diff options
Diffstat (limited to 'pkg/tcpip/stack/nic.go')
| -rw-r--r-- | pkg/tcpip/stack/nic.go | 1379 |
1 files changed, 200 insertions, 1179 deletions
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 863ef6bee..6cf54cc89 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -18,7 +18,6 @@ import ( "fmt" "math/rand" "reflect" - "sort" "sync/atomic" "gvisor.dev/gvisor/pkg/sleep" @@ -28,13 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) -var ipv4BroadcastAddr = tcpip.ProtocolAddress{ - Protocol: header.IPv4ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: header.IPv4Broadcast, - PrefixLen: 8 * header.IPv4AddressSize, - }, -} +var _ NetworkInterface = (*NIC)(nil) // NIC represents a "network interface card" to which the networking stack is // attached. @@ -45,22 +38,25 @@ type NIC struct { linkEP LinkEndpoint context NICContext - stats NICStats - neigh *neighborCache + stats NICStats + neigh *neighborCache + + // The network endpoints themselves may be modified by calling the interface's + // methods, but the map reference and entries must be constant. networkEndpoints map[tcpip.NetworkProtocolNumber]NetworkEndpoint + // enabled is set to 1 when the NIC is enabled and 0 when it is disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + 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 } } @@ -84,25 +80,6 @@ type DirectionStats struct { Bytes *tcpip.StatCounter } -// PrimaryEndpointBehavior is an enumeration of an endpoint's primacy behavior. -type PrimaryEndpointBehavior int - -const ( - // CanBePrimaryEndpoint indicates the endpoint can be used as a primary - // endpoint for new connections with no local address. This is the - // default when calling NIC.AddAddress. - CanBePrimaryEndpoint PrimaryEndpointBehavior = iota - - // FirstPrimaryEndpoint indicates the endpoint should be the first - // primary endpoint considered. If there are multiple endpoints with - // this behavior, the most recently-added one will be first. - FirstPrimaryEndpoint - - // NeverPrimaryEndpoint indicates the endpoint should never be a - // primary endpoint. - NeverPrimaryEndpoint -) - // newNIC returns a new NIC using the default NDP configurations from stack. func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICContext) *NIC { // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For @@ -122,19 +99,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC 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() // Check for Neighbor Unreachability Detection support. var nud NUDHandler @@ -162,7 +127,7 @@ 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] = nil - nic.networkEndpoints[netNum] = netProto.NewEndpoint(id, stack, nud, nic, ep, stack) + nic.networkEndpoints[netNum] = netProto.NewEndpoint(nic, stack, nud, nic) } nic.linkEP.Attach(nic) @@ -170,29 +135,32 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC return nic } -// enabled returns true if n is enabled. -func (n *NIC) enabled() bool { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - return enabled +func (n *NIC) getNetworkEndpoint(proto tcpip.NetworkProtocolNumber) NetworkEndpoint { + return n.networkEndpoints[proto] } -// disable disables n. +// Enabled implements NetworkInterface. +func (n *NIC) Enabled() bool { + return atomic.LoadUint32(&n.enabled) == 1 +} + +// setEnabled sets the enabled status for the NIC. // -// 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 +// Returns true if the enabled status was updated. +func (n *NIC) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&n.enabled, 1) == 0 } + return atomic.SwapUint32(&n.enabled, 0) == 1 +} +// disable disables n. +// +// It undoes the work done by enable. +func (n *NIC) disable() { n.mu.Lock() - err := n.disableLocked() + n.disableLocked() n.mu.Unlock() - return err } // disableLocked disables n. @@ -200,9 +168,9 @@ func (n *NIC) disable() *tcpip.Error { // It undoes the work done by enable. // // n MUST be locked. -func (n *NIC) disableLocked() *tcpip.Error { - if !n.mu.enabled { - return nil +func (n *NIC) disableLocked() { + if !n.setEnabled(false) { + return } // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be @@ -210,38 +178,9 @@ func (n *NIC) disableLocked() *tcpip.Error { // 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 - } + for _, ep := range n.networkEndpoints { + ep.Disable() } - - n.mu.enabled = false - return nil } // enable enables n. @@ -251,162 +190,38 @@ func (n *NIC) disableLocked() *tcpip.Error { // 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.