diff options
Diffstat (limited to 'pkg/tcpip')
54 files changed, 1046 insertions, 414 deletions
diff --git a/pkg/tcpip/header/icmpv4.go b/pkg/tcpip/header/icmpv4.go index 7908c5744..1a631b31a 100644 --- a/pkg/tcpip/header/icmpv4.go +++ b/pkg/tcpip/header/icmpv4.go @@ -72,6 +72,7 @@ const ( // Values for ICMP code as defined in RFC 792. const ( ICMPv4TTLExceeded = 0 + ICMPv4HostUnreachable = 1 ICMPv4PortUnreachable = 3 ICMPv4FragmentationNeeded = 4 ) diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index c7ee2de57..a13b4b809 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -110,9 +110,16 @@ const ( ICMPv6RedirectMsg ICMPv6Type = 137 ) -// Values for ICMP code as defined in RFC 4443. +// Values for ICMP destination unreachable code as defined in RFC 4443 section +// 3.1. const ( - ICMPv6PortUnreachable = 4 + ICMPv6NetworkUnreachable = 0 + ICMPv6Prohibited = 1 + ICMPv6BeyondScope = 2 + ICMPv6AddressUnreachable = 3 + ICMPv6PortUnreachable = 4 + ICMPv6Policy = 5 + ICMPv6RejectRoute = 6 ) // Type is the ICMP type field. diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index a2bb773d4..e12a5929b 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -302,3 +302,7 @@ func (e *Endpoint) RemoveNotify(handle *NotificationHandle) { func (*Endpoint) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareNone } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 6aa1badc7..c18bb91fb 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -386,26 +386,33 @@ const ( _VIRTIO_NET_HDR_GSO_TCPV6 = 4 ) -// WritePacket writes outbound packets to the file descriptor. If it is not -// currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { if e.hdrSize > 0 { // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, + DstAddr: remote, Type: protocol, } // Preserve the src address if it's set in the route. - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress + if local != "" { + ethHdr.SrcAddr = local } else { ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) } +} + +// WritePacket writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + if e.hdrSize > 0 { + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) + } var builder iovec.Builder @@ -448,22 +455,8 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc // Send a batch of packets through batchFD. mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch)) for _, pkt := range batch { - var eth header.Ethernet if e.hdrSize > 0 { - // Add ethernet header if needed. - eth = make(header.Ethernet, header.EthernetMinimumSize) - ethHdr := &header.EthernetFields{ - DstAddr: pkt.EgressRoute.RemoteLinkAddress, - Type: pkt.NetworkProtocolNumber, - } - - // Preserve the src address if it's set in the route. - if pkt.EgressRoute.LocalLinkAddress != "" { - ethHdr.SrcAddr = pkt.EgressRoute.LocalLinkAddress - } else { - ethHdr.SrcAddr = e.addr - } - eth.Encode(ethHdr) + e.AddHeader(pkt.EgressRoute.LocalLinkAddress, pkt.EgressRoute.RemoteLinkAddress, pkt.NetworkProtocolNumber, pkt) } var vnetHdrBuf []byte @@ -493,7 +486,6 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc var builder iovec.Builder builder.Add(vnetHdrBuf) - builder.Add(eth) builder.Add(pkt.Header.View()) for _, v := range pkt.Data.Views() { builder.Add(v) diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 4bad930c7..7b995b85a 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -107,6 +107,10 @@ func (c *context) DeliverNetworkPacket(remote tcpip.LinkAddress, local tcpip.Lin c.ch <- packetInfo{remote, protocol, pkt} } +func (c *context) DeliverOutboundPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestNoEthernetProperties(t *testing.T) { c := newContext(t, &Options{MTU: mtu}) defer c.cleanup() @@ -510,6 +514,10 @@ func (d *fakeNetworkDispatcher) DeliverNetworkPacket(remote, local tcpip.LinkAdd d.pkts = append(d.pkts, pkt) } +func (d *fakeNetworkDispatcher) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestDispatchPacketFormat(t *testing.T) { for _, test := range []struct { name string diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 3b17d8c28..781cdd317 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -118,3 +118,6 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { func (*endpoint) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareLoopback } + +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index c305d9e86..56a611825 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -135,6 +135,10 @@ func (*InjectableEndpoint) ARPHardwareType() header.ARPHardwareType { panic("unsupported operation") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (*InjectableEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} + // NewInjectableEndpoint creates a new multi-endpoint injectable endpoint. func NewInjectableEndpoint(routes map[tcpip.Address]stack.InjectableLinkEndpoint) *InjectableEndpoint { return &InjectableEndpoint{ diff --git a/pkg/tcpip/link/nested/nested.go b/pkg/tcpip/link/nested/nested.go index 328bd048e..d40de54df 100644 --- a/pkg/tcpip/link/nested/nested.go +++ b/pkg/tcpip/link/nested/nested.go @@ -61,6 +61,16 @@ func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco } } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *Endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.mu.RLock() + d := e.dispatcher + e.mu.RUnlock() + if d != nil { + d.DeliverOutboundPacket(remote, local, protocol, pkt) + } +} + // Attach implements stack.LinkEndpoint. func (e *Endpoint) Attach(dispatcher stack.NetworkDispatcher) { e.mu.Lock() @@ -135,3 +145,8 @@ func (e *Endpoint) GSOMaxSize() uint32 { func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { return e.child.ARPHardwareType() } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.child.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/nested/nested_test.go b/pkg/tcpip/link/nested/nested_test.go index c1a219f02..7d9249c1c 100644 --- a/pkg/tcpip/link/nested/nested_test.go +++ b/pkg/tcpip/link/nested/nested_test.go @@ -55,6 +55,10 @@ func (d *counterDispatcher) DeliverNetworkPacket(tcpip.LinkAddress, tcpip.LinkAd d.count++ } +func (d *counterDispatcher) DeliverOutboundPacket(tcpip.LinkAddress, tcpip.LinkAddress, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) { + panic("unimplemented") +} + func TestNestedLinkEndpoint(t *testing.T) { const emptyAddress = tcpip.LinkAddress("") diff --git a/pkg/tcpip/link/packetsocket/BUILD b/pkg/tcpip/link/packetsocket/BUILD new file mode 100644 index 000000000..6fff160ce --- /dev/null +++ b/pkg/tcpip/link/packetsocket/BUILD @@ -0,0 +1,14 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "packetsocket", + srcs = ["endpoint.go"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/link/nested", + "//pkg/tcpip/stack", + ], +) diff --git a/pkg/tcpip/link/packetsocket/endpoint.go b/pkg/tcpip/link/packetsocket/endpoint.go new file mode 100644 index 000000000..3922c2a04 --- /dev/null +++ b/pkg/tcpip/link/packetsocket/endpoint.go @@ -0,0 +1,50 @@ +// Copyright 2020 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 packetsocket provides a link layer endpoint that provides the ability +// to loop outbound packets to any AF_PACKET sockets that may be interested in +// the outgoing packet. +package packetsocket + +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/link/nested" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +type endpoint struct { + nested.Endpoint +} + +// New creates a new packetsocket LinkEndpoint. +func New(lower stack.LinkEndpoint) stack.LinkEndpoint { + e := &endpoint{} + e.Endpoint.Init(lower, e) + return e +} + +// WritePacket implements stack.LinkEndpoint.WritePacket. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.Endpoint.DeliverOutboundPacket(r.RemoteLinkAddress, r.LocalLinkAddress, protocol, pkt) + return e.Endpoint.WritePacket(r, gso, protocol, pkt) +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, proto tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + e.Endpoint.DeliverOutboundPacket(pkt.EgressRoute.RemoteLinkAddress, pkt.EgressRoute.LocalLinkAddress, pkt.NetworkProtocolNumber, pkt) + } + + return e.Endpoint.WritePackets(r, gso, pkts, proto) +} diff --git a/pkg/tcpip/link/qdisc/fifo/endpoint.go b/pkg/tcpip/link/qdisc/fifo/endpoint.go index c84fe1bb9..467083239 100644 --- a/pkg/tcpip/link/qdisc/fifo/endpoint.go +++ b/pkg/tcpip/link/qdisc/fifo/endpoint.go @@ -107,6 +107,11 @@ func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.dispatcher.DeliverNetworkPacket(remote, local, protocol, pkt) } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.dispatcher.DeliverOutboundPacket(remote, local, protocol, pkt) +} + // Attach implements stack.LinkEndpoint.Attach. func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher @@ -194,6 +199,8 @@ func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketB // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + // TODO(gvisor.dev/issue/3267/): Queue these packets as well once + // WriteRawPacket takes PacketBuffer instead of VectorisedView. return e.lower.WriteRawPacket(vv) } @@ -213,3 +220,8 @@ func (e *endpoint) Wait() { func (e *endpoint) ARPHardwareType() header.ARPHardwareType { return e.lower.ARPHardwareType() } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.lower.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index a36862c67..507c76b76 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -183,22 +183,29 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { return e.addr } -// WritePacket writes outbound packets to the file descriptor. If it is not -// currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - // Add the ethernet header here. +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, + DstAddr: remote, Type: protocol, } - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress + + // Preserve the src address if it's set in the route. + if local != "" { + ethHdr.SrcAddr = local } else { ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) +} + +// WritePacket writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) v := pkt.Data.ToView() // Transmit the packet. diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 28a2e88ba..8f3cd9449 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -143,6 +143,10 @@ func (c *testContext) DeliverNetworkPacket(remoteLinkAddr, localLinkAddr tcpip.L c.packetCh <- struct{}{} } +func (c *testContext) DeliverOutboundPacket(remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func (c *testContext) cleanup() { c.ep.Close() closeFDs(&c.txCfg) diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index d9cd4e83a..509076643 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -123,6 +123,11 @@ func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.Endpoint.DeliverNetworkPacket(remote, local, protocol, pkt) } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.Endpoint.DeliverOutboundPacket(remote, local, protocol, pkt) +} + func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 47446efec..04ae58e59 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -272,21 +272,9 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { if d.hasFlags(linux.IFF_TAP) { // Add ethernet header if not provided. if info.Pkt.LinkHeader == nil { - hdr := &header.EthernetFields{ - SrcAddr: info.Route.LocalLinkAddress, - DstAddr: info.Route.RemoteLinkAddress, - Type: info.Proto, - } - if hdr.SrcAddr == "" { - hdr.SrcAddr = d.endpoint.LinkAddress() - } - - eth := make(header.Ethernet, header.EthernetMinimumSize) - eth.Encode(hdr) - vv.AppendView(buffer.View(eth)) - } else { - vv.AppendView(info.Pkt.LinkHeader) + d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress, info.Proto, info.Pkt) } + vv.AppendView(info.Pkt.LinkHeader) } // Append upper headers. @@ -366,3 +354,30 @@ func (e *tunEndpoint) ARPHardwareType() header.ARPHardwareType { } return header.ARPHardwareNone } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *tunEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + if !e.isTap { + return + } + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) + hdr := &header.EthernetFields{ + SrcAddr: local, + DstAddr: remote, + Type: protocol, + } + if hdr.SrcAddr == "" { + hdr.SrcAddr = e.LinkAddress() + } + + eth.Encode(hdr) +} + +// MaxHeaderLength returns the maximum size of the link layer header. +func (e *tunEndpoint) MaxHeaderLength() uint16 { + if e.isTap { + return header.EthernetMinimumSize + } + return 0 +} diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 24a8dc2eb..b152a0f26 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -60,6 +60,15 @@ func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.dispatchGate.Leave() } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *Endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + if !e.dispatchGate.Enter() { + return + } + e.dispatcher.DeliverOutboundPacket(remote, local, protocol, pkt) + e.dispatchGate.Leave() +} + // Attach implements stack.LinkEndpoint.Attach. It saves the dispatcher and // registers with the lower endpoint as its dispatcher so that "e" is called // for inbound packets. @@ -153,3 +162,8 @@ func (e *Endpoint) Wait() {} func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { return e.lower.ARPHardwareType() } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.lower.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index ffb2354be..c448a888f 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -40,6 +40,10 @@ func (e *countedEndpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, e.dispatchCount++ } +func (e *countedEndpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func (e *countedEndpoint) Attach(dispatcher stack.NetworkDispatcher) { e.attachCount++ e.dispatcher = dispatcher @@ -90,6 +94,11 @@ func (*countedEndpoint) ARPHardwareType() header.ARPHardwareType { // Wait implements stack.LinkEndpoint.Wait. func (*countedEndpoint) Wait() {} +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *countedEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestWaitWrite(t *testing.T) { ep := &countedEndpoint{} wep := New(ep) diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index b0f57040c..31a242482 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -160,9 +160,12 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { } // LinkAddressRequest implements stack.LinkAddressResolver.LinkAddressRequest. -func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack.LinkEndpoint) *tcpip.Error { +func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP stack.LinkEndpoint) *tcpip.Error { r := &stack.Route{ - RemoteLinkAddress: header.EthernetBroadcastAddress, + RemoteLinkAddress: remoteLinkAddr, + } + if len(r.RemoteLinkAddress) == 0 { + r.RemoteLinkAddress = header.EthernetBroadcastAddress } hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.ARPSize) diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 66e67429c..a35a64a0f 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -32,10 +32,14 @@ import ( ) const ( - stackLinkAddr = tcpip.LinkAddress("\x0a\x0a\x0b\x0b\x0c\x0c") - stackAddr1 = tcpip.Address("\x0a\x00\x00\x01") - stackAddr2 = tcpip.Address("\x0a\x00\x00\x02") - stackAddrBad = tcpip.Address("\x0a\x00\x00\x03") + stackLinkAddr1 = tcpip.LinkAddress("\x0a\x0a\x0b\x0b\x0c\x0c") + stackLinkAddr2 = tcpip.LinkAddress("\x0b\x0b\x0c\x0c\x0d\x0d") + stackAddr1 = tcpip.Address("\x0a\x00\x00\x01") + stackAddr2 = tcpip.Address("\x0a\x00\x00\x02") + stackAddrBad = tcpip.Address("\x0a\x00\x00\x03") + + defaultChannelSize = 1 + defaultMTU = 65536 ) type testContext struct { @@ -50,8 +54,7 @@ func newTestContext(t *testing.T) *testContext { TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, }) - const defaultMTU = 65536 - ep := channel.New(256, defaultMTU, stackLinkAddr) + ep := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr1) wep := stack.LinkEndpoint(ep) if testing.Verbose() { @@ -119,7 +122,7 @@ func TestDirectRequest(t *testing.T) { if !rep.IsValid() { t.Fatalf("invalid ARP response pi.Pkt.Header.UsedLength()=%d", pi.Pkt.Header.UsedLength()) } - if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr; got != want { + if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr1; got != want { t.Errorf("got HardwareAddressSender = %s, want = %s", got, want) } if got, want := tcpip.Address(rep.ProtocolAddressSender()), tcpip.Address(h.ProtocolAddressTarget()); got != want { @@ -144,3 +147,44 @@ func TestDirectRequest(t *testing.T) { t.Errorf("stackAddrBad: unexpected packet sent, Proto=%v", pkt.Proto) } } + +func TestLinkAddressRequest(t *testing.T) { + tests := []struct { + name string + remoteLinkAddr tcpip.LinkAddress + expectLinkAddr tcpip.LinkAddress + }{ + { + name: "Unicast", + remoteLinkAddr: stackLinkAddr2, + expectLinkAddr: stackLinkAddr2, + }, + { + name: "Multicast", + remoteLinkAddr: "", + expectLinkAddr: header.EthernetBroadcastAddress, + }, + } + + for _, test := range tests { + p := arp.NewProtocol() + linkRes, ok := p.(stack.LinkAddressResolver) + if !ok { + t.Fatal("expected ARP protocol to implement stack.LinkAddressResolver") + } + + linkEP := channel.New(defaultChannelSize, defaultMTU, stackLinkAddr1) + if err := linkRes.LinkAddressRequest(stackAddr1, stackAddr2, test.remoteLinkAddr, linkEP); err != nil { + t.Errorf("got p.LinkAddressRequest(%s, %s, %s, _) = %s", stackAddr1, stackAddr2, test.remoteLinkAddr, err) + } + + pkt, ok := linkEP.Read() + if !ok { + t.Fatal("expected to send a link address request") + } + + if got, want := pkt.Route.RemoteLinkAddress, test.expectLinkAddr; got != want { + t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", got, want) + } + } +} diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index a5b780ca2..615bae648 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -185,6 +185,11 @@ func (*testObject) ARPHardwareType() header.ARPHardwareType { panic("not implemented") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (*testObject) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("not implemented") +} + func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 1b67aa066..83e71cb8c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -129,6 +129,9 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { pkt.Data.TrimFront(header.ICMPv4MinimumSize) switch h.Code() { + case header.ICMPv4HostUnreachable: + e.handleControl(stack.ControlNoRoute, 0, pkt) + case header.ICMPv4PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 3b79749b5..24600d877 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -128,6 +128,8 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) switch header.ICMPv6(hdr).Code() { + case header.ICMPv6NetworkUnreachable: + e.handleControl(stack.ControlNetworkUnreachable, 0, pkt) case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } @@ -502,7 +504,7 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { } // LinkAddressRequest implements stack.LinkAddressResolver. -func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack.LinkEndpoint) *tcpip.Error { +func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP stack.LinkEndpoint) *tcpip.Error { snaddr := header.SolicitedNodeAddr(addr) // TODO(b/148672031): Use stack.FindRoute instead of manually creating the @@ -511,8 +513,12 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. r := &stack.Route{ LocalAddress: localAddr, RemoteAddress: snaddr, - RemoteLinkAddress: header.EthernetAddressFromMulticastIPv6Address(snaddr), + RemoteLinkAddress: remoteLinkAddr, } + if len(r.RemoteLinkAddress) == 0 { + r.RemoteLinkAddress = header.EthernetAddressFromMulticastIPv6Address(snaddr) + } + hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize) pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) pkt.SetType(header.ICMPv6NeighborSolicit) diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 52a01b44e..f86aaed1d 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -34,6 +34,9 @@ const ( linkAddr0 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") linkAddr1 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0e") linkAddr2 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0f") + + defaultChannelSize = 1 + defaultMTU = 65536 ) var ( @@ -257,8 +260,7 @@ func newTestContext(t *testing.T) *testContext { }), } - const defaultMTU = 65536 - c.linkEP0 = channel.New(256, defaultMTU, linkAddr0) + c.linkEP0 = channel.New(defaultChannelSize, defaultMTU, linkAddr0) wrappedEP0 := stack.LinkEndpoint(endpointWithResolutionCapability{LinkEndpoint: c.linkEP0}) if testing.Verbose() { @@ -271,7 +273,7 @@ func newTestContext(t *testing.T) *testContext { t.Fatalf("AddAddress lladdr0: %v", err) } - c.linkEP1 = channel.New(256, defaultMTU, linkAddr1) + c.linkEP1 = channel.New(defaultChannelSize, defaultMTU, linkAddr1) wrappedEP1 := stack.LinkEndpoint(endpointWithResolutionCapability{LinkEndpoint: c.linkEP1}) if err := c.s1.CreateNIC(1, wrappedEP1); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -951,3 +953,47 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { }) } } + +func TestLinkAddressRequest(t *testing.T) { + snaddr := header.SolicitedNodeAddr(lladdr0) + mcaddr := header.EthernetAddressFromMulticastIPv6Address(snaddr) + + tests := []struct { + name string + remoteLinkAddr tcpip.LinkAddress + expectLinkAddr tcpip.LinkAddress + }{ + { + name: "Unicast", + remoteLinkAddr: linkAddr1, + expectLinkAddr: linkAddr1, + }, + { + name: "Multicast", + remoteLinkAddr: "", + expectLinkAddr: mcaddr, + }, + } + + for _, test := range tests { + p := NewProtocol() + linkRes, ok := p.(stack.LinkAddressResolver) + if !ok { + t.Fatalf("expected IPv6 protocol to implement stack.LinkAddressResolver") + } + + linkEP := channel.New(defaultChannelSize, defaultMTU, linkAddr0) + if err := linkRes.LinkAddressRequest(lladdr0, lladdr1, test.remoteLinkAddr, linkEP); err != nil { + t.Errorf("got p.LinkAddressRequest(%s, %s, %s, _) = %s", lladdr0, lladdr1, test.remoteLinkAddr, err) + } + + pkt, ok := linkEP.Read() + if !ok { + t.Fatal("expected to send a link address request") + } + + if got, want := pkt.Route.RemoteLinkAddress, test.expectLinkAddr; got != want { + t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", got, want) + } + } +} diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index d39baf620..559a1c4dd 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -49,7 +49,8 @@ const ( type manipType int const ( - manipDstPrerouting manipType = iota + manipNone manipType = iota + manipDstPrerouting manipDstOutput ) @@ -113,13 +114,11 @@ type conn struct { // update the state of tcb. It is immutable. tcbHook Hook - // mu protects tcb. + // mu protects all mutable state. mu sync.Mutex `state:"nosave"` - // tcb is TCB control block. It is used to keep track of states // of tcp connection and is protected by mu. tcb tcpconntrack.TCB - // lastUsed is the last time the connection saw a relevant packet, and // is updated by each packet on the connection. It is protected by mu. lastUsed time.Time `state:".