summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/stack
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/stack')
-rw-r--r--pkg/tcpip/stack/BUILD5
-rw-r--r--pkg/tcpip/stack/nic.go15
-rw-r--r--pkg/tcpip/stack/stack.go83
-rw-r--r--pkg/tcpip/stack/transport_demuxer.go227
-rw-r--r--pkg/tcpip/stack/transport_demuxer_test.go352
-rw-r--r--pkg/tcpip/stack/transport_test.go5
6 files changed, 543 insertions, 144 deletions
diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD
index 28c49e8ff..baf88bfab 100644
--- a/pkg/tcpip/stack/BUILD
+++ b/pkg/tcpip/stack/BUILD
@@ -36,6 +36,7 @@ go_library(
],
deps = [
"//pkg/ilist",
+ "//pkg/rand",
"//pkg/sleep",
"//pkg/tcpip",
"//pkg/tcpip/buffer",
@@ -54,6 +55,7 @@ go_test(
size = "small",
srcs = [
"stack_test.go",
+ "transport_demuxer_test.go",
"transport_test.go",
],
deps = [
@@ -64,6 +66,9 @@ go_test(
"//pkg/tcpip/iptables",
"//pkg/tcpip/link/channel",
"//pkg/tcpip/link/loopback",
+ "//pkg/tcpip/network/ipv4",
+ "//pkg/tcpip/network/ipv6",
+ "//pkg/tcpip/transport/udp",
"//pkg/waiter",
],
)
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 0e8a23f00..f6106f762 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -34,8 +34,6 @@ type NIC struct {
linkEP LinkEndpoint
loopback bool
- demux *transportDemuxer
-
mu sync.RWMutex
spoofing bool
promiscuous bool
@@ -85,7 +83,6 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback
name: name,
linkEP: ep,
loopback: loopback,
- demux: newTransportDemuxer(stack),
primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List),
endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint),
mcastJoins: make(map[NetworkEndpointID]int32),
@@ -707,9 +704,7 @@ 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, netHeader, vv)
if len(vv.First()) < transProto.MinimumPacketSize() {
n.stack.stats.MalformedRcvdPackets.Increment()
@@ -723,9 +718,6 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN
}
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) {
return
}
@@ -767,10 +759,7 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp
}
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, vv, id) {
return
}
}
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 18d1704a5..90c2cf1be 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -20,10 +20,12 @@
package stack
import (
+ "encoding/binary"
"sync"
"time"
"golang.org/x/time/rate"
+ "gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sleep"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/tcpip/buffer"
@@ -388,6 +390,12 @@ type Stack struct {
// icmpRateLimiter is a global rate limiter for all ICMP messages generated
// by the stack.
icmpRateLimiter *ICMPRateLimiter
+
+ // portSeed is a one-time random value initialized at stack startup
+ // and is used to seed the TCP port picking on active connections
+ //
+ // TODO(gvisor.dev/issues/940): S/R this field.
+ portSeed uint32
}
// Options contains optional Stack configuration.
@@ -440,6 +448,7 @@ func New(opts Options) *Stack {
stats: opts.Stats.FillIn(),
handleLocal: opts.HandleLocal,
icmpRateLimiter: NewICMPRateLimiter(),
+ portSeed: generateRandUint32(),
}
// Add specified network protocols.
@@ -1033,73 +1042,27 @@ func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.
// transport dispatcher. Received packets that match the provided id will be
// delivered to the given endpoint; specifying a nic is optional, but
// nic-specific IDs have precedence over global ones.
-func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error {
- if nicID == 0 {
- return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort)
- }
-
- s.mu.RLock()
- defer s.mu.RUnlock()
-
- nic := s.nics[nicID]
- if nic == nil {
- return tcpip.ErrUnknownNICID
- }
-
- return nic.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort)
+func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
+ return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort, bindToDevice)
}
// UnregisterTransportEndpoint removes the endpoint with the given id from the
// stack transport dispatcher.
