summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/stack
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/stack')
-rw-r--r--pkg/tcpip/stack/BUILD1
-rw-r--r--pkg/tcpip/stack/ndp_test.go3
-rw-r--r--pkg/tcpip/stack/nic.go175
-rw-r--r--pkg/tcpip/stack/registration.go83
-rw-r--r--pkg/tcpip/stack/route.go48
-rw-r--r--pkg/tcpip/stack/stack.go178
-rw-r--r--pkg/tcpip/stack/stack_test.go286
-rw-r--r--pkg/tcpip/stack/transport_demuxer.go2
8 files changed, 714 insertions, 62 deletions
diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD
index bfc03e90b..460db3cf8 100644
--- a/pkg/tcpip/stack/BUILD
+++ b/pkg/tcpip/stack/BUILD
@@ -73,6 +73,7 @@ go_test(
"//pkg/tcpip/transport/icmp",
"//pkg/tcpip/transport/udp",
"//pkg/waiter",
+ "@com_github_google_go-cmp//cmp:go_default_library",
],
)
diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go
index 849c2f35f..8995fbfc3 100644
--- a/pkg/tcpip/stack/ndp_test.go
+++ b/pkg/tcpip/stack/ndp_test.go
@@ -31,8 +31,7 @@ import (
const (
addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"
addr2 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"
- linkAddr1 = "\x01\x02\x03\x04\x05\x06"
- linkAddr2 = "\x01\x02\x03\x04\x05\x07"
+ linkAddr1 = "\x02\x02\x03\x04\x05\x06"
)
// TestDADDisabled tests that an address successfully resolves immediately
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 38d066cd1..e456e05f4 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -40,6 +40,9 @@ type NIC struct {
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
@@ -78,7 +81,10 @@ const (
)
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.
+ nic := &NIC{
stack: stack,
id: id,
name: name,
@@ -87,6 +93,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, 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{},
@@ -101,6 +108,16 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback
dad: make(map[tcpip.Address]dadState),
},
}
+
+ // 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
@@ -125,11 +142,50 @@ func (n *NIC) enable() *tcpip.Error {
// 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.
- if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok {
- return n.joinGroup(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress)
+ //
+ // 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
}
- 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
@@ -279,9 +335,31 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p
// 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
@@ -350,12 +428,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar
n.endpoints[id] = ref
- switch peb {
- case CanBePrimaryEndpoint:
- n.primary[protocolAddress.Protocol] = append(n.primary[protocolAddress.Protocol], ref)
- case FirstPrimaryEndpoint:
- n.primary[protocolAddress.Protocol] = append([]*referencedNetworkEndpoint{ref}, n.primary[protocolAddress.Protocol]...)
- }
+ n.insertPrimaryEndpointLocked(ref, peb)
// If we are adding a tentative IPv6 address, start DAD.
if isIPv6Unicast && kind == permanentTentative {
@@ -477,6 +550,19 @@ 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()
@@ -494,9 +580,10 @@ func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) {
}
delete(n.endpoints, id)
- for i, ref := range n.primary[r.protocol] {
+ refs := n.primary[r.protocol]
+ for i, ref := range refs {
if ref == r {
- n.primary[r.protocol] = append(n.primary[r.protocol][:i], n.primary[r.protocol][i+1:]...)
+ n.primary[r.protocol] = append(refs[:i], refs[i+1:]...)
break
}
}
@@ -568,6 +655,11 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address
// 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 {
@@ -631,7 +723,7 @@ 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, vv buffer.VectorisedView, linkHeader buffer.View) {
n.stats.Rx.Packets.Increment()
n.stats.Rx.Bytes.IncrementBy(uint64(vv.Size()))
@@ -641,6 +733,26 @@ 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, vv, linkHeader)
+ }
+
if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber {
n.stack.stats.IP.PacketsReceived.Increment()
}
@@ -700,7 +812,10 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr
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
@@ -856,6 +971,36 @@ 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 referencedNetworkEndpoint struct {
ep NetworkEndpoint
nic *NIC
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index 9d6157f22..0869fb084 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -71,8 +71,8 @@ type TransportEndpoint interface {
// RawTransportEndpoint is the interface that needs to be implemented by raw
// transport protocol endpoints. RawTransportEndpoints receive the entire
-// packet - including the link, network, and transport headers - as delivered
-// to netstack.