(unixTime)"` @@ -141,8 +140,26 @@ func (cn *conn) timedOut(now time.Time) bool { return now.Sub(cn.lastUsed) > defaultTimeout } +// update the connection tracking state. +// +// Precondition: ct.mu must be held. +func (ct *conn) updateLocked(tcpHeader header.TCP, hook Hook) { + // Update the state of tcb. tcb assumes it's always initialized on the + // client. However, we only need to know whether the connection is + // established or not, so the client/server distinction isn't important. + // TODO(gvisor.dev/issue/170): Add support in tcpconntrack to handle + // other tcp states. + if ct.tcb.IsEmpty() { + ct.tcb.Init(tcpHeader) + } else if hook == ct.tcbHook { + ct.tcb.UpdateStateOutbound(tcpHeader) + } else { + ct.tcb.UpdateStateInbound(tcpHeader) + } +} + // ConnTrack tracks all connections created for NAT rules. Most users are -// expected to only call handlePacket and createConnFor. +// expected to only call handlePacket, insertRedirectConn, and maybeInsertNoop. // // ConnTrack keeps all connections in a slice of buckets, each of which holds a // linked list of tuples. This gives us some desirable properties: @@ -248,8 +265,7 @@ func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) { return nil, dirOriginal } -// createConnFor creates a new conn for pkt. -func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { +func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { tid, err := packetToTupleID(pkt) if err != nil { return nil @@ -272,10 +288,15 @@ func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarg manip = manipDstOutput } conn := newConn(tid, replyTID, manip, hook) + ct.insertConn(conn) + return conn +} +// insertConn inserts conn into the appropriate table bucket. +func (ct *ConnTrack) insertConn(conn *conn) { // Lock the buckets in the correct order. - tupleBucket := ct.bucket(tid) - replyBucket := ct.bucket(replyTID) + tupleBucket := ct.bucket(conn.original.tupleID) + replyBucket := ct.bucket(conn.reply.tupleID) ct.mu.RLock() defer ct.mu.RUnlock() if tupleBucket < replyBucket { @@ -289,22 +310,37 @@ func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarg ct.buckets[tupleBucket].mu.Lock() } - // Add the tuple to the map. - ct.buckets[tupleBucket].tuples.PushFront(&conn.original) - ct.buckets[replyBucket].tuples.PushFront(&conn.reply) + // Now that we hold the locks, ensure the tuple hasn't been inserted by + // another thread. + alreadyInserted := false + for other := ct.buckets[tupleBucket].tuples.Front(); other != nil; other = other.Next() { + if other.tupleID == conn.original.tupleID { + alreadyInserted = true + break + } + } + + if !alreadyInserted { + // Add the tuple to the map. + ct.buckets[tupleBucket].tuples.PushFront(&conn.original) + ct.buckets[replyBucket].tuples.PushFront(&conn.reply) + } // Unlocking can happen in any order. ct.buckets[tupleBucket].mu.Unlock() if tupleBucket != replyBucket { ct.buckets[replyBucket].mu.Unlock() } - - return conn } // handlePacketPrerouting manipulates ports for packets in Prerouting hook. // TODO(gvisor.dev/issue/170): Change address for Prerouting hook. func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { + // If this is a noop entry, don't do anything. + if conn.manip == manipNone { + return + } + netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -322,12 +358,22 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { netHeader.SetSourceAddress(conn.original.dstAddr) } + // TODO(gvisor.dev/issue/170): TCP checksums aren't usually validated + // on inbound packets, so we don't recalculate them. However, we should + // support cases when they are validated, e.g. when we can't offload + // receive checksumming. + netHeader.SetChecksum(0) netHeader.SetChecksum(^netHeader.CalculateChecksum()) } // handlePacketOutput manipulates ports for packets in Output hook. func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir direction) { + // If this is a noop entry, don't do anything. + if conn.manip == manipNone { + return + } + netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -362,20 +408,31 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d } // handlePacket will manipulate the port and address of the packet if the -// connection exists. -func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) { +// connection exists. Returns whether, after the packet traverses the tables, +// it should create a new entry in the table. +func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) bool { if pkt.NatDone { - return + return false } if hook != Prerouting && hook != Output { - return + return false + } + + // TODO(gvisor.dev/issue/170): Support other transport protocols. + if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + return false } conn, dir := ct.connFor(pkt) + // Connection or Rule not found for the packet. if conn == nil { - // Connection not found for the packet or the packet is invalid. - return + return true + } + + tcpHeader := header.TCP(pkt.TransportHeader) + if tcpHeader == nil { + return false } switch hook { @@ -395,14 +452,39 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou // Mark the connection as having been used recently so it isn't reaped. conn.lastUsed = time.Now() // Update connection state. - if tcpHeader := header.TCP(pkt.TransportHeader); conn.tcb.IsEmpty() { - conn.tcb.Init(tcpHeader) - conn.tcbHook = hook - } else if hook == conn.tcbHook { - conn.tcb.UpdateStateOutbound(tcpHeader) - } else { - conn.tcb.UpdateStateInbound(tcpHeader) + conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + + return false +} + +// maybeInsertNoop tries to insert a no-op connection entry to keep connections +// from getting clobbered when replies arrive. It only inserts if there isn't +// already a connection for pkt. +// +// This should be called after traversing iptables rules only, to ensure that +// pkt.NatDone is set correctly. +func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { + // If there were a rule applying to this packet, it would be marked + // with NatDone. + if pkt.NatDone { + return + } + + // We only track TCP connections. + if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + return + } + + // This is the first packet we're seeing for the TCP connection. Insert + // the noop entry (an identity mapping) so that the response doesn't + // get NATed, breaking the connection. + tid, err := packetToTupleID(pkt) + if err != nil { + return } + conn := newConn(tid, tid.reply(), manipNone, hook) + conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + ct.insertConn(conn) } // bucket gets the conntrack bucket for a tupleID. diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index eefb4b07f..c962693f5 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -121,10 +121,12 @@ func (*fwdTestNetworkEndpoint) Close() {} type fwdTestNetworkProtocol struct { addrCache *linkAddrCache addrResolveDelay time.Duration - onLinkAddressResolved func(cache *linkAddrCache, addr tcpip.Address) + onLinkAddressResolved func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) onResolveStaticAddress func(tcpip.Address) (tcpip.LinkAddress, bool) } +var _ LinkAddressResolver = (*fwdTestNetworkProtocol)(nil) + func (f *fwdTestNetworkProtocol) Number() tcpip.NetworkProtocolNumber { return fwdTestNetNumber } @@ -174,10 +176,10 @@ func (f *fwdTestNetworkProtocol) Close() {} func (f *fwdTestNetworkProtocol) Wait() {} -func (f *fwdTestNetworkProtocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP LinkEndpoint) *tcpip.Error { +func (f *fwdTestNetworkProtocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error { if f.addrCache != nil && f.onLinkAddressResolved != nil { time.AfterFunc(f.addrResolveDelay, func() { - f.onLinkAddressResolved(f.addrCache, addr) + f.onLinkAddressResolved(f.addrCache, addr, remoteLinkAddr) }) } return nil @@ -307,6 +309,11 @@ func (*fwdTestLinkEndpoint) ARPHardwareType() header.ARPHardwareType { panic("not implemented") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *fwdTestLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + panic("not implemented") +} + func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol) (ep1, ep2 *fwdTestLinkEndpoint) { // Create a stack with the network protocol and two NICs. s := New(Options{ @@ -400,7 +407,7 @@ func TestForwardingWithFakeResolver(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any address will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, @@ -458,7 +465,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Only packets to address 3 will be resolved to the // link address "c". if addr == "\x03" { @@ -510,7 +517,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any packets will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, @@ -554,7 +561,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any packets will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, @@ -611,7 +618,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // Create a network protocol with a fake resolver. proto := &fwdTestNetworkProtocol{ addrResolveDelay: 500 * time.Millisecond, - onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address) { + onLinkAddressResolved: func(cache *linkAddrCache, addr tcpip.Address, _ tcpip.LinkAddress) { // Any packets will be resolved to the link address "c". cache.add(tcpip.FullAddress{NIC: 2, Addr: addr}, "c") }, diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index bbf3b60e8..cbbae4224 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -22,22 +22,30 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) -// Table names. +// tableID is an index into IPTables.tables. +type tableID int + const ( - TablenameNat = "nat" - TablenameMangle = "mangle" - TablenameFilter = "filter" + natID tableID = iota + mangleID + filterID + numTables ) -// Chain names as defined by net/ipv4/netfilter/ip_tables.c. +// Table names. const ( - ChainNamePrerouting = "PREROUTING" - ChainNameInput = "INPUT" - ChainNameForward = "FORWARD" - ChainNameOutput = "OUTPUT" - ChainNamePostrouting = "POSTROUTING" + NATTable = "nat" + MangleTable = "mangle" + FilterTable = "filter" ) +// nameToID is immutable. +var nameToID = map[string]tableID{ + NATTable: natID, + MangleTable: mangleID, + FilterTable: filterID, +} + // HookUnset indicates that there is no hook set for an entrypoint or // underflow. const HookUnset = -1 @@ -48,11 +56,9 @@ const reaperDelay = 5 * time.Second // DefaultTables returns a default set of tables. Each chain is set to accept // all packets. func DefaultTables() *IPTables { - // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for - // iotas. return &IPTables{ - tables: map[string]Table{ - TablenameNat: Table{ + tables: [numTables]Table{ + natID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, @@ -60,60 +66,66 @@ func DefaultTables() *IPTables { Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, - BuiltinChains: map[Hook]int{ + BuiltinChains: [NumHooks]int{ Prerouting: 0, Input: 1, + Forward: HookUnset, Output: 2, Postrouting: 3, }, - Underflows: map[Hook]int{ + Underflows: [NumHooks]int{ Prerouting: 0, Input: 1, + Forward: HookUnset, Output: 2, Postrouting: 3, }, - UserChains: map[string]int{}, }, - TablenameMangle: Table{ + mangleID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, - BuiltinChains: map[Hook]int{ + BuiltinChains: [NumHooks]int{ Prerouting: 0, Output: 1, }, - Underflows: map[Hook]int{ - Prerouting: 0, - Output: 1, + Underflows: [NumHooks]int{ + Prerouting: 0, + Input: HookUnset, + Forward: HookUnset, + Output: 1, + Postrouting: HookUnset, }, - UserChains: map[string]int{}, }, - TablenameFilter: Table{ + filterID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, - BuiltinChains: map[Hook]int{ - Input: 0, - Forward: 1, - Output: 2, + BuiltinChains: [NumHooks]int{ + Prerouting: HookUnset, + Input: 0, + Forward: 1, + Output: 2, + Postrouting: HookUnset, }, - Underflows: map[Hook]int{ - Input: 0, - Forward: 1, - Output: 2, + Underflows: [NumHooks]int{ + Prerouting: HookUnset, + Input: 0, + Forward: 1, + Output: 2, + Postrouting: HookUnset, }, - UserChains: map[string]int{}, }, }, - priorities: map[Hook][]string{ - Input: []string{TablenameNat, TablenameFilter}, - Prerouting: []string{TablenameMangle, TablenameNat}, - Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, + priorities: [NumHooks][]tableID{ + Prerouting: []tableID{mangleID, natID}, + Input: []tableID{natID, filterID}, + Output: []tableID{mangleID, natID, filterID}, }, connections: ConnTrack{ seed: generateRandUint32(), @@ -127,51 +139,48 @@ func DefaultTables() *IPTables { func EmptyFilterTable() Table { return Table{ Rules: []Rule{}, - BuiltinChains: map[Hook]int{ - Input: HookUnset, - Forward: HookUnset, - Output: HookUnset, + BuiltinChains: [NumHooks]int{ + Prerouting: HookUnset, + Postrouting: HookUnset, }, - Underflows: map[Hook]int{ - Input: HookUnset, - Forward: HookUnset, - Output: HookUnset, + Underflows: [NumHooks]int{ + Prerouting: HookUnset, + Postrouting: HookUnset, }, - UserChains: map[string]int{}, } } -// EmptyNatTable returns a Table with no rules and the filter table chains +// EmptyNATTable returns a Table with no rules and the filter table chains // mapped to HookUnset. -func EmptyNatTable() Table { +func EmptyNATTable() Table { return Table{ Rules: []Rule{}, - BuiltinChains: map[Hook]int{ - Prerouting: HookUnset, - Input: HookUnset, - Output: HookUnset, - Postrouting: HookUnset, + BuiltinChains: [NumHooks]int{ + Forward: HookUnset, }, - Underflows: map[Hook]int{ - Prerouting: HookUnset, - Input: HookUnset, - Output: HookUnset, - Postrouting: HookUnset, + Underflows: [NumHooks]int{ + Forward: HookUnset, }, - UserChains: map[string]int{}, } } -// GetTable returns table by name. +// GetTable returns a table by name. func (it *IPTables) GetTable(name string) (Table, bool) { + id, ok := nameToID[name] + if !ok { + return Table{}, false + } it.mu.RLock() defer it.mu.RUnlock() - t, ok := it.tables[name] - return t, ok + return it.tables[id], true } // ReplaceTable replaces or inserts table by name. -func (it *IPTables) ReplaceTable(name string, table Table) { +func (it *IPTables) ReplaceTable(name string, table Table) *tcpip.Error { + id, ok := nameToID[name] + if !ok { + return tcpip.ErrInvalidOptionValue + } it.mu.Lock() defer it.mu.Unlock() // If iptables is being enabled, initialize the conntrack table and @@ -181,14 +190,8 @@ func (it *IPTables) ReplaceTable(name string, table Table) { it.startReaper(reaperDelay) } it.modified = true - it.tables[name] = table -} - -// GetPriorities returns slice of priorities associated with hook. -func (it *IPTables) GetPriorities(hook Hook) []string { - it.mu.RLock() - defer it.mu.RUnlock() - return it.priorities[hook] + it.tables[id] = table + return nil } // A chainVerdict is what a table decides should be done with a packet. @@ -223,11 +226,19 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // Packets are manipulated only if connection and matching // NAT rule exists. - it.connections.handlePacket(pkt, hook, gso, r) + shouldTrack := it.connections.handlePacket(pkt, hook, gso, r) // Go through each table containing the hook. - for _, tablename := range it.GetPriorities(hook) { - table, _ := it.GetTable(tablename) + it.mu.RLock() + defer it.mu.RUnlock() + priorities := it.priorities[hook] + for _, tableID := range priorities { + // If handlePacket already NATed the packet, we don't need to + // check the NAT table. + if tableID == natID && pkt.NatDone { + continue + } + table := it.tables[tableID] ruleIdx := table.BuiltinChains[hook] switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict { // If the table returns Accept, move on to the next table. @@ -256,6 +267,20 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr } } + // If this connection should be tracked, try to add an entry for it. If + // traversing the nat table didn't end in adding an entry, + // maybeInsertNoop will add a no-op entry for the connection. This is + // needeed when establishing connections so that the SYN/ACK reply to an + // outgoing SYN is delivered to the correct endpoint rather than being + // redirected by a prerouting rule. + // + // From the iptables documentation: "If there is no rule, a `null' + // binding is created: this usually does not map the packet, but exists + // to ensure we don't map another stream over an existing one." + if shouldTrack { + it.connections.maybeInsertNoop(pkt, hook) + } + // Every table returned Accept. return true } diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index d43f60c67..dc88033c7 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -153,7 +153,7 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso // Set up conection for matching NAT rule. Only the first // packet of the connection comes here. Other packets will be // manipulated in connection tracking. - if conn := ct.createConnFor(pkt, hook, rt); conn != nil { + if conn := ct.insertRedirectConn(pkt, hook, rt); conn != nil { ct.handlePacket(pkt, hook, gso, r) } default: diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index eb70e3104..73274ada9 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -84,14 +84,14 @@ type IPTables struct { // mu protects tables, priorities, and modified. mu sync.RWMutex - // tables maps table names to tables. User tables have arbitrary names. - // mu needs to be locked for accessing. - tables map[string]Table + // tables maps tableIDs to tables. Holds builtin tables only, not user + // tables. mu must be locked for accessing. + tables [numTables]Table // priorities maps each hook to a list of table names. The order of the // list is the order in which each table should be visited for that // hook. mu needs to be locked for accessing. - priorities map[Hook][]string + priorities [NumHooks][]tableID // modified is whether tables have been modified at least once. It is // used to elide the iptables performance overhead for workloads that @@ -113,22 +113,20 @@ type Table struct { Rules []Rule // BuiltinChains maps builtin chains to their entrypoint rule in Rules. - BuiltinChains map[Hook]int + BuiltinChains [NumHooks]int // Underflows maps builtin chains to their underflow rule in Rules // (i.e. the rule to execute if the chain returns without a verdict). - Underflows map[Hook]int - - // UserChains holds user-defined chains for the keyed by name. Users - // can give their chains arbitrary names. - UserChains map[string]int + Underflows [NumHooks]int } // ValidHooks returns a bitmap of the builtin hooks for the given table. func (table *Table) ValidHooks() uint32 { hooks := uint32(0) - for hook := range table.BuiltinChains { - hooks |= 1 << hook + for hook, ruleIdx := range table.BuiltinChains { + if ruleIdx != HookUnset { + hooks |= 1 << hook + } } return hooks } diff --git a/pkg/tcpip/stack/linkaddrcache.go b/pkg/tcpip/stack/linkaddrcache.go index 403557fd7..6f73a0ce4 100644 --- a/pkg/tcpip/stack/linkaddrcache.go +++ b/pkg/tcpip/stack/linkaddrcache.go @@ -244,7 +244,7 @@ func (c *linkAddrCache) startAddressResolution(k tcpip.FullAddress, linkRes Link for i := 0; ; i++ { // Send link request, then wait for the timeout limit and check // whether the request succeeded. - linkRes.LinkAddressRequest(k.Addr, localAddr, linkEP) + linkRes.LinkAddressRequest(k.Addr, localAddr, "" /* linkAddr */, linkEP) select { case now := <-time.After(c.resolutionTimeout): diff --git a/pkg/tcpip/stack/linkaddrcache_test.go b/pkg/tcpip/stack/linkaddrcache_test.go index 1baa498d0..b15b8d1cb 100644 --- a/pkg/tcpip/stack/linkaddrcache_test.go +++ b/pkg/tcpip/stack/linkaddrcache_test.go @@ -48,7 +48,7 @@ type testLinkAddressResolver struct { onLinkAddressRequest func() } -func (r *testLinkAddressResolver) LinkAddressRequest(addr, _ tcpip.Address, _ LinkEndpoint) *tcpip.Error { +func (r *testLinkAddressResolver) LinkAddressRequest(addr, _ tcpip.Address, _ tcpip.LinkAddress, _ LinkEndpoint) *tcpip.Error { time.AfterFunc(r.delay, func() { r.fakeRequest(addr) }) if f := r.onLinkAddressRequest; f != nil { f() diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index e28c23d66..9dce11a97 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -469,7 +469,7 @@ type ndpState struct { rtrSolicit struct { // The timer used to send the next router solicitation message. - timer *time.Timer + timer tcpip.Timer // Used to let the Router Solicitation timer know that it has been stopped. // @@ -503,7 +503,7 @@ type ndpState struct { // to the DAD goroutine that DAD should stop. type dadState struct { // The DAD timer to send the next NS message, or resolve the address. - timer *time.Timer + timer tcpip.Timer // Used to let the DAD timer know that it has been stopped. // @@ -515,38 +515,38 @@ type dadState struct { // defaultRouterState holds data associated with a default router discovered by // a Router Advertisement (RA). type defaultRouterState struct { - // Timer to invalidate the default router. + // Job to invalidate the default router. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // onLinkPrefixState holds data associated with an on-link prefix discovered by // a Router Advertisement's Prefix Information option (PI) when the NDP // configurations was configured to do so. type onLinkPrefixState struct { - // Timer to invalidate the on-link prefix. + // Job to invalidate the on-link prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // tempSLAACAddrState holds state associated with a temporary SLAAC address. type tempSLAACAddrState struct { - // Timer to deprecate the temporary SLAAC address. + // Job to deprecate the temporary SLAAC address. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the temporary SLAAC address. + // Job to invalidate the temporary SLAAC address. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job - // Timer to regenerate the temporary SLAAC address. + // Job to regenerate the temporary SLAAC address. // // Must not be nil. - regenTimer *tcpip.CancellableTimer + regenJob *tcpip.Job createdAt time.Time @@ -561,15 +561,15 @@ type tempSLAACAddrState struct { // slaacPrefixState holds state associated with a SLAAC prefix. type slaacPrefixState struct { - // Timer to deprecate the prefix. + // Job to deprecate the prefix. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the prefix. + // Job to invalidate the prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job // Nonzero only when the address is not valid forever. validUntil time.Time @@ -651,12 +651,12 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref } var done bool - var timer *time.Timer + var timer tcpip.Timer // We initially start a timer to fire immediately because some of the DAD work // cannot be done while holding the NIC's lock. This is effectively the same // as starting a goroutine but we use a timer that fires immediately so we can // reset it for the next DAD iteration. - timer = time.AfterFunc(0, func() { + timer = ndp.nic.stack.Clock().AfterFunc(0, func() { ndp.nic.mu.Lock() defer ndp.nic.mu.Unlock() @@ -871,9 +871,9 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { case ok && rl != 0: // This is an already discovered default router. Update - // the invalidation timer. - rtr.invalidationTimer.StopLocked() - rtr.invalidationTimer.Reset(rl) + // the invalidation job. + rtr.invalidationJob.Cancel() + rtr.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = rtr case ok && rl == 0: @@ -950,7 +950,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { return } - rtr.invalidationTimer.StopLocked() + rtr.invalidationJob.Cancel() delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. @@ -979,12 +979,12 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { } state := defaultRouterState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateDefaultRouter(ip) }), } - state.invalidationTimer.Reset(rl) + state.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = state } @@ -1009,13 +1009,13 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) } state := onLinkPrefixState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateOnLinkPrefix(prefix) }), } if l < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(l) + state.invalidationJob.Schedule(l) } ndp.onLinkPrefixes[prefix] = state @@ -1033,7 +1033,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { return } - s.invalidationTimer.StopLocked() + s.invalidationJob.Cancel() delete(ndp.onLinkPrefixes, prefix) // Let the integrator know a discovered on-link prefix is invalidated. @@ -1082,14 +1082,14 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio // This is an already discovered on-link prefix with a // new non-zero valid lifetime. // - // Update the invalidation timer. + // Update the invalidation job. - prefixState.invalidationTimer.StopLocked() + prefixState.invalidationJob.Cancel() if vl < header.NDPInfiniteLifetime { - // Prefix is valid for a finite lifetime, reset the timer to expire after + // Prefix is valid for a finite lifetime, schedule the job to execute after // the new valid lifetime. - prefixState.invalidationTimer.Reset(vl) + prefixState.invalidationJob.Schedule(vl) } ndp.onLinkPrefixes[prefix] = prefixState @@ -1154,7 +1154,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } state := slaacPrefixState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix)) @@ -1162,7 +1162,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { ndp.deprecateSLAACAddress(state.stableAddr.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix)) @@ -1184,19 +1184,19 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { if !ndp.generateSLAACAddr(prefix, &state) { // We were unable to generate an address for the prefix, we do not nothing - // further as there is no reason to maintain state or timers for a prefix we + // further as there is no reason to maintain state or jobs for a prefix we // do not have an address for. return } - // Setup the initial timers to deprecate and invalidate prefix. + // Setup the initial jobs to deprecate and invalidate prefix. if pl < header.NDPInfiniteLifetime && pl != 0 { - state.deprecationTimer.Reset(pl) + state.deprecationJob.Schedule(pl) } if vl < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(vl) + state.invalidationJob.Schedule(vl) state.validUntil = now.Add(vl) } @@ -1428,7 +1428,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla } state := tempSLAACAddrState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr)) @@ -1441,7 +1441,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.deprecateSLAACAddress(tempAddrState.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr)) @@ -1454,7 +1454,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState) }), - regenTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr)) @@ -1481,9 +1481,9 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ref: ref, } - state.deprecationTimer.Reset(pl) - state.invalidationTimer.Reset(vl) - state.regenTimer.Reset(pl - ndp.configs.RegenAdvanceDuration) + state.deprecationJob.Schedule(pl) + state.invalidationJob.Schedule(vl) + state.regenJob.Schedule(pl - ndp.configs.RegenAdvanceDuration) prefixState.generationAttempts++ prefixState.tempAddrs[generatedAddr.Address] = state @@ -1518,16 +1518,16 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat prefixState.stableAddr.ref.deprecated = false } - // If prefix was preferred for some finite lifetime before, stop the - // deprecation timer so it can be reset. - prefixState.deprecationTimer.StopLocked() + // If prefix was preferred for some finite lifetime before, cancel the + // deprecation job so it can be reset. + prefixState.deprecationJob.Cancel() now := time.Now() - // Reset the deprecation timer if prefix has a finite preferred lifetime. + // Schedule the deprecation job if prefix has a finite preferred lifetime. if pl < header.NDPInfiniteLifetime { if !deprecated { - prefixState.deprecationTimer.Reset(pl) + prefixState.deprecationJob.Schedule(pl) } prefixState.preferredUntil = now.Add(pl) } else { @@ -1546,9 +1546,9 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // 3) Otherwise, reset the valid lifetime of the prefix to 2 hours. if vl >= header.NDPInfiniteLifetime { - // Handle the infinite valid lifetime separately as we do not keep a timer - // in this case. - prefixState.invalidationTimer.StopLocked() + // Handle the infinite valid lifetime separately as we do not schedule a + // job in this case. + prefixState.invalidationJob.Cancel() prefixState.validUntil = time.Time{} } else { var effectiveVl time.Duration @@ -1569,8 +1569,8 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } if effectiveVl != 0 { - prefixState.invalidationTimer.StopLocked() - prefixState.invalidationTimer.Reset(effectiveVl) + prefixState.invalidationJob.Cancel() + prefixState.invalidationJob.Schedule(effectiveVl) prefixState.validUntil = now.Add(effectiveVl) } } @@ -1582,7 +1582,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // Note, we do not need to update the entries in the temporary address map - // after updating the timers because the timers are held as pointers. + // after updating the jobs because the jobs are held as pointers. var regenForAddr tcpip.Address allAddressesRegenerated := true for tempAddr, tempAddrState := range prefixState.tempAddrs { @@ -1596,14 +1596,14 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer valid, invalidate it immediately. Otherwise, - // reset the invalidation timer. + // reset the invalidation job. newValidLifetime := validUntil.Sub(now) if newValidLifetime <= 0 { ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, tempAddr, tempAddrState) continue } - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.invalidationTimer.Reset(newValidLifetime) + tempAddrState.invalidationJob.Cancel() + tempAddrState.invalidationJob.Schedule(newValidLifetime) // As per RFC 4941 section 3.3 step 4, the preferred lifetime of a temporary // address is the lower of the preferred lifetime of the stable address or @@ -1616,17 +1616,17 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer preferred, deprecate it immediately. - // Otherwise, reset the deprecation timer. + // Otherwise, schedule the deprecation job again. newPreferredLifetime := preferredUntil.Sub(now) - tempAddrState.deprecationTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() if newPreferredLifetime <= 0 { ndp.deprecateSLAACAddress(tempAddrState.ref) } else { tempAddrState.ref.deprecated = false - tempAddrState.deprecationTimer.Reset(newPreferredLifetime) + tempAddrState.deprecationJob.Schedule(newPreferredLifetime) } - tempAddrState.regenTimer.StopLocked() + tempAddrState.regenJob.Cancel() if tempAddrState.regenerated { } else { allAddressesRegenerated = false @@ -1637,7 +1637,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // immediately after we finish iterating over the temporary addresses. regenForAddr = tempAddr } else { - tempAddrState.regenTimer.Reset(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) + tempAddrState.regenJob.Schedule(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) } } } @@ -1717,7 +1717,7 @@ func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPr ndp.cleanupSLAACPrefixResources(prefix, state) } -// cleanupSLAACPrefixResources cleansup a SLAAC prefix's timers and entry. +// cleanupSLAACPrefixResources cleans up a SLAAC prefix's jobs and entry. // // Panics if the SLAAC prefix is not known. // @@ -1729,8 +1729,8 @@ func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaa } state.stableAddr.ref = nil - state.deprecationTimer.StopLocked() - state.invalidationTimer.StopLocked() + state.deprecationJob.Cancel() + state.invalidationJob.Cancel() delete(ndp.slaacPrefixes, prefix) } @@ -1775,13 +1775,13 @@ func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWi } // cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's -// timers and entry. +// jobs and entry. // // The NIC that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { - tempAddrState.deprecationTimer.StopLocked() - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.regenTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() + tempAddrState.invalidationJob.Cancel() + tempAddrState.regenJob.Cancel() delete(tempAddrs, tempAddr) } @@ -1860,7 +1860,7 @@ func (ndp *ndpState) startSolicitingRouters() { var done bool ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = time.AfterFunc(delay, func() { + ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() { ndp.nic.mu.Lock() if done { // If we reach this point, it means that the RS timer fired after another diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 6f86abc98..644ba7c33 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -1254,7 +1254,7 @@ func TestRouterDiscovery(t *testing.T) { default: } - // Wait for lladdr2's router invalidation timer to fire. The lifetime + // Wait for lladdr2's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1271,7 +1271,7 @@ func TestRouterDiscovery(t *testing.T) { e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) expectRouterEvent(llAddr2, false) - // Wait for lladdr3's router invalidation timer to fire. The lifetime + // Wait for lladdr3's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1502,7 +1502,7 @@ func TestPrefixDiscovery(t *testing.T) { default: } - // Wait for prefix2's most recent invalidation timer plus some buffer to + // Wait for prefix2's most recent invalidation job plus some buffer to // expire. select { case e := <-ndpDisp.prefixC: @@ -2395,7 +2395,7 @@ func TestAutoGenTempAddrRegen(t *testing.T) { for _, addr := range tempAddrs { // Wait for a deprecation then invalidation event, or just an invalidation // event. We need to cover both cases but cannot deterministically hit both - // cases because the deprecation and invalidation timers could fire in any + // cases because the deprecation and invalidation jobs could execute in any // order. select { case e := <-ndpDisp.autoGenAddrC: @@ -2432,9 +2432,9 @@ func TestAutoGenTempAddrRegen(t *testing.T) { } } -// TestAutoGenTempAddrRegenTimerUpdates tests that a temporary address's -// regeneration timer gets updated when refreshing the address's lifetimes. -func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { +// TestAutoGenTempAddrRegenJobUpdates tests that a temporary address's +// regeneration job gets updated when refreshing the address's lifetimes. +func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { const ( nicID = 1 regenAfter = 2 * time.Second @@ -2533,7 +2533,7 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { // // A new temporary address should immediately be generated since the // regeneration time has already passed since the last address was generated - // - this regeneration does not depend on a timer. + // - this regeneration does not depend on a job. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) expectAutoGenAddrEvent(tempAddr2, newAddr) @@ -2559,11 +2559,11 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { } // Set the maximum lifetimes for temporary addresses such that on the next - // RA, the regeneration timer gets reset. + // RA, the regeneration job gets scheduled again. // // The maximum lifetime is the sum of the minimum lifetimes for temporary // addresses + the time that has already passed since the last address was - // generated so that the regeneration timer is needed to generate the next + // generated so that the regeneration job is needed to generate the next // address. newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout ndpConfigs.MaxTempAddrValidLifetime = newLifetimes @@ -2993,9 +2993,9 @@ func TestAutoGenAddrDeprecateFromPI(t *testing.T) { expectPrimaryAddr(addr2) } -// TestAutoGenAddrTimerDeprecation tests that an address is properly deprecated +// TestAutoGenAddrJobDeprecation tests that an address is properly deprecated // when its preferred lifetime expires. -func TestAutoGenAddrTimerDeprecation(t *testing.T) { +func TestAutoGenAddrJobDeprecation(t *testing.T) { const nicID = 1 const newMinVL = 2 newMinVLDuration := newMinVL * time.Second @@ -3513,8 +3513,8 @@ func TestAutoGenAddrRemoval(t *testing.T) { } expectAutoGenAddrEvent(addr, invalidatedAddr) - // Wait for the original valid lifetime to make sure the original timer - // got stopped/cleaned up. + // Wait for the original valid lifetime to make sure the original job got + // cancelled/cleaned up. select { case <-ndpDisp.autoGenAddrC: t.Fatal("unexpectedly received an auto gen addr event") diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 7b80534e6..fea0ce7e8 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1200,15 +1200,13 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp // Are any packet sockets listening for this network protocol? packetEPs := n.mu.packetEPs[protocol] - // Check whether there are packet sockets listening for every protocol. - // If we received a packet with protocol EthernetProtocolAll, then the - // previous for loop will have handled it. - if protocol != header.EthernetProtocolAll { - packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) - } + // Add any other packet sockets that maybe listening for all protocols. + packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) n.mu.RUnlock() for _, ep := range packetEPs { - ep.HandlePacket(n.id, local, protocol, pkt.Clone()) + p := pkt.Clone() + p.PktType = tcpip.PacketHost + ep.HandlePacket(n.id, local, protocol, p) } if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { @@ -1311,6 +1309,24 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } +// DeliverOutboundPacket implements NetworkDispatcher.DeliverOutboundPacket. +func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + n.mu.RLock() + // We do not deliver to protocol specific packet endpoints as on Linux + // only ETH_P_ALL endpoints get outbound packets. + // Add any other packet sockets that maybe listening for all protocols. + packetEPs := n.mu.packetEPs[header.EthernetProtocolAll] + n.mu.RUnlock() + for _, ep := range packetEPs { + p := pkt.Clone() + p.PktType = tcpip.PacketOutgoing + // Add the link layer header as outgoing packets are intercepted + // before the link layer header is created. + n.linkEP.AddHeader(local, remote, protocol, p) + ep.HandlePacket(n.id, local, protocol, p) + } +} + func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. // TODO(b/151227689): Avoid copying the packet when forwarding. We can do this diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index 3bc9fd831..a70792b50 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -89,6 +89,11 @@ func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType { panic("not implemented") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *testLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + panic("not implemented") +} + var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) // An IPv6 NetworkEndpoint that throws away outgoing packets. @@ -238,7 +243,7 @@ func (*testIPv6Protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { } // LinkAddressRequest implements LinkAddressResolver. -func (*testIPv6Protocol) LinkAddressRequest(_, _ tcpip.Address, _ LinkEndpoint) *tcpip.Error { +func (*testIPv6Protocol) LinkAddressRequest(_, _ tcpip.Address, _ tcpip.LinkAddress, _ LinkEndpoint) *tcpip.Error { return nil } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index e3556d5d2..5d6865e35 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -79,6 +79,10 @@ type PacketBuffer struct { // NatDone indicates if the packet has been manipulated as per NAT // iptables rule. NatDone bool + + // PktType indicates the SockAddrLink.PacketType of the packet as defined in + // https://www.man7.org/linux/man-pages/man7/packet.7.html. + PktType tcpip.PacketType } // Clone makes a copy of pk. It clones the Data field, which creates a new diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index f260eeb7f..8604c4259 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -52,8 +52,11 @@ type TransportEndpointID struct { type ControlType int // The following are the allowed values for ControlType values. +// TODO(http://gvisor.dev/issue/3210): Support time exceeded messages. const ( - ControlPacketTooBig ControlType = iota + ControlNetworkUnreachable ControlType = iota + ControlNoRoute + ControlPacketTooBig ControlPortUnreachable ControlUnknown ) @@ -330,8 +333,7 @@ type NetworkProtocol interface { } // NetworkDispatcher contains the methods used by the network stack to deliver -// packets to the appropriate network endpoint after it has been handled by -// the data link layer. +// inbound/outbound packets to the appropriate network/packet(if any) endpoints. type NetworkDispatcher interface { // DeliverNetworkPacket finds the appropriate network protocol endpoint // and hands the packet over for further processing. @@ -342,6 +344,16 @@ type NetworkDispatcher interface { // // DeliverNetworkPacket takes ownership of pkt. DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) + + // DeliverOutboundPacket is called by link layer when a packet is being + // sent out. + // + // pkt.LinkHeader may or may not be set before calling + // DeliverOutboundPacket. Some packets do not have link headers (e.g. + // packets sent via loopback), and won't have the field set. + // + // DeliverOutboundPacket takes ownership of pkt. + DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -443,6 +455,9 @@ type LinkEndpoint interface { // See: // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/include/uapi/linux/if_arp.h#L30 ARPHardwareType() header.ARPHardwareType + + // AddHeader adds a link layer header to pkt if required. + AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // InjectableLinkEndpoint is a LinkEndpoint where inbound packets are @@ -463,12 +478,13 @@ type InjectableLinkEndpoint interface { // A LinkAddressResolver is an extension to a NetworkProtocol that // can resolve link addresses. type LinkAddressResolver interface { - // LinkAddressRequest sends a request for the LinkAddress of addr. - // The request is sent on linkEP with localAddr as the source. + // LinkAddressRequest sends a request for the LinkAddress of addr. Broadcasts + // the request on the local network if remoteLinkAddr is the zero value. The + // request is sent on linkEP with localAddr as the source. // // A valid response will cause the discovery protocol's network // endpoint to call AddLinkAddress. - LinkAddressRequest(addr, localAddr tcpip.Address, linkEP LinkEndpoint) *tcpip.Error + LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP LinkEndpoint) *tcpip.Error // ResolveStaticAddress attempts to resolve address without sending // requests. It either resolves the name immediately or returns the diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 2b7ece851..a6faa22c2 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -728,6 +728,11 @@ func New(opts Options) *Stack { return s } +// newJob returns a tcpip.Job using the Stack clock. +func (s *Stack) newJob(l sync.Locker, f func()) *tcpip.Job { + return tcpip.NewJob(s.clock, l, f) +} + // UniqueID returns a unique identifier. func (s *Stack) UniqueID() uint64 { return s.uniqueIDGenerator.UniqueID() @@ -801,9 +806,10 @@ func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h f } } -// NowNanoseconds implements tcpip.Clock.NowNanoseconds. -func (s *Stack) NowNanoseconds() int64 { - return s.clock.NowNanoseconds() +// Clock returns the Stack's clock for retrieving the current time and +// scheduling work. +func (s *Stack) Clock() tcpip.Clock { + return s.clock } // Stats returns a mutable copy of the current stats. diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 48ad56d4d..21aafb0a2 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -192,7 +192,7 @@ func (e ErrSaveRejection) Error() string { return "save rejected due to unsupported networking state: " + e.Err.Error() } -// A Clock provides the current time. +// A Clock provides the current time and schedules work for execution. // // Times returned by a Clock should always be used for application-visible // time. Only monotonic times should be used for netstack internal timekeeping. @@ -203,6 +203,31 @@ type Clock interface { // NowMonotonic returns a monotonic time value. NowMonotonic() int64 + + // AfterFunc waits for the duration to elapse and then calls f in its own + // goroutine. It returns a Timer that can be used to cancel the call using + // its Stop method. + AfterFunc(d time.Duration, f func()) Timer +} + +// Timer represents a single event. A Timer must be created with +// Clock.AfterFunc. +type Timer interface { + // Stop prevents the Timer from firing. It returns true if the call stops the + // timer, false if the timer has already expired or been stopped. + // + // If Stop returns false, then the timer has already expired and the function + // f of Clock.AfterFunc(d, f) has been started in its own goroutine; Stop + // does not wait for f to complete before returning. If the caller needs to + // know whether f is completed, it must coordinate with f explicitly. + Stop() bool + + // Reset changes the timer to expire after duration d. + // + // Reset should be invoked only on stopped or expired timers. If the timer is + // known to have expired, Reset can be used directly. Otherwise, the caller + // must coordinate with the function f of Clock.AfterFunc(d, f). + Reset(d time.Duration) } // Address is a byte slice cast