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - if enabled { - return nil - } - n.mu.Lock() defer n.mu.Unlock() - if n.mu.enabled { + if !n.setEnabled(true) { 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.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 - } - } - - // Join the IPv6 All-Nodes Multicast group if the stack is configured to - // use IPv6. This is required to ensure that this node properly receives - // and responds to the various NDP messages that are destined to the - // all-nodes multicast address. An example is the Neighbor Advertisement - // when we perform Duplicate Address Detection, or Router Advertisement - // when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861 - // section 4.2 for more information. - // - // Also auto-generate an IPv6 link-local address based on the NIC's - // link address if it is configured to do so. Note, each interface is - // required to have IPv6 link-local unicast address, as per RFC 4291 - // section 2.1. - _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber] - if !ok { - return nil - } - - // 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 - } - - // 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 { + for _, ep := range n.networkEndpoints { + if err := ep.Enable(); err != nil { return err } } - // 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) - } - - // If we are operating as a router, then do not solicit routers since we - // won't process the RAs anyways. - // - // 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 { - n.mu.ndp.startSolicitingRouters() - } - return nil } -// 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. +// 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 { 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 -} - -// 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() + return nil } // setPromiscuousMode enables or disables promiscuous mode. @@ -423,7 +238,8 @@ func (n *NIC) isPromiscuousMode() bool { return rv } -func (n *NIC) isLoopback() bool { +// IsLoopback implements NetworkInterface. +func (n *NIC) IsLoopback() bool { return n.linkEP.Capabilities()&CapabilityLoopback != 0 } @@ -434,206 +250,44 @@ func (n *NIC) setSpoofing(enable bool) { n.mu.Unlock() } -// 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 && len(remoteAddr) != 0 { - return n.primaryIPv6Endpoint(remoteAddr) - } - +// primaryAddress returns an address that can be used to communicate with +// remoteAddr. +func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) AssignableAddressEndpoint { n.mu.RLock() - defer n.mu.RUnlock() - - 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) + spoofing := n.mu.spoofing 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 - } - } - - 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}] + ep, ok := n.networkEndpoints[protocol] if !ok { - return false + return nil } - kind := ref.getKind() - - return kind == permanent || kind == permanentTentative + return ep.AcquireOutgoingPrimaryAddress(remoteAddr, spoofing) } -type getRefBehaviour int +type getAddressBehaviour int const ( // spoofing indicates that the NIC's spoofing flag should be observed when - // getting a NIC's referenced network endpoint. - spoofing getRefBehaviour = iota + // getting a NIC's address endpoint. + spoofing getAddressBehaviour = iota // promiscuous indicates that the NIC's promiscuous flag should be observed - // when getting a NIC's referenced network endpoint. + // when getting a NIC's address endpoint. promiscuous ) -func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous) +func (n *NIC) getAddress(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) AssignableAddressEndpoint { + return n.getAddressOrCreateTemp(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, spoofing) +func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) AssignableAddressEndpoint { + return n.getAddressOrCreateTemp(protocol, address, peb, spoofing) } -// getRefEpOrCreateTemp returns the referenced network endpoint for the given -// protocol and address. +// getAddressEpOrCreateTemp returns the address endpoint for the given 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. @@ -641,9 +295,8 @@ func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.A // // 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 { +func (n *NIC) getAddressOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getAddressBehaviour) AssignableAddressEndpoint { n.mu.RLock() - var spoofingOrPromiscuous bool switch tempRef { case spoofing: @@ -651,294 +304,54 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t 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. - if !ref.isAssignedRLocked(spoofingOrPromiscuous) { - n.mu.RUnlock() - return