diff options
Diffstat (limited to 'pkg/tcpip/stack/nic.go')
-rw-r--r-- | pkg/tcpip/stack/nic.go | 1346 |
1 files changed, 1004 insertions, 342 deletions
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index ab6798aa6..e74d2562a 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -15,48 +15,66 @@ package stack import ( - "strings" - "sync" + "fmt" + "math/rand" + "reflect" + "sort" "sync/atomic" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" ) +var ipv4BroadcastAddr = tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: header.IPv4Broadcast, + PrefixLen: 8 * header.IPv4AddressSize, + }, +} + // NIC represents a "network interface card" to which the networking stack is // attached. type NIC struct { - stack *Stack - id tcpip.NICID - name string - linkEP LinkEndpoint - loopback bool - - mu sync.RWMutex - spoofing bool - promiscuous bool - primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint - endpoints map[NetworkEndpointID]*referencedNetworkEndpoint - addressRanges []tcpip.Subnet - mcastJoins map[NetworkEndpointID]int32 - // packetEPs is protected by mu, but the contained PacketEndpoint - // values are not. - packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint - - stats NICStats - - // ndp is the NDP related state for NIC. - // - // Note, read and write operations on ndp require that the NIC is - // appropriately locked. - ndp ndpState + stack *Stack + id tcpip.NICID + name string + linkEP LinkEndpoint + context NICContext + + stats NICStats + neigh *neighborCache + networkEndpoints map[tcpip.NetworkProtocolNumber]NetworkEndpoint + + mu struct { + sync.RWMutex + enabled bool + spoofing bool + promiscuous bool + primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint + endpoints map[NetworkEndpointID]*referencedNetworkEndpoint + mcastJoins map[NetworkEndpointID]uint32 + // packetEPs is protected by mu, but the contained PacketEndpoint + // values are not. + packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint + ndp ndpState + } } // NICStats includes transmitted and received stats. type NICStats struct { Tx DirectionStats Rx DirectionStats + + DisabledRx DirectionStats +} + +func makeNICStats() NICStats { + var s NICStats + tcpip.InitStatCounters(reflect.ValueOf(&s).Elem()) + return s } // DirectionStats includes packet and byte counts. @@ -85,59 +103,170 @@ const ( ) // newNIC returns a new NIC using the default NDP configurations from stack. -func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback bool) *NIC { +func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICContext) *NIC { // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For // example, make sure that the link address it provides is a valid // unicast ethernet address. + + // TODO(b/143357959): RFC 8200 section 5 requires that IPv6 endpoints + // observe an MTU of at least 1280 bytes. Ensure that this requirement + // of IPv6 is supported on this endpoint's LinkEndpoint. + nic := &NIC{ - stack: stack, - id: id, - name: name, - linkEP: ep, - loopback: loopback, - primary: make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint), - endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint), - mcastJoins: make(map[NetworkEndpointID]int32), - packetEPs: make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint), - stats: NICStats{ - Tx: DirectionStats{ - Packets: &tcpip.StatCounter{}, - Bytes: &tcpip.StatCounter{}, - }, - Rx: DirectionStats{ - Packets: &tcpip.StatCounter{}, - Bytes: &tcpip.StatCounter{}, - }, - }, - ndp: ndpState{ - configs: stack.ndpConfigs, - dad: make(map[tcpip.Address]dadState), - }, - } - nic.ndp.nic = nic + stack: stack, + id: id, + name: name, + linkEP: ep, + context: ctx, + stats: makeNICStats(), + networkEndpoints: make(map[tcpip.NetworkProtocolNumber]NetworkEndpoint), + } + nic.mu.primary = make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint) + nic.mu.endpoints = make(map[NetworkEndpointID]*referencedNetworkEndpoint) + nic.mu.mcastJoins = make(map[NetworkEndpointID]uint32) + nic.mu.packetEPs = make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint) + nic.mu.ndp = ndpState{ + nic: nic, + configs: stack.ndpConfigs, + dad: make(map[tcpip.Address]dadState), + defaultRouters: make(map[tcpip.Address]defaultRouterState), + onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), + slaacPrefixes: make(map[tcpip.Subnet]slaacPrefixState), + } + nic.mu.ndp.initializeTempAddrState() // Register supported packet endpoint protocols. for _, netProto := range header.Ethertypes { - nic.packetEPs[netProto] = []PacketEndpoint{} + nic.mu.packetEPs[netProto] = []PacketEndpoint{} } for _, netProto := range stack.networkProtocols { - nic.packetEPs[netProto.Number()] = []PacketEndpoint{} + netNum := netProto.Number() + nic.mu.packetEPs[netNum] = nil + nic.networkEndpoints[netNum] = netProto.NewEndpoint(id, stack, nic, ep, stack) + } + + // Check for Neighbor Unreachability Detection support. + if ep.Capabilities()&CapabilityResolutionRequired != 0 && len(stack.linkAddrResolvers) != 0 { + rng := rand.New(rand.NewSource(stack.clock.NowNanoseconds())) + nic.neigh = &neighborCache{ + nic: nic, + state: NewNUDState(stack.nudConfigs, rng), + cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize), + } } + nic.linkEP.Attach(nic) + return nic } -// enable enables the NIC. enable will attach the link to its LinkEndpoint and -// join the IPv6 All-Nodes Multicast address (ff02::1). +// enabled returns true if n is enabled. +func (n *NIC) enabled() bool { + n.mu.RLock() + enabled := n.mu.enabled + n.mu.RUnlock() + return enabled +} + +// disable disables n. +// +// It undoes the work done by enable. +func (n *NIC) disable() *tcpip.Error { + n.mu.RLock() + enabled := n.mu.enabled + n.mu.RUnlock() + if !enabled { + return nil + } + + n.mu.Lock() + err := n.disableLocked() + n.mu.Unlock() + return err +} + +// disableLocked disables n. +// +// It undoes the work done by enable. +// +// n MUST be locked. +func (n *NIC) disableLocked() *tcpip.Error { + if !n.mu.enabled { + return nil + } + + // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be + // invalidated? Currently, Routes will continue to work when a NIC is enabled + // again, and applications may not know that the underlying NIC was ever + // disabled. + + if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok { + n.mu.ndp.stopSolicitingRouters() + n.mu.ndp.cleanupState(false /* hostOnly */) + + // Stop DAD for all the unicast IPv6 endpoints that are in the + // permanentTentative state. + for _, r := range n.mu.endpoints { + if addr := r.address(); r.getKind() == permanentTentative && header.IsV6UnicastAddress(addr) { + n.mu.ndp.stopDuplicateAddressDetection(addr) + } + } + + // The NIC may have already left the multicast group. + if err := n.leaveGroupLocked(header.IPv6AllNodesMulticastAddress, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { + return err + } + } + + if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { + // The NIC may have already left the multicast group. + if err := n.leaveGroupLocked(header.IPv4AllSystems, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { + return err + } + + // The address may have already been removed. + if