Add IPPROTO_RAW, which allows raw sockets to write IP headers.

iptables also relies on IPPROTO_RAW in a way. It opens such a socket to
manipulate the kernel's tables, but it doesn't actually use any of the
functionality. Blegh.

PiperOrigin-RevId: 257903078

10 files changed, 215 insertions, 24 deletions

diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go
index ca3d6c0bf..cb35635fc 100644
--- a/pkg/tcpip/network/arp/arp.go
+++ b/pkg/tcpip/network/arp/arp.go
@@ -83,6 +83,10 @@ func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buf
 	return tcpip.ErrNotSupported
 }
 
+func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
+	return tcpip.ErrNotSupported
+}
+
 func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) {
 	v := vv.First()
 	h := header.ARP(v)
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 1e3a7425a..e44a73d96 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -232,6 +232,55 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen
 	return nil
 }
 
+// WriteHeaderIncludedPacket writes a packet already containing a network
+// header through the given route.
+func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
+	// The packet already has an IP header, but there are a few required
+	// checks.
+	ip := header.IPv4(payload.First())
+	if !ip.IsValid(payload.Size()) {
+		return tcpip.ErrInvalidOptionValue
+	}
+
+	// Always set the total length.
+	ip.SetTotalLength(uint16(payload.Size()))
+
+	// Set the source address when zero.
+	if ip.SourceAddress() == tcpip.Address(([]byte{0, 0, 0, 0})) {
+		ip.SetSourceAddress(r.LocalAddress)
+	}
+
+	// Set the destination. If the packet already included a destination,
+	// it will be part of the route.
+	ip.SetDestinationAddress(r.RemoteAddress)
+
+	// Set the packet ID when zero.
+	if ip.ID() == 0 {
+		id := uint32(0)
+		if payload.Size() > header.IPv4MaximumHeaderSize+8 {
+			// Packets of 68 bytes or less are required by RFC 791 to not be
+			// fragmented, so we only assign ids to larger packets.
+			id = atomic.AddUint32(&ids[hashRoute(r, 0 /* protocol */)%buckets], 1)
+		}
+		ip.SetID(uint16(id))
+	}
+
+	// Always set the checksum.
+	ip.SetChecksum(0)
+	ip.SetChecksum(^ip.CalculateChecksum())
+
+	if loop&stack.PacketLoop != 0 {
+		e.HandlePacket(r, payload)
+	}
+	if loop&stack.PacketOut == 0 {
+		return nil
+	}
+
+	hdr := buffer.NewPrependableFromView(payload.ToView())
+	r.Stats().IP.PacketsSent.Increment()
+	return e.linkEP.WritePacket(r, nil /* gso */, hdr, buffer.VectorisedView{}, ProtocolNumber)
+}
+
 // HandlePacket is called by the link layer when new ipv4 packets arrive for
 // this endpoint.
 func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) {
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 27367d6c5..e3e8739fd 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -120,6 +120,13 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen
 	return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber)
 }
 
+// WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet
+// supported by IPv6.
+func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
+	// TODO(b/119580726): Support IPv6 header-included packets.
+	return tcpip.ErrNotSupported
+}
+
 // HandlePacket is called by the link layer when new ipv6 packets arrive for
 // this endpoint.
 func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) {
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index 0ecaa0833..462265281 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -174,6 +174,10 @@ type NetworkEndpoint interface {
 	// protocol.
 	WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, loop PacketLooping) *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
+
 	// ID returns the network protocol endpoint ID.
 	ID() *NetworkEndpointID
 
@@ -357,10 +361,19 @@ type TransportProtocolFactory func() TransportProtocol
 // instantiate network protocols.
 type NetworkProtocolFactory func() NetworkProtocol
 
+// 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)
+}
+
 var (
 	transportProtocols = make(map[string]TransportProtocolFactory)
 	networkProtocols   = make(map[string]NetworkProtocolFactory)
 
