diff options
Diffstat (limited to 'pkg/tcpip/stack/nic.go')
-rw-r--r-- | pkg/tcpip/stack/nic.go | 173 |
1 files changed, 107 insertions, 66 deletions
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 0f545f255..e56a624fe 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -53,6 +53,8 @@ type NIC struct { // complete. linkResQueue packetsPendingLinkResolution + linkAddrCache *linkAddrCache + mu struct { sync.RWMutex spoofing bool @@ -138,7 +140,8 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC stats: makeNICStats(), networkEndpoints: make(map[tcpip.NetworkProtocolNumber]NetworkEndpoint), } - nic.linkResQueue.init() + nic.linkResQueue.init(nic) + nic.linkAddrCache = newLinkAddrCache(nic, ageLimit, resolutionTimeout, resolutionAttempts) nic.mu.packetEPs = make(map[tcpip.NetworkProtocolNumber]*packetEndpointList) // Check for Neighbor Unreachability Detection support. @@ -167,7 +170,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC for _, netProto := range stack.networkProtocols { netNum := netProto.Number() nic.mu.packetEPs[netNum] = new(packetEndpointList) - nic.networkEndpoints[netNum] = netProto.NewEndpoint(nic, stack, nud, nic) + nic.networkEndpoints[netNum] = netProto.NewEndpoint(nic, nic.linkAddrCache, nud, nic) } nic.LinkEndpoint.Attach(nic) @@ -228,7 +231,9 @@ func (n *NIC) disableLocked() { // // This matches linux's behaviour at the time of writing: // https://github.com/torvalds/linux/blob/71c061d2443814de15e177489d5cc00a4a253ef3/net/core/neighbour.c#L371 - if err := n.clearNeighbors(); err != nil && err != tcpip.ErrNotSupported { + switch err := n.clearNeighbors(); err.(type) { + case nil, *tcpip.ErrNotSupported: + default: panic(fmt.Sprintf("n.clearNeighbors(): %s", err)) } @@ -243,7 +248,7 @@ func (n *NIC) disableLocked() { // address (ff02::1), start DAD for permanent addresses, and start soliciting // routers if the stack is not operating as a router. If the stack is also // configured to auto-generate a link-local address, one will be generated. -func (n *NIC) enable() *tcpip.Error { +func (n *NIC) enable() tcpip.Error { n.mu.Lock() defer n.mu.Unlock() @@ -263,7 +268,7 @@ func (n *NIC) enable() *tcpip.Error { // remove detaches NIC from the link endpoint and releases network endpoint // resources. This guarantees no packets between this NIC and the network // stack. -func (n *NIC) remove() *tcpip.Error { +func (n *NIC) remove() tcpip.Error { n.mu.Lock() defer n.mu.Unlock() @@ -299,48 +304,69 @@ func (n *NIC) IsLoopback() bool { } // WritePacket implements NetworkLinkEndpoint. -func (n *NIC) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { - // As per relevant RFCs, we should queue packets while we wait for link - // resolution to complete. - // - // RFC 1122 section 2.3.2.2 (for IPv4): - // The link layer SHOULD save (rather than discard) at least - // one (the latest) packet of each set of packets destined to - // the same unresolved IP address, and transmit the saved - // packet when the address has been resolved. - // - // RFC 4861 section 7.2.2 (for IPv6): - // While waiting for address resolution to complete, the sender MUST, for - // each neighbor, retain a small queue of packets waiting for address - // resolution to complete. The queue MUST hold at least one packet, and MAY - // contain more. However, the number of queued packets per neighbor SHOULD - // be limited to some small value. When a queue overflows, the new arrival - // SHOULD replace the oldest entry. Once address resolution completes, the - // node transmits any queued packets. - if ch, err := r.Resolve(nil); err != nil { - if err == tcpip.ErrWouldBlock { - r.Acquire() - n.linkResQueue.enqueue(ch, r, protocol, pkt) - return nil +func (n *NIC) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) tcpip.Error { + _, err := n.enqueuePacketBuffer(r, gso, protocol, pkt) + return err +} + +func (n *NIC) writePacketBuffer(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt pendingPacketBuffer) (int, tcpip.Error) { + switch pkt := pkt.(type) { + case *PacketBuffer: + if err := n.writePacket(r, gso, protocol, pkt); err != nil { + return 0, err } - return err + return 1, nil + case *PacketBufferList: + return n.writePackets(r, gso, protocol, *pkt) + default: + panic(fmt.Sprintf("unrecognized pending packet buffer type = %T", pkt)) } +} - return n.writePacket(r.Fields(), gso, protocol, pkt) +func (n *NIC) enqueuePacketBuffer(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt pendingPacketBuffer) (int, tcpip.Error) { + routeInfo, _, err := r.resolvedFields(nil) + switch err.