err := n.removePermanentAddressLocked(ipv4BroadcastAddr.AddressWithPrefix.Address); err != nil && err != tcpip.ErrBadLocalAddress { + return err + } + } + + n.mu.enabled = false + return nil +} + +// enable enables n. +// +// If the stack has IPv6 enabled, enable will join the IPv6 All-Nodes Multicast +// 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 { - n.attachLinkEndpoint() + n.mu.RLock() + enabled := n.mu.enabled + n.mu.RUnlock() + if enabled { + return nil + } + + n.mu.Lock() + defer n.mu.Unlock() + + if n.mu.enabled { + return nil + } + + n.mu.enabled = true // Create an endpoint to receive broadcast packets on this interface. if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { - if err := n.AddAddress(tcpip.ProtocolAddress{ - Protocol: header.IPv4ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Broadcast, 8 * header.IPv4AddressSize}, - }, NeverPrimaryEndpoint); err != nil { + if _, err := n.addAddressLocked(ipv4BroadcastAddr, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { + return err + } + + // As per RFC 1122 section 3.3.7, all hosts should join the all-hosts + // multicast group. Note, the IANA calls the all-hosts multicast group the + // all-systems multicast group. + if err := n.joinGroupLocked(header.IPv4ProtocolNumber, header.IPv4AllSystems); err != nil { return err } } @@ -159,77 +288,298 @@ func (n *NIC) enable() *tcpip.Error { return nil } - n.mu.Lock() - defer n.mu.Unlock() - + // Join the All-Nodes multicast group before starting DAD as responses to DAD + // (NDP NS) messages may be sent to the All-Nodes multicast group if the + // source address of the NDP NS is the unspecified address, as per RFC 4861 + // section 7.2.4. if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil { return err } - if !n.stack.autoGenIPv6LinkLocal { - return nil + // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent + // state. + // + // Addresses may have aleady completed DAD but in the time since the NIC was + // last enabled, other devices may have acquired the same addresses. + for _, r := range n.mu.endpoints { + addr := r.address() + if k := r.getKind(); (k != permanent && k != permanentTentative) || !header.IsV6UnicastAddress(addr) { + continue + } + + r.setKind(permanentTentative) + if err := n.mu.ndp.startDuplicateAddressDetection(addr, r); err != nil { + return err + } } - l2addr := n.linkEP.LinkAddress() + // Do not auto-generate an IPv6 link-local address for loopback devices. + if n.stack.autoGenIPv6LinkLocal && !n.isLoopback() { + // The valid and preferred lifetime is infinite for the auto-generated + // link-local address. + n.mu.ndp.doSLAAC(header.IPv6LinkLocalPrefix.Subnet(), header.NDPInfiniteLifetime, header.NDPInfiniteLifetime) + } - // Only attempt to generate the link-local address if we have a - // valid MAC address. + // If we are operating as a router, then do not solicit routers since we + // won't process the RAs anyways. // - // TODO(b/141011931): Validate a LinkEndpoint's link address - // (provided by LinkEndpoint.LinkAddress) before reaching this - // point. - if !header.IsValidUnicastEthernetAddress(l2addr) { - return nil + // Routers do not process Router Advertisements (RA) the same way a host + // does. That is, routers do not learn from RAs (e.g. on-link prefixes + // and default routers). Therefore, soliciting RAs from other routers on + // a link is unnecessary for routers. + if !n.stack.Forwarding(header.IPv6ProtocolNumber) { + n.mu.ndp.startSolicitingRouters() } - addr := header.LinkLocalAddr(l2addr) + return nil +} - _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{ - Protocol: header.IPv6ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: addr, - PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen, - }, - }, CanBePrimaryEndpoint) +// remove detaches NIC from the link endpoint, and marks existing referenced +// network endpoints expired. This guarantees no packets between this NIC and +// the network stack. +func (n *NIC) remove() *tcpip.Error { + n.mu.Lock() + defer n.mu.Unlock() + + n.disableLocked() + + // TODO(b/151378115): come up with a better way to pick an error than the + // first one. + var err *tcpip.Error + + // Forcefully leave multicast groups. + for nid := range n.mu.mcastJoins { + if tempErr := n.leaveGroupLocked(nid.LocalAddress, true /* force */); tempErr != nil && err == nil { + err = tempErr + } + } + + // Remove permanent and permanentTentative addresses, so no packet goes out. + for nid, ref := range n.mu.endpoints { + switch ref.getKind() { + case permanentTentative, permanent: + if tempErr := n.removePermanentAddressLocked(nid.LocalAddress); tempErr != nil && err == nil { + err = tempErr + } + } + } + + // Release any resources the network endpoint may hold. + for _, ep := range n.networkEndpoints { + ep.Close() + } + + // Detach from link endpoint, so no packet comes in. + n.linkEP.Attach(nil) return err } -// attachLinkEndpoint attaches the NIC to the endpoint, which will enable it -// to start delivering packets. -func (n *NIC) attachLinkEndpoint() { - n.linkEP.Attach(n) +// becomeIPv6Router transitions n into an IPv6 router. +// +// When transitioning into an IPv6 router, host-only state (NDP discovered +// routers, discovered on-link prefixes, and auto-generated addresses) will +// be cleaned up/invalidated and NDP router solicitations will be stopped. +func (n *NIC) becomeIPv6Router() { + n.mu.Lock() + defer n.mu.Unlock() + + n.mu.ndp.cleanupState(true /* hostOnly */) + n.mu.ndp.stopSolicitingRouters() +} + +// becomeIPv6Host transitions n into an IPv6 host. +// +// When transitioning into an IPv6 host, NDP router solicitations will be +// started. +func (n *NIC) becomeIPv6Host() { + n.mu.Lock() + defer n.mu.Unlock() + + n.mu.ndp.startSolicitingRouters() } // setPromiscuousMode enables or disables promiscuous mode. func (n *NIC) setPromiscuousMode(enable bool) { n.mu.Lock() - n.promiscuous = enable + n.mu.promiscuous = enable n.mu.Unlock() } func (n *NIC) isPromiscuousMode() bool { n.mu.RLock() - rv := n.promiscuous + rv := n.mu.promiscuous n.mu.RUnlock() return rv } +func (n *NIC) isLoopback() bool { + return n.linkEP.Capabilities()&CapabilityLoopback != 0 +} + // setSpoofing enables or disables address spoofing. func (n *NIC) setSpoofing(enable bool) { n.mu.Lock() - n.spoofing = enable + n.mu.spoofing = enable n.mu.Unlock() } -// primaryEndpoint returns the primary endpoint of n for the given network -// protocol. -func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedNetworkEndpoint { +// primaryEndpoint will return the first non-deprecated endpoint if such an +// endpoint exists for the given protocol and remoteAddr. If no non-deprecated +// endpoint exists, the first deprecated endpoint will be returned. +// +// If an IPv6 primary endpoint is requested, Source Address Selection (as +// defined by RFC 6724 section 5) will be performed. +func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) *referencedNetworkEndpoint { + if protocol == header.IPv6ProtocolNumber && remoteAddr != "" { + return n.primaryIPv6Endpoint(remoteAddr) + } + n.mu.RLock() defer n.mu.RUnlock() - for _, r := range n.primary[protocol] { - if