen int) tcpip.Address { } func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.ProtocolAddress) { + t.Helper() + if len(gotAddresses) != len(expectedAddresses) { - t.Fatalf("got len(addresses) = %d, wanted = %d", len(gotAddresses), len(expectedAddresses)) + t.Fatalf("got len(addresses) = %d, want = %d", len(gotAddresses), len(expectedAddresses)) } sort.Slice(gotAddresses, func(i, j int) bool { @@ -1500,7 +1551,9 @@ func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.Proto func TestAddAddress(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1519,13 +1572,15 @@ func TestAddAddress(t *testing.T) { }) } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddProtocolAddress(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1551,13 +1606,15 @@ func TestAddProtocolAddress(t *testing.T) { } } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddAddressWithOptions(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1580,13 +1637,15 @@ func TestAddAddressWithOptions(t *testing.T) { } } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddProtocolAddressWithOptions(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1615,12 +1674,14 @@ func TestAddProtocolAddressWithOptions(t *testing.T) { } } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestNICStats(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { t.Fatal("CreateNIC failed: ", err) @@ -1666,7 +1727,9 @@ func TestNICStats(t *testing.T) { func TestNICForwarding(t *testing.T) { // Create a stack with the fake network protocol, two NICs, each with // an address. - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) s.SetForwarding(true) ep1 := channel.New(10, defaultMTU, "") @@ -1714,9 +1777,3 @@ func TestNICForwarding(t *testing.T) { t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want) } } - -func init() { - stack.RegisterNetworkProtocolFactory("fakeNet", func() stack.NetworkProtocol { - return &fakeNetworkProtocol{} - }) -} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index cf8a6d129..8c768c299 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -35,25 +35,109 @@ type protocolIDs struct { type transportEndpoints struct { // mu protects all fields of the transportEndpoints. mu sync.RWMutex - endpoints map[TransportEndpointID]TransportEndpoint + endpoints map[TransportEndpointID]*endpointsByNic // rawEndpoints contains endpoints for raw sockets, which receive all // traffic of a given protocol regardless of port. rawEndpoints []RawTransportEndpoint } +type endpointsByNic struct { + mu sync.RWMutex + endpoints map[tcpip.NICID]*multiPortEndpoint + // seed is a random secret for a jenkins hash. + seed uint32 +} + +// HandlePacket is called by the stack when new packets arrive to this transport +// endpoint. +func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { + epsByNic.mu.RLock() + + mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] + if !ok { + if mpep, ok = epsByNic.endpoints[0]; !ok { + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + } + + // If this is a broadcast or multicast datagram, deliver the datagram to all + // endpoints bound to the right device. + if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) { + mpep.handlePacketAll(r, id, vv) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, vv) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. +} + +// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. +func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { + epsByNic.mu.RLock() + defer epsByNic.mu.RUnlock() + + mpep, ok := epsByNic.endpoints[n.ID()] + if !ok { + mpep, ok = epsByNic.endpoints[0] + } + if !ok { + return + } + + // TODO(eyalsoha): Why don't we look at id to see if this packet needs to + // broadcast like we are doing with handlePacket above? + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, vv) +} + +// registerEndpoint returns true if it succeeds. It fails and returns +// false if ep already has an element with the same key. +func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + + if multiPortEp, ok := epsByNic.endpoints[bindToDevice]; ok { + // There was already a bind. + return multiPortEp.singleRegisterEndpoint(t, reusePort) + } + + // This is a new binding. + multiPortEp := &multiPortEndpoint{} + multiPortEp.endpointsMap = make(map[TransportEndpoint]int) + multiPortEp.reuse = reusePort + epsByNic.endpoints[bindToDevice] = multiPortEp + return multiPortEp.singleRegisterEndpoint(t, reusePort) +} + +// unregisterEndpoint returns true if endpointsByNic has to be unregistered. +func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint) bool { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + multiPortEp, ok := epsByNic.endpoints[bindToDevice] + if !ok { + return false + } + if multiPortEp.unregisterEndpoint(t) { + delete(epsByNic.endpoints, bindToDevice) + } + return len(epsByNic.endpoints) == 0 +} + // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint) { +func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { eps.mu.Lock() defer eps.mu.Unlock() - e, ok := eps.endpoints[id] + epsByNic, ok := eps.endpoints[id] if !ok { return } - if multiPortEp, ok := e.