-func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) {
- if nicID == 0 {
- s.demux.unregisterEndpoint(netProtos, protocol, id, ep)
- return
- }
-
- s.mu.RLock()
- defer s.mu.RUnlock()
-
- nic := s.nics[nicID]
- if nic != nil {
- nic.demux.unregisterEndpoint(netProtos, protocol, id, ep)
- }
+func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) {
+ s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice)
}
// RegisterRawTransportEndpoint registers the given endpoint with the stack
// transport dispatcher. Received packets that match the provided transport
// protocol will be delivered to the given endpoint.
func (s *Stack) RegisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error {
- if nicID == 0 {
- return s.demux.registerRawEndpoint(netProto, transProto, ep)
- }
-
- s.mu.RLock()
- defer s.mu.RUnlock()
-
- nic := s.nics[nicID]
- if nic == nil {
- return tcpip.ErrUnknownNICID
- }
-
- return nic.demux.registerRawEndpoint(netProto, transProto, ep)
+ return s.demux.registerRawEndpoint(netProto, transProto, ep)
}
// UnregisterRawTransportEndpoint removes the endpoint for the transport
// protocol from the stack transport dispatcher.
func (s *Stack) UnregisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) {
- if nicID == 0 {
- s.demux.unregisterRawEndpoint(netProto, transProto, ep)
- return
- }
-
- s.mu.RLock()
- defer s.mu.RUnlock()
-
- nic := s.nics[nicID]
- if nic != nil {
- nic.demux.unregisterRawEndpoint(netProto, transProto, ep)
- }
+ s.demux.unregisterRawEndpoint(netProto, transProto, ep)
}
// RegisterRestoredEndpoint records e as an endpoint that has been restored on
@@ -1243,3 +1206,19 @@ func (s *Stack) SetICMPBurst(burst int) {
func (s *Stack) AllowICMPMessage() bool {
return s.icmpRateLimiter.Allow()
}
+
+// PortSeed returns a 32 bit value that can be used as a seed value for port
+// picking.
+//
+// NOTE: The seed is generated once during stack initialization only.
+func (s *Stack) PortSeed() uint32 {
+ return s.portSeed
+}
+
+func generateRandUint32() uint32 {
+ b := make([]byte, 4)
+ if _, err := rand.Read(b); err != nil {
+ panic(err)
+ }
+ return binary.LittleEndian.Uint32(b)
+}
diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go
index cf8a6d129..8c768c299 100644
--- a/pkg/tcpip/stack/transport_demuxer.go
+++ b/pkg/tcpip/stack/transport_demuxer.go
@@ -35,25 +35,109 @@ type protocolIDs struct {
type transportEndpoints struct {
// mu protects all fields of the transportEndpoints.
mu sync.RWMutex
- endpoints map[TransportEndpointID]TransportEndpoint
+ endpoints map[TransportEndpointID]*endpointsByNic
// rawEndpoints contains endpoints for raw sockets, which receive all
// traffic of a given protocol regardless of port.
rawEndpoints []RawTransportEndpoint
}
+type endpointsByNic struct {
+ mu sync.RWMutex
+ endpoints map[tcpip.NICID]*multiPortEndpoint
+ // seed is a random secret for a jenkins hash.
+ seed uint32
+}
+
+// HandlePacket is called by the stack when new packets arrive to this transport
+// endpoint.
+func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) {
+ epsByNic.mu.RLock()
+
+ mpep, ok := epsByNic.endpoints[r.ref.nic.ID()]
+ if !ok {
+ if mpep, ok = epsByNic.endpoints[0]; !ok {
+ epsByNic.mu.RUnlock() // Don't use defer for performance reasons.
+ return
+ }
+ }
+
+ // If this is a broadcast or multicast datagram, deliver the datagram to all
+ // endpoints bound to the right device.
+ if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) {
+ mpep.handlePacketAll(r, id, vv)
+ epsByNic.mu.RUnlock() // Don't use defer for performance reasons.
+ return
+ }
+
+ // multiPortEndpoints are guaranteed to have at least one element.
+ selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, vv)
+ epsByNic.mu.RUnlock() // Don't use defer for performance reasons.
+}
+
+// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
+func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) {
+ epsByNic.mu.RLock()
+ defer epsByNic.mu.RUnlock()
+
+ mpep, ok := epsByNic.endpoints[n.ID()]
+ if !ok {
+ mpep, ok = epsByNic.endpoints[0]
+ }
+ if !ok {
+ return
+ }
+
+ // TODO(eyalsoha): Why don't we look at id to see if this packet needs to
+ // broadcast like we are doing with handlePacket above?
+
+ // multiPortEndpoints are guaranteed to have at least one element.
+ selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, vv)
+}
+
+// registerEndpoint returns true if it succeeds. It fails and returns
+// false if ep already has an element with the same key.