r.isValidForOutgoing() && r.tryIncRef() { + var deprecatedEndpoint *referencedNetworkEndpoint + for _, r := range n.mu.primary[protocol] { + if !r.isValidForOutgoingRLocked() { + continue + } + + if !r.deprecated { + if r.tryIncRef() { + // r is not deprecated, so return it immediately. + // + // If we kept track of a deprecated endpoint, decrement its reference + // count since it was incremented when we decided to keep track of it. + if deprecatedEndpoint != nil { + deprecatedEndpoint.decRefLocked() + deprecatedEndpoint = nil + } + + return r + } + } else if deprecatedEndpoint == nil && r.tryIncRef() { + // We prefer an endpoint that is not deprecated, but we keep track of r in + // case n doesn't have any non-deprecated endpoints. + // + // If we end up finding a more preferred endpoint, r's reference count + // will be decremented when such an endpoint is found. + deprecatedEndpoint = r + } + } + + // n doesn't have any valid non-deprecated endpoints, so return + // deprecatedEndpoint (which may be nil if n doesn't have any valid deprecated + // endpoints either). + return deprecatedEndpoint +} + +// ipv6AddrCandidate is an IPv6 candidate for Source Address Selection (RFC +// 6724 section 5). +type ipv6AddrCandidate struct { + ref *referencedNetworkEndpoint + scope header.IPv6AddressScope +} + +// primaryIPv6Endpoint returns an IPv6 endpoint following Source Address +// Selection (RFC 6724 section 5). +// +// Note, only rules 1-3 and 7 are followed. +// +// remoteAddr must be a valid IPv6 address. +func (n *NIC) primaryIPv6Endpoint(remoteAddr tcpip.Address) *referencedNetworkEndpoint { + n.mu.RLock() + ref := n.primaryIPv6EndpointRLocked(remoteAddr) + n.mu.RUnlock() + return ref +} + +// primaryIPv6EndpointLocked returns an IPv6 endpoint following Source Address +// Selection (RFC 6724 section 5). +// +// Note, only rules 1-3 and 7 are followed. +// +// remoteAddr must be a valid IPv6 address. +// +// n.mu MUST be read locked. +func (n *NIC) primaryIPv6EndpointRLocked(remoteAddr tcpip.Address) *referencedNetworkEndpoint { + primaryAddrs := n.mu.primary[header.IPv6ProtocolNumber] + + if len(primaryAddrs) == 0 { + return nil + } + + // Create a candidate set of available addresses we can potentially use as a + // source address. + cs := make([]ipv6AddrCandidate, 0, len(primaryAddrs)) + for _, r := range primaryAddrs { + // If r is not valid for outgoing connections, it is not a valid endpoint. + if !r.isValidForOutgoingRLocked() { + continue + } + + addr := r.address() + scope, err := header.ScopeForIPv6Address(addr) + if err != nil { + // Should never happen as we got r from the primary IPv6 endpoint list and + // ScopeForIPv6Address only returns an error if addr is not an IPv6 + // address. + panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", addr, err)) + } + + cs = append(cs, ipv6AddrCandidate{ + ref: r, + scope: scope, + }) + } + + remoteScope, err := header.ScopeForIPv6Address(remoteAddr) + if err != nil { + // primaryIPv6Endpoint should never be called with an invalid IPv6 address. + panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err)) + } + + // Sort the addresses as per RFC 6724 section 5 rules 1-3. + // + // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5. + sort.Slice(cs, func(i, j int) bool { + sa := cs[i] + sb := cs[j] + + // Prefer same address as per RFC 6724 section 5 rule 1. + if sa.ref.address() == remoteAddr { + return true + } + if sb.ref.address() == remoteAddr { + return false + } + + // Prefer appropriate scope as per RFC 6724 section 5 rule 2. + if sa.scope < sb.scope { + return sa.scope >= remoteScope + } else if sb.scope < sa.scope { + return sb.scope < remoteScope + } + + // Avoid deprecated addresses as per RFC 6724 section 5 rule 3. + if saDep, sbDep := sa.ref.deprecated, sb.ref.deprecated; saDep != sbDep { + // If sa is not deprecated, it is preferred over sb. + return sbDep + } + + // Prefer temporary addresses as per RFC 6724 section 5 rule 7. + if saTemp, sbTemp := sa.ref.configType == slaacTemp, sb.ref.configType == slaacTemp; saTemp != sbTemp { + return saTemp + } + + // sa and sb are equal, return the endpoint that is closest to the front of + // the primary endpoint list. + return i < j + }) + + // Return the most preferred address that can have its reference count + // incremented. + for _, c := range cs { + if r := c.ref; r.tryIncRef() { return r } } @@ -237,62 +587,87 @@ func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedN return nil } +// hasPermanentAddrLocked returns true if n has a permanent (including currently +// tentative) address, addr. +func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool { + ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] + + if !ok { + return false + } + + kind := ref.getKind() + + return kind == permanent || kind == permanentTentative +} + +type getRefBehaviour int + +const ( + // spoofing indicates that the NIC's spoofing flag should be observed when + // getting a NIC's referenced network endpoint. + spoofing getRefBehaviour = iota + + // promiscuous indicates that the NIC's promiscuous flag should be observed + // when getting a NIC's referenced network endpoint. + promiscuous +) + func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, n.promiscuous) + return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous) } // findEndpoint finds the endpoint, if any, with the given address. func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, address, peb, n.spoofing) + return n.getRefOrCreateTemp(protocol, address, peb, spoofing) } // getRefEpOrCreateTemp returns the referenced network endpoint for the given -// protocol and address. If none exists a temporary one may be created if -// we are in promiscuous mode or spoofing. -func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, spoofingOrPromiscuous bool) *referencedNetworkEndpoint { - id := NetworkEndpointID{address} - +// protocol and address. +// +// If none exists a temporary one may be created if we are in promiscuous mode +// or spoofing. Promiscuous mode will only be checked if promiscuous is true. +// Similarly, spoofing will only be checked if spoofing is true. +// +// If the address is the IPv4 broadcast address for an endpoint's network, that +// endpoint will be returned. +func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getRefBehaviour) *referencedNetworkEndpoint { n.mu.RLock() - if ref, ok := n.endpoints[id]; ok { + var spoofingOrPromiscuous bool + switch tempRef { + case spoofing: + spoofingOrPromiscuous = n.mu.spoofing + case promiscuous: + spoofingOrPromiscuous = n.mu.promiscuous + } + + if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok { // An endpoint with this id exists, check if it can be used and return it. - switch ref.getKind() { - case permanentExpired: - if !spoofingOrPromiscuous { - n.mu.RUnlock() - return nil - } - fallthrough - case temporary, permanent: - if ref.tryIncRef() { - n.mu.RUnlock() - return ref - } + if !ref.isAssignedRLocked(spoofingOrPromiscuous) { + n.mu.RUnlock() + return nil + } + + if ref.tryIncRef() { + n.mu.RUnlock() + return ref } } - // A usable reference was not found, create a temporary one if requested by - // the caller or if the