summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/stack/nic.go
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/stack/nic.go')
-rw-r--r--pkg/tcpip/stack/nic.go596
1 files changed, 461 insertions, 135 deletions
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 43719085e..e8401c673 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -19,7 +19,6 @@ import (
"sync"
"sync/atomic"
- "gvisor.dev/gvisor/pkg/ilist"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
@@ -34,17 +33,24 @@ type NIC struct {
linkEP LinkEndpoint
loopback bool
- demux *transportDemuxer
-
mu sync.RWMutex
spoofing bool
promiscuous bool
- primary map[tcpip.NetworkProtocolNumber]*ilist.List
+ 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
}
// NICStats includes transmitted and received stats.
@@ -78,17 +84,26 @@ const (
NeverPrimaryEndpoint
)
+// 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 {
- return &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,
- demux: newTransportDemuxer(stack),
- primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List),
+ 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{},
@@ -99,7 +114,93 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback
Bytes: &tcpip.StatCounter{},
},
},
+ ndp: ndpState{
+ configs: stack.ndpConfigs,
+ dad: make(map[tcpip.Address]dadState),
+ defaultRouters: make(map[tcpip.Address]defaultRouterState),
+ onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState),
+ autoGenAddresses: make(map[tcpip.Address]autoGenAddressState),
+ },
+ }
+ nic.ndp.nic = nic
+
+ // Register supported packet endpoint protocols.
+ for _, netProto := range header.Ethertypes {
+ nic.packetEPs[netProto] = []PacketEndpoint{}
+ }
+ for _, netProto := range stack.networkProtocols {
+ nic.packetEPs[netProto.Number()] = []PacketEndpoint{}
+ }
+
+ return nic
+}
+
+// enable enables the NIC. enable will attach the link to its LinkEndpoint and
+// join the IPv6 All-Nodes Multicast address (ff02::1).
+func (n *NIC) enable() *tcpip.Error {
+ n.attachLinkEndpoint()
+
+ // 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 {
+ 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
+ }
+
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil {
+ return err
+ }
+
+ if !n.stack.autoGenIPv6LinkLocal {
+ return nil
}
+
+ l2addr := n.linkEP.LinkAddress()
+
+ // Only attempt to generate the link-local address if we have a
+ // valid MAC address.
+ //
+ // TODO(b/141011931): Validate a LinkEndpoint's link address
+ // (provided by LinkEndpoint.LinkAddress) before reaching this
+ // point.
+ if !header.IsValidUnicastEthernetAddress(l2addr) {
+ return nil
+ }
+
+ addr := header.LinkLocalAddr(l2addr)
+
+ _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{
+ Protocol: header.IPv6ProtocolNumber,
+ AddressWithPrefix: tcpip.AddressWithPrefix{
+ Address: addr,
+ PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen,
+ },
+ }, CanBePrimaryEndpoint)
+
+ return err
}
// attachLinkEndpoint attaches the NIC to the endpoint, which will enable it
@@ -129,55 +230,13 @@ func (n *NIC) setSpoofing(enable bool) {
n.mu.Unlock()
}
-func (n *NIC) getMainNICAddress(protocol tcpip.NetworkProtocolNumber) (tcpip.AddressWithPrefix, *tcpip.Error) {
- n.mu.RLock()
- defer n.mu.RUnlock()
-
- var r *referencedNetworkEndpoint
-
- // Check for a primary endpoint.
- if list, ok := n.primary[protocol]; ok {
- for e := list.Front(); e != nil; e = e.Next() {
- ref := e.(*referencedNetworkEndpoint)
- if ref.getKind() == permanent && ref.tryIncRef() {
- r = ref
- break
- }
- }
-
- }
-
- if r == nil {
- return tcpip.AddressWithPrefix{}, tcpip.ErrNoLinkAddress
- }
-
- addressWithPrefix := tcpip.AddressWithPrefix{
- Address: r.ep.ID().LocalAddress,
- PrefixLen: r.ep.PrefixLen(),
- }
- r.decRef()
-
- return addressWithPrefix, nil
-}
-
// primaryEndpoint returns the primary endpoint of n for the given network
// protocol.
func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedNetworkEndpoint {
n.mu.RLock()
defer n.mu.RUnlock()
- list := n.primary[protocol]
- if list == nil {
- return nil
- }
-
- for e := list.Front(); e != nil; e = e.Next() {
- r := e.(*referencedNetworkEndpoint)
- // TODO(crawshaw): allow broadcast address when SO_BROADCAST is set.