nil - } - - if ref.tryIncRef() { - n.mu.RUnlock() - return ref - } - } - - // 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 IPv4 address is found in the NIC's subnets and the NIC - // is a loopback interface. - createTempEP := spoofingOrPromiscuous - if !createTempEP && n.isLoopback() && protocol == header.IPv4ProtocolNumber { - for _, r := range n.mu.endpoints { - addr := r.addrWithPrefix() - subnet := addr.Subnet() - if subnet.Contains(address) { - createTempEP = true - break - } - } - } n.mu.RUnlock() - - if !createTempEP { - return nil - } - - // Try again with the lock in exclusive mode. If we still can't get the - // endpoint, create a new "temporary" endpoint. It will only exist while - // there's a route through it. - n.mu.Lock() - ref := n.getRefOrCreateTempLocked(protocol, address, peb) - n.mu.Unlock() - return ref + return n.getAddressOrCreateTempInner(protocol, address, spoofingOrPromiscuous, peb) } -// 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 - } +// getAddressOrCreateTempInner is like getAddressEpOrCreateTemp except a boolean +// is passed to indicate whether or not we should generate temporary endpoints. +func (n *NIC) getAddressOrCreateTempInner(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, createTemp bool, peb PrimaryEndpointBehavior) AssignableAddressEndpoint { + if ep, ok := n.networkEndpoints[protocol]; ok { + return ep.AcquireAssignedAddress(address, createTemp, peb) } 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() { - return ref - } - // tryIncRef failing means the endpoint is scheduled to be removed once the - // lock is released. Remove it here so we can create a new (temporary) one. - // The removal logic waiting for the lock handles this case. - 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 { - return nil - } - ref, _ := n.addAddressLocked(tcpip.ProtocolAddress{ - Protocol: protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: address, - PrefixLen: netProto.DefaultPrefixLen(), - }, - }, peb, temporary, static, false) - return ref -} - -// 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, - // 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.mu.primary[ref.protocol] - for i, r := range refs { - if r == ref { - switch peb { - case CanBePrimaryEndpoint: - return ref, nil - case FirstPrimaryEndpoint: - if i == 0 { - return ref, nil - } - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - case NeverPrimaryEndpoint: - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - return ref, nil - } - } - } - - n.insertPrimaryEndpointLocked(ref, peb) - - return ref, nil - } - // tryIncRef failing means the endpoint is scheduled to be removed once - // the lock is released. Remove it here so we can create a new - // (permanent) one. The removal logic waiting for the lock handles this - // case. - n.removeEndpointLocked(ref) - } - } - +// 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 { ep, ok := n.networkEndpoints[protocolAddress.Protocol] if !ok { - return nil, tcpip.ErrUnknownProtocol - } - - 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 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, - addr: protocolAddress.AddressWithPrefix, - ep: ep, - nic: n, - protocol: protocolAddress.Protocol, - kind: kind, - configType: configType, - deprecated: deprecated, + return tcpip.ErrUnknownProtocol } - // Set up resolver if link address resolution exists for this protocol. - if n.linkEP.Capabilities()&CapabilityResolutionRequired != 0 { - if linkRes, ok := n.stack.linkAddrResolvers[protocolAddress.Protocol]; ok { - ref.linkCache = n.stack - ref.linkRes = linkRes - } - } - - // If we are adding an IPv6 unicast address, join the solicited-node - // multicast address. - if isIPv6Unicast { - snmc := header.SolicitedNodeAddr(protocolAddress.AddressWithPrefix.Address) - if err := n.joinGroupLocked(protocolAddress.Protocol, snmc); err != nil { - return nil, err - } - } - - n.mu.endpoints[id] = ref - - n.insertPrimaryEndpointLocked(ref, peb) - - // 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 - } + addressEndpoint, err := ep.AddAndAcquirePermanentAddress(protocolAddress.AddressWithPrefix, peb, AddressConfigStatic, false /* deprecated */) + if err == nil { + // We have no need for the address endpoint. + addressEndpoint.DecRef() } - - return ref, nil -} - -// 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 { - // Add the endpoint. - n.mu.Lock() - _, err := n.addAddressLocked(protocolAddress, peb, permanent, static, false /* deprecated */) - n.mu.Unlock() - return err } -// AllAddresses returns all addresses (primary and non-primary) associated with +// allPermanentAddresses returns all permanent addresses associated with // this NIC. -func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { - n.mu.RLock() - defer n.mu.RUnlock() - - 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 permanentExpired, temporary: - continue +func (n *NIC) allPermanentAddresses() []tcpip.ProtocolAddress { + var addrs []tcpip.ProtocolAddress + for p, ep := range n.networkEndpoints { + for _, a := range ep.PermanentAddresses() { + addrs = append(addrs, tcpip.ProtocolAddress{Protocol: p, AddressWithPrefix: a}) } - - addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: ref.protocol, - AddressWithPrefix: ref.addrWithPrefix(), - }) } return addrs } -// PrimaryAddresses returns the primary addresses associated with this NIC. -func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { - n.mu.RLock() - defer n.mu.RUnlock() - +// primaryAddresses returns the primary addresses associated with this NIC. +func (n *NIC) primaryAddresses() []tcpip.ProtocolAddress { var addrs []tcpip.ProtocolAddress - 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 - // those. - switch ref.getKind() { - case permanentTentative, permanentExpired, temporary: - continue - } - - addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: proto, - AddressWithPrefix: ref.addrWithPrefix(), - }) + for p, ep := range n.networkEndpoints { + for _, a := range ep.PrimaryAddresses() { + addrs = append(addrs, tcpip.ProtocolAddress{Protocol: p, AddressWithPrefix: a}) } } return addrs @@ -950,147 +363,25 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { // 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() - - list, ok := n.mu.primary[proto] + ep, ok := n.networkEndpoints[proto] if !ok { return tcpip.AddressWithPrefix{} } - 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 - } - - if !ref.deprecated { - return ref.addrWithPrefix() - } - - if deprecatedEndpoint == nil { - deprecatedEndpoint = ref - } - } - - if deprecatedEndpoint != nil { - return deprecatedEndpoint.addrWithPrefix() - } - - return tcpip.AddressWithPrefix{} + return ep.MainAddress() } -// insertPrimaryEndpointLocked adds r to n's primary endpoint list as required -// by peb. -// -// n MUST be locked. -func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) { - switch peb { - case CanBePrimaryEndpoint: - n.mu.primary[r.protocol] = append(n.mu.primary[r.protocol], r) - case FirstPrimaryEndpoint: - n.mu.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.mu.primary[r.protocol]...) - } -} - -func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { - 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.mu.endpoints[id] != r { - return - } - - if r.getKind() == permanent { - panic("Reference count dropped to zero before being removed") - } - - delete(n.mu.endpoints, id) - refs := n.mu.primary[r.protocol] - for i, ref := range refs { - if ref == r { - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - refs[len(refs)-1] = nil - break - } - } -} - -func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) { - n.mu.Lock() - n.removeEndpointLocked(r) - n.mu.Unlock() -} - -func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { - r, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return tcpip.ErrBadLocalAddress - } - - kind := r.getKind() - if kind != permanent && kind != permanentTentative { - return tcpip.ErrBadLocalAddress - } - - switch r.protocol { - case header.IPv6ProtocolNumber: - return n.removePermanentIPv6EndpointLocked(r, true /* allowSLAACInvalidation */) - default: - r.expireLocked() - 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.Address) - if err := n.leaveGroupLocked(snmc, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { +// removeAddress removes an address from n. +func (n *NIC) removeAddress(addr tcpip.Address) *tcpip.Error { + for _, ep := range n.networkEndpoints { + if err := ep.RemovePermanentAddress(addr); err == tcpip.ErrBadLocalAddress { + continue + } else { return err } } - return nil -} - -// RemoveAddress removes an address from n. -func (n *NIC) RemoveAddress(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - return n.removePermanentAddressLocked(addr) + return tcpip.ErrBadLocalAddress } func (n *NIC) neighbors() ([]NeighborEntry, *tcpip.Error) { @@ -1141,91 +432,66 @@ 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 { - n.mu.Lock() - defer n.mu.Unlock() - - return n.joinGroupLocked(protocol, addr) -} - -// joinGroupLocked adds a new endpoint for the given multicast address, if none -// exists yet. Otherwise it just increments its count. n MUST