address is found in the NIC's subnets. - createTempEP := spoofingOrPromiscuous - if !createTempEP { - for _, sn := range n.addressRanges { - // Skip the subnet address. - if address == sn.ID() { - continue - } - // For now just skip the broadcast address, until we support it. - // FIXME(b/137608825): Add support for sending/receiving directed - // (subnet) broadcast. - if address == sn.Broadcast() { - continue - } - if sn.Contains(address) { - createTempEP = true - break - } + // Check if address is a broadcast address for the endpoint's network. + // + // Only IPv4 has a notion of broadcast addresses. + if protocol == header.IPv4ProtocolNumber { + if ref := n.getRefForBroadcastRLocked(address); ref != nil { + n.mu.RUnlock() + return ref } } + // A usable reference was not found, create a temporary one if requested by + // the caller or if the address is found in the NIC's subnets. + createTempEP := spoofingOrPromiscuous n.mu.RUnlock() if !createTempEP { @@ -303,11 +678,44 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t // endpoint, create a new "temporary" endpoint. It will only exist while // there's a route through it. n.mu.Lock() - if ref, ok := n.endpoints[id]; ok { + ref := n.getRefOrCreateTempLocked(protocol, address, peb) + n.mu.Unlock() + return ref +} + +// getRefForBroadcastLocked returns an endpoint where address is the IPv4 +// broadcast address for the endpoint's network. +// +// n.mu MUST be read locked. +func (n *NIC) getRefForBroadcastRLocked(address tcpip.Address) *referencedNetworkEndpoint { + for _, ref := range n.mu.endpoints { + // Only IPv4 has a notion of broadcast addresses. + if ref.protocol != header.IPv4ProtocolNumber { + continue + } + + addr := ref.addrWithPrefix() + subnet := addr.Subnet() + if subnet.IsBroadcast(address) && ref.tryIncRef() { + return ref + } + } + + return nil +} + +/// getRefOrCreateTempLocked returns an existing endpoint for address or creates +/// and returns a temporary endpoint. +// +// If the address is the IPv4 broadcast address for an endpoint's network, that +// endpoint will be returned. +// +// n.mu must be write locked. +func (n *NIC) getRefOrCreateTempLocked(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint { + if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok { // No need to check the type as we are ok with expired endpoints at this // point. if ref.tryIncRef() { - n.mu.Unlock() return ref } // tryIncRef failing means the endpoint is scheduled to be removed once the @@ -316,10 +724,18 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t n.removeEndpointLocked(ref) } + // Check if address is a broadcast address for an endpoint's network. + // + // Only IPv4 has a notion of broadcast addresses. + if protocol == header.IPv4ProtocolNumber { + if ref := n.getRefForBroadcastRLocked(address); ref != nil { + return ref + } + } + // Add a new temporary endpoint. netProto, ok := n.stack.networkProtocols[protocol] if !ok { - n.mu.Unlock() return nil } ref, _ := n.addAddressLocked(tcpip.ProtocolAddress{ @@ -328,26 +744,40 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t Address: address, PrefixLen: netProto.DefaultPrefixLen(), }, - }, peb, temporary) - - n.mu.Unlock() + }, peb, temporary, static, false) return ref } -func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) (*referencedNetworkEndpoint, *tcpip.Error) { - id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address} - if ref, ok := n.endpoints[id]; ok { +// addAddressLocked adds a new protocolAddress to n. +// +// If n already has the address in a non-permanent state, and the kind given is +// permanent, that address will be promoted in place and its properties set to +// the properties provided. Otherwise, it returns tcpip.ErrDuplicateAddress. +func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType, deprecated bool) (*referencedNetworkEndpoint, *tcpip.Error) { + // TODO(b/141022673): Validate IP addresses before adding them. + + // Sanity check. + id := NetworkEndpointID{LocalAddress: protocolAddress.AddressWithPrefix.Address} + if ref, ok := n.mu.endpoints[id]; ok { + // Endpoint already exists. + if kind != permanent { + return nil, tcpip.ErrDuplicateAddress + } switch ref.getKind() { case permanentTentative, permanent: // The NIC already have a permanent endpoint with that address. return nil, tcpip.ErrDuplicateAddress case permanentExpired, temporary: - // Promote the endpoint to become permanent and respect - // the new peb. + // Promote the endpoint to become permanent and respect the new peb, + // configType and deprecated status. if ref.tryIncRef() { + // TODO(b/147748385): Perform Duplicate Address Detection when promoting + // an IPv6 endpoint to permanent. ref.setKind(permanent) + ref.deprecated = deprecated + ref.configType = configType - refs := n.primary[ref.protocol] + refs := n.mu.primary[ref.protocol] for i, r := range refs { if r == ref { switch peb { @@ -357,9 +787,9 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p if i == 0 { return ref, nil } - n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) case NeverPrimaryEndpoint: - n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) return ref, nil } } @@ -377,44 +807,30 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p } } - return n.addAddressLocked(protocolAddress, peb, permanent) -} - -func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind) (*referencedNetworkEndpoint, *tcpip.Error) { - // TODO(b/141022673): Validate IP address before adding them. - - // Sanity check. - id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address} - if _, ok := n.endpoints[id]; ok { - // Endpoint already exists. - return nil, tcpip.ErrDuplicateAddress - } - - netProto, ok := n.stack.networkProtocols[protocolAddress.Protocol] + ep, ok := n.networkEndpoints[protocolAddress.Protocol] if !ok { return nil, tcpip.ErrUnknownProtocol } - // Create the new network endpoint. - ep, err := netProto.NewEndpoint(n.id, protocolAddress.AddressWithPrefix, n.stack, n, n.linkEP) - if err != nil { - return nil, err - } - isIPv6Unicast := protocolAddress.Protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(protocolAddress.AddressWithPrefix.Address) // If the address is an IPv6 address and it is a permanent address, - // mark it as tentative so it goes through the DAD process. + // mark it as tentative so it goes through the DAD process if the NIC is + // enabled. If the NIC is not enabled, DAD will be started when the NIC is + // enabled. if isIPv6Unicast && kind == permanent { kind = permanentTentative } ref := &referencedNetworkEndpoint{ - refs: 1, - ep: ep, - nic: n, - protocol: protocolAddress.Protocol, - kind: kind, + refs: 1, + addr: protocolAddress.AddressWithPrefix, + ep: ep, + nic: n, + protocol: protocolAddress.Protocol, + kind: kind, + configType: configType, + deprecated: deprecated, } // Set up cache if link address resolution exists for this protocol. @@ -433,13 +849,13 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar } } - n.endpoints[id] = ref + n.mu.endpoints[id] = ref n.insertPrimaryEndpointLocked(ref, peb) - // If we are adding a tentative IPv6 address, start DAD. - if isIPv6Unicast && kind == permanentTentative { - if err := n.