- switch r.ep.ID().LocalAddress {
- case header.IPv4Broadcast, header.IPv4Any:
- continue
- }
+ for _, r := range n.primary[protocol] {
if r.isValidForOutgoing() && r.tryIncRef() {
return r
}
@@ -186,6 +245,20 @@ 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.endpoints[NetworkEndpointID{addr}]
+
+ if !ok {
+ return false
+ }
+
+ kind := ref.getKind()
+
+ return kind == permanent || kind == permanentTentative
+}
+
func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint {
return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, n.promiscuous)
}
@@ -277,7 +350,7 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t
Address: address,
PrefixLen: netProto.DefaultPrefixLen(),
},
- }, peb, temporary)
+ }, peb, temporary, static)
n.mu.Unlock()
return ref
@@ -287,13 +360,35 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p
id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address}
if ref, ok := n.endpoints[id]; ok {
switch ref.getKind() {
- case permanent:
+ 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.
+ // Promote the endpoint to become permanent and respect
+ // the new peb.
if ref.tryIncRef() {
ref.setKind(permanent)
+
+ refs := n.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.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ case NeverPrimaryEndpoint:
+ n.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
@@ -303,10 +398,13 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p
n.removeEndpointLocked(ref)
}
}
- return n.addAddressLocked(protocolAddress, peb, permanent)
+
+ return n.addAddressLocked(protocolAddress, peb, permanent, static)
}
-func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind) (*referencedNetworkEndpoint, *tcpip.Error) {
+func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType) (*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 {
@@ -324,12 +422,22 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar
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.
+ if isIPv6Unicast && kind == permanent {
+ kind = permanentTentative
+ }
+
ref := &referencedNetworkEndpoint{
- refs: 1,
- ep: ep,
- nic: n,
- protocol: protocolAddress.Protocol,
- kind: kind,
+ refs: 1,
+ ep: ep,
+ nic: n,
+ protocol: protocolAddress.Protocol,
+ kind: kind,
+ configType: configType,
}
// Set up cache if link address resolution exists for this protocol.
@@ -339,19 +447,24 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar
}
}
+ // 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.endpoints[id] = ref
- l, ok := n.primary[protocolAddress.Protocol]
- if !ok {
- l = &ilist.List{}
- n.primary[protocolAddress.Protocol] = l
- }
+ n.insertPrimaryEndpointLocked(ref, peb)
- switch peb {
- case CanBePrimaryEndpoint:
- l.PushBack(ref)
- case FirstPrimaryEndpoint:
- l.PushFront(ref)
+ // 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 {
+ return nil, err
+ }
}
return ref, nil
@@ -368,16 +481,20 @@ func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpo
return err
}
-// Addresses returns the addresses associated with this NIC.
-func (n *NIC) Addresses() []tcpip.ProtocolAddress {
+// AllAddresses returns all addresses (primary and non-primary) associated with
+// this NIC.
+func (n *NIC) AllAddresses() []tcpip.ProtocolAddress {
n.mu.RLock()
defer n.mu.RUnlock()
+
addrs := make([]tcpip.ProtocolAddress, 0, len(n.endpoints))
for nid, ref := range n.endpoints {
- // Don't include expired or tempory endpoints to avoid confusion and
- // prevent the caller from using those.
+ // Don't include tentative, expired or temporary endpoints to
+ // avoid confusion and prevent the caller from using those.
switch ref.getKind() {
- case permanentExpired, temporary:
+ case permanentTentative, permanentExpired, temporary:
+ // TODO(b/140898488): Should tentative addresses be
+ // returned?
continue
}
addrs = append(addrs, tcpip.ProtocolAddress{
@@ -391,6 +508,34 @@ func (n *NIC) Addresses() []tcpip.ProtocolAddress {
return addrs
}
+// PrimaryAddresses returns the primary addresses associated with this NIC.
+func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress {
+ n.mu.RLock()
+ defer n.mu.RUnlock()
+
+ var addrs []tcpip.ProtocolAddress
+ for proto, list := range n.primary {
+ for _, 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: tcpip.AddressWithPrefix{
+ Address: ref.ep.ID().LocalAddress,
+ PrefixLen: ref.ep.PrefixLen(),
+ },
+ })
+ }
+ }
+ return addrs
+}
+
// AddAddressRange adds a range of addresses to n, so that it starts accepting
// packets targeted at the given addresses and network protocol. The range is
// given by a subnet address, and all addresses contained in the subnet are
@@ -435,6 +580,19 @@ func (n *NIC) AddressRanges() []tcpip.Subnet {
return append(sns, n.addressRanges...)