(*multiPortEndpoint); ok { - if !multiPortEp.unregisterEndpoint(ep) { - return - } + if !epsByNic.unregisterEndpoint(bindToDevice, ep) { + return } delete(eps.endpoints, id) } @@ -75,7 +159,7 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { for netProto := range stack.networkProtocols { for proto := range stack.transportProtocols { d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{ - endpoints: make(map[TransportEndpointID]TransportEndpoint), + endpoints: make(map[TransportEndpointID]*endpointsByNic), } } } @@ -85,10 +169,10 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { // registerEndpoint registers the given endpoint with the dispatcher such that // packets that match the endpoint ID are delivered to it. -func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { for i, n := range netProtos { - if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort); err != nil { - d.unregisterEndpoint(netProtos[:i], protocol, id, ep) + if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort, bindToDevice); err != nil { + d.unregisterEndpoint(netProtos[:i], protocol, id, ep, bindToDevice) return err } } @@ -97,13 +181,14 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum } // multiPortEndpoint is a container for TransportEndpoints which are bound to -// the same pair of address and port. +// the same pair of address and port. endpointsArr always has at least one +// element. type multiPortEndpoint struct { mu sync.RWMutex endpointsArr []TransportEndpoint endpointsMap map[TransportEndpoint]int - // seed is a random secret for a jenkins hash. - seed uint32 + // reuse indicates if more than one endpoint is allowed. + reuse bool } // reciprocalScale scales a value into range [0, n). @@ -117,9 +202,10 @@ func reciprocalScale(val, n uint32) uint32 { // selectEndpoint calculates a hash of destination and source addresses and // ports then uses it to select a socket. In this case, all packets from one // address will be sent to same endpoint. -func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEndpoint { - ep.mu.RLock() - defer ep.mu.RUnlock() +func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32) TransportEndpoint { + if len(mpep.endpointsArr) == 1 { + return mpep.endpointsArr[0] + } payload := []byte{ byte(id.LocalPort), @@ -128,51 +214,50 @@ func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEnd byte(id.RemotePort >> 8), } - h := jenkins.Sum32(ep.seed) + h := jenkins.Sum32(seed) h.Write(payload) h.Write([]byte(id.LocalAddress)) h.Write([]byte(id.RemoteAddress)) hash := h.Sum32() - idx := reciprocalScale(hash, uint32(len(ep.endpointsArr))) - return ep.endpointsArr[idx] + idx := reciprocalScale(hash, uint32(len(mpep.endpointsArr))) + return mpep.endpointsArr[idx] } -// HandlePacket is called by the stack when new packets arrive to this transport -// endpoint. -func (ep *multiPortEndpoint) HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { - // If this is a broadcast or multicast datagram, deliver the datagram to all - // endpoints managed by ep. - if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) { - for i, endpoint := range ep.endpointsArr { - // HandlePacket modifies vv, so each endpoint needs its own copy. - if i == len(ep.endpointsArr)-1 { - endpoint.HandlePacket(r, id, vv) - break - } - vvCopy := buffer.NewView(vv.Size()) - copy(vvCopy, vv.ToView()) - endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { + ep.mu.RLock() + for i, endpoint := range ep.endpointsArr { + // HandlePacket modifies vv, so each endpoint needs its own copy except for + // the final one. + if i == len(ep.endpointsArr)-1 { + endpoint.HandlePacket(r, id, vv) + break } - } else { - ep.selectEndpoint(id).HandlePacket(r, id, vv) + vvCopy := buffer.NewView(vv.Size()) + copy(vvCopy, vv.ToView()) + endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) } + ep.mu.RUnlock() // Don't use defer for performance reasons. } -// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (ep *multiPortEndpoint) HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { - ep.selectEndpoint(id).HandleControlPacket(id, typ, extra, vv) -} - -func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint) { +// singleRegisterEndpoint tries to add an endpoint to the multiPortEndpoint +// list. The list might be empty already. +func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePort bool) *tcpip.Error { ep.mu.Lock() defer ep.mu.Unlock() - // A