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil { + // If we are adding a tentative IPv6 address, start DAD if the NIC is enabled. + if isIPv6Unicast && kind == permanentTentative && n.mu.enabled { + if err := n.mu.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil { return nil, err } } @@ -452,7 +868,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { // Add the endpoint. n.mu.Lock() - _, err := n.addPermanentAddressLocked(protocolAddress, peb) + _, err := n.addAddressLocked(protocolAddress, peb, permanent, static, false /* deprecated */) n.mu.Unlock() return err @@ -464,22 +880,18 @@ func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { n.mu.RLock() defer n.mu.RUnlock() - addrs := make([]tcpip.ProtocolAddress, 0, len(n.endpoints)) - for nid, ref := range n.endpoints { + addrs := make([]tcpip.ProtocolAddress, 0, len(n.mu.endpoints)) + for _, ref := range n.mu.endpoints { // Don't include tentative, expired or temporary endpoints to // avoid confusion and prevent the caller from using those. switch ref.getKind() { - case permanentTentative, permanentExpired, temporary: - // TODO(b/140898488): Should tentative addresses be - // returned? + case permanentExpired, temporary: continue } + addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: ref.protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: nid.LocalAddress, - PrefixLen: ref.ep.PrefixLen(), - }, + Protocol: ref.protocol, + AddressWithPrefix: ref.addrWithPrefix(), }) } return addrs @@ -491,7 +903,7 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { defer n.mu.RUnlock() var addrs []tcpip.ProtocolAddress - for proto, list := range n.primary { + for proto, list := range n.mu.primary { for _, ref := range list { // Don't include tentative, expired or tempory endpoints // to avoid confusion and prevent the caller from using @@ -502,59 +914,51 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { } addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: proto, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: ref.ep.ID().LocalAddress, - PrefixLen: ref.ep.PrefixLen(), - }, + Protocol: proto, + AddressWithPrefix: ref.addrWithPrefix(), }) } } return addrs } -// AddAddressRange adds a range of addresses to n, so that it starts accepting -// packets targeted at the given addresses and network protocol. The range is -// given by a subnet address, and all addresses contained in the subnet are -// used except for the subnet address itself and the subnet's broadcast -// address. -func (n *NIC) AddAddressRange(protocol tcpip.NetworkProtocolNumber, subnet tcpip.Subnet) { - n.mu.Lock() - n.addressRanges = append(n.addressRanges, subnet) - n.mu.Unlock() -} +// primaryAddress returns the primary address associated with this NIC. +// +// primaryAddress will return the first non-deprecated address if such an +// address exists. If no non-deprecated address exists, the first deprecated +// address will be returned. +func (n *NIC) primaryAddress(proto tcpip.NetworkProtocolNumber) tcpip.AddressWithPrefix { + n.mu.RLock() + defer n.mu.RUnlock() -// RemoveAddressRange removes the given address range from n. -func (n *NIC) RemoveAddressRange(subnet tcpip.Subnet) { - n.mu.Lock() + list, ok := n.mu.primary[proto] + if !ok { + return tcpip.AddressWithPrefix{} + } - // Use the same underlying array. - tmp := n.addressRanges[:0] - for _, sub := range n.addressRanges { - if sub != subnet { - tmp = append(tmp, sub) + var deprecatedEndpoint *referencedNetworkEndpoint + for _, ref := range list { + // Don't include tentative, expired or tempory endpoints to avoid confusion + // and prevent the caller from using those. + switch ref.getKind() { + case permanentTentative, permanentExpired, temporary: + continue } - } - n.addressRanges = tmp - n.mu.Unlock() -} + if !ref.deprecated { + return ref.addrWithPrefix() + } -// Subnets returns the Subnets associated with this NIC. -func (n *NIC) AddressRanges() []tcpip.Subnet { - n.mu.RLock() - defer n.mu.RUnlock() - sns := make([]tcpip.Subnet, 0, len(n.addressRanges)+len(n.endpoints)) - for nid := range n.endpoints { - sn, err := tcpip.NewSubnet(nid.LocalAddress, tcpip.AddressMask(strings.Repeat("\xff", len(nid.LocalAddress)))) - if err != nil { - // This should never happen as the mask has been carefully crafted to - // match the address. - panic("Invalid endpoint subnet: " + err.Error()) + if deprecatedEndpoint == nil { + deprecatedEndpoint = ref } - sns = append(sns, sn) } - return append(sns, n.addressRanges...) + + if deprecatedEndpoint != nil { + return deprecatedEndpoint.addrWithPrefix() + } + + return tcpip.AddressWithPrefix{} } // insertPrimaryEndpointLocked adds r to n's primary endpoint list as required @@ -564,21 +968,21 @@ func (n *NIC) AddressRanges() []tcpip.Subnet { func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) { switch peb { case CanBePrimaryEndpoint: - n.primary[r.protocol] = append(n.primary[r.protocol], r) + n.mu.primary[r.protocol] = append(n.mu.primary[r.protocol], r) case FirstPrimaryEndpoint: - n.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.primary[r.protocol]...) + n.mu.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.mu.primary[r.protocol]...) } } func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { - id := *r.ep.ID() + id := NetworkEndpointID{LocalAddress: r.address()} // Nothing to do if the reference has already been replaced with a different // one. This happens in the case where 1) this endpoint's ref count hit zero // and was waiting (on the lock) to be removed and 2) the same address was // re-added in the meantime by removing this endpoint from the list and // adding a new one. - if n.endpoints[id] != r { + if n.mu.endpoints[id] != r { return } @@ -586,16 +990,15 @@ func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { panic("Reference count dropped to zero before being removed") } - delete(n.endpoints, id) - refs := n.primary[r.protocol] + delete(n.mu.endpoints, id) + refs := n.mu.primary[r.protocol] for i, ref := range refs { if ref == r { - n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + refs[len(refs)-1] = nil break } } - - r.ep.Close() } func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) { @@ -605,7 +1008,7 @@ func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) { } func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { - r, ok := n.endpoints[NetworkEndpointID{addr}] + r, ok := n.mu.endpoints[NetworkEndpointID{addr}] if !ok { return tcpip.ErrBadLocalAddress } @@ -615,26 +1018,45 @@ func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { return tcpip.ErrBadLocalAddress } - isIPv6Unicast := r.protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(addr) - - // If we are removing a tentative IPv6 unicast address, stop DAD. - if isIPv6Unicast && kind == permanentTentative { - n.ndp.stopDuplicateAddressDetection(addr) + switch r.protocol { + case header.IPv6ProtocolNumber: + return n.removePermanentIPv6EndpointLocked(r, true /* allowSLAACInvalidation */) + default: + r.expireLocked() + return nil } +} - r.setKind(permanentExpired) - if !r.decRefLocked() { - // The endpoint still has references to it. - return nil +func (n *NIC) removePermanentIPv6EndpointLocked(r *referencedNetworkEndpoint, allowSLAACInvalidation bool) *tcpip.Error { + addr := r.addrWithPrefix() + + isIPv6Unicast := header.IsV6UnicastAddress(addr.Address) + + if isIPv6Unicast { + n.mu.ndp.stopDuplicateAddressDetection(addr.Address) + + // If we are removing an address generated via SLAAC, cleanup + // its SLAAC resources and notify the integrator. + switch r.configType { + case slaac: + n.mu.ndp.cleanupSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) + case slaacTemp: + n.mu.ndp.cleanupTempSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) + } } + r.expireLocked() + // At this point the endpoint is deleted. // If we are removing an IPv6 unicast address, leave the solicited-node // multicast address. + // + // We ignore the tcpip.ErrBadLocalAddress error because the solicited-node + // multicast group may be left by user action. if isIPv6Unicast { - snmc := header.SolicitedNodeAddr(addr) - if err := n.leaveGroupLocked(snmc); err != nil { + snmc := header.SolicitedNodeAddr(addr.Address) + if err := n.leaveGroupLocked(snmc, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { return err } } @@ -668,23 +1090,23 @@ func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.A // outlined in RFC 3810 section 5. id := NetworkEndpointID{addr} - joins := n.mcastJoins[id] + joins := n.mu.mcastJoins[id] if joins == 0 { netProto, ok := n.stack.networkProtocols[protocol] if !ok { return tcpip.ErrUnknownProtocol } - if _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{ + if _, err := n.addAddressLocked(tcpip.ProtocolAddress{ Protocol: protocol, AddressWithPrefix: tcpip.AddressWithPrefix{ Address: addr, PrefixLen: netProto.DefaultPrefixLen(), }, - }, NeverPrimaryEndpoint); err != nil { + }, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { return err } } - n.mcastJoins[id] = joins + 1 + n.mu.mcastJoins[id] = joins + 1 return nil } @@ -694,48 +1116,75 @@ func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() - return n.leaveGroupLocked(addr) + return n.leaveGroupLocked(addr, false /* force */) } // leaveGroupLocked decrements the count for the given multicast address, and // when it reaches zero removes the endpoint for this address. n MUST be locked // before leaveGroupLocked is called. -func (n *NIC) leaveGroupLocked(addr tcpip.Address) *tcpip.Error { +// +// If force is true, then the count for the multicast addres is ignored and the +// endpoint will be removed immediately. +func (n *NIC) leaveGroupLocked(addr tcpip.Address, force bool) *tcpip.Error { id := NetworkEndpointID{addr} - joins := n.mcastJoins[id] - switch joins { - case 0: + joins, ok := n.mu.mcastJoins[id] + if !ok { // There are no joins with this address on this NIC. return tcpip.ErrBadLocalAddress - case 1: - // This is the last one, clean up. - if err := n.removePermanentAddressLocked(addr); err != nil { - return err - } } - n.mcastJoins[id] = joins - 1 + + joins-- + if force || joins == 0 { + // There are no outstanding joins or we are forced to leave, clean up. + delete(n.mu.mcastJoins, id) + return n.removePermanentAddressLocked(addr) + } + + n.mu.mcastJoins[id] = joins return nil } -func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, vv buffer.VectorisedView) { +// isInGroup returns true if n has joined the multicast group addr. +func (n *NIC) isInGroup(addr tcpip.Address) bool { + n.mu.RLock() + joins := n.mu.mcastJoins[NetworkEndpointID{addr}] + n.mu.RUnlock() + + return joins != 0 +} + +func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt *PacketBuffer) { r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) r.RemoteLinkAddress = remotelinkAddr - ref.ep.HandlePacket(&r, vv) + + ref.ep.HandlePacket(&r, pkt) ref.decRef() } // DeliverNetworkPacket finds the appropriate network protocol endpoint and // hands the packet over for further processing. This function is called when -// the NIC receives a packet from the physical interface. +// the NIC receives a packet from the link endpoint. // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. -func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { +func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + n.mu.RLock() + enabled := n.mu.enabled + // If the NIC is not yet enabled, don't receive any packets. + if !enabled { + n.mu.RUnlock() + + n.stats.DisabledRx.Packets.Increment() + n.stats.DisabledRx.Bytes.IncrementBy(uint64(pkt.Data.Size())) + return + } + n.stats.Rx.Packets.Increment() - n.stats.Rx.Bytes.IncrementBy(uint64(vv.Size())) + n.stats.Rx.Bytes.IncrementBy(uint64(pkt.Data.Size())) netProto, ok := n.stack.networkProtocols[protocol] if !ok { + n.mu.RUnlock() n.stack.stats.UnknownProtocolRcvdPackets.Increment() return } @@ -747,32 +1196,59 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link } // Are any packet sockets listening for this network protocol? - n.mu.RLock() - packetEPs := n.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.packetEPs[header.EthernetProtocolAll]...) - } + packetEPs := n.mu.packetEPs[protocol] + // 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, vv, linkHeader) + p := pkt.Clone() + p.PktType = tcpip.PacketHost + ep.HandlePacket(n.id, local, protocol, p) } if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { n.stack.stats.IP.PacketsReceived.Increment() } - if len(vv.First()) < netProto.MinimumPacketSize() { + // Parse headers. + transProtoNum, hasTransportHdr, ok := netProto.Parse(pkt) + if !ok { + // The packet is too small to contain a network header. n.stack.stats.MalformedRcvdPackets.Increment() return } + if hasTransportHdr { + // Parse the transport header if present. + if state, ok := n.stack.transportProtocols[transProtoNum]; ok { + state.proto.Parse(pkt) + } + } + + src, dst := netProto.ParseAddresses(pkt.NetworkHeader().View()) + + if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { + // The source address is one of our own, so we never should have gotten a + // packet like this unless handleLocal is false. Loopback also calls this + // function even though the packets didn't come from the physical interface + // so don't drop those. + n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() + return + } - src, dst := netProto.ParseAddresses(vv.First()) + // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet. + // Loopback traffic skips the prerouting chain. + if protocol == header.IPv4ProtocolNumber && !n.isLoopback() { + // iptables filtering. + ipt := n.stack.IPTables() + address := n.primaryAddress(protocol) + if ok := ipt.Check(Prerouting, pkt, nil, nil, address.Address, ""); !ok { + // iptables is telling us to drop the packet. + return + } + } if ref := n.getRef(protocol, dst); ref != nil { - handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv) + handlePacket(protocol, dst, src, n.linkEP.LinkAddress(), remote, ref, pkt) return } @@ -783,54 +1259,98 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link if n.stack.Forwarding(protocol) { r, err := n.stack.FindRoute(0, "", dst, protocol, false /* multicastLoop */) if err != nil { - n.stack.stats.IP.InvalidAddressesReceived.Increment() + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() return } - defer r.Release() - - r.LocalLinkAddress = n.linkEP.LinkAddress() - r.RemoteLinkAddress = remote // Found a NIC. n := r.ref.nic n.mu.RLock() - ref, ok := n.endpoints[NetworkEndpointID{dst}] - ok = ok && ref.isValidForOutgoing() && ref.tryIncRef() + ref, ok := n.mu.endpoints[NetworkEndpointID{dst}] + ok = ok && ref.isValidForOutgoingRLocked() && ref.tryIncRef() n.mu.RUnlock() if ok { + r.LocalLinkAddress = n.linkEP.LinkAddress() + r.RemoteLinkAddress = remote r.RemoteAddress = src // TODO(b/123449044): Update the source NIC as well. - ref.ep.HandlePacket(&r, vv) + ref.ep.HandlePacket(&r, pkt) ref.decRef() - } else { - // n doesn't have a destination endpoint. - // Send the packet out of n. - // If we want to send the packet to a link-layer, - // we have to reserve space for an Ethernet header. - hdr := buffer.NewPrependableFromView(vv.First(), int(n.linkEP.MaxHeaderLength())) - vv.RemoveFirst() - - // TODO(gvisor.dev/issue/1085): According to the RFC, we must decrease the TTL field for ipv4/ipv6. - // TODO(b/128629022): use route.WritePacket. - if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, vv, protocol); err != nil { - r.Stats().IP.OutgoingPacketErrors.Increment() - } else { - n.stats.Tx.Packets.Increment() - n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + vv.Size())) + r.Release() + return + } + + // n doesn't have a destination endpoint. + // Send the packet out of n. + // TODO(b/128629022): move this logic to route.WritePacket. + // TODO(gvisor.dev/issue/1085): According to the RFC, we must decrease the TTL field for ipv4/ipv6. + if ch, err := r.Resolve(nil); err != nil { + if err == tcpip.ErrWouldBlock { + n.stack.forwarder.enqueue(ch, n, &r, protocol, pkt) + // forwarder will release route. + return } + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() + r.Release() + return } + + // The link-address resolution finished immediately. + n.forwardPacket(&r, protocol, pkt) + r.Release() return } // If a packet socket handled the packet, don't treat it as invalid. if len(packetEPs) == 0 { - n.stack.stats.IP.InvalidAddressesReceived.Increment() + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() + } +} + +// 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. + + // pkt may have set its header and may not have enough headroom for link-layer + // header for the other link to prepend. Here we create a new packet to + // forward. + fwdPkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(n.linkEP.MaxHeaderLength()), + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + }) + + // WritePacket takes ownership of fwdPkt, calculate numBytes first. + numBytes := fwdPkt.Size() + + if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, fwdPkt); err != nil { + r.Stats().IP.OutgoingPacketErrors.Increment() + return + } + + n.stats.Tx.Packets.Increment() + n.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) +} + // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, netHeader buffer.View, vv buffer.VectorisedView) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) { state, ok := n.stack.transportProtocols[protocol] if !ok { n.stack.stats.UnknownProtocolRcvdPackets.Increment() @@ -842,41 +1362,60 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // Raw socket packets are delivered based solely on the transport // protocol number. We do not inspect the payload to ensure it's // validly formed. - n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) + n.stack.demux.deliverRawPacket(r, protocol, pkt) + + // TransportHeader is empty only when pkt is an ICMP packet or was reassembled + // from fragments. + if pkt.TransportHeader().View().IsEmpty() { + // TODO(gvisor.dev/issue/170): ICMP packets don't have their TransportHeader + // fields set yet, parse it here. See icmp/protocol.go:protocol.Parse for a + // full explanation. + if protocol == header.ICMPv4ProtocolNumber || protocol == header.ICMPv6ProtocolNumber { + // ICMP packets may be longer, but until icmp.Parse is implemented, here + // we parse it using the minimum size. + if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok { + n.stack.stats.MalformedRcvdPackets.Increment() + return + } + } else { + // This is either a bad packet or was re-assembled from fragments. + transProto.Parse(pkt) + } + } - if len(vv.First()) < transProto.MinimumPacketSize() { + if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(vv.First()) + srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} - if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) { + if n.stack.demux.deliverPacket(r, protocol, pkt, id) { return } // Try to deliver to per-stack default handler. if state.defaultHandler != nil { - if state.defaultHandler(r, id, netHeader, vv) { + if state.defaultHandler(r, id, pkt) { return } } // We could not find an appropriate destination for this packet, so // deliver it to the global handler. - if !transProto.HandleUnknownDestinationPacket(r, id, netHeader, vv) { + if !transProto.HandleUnknownDestinationPacket(r, id, pkt) { n.stack.stats.MalformedRcvdPackets.Increment() } } // DeliverTransportControlPacket delivers control packets to the appropriate // transport protocol endpoint. -func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView) { +func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer) { state, ok := n.stack.transportProtocols[trans] if !ok { return @@ -887,17 +1426,18 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - if len(vv.First()) < 8 { + transHeader, ok := pkt.Data.PullUp(8) + if !ok { return } - srcPort, dstPort, err := transProto.ParsePorts(vv.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { return } id := TransportEndpointID{srcPort, local, dstPort, remote} - if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, vv, id) { + if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, pkt, id) { return } } @@ -907,18 +1447,31 @@ func (n *NIC) ID() tcpip.NICID { return n.id } +// Name returns the name of n. +func (n *NIC) Name() string { + return n.name +} + // Stack returns the instance of the Stack that owns this NIC. func (n *NIC) Stack() *Stack { return n.stack } +// LinkEndpoint returns the link endpoint of n. +func (n *NIC) LinkEndpoint() LinkEndpoint { + return n.linkEP +} + // isAddrTentative returns true if addr is tentative on n. // // Note that if addr is not associated with n, then this function will return // false. It will only return true if the address is associated with the NIC // AND it is tentative. func (n *NIC) isAddrTentative(addr tcpip.Address) bool { - ref, ok := n.endpoints[NetworkEndpointID{addr}] + n.mu.RLock() + defer n.mu.RUnlock() + + ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] if !ok { return false } @@ -926,15 +1479,17 @@ func (n *NIC) isAddrTentative(addr tcpip.Address) bool { return ref.getKind() == permanentTentative } -// dupTentativeAddrDetected attempts to inform n that a tentative addr -// is a duplicate on a link. +// dupTentativeAddrDetected attempts to inform n that a tentative addr is a +// duplicate on a link. // -// dupTentativeAddrDetected will delete the tentative address if it exists. +// dupTentativeAddrDetected will remove the tentative address if it exists. If +// the address was generated via SLAAC, an attempt will be