}
+// 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.primary[r.protocol] = append(n.primary[r.protocol], r)
+ case FirstPrimaryEndpoint:
+ n.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.primary[r.protocol]...)
+ }
+}
+
func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) {
id := *r.ep.ID()
@@ -452,9 +610,12 @@ func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) {
}
delete(n.endpoints, id)
- wasInList := r.Next() != nil || r.Prev() != nil || r == n.primary[r.protocol].Front()
- if wasInList {
- n.primary[r.protocol].Remove(r)
+ refs := n.primary[r.protocol]
+ for i, ref := range refs {
+ if ref == r {
+ n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
+ break
+ }
}
r.ep.Close()
@@ -467,13 +628,48 @@ func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) {
}
func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error {
- r := n.endpoints[NetworkEndpointID{addr}]
- if r == nil || r.getKind() != permanent {
+ r, ok := n.endpoints[NetworkEndpointID{addr}]
+ if !ok {
+ return tcpip.ErrBadLocalAddress
+ }
+
+ kind := r.getKind()
+ if kind != permanent && kind != permanentTentative {
return tcpip.ErrBadLocalAddress
}
+ isIPv6Unicast := r.protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(addr)
+
+ if isIPv6Unicast {
+ // If we are removing a tentative IPv6 unicast address, stop
+ // DAD.
+ if kind == permanentTentative {
+ n.ndp.stopDuplicateAddressDetection(addr)
+ }
+
+ // If we are removing an address generated via SLAAC, cleanup
+ // its SLAAC resources and notify the integrator.
+ if r.configType == slaac {
+ n.ndp.cleanupAutoGenAddrResourcesAndNotify(addr)
+ }
+ }
+
r.setKind(permanentExpired)
- r.decRefLocked()
+ if !r.decRefLocked() {
+ // The endpoint still has references to it.
+ return nil
+ }
+
+ // At this point the endpoint is deleted.
+
+ // If we are removing an IPv6 unicast address, leave the solicited-node
+ // multicast address.
+ if isIPv6Unicast {
+ snmc := header.SolicitedNodeAddr(addr)
+ if err := n.leaveGroupLocked(snmc); err != nil {
+ return err
+ }
+ }
return nil
}
@@ -491,6 +687,18 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address
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.mcastJoins[id]
if joins == 0 {
@@ -518,6 +726,13 @@ func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error {
n.mu.Lock()
defer n.mu.Unlock()
+ return n.leaveGroupLocked(addr)
+}
+
+// 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 {
id := NetworkEndpointID{addr}
joins := n.mcastJoins[id]
switch joins {
@@ -534,10 +749,10 @@ func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error {
return nil
}
-func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, vv buffer.VectorisedView) {
+func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt tcpip.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()
}
@@ -547,9 +762,9 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address,
// 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, _ tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
+func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) {
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 {
@@ -557,36 +772,39 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr
return
}
+ // If no local link layer address is provided, assume it was sent
+ // directly to this NIC.
+ if local == "" {
+ local = n.linkEP.LinkAddress()
+ }
+
+ // 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]...)
+ }
+ n.mu.RUnlock()
+ for _, ep := range packetEPs {
+ ep.HandlePacket(n.id, local, protocol, pkt.Clone())
+ }
+
if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber {
n.stack.stats.IP.PacketsReceived.Increment()
}
- if len(vv.First()) < netProto.MinimumPacketSize() {
+ if len(pkt.Data.First()) < netProto.MinimumPacketSize() {
n.stack.stats.MalformedRcvdPackets.Increment()
return
}
- src, dst := netProto.ParseAddresses(vv.First())
-
- n.stack.AddLinkAddress(n.id, src, remote)
-
- // If the packet is destined to the IPv4 Broadcast address, then make a
- // route to each IPv4 network endpoint and let each endpoint handle the
- // packet.
- if dst == header.IPv4Broadcast {
- // n.endpoints is mutex protected so acquire lock.
- n.mu.RLock()
- for _, ref := range n.endpoints {
- if ref.isValidForIncoming() && ref.protocol == header.IPv4ProtocolNumber && ref.tryIncRef() {
- handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv)
- }
- }
- n.mu.RUnlock()
- return
- }
+ src, dst := netProto.ParseAddresses(pkt.Data.First())
if ref := n.getRef(protocol, dst); ref != nil {
- handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv)
+ handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, pkt)
return
}
@@ -614,31 +832,34 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr
if ok {
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.
- hdr := buffer.NewPrependableFromView(vv.First())
- vv.RemoveFirst()
+ pkt.Header = buffer.NewPrependableFromView(pkt.Data.First())
+ pkt.Data.RemoveFirst()
// TODO(b/128629022): use route.WritePacket.