+	unassociatedFactory UnassociatedEndpointFactory
+
 	linkEPMu           sync.RWMutex
 	nextLinkEndpointID tcpip.LinkEndpointID = 1
 	linkEndpoints                           = make(map[tcpip.LinkEndpointID]LinkEndpoint)
@@ -380,6 +393,13 @@ func RegisterNetworkProtocolFactory(name string, p NetworkProtocolFactory) {
 	networkProtocols[name] = p
 }
 
+// RegisterUnassociatedFactory registers a factory to produce endpoints not
+// associated with any particular transport protocol. This function is intended
+// to be called by init() functions of the protocols.
+func RegisterUnassociatedFactory(f UnassociatedEndpointFactory) {
+	unassociatedFactory = f
+}
+
 // RegisterLinkEndpoint register a link-layer protocol endpoint and returns an
 // ID that can be used to refer to it.
 func RegisterLinkEndpoint(linkEP LinkEndpoint) tcpip.LinkEndpointID {
diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go
index 36d7b6ac7..391ab4344 100644
--- a/pkg/tcpip/stack/route.go
+++ b/pkg/tcpip/stack/route.go
@@ -163,6 +163,18 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec
 	return err
 }
 
+// WriteHeaderIncludedPacket writes a packet already containing a network
+// header through the given route.
+func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error {
+	if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil {
+		r.Stats().IP.OutgoingPacketErrors.Increment()
+		return err
+	}
+	r.ref.nic.stats.Tx.Packets.Increment()
+	r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payload.Size()))
+	return nil
+}
+
 // DefaultTTL returns the default TTL of the underlying network endpoint.
 func (r *Route) DefaultTTL() uint8 {
 	return r.ref.ep.DefaultTTL()
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index 2d7f56ca9..3e8fb2a6c 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -340,6 +340,8 @@ type Stack struct {
 	networkProtocols   map[tcpip.NetworkProtocolNumber]NetworkProtocol
 	linkAddrResolvers  map[tcpip.NetworkProtocolNumber]LinkAddressResolver
 
+	unassociatedFactory UnassociatedEndpointFactory
+
 	demux *transportDemuxer
 
 	stats tcpip.Stats
@@ -442,6 +444,8 @@ func New(network []string, transport []string, opts Options) *Stack {
 		}
 	}
 
+	s.unassociatedFactory = unassociatedFactory
+
 	// Create the global transport demuxer.
 	s.demux = newTransportDemuxer(s)
 
@@ -574,11 +578,15 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp
 // NewRawEndpoint creates a new raw transport layer endpoint of the given
 // 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) (tcpip.Endpoint, *tcpip.Error) {
+func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) {
 	if !s.raw {
 		return nil, tcpip.ErrNotPermitted
 	}
 
+	if !associated {
+		return s.unassociatedFactory.NewUnassociatedRawEndpoint(s, network, transport, waiterQueue)
+	}
+
 	t, ok := s.transportProtocols[transport]
 	if !ok {
 		return nil, tcpip.ErrUnknownProtocol
diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go
index 69884af03..959071dbe 100644
--- a/pkg/tcpip/stack/stack_test.go
+++ b/pkg/tcpip/stack/stack_test.go
@@ -137,6 +137,10 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr bu
 	return f.linkEP.WritePacket(r, gso, hdr, payload, fakeNetNumber)
 }
 
+func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error {
+	return tcpip.ErrNotSupported
+}
+
 func (*fakeNetworkEndpoint) Close() {}
 
 type fakeNetGoodOption bool
diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD
index 34a14bf7f..bc4b255b4 100644
--- a/pkg/tcpip/transport/raw/BUILD
+++ b/pkg/tcpip/transport/raw/BUILD
@@ -21,6 +21,7 @@ go_library(
         "endpoint.go",
         "endpoint_state.go",
         "packet_list.go",
+        "protocol.go",
     ],
     importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/raw",
     imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"],
diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go
index 42aded77f..a29587658 100644
--- a/pkg/tcpip/transport/raw/endpoint.go
+++ b/pkg/tcpip/transport/raw/endpoint.go
@@ -67,6 +67,7 @@ type endpoint struct {
 	netProto    tcpip.NetworkProtocolNumber
 	transProto  tcpip.TransportProtocolNumber
 	waiterQueue *waiter.Queue
+	associated  bool
 