as a string that represents the address of a @@ -316,6 +341,28 @@ const ( ShutdownWrite ) +// PacketType is used to indicate the destination of the packet. +type PacketType uint8 + +const ( + // PacketHost indicates a packet addressed to the local host. + PacketHost PacketType = iota + + // PacketOtherHost indicates an outgoing packet addressed to + // another host caught by a NIC in promiscuous mode. + PacketOtherHost + + // PacketOutgoing for a packet originating from the local host + // that is looped back to a packet socket. + PacketOutgoing + + // PacketBroadcast indicates a link layer broadcast packet. + PacketBroadcast + + // PacketMulticast indicates a link layer multicast packet. + PacketMulticast +) + // FullAddress represents a full transport node address, as required by the // Connect() and Bind() methods. // @@ -555,6 +602,9 @@ type Endpoint interface { type LinkPacketInfo struct { // Protocol is the NetworkProtocolNumber for the packet. Protocol NetworkProtocolNumber + + // PktType is used to indicate the destination of the packet. + PktType PacketType } // PacketEndpoint are additional methods that are only implemented by Packet diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index 7f172f978..f32d58091 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -20,7 +20,7 @@ package tcpip import ( - _ "time" // Used with go:linkname. + "time" // Used with go:linkname. _ "unsafe" // Required for go:linkname. ) @@ -45,3 +45,31 @@ func (*StdClock) NowMonotonic() int64 { _, _, mono := now() return mono } + +// AfterFunc implements Clock.AfterFunc. +func (*StdClock) AfterFunc(d time.Duration, f func()) Timer { + return &stdTimer{ + t: time.AfterFunc(d, f), + } +} + +type stdTimer struct { + t *time.Timer +} + +var _ Timer = (*stdTimer)(nil) + +// Stop implements Timer.Stop. +func (st *stdTimer) Stop() bool { + return st.t.Stop() +} + +// Reset implements Timer.Reset. +func (st *stdTimer) Reset(d time.Duration) { + st.t.Reset(d) +} + +// NewStdTimer returns a Timer implemented with the time package. +func NewStdTimer(t *time.Timer) Timer { + return &stdTimer{t: t} +} diff --git a/pkg/tcpip/timer.go b/pkg/tcpip/timer.go index 5554c573f..f1dd7c310 100644 --- a/pkg/tcpip/timer.go +++ b/pkg/tcpip/timer.go @@ -20,50 +20,49 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) -// cancellableTimerInstance is a specific instance of CancellableTimer. +// jobInstance is a specific instance of Job. // -// Different instances are created each time CancellableTimer is Reset so each -// timer has its own earlyReturn signal. This is to address a bug when a -// CancellableTimer is stopped and reset in quick succession resulting in a -// timer instance's earlyReturn signal being affected or seen by another timer -// instance. +// Different instances are created each time Job is scheduled so each timer has +// its own earlyReturn signal. This is to address a bug when a Job is stopped +// and reset in quick succession resulting in a timer instance's earlyReturn +// signal being affected or seen by another timer instance. // // Consider the following sceneario where timer instances share a common // earlyReturn signal (T1 creates, stops and resets a Cancellable timer under a // lock L; T2, T3, T4 and T5 are goroutines that handle the first (A), second // (B), third (C), and fourth (D) instance of the timer firing, respectively): // T1: Obtain L -// T1: Create a new CancellableTimer w/ lock L (create instance A) +// T1: Create a new Job w/ lock L (create instance A) // T2: instance A fires, blocked trying to obtain L. // T1: Attempt to stop instance A (set earlyReturn = true) -// T1: Reset timer (create instance B) +// T1: Schedule timer (create instance B) // T3: instance B fires, blocked trying to obtain L. // T1: Attempt to stop instance B (set earlyReturn = true) -// T1: Reset timer (create instance C) +// T1: Schedule timer (create instance C) // T4: instance C fires, blocked trying to obtain L. // T1: Attempt to stop instance C (set earlyReturn = true) -// T1: Reset timer (create instance D) +// T1: Schedule timer (create instance D) // T5: instance D fires, blocked trying to obtain L. // T1: Release L // -// Now that T1 has released L, any of the 4 timer instances can take L and check -// earlyReturn. If the timers simply check earlyReturn and then do nothing -// further, then instance D will never early return even though it was not -// requested to stop. If the timers reset earlyReturn before early returning, -// then all but one of the timers will do work when only one was expected to. -// If CancellableTimer resets earlyReturn when resetting, then all the timers +// Now that T1 has released L, any of the 4 timer instances can take L and +// check earlyReturn. If the timers simply check earlyReturn and then do +// nothing further, then instance D will never early return even though it was +// not requested to stop. If the timers reset earlyReturn before early +// returning, then all but one of the timers will do work when only one was +// expected to. If Job resets earlyReturn when resetting, then all the timers // will fire (again, when only one was expected to). // // To address the above concerns the simplest solution was to give each timer // its own earlyReturn signal. -type cancellableTimerInstance struct { - timer *time.Timer +type jobInstance struct { + timer Timer // Used to inform the timer to early return when it gets stopped while the // lock the timer tries to obtain when fired is held (T1 is a goroutine that // tries to cancel the timer and T2 is the goroutine that handles the timer // firing): - // T1: Obtain the lock, then call StopLocked() + // T1: Obtain the lock, then call Cancel() // T2: timer fires, and gets blocked on obtaining the lock // T1: Releases lock // T2: Obtains lock does unintended work @@ -74,29 +73,33 @@ type cancellableTimerInstance struct { earlyReturn *bool } -// stop stops the timer instance t from firing if it hasn't fired already. If it +// stop stops the job instance j from firing if it hasn't fired already. If it // has fired and is blocked at obtaining the lock, earlyReturn will be set to // true so that it will early return when it obtains the lock. -func (t *cancellableTimerInstance) stop() { - if t.timer != nil { - t.timer.Stop() - *t.earlyReturn = true +func (j *jobInstance) stop() { + if j.timer != nil { + j.timer.Stop() + *j.earlyReturn = true } } -// CancellableTimer is a timer that does some work and can be safely cancelled -// when it fires at the same time some "related work" is being done. +// Job represents some work that can be scheduled for execution. The work can +// be safely cancelled when it fires at the same time some "related work" is +// being done. // // The term "related work" is defined as some work that needs to be done while // holding some lock that the timer must also hold while doing some work. // -// Note, it is not safe to copy a CancellableTimer as its timer instance creates -// a closure over the address of the CancellableTimer. -type CancellableTimer struct { +// Note, it is not safe to copy a Job as its timer instance creates +// a closure over the address of the Job. +type Job struct { _ sync.NoCopy + // The clock used to schedule the backing timer + clock Clock + // The active instance of a cancellable timer. - instance cancellableTimerInstance + instance jobInstance // locker is the lock taken by the timer immediately after it fires and must // be held when attempting to stop the timer. @@ -113,59 +116,91 @@ type CancellableTimer struct { fn func() } -// StopLocked prevents the Timer from firing if it has not fired already. +// Cancel prevents the Job from executing if it has not executed already. // -// If the timer is blocked on obtaining the t.locker lock when StopLocked is -// called, it will early return instead of calling t.fn. +// Cancel requires appropriate locking to be in place for any resources managed +// by the Job. If the Job is blocked on obtaining the lock when Cancel is +// called, it will early return. // // Note, t will be modified. // -// t.locker MUST be locked. -func (t *CancellableTimer) StopLocked() { - t.instance.stop() +// j.locker MUST be locked. +func (j *Job) Cancel() { + j.instance.stop() // Nothing to do with the stopped instance anymore. - t.instance = cancellableTimerInstance{} + j.instance = jobInstance{} } -// Reset changes the timer to expire after duration d. +// Schedule schedules the Job for execution after duration d. This can be +// called on cancelled or completed Jobs to schedule them again. // -// Note, t will be modified. +// Schedule should be invoked only on unscheduled, cancelled, or completed +// Jobs. To be safe, callers should always call Cancel before calling Schedule. // -// Reset should only be called on stopped or expired timers. To be safe, callers -// should always call StopLocked before calling Reset. -func (t *CancellableTimer) Reset(d time.Duration) { +// Note, j will be modified. +func (j *Job) Schedule(d time.Duration) { // Create a new instance. earlyReturn := false // Capture the locker so that updating the timer does not cause a data race // when a timer fires and tries to obtain the lock (read the timer's locker). - locker := t.locker - t.instance = cancellableTimerInstance{ - timer: time.AfterFunc(d, func() { + locker := j.locker + j.instance = jobInstance{ + timer: j.clock.AfterFunc(d, func() { locker.Lock() defer locker.Unlock() if earlyReturn { // If we reach this point, it means that the timer fired while another - // goroutine called StopLocked while it had the lock. Simply return - // here and do nothing further. + // goroutine called Cancel while it had the lock. Simply return here + // and do nothing further. earlyReturn = false return } - t.fn() + j.fn() }), earlyReturn: &earlyReturn, } } -// NewCancellableTimer returns an unscheduled CancellableTimer with the given -// locker and fn. -// -// fn MUST NOT attempt to lock locker. -// -// Callers must call Reset to schedule the timer to fire. -func NewCancellableTimer(locker sync.Locker, fn func()) *CancellableTimer { - return &CancellableTimer{locker: locker, fn: fn} +// NewJob returns a new Job that can be used to schedule f to run in its own +// gorountine. l will be locked before calling f then unlocked after f returns. +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// message = "bar" +// mu.Unlock() +// +// // Output: bar +// +// f MUST NOT attempt to lock l. +// +// l MUST be locked prior to calling the returned job's Cancel(). +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// job.Cancel() +// mu.Unlock() +func NewJob(c Clock, l sync.Locker, f func()) *Job { + return &Job{ + clock: c, + locker: l, + fn: f, + } } diff --git a/pkg/tcpip/timer_test.go b/pkg/tcpip/timer_test.go index b4940e397..a82384c49 100644 --- a/pkg/tcpip/timer_test.go +++ b/pkg/tcpip/timer_test.go @@ -28,8 +28,8 @@ const ( longDuration = 1 * time.Second ) -func TestCancellableTimerReassignment(t *testing.T) { - var timer tcpip.CancellableTimer +func TestJobReschedule(t *testing.T) { + var clock tcpip.StdClock var wg sync.WaitGroup var lock sync.Mutex @@ -43,26 +43,27 @@ func TestCancellableTimerReassignment(t *testing.T) { // that has an active timer (even if it has been stopped as a stopped // timer may be blocked on a lock before it