made to generate a +// new address. func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() - ref, ok := n.endpoints[NetworkEndpointID{addr}] + ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] if !ok { return tcpip.ErrBadAddress } @@ -943,7 +1498,24 @@ func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - return n.removePermanentAddressLocked(addr) + // If the address is a SLAAC address, do not invalidate its SLAAC prefix as a + // new address will be generated for it. + if err := n.removePermanentIPv6EndpointLocked(ref, false /* allowSLAACInvalidation */); err != nil { + return err + } + + prefix := ref.addrWithPrefix().Subnet() + + switch ref.configType { + case slaac: + n.mu.ndp.regenerateSLAACAddr(prefix) + case slaacTemp: + // Do not reset the generation attempts counter for the prefix as the + // temporary address is being regenerated in response to a DAD conflict. + n.mu.ndp.regenerateTempSLAACAddr(prefix, false /* resetGenAttempts */) + } + + return nil } // setNDPConfigs sets the NDP configurations for n. @@ -954,10 +1526,39 @@ func (n *NIC) setNDPConfigs(c NDPConfigurations) { c.validate() n.mu.Lock() - n.ndp.configs = c + n.mu.ndp.configs = c n.mu.Unlock() } +// NUDConfigs gets the NUD configurations for n. +func (n *NIC) NUDConfigs() (NUDConfigurations, *tcpip.Error) { + if n.neigh == nil { + return NUDConfigurations{}, tcpip.ErrNotSupported + } + return n.neigh.config(), nil +} + +// setNUDConfigs sets the NUD configurations for n. +// +// 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 { + if n.neigh == nil { + return tcpip.ErrNotSupported + } + c.resetInvalidFields() + n.neigh.setConfig(c) + return nil +} + +// handleNDPRA handles an NDP Router Advertisement message that arrived on n. +func (n *NIC) handleNDPRA(ip tcpip.Address, ra header.NDPRouterAdvert) { + n.mu.Lock() + defer n.mu.Unlock() + + n.mu.ndp.handleRA(ip, ra) +} + type networkEndpointKind int32 const ( @@ -977,7 +1578,7 @@ const ( // removing the permanent address from the NIC. permanent - // An expired permanent endoint is a permanent endoint that had its address + // An expired permanent endpoint is a permanent endpoint that had its address // removed from the NIC, and it is waiting to be removed once no more routes // hold a reference to it. This is achieved by decreasing its reference count // by 1. If its address is re-added before the endpoint is removed, its type @@ -997,11 +1598,11 @@ func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep Pa n.mu.Lock() defer n.mu.Unlock() - eps, ok := n.packetEPs[netProto] + eps, ok := n.mu.packetEPs[netProto] if !ok { return tcpip.ErrNotSupported } - n.packetEPs[netProto] = append(eps, ep) + n.mu.packetEPs[netProto] = append(eps, ep) return nil } @@ -1010,21 +1611,40 @@ func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep n.mu.Lock() defer n.mu.Unlock() - eps, ok := n.packetEPs[netProto] + eps, ok := n.mu.packetEPs[netProto] if !ok { return } for i, epOther := range eps { if epOther == ep { - n.packetEPs[netProto] = append(eps[:i], eps[i+1:]...) + n.mu.packetEPs[netProto] = append(eps[:i], eps[i+1:]...) return } } } +type networkEndpointConfigType int32 + +const ( + // A statically configured endpoint is an address that was added by + // some user-specified action (adding an explicit address, joining a + // multicast group). + static networkEndpointConfigType = iota + + // A SLAAC configured endpoint is an IPv6 endpoint that was added by + // SLAAC as per RFC 4862 section 5.5.3. + slaac + + // A temporary SLAAC configured endpoint is an IPv6 endpoint that was added by + // SLAAC as per RFC 4941. Temporary SLAAC addresses are short-lived and are + // not expected to be valid (or preferred) forever; hence the term temporary. + slaacTemp +) + type referencedNetworkEndpoint struct { ep NetworkEndpoint + addr tcpip.AddressWithPrefix nic *NIC protocol tcpip.NetworkProtocolNumber @@ -1038,6 +1658,24 @@ type referencedNetworkEndpoint struct { // networkEndpointKind must only be accessed using {get,set}Kind(). kind networkEndpointKind + + // configType is the method that was used to configure this endpoint. + // This must never change except during endpoint creation and promotion to + // permanent. + configType networkEndpointConfigType + + // deprecated indicates whether or not the endpoint should be considered + // deprecated. That is, when deprecated is true, other endpoints that are not + // deprecated should be preferred. + deprecated bool +} + +func (r *referencedNetworkEndpoint) address() tcpip.Address { + return r.addr.Address +} + +func (r *referencedNetworkEndpoint) addrWithPrefix() tcpip.AddressWithPrefix { + return r.addr } func (r *referencedNetworkEndpoint) getKind() networkEndpointKind { @@ -1049,17 +1687,44 @@ func (r *referencedNetworkEndpoint) setKind(kind networkEndpointKind) { } // isValidForOutgoing returns true if the endpoint can be used to send out a -// packet. It requires the endpoint to not be marked expired (i.e., its address -// has been removed), or the NIC to be in spoofing mode. +// packet. It requires the endpoint to not be marked expired (i.e., its address) +// has been removed) unless the NIC is in spoofing mode, or temporary. func (r *referencedNetworkEndpoint) isValidForOutgoing() bool { - return r.getKind() != permanentExpired || r.nic.spoofing + r.nic.mu.RLock() + defer r.nic.mu.RUnlock() + + return r.isValidForOutgoingRLocked() } -// isValidForIncoming returns true if the endpoint can accept an incoming -// packet. It requires the endpoint to not be marked expired (i.e., its address -// has been removed), or the NIC to be in promiscuous mode. -func (r *referencedNetworkEndpoint) isValidForIncoming() bool { - return r.getKind() != permanentExpired || r.nic.promiscuous +// isValidForOutgoingRLocked is the same as isValidForOutgoing but requires +// r.nic.mu to be read locked. +func (r *referencedNetworkEndpoint) isValidForOutgoingRLocked() bool { + if !r.nic.mu.enabled { + return false + } + + return r.isAssignedRLocked(r.nic.mu.spoofing) +} + +// isAssignedRLocked returns true if r is considered to be assigned to the NIC. +// +// r.nic.mu must be read locked. +func (r *referencedNetworkEndpoint) isAssignedRLocked(spoofingOrPromiscuous bool) bool { + switch r.getKind() { + case permanentTentative: + return false + case permanentExpired: + return spoofingOrPromiscuous + default: + return true + } +} + +// expireLocked decrements the reference count and marks the permanent endpoint +// as expired. +func (r *referencedNetworkEndpoint) expireLocked() { + r.setKind(permanentExpired) + r.decRefLocked() } // decRef decrements the ref count and cleans up the endpoint once it reaches @@ -1071,14 +1736,11 @@ func (r *referencedNetworkEndpoint) decRef() { } // decRefLocked is the same as decRef but assumes that the NIC.mu mutex is -// locked. Returns true if the endpoint was removed. -func (r *referencedNetworkEndpoint) decRefLocked() bool { +// locked. +func (r *referencedNetworkEndpoint) decRefLocked() { if atomic.AddInt32(&r.refs, -1) == 0 { r.nic.removeEndpointLocked(r) - return true } - - return false } // incRef increments the ref count. It must only be called when the caller is |