netstack/fdbased: add generic segmentation offload (GSO) support

The linux packet socket can handle GSO packets, so we can segment packets to
64K instead of the MTU which is usually 1500.

Here are numbers for the nginx-1m test:
runsc:		579330.01 [Kbytes/sec] received
runsc-gso:	1794121.66 [Kbytes/sec] received
runc:		2122139.06 [Kbytes/sec] received

and for tcp_benchmark:

$ tcp_benchmark  --duration 15   --ideal
[  4]  0.0-15.0 sec  86647 MBytes  48456 Mbits/sec

$ tcp_benchmark --client --duration 15   --ideal
[  4]  0.0-15.0 sec  2173 MBytes  1214 Mbits/sec

$ tcp_benchmark --client --duration 15   --ideal --gso 65536
[  4]  0.0-15.0 sec  19357 MBytes  10825 Mbits/sec

PiperOrigin-RevId: 240809103
Change-Id: I2637f104db28b5d4c64e1e766c610162a195775a

5 files changed, 56 insertions, 9 deletions

diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 1d032ebf8..8b6c17a90 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -486,7 +486,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr
 			vv.RemoveFirst()
 
 			// TODO: use route.WritePacket.
-			if err := n.linkEP.WritePacket(&r, hdr, vv, protocol); err != nil {
+			if err := n.linkEP.WritePacket(&r, nil /* gso */, hdr, vv, protocol); err != nil {
 				r.Stats().IP.OutgoingPacketErrors.Increment()
 			} else {
 				n.stats.Tx.Packets.Increment()
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index cf4d52fe9..ff356ea22 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -161,7 +161,7 @@ type NetworkEndpoint interface {
 
 	// WritePacket writes a packet to the given destination address and
 	// protocol.
-	WritePacket(r *Route, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, loop PacketLooping) *tcpip.Error
+	WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, loop PacketLooping) *tcpip.Error
 
 	// ID returns the network protocol endpoint ID.
 	ID() *NetworkEndpointID
@@ -226,6 +226,7 @@ const (
 	CapabilitySaveRestore
 	CapabilityDisconnectOk
 	CapabilityLoopback
+	CapabilityGSO
 )
 
 // LinkEndpoint is the interface implemented by data link layer protocols (e.g.,
@@ -258,7 +259,7 @@ type LinkEndpoint interface {
 	// To participate in transparent bridging, a LinkEndpoint implementation
 	// should call eth.Encode with header.EthernetFields.SrcAddr set to
 	// r.LocalLinkAddress if it is provided.
-	WritePacket(r *Route, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error
+	WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error
 
 	// Attach attaches the data link layer endpoint to the network-layer
 	// dispatcher of the stack.
@@ -381,3 +382,41 @@ func FindLinkEndpoint(id tcpip.LinkEndpointID) LinkEndpoint {
 
 	return linkEndpoints[id]
 }
+
+// GSOType is the type of GSO segments.
+//
+// +stateify savable
+type GSOType int
+
+// Types of gso segments.
+const (
+	GSONone GSOType = iota
+	GSOTCPv4
+	GSOTCPv6
+)
+
+// GSO contains generic segmentation offload properties.
+//
+// +stateify savable
+type GSO struct {
+	// Type is one of GSONone, GSOTCPv4, etc.
+	Type GSOType
+	// NeedsCsum is set if the checksum offload is enabled.
+	NeedsCsum bool
+	// CsumOffset is offset after that to place checksum.
+	CsumOffset uint16
+
+	// Mss is maximum segment size.
+	MSS uint16
+	// L3Len is L3 (IP) header length.
+	L3HdrLen uint16
+
+	// MaxSize is maximum GSO packet size.
+	MaxSize uint32
+}
+
+// GSOEndpoint provides access to GSO properties.
+type GSOEndpoint interface {
+	// GSOMaxSize returns the maximum GSO packet size.
+	GSOMaxSize() uint32
+}
diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go
index ee860eafe..8ae562dcd 100644
--- a/pkg/tcpip/stack/route.go
+++ b/pkg/tcpip/stack/route.go
@@ -97,6 +97,14 @@ func (r *Route) Capabilities() LinkEndpointCapabilities {
 	return r.ref.ep.Capabilities()
 }
 
+// GSOMaxSize returns the maximum GSO packet size.
+func (r *Route) GSOMaxSize() uint32 {
+	if gso, ok := r.ref.ep.(GSOEndpoint); ok {
+		return gso.GSOMaxSize()
+	}
+	return 0
+}
+
 // Resolve attempts to resolve the link address if necessary. Returns ErrWouldBlock in
 // case address resolution requires blocking, e.g. wait for ARP reply. Waker is
 // notified when address resolution is complete (success or not).
@@ -144,8 +152,8 @@ func (r *Route) IsResolutionRequired() bool {
 }
 
 // WritePacket writes the packet through the given route.
-func (r *Route) WritePacket(hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8) *tcpip.Error {
-	err := r.ref.ep.WritePacket(r, hdr, payload, protocol, ttl, r.loop)
+func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8) *tcpip.Error {
+	err := r.ref.ep.WritePacket(r, gso, hdr, payload, protocol, ttl, r.loop)
 	if err != nil {
 		r.Stats().IP.OutgoingPacketErrors.Increment()
 	} else {
diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go
index da8269999..b5375df3c 100644
--- a/pkg/tcpip/stack/stack_test.go
+++ b/pkg/tcpip/stack/stack_test.go
@@ -112,7 +112,7 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities {
 	return f.linkEP.Capabilities()
 }
 
-func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, _ uint8, loop stack.PacketLooping) *tcpip.Error {
+func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, _ uint8, loop stack.PacketLooping) *tcpip.Error {
 	// Increment the sent packet count in the protocol descriptor.
 	f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++
 
@@ -134,7 +134,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, hdr buffer.Prependable
 		return nil
 	}
 
-	return f.linkEP.WritePacket(r, hdr, payload, fakeNetNumber)
+	return f.linkEP.WritePacket(r, gso, hdr, payload, fakeNetNumber)
 }
 
 func (*fakeNetworkEndpoint) Close() {}
@@ -281,7 +281,7 @@ func sendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, payload buffer.Vie
 	defer r.Release()
 
 	hdr := buffer.NewPrependable(int(r.MaxHeaderLength()))
-	if err := r.WritePacket(hdr, payload.ToVectorisedView(), fakeTransNumber, 123); err != nil {
+	if err := r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), fakeTransNumber, 123); err != nil {
 		t.Errorf("WritePacket failed: %v", err)
 	}
 }
diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go
index 279ab3c56..dfd31557a 100644
--- a/pkg/tcpip/stack/transport_test.go
+++ b/pkg/tcpip/stack/transport_test.go
@@ -74,7 +74,7 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions)
 	if err != nil {
 		return 0, nil, err
 	}
-	if err := f.route.WritePacket(hdr, buffer.View(v).ToVectorisedView(), fakeTransNumber, 123); err != nil {
+	if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), fakeTransNumber, 123); err != nil {
 		return 0, nil, err
 	}