be locked before -// joinGroupLocked is called. -func (n *NIC) joinGroupLocked(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 // outlined in RFC 3810 section 5. - id := NetworkEndpointID{addr} - joins := n.mu.mcastJoins[id] - if joins == 0 { - netProto, ok := n.stack.networkProtocols[protocol] - if !ok { - return tcpip.ErrUnknownProtocol - } - if _, err := n.addAddressLocked(tcpip.ProtocolAddress{ - Protocol: protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: addr, - PrefixLen: netProto.DefaultPrefixLen(), - }, - }, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { - return err - } + ep, ok := n.networkEndpoints[protocol] + if !ok { + return tcpip.ErrNotSupported } - n.mu.mcastJoins[id] = joins + 1 - return nil + + gep, ok := ep.(GroupAddressableEndpoint) + if !ok { + return tcpip.ErrNotSupported + } + + _, err := gep.JoinGroup(addr) + return err } // leaveGroup decrements the count for the given multicast address, and when it // reaches zero removes the endpoint for this address. -func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - return n.leaveGroupLocked(addr, false /* force */) -} +func (n *NIC) leaveGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { + ep, ok := n.networkEndpoints[protocol] + if !ok { + return tcpip.ErrNotSupported + } -// 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. -// -// 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, ok := n.mu.mcastJoins[id] + gep, ok := ep.(GroupAddressableEndpoint) if !ok { - // There are no joins with this address on this NIC. - return tcpip.ErrBadLocalAddress + return tcpip.ErrNotSupported } - 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) + if _, err := gep.LeaveGroup(addr); err != nil { + return err } - n.mu.mcastJoins[id] = joins return nil } // 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() + for _, ep := range n.networkEndpoints { + gep, ok := ep.(GroupAddressableEndpoint) + if !ok { + continue + } + + if gep.IsInGroup(addr) { + return true + } + } - return joins != 0 + return false } -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 */) +func (n *NIC) handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, remotelinkAddr tcpip.LinkAddress, addressEndpoint AssignableAddressEndpoint, pkt *PacketBuffer) { + r := makeRoute(protocol, dst, src, n, addressEndpoint, false /* handleLocal */, false /* multicastLoop */) + defer r.Release() r.RemoteLinkAddress = remotelinkAddr - - ref.ep.HandlePacket(&r, pkt) - ref.decRef() + n.getNetworkEndpoint(protocol).HandlePacket(&r, pkt) } // DeliverNetworkPacket finds the appropriate network protocol endpoint and @@ -1236,7 +502,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // the ownership of the items is not retained by the caller. func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { n.mu.RLock() - enabled := n.mu.enabled + enabled := n.Enabled() // If the NIC is not yet enabled, don't receive any packets. if !enabled { n.mu.RUnlock() @@ -1262,9 +528,9 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp local = n.linkEP.LinkAddress() } - // Are any packet sockets listening for this network protocol? + // Are any packet type sockets listening for this network protocol? packetEPs := n.mu.packetEPs[protocol] - // Add any other packet sockets that maybe listening for all protocols. + // Add any other packet type sockets that may be listening for all protocols. packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) n.mu.RUnlock() for _, ep := range packetEPs { @@ -1285,6 +551,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp return } if hasTransportHdr { + pkt.TransportProtocolNumber = transProtoNum // Parse the transport header if present. if state, ok := n.stack.transportProtocols[transProtoNum]; ok { state.proto.Parse(pkt) @@ -1293,29 +560,33 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp 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 + if n.stack.handleLocal && !n.IsLoopback() { + if r := n.getAddress(protocol, src); r != nil { + r.DecRef() + + // 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 + } } - // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet. // Loopback traffic skips the prerouting chain. - if protocol == header.IPv4ProtocolNumber && !n.isLoopback() { + if !