+// packet - including the network and transport headers - as delivered to
+// netstack.
type RawTransportEndpoint interface {
// HandlePacket is called by the stack when new packets arrive to
// this transport endpoint. The packet contains all data from the link
@@ -80,6 +80,22 @@ type RawTransportEndpoint interface {
HandlePacket(r *Route, netHeader buffer.View, packet buffer.VectorisedView)
}
+// PacketEndpoint is the interface that needs to be implemented by packet
+// transport protocol endpoints. These endpoints receive link layer headers in
+// addition to whatever they contain (usually network and transport layer
+// headers and a payload).
+type PacketEndpoint interface {
+ // HandlePacket is called by the stack when new packets arrive that
+ // match the endpoint.
+ //
+ // Implementers should treat packet as immutable and should copy it
+ // before before modification.
+ //
+ // linkHeader may have a length of 0, in which case the PacketEndpoint
+ // should construct its own ethernet header for applications.
+ HandlePacket(nicid tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, packet buffer.VectorisedView, linkHeader buffer.View)
+}
+
// TransportProtocol is the interface that needs to be implemented by transport
// protocols (e.g., tcp, udp) that want to be part of the networking stack.
type TransportProtocol interface {
@@ -185,6 +201,10 @@ type NetworkEndpoint interface {
// protocol.
WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error
+ // WritePackets writes packets to the given destination address and
+ // protocol.
+ WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error)
+
// WriteHeaderIncludedPacket writes a packet that includes a network
// header to the given destination address.
WriteHeaderIncludedPacket(r *Route, payload buffer.VectorisedView, loop PacketLooping) *tcpip.Error
@@ -242,9 +262,10 @@ type NetworkProtocol interface {
// packets to the appropriate network endpoint after it has been handled by
// the data link layer.
type NetworkDispatcher interface {
- // DeliverNetworkPacket finds the appropriate network protocol
- // endpoint and hands the packet over for further processing.
- DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
+ // DeliverNetworkPacket finds the appropriate network protocol endpoint
+ // and hands the packet over for further processing. linkHeader may have
+ // length 0 when the caller does not have ethernet data.
+ DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View)
}
// LinkEndpointCapabilities is the type associated with the capabilities
@@ -266,7 +287,11 @@ const (
CapabilitySaveRestore
CapabilityDisconnectOk
CapabilityLoopback
- CapabilityGSO
+ CapabilityHardwareGSO
+
+ // CapabilitySoftwareGSO indicates the link endpoint supports of sending
+ // multiple packets using a single call (LinkEndpoint.WritePackets).
+ CapabilitySoftwareGSO
)
// LinkEndpoint is the interface implemented by data link layer protocols (e.g.,
@@ -301,6 +326,18 @@ type LinkEndpoint interface {
// r.LocalLinkAddress if it is provided.
WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error
+ // WritePackets writes packets with the given protocol through the
+ // given route.
+ //
+ // Right now, WritePackets is used only when the software segmentation
+ // offload is enabled. If it will be used for something else, it may
+ // require to change syscall filters.
+ WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error)
+
+ // WriteRawPacket writes a packet directly to the link. The packet
+ // should already have an ethernet header.
+ WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error
+
// Attach attaches the data link layer endpoint to the network-layer
// dispatcher of the stack.
Attach(dispatcher NetworkDispatcher)
@@ -324,13 +361,14 @@ type LinkEndpoint interface {
type InjectableLinkEndpoint interface {
LinkEndpoint
- // Inject injects an inbound packet.
- Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
+ // InjectInbound injects an inbound packet.
+ InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView)
- // WriteRawPacket writes a fully formed outbound packet directly to the link.
+ // InjectOutbound writes a fully formed outbound packet directly to the
+ // link.
//
// dest is used by endpoints with multiple raw destinations.
- WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error
+ InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error
}
// A LinkAddressResolver is an extension to a NetworkProtocol that
@@ -379,11 +417,16 @@ type LinkAddressCache interface {
RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker)
}
-// UnassociatedEndpointFactory produces endpoints for writing packets not
-// associated with a particular transport protocol. Such endpoints can be used
-// to write arbitrary packets that include the IP header.
-type UnassociatedEndpointFactory interface {
- NewUnassociatedRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error)
+// RawFactory produces endpoints for writing various types of raw packets.