 	// The following fields are used to manage the receive queue and are
 	// protected by rcvMu.
@@ -97,8 +98,12 @@ type endpoint struct {
 }
 
 // NewEndpoint returns a raw  endpoint for the given protocols.
-// TODO(b/129292371): IP_HDRINCL, IPPROTO_RAW, and AF_PACKET.
+// TODO(b/129292371): IP_HDRINCL and AF_PACKET.
 func NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+	return newEndpoint(stack, netProto, transProto, waiterQueue, true /* associated */)
+}
+
+func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) {
 	if netProto != header.IPv4ProtocolNumber {
 		return nil, tcpip.ErrUnknownProtocol
 	}
@@ -110,6 +115,16 @@ func NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, trans
 		waiterQueue:   waiterQueue,
 		rcvBufSizeMax: 32 * 1024,
 		sndBufSize:    32 * 1024,
+		associated:    associated,
+	}
+
+	// Unassociated endpoints are write-only and users call Write() with IP
+	// headers included. Because they're write-only, We don't need to
+	// register with the stack.
+	if !associated {
+		ep.rcvBufSizeMax = 0
+		ep.waiterQueue = nil
+		return ep, nil
 	}
 
 	if err := ep.stack.RegisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep); err != nil {
@@ -124,7 +139,7 @@ func (ep *endpoint) Close() {
 	ep.mu.Lock()
 	defer ep.mu.Unlock()
 
-	if ep.closed {
+	if ep.closed || !ep.associated {
 		return
 	}
 
@@ -142,8 +157,11 @@ func (ep *endpoint) Close() {
 
 	if ep.connected {
 		ep.route.Release()
+		ep.connected = false
 	}
 
+	ep.closed = true
+
 	ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut)
 }
 
@@ -152,6 +170,10 @@ func (ep *endpoint) ModerateRecvBuf(copied int) {}
 
 // Read implements tcpip.Endpoint.Read.
 func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
+	if !ep.associated {
+		return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidOptionValue
+	}
+
 	ep.rcvMu.Lock()
 
 	// If there's no data to read, return that read would block or that the
@@ -192,6 +214,33 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (uintp
 		return 0, nil, tcpip.ErrInvalidEndpointState
 	}
 
+	payloadBytes, err := payload.Get(payload.Size())
+	if err != nil {
+		ep.mu.RUnlock()
+		return 0, nil, err
+	}
+
+	// If this is an unassociated socket and callee provided a nonzero
+	// destination address, route using that address.
+	if !ep.associated {
+		ip := header.IPv4(payloadBytes)
+		if !ip.IsValid(payload.Size()) {
+			ep.mu.RUnlock()
+			return 0, nil, tcpip.ErrInvalidOptionValue
+		}
+		dstAddr := ip.DestinationAddress()
+		// Update dstAddr with the address in the IP header, unless
+		// opts.To is set (e.g. if sendto specifies a specific
+		// address).
+		if dstAddr != tcpip.Address([]byte{0, 0, 0, 0}) && opts.To == nil {
+			opts.To = &tcpip.FullAddress{
+				NIC:  0,       // NIC is unset.
+				Addr: dstAddr, // The address from the payload.
+				Port: 0,       // There are no ports here.
+			}
+		}
+	}
+
 	// Did the user caller provide a destination? If not, use the connected
 	// destination.
 	if opts.To == nil {
@@ -216,12 +265,12 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (uintp
 				return 0, nil, tcpip.ErrInvalidEndpointState
 			}
 
-			n, ch, err := ep.finishWrite(payload, savedRoute)
+			n, ch, err := ep.finishWrite(payloadBytes, savedRoute)
 			ep.mu.Unlock()
 			return n, ch, err
 		}
 
-		n, ch, err := ep.finishWrite(payload, &ep.route)
+		n, ch, err := ep.finishWrite(payloadBytes, &ep.route)
 		ep.mu.RUnlock()
 		return n, ch, err
 	}
@@ -248,7 +297,7 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (uintp
 		return 0, nil, err
 	}
 