+func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
+ epsByNic.mu.Lock()
+ defer epsByNic.mu.Unlock()
+
+ if multiPortEp, ok := epsByNic.endpoints[bindToDevice]; ok {
+ // There was already a bind.
+ return multiPortEp.singleRegisterEndpoint(t, reusePort)
+ }
+
+ // This is a new binding.
+ multiPortEp := &multiPortEndpoint{}
+ multiPortEp.endpointsMap = make(map[TransportEndpoint]int)
+ multiPortEp.reuse = reusePort
+ epsByNic.endpoints[bindToDevice] = multiPortEp
+ return multiPortEp.singleRegisterEndpoint(t, reusePort)
+}
+
+// unregisterEndpoint returns true if endpointsByNic has to be unregistered.
+func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint) bool {
+ epsByNic.mu.Lock()
+ defer epsByNic.mu.Unlock()
+ multiPortEp, ok := epsByNic.endpoints[bindToDevice]
+ if !ok {
+ return false
+ }
+ if multiPortEp.unregisterEndpoint(t) {
+ delete(epsByNic.endpoints, bindToDevice)
+ }
+ return len(epsByNic.endpoints) == 0
+}
+
// unregisterEndpoint unregisters the endpoint with the given id such that it
// won't receive any more packets.
-func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint) {
+func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) {
eps.mu.Lock()
defer eps.mu.Unlock()
- e, ok := eps.endpoints[id]
+ epsByNic, ok := eps.endpoints[id]
if !ok {
return
}
- if multiPortEp, ok := e.(*multiPortEndpoint); ok {
- if !multiPortEp.unregisterEndpoint(ep) {
- return
- }
+ if !epsByNic.unregisterEndpoint(bindToDevice, ep) {
+ return
}
delete(eps.endpoints, id)
}
@@ -75,7 +159,7 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer {
for netProto := range stack.networkProtocols {
for proto := range stack.transportProtocols {
d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{
- endpoints: make(map[TransportEndpointID]TransportEndpoint),
+ endpoints: make(map[TransportEndpointID]*endpointsByNic),
}
}
}
@@ -85,10 +169,10 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer {
// registerEndpoint registers the given endpoint with the dispatcher such that
// packets that match the endpoint ID are delivered to it.
-func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error {
+func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
for i, n := range netProtos {
- if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort); err != nil {
- d.unregisterEndpoint(netProtos[:i], protocol, id, ep)
+ if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort, bindToDevice); err != nil {
+ d.unregisterEndpoint(netProtos[:i], protocol, id, ep, bindToDevice)
return err
}
}
@@ -97,13 +181,14 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum
}
// multiPortEndpoint is a container for TransportEndpoints which are bound to
-// the same pair of address and port.
+// the same pair of address and port. endpointsArr always has at least one
+// element.
type multiPortEndpoint struct {
mu sync.RWMutex
endpointsArr []TransportEndpoint
endpointsMap map[TransportEndpoint]int
- // seed is a random secret for a jenkins hash.
- seed uint32
+ // reuse indicates if more than one endpoint is allowed.
+ reuse bool
}
// reciprocalScale scales a value into range [0, n).
@@ -117,9 +202,10 @@ func reciprocalScale(val, n uint32) uint32 {
// selectEndpoint calculates a hash of destination and source addresses and
// ports then uses it to select a socket. In this case, all packets from one
// address will be sent to same endpoint.
-func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEndpoint {
- ep.mu.RLock()
- defer ep.mu.RUnlock()
+func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32) TransportEndpoint {
+ if len(mpep.endpointsArr) == 1 {
+ return mpep.endpointsArr[0]
+ }
payload := []byte{
byte(id.LocalPort),
@@ -128,51 +214,50 @@ func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEnd
byte(id.RemotePort >> 8),
}
- h := jenkins.Sum32(ep.seed)
+ h := jenkins.Sum32(seed)
h.Write(payload)
h.Write([]byte(id.LocalAddress))
h.Write([]byte(id.RemoteAddress))
hash := h.Sum32()
- idx := reciprocalScale(hash, uint32(len(ep.endpointsArr)))
- return ep.endpointsArr[idx]
+ idx := reciprocalScale(hash, uint32(len(mpep.endpointsArr)))
+ return mpep.endpointsArr[idx]
}
-// HandlePacket is called by the stack when new packets arrive to this transport
-// endpoint.
-func (ep *multiPortEndpoint) HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) {
- // If this is a broadcast or multicast datagram, deliver the datagram to all
- // endpoints managed by ep.