+type RawFactory interface {
+ // NewUnassociatedEndpoint produces endpoints for writing packets not
+ // associated with a particular transport protocol. Such endpoints can
+ // be used to write arbitrary packets that include the network header.
+ NewUnassociatedEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error)
+
+ // NewPacketEndpoint produces endpoints for reading and writing packets
+ // that include network and (when cooked is false) link layer headers.
+ NewPacketEndpoint(stack *Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error)
}
// GSOType is the type of GSO segments.
@@ -394,8 +437,14 @@ type GSOType int
// Types of gso segments.
const (
GSONone GSOType = iota
+
+ // Hardware GSO types:
GSOTCPv4
GSOTCPv6
+
+ // GSOSW is used for software GSO segments which have to be sent by
+ // endpoint.WritePackets.
+ GSOSW
)
// GSO contains generic segmentation offload properties.
@@ -423,3 +472,7 @@ type GSOEndpoint interface {
// GSOMaxSize returns the maximum GSO packet size.
GSOMaxSize() uint32
}
+
+// SoftwareGSOMaxSize is a maximum allowed size of a software GSO segment.
+// This isn't a hard limit, because it is never set into packet headers.
+const SoftwareGSOMaxSize = (1 << 16)
diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go
index e72373964..1a0a51b57 100644
--- a/pkg/tcpip/stack/route.go
+++ b/pkg/tcpip/stack/route.go
@@ -47,8 +47,8 @@ type Route struct {
// starts.
ref *referencedNetworkEndpoint
- // loop controls where WritePacket should send packets.
- loop PacketLooping
+ // Loop controls where WritePacket should send packets.
+ Loop PacketLooping
}
// makeRoute initializes a new route. It takes ownership of the provided
@@ -69,7 +69,7 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip
LocalLinkAddress: localLinkAddr,
RemoteAddress: remoteAddr,
ref: ref,
- loop: loop,
+ Loop: loop,
}
}
@@ -159,7 +159,7 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec
return tcpip.ErrInvalidEndpointState
}
- err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.loop)
+ err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop)
if err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
} else {
@@ -169,6 +169,44 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec
return err
}
+// PacketDescriptor is a packet descriptor which contains a packet header and
+// offset and size of packet data in a payload view.
+type PacketDescriptor struct {
+ Hdr buffer.Prependable
+ Off int
+ Size int
+}
+
+// NewPacketDescriptors allocates a set of packet descriptors.
+func NewPacketDescriptors(n int, hdrSize int) []PacketDescriptor {
+ buf := make([]byte, n*hdrSize)
+ hdrs := make([]PacketDescriptor, n)
+ for i := range hdrs {
+ hdrs[i].Hdr = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize])
+ }
+ return hdrs
+}
+
+// WritePackets writes the set of packets through the given route.
+func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams) (int, *tcpip.Error) {
+ if !r.ref.isValidForOutgoing() {
+ return 0, tcpip.ErrInvalidEndpointState
+ }
+
+ n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop)
+ if err != nil {
+ r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n))
+ }
+ r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n))
+ payloadSize := 0
+ for i := 0; i < n; i++ {
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdrs[i].Hdr.UsedLength()))
+ payloadSize += hdrs[i].Size
+ }
+ r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize))
+ return n, err
+}
+
// WriteHeaderIncludedPacket writes a packet already containing a network
// header through the given route.
func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error {
@@ -176,7 +214,7 @@ func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.
return tcpip.ErrInvalidEndpointState
}
- if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil {
+ if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil {
r.Stats().IP.OutgoingPacketErrors.Increment()
return err
}
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index a199bc1cc..284280917 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -351,10 +351,9 @@ type Stack struct {
networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol
linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver
- // unassociatedFactory creates unassociated endpoints. If nil, raw
- // endpoints are disabled. It is set during Stack creation and is
- // immutable.
- unassociatedFactory UnassociatedEndpointFactory
+ // rawFactory creates raw endpoints. If nil, raw endpoints are
+ // disabled. It is set during Stack creation and is immutable.
+ rawFactory RawFactory
demux *transportDemuxer
@@ -402,6 +401,11 @@ type Stack struct {
// ndpConfigs is the NDP configurations used by interfaces.
ndpConfigs NDPConfigurations
+
+ // autoGenIPv6LinkLocal determines whether or not the stack will attempt
+ // to auto-generate an IPv6 link-local address for newly enabled NICs.