can check if it has been // stopped while another goroutine holds the same lock). - timer = *tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { wg.Done() }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) lock.Unlock() }() } wg.Wait() } -func TestCancellableTimerFire(t *testing.T) { +func TestJobExecution(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -82,17 +83,18 @@ func TestCancellableTimerFire(t *testing.T) { func TestCancellableTimerResetFromLongDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(middleDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(middleDuration) lock.Lock() - timer.StopLocked() + job.Cancel() lock.Unlock() - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -109,16 +111,17 @@ func TestCancellableTimerResetFromLongDuration(t *testing.T) { } } -func TestCancellableTimerResetFromShortDuration(t *testing.T) { +func TestJobRescheduleFromShortDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() // Wait for timer to fire if it wasn't correctly stopped. @@ -128,7 +131,7 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { case <-time.After(middleDuration): } - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -145,17 +148,18 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { } } -func TestCancellableTimerImmediatelyStop(t *testing.T) { +func TestJobImmediatelyCancel(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) for i := 0; i < 1000; i++ { lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() } @@ -167,25 +171,26 @@ func TestCancellableTimerImmediatelyStop(t *testing.T) { } } -func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { +func TestJobCancelledRescheduleWithoutLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() for i := 0; i < 10; i++ { - timer.Reset(middleDuration) + job.Schedule(middleDuration) lock.Lock() // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration * 2) - timer.StopLocked() + job.Cancel() lock.Unlock() } @@ -201,17 +206,18 @@ func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration) - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() @@ -230,18 +236,19 @@ func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { } } -func TestManyCancellableTimerResetUnderLock(t *testing.T) { +func TestManyJobReschedulesUnderLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 678f4e016..4612be4e7 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -797,7 +797,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 7b2083a09..0e46e6355 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -441,6 +441,7 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, Addr: tcpip.Address(hdr.SourceAddress()), } packet.packetInfo.Protocol = netProto + packet.packetInfo.PktType = pkt.PktType } else { // Guess the would-be ethernet header. packet.senderAddr = tcpip.FullAddress{ @@ -448,34 +449,57 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, Addr: tcpip.Address(localAddr), } packet.packetInfo.Protocol = netProto + packet.packetInfo.PktType = pkt.PktType } if ep.cooked { // Cooked packets can simply be queued. - packet.data = pkt.Data + switch pkt.PktType { + case tcpip.PacketHost: + packet.data = pkt.Data + case tcpip.PacketOutgoing: + // Strip Link Header from the Header. + pkt.Header = buffer.NewPrependableFromView(pkt.Header.View()[len(pkt.LinkHeader):]) + combinedVV := pkt.Header.View().ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + default: + panic(fmt.Sprintf("unexpected PktType in pkt: %+v", pkt)) + } + } else { // Raw packets need their ethernet headers prepended before // queueing. var linkHeader buffer.View - if len(pkt.LinkHeader) == 0 { - // We weren't provided with an actual ethernet header, - // so fake one. - ethFields := header.EthernetFields{ - SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), - DstAddr: localAddr, - Type: netProto, + var combinedVV buffer.VectorisedView + if pkt.PktType != tcpip.PacketOutgoing { + if len(pkt.LinkHeader) == 0 { + // We weren't provided with an actual ethernet header, + // so fake one. + ethFields := header.EthernetFields{ + SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), + DstAddr: localAddr, + Type: netProto, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + linkHeader = buffer.View(fakeHeader) + } else { + linkHeader = append(buffer.View(nil), pkt.LinkHeader...) } - fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) - fakeHeader.Encode(ðFields) - linkHeader = buffer.View(fakeHeader) - } else { - linkHeader = append(buffer.View(nil), pkt.LinkHeader...) + combinedVV = linkHeader.ToVectorisedView() + } + if pkt.PktType == tcpip.PacketOutgoing { + // For outgoing packets the Link, Network and Transport + // headers are in the pkt.Header fields normally unless + // a Raw socket is in use in which case pkt.Header could + // be nil. + combinedVV.AppendView(pkt.Header.View()) } - combinedVV := linkHeader.ToVectorisedView() combinedVV.Append(pkt.Data) packet.data = combinedVV } - packet.timestampNS = ep.stack.NowNanoseconds() + packet.timestampNS = ep.stack.Clock().NowNanoseconds() ep.rcvList.PushBack(&packet) ep.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index c2e9fd29f..f85a68554 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -456,7 +456,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { defer e.mu.Unlock() // If a local address was specified, verify that it's valid. - if e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { + if len(addr.Addr) != 0 && e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { return tcpip.ErrBadLocalAddress } @@ -700,7 +700,7 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { } combinedVV.Append(pkt.Data) packet.data = combinedVV - packet.timestampNS = e.stack.NowNanoseconds() + packet.timestampNS = e.stack.Clock().NowNanoseconds() e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 18ff89ffc..e860ee484 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -49,6 +49,7 @@ go_library( "segment_heap.go", "segment_queue.go", "segment_state.go", + "segment_unsafe.go", "snd.go", "snd_state.go", "tcp_endpoint_list.go", diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 81b740115..1798510bc 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -490,6 +490,9 @@ func (h *handshake) resolveRoute() *tcpip.Error { <-h.ep.undrain h.ep.mu.Lock() } + if n¬ifyError != 0 { + return h.ep.takeLastError() + } } // Wait for notification. @@ -616,6 +619,9 @@ func (h *handshake) execute() *tcpip.Error { <-h.ep.undrain h.ep.mu.Lock() } + if n¬ifyError != 0 { + return h.ep.takeLastError() + } case wakerForNewSegment: if err := h.processSegments(); err != nil { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 83dc10ed0..0f7487963 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1209,6 +1209,14 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } +func (e *endpoint) takeLastError() *tcpip.Error { + e.lastErrorMu.Lock() + defer e.lastErrorMu.Unlock() + err := e.lastError + e.lastError = nil + return err +} + // Read reads data from the endpoint. func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { e.LockUser() @@ -1956,11 +1964,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { switch o := opt.(type) { case tcpip.ErrorOption: - e.lastErrorMu.Lock() - err := e.lastError - e.lastError = nil - e.lastErrorMu.Unlock() - return err + return e.takeLastError() case *tcpip.BindToDeviceOption: e.LockUser() @@ -2546,6 +2550,18 @@ func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.C e.sndBufMu.Unlock() e.notifyProtocolGoroutine(notifyMTUChanged) + + case stack.ControlNoRoute: + e.lastErrorMu.Lock() + e.lastError = tcpip.ErrNoRoute + e.lastErrorMu.Unlock() + e.notifyProtocolGoroutine(notifyError) + + case stack.ControlNetworkUnreachable: + e.lastErrorMu.Lock() + e.lastError = tcpip.ErrNetworkUnreachable + e.lastErrorMu.Unlock() + e.notifyProtocolGoroutine(notifyError) } } diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index b34e47bbd..5d6174a59 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -49,7 +49,7 @@ const ( // DefaultReceiveBufferSize is the default size of the receive buffer // for an endpoint. - DefaultReceiveBufferSize = 1 << 20 // 1MB + DefaultReceiveBufferSize = 32 << 10 // 32KB // MaxBufferSize is the largest size a receive/send buffer can grow to. MaxBufferSize = 4 << 20 // 4MB diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index dd89a292a..5e0bfe585 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -372,7 +372,7 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { // We only store the segment if it's within our buffer // size limit. if r.pendingBufUsed < r.pendingBufSize { - r.pendingBufUsed += s.logicalLen() + r.pendingBufUsed += seqnum.Size(s.segMemSize()) s.incRef() heap.Push(&r.pendingRcvdSegments, s) UpdateSACKBlocks(&r.ep.sack, segSeq, segSeq.Add(segLen), r.rcvNxt) @@ -406,7 +406,7 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { } heap.Pop(&r.pendingRcvdSegments) - r.pendingBufUsed -= s.logicalLen() + r.pendingBufUsed -= seqnum.Size(s.segMemSize()) s.decRef() } return false, nil diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 0280892a8..bb60dc29d 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -138,6 +138,12 @@ func (s *segment) logicalLen() seqnum.Size { return l } +// segMemSize is the amount of memory used to hold the segment data and +// the associated metadata. +func (s *segment) segMemSize() int { + return segSize + s.data.Size() +} + // parse populates the sequence & ack numbers, flags, and window fields of the // segment from the TCP header stored in the data. It then updates the view to // skip the header. diff --git a/pkg/tcpip/transport/tcp/segment_unsafe.go b/pkg/tcpip/transport/tcp/segment_unsafe.go new file mode 100644 index 000000000..0ab7b8f56 --- /dev/null +++ b/pkg/tcpip/transport/tcp/segment_unsafe.go @@ -0,0 +1,23 @@ +// Copyright 2020 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 tcp + +import ( + "unsafe" +) + +const ( + segSize = int(unsafe.Sizeof(segment{})) +) diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 06fde2a79..37e7767d6 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -143,12 +143,14 @@ func New(t *testing.T, mtu uint32) *Context { TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, }) + const sendBufferSize = 1 << 20 // 1 MiB + const recvBufferSize = 1 << 20 // 1 MiB // Allow minimum send/receive buffer sizes to be 1 during tests. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{Min: 1, Default: tcp.DefaultSendBufferSize, Max: 10 * tcp.DefaultSendBufferSize}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{Min: 1, Default: sendBufferSize, Max: 10 * sendBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: tcp.DefaultReceiveBufferSize, Max: 10 * tcp.DefaultReceiveBufferSize}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: recvBufferSize, Max: 10 * recvBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -202,7 +204,7 @@ func New(t *testing.T, mtu uint32) *Context { t: t, s: s, linkEP: ep, - WindowScale: uint8(tcp.FindWndScale(tcp.DefaultReceiveBufferSize)), + WindowScale: uint8(tcp.FindWndScale(recvBufferSize)), } } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index a14643ae8..6e692da07 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1451,7 +1451,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() } - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() |