- if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) {
- for i, endpoint := range ep.endpointsArr {
- // HandlePacket modifies vv, so each endpoint needs its own copy.
- if i == len(ep.endpointsArr)-1 {
- endpoint.HandlePacket(r, id, vv)
- break
- }
- vvCopy := buffer.NewView(vv.Size())
- copy(vvCopy, vv.ToView())
- endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView())
+func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, vv buffer.VectorisedView) {
+ ep.mu.RLock()
+ for i, endpoint := range ep.endpointsArr {
+ // HandlePacket modifies vv, so each endpoint needs its own copy except for
+ // the final one.
+ if i == len(ep.endpointsArr)-1 {
+ endpoint.HandlePacket(r, id, vv)
+ break
}
- } else {
- ep.selectEndpoint(id).HandlePacket(r, id, vv)
+ vvCopy := buffer.NewView(vv.Size())
+ copy(vvCopy, vv.ToView())
+ endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView())
}
+ ep.mu.RUnlock() // Don't use defer for performance reasons.
}
-// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
-func (ep *multiPortEndpoint) HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) {
- ep.selectEndpoint(id).HandleControlPacket(id, typ, extra, vv)
-}
-
-func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint) {
+// singleRegisterEndpoint tries to add an endpoint to the multiPortEndpoint
+// list. The list might be empty already.
+func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePort bool) *tcpip.Error {
ep.mu.Lock()
defer ep.mu.Unlock()
- // A new endpoint is added into endpointsArr and its index there is
- // saved in endpointsMap. This will allows to remove endpoint from
- // the array fast.
+ if len(ep.endpointsArr) > 0 {
+ // If it was previously bound, we need to check if we can bind again.
+ if !ep.reuse || !reusePort {
+ return tcpip.ErrPortInUse
+ }
+ }
+
+ // A new endpoint is added into endpointsArr and its index there is saved in
+ // endpointsMap. This will allow us to remove endpoint from the array fast.
ep.endpointsMap[t] = len(ep.endpointsArr)
ep.endpointsArr = append(ep.endpointsArr, t)
+ return nil
}
// unregisterEndpoint returns true if multiPortEndpoint has to be unregistered.
@@ -197,53 +282,41 @@ func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint) bool {
return true
}
-func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error {
+func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error {
if id.RemotePort != 0 {
+ // TODO(eyalsoha): Why?
reusePort = false
}
eps, ok := d.protocol[protocolIDs{netProto, protocol}]
if !ok {
- return nil
+ return tcpip.ErrUnknownProtocol
}
eps.mu.Lock()
defer eps.mu.Unlock()
- var multiPortEp *multiPortEndpoint
- if _, ok := eps.endpoints[id]; ok {
- if !reusePort {
- return tcpip.ErrPortInUse
- }
- multiPortEp, ok = eps.endpoints[id].(*multiPortEndpoint)
- if !ok {
- return tcpip.ErrPortInUse
- }
+ if epsByNic, ok := eps.endpoints[id]; ok {
+ // There was already a binding.
+ return epsByNic.registerEndpoint(ep, reusePort, bindToDevice)
}
- if reusePort {
- if multiPortEp == nil {
- multiPortEp = &multiPortEndpoint{}
- multiPortEp.endpointsMap = make(map[TransportEndpoint]int)
- multiPortEp.seed = rand.Uint32()
- eps.endpoints[id] = multiPortEp
- }
-
- multiPortEp.singleRegisterEndpoint(ep)
-
- return nil
+ // This is a new binding.
+ epsByNic := &endpointsByNic{
+ endpoints: make(map[tcpip.NICID]*multiPortEndpoint),
+ seed: rand.Uint32(),
}
- eps.endpoints[id] = ep
+ eps.endpoints[id] = epsByNic
- return nil
+ return epsByNic.registerEndpoint(ep, reusePort, bindToDevice)
}
// unregisterEndpoint unregisters the endpoint with the given id such that it
// won't receive any more packets.
-func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) {
+func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) {
for _, n := range netProtos {
if eps, ok := d.protocol[protocolIDs{n, protocol}]; ok {
- eps.unregisterEndpoint(id, ep)
+ eps.unregisterEndpoint(id, ep, bindToDevice)
}
}
}
@@ -273,7 +346,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto
// If the packet is a broadcast, then find all matching transport endpoints.
// Otherwise, try to find a single matching transport endpoint.