+ // See the AutoGenIPv6LinkLocal field of Options for more details.
+ autoGenIPv6LinkLocal bool
}
// Options contains optional Stack configuration.
@@ -425,16 +429,28 @@ type Options struct {
// stack (false).
HandleLocal bool
- // UnassociatedFactory produces unassociated endpoints raw endpoints.
- // Raw endpoints are enabled only if this is non-nil.
- UnassociatedFactory UnassociatedEndpointFactory
-
// NDPConfigs is the NDP configurations used by interfaces.
//
// By default, NDPConfigs will have a zero value for its
// DupAddrDetectTransmits field, implying that DAD will not be performed
// before assigning an address to a NIC.
NDPConfigs NDPConfigurations
+
+ // AutoGenIPv6LinkLocal determins whether or not the stack will attempt
+ // to auto-generate an IPv6 link-local address for newly enabled NICs.
+ // Note, setting this to true does not mean that a link-local address
+ // will be assigned right away, or at all. If Duplicate Address
+ // Detection is enabled, an address will only be assigned if it
+ // successfully resolves. If it fails, no further attempt will be made
+ // to auto-generate an IPv6 link-local address.
+ //
+ // The generated link-local address will follow RFC 4291 Appendix A
+ // guidelines.
+ AutoGenIPv6LinkLocal bool
+
+ // RawFactory produces raw endpoints. Raw endpoints are enabled only if
+ // this is non-nil.
+ RawFactory RawFactory
}
// TransportEndpointInfo holds useful information about a transport endpoint
@@ -485,18 +501,19 @@ func New(opts Options) *Stack {
opts.NDPConfigs.validate()
s := &Stack{
- transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState),
- networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol),
- linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver),
- nics: make(map[tcpip.NICID]*NIC),
- linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts),
- PortManager: ports.NewPortManager(),
- clock: clock,
- stats: opts.Stats.FillIn(),
- handleLocal: opts.HandleLocal,
- icmpRateLimiter: NewICMPRateLimiter(),
- portSeed: generateRandUint32(),
- ndpConfigs: opts.NDPConfigs,
+ transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState),
+ networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol),
+ linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver),
+ nics: make(map[tcpip.NICID]*NIC),
+ linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts),
+ PortManager: ports.NewPortManager(),
+ clock: clock,
+ stats: opts.Stats.FillIn(),
+ handleLocal: opts.HandleLocal,
+ icmpRateLimiter: NewICMPRateLimiter(),
+ portSeed: generateRandUint32(),
+ ndpConfigs: opts.NDPConfigs,
+ autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal,
}
// Add specified network protocols.
@@ -514,8 +531,8 @@ func New(opts Options) *Stack {
}
}
- // Add the factory for unassociated endpoints, if present.
- s.unassociatedFactory = opts.UnassociatedFactory
+ // Add the factory for raw endpoints, if present.
+ s.rawFactory = opts.RawFactory
// Create the global transport demuxer.
s.demux = newTransportDemuxer(s)
@@ -650,12 +667,12 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp
// protocol. Raw endpoints receive all traffic for a given protocol regardless
// of address.
func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) {
- if s.unassociatedFactory == nil {
+ if s.rawFactory == nil {
return nil, tcpip.ErrNotPermitted
}
if !associated {
- return s.unassociatedFactory.NewUnassociatedRawEndpoint(s, network, transport, waiterQueue)
+ return s.rawFactory.NewUnassociatedEndpoint(s, network, transport, waiterQueue)
}
t, ok := s.transportProtocols[transport]
@@ -666,6 +683,16 @@ func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network
return t.proto.NewRawEndpoint(s, network, waiterQueue)
}
+// NewPacketEndpoint creates a new packet endpoint listening for the given
+// netProto.
+func (s *Stack) NewPacketEndpoint(cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+ if s.rawFactory == nil {
+ return nil, tcpip.ErrNotPermitted
+ }
+
+ return s.rawFactory.NewPacketEndpoint(s, cooked, netProto, waiterQueue)
+}
+
// createNIC creates a NIC with the provided id and link-layer endpoint, and
// optionally enable it.
func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled, loopback bool) *tcpip.Error {
@@ -1135,6 +1162,109 @@ func (s *Stack) Resume() {
}
}
+// RegisterPacketEndpoint registers ep with the stack, causing it to receive
+// all traffic of the specified netProto on the given NIC. If nicID is 0, it
+// receives traffic from every NIC.
+func (s *Stack) RegisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ // If no NIC is specified, capture on all devices.
+ if nicID == 0 {
+ // Register with each NIC.
+ for _, nic := range s.nics {
+ if err := nic.registerPacketEndpoint(netProto, ep); err != nil {
+ s.unregisterPacketEndpointLocked(0, netProto, ep)
+ return err
+ }
+ }
+ return nil
+ }
+
+ // Capture on a specific device.
+ nic, ok := s.nics[nicID]
+ if !ok {
+ return tcpip.ErrUnknownNICID
+ }
+ if err := nic.registerPacketEndpoint(netProto, ep); err != nil {
+ return err
+ }
+
+ return nil
+}
+
+// UnregisterPacketEndpoint unregisters ep for packets of the specified
+// netProto from the specified NIC. If nicID is 0, ep is unregistered from all
+// NICs.
+func (s *Stack) UnregisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ s.unregisterPacketEndpointLocked(nicID, netProto, ep)
+}
+
+func (s *Stack) unregisterPacketEndpointLocked(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) {
+ // If no NIC is specified, unregister on all devices.
+ if nicID == 0 {
+ // Unregister with each NIC.
+ for _, nic := range s.nics {
+ nic.unregisterPacketEndpoint(netProto, ep)
+ }
+ return
+ }
+
+ // Unregister in a single device.
+ nic, ok := s.nics[nicID]
+ if !ok {
+ return
+ }
+ nic.unregisterPacketEndpoint(netProto, ep)
+}
+
+// WritePacket writes data directly to the specified NIC. It adds an ethernet
+// header based on the arguments.
+func (s *Stack) WritePacket(nicid tcpip.NICID, dst tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, payload buffer.VectorisedView) *tcpip.Error {
+ s.mu.Lock()
+ nic, ok := s.nics[nicid]
+ s.mu.Unlock()
+ if !ok {
+ return tcpip.ErrUnknownDevice
+ }
+
+ // Add our own fake ethernet header.
+ ethFields := header.EthernetFields{
+ SrcAddr: nic.linkEP.LinkAddress(),
+ DstAddr: dst,
+ Type: netProto,
+ }
+ fakeHeader := make(header.Ethernet, header.EthernetMinimumSize)
+ fakeHeader.Encode(&ethFields)
+ ethHeader := buffer.View(fakeHeader).ToVectorisedView()
+ ethHeader.Append(payload)
+
+ if err := nic.linkEP.WriteRawPacket(ethHeader); err != nil {
+ return err
+ }
+
+ return nil
+}
+
+// WriteRawPacket writes data directly to the specified NIC without adding any
+// headers.
+func (s *Stack) WriteRawPacket(nicid tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error {
+ s.mu.Lock()
+ nic, ok := s.nics[nicid]
+ s.mu.Unlock()
+ if !ok {
+ return tcpip.ErrUnknownDevice
+ }
+
+ if err := nic.linkEP.WriteRawPacket(payload); err != nil {
+ return err
+ }
+
+ return nil
+}
+
// NetworkProtocolInstance returns the protocol instance in the stack for the
// specified network protocol. This method is public for protocol implementers
// and tests to use.
diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go
index 10fd1065f..9a8906a0d 100644
--- a/pkg/tcpip/stack/stack_test.go
+++ b/pkg/tcpip/stack/stack_test.go
@@ -24,11 +24,14 @@ import (
"sort"
"strings"
"testing"
+ "time"
+ "github.com/google/go-cmp/cmp"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
"gvisor.dev/gvisor/pkg/tcpip/header"
"gvisor.dev/gvisor/pkg/tcpip/link/channel"
+ "gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
"gvisor.dev/gvisor/pkg/tcpip/stack"
)
@@ -144,6 +147,11 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr bu
return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber)
}
+// WritePackets implements stack.LinkEndpoint.WritePackets.
+func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) {
+ panic("not implemented")
+}
+
func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
return tcpip.ErrNotSupported
}
@@ -1864,3 +1872,281 @@ func TestNICForwarding(t *testing.T) {
t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want)
}
}
+
+// TestNICAutoGenAddr tests the auto-generation of IPv6 link-local addresses
+// (or lack there-of if disabled (default)). Note, DAD will be disabled in
+// these tests.
+func TestNICAutoGenAddr(t *testing.T) {
+ tests := []struct {
+ name string
+ autoGen bool
+ linkAddr tcpip.LinkAddress
+ shouldGen bool
+ }{
+ {
+ "Disabled",
+ false,
+ linkAddr1,
+ false,
+ },
+ {
+ "Enabled",
+ true,
+ linkAddr1,
+ true,
+ },
+ {
+ "Nil MAC",
+ true,
+ tcpip.LinkAddress([]byte(nil)),
+ false,
+ },
+ {
+ "Empty MAC",
+ true,
+ tcpip.LinkAddress(""),
+ false,
+ },
+ {
+ "Invalid MAC",
+ true,
+ tcpip.LinkAddress("\x01\x02\x03"),
+ false,
+ },
+ {
+ "Multicast MAC",
+ true,
+ tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"),
+ false,
+ },
+ {
+ "Unspecified MAC",
+ true,
+ tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"),
+ false,
+ },
+ }
+
+ for _, test := range tests {
+ t.Run(test.name, func(t *testing.T) {
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ }
+
+ if test.autoGen {
+ // Only set opts.AutoGenIPv6LinkLocal when
+ // test.autoGen is true because
+ // opts.AutoGenIPv6LinkLocal should be false by
+ // default.
+ opts.AutoGenIPv6LinkLocal = true
+ }
+
+ e := channel.New(10, 1280, test.linkAddr)
+ s := stack.New(opts)
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(_) = %s", err)
+ }
+
+ addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ }
+
+ if test.shouldGen {
+ // Should have auto-generated an address and
+ // resolved immediately (DAD is disabled).
+ if want := (tcpip.AddressWithPrefix{Address: header.LinkLocalAddr(test.linkAddr), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want)
+ }
+ } else {
+ // Should not have auto-generated an address.
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ }
+ }
+ })
+ }
+}
+
+// TestNICAutoGenAddrDoesDAD tests that the successful auto-generation of IPv6
+// link-local addresses will only be assigned after the DAD process resolves.
+func TestNICAutoGenAddrDoesDAD(t *testing.T) {
+ opts := stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()},
+ NDPConfigs: stack.NDPConfigurations{
+ RetransmitTimer: time.Second,
+ DupAddrDetectTransmits: 1,
+ },
+ AutoGenIPv6LinkLocal: true,
+ }
+
+ e := channel.New(10, 1280, linkAddr1)
+ s := stack.New(opts)
+ if err := s.CreateNIC(1, e); err != nil {
+ t.Fatalf("CreateNIC(_) = %s", err)
+ }
+
+ // Address should not be considered bound to the
+ // NIC yet (DAD ongoing).
+ addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err)
+ }
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want)
+ }
+
+ // Wait for the address to resolve (an extra
+ // 250ms to make sure the address resolves).
+ //
+ // TODO(b/140896005): Use events from the
+ // netstack to know immediately when DAD
+ // completes.
+ time.Sleep(time.Second + 250*time.Millisecond)
+
+ // Should have auto-generated an address and
+ // resolved (if DAD).
+ addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber)
+ if err != nil {
+ t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err)
+ }
+ if want := (tcpip.AddressWithPrefix{Address: header.LinkLocalAddr(linkAddr1), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want {
+ t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want)
+ }
+}
+
+// TestNewPEB tests that a new PrimaryEndpointBehavior value (peb) is respected
+// when an address's kind gets "promoted" to permanent from permanentExpired.
+func TestNewPEBOnPromotionToPermanent(t *testing.T) {
+ pebs := []stack.PrimaryEndpointBehavior{
+ stack.NeverPrimaryEndpoint,
+ stack.CanBePrimaryEndpoint,
+ stack.FirstPrimaryEndpoint,
+ }
+
+ for _, pi := range pebs {
+ for _, ps := range pebs {
+ t.Run(fmt.Sprintf("%d-to-%d", pi, ps), func(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()},
+ })
+ ep1 := channel.New(10, defaultMTU, "")
+ if err := s.CreateNIC(1, ep1); err != nil {
+ t.Fatal("CreateNIC failed:", err)
+ }
+
+ // Add a permanent address with initial
+ // PrimaryEndpointBehavior (peb), pi. If pi is
+ // NeverPrimaryEndpoint, the address should not
+ // be returned by a call to GetMainNICAddress;
+ // else, it should.