(type) { + case nil: + return n.writePacketBuffer(routeInfo, gso, protocol, pkt) + case *tcpip.ErrWouldBlock: + // As per relevant RFCs, we should queue packets while we wait for link + // resolution to complete. + // + // RFC 1122 section 2.3.2.2 (for IPv4): + // The link layer SHOULD save (rather than discard) at least + // one (the latest) packet of each set of packets destined to + // the same unresolved IP address, and transmit the saved + // packet when the address has been resolved. + // + // RFC 4861 section 7.2.2 (for IPv6): + // While waiting for address resolution to complete, the sender MUST, for + // each neighbor, retain a small queue of packets waiting for address + // resolution to complete. The queue MUST hold at least one packet, and + // MAY contain more. However, the number of queued packets per neighbor + // SHOULD be limited to some small value. When a queue overflows, the new + // arrival SHOULD replace the oldest entry. Once address resolution + // completes, the node transmits any queued packets. + return n.linkResQueue.enqueue(r, gso, protocol, pkt) + default: + return 0, err + } } // WritePacketToRemote implements NetworkInterface. -func (n *NIC) WritePacketToRemote(remoteLinkAddr tcpip.LinkAddress, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { +func (n *NIC) WritePacketToRemote(remoteLinkAddr tcpip.LinkAddress, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) tcpip.Error { var r RouteInfo r.NetProto = protocol r.RemoteLinkAddress = remoteLinkAddr return n.writePacket(r, gso, protocol, pkt) } -func (n *NIC) writePacket(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { +func (n *NIC) writePacket(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) tcpip.Error { // WritePacket takes ownership of pkt, calculate numBytes first. numBytes := pkt.Size() + pkt.EgressRoute = r + pkt.GSOOptions = gso + pkt.NetworkProtocolNumber = protocol if err := n.LinkEndpoint.WritePacket(r, gso, protocol, pkt); err != nil { return err } @@ -351,10 +377,18 @@ func (n *NIC) writePacket(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolN } // WritePackets implements NetworkLinkEndpoint. -func (n *NIC) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - // TODO(gvisor.dev/issue/4458): Queue packets whie link address resolution - // is being peformed like WritePacket. - writtenPackets, err := n.LinkEndpoint.WritePackets(r.Fields(), gso, pkts, protocol) +func (n *NIC) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, tcpip.Error) { + return n.enqueuePacketBuffer(r, gso, protocol, &pkts) +} + +func (n *NIC) writePackets(r RouteInfo, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkts PacketBufferList) (int, tcpip.Error) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + pkt.EgressRoute = r + pkt.GSOOptions = gso + pkt.NetworkProtocolNumber = protocol + } + + writtenPackets, err := n.LinkEndpoint.WritePackets(r, gso, pkts, protocol) n.stats.Tx.Packets.IncrementBy(uint64(writtenPackets)) writtenBytes := 0 for i, pb := 0, pkts.Front(); i < writtenPackets && pb != nil; i, pb = i+1, pb.Next() { @@ -463,15 +497,15 @@ func (n *NIC) getAddressOrCreateTempInner(protocol tcpip.NetworkProtocolNumber, // addAddress adds a new address to n, so that it starts accepting packets // targeted at the given address (and network protocol). -func (n *NIC) addAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { +func (n *NIC) addAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) tcpip.Error { ep, ok := n.networkEndpoints[protocolAddress.Protocol] if !ok { - return tcpip.ErrUnknownProtocol + return &tcpip.ErrUnknownProtocol{} } addressableEndpoint, ok := ep.(AddressableEndpoint) if !ok { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } addressEndpoint, err := addressableEndpoint.AddAndAcquirePermanentAddress(protocolAddress.AddressWithPrefix, peb, AddressConfigStatic, false /* deprecated */) @@ -535,63 +569,70 @@ func (n *NIC) primaryAddress(proto tcpip.NetworkProtocolNumber) tcpip.AddressWit } // removeAddress removes an address from n. -func (n *NIC) removeAddress(addr tcpip.Address) *tcpip.Error { +func (n *NIC) removeAddress(addr tcpip.Address) tcpip.Error { for _, ep := range n.networkEndpoints { addressableEndpoint, ok := ep.(AddressableEndpoint) if !ok { continue } - if err := addressableEndpoint.RemovePermanentAddress(addr); err == tcpip.ErrBadLocalAddress { + switch err := addressableEndpoint.RemovePermanentAddress(addr); err.