- destEps := make([]TransportEndpoint, 0, 1)
+ destEps := make([]*endpointsByNic, 0, 1)
eps.mu.RLock()
if protocol == header.UDPProtocolNumber && id.LocalAddress == header.IPv4Broadcast {
@@ -299,7 +372,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto
// Deliver the packet.
for _, ep := range destEps {
- ep.HandlePacket(r, id, vv)
+ ep.handlePacket(r, id, vv)
}
return true
@@ -331,7 +404,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr
// deliverControlPacket attempts to deliver the given control packet. Returns
// true if it found an endpoint, false otherwise.
-func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool {
+func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool {
eps, ok := d.protocol[protocolIDs{net, trans}]
if !ok {
return false
@@ -348,12 +421,12 @@ func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber,
}
// Deliver the packet.
- ep.HandleControlPacket(id, typ, extra, vv)
+ ep.handleControlPacket(n, id, typ, extra, vv)
return true
}
-func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) TransportEndpoint {
+func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) *endpointsByNic {
// Try to find a match with the id as provided.
if ep, ok := eps.endpoints[id]; ok {
return ep
diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go
new file mode 100644
index 000000000..210233dc0
--- /dev/null
+++ b/pkg/tcpip/stack/transport_demuxer_test.go
@@ -0,0 +1,352 @@
+// Copyright 2018 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package stack_test
+
+import (
+ "math"
+ "math/rand"
+ "testing"
+
+ "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/ipv4"
+ "gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
+ "gvisor.dev/gvisor/pkg/tcpip/stack"
+ "gvisor.dev/gvisor/pkg/tcpip/transport/udp"
+ "gvisor.dev/gvisor/pkg/waiter"
+)
+
+const (
+ stackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"
+ testV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"
+
+ stackAddr = "\x0a\x00\x00\x01"
+ stackPort = 1234
+ testPort = 4096
+)
+
+type testContext struct {
+ t *testing.T
+ linkEPs map[string]*channel.Endpoint
+ s *stack.Stack
+
+ ep tcpip.Endpoint
+ wq waiter.Queue
+}
+
+func (c *testContext) cleanup() {
+ if c.ep != nil {
+ c.ep.Close()
+ }
+}
+
+func (c *testContext) createV6Endpoint(v6only bool) {
+ var err *tcpip.Error
+ c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq)
+ if err != nil {
+ c.t.Fatalf("NewEndpoint failed: %v", err)
+ }
+
+ var v tcpip.V6OnlyOption
+ if v6only {
+ v = 1
+ }
+ if err := c.ep.SetSockOpt(v); err != nil {
+ c.t.Fatalf("SetSockOpt failed: %v", err)
+ }
+}
+
+// newDualTestContextMultiNic creates the testing context and also linkEpNames
+// named NICs.
+func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()},
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}})
+ linkEPs := make(map[string]*channel.Endpoint)
+ for i, linkEpName := range linkEpNames {
+ channelEP := channel.New(256, mtu, "")
+ nicid := tcpip.NICID(i + 1)
+ if err := s.CreateNamedNIC(nicid, linkEpName, channelEP); err != nil {
+ t.Fatalf("CreateNIC failed: %v", err)
+ }
+ linkEPs[linkEpName] = channelEP
+
+ if err := s.AddAddress(nicid, ipv4.ProtocolNumber, stackAddr); err != nil {
+ t.Fatalf("AddAddress IPv4 failed: %v", err)
+ }
+
+ if err := s.AddAddress(nicid, ipv6.ProtocolNumber, stackV6Addr); err != nil {
+ t.Fatalf("AddAddress IPv6 failed: %v", err)
+ }
+ }
+
+ s.SetRouteTable([]tcpip.Route{
+ {
+ Destination: header.IPv4EmptySubnet,
+ NIC: 1,
+ },
+ {
+ Destination: header.IPv6EmptySubnet,
+ NIC: 1,
+ },
+ })
+
+ return &testContext{
+ t: t,
+ s: s,
+ linkEPs: linkEPs,
+ }
+}
+
+type headers struct {
+ srcPort uint16
+ dstPort uint16
+}
+
+func newPayload() []byte {
+ b := make([]byte, 30+rand.Intn(100))
+ for i := range b {
+ b[i] = byte(rand.Intn(256))
+ }
+ return b
+}
+
+func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) {
+ // Allocate a buffer for data and headers.
+ buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload))
+ copy(buf[len(buf)-len(payload):], payload)
+
+ // Initialize the IP header.
+ ip := header.IPv6(buf)
+ ip.Encode(&header.IPv6Fields{
+ PayloadLength: uint16(header.UDPMinimumSize + len(payload)),
+ NextHeader: uint8(udp.ProtocolNumber),
+ HopLimit: 65,
+ SrcAddr: testV6Addr,
+ DstAddr: stackV6Addr,
+ })
+
+ // Initialize the UDP header.
+ u := header.UDP(buf[header.IPv6MinimumSize:])
+ u.Encode(&header.UDPFields{
+ SrcPort: h.srcPort,
+ DstPort: h.dstPort,
+ Length: uint16(header.UDPMinimumSize + len(payload)),
+ })
+
+ // Calculate the UDP pseudo-header checksum.
+ xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testV6Addr, stackV6Addr, uint16(len(u)))
+
+ // Calculate the UDP checksum and set it.
+ xsum = header.Checksum(payload, xsum)
+ u.SetChecksum(^u.CalculateChecksum(xsum))
+
+ // Inject packet.
+ c.linkEPs[linkEpName].Inject(ipv6.ProtocolNumber, buf.ToVectorisedView())
+}
+
+func TestTransportDemuxerRegister(t *testing.T) {
+ for _, test := range []struct {
+ name string
+ proto tcpip.NetworkProtocolNumber
+ want *tcpip.Error
+ }{
+ {"failure", ipv6.ProtocolNumber, tcpip.ErrUnknownProtocol},
+ {"success", ipv4.ProtocolNumber, nil},
+ } {
+ t.Run(test.name, func(t *testing.T) {
+ s := stack.New(stack.Options{
+ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()},
+ TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}})
+ if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, nil, false, 0), test.want; got != want {
+ t.Fatalf("s.RegisterTransportEndpoint(...) = %v, want %v", got, want)
+ }
+ })
+ }
+}
+
+// TestReuseBindToDevice injects varied packets on input devices and checks that
+// the distribution of packets received matches expectations.
+func TestDistribution(t *testing.T) {
+ type endpointSockopts struct {
+ reuse int
+ bindToDevice string
+ }
+ for _, test := range []struct {
+ name string
+ // endpoints will received the inject packets.
+ endpoints []endpointSockopts
+ // wantedDistribution is the wanted ratio of packets received on each
+ // endpoint for each NIC on which packets are injected.
+ wantedDistributions map[string][]float64
+ }{
+ {
+ "BindPortReuse",
+ // 5 endpoints that all have reuse set.
+ []endpointSockopts{
+ endpointSockopts{1, ""},
+ endpointSockopts{1, ""},
+ endpointSockopts{1, ""},
+ endpointSockopts{1, ""},
+ endpointSockopts{1, ""},
+ },
+ map[string][]float64{
+ // Injected packets on dev0 get distributed evenly.
+ "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2},
+ },
+ },
+ {
+ "BindToDevice",
+ // 3 endpoints with various bindings.
+ []endpointSockopts{
+ endpointSockopts{0, "dev0"},
+ endpointSockopts{0, "dev1"},
+ endpointSockopts{0, "dev2"},
+ },
+ map[string][]float64{
+ // Injected packets on dev0 go only to the endpoint bound to dev0.
+ "dev0": []float64{1, 0, 0},
+ // Injected packets on dev1 go only to the endpoint bound to dev1.
+ "dev1": []float64{0, 1, 0},
+ // Injected packets on dev2 go only to the endpoint bound to dev2.
+ "dev2": []float64{0, 0, 1},
+ },
+ },
+ {
+ "ReuseAndBindToDevice",
+ // 6 endpoints with various bindings.
+ []endpointSockopts{
+ endpointSockopts{1, "dev0"},
+ endpointSockopts{1, "dev0"},
+ endpointSockopts{1, "dev1"},
+ endpointSockopts{1, "dev1"},
+ endpointSockopts{1, "dev1"},
+ endpointSockopts{1, ""},
+ },
+ map[string][]float64{
+ // Injected packets on dev0 get distributed among endpoints bound to
+ // dev0.
+ "dev0": []float64{0.5, 0.5, 0, 0, 0, 0},
+ // Injected packets on dev1 get distributed among endpoints bound to
+ // dev1 or unbound.
+ "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0},
+ // Injected packets on dev999 go only to the unbound.