new endpoint is added into endpointsArr and its index there is - // saved in endpointsMap. This will allows to remove endpoint from - // the array fast. + if len(ep.endpointsArr) > 0 { + // If it was previously bound, we need to check if we can bind again. + if !ep.reuse || !reusePort { + return tcpip.ErrPortInUse + } + } + + // A new endpoint is added into endpointsArr and its index there is saved in + // endpointsMap. This will allow us to remove endpoint from the array fast. ep.endpointsMap[t] = len(ep.endpointsArr) ep.endpointsArr = append(ep.endpointsArr, t) + return nil } // unregisterEndpoint returns true if multiPortEndpoint has to be unregistered. @@ -197,53 +282,41 @@ func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint) bool { return true } -func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { if id.RemotePort != 0 { + // TODO(eyalsoha): Why? reusePort = false } eps, ok := d.protocol[protocolIDs{netProto, protocol}] if !ok { - return nil + return tcpip.ErrUnknownProtocol } eps.mu.Lock() defer eps.mu.Unlock() - var multiPortEp *multiPortEndpoint - if _, ok := eps.endpoints[id]; ok { - if !reusePort { - return tcpip.ErrPortInUse - } - multiPortEp, ok = eps.endpoints[id].(*multiPortEndpoint) - if !ok { - return tcpip.ErrPortInUse - } + if epsByNic, ok := eps.endpoints[id]; ok { + // There was already a binding. + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } - if reusePort { - if multiPortEp == nil { - multiPortEp = &multiPortEndpoint{} - multiPortEp.endpointsMap = make(map[TransportEndpoint]int) - multiPortEp.seed = rand.Uint32() - eps.endpoints[id] = multiPortEp - } - - multiPortEp.singleRegisterEndpoint(ep) - - return nil + // This is a new binding. + epsByNic := &endpointsByNic{ + endpoints: make(map[tcpip.NICID]*multiPortEndpoint), + seed: rand.Uint32(), } - eps.endpoints[id] = ep + eps.endpoints[id] = epsByNic - return nil + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) { +func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { for _, n := range netProtos { if eps, ok := d.protocol[protocolIDs{n, protocol}]; ok { - eps.unregisterEndpoint(id, ep) + eps.unregisterEndpoint(id, ep, bindToDevice) } } } @@ -273,7 +346,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // If the packet is a broadcast, then find all matching transport endpoints. // Otherwise, try to find a single matching transport endpoint. - destEps := make([]TransportEndpoint, 0, 1) + destEps := make([]*endpointsByNic, 0, 1) eps.mu.RLock() if protocol == header.UDPProtocolNumber && id.LocalAddress == header.IPv4Broadcast { @@ -299,7 +372,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // Deliver the packet. for _, ep := range destEps { - ep.HandlePacket(r, id, vv) + ep.handlePacket(r, id, vv) } return true @@ -331,7 +404,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr // deliverControlPacket attempts to deliver the given control packet. Returns // true if it found an endpoint, false otherwise. -func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{net, trans}] if !ok { return false @@ -348,12 +421,12 @@ func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, } // Deliver the packet. - ep.HandleControlPacket(id, typ, extra, vv) + ep.handleControlPacket(n, id, typ, extra, vv) return true } -func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) TransportEndpoint { +func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) *endpointsByNic { // Try to find a match with the id as provided. if ep, ok := eps.endpoints[id]; ok { return ep diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go new file mode 100644 index 000000000..210233dc0 --- /dev/null +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -0,0 +1,352 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack_test + +import ( + "math" + "math/rand" + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/udp" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + stackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + testV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + + stackAddr = "\x0a\x00\x00\x01" + stackPort = 1234 + testPort = 4096 +) + +type testContext struct { + t *testing.T + linkEPs map[string]*channel.Endpoint + s *stack.Stack + + ep tcpip.Endpoint + wq waiter.Queue +} + +func (c *testContext) cleanup() { + if c.ep != nil { + c.ep.Close() + } +} + +func (c *testContext) createV6Endpoint(v6only bool) { + var err *tcpip.Error + c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + + var v tcpip.V6OnlyOption + if v6only { + v = 1 + } + if err := c.ep.SetSockOpt(v); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } +} + +// newDualTestContextMultiNic creates the testing context and also linkEpNames +// named NICs. +func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + linkEPs := make(map[string]*channel.Endpoint) + for i, linkEpName := range linkEpNames { + channelEP := channel.New(256, mtu, "") + nicid := tcpip.NICID(i + 1) + if err := s.CreateNamedNIC(nicid, linkEpName, channelEP); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + linkEPs[linkEpName] = channelEP + + if err := s.AddAddress(nicid, ipv4.ProtocolNumber, stackAddr); err != nil { + t.Fatalf("AddAddress IPv4 failed: %v", err) + } + + if err := s.AddAddress(nicid, ipv6.ProtocolNumber, stackV6Addr); err != nil { + t.Fatalf("AddAddress IPv6 failed: %v", err) + } + } + + s.SetRouteTable([]tcpip.Route{ + { + Destination: header.IPv4EmptySubnet, + NIC: 1, + }, + { + Destination: header.IPv6EmptySubnet, + NIC: 1, + }, + }) + + return &testContext{ + t: t, + s: s, + linkEPs: linkEPs, + } +} + +type headers struct { + srcPort uint16 + dstPort uint16 +} + +func newPayload() []byte { + b := make([]byte, 30+rand.Intn(100)) + for i := range b { + b[i] = byte(rand.Intn(256)) + } + return b +} + +func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) { + // Allocate a buffer for data and headers. + buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) + copy(buf[len(buf)-len(payload):], payload) + + // Initialize the IP header. + ip := header.IPv6(buf) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(header.UDPMinimumSize + len(payload)), + NextHeader: uint8(udp.ProtocolNumber), + HopLimit: 65, + SrcAddr: testV6Addr, + DstAddr: stackV6Addr, + }) + + // Initialize the UDP header. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.Encode(&header.UDPFields{ + SrcPort: h.srcPort, + DstPort: h.dstPort, + Length: uint16(header.UDPMinimumSize + len(payload)), + }) + + // Calculate the UDP pseudo-header checksum. + xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testV6Addr, stackV6Addr, uint16(len(u))) + + // Calculate the UDP checksum and set it. + xsum = header.Checksum(payload, xsum) + u.SetChecksum(^u.CalculateChecksum(xsum)) + + // Inject packet. + c.linkEPs[linkEpName].Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) +} + +func TestTransportDemuxerRegister(t *testing.T) { + for _, test := range []struct { + name string + proto tcpip.NetworkProtocolNumber + want *tcpip.Error + }{ + {"failure", ipv6.ProtocolNumber, tcpip.ErrUnknownProtocol}, + {"success", ipv4.ProtocolNumber, nil}, + } { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, nil, false, 0), test.want; got != want { + t.Fatalf("s.RegisterTransportEndpoint(...) = %v, want %v", got, want) + } + }) + } +} + +// TestReuseBindToDevice injects varied packets on input devices and checks that +// the distribution of packets received matches expectations. +func TestDistribution(t *testing.T) { + type endpointSockopts struct { + reuse int + bindToDevice string + } + for _, test := range []struct { + name string + // endpoints will received the inject packets. + endpoints []endpointSockopts + // wantedDistribution is the wanted ratio of packets received on each + // endpoint for each NIC on which packets are injected. + wantedDistributions map[string][]float64 + }{ + { + "BindPortReuse", + // 5 endpoints that all have reuse set. + []endpointSockopts{ + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed evenly. + "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2}, + }, + }, + { + "BindToDevice", + // 3 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{0, "dev0"}, + endpointSockopts{0, "dev1"}, + endpointSockopts{0, "dev2"}, + }, + map[string][]float64{ + // Injected packets on dev0 go only to the endpoint bound to dev0. + "dev0": []float64{1, 0, 0}, + // Injected packets on dev1 go only to the endpoint bound to dev1. + "dev1": []float64{0, 1, 0}, + // Injected packets on dev2 go only to the endpoint bound to dev2. + "dev2": []float64{0, 0, 1}, + }, + }, + { + "ReuseAndBindToDevice", + // 6 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed among endpoints bound to + // dev0. + "dev0": []float64{0.5, 0.5, 0, 0, 0, 0}, + // Injected packets on dev1 get distributed among endpoints bound to + // dev1 or unbound. + "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, + // Injected packets on dev999 go only to the unbound. + "dev999": []float64{0, 0, 0, 0, 0, 1}, + }, + }, + } { + t.Run(test.name, func(t *testing.T) { + for device, wantedDistribution := range test.wantedDistributions { + t.Run(device, func(t *testing.T) { + var devices []string + for d := range test.wantedDistributions { + devices = append(devices, d) + } + c := newDualTestContextMultiNic(t, defaultMTU, devices) + defer c.cleanup() + + c.createV6Endpoint(false) + + eps := make(map[tcpip.Endpoint]int) + + pollChannel := make(chan tcpip.Endpoint) + for i, endpoint := range test.endpoints { + // 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 + ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + eps[ep] = i + + go func(ep tcpip.Endpoint) { + for range ch { + pollChannel <- ep + } + }(ep) + + defer ep.Close() + reusePortOption := tcpip.ReusePortOption(endpoint.reuse) + if err := ep.SetSockOpt(reusePortOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", reusePortOption, i, err) + } + bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) + if err := ep.SetSockOpt(bindToDeviceOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", bindToDeviceOption, i, err) + } + if err := ep.Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { + t.Fatalf("ep.Bind(...) on endpoint %d failed: %v", i, err) + } + } + + npackets := 100000 + nports := 10000 + if got, want := len(test.endpoints), len(wantedDistribution); got != want { + t.Fatalf("got len(test.endpoints) = %d, want %d", got, want) + } + 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.sendV6Packet(payload, + &headers{ + srcPort: testPort + port, + dstPort: stackPort}, + device) + + var addr tcpip.FullAddress + ep := <-pollChannel + _, _, err := ep.Read(&addr) + if err != nil { + c.t.Fatalf("Read on endpoint %d failed: %v", eps[ep], 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 want, got := ports[port], ep; want != got { + t.Fatalf("Packet sent on port %d expected on endpoint %d but received on endpoint %d", port, eps[want], eps[got]) + } + } + } + + // Check that a packet distribution is as expected. + for ep, i := range eps { + wantedRatio := wantedDistribution[i] + wantedRecv := wantedRatio * float64(npackets) + actualRecv := stats[ep] + actualRatio := float64(stats[ep]) / float64(npackets) + // The deviation is less than 10%. + if math.Abs(actualRatio-wantedRatio) > 0.05 { + t.Errorf("wanted about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantedRatio*100, wantedRecv, npackets, i, actualRatio*100, actualRecv, npackets) + } + } + }) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 0e69ac7c8..842a16277 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -127,7 +127,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Try to register so that we can start receiving packets. f.id.RemoteAddress = addr.Addr - err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false) + err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false /* reuse */, 0 /* bindToDevice */) if err != nil { return err } @@ -168,7 +168,8 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { fakeTransNumber, stack.TransportEndpointID{LocalAddress: a.Addr}, f, - false, + false, /* reuse */ + 0, /* bindtoDevice */ ); err != nil { return err } @@ -282,9 +283,16 @@ func (f *fakeTransportProtocol) Option(option interface{}) *tcpip.Error { } } +func fakeTransFactory() stack.TransportProtocol { + return &fakeTransportProtocol{} +} + func TestTransportReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -346,7 +354,10 @@ func TestTransportReceive(t *testing.T) { func TestTransportControlReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -414,7 +425,10 @@ func TestTransportControlReceive(t *testing.T) { func TestTransportSend(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -457,7 +471,10 @@ func TestTransportSend(t *testing.T) { } func TestTransportOptions(t *testing.T) { - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) // Try an unsupported transport protocol. if err := s.SetTransportProtocolOption(tcpip.TransportProtocolNumber(99999), fakeTransportGoodOption(false)); err != tcpip.ErrUnknownProtocol { @@ -498,7 +515,10 @@ func TestTransportOptions(t *testing.T) { } func TestTransportForwarding(t *testing.T) { - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) s.SetForwarding(true) // TODO(b/123449044): Change this to a channel NIC. @@ -576,9 +596,3 @@ func TestTransportForwarding(t *testing.T) { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } - -func init() { - stack.RegisterTransportProtocolFactory("fakeTrans", func() stack.TransportProtocol { - return &fakeTransportProtocol{} - }) -} diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index c021c67ac..faaa4a4e3 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -495,6 +495,10 @@ type ReuseAddressOption int // to be bound to an identical socket address. type ReusePortOption int +// BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets +// should bind only on a specific NIC. +type BindToDeviceOption string + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. type QuickAckOption int diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index a111fdb2a..a3a910d41 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -104,7 +104,7 @@ 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. @@ -543,14 +543,14 @@ 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 