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. + n.stack.stats.IP.IPTablesPreroutingDropped.Increment() return } } - if ref := n.getRef(protocol, dst); ref != nil { - handlePacket(protocol, dst, src, n.linkEP.LinkAddress(), remote, ref, pkt) + if addressEndpoint := n.getAddress(protocol, dst); addressEndpoint != nil { + n.handlePacket(protocol, dst, src, remote, addressEndpoint, pkt) return } @@ -1323,7 +594,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp // packet and forward it to the NIC. // // TODO: Should we be forwarding the packet even if promiscuous? - if n.stack.Forwarding() { + if n.stack.Forwarding(protocol) { r, err := n.stack.FindRoute(0, "", dst, protocol, false /* multicastLoop */) if err != nil { n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() @@ -1331,25 +602,26 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } // Found a NIC. - n := r.ref.nic - n.mu.RLock() - 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, pkt) - ref.decRef() - r.Release() - return + n := r.nic + if addressEndpoint := n.getAddressOrCreateTempInner(protocol, dst, false, NeverPrimaryEndpoint); addressEndpoint != nil { + if n.isValidForOutgoing(addressEndpoint) { + r.LocalLinkAddress = n.linkEP.LinkAddress() + r.RemoteLinkAddress = remote + r.RemoteAddress = src + // TODO(b/123449044): Update the source NIC as well. + n.getNetworkEndpoint(protocol).HandlePacket(&r, pkt) + addressEndpoint.DecRef() + r.Release() + return + } + + addressEndpoint.DecRef() } // 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) @@ -1416,11 +688,11 @@ func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) TransportPacketDisposition { state, ok := n.stack.transportProtocols[protocol] if !ok { n.stack.stats.UnknownProtocolRcvdPackets.Increment() - return + return TransportPacketProtocolUnreachable } transProto := state.proto @@ -1441,41 +713,47 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // we parse it using the minimum size. if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok { n.stack.stats.MalformedRcvdPackets.Increment() - return + // We consider a malformed transport packet handled because there is + // nothing the caller can do. + return TransportPacketHandled } - } else { - // This is either a bad packet or was re-assembled from fragments. - transProto.Parse(pkt) + } else if !transProto.Parse(pkt) { + n.stack.stats.MalformedRcvdPackets.Increment() + return TransportPacketHandled } } - if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() { - n.stack.stats.MalformedRcvdPackets.Increment() - return - } - srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() - return + return TransportPacketHandled } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} if n.stack.demux.deliverPacket(r, protocol, pkt, id) { - return + return TransportPacketHandled } // Try to deliver to per-stack default handler. if state.defaultHandler != nil { if state.defaultHandler(r, id, pkt) { - return + return TransportPacketHandled } } - // We could not find an appropriate destination for this packet, so - // deliver it to the global handler. - if !transProto.HandleUnknownDestinationPacket(r, id, pkt) { + // We could not find an appropriate destination for this packet so + // give the protocol specific error handler a chance to handle it. + // If it doesn't handle it then we should do so. + switch res := transProto.HandleUnknownDestinationPacket(r, id, pkt); res { + case UnknownDestinationPacketMalformed: n.stack.stats.MalformedRcvdPackets.Increment() + return TransportPacketHandled + case UnknownDestinationPacketUnhandled: + return TransportPacketDestinationPortUnreachable + case UnknownDestinationPacketHandled: + return TransportPacketHandled + default: + panic(fmt.Sprintf("unrecognized result from HandleUnknownDestinationPacket = %d", res)) } } @@ -1508,96 +786,23 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp } } -// ID returns the identifier of n. +// ID implements NetworkInterface. func (n *NIC) ID() tcpip.NICID { return n.id } -// Name returns the name of n. +// Name implements NetworkInterface. 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. +// LinkEndpoint implements NetworkInterface. 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 { - n.mu.RLock() - defer n.mu.RUnlock() - - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return false - } - - return ref.getKind() == permanentTentative -} - -// dupTentativeAddrDetected attempts to inform n that a tentative addr is a -// duplicate on a link. -// -// 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.