(type) { + case *tcpip.ErrBadLocalAddress: continue - } else { + default: return err } } - return tcpip.ErrBadLocalAddress + return &tcpip.ErrBadLocalAddress{} +} + +func (n *NIC) confirmReachable(addr tcpip.Address) { + if n := n.neigh; n != nil { + n.handleUpperLevelConfirmation(addr) + } } -func (n *NIC) getNeighborLinkAddress(addr, localAddr tcpip.Address, linkRes LinkAddressResolver, onResolve func(tcpip.LinkAddress, bool)) (tcpip.LinkAddress, <-chan struct{}, *tcpip.Error) { +func (n *NIC) getNeighborLinkAddress(addr, localAddr tcpip.Address, linkRes LinkAddressResolver, onResolve func(LinkResolutionResult)) (tcpip.LinkAddress, <-chan struct{}, tcpip.Error) { if n.neigh != nil { entry, ch, err := n.neigh.entry(addr, localAddr, linkRes, onResolve) return entry.LinkAddr, ch, err } - return n.stack.linkAddrCache.get(tcpip.FullAddress{NIC: n.ID(), Addr: addr}, linkRes, localAddr, n, onResolve) + return n.linkAddrCache.get(addr, linkRes, localAddr, n, onResolve) } -func (n *NIC) neighbors() ([]NeighborEntry, *tcpip.Error) { +func (n *NIC) neighbors() ([]NeighborEntry, tcpip.Error) { if n.neigh == nil { - return nil, tcpip.ErrNotSupported + return nil, &tcpip.ErrNotSupported{} } return n.neigh.entries(), nil } -func (n *NIC) addStaticNeighbor(addr tcpip.Address, linkAddress tcpip.LinkAddress) *tcpip.Error { +func (n *NIC) addStaticNeighbor(addr tcpip.Address, linkAddress tcpip.LinkAddress) tcpip.Error { if n.neigh == nil { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } n.neigh.addStaticEntry(addr, linkAddress) return nil } -func (n *NIC) removeNeighbor(addr tcpip.Address) *tcpip.Error { +func (n *NIC) removeNeighbor(addr tcpip.Address) tcpip.Error { if n.neigh == nil { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } if !n.neigh.removeEntry(addr) { - return tcpip.ErrBadAddress + return &tcpip.ErrBadAddress{} } return nil } -func (n *NIC) clearNeighbors() *tcpip.Error { +func (n *NIC) clearNeighbors() tcpip.Error { if n.neigh == nil { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } n.neigh.clear() @@ -600,7 +641,7 @@ func (n *NIC) clearNeighbors() *tcpip.Error { // joinGroup adds a new endpoint for the given multicast address, if none // exists yet. Otherwise it just increments its count. -func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { +func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.Error { // TODO(b/143102137): When implementing MLD, make sure MLD packets are // not sent unless a valid link-local address is available for use on n // as an MLD packet's source address must be a link-local address as @@ -608,12 +649,12 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address ep, ok := n.networkEndpoints[protocol] if !ok { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } gep, ok := ep.(GroupAddressableEndpoint) if !ok { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } return gep.JoinGroup(addr) @@ -621,15 +662,15 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address // leaveGroup decrements the count for the given multicast address, and when it // reaches zero removes the endpoint for this address. -func (n *NIC) leaveGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { +func (n *NIC) leaveGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) tcpip.Error { ep, ok := n.networkEndpoints[protocol] if !ok { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } gep, ok := ep.(GroupAddressableEndpoint) if !ok { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } return gep.LeaveGroup(addr) @@ -879,9 +920,9 @@ func (n *NIC) Name() string { } // nudConfigs gets the NUD configurations for n. -func (n *NIC) nudConfigs() (NUDConfigurations, *tcpip.Error) { +func (n *NIC) nudConfigs() (NUDConfigurations, tcpip.Error) { if n.neigh == nil { - return NUDConfigurations{}, tcpip.ErrNotSupported + return NUDConfigurations{}, &tcpip.ErrNotSupported{} } return n.neigh.config(), nil } @@ -890,22 +931,22 @@ func (n *NIC) nudConfigs() (NUDConfigurations, *tcpip.Error) { // // Note, if c contains invalid NUD configuration values, it will be fixed to // use default values for the erroneous values. -func (n *NIC) setNUDConfigs(c NUDConfigurations) *tcpip.Error { +func (n *NIC) setNUDConfigs(c NUDConfigurations) tcpip.Error { if n.neigh == nil { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } c.resetInvalidFields() n.neigh.setConfig(c) return nil } -func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { +func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) tcpip.Error { n.mu.Lock() defer n.mu.Unlock() eps, ok := n.mu.packetEPs[netProto] if !ok { - return tcpip.ErrNotSupported + return &tcpip.ErrNotSupported{} } eps.add(ep) |