1eb790932..bfb16f7c3 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -14,10 +14,9 @@ // 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 @@ -34,15 +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 ) @@ -125,12 +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/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 39a839ab7..a42e1f4a2 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -49,6 +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 0802e984e..3ae4a5426 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -242,7 +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 } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 35b489c68..f9d5e0085 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" @@ -280,6 +282,9 @@ 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 +569,11 @@ 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 +630,12 @@ 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 } @@ -1060,6 +1065,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { 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) @@ -1260,6 +1280,16 @@ 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 { @@ -1466,7 +1496,7 @@ 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 } @@ -1476,17 +1506,32 @@ 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 @@ -1504,7 +1549,7 @@ 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 } @@ -1648,7 +1693,7 @@ 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 } @@ -1729,7 +1774,7 @@ 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 } @@ -1739,16 +1784,16 @@ 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 2a13b2022..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 @@ -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/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 7fa5cfb6e..089826a88 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -465,6 +465,66 @@ 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() @@ -2873,7 +2933,10 @@ 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{}) @@ -2919,7 +2982,10 @@ 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{}) @@ -2964,11 +3030,70 @@ func TestMinMaxBufferSizes(t *testing.T) { 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() { diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 78eff5c3a..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 { @@ -155,7 +158,14 @@ func New(t *testing.T, mtu uint32) *Context { if testing.Verbose() { wep = sniffer.New(ep) } - if err := s.CreateNIC(1, wep); err != nil { + 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() { + wep2 = sniffer.New(channel.New(1000, mtu, "")) + } + if err := s.CreateNamedNIC(2, "nic2", wep2); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -585,12 +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 int, 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) @@ -603,6 +609,15 @@ func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum. 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/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c1ca22b35..7a635ab8d 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -52,6 +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 0bec7e62d..52f5af777 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -88,6 +88,7 @@ type endpoint struct { multicastNICID tcpip.NICID multicastLoop bool reusePort bool + bindToDevice tcpip.NICID broadcast bool // shutdownFlags represent the current shutdown state of the endpoint. @@ -144,8 +145,8 @@ func (e *endpoint) Close() { 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) + 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 { @@ -551,6 +552,21 @@ 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 @@ -646,6 +662,16 @@ 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 @@ -753,12 +779,12 @@ func (e *endpoint) Disconnect() *tcpip.Error { } else { if e.id.LocalPort != 0 { // Release the ephemeral port. - 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.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{} @@ -835,7 +861,7 @@ 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 @@ -898,16 +924,16 @@ 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 } diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a9edc2c8d..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 } diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 068d9a272..f5cc932dd 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 ) @@ -182,8 +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 c6deab892..5059ca22d 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" @@ -274,7 +274,10 @@ 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) @@ -473,87 +476,59 @@ func newMinPayload(minSize int) []byte { 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) + } + }) } } |