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return tcpip.ErrBadAddress - } - - if ref.getKind() != permanentTentative { - return tcpip.ErrInvalidEndpointState - } - - // 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. -// -// Note, if c contains invalid NDP configuration values, it will be fixed to -// use default values for the erroneous values. -func (n *NIC) setNDPConfigs(c NDPConfigurations) { - c.validate() - - n.mu.Lock() - n.mu.ndp.configs = c - n.mu.Unlock() -} - -// NUDConfigs gets the NUD configurations for n. -func (n *NIC) NUDConfigs() (NUDConfigurations, *tcpip.Error) { +// nudConfigs gets the NUD configurations for n. +func (n *NIC) nudConfigs() (NUDConfigurations, *tcpip.Error) { if n.neigh == nil { return NUDConfigurations{}, tcpip.ErrNotSupported } @@ -1617,49 +822,6 @@ func (n *NIC) setNUDConfigs(c NUDConfigurations) *tcpip.Error { 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 ( - // A permanentTentative endpoint is a permanent address that is not yet - // considered to be fully bound to an interface in the traditional - // sense. That is, the address is associated with a NIC, but packets - // destined to the address MUST NOT be accepted and MUST be silently - // dropped, and the address MUST NOT be used as a source address for - // outgoing packets. For IPv6, addresses will be of this kind until - // NDP's Duplicate Address Detection has resolved, or be deleted if - // the process results in detecting a duplicate address. - permanentTentative networkEndpointKind = iota - - // A permanent endpoint is created by adding a permanent address (vs. a - // temporary one) to the NIC. Its reference count is biased by 1 to avoid - // removal when no route holds a reference to it. It is removed by explicitly - // removing the permanent address from the NIC. - permanent - - // 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 - // changes back to permanent and its reference count increases by 1 again. - permanentExpired - - // A temporary endpoint is created for spoofing outgoing packets, or when in - // promiscuous mode and accepting incoming packets that don't match any - // permanent endpoint. Its reference count is not biased by 1 and the - // endpoint is removed immediately when no more route holds a reference to - // it. A temporary endpoint can be promoted to permanent if its address - // is added permanently. - temporary -) - func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() @@ -1690,153 +852,12 @@ func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep } } -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 - - // linkCache is set if link address resolution is enabled for this - // protocol. Set to nil otherwise. - linkCache LinkAddressCache - - // linkRes is set if link address resolution is enabled for this protocol. - // Set to nil otherwise. - linkRes LinkAddressResolver - - // refs is counting references held for this endpoint. When refs hits zero it - // triggers the automatic removal of the endpoint from the NIC. - refs int32 - - // 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 { - return networkEndpointKind(atomic.LoadInt32((*int32)(&r.kind))) -} - -func (r *referencedNetworkEndpoint) setKind(kind networkEndpointKind) { - atomic.StoreInt32((*int32)(&r.kind), int32(kind)) -} - // 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) unless the NIC is in spoofing mode, or temporary. -func (r *referencedNetworkEndpoint) isValidForOutgoing() bool { - r.nic.mu.RLock() - defer r.nic.mu.RUnlock() - - return r.isValidForOutgoingRLocked() -} - -// 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 -// zero. -func (r *referencedNetworkEndpoint) decRef() { - if atomic.AddInt32(&r.refs, -1) == 0 { - r.nic.removeEndpoint(r) - } -} - -// decRefLocked is the same as decRef but assumes that the NIC.mu mutex is -// locked. -func (r *referencedNetworkEndpoint) decRefLocked() { - if atomic.AddInt32(&r.refs, -1) == 0 { - r.nic.removeEndpointLocked(r) - } -} - -// incRef increments the ref count. It must only be called when the caller is -// known to be holding a reference to the endpoint, otherwise tryIncRef should -// be used. -func (r *referencedNetworkEndpoint) incRef() { - atomic.AddInt32(&r.refs, 1) -} - -// tryIncRef attempts to increment the ref count from n to n+1, but only if n is -// not zero. That is, it will increment the count if the endpoint is still -// alive, and do nothing if it has already been clean up. -func (r *referencedNetworkEndpoint) tryIncRef() bool { - for { - v := atomic.LoadInt32(&r.refs) - if v == 0 { - return false - } - - if atomic.CompareAndSwapInt32(&r.refs, v, v+1) { - return true - } - } -} - -// stack returns the Stack instance that owns the underlying endpoint. -func (r *referencedNetworkEndpoint) stack() *Stack { - return r.nic.stack +func (n *NIC) isValidForOutgoing(ep AssignableAddressEndpoint) bool { + n.mu.RLock() + spoofing := n.mu.spoofing + n.mu.RUnlock() + return n.Enabled() && ep.IsAssigned(spoofing) } |