- if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, vv, protocol); err != nil {
+ if err := n.linkEP.WritePacket(&r, nil /* gso */, protocol, pkt); err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
} else {
n.stats.Tx.Packets.Increment()
- n.stats.Tx.Bytes.IncrementBy(uint64(hdr.UsedLength() + vv.Size()))
+ n.stats.Tx.Bytes.IncrementBy(uint64(pkt.Header.UsedLength() + pkt.Data.Size()))
}
}
return
}
- n.stack.stats.IP.InvalidAddressesReceived.Increment()
+ // If a packet socket handled the packet, don't treat it as invalid.
+ if len(packetEPs) == 0 {
+ n.stack.stats.IP.InvalidAddressesReceived.Increment()
+ }
}
// 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 tcpip.PacketBuffer) {
state, ok := n.stack.transportProtocols[protocol]
if !ok {
n.stack.stats.UnknownProtocolRcvdPackets.Increment()
@@ -650,46 +871,41 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN
// Raw socket packets are delivered based solely on the transport
// protocol number. We do not inspect the payload to ensure it's
// validly formed.
- if !n.demux.deliverRawPacket(r, protocol, netHeader, vv) {
- n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv)
- }
+ n.stack.demux.deliverRawPacket(r, protocol, pkt)
- if len(vv.First()) < transProto.MinimumPacketSize() {
+ if len(pkt.Data.First()) < transProto.MinimumPacketSize() {
n.stack.stats.MalformedRcvdPackets.Increment()
return
}
- srcPort, dstPort, err := transProto.ParsePorts(vv.First())
+ srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First())
if err != nil {
n.stack.stats.MalformedRcvdPackets.Increment()
return
}
id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress}
- if n.demux.deliverPacket(r, protocol, netHeader, vv, id) {
- return
- }
- if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) {
+ 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 tcpip.PacketBuffer) {
state, ok := n.stack.transportProtocols[trans]
if !ok {
return
@@ -700,20 +916,17 @@ 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 {
+ if len(pkt.Data.First()) < 8 {
return
}
- srcPort, dstPort, err := transProto.ParsePorts(vv.First())
+ srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First())
if err != nil {
return
}
id := TransportEndpointID{srcPort, local, dstPort, remote}
- if n.demux.deliverControlPacket(net, trans, typ, extra, vv, id) {
- return
- }
- if n.stack.demux.deliverControlPacket(net, trans, typ, extra, vv, id) {
+ if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, pkt, id) {
return
}
}
@@ -728,16 +941,80 @@ func (n *NIC) Stack() *Stack {
return n.stack
}
+// 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}]
+ 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 delete the tentative address if it exists.
+func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ ref, ok := n.endpoints[NetworkEndpointID{addr}]
+ if !ok {
+ return tcpip.ErrBadAddress
+ }
+
+ if ref.getKind() != permanentTentative {
+ return tcpip.ErrInvalidEndpointState
+ }
+
+ return n.removePermanentAddressLocked(addr)
+}
+
+// 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.ndp.configs = c
+ n.mu.Unlock()
+}
+
+// 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.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 networkEndpointKind = iota
+ 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
@@ -753,8 +1030,50 @@ const (
temporary
)
+func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ eps, ok := n.packetEPs[netProto]
+ if !ok {
+ return tcpip.ErrNotSupported
+ }
+ n.packetEPs[netProto] = append(eps, ep)
+
+ return nil
+}
+
+func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) {
+ n.mu.Lock()
+ defer n.mu.Unlock()
+
+ eps, ok := n.packetEPs[netProto]
+ if !ok {
+ return
+ }
+
+ for i, epOther := range eps {
+ if epOther == ep {
+ n.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
+)
+
type referencedNetworkEndpoint struct {
- ilist.Entry
ep NetworkEndpoint
nic *NIC
protocol tcpip.NetworkProtocolNumber
@@ -769,6 +1088,10 @@ 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 after the endpoint is added to a NIC.
+ configType networkEndpointConfigType
}
func (r *referencedNetworkEndpoint) getKind() networkEndpointKind {
@@ -802,11 +1125,14 @@ func (r *referencedNetworkEndpoint) decRef() {
}
// decRefLocked is the same as decRef but assumes that the NIC.mu mutex is
-// locked.
-func (r *referencedNetworkEndpoint) decRefLocked() {
+// locked. Returns true if the endpoint was removed.
+func (r *referencedNetworkEndpoint) decRefLocked() bool {
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
generated by cgit v1.2.3 (git 2.39.1) at 2025-03-10 08:12:00 +0000