-	n, ch, err := ep.finishWrite(payload, &route)
+	n, ch, err := ep.finishWrite(payloadBytes, &route)
 	route.Release()
 	ep.mu.RUnlock()
 	return n, ch, err
@@ -256,7 +305,7 @@ func (ep *endpoint) Write(payload tcpip.Payload, opts tcpip.WriteOptions) (uintp
 
 // finishWrite writes the payload to a route. It resolves the route if
 // necessary. It's really just a helper to make defer unnecessary in Write.
-func (ep *endpoint) finishWrite(payload tcpip.Payload, route *stack.Route) (uintptr, <-chan struct{}, *tcpip.Error) {
+func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (uintptr, <-chan struct{}, *tcpip.Error) {
 	// We may need to resolve the route (match a link layer address to the
 	// network address). If that requires blocking (e.g. to use ARP),
 	// return a channel on which the caller can wait.
@@ -269,13 +318,14 @@ func (ep *endpoint) finishWrite(payload tcpip.Payload, route *stack.Route) (uint
 		}
 	}
 
-	payloadBytes, err := payload.Get(payload.Size())
-	if err != nil {
-		return 0, nil, err
-	}
-
 	switch ep.netProto {
 	case header.IPv4ProtocolNumber:
+		if !ep.associated {
+			if err := route.WriteHeaderIncludedPacket(buffer.View(payloadBytes).ToVectorisedView()); err != nil {
+				return 0, nil, err
+			}
+			break
+		}
 		hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength()))
 		if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, route.DefaultTTL()); err != nil {
 			return 0, nil, err
@@ -335,15 +385,17 @@ func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
 	}
 	defer route.Release()
 
-	// Re-register the endpoint with the appropriate NIC.
-	if err := ep.stack.RegisterRawTransportEndpoint(addr.NIC, ep.netProto, ep.transProto, ep); err != nil {
-		return err
+	if ep.associated {
+		// Re-register the endpoint with the appropriate NIC.
+		if err := ep.stack.RegisterRawTransportEndpoint(addr.NIC, ep.netProto, ep.transProto, ep); err != nil {
+			return err
+		}
+		ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep)
+		ep.registeredNIC = nic
 	}
-	ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep)
 
-	// Save the route and NIC we've connected via.
+	// Save the route we've connected via.
 	ep.route = route.Clone()
-	ep.registeredNIC = nic
 	ep.connected = true
 
 	return nil
@@ -386,14 +438,16 @@ func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error {
 		return tcpip.ErrBadLocalAddress
 	}
 
-	// Re-register the endpoint with the appropriate NIC.
-	if err := ep.stack.RegisterRawTransportEndpoint(addr.NIC, ep.netProto, ep.transProto, ep); err != nil {
-		return err
+	if ep.associated {
+		// Re-register the endpoint with the appropriate NIC.
+		if err := ep.stack.RegisterRawTransportEndpoint(addr.NIC, ep.netProto, ep.transProto, ep); err != nil {
+			return err
+		}
+		ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep)
+		ep.registeredNIC = addr.NIC
+		ep.boundNIC = addr.NIC
 	}
-	ep.stack.UnregisterRawTransportEndpoint(ep.registeredNIC, ep.netProto, ep.transProto, ep)
 
-	ep.registeredNIC = addr.NIC
-	ep.boundNIC = addr.NIC
 	ep.boundAddr = addr.Addr
 	ep.bound = true
 
diff --git a/pkg/tcpip/transport/raw/protocol.go b/pkg/tcpip/transport/raw/protocol.go
new file mode 100644
index 000000000..783c21e6b
--- /dev/null
+++ b/pkg/tcpip/transport/raw/protocol.go
@@ -0,0 +1,32 @@
+// Copyright 2019 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 raw
+
+import (
+	"gvisor.dev/gvisor/pkg/tcpip"
+	"gvisor.dev/gvisor/pkg/tcpip/stack"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+type factory struct{}
+
+// NewUnassociatedRawEndpoint implements stack.UnassociatedEndpointFactory.
+func (factory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
+	return newEndpoint(stack, netProto, transProto, waiterQueue, false /* associated */)
+}
+
+func init() {
+	stack.RegisterUnassociatedFactory(factory{})
+}