+ if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", pi); err != nil {
+ t.Fatal("AddAddressWithOptions failed:", err)
+ }
+ addr, err := s.GetMainNICAddress(1, fakeNetNumber)
+ if err != nil {
+ t.Fatal("s.GetMainNICAddress failed:", err)
+ }
+ if pi == stack.NeverPrimaryEndpoint {
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want)
+
+ }
+ } else if addr.Address != "\x01" {
+ t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address)
+ }
+
+ {
+ subnet, err := tcpip.NewSubnet("\x00", "\x00")
+ if err != nil {
+ t.Fatalf("NewSubnet failed:", err)
+ }
+ s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}})
+ }
+
+ // Take a route through the address so its ref
+ // count gets incremented and does not actually
+ // get deleted when RemoveAddress is called
+ // below. This is because we want to test that a
+ // new peb is respected when an address gets
+ // "promoted" to permanent from a
+ // permanentExpired kind.
+ r, err := s.FindRoute(1, "\x01", "\x02", fakeNetNumber, false)
+ if err != nil {
+ t.Fatal("FindRoute failed:", err)
+ }
+ defer r.Release()
+ if err := s.RemoveAddress(1, "\x01"); err != nil {
+ t.Fatalf("RemoveAddress failed:", err)
+ }
+
+ //
+ // At this point, the address should still be
+ // known by the NIC, but have its
+ // kind = permanentExpired.
+ //
+
+ // Add some other address with peb set to
+ // FirstPrimaryEndpoint.
+ if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x03", stack.FirstPrimaryEndpoint); err != nil {
+ t.Fatal("AddAddressWithOptions failed:", err)
+
+ }
+
+ // Add back the address we removed earlier and
+ // make sure the new peb was respected.
+ // (The address should just be promoted now).
+ if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", ps); err != nil {
+ t.Fatal("AddAddressWithOptions failed:", err)
+ }
+ var primaryAddrs []tcpip.Address
+ for _, pa := range s.NICInfo()[1].ProtocolAddresses {
+ primaryAddrs = append(primaryAddrs, pa.AddressWithPrefix.Address)
+ }
+ var expectedList []tcpip.Address
+ switch ps {
+ case stack.FirstPrimaryEndpoint:
+ expectedList = []tcpip.Address{
+ "\x01",
+ "\x03",
+ }
+ case stack.CanBePrimaryEndpoint:
+ expectedList = []tcpip.Address{
+ "\x03",
+ "\x01",
+ }
+ case stack.NeverPrimaryEndpoint:
+ expectedList = []tcpip.Address{
+ "\x03",
+ }
+ }
+ if !cmp.Equal(primaryAddrs, expectedList) {
+ t.Fatalf("got NIC's primary addresses = %v, want = %v", primaryAddrs, expectedList)
+ }
+
+ // Once we remove the other address, if the new
+ // peb, ps, was NeverPrimaryEndpoint, no address
+ // should be returned by a call to
+ // GetMainNICAddress; else, our original address
+ // should be returned.
+ if err := s.RemoveAddress(1, "\x03"); err != nil {
+ t.Fatalf("RemoveAddress failed:", err)
+ }
+ addr, err = s.GetMainNICAddress(1, fakeNetNumber)
+ if err != nil {
+ t.Fatal("s.GetMainNICAddress failed:", err)
+ }
+ if ps == stack.NeverPrimaryEndpoint {
+ if want := (tcpip.AddressWithPrefix{}); addr != want {
+ t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want)
+
+ }
+ } else {
+ if addr.Address != "\x01" {
+ t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address)
+ }
+ }
+ })
+ }
+ }
+}
diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go
index 92267ce4d..97a1aec4b 100644
--- a/pkg/tcpip/stack/transport_demuxer.go
+++ b/pkg/tcpip/stack/transport_demuxer.go
@@ -465,7 +465,7 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer
func (d *transportDemuxer) registerRawEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error {
eps, ok := d.protocol[protocolIDs{netProto, transProto}]
if !ok {
- return nil
+ return tcpip.ErrNotSupported
}
eps.mu.Lock()
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-13 16:19:34 +0000