+ "dev999": []float64{0, 0, 0, 0, 0, 1},
+ },
+ },
+ } {
+ t.Run(test.name, func(t *testing.T) {
+ for device, wantedDistribution := range test.wantedDistributions {
+ t.Run(device, func(t *testing.T) {
+ var devices []string
+ for d := range test.wantedDistributions {
+ devices = append(devices, d)
+ }
+ c := newDualTestContextMultiNic(t, defaultMTU, devices)
+ defer c.cleanup()
+
+ c.createV6Endpoint(false)
+
+ eps := make(map[tcpip.Endpoint]int)
+
+ pollChannel := make(chan tcpip.Endpoint)
+ for i, endpoint := range test.endpoints {
+ // Try to receive the data.
+ wq := waiter.Queue{}
+ we, ch := waiter.NewChannelEntry(nil)
+ wq.EventRegister(&we, waiter.EventIn)
+ defer wq.EventUnregister(&we)
+ defer close(ch)
+
+ var err *tcpip.Error
+ ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq)
+ if err != nil {
+ c.t.Fatalf("NewEndpoint failed: %v", err)
+ }
+ eps[ep] = i
+
+ go func(ep tcpip.Endpoint) {
+ for range ch {
+ pollChannel <- ep
+ }
+ }(ep)
+
+ defer ep.Close()
+ reusePortOption := tcpip.ReusePortOption(endpoint.reuse)
+ if err := ep.SetSockOpt(reusePortOption); err != nil {
+ c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", reusePortOption, i, err)
+ }
+ bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice)
+ if err := ep.SetSockOpt(bindToDeviceOption); err != nil {
+ c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", bindToDeviceOption, i, err)
+ }
+ if err := ep.Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil {
+ t.Fatalf("ep.Bind(...) on endpoint %d failed: %v", i, err)
+ }
+ }
+
+ npackets := 100000
+ nports := 10000
+ if got, want := len(test.endpoints), len(wantedDistribution); got != want {
+ t.Fatalf("got len(test.endpoints) = %d, want %d", got, want)
+ }
+ ports := make(map[uint16]tcpip.Endpoint)
+ stats := make(map[tcpip.Endpoint]int)
+ for i := 0; i < npackets; i++ {
+ // Send a packet.
+ port := uint16(i % nports)
+ payload := newPayload()
+ c.sendV6Packet(payload,
+ &headers{
+ srcPort: testPort + port,
+ dstPort: stackPort},
+ device)
+
+ var addr tcpip.FullAddress
+ ep := <-pollChannel
+ _, _, err := ep.Read(&addr)
+ if err != nil {
+ c.t.Fatalf("Read on endpoint %d failed: %v", eps[ep], err)
+ }
+ stats[ep]++
+ if i < nports {
+ ports[uint16(i)] = ep
+ } else {
+ // Check that all packets from one client are handled by the same
+ // socket.
+ if want, got := ports[port], ep; want != got {
+ t.Fatalf("Packet sent on port %d expected on endpoint %d but received on endpoint %d", port, eps[want], eps[got])
+ }
+ }
+ }
+
+ // Check that a packet distribution is as expected.
+ for ep, i := range eps {
+ wantedRatio := wantedDistribution[i]
+ wantedRecv := wantedRatio * float64(npackets)
+ actualRecv := stats[ep]
+ actualRatio := float64(stats[ep]) / float64(npackets)
+ // The deviation is less than 10%.
+ if math.Abs(actualRatio-wantedRatio) > 0.05 {
+ t.Errorf("wanted about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantedRatio*100, wantedRecv, npackets, i, actualRatio*100, actualRecv, npackets)
+ }
+ }
+ })
+ }
+ })
+ }
+}
diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go
index 56e8a5d9b..842a16277 100644
--- a/pkg/tcpip/stack/transport_test.go
+++ b/pkg/tcpip/stack/transport_test.go
@@ -127,7 +127,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
// Try to register so that we can start receiving packets.
f.id.RemoteAddress = addr.Addr
- err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false)
+ err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false /* reuse */, 0 /* bindToDevice */)
if err != nil {
return err
}
@@ -168,7 +168,8 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error {
fakeTransNumber,
stack.TransportEndpointID{LocalAddress: a.Addr},
f,
- false,
+ false, /* reuse */
+ 0, /* bindtoDevice */
); err != nil {
return err
}
generated by cgit v1.2.3 (git 2.39.